Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * xlog.c
4 : * PostgreSQL write-ahead log manager
5 : *
6 : * The Write-Ahead Log (WAL) functionality is split into several source
7 : * files, in addition to this one:
8 : *
9 : * xloginsert.c - Functions for constructing WAL records
10 : * xlogrecovery.c - WAL recovery and standby code
11 : * xlogreader.c - Facility for reading WAL files and parsing WAL records
12 : * xlogutils.c - Helper functions for WAL redo routines
13 : *
14 : * This file contains functions for coordinating database startup and
15 : * checkpointing, and managing the write-ahead log buffers when the
16 : * system is running.
17 : *
18 : * StartupXLOG() is the main entry point of the startup process. It
19 : * coordinates database startup, performing WAL recovery, and the
20 : * transition from WAL recovery into normal operations.
21 : *
22 : * XLogInsertRecord() inserts a WAL record into the WAL buffers. Most
23 : * callers should not call this directly, but use the functions in
24 : * xloginsert.c to construct the WAL record. XLogFlush() can be used
25 : * to force the WAL to disk.
26 : *
27 : * In addition to those, there are many other functions for interrogating
28 : * the current system state, and for starting/stopping backups.
29 : *
30 : *
31 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
32 : * Portions Copyright (c) 1994, Regents of the University of California
33 : *
34 : * src/backend/access/transam/xlog.c
35 : *
36 : *-------------------------------------------------------------------------
37 : */
38 :
39 : #include "postgres.h"
40 :
41 : #include <ctype.h>
42 : #include <math.h>
43 : #include <time.h>
44 : #include <fcntl.h>
45 : #include <sys/stat.h>
46 : #include <sys/time.h>
47 : #include <unistd.h>
48 :
49 : #include "access/clog.h"
50 : #include "access/commit_ts.h"
51 : #include "access/heaptoast.h"
52 : #include "access/multixact.h"
53 : #include "access/rewriteheap.h"
54 : #include "access/subtrans.h"
55 : #include "access/timeline.h"
56 : #include "access/transam.h"
57 : #include "access/twophase.h"
58 : #include "access/xact.h"
59 : #include "access/xlog_internal.h"
60 : #include "access/xlogarchive.h"
61 : #include "access/xloginsert.h"
62 : #include "access/xlogreader.h"
63 : #include "access/xlogrecovery.h"
64 : #include "access/xlogutils.h"
65 : #include "backup/basebackup.h"
66 : #include "catalog/catversion.h"
67 : #include "catalog/pg_control.h"
68 : #include "catalog/pg_database.h"
69 : #include "common/controldata_utils.h"
70 : #include "common/file_utils.h"
71 : #include "executor/instrument.h"
72 : #include "miscadmin.h"
73 : #include "pg_trace.h"
74 : #include "pgstat.h"
75 : #include "port/atomics.h"
76 : #include "port/pg_iovec.h"
77 : #include "postmaster/bgwriter.h"
78 : #include "postmaster/startup.h"
79 : #include "postmaster/walsummarizer.h"
80 : #include "postmaster/walwriter.h"
81 : #include "replication/origin.h"
82 : #include "replication/slot.h"
83 : #include "replication/snapbuild.h"
84 : #include "replication/walreceiver.h"
85 : #include "replication/walsender.h"
86 : #include "storage/bufmgr.h"
87 : #include "storage/fd.h"
88 : #include "storage/ipc.h"
89 : #include "storage/large_object.h"
90 : #include "storage/latch.h"
91 : #include "storage/predicate.h"
92 : #include "storage/proc.h"
93 : #include "storage/procarray.h"
94 : #include "storage/reinit.h"
95 : #include "storage/spin.h"
96 : #include "storage/sync.h"
97 : #include "utils/guc_hooks.h"
98 : #include "utils/guc_tables.h"
99 : #include "utils/injection_point.h"
100 : #include "utils/memutils.h"
101 : #include "utils/ps_status.h"
102 : #include "utils/relmapper.h"
103 : #include "utils/snapmgr.h"
104 : #include "utils/timeout.h"
105 : #include "utils/timestamp.h"
106 : #include "utils/varlena.h"
107 :
108 : extern uint32 bootstrap_data_checksum_version;
109 :
110 : /* timeline ID to be used when bootstrapping */
111 : #define BootstrapTimeLineID 1
112 :
113 : /* User-settable parameters */
114 : int max_wal_size_mb = 1024; /* 1 GB */
115 : int min_wal_size_mb = 80; /* 80 MB */
116 : int wal_keep_size_mb = 0;
117 : int XLOGbuffers = -1;
118 : int XLogArchiveTimeout = 0;
119 : int XLogArchiveMode = ARCHIVE_MODE_OFF;
120 : char *XLogArchiveCommand = NULL;
121 : bool EnableHotStandby = false;
122 : bool fullPageWrites = true;
123 : bool wal_log_hints = false;
124 : int wal_compression = WAL_COMPRESSION_NONE;
125 : char *wal_consistency_checking_string = NULL;
126 : bool *wal_consistency_checking = NULL;
127 : bool wal_init_zero = true;
128 : bool wal_recycle = true;
129 : bool log_checkpoints = true;
130 : int wal_sync_method = DEFAULT_WAL_SYNC_METHOD;
131 : int wal_level = WAL_LEVEL_REPLICA;
132 : int CommitDelay = 0; /* precommit delay in microseconds */
133 : int CommitSiblings = 5; /* # concurrent xacts needed to sleep */
134 : int wal_retrieve_retry_interval = 5000;
135 : int max_slot_wal_keep_size_mb = -1;
136 : int wal_decode_buffer_size = 512 * 1024;
137 : bool track_wal_io_timing = false;
138 :
139 : #ifdef WAL_DEBUG
140 : bool XLOG_DEBUG = false;
141 : #endif
142 :
143 : int wal_segment_size = DEFAULT_XLOG_SEG_SIZE;
144 :
145 : /*
146 : * Number of WAL insertion locks to use. A higher value allows more insertions
147 : * to happen concurrently, but adds some CPU overhead to flushing the WAL,
148 : * which needs to iterate all the locks.
149 : */
150 : #define NUM_XLOGINSERT_LOCKS 8
151 :
152 : /*
153 : * Max distance from last checkpoint, before triggering a new xlog-based
154 : * checkpoint.
155 : */
156 : int CheckPointSegments;
157 :
158 : /* Estimated distance between checkpoints, in bytes */
159 : static double CheckPointDistanceEstimate = 0;
160 : static double PrevCheckPointDistance = 0;
161 :
162 : /*
163 : * Track whether there were any deferred checks for custom resource managers
164 : * specified in wal_consistency_checking.
165 : */
166 : static bool check_wal_consistency_checking_deferred = false;
167 :
168 : /*
169 : * GUC support
170 : */
171 : const struct config_enum_entry wal_sync_method_options[] = {
172 : {"fsync", WAL_SYNC_METHOD_FSYNC, false},
173 : #ifdef HAVE_FSYNC_WRITETHROUGH
174 : {"fsync_writethrough", WAL_SYNC_METHOD_FSYNC_WRITETHROUGH, false},
175 : #endif
176 : {"fdatasync", WAL_SYNC_METHOD_FDATASYNC, false},
177 : #ifdef O_SYNC
178 : {"open_sync", WAL_SYNC_METHOD_OPEN, false},
179 : #endif
180 : #ifdef O_DSYNC
181 : {"open_datasync", WAL_SYNC_METHOD_OPEN_DSYNC, false},
182 : #endif
183 : {NULL, 0, false}
184 : };
185 :
186 :
187 : /*
188 : * Although only "on", "off", and "always" are documented,
189 : * we accept all the likely variants of "on" and "off".
190 : */
191 : const struct config_enum_entry archive_mode_options[] = {
192 : {"always", ARCHIVE_MODE_ALWAYS, false},
193 : {"on", ARCHIVE_MODE_ON, false},
194 : {"off", ARCHIVE_MODE_OFF, false},
195 : {"true", ARCHIVE_MODE_ON, true},
196 : {"false", ARCHIVE_MODE_OFF, true},
197 : {"yes", ARCHIVE_MODE_ON, true},
198 : {"no", ARCHIVE_MODE_OFF, true},
199 : {"1", ARCHIVE_MODE_ON, true},
200 : {"0", ARCHIVE_MODE_OFF, true},
201 : {NULL, 0, false}
202 : };
203 :
204 : /*
205 : * Statistics for current checkpoint are collected in this global struct.
206 : * Because only the checkpointer or a stand-alone backend can perform
207 : * checkpoints, this will be unused in normal backends.
208 : */
209 : CheckpointStatsData CheckpointStats;
210 :
211 : /*
212 : * During recovery, lastFullPageWrites keeps track of full_page_writes that
213 : * the replayed WAL records indicate. It's initialized with full_page_writes
214 : * that the recovery starting checkpoint record indicates, and then updated
215 : * each time XLOG_FPW_CHANGE record is replayed.
216 : */
217 : static bool lastFullPageWrites;
218 :
219 : /*
220 : * Local copy of the state tracked by SharedRecoveryState in shared memory,
221 : * It is false if SharedRecoveryState is RECOVERY_STATE_DONE. True actually
222 : * means "not known, need to check the shared state".
223 : */
224 : static bool LocalRecoveryInProgress = true;
225 :
226 : /*
227 : * Local state for XLogInsertAllowed():
228 : * 1: unconditionally allowed to insert XLOG
229 : * 0: unconditionally not allowed to insert XLOG
230 : * -1: must check RecoveryInProgress(); disallow until it is false
231 : * Most processes start with -1 and transition to 1 after seeing that recovery
232 : * is not in progress. But we can also force the value for special cases.
233 : * The coding in XLogInsertAllowed() depends on the first two of these states
234 : * being numerically the same as bool true and false.
235 : */
236 : static int LocalXLogInsertAllowed = -1;
237 :
238 : /*
239 : * ProcLastRecPtr points to the start of the last XLOG record inserted by the
240 : * current backend. It is updated for all inserts. XactLastRecEnd points to
241 : * end+1 of the last record, and is reset when we end a top-level transaction,
242 : * or start a new one; so it can be used to tell if the current transaction has
243 : * created any XLOG records.
244 : *
245 : * While in parallel mode, this may not be fully up to date. When committing,
246 : * a transaction can assume this covers all xlog records written either by the
247 : * user backend or by any parallel worker which was present at any point during
248 : * the transaction. But when aborting, or when still in parallel mode, other
249 : * parallel backends may have written WAL records at later LSNs than the value
250 : * stored here. The parallel leader advances its own copy, when necessary,
251 : * in WaitForParallelWorkersToFinish.
252 : */
253 : XLogRecPtr ProcLastRecPtr = InvalidXLogRecPtr;
254 : XLogRecPtr XactLastRecEnd = InvalidXLogRecPtr;
255 : XLogRecPtr XactLastCommitEnd = InvalidXLogRecPtr;
256 :
257 : /*
258 : * RedoRecPtr is this backend's local copy of the REDO record pointer
259 : * (which is almost but not quite the same as a pointer to the most recent
260 : * CHECKPOINT record). We update this from the shared-memory copy,
261 : * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
262 : * hold an insertion lock). See XLogInsertRecord for details. We are also
263 : * allowed to update from XLogCtl->RedoRecPtr if we hold the info_lck;
264 : * see GetRedoRecPtr.
265 : *
266 : * NB: Code that uses this variable must be prepared not only for the
267 : * possibility that it may be arbitrarily out of date, but also for the
268 : * possibility that it might be set to InvalidXLogRecPtr. We used to
269 : * initialize it as a side effect of the first call to RecoveryInProgress(),
270 : * which meant that most code that might use it could assume that it had a
271 : * real if perhaps stale value. That's no longer the case.
272 : */
273 : static XLogRecPtr RedoRecPtr;
274 :
275 : /*
276 : * doPageWrites is this backend's local copy of (fullPageWrites ||
277 : * runningBackups > 0). It is used together with RedoRecPtr to decide whether
278 : * a full-page image of a page need to be taken.
279 : *
280 : * NB: Initially this is false, and there's no guarantee that it will be
281 : * initialized to any other value before it is first used. Any code that
282 : * makes use of it must recheck the value after obtaining a WALInsertLock,
283 : * and respond appropriately if it turns out that the previous value wasn't
284 : * accurate.
285 : */
286 : static bool doPageWrites;
287 :
288 : /*----------
289 : * Shared-memory data structures for XLOG control
290 : *
291 : * LogwrtRqst indicates a byte position that we need to write and/or fsync
292 : * the log up to (all records before that point must be written or fsynced).
293 : * The positions already written/fsynced are maintained in logWriteResult
294 : * and logFlushResult using atomic access.
295 : * In addition to the shared variable, each backend has a private copy of
296 : * both in LogwrtResult, which is updated when convenient.
297 : *
298 : * The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
299 : * (protected by info_lck), but we don't need to cache any copies of it.
300 : *
301 : * info_lck is only held long enough to read/update the protected variables,
302 : * so it's a plain spinlock. The other locks are held longer (potentially
303 : * over I/O operations), so we use LWLocks for them. These locks are:
304 : *
305 : * WALBufMappingLock: must be held to replace a page in the WAL buffer cache.
306 : * It is only held while initializing and changing the mapping. If the
307 : * contents of the buffer being replaced haven't been written yet, the mapping
308 : * lock is released while the write is done, and reacquired afterwards.
309 : *
310 : * WALWriteLock: must be held to write WAL buffers to disk (XLogWrite or
311 : * XLogFlush).
312 : *
313 : * ControlFileLock: must be held to read/update control file or create
314 : * new log file.
315 : *
316 : *----------
317 : */
318 :
319 : typedef struct XLogwrtRqst
320 : {
321 : XLogRecPtr Write; /* last byte + 1 to write out */
322 : XLogRecPtr Flush; /* last byte + 1 to flush */
323 : } XLogwrtRqst;
324 :
325 : typedef struct XLogwrtResult
326 : {
327 : XLogRecPtr Write; /* last byte + 1 written out */
328 : XLogRecPtr Flush; /* last byte + 1 flushed */
329 : } XLogwrtResult;
330 :
331 : /*
332 : * Inserting to WAL is protected by a small fixed number of WAL insertion
333 : * locks. To insert to the WAL, you must hold one of the locks - it doesn't
334 : * matter which one. To lock out other concurrent insertions, you must hold
335 : * of them. Each WAL insertion lock consists of a lightweight lock, plus an
336 : * indicator of how far the insertion has progressed (insertingAt).
337 : *
338 : * The insertingAt values are read when a process wants to flush WAL from
339 : * the in-memory buffers to disk, to check that all the insertions to the
340 : * region the process is about to write out have finished. You could simply
341 : * wait for all currently in-progress insertions to finish, but the
342 : * insertingAt indicator allows you to ignore insertions to later in the WAL,
343 : * so that you only wait for the insertions that are modifying the buffers
344 : * you're about to write out.
345 : *
346 : * This isn't just an optimization. If all the WAL buffers are dirty, an
347 : * inserter that's holding a WAL insert lock might need to evict an old WAL
348 : * buffer, which requires flushing the WAL. If it's possible for an inserter
349 : * to block on another inserter unnecessarily, deadlock can arise when two
350 : * inserters holding a WAL insert lock wait for each other to finish their
351 : * insertion.
352 : *
353 : * Small WAL records that don't cross a page boundary never update the value,
354 : * the WAL record is just copied to the page and the lock is released. But
355 : * to avoid the deadlock-scenario explained above, the indicator is always
356 : * updated before sleeping while holding an insertion lock.
357 : *
358 : * lastImportantAt contains the LSN of the last important WAL record inserted
359 : * using a given lock. This value is used to detect if there has been
360 : * important WAL activity since the last time some action, like a checkpoint,
361 : * was performed - allowing to not repeat the action if not. The LSN is
362 : * updated for all insertions, unless the XLOG_MARK_UNIMPORTANT flag was
363 : * set. lastImportantAt is never cleared, only overwritten by the LSN of newer
364 : * records. Tracking the WAL activity directly in WALInsertLock has the
365 : * advantage of not needing any additional locks to update the value.
366 : */
367 : typedef struct
368 : {
369 : LWLock lock;
370 : pg_atomic_uint64 insertingAt;
371 : XLogRecPtr lastImportantAt;
372 : } WALInsertLock;
373 :
374 : /*
375 : * All the WAL insertion locks are allocated as an array in shared memory. We
376 : * force the array stride to be a power of 2, which saves a few cycles in
377 : * indexing, but more importantly also ensures that individual slots don't
378 : * cross cache line boundaries. (Of course, we have to also ensure that the
379 : * array start address is suitably aligned.)
380 : */
381 : typedef union WALInsertLockPadded
382 : {
383 : WALInsertLock l;
384 : char pad[PG_CACHE_LINE_SIZE];
385 : } WALInsertLockPadded;
386 :
387 : /*
388 : * Session status of running backup, used for sanity checks in SQL-callable
389 : * functions to start and stop backups.
390 : */
391 : static SessionBackupState sessionBackupState = SESSION_BACKUP_NONE;
392 :
393 : /*
394 : * Shared state data for WAL insertion.
395 : */
396 : typedef struct XLogCtlInsert
397 : {
398 : slock_t insertpos_lck; /* protects CurrBytePos and PrevBytePos */
399 :
400 : /*
401 : * CurrBytePos is the end of reserved WAL. The next record will be
402 : * inserted at that position. PrevBytePos is the start position of the
403 : * previously inserted (or rather, reserved) record - it is copied to the
404 : * prev-link of the next record. These are stored as "usable byte
405 : * positions" rather than XLogRecPtrs (see XLogBytePosToRecPtr()).
406 : */
407 : uint64 CurrBytePos;
408 : uint64 PrevBytePos;
409 :
410 : /*
411 : * Make sure the above heavily-contended spinlock and byte positions are
412 : * on their own cache line. In particular, the RedoRecPtr and full page
413 : * write variables below should be on a different cache line. They are
414 : * read on every WAL insertion, but updated rarely, and we don't want
415 : * those reads to steal the cache line containing Curr/PrevBytePos.
416 : */
417 : char pad[PG_CACHE_LINE_SIZE];
418 :
419 : /*
420 : * fullPageWrites is the authoritative value used by all backends to
421 : * determine whether to write full-page image to WAL. This shared value,
422 : * instead of the process-local fullPageWrites, is required because, when
423 : * full_page_writes is changed by SIGHUP, we must WAL-log it before it
424 : * actually affects WAL-logging by backends. Checkpointer sets at startup
425 : * or after SIGHUP.
426 : *
427 : * To read these fields, you must hold an insertion lock. To modify them,
428 : * you must hold ALL the locks.
429 : */
430 : XLogRecPtr RedoRecPtr; /* current redo point for insertions */
431 : bool fullPageWrites;
432 :
433 : /*
434 : * runningBackups is a counter indicating the number of backups currently
435 : * in progress. lastBackupStart is the latest checkpoint redo location
436 : * used as a starting point for an online backup.
437 : */
438 : int runningBackups;
439 : XLogRecPtr lastBackupStart;
440 :
441 : /*
442 : * WAL insertion locks.
443 : */
444 : WALInsertLockPadded *WALInsertLocks;
445 : } XLogCtlInsert;
446 :
447 : /*
448 : * Total shared-memory state for XLOG.
449 : */
450 : typedef struct XLogCtlData
451 : {
452 : XLogCtlInsert Insert;
453 :
454 : /* Protected by info_lck: */
455 : XLogwrtRqst LogwrtRqst;
456 : XLogRecPtr RedoRecPtr; /* a recent copy of Insert->RedoRecPtr */
457 : FullTransactionId ckptFullXid; /* nextXid of latest checkpoint */
458 : XLogRecPtr asyncXactLSN; /* LSN of newest async commit/abort */
459 : XLogRecPtr replicationSlotMinLSN; /* oldest LSN needed by any slot */
460 :
461 : XLogSegNo lastRemovedSegNo; /* latest removed/recycled XLOG segment */
462 :
463 : /* Fake LSN counter, for unlogged relations. */
464 : pg_atomic_uint64 unloggedLSN;
465 :
466 : /* Time and LSN of last xlog segment switch. Protected by WALWriteLock. */
467 : pg_time_t lastSegSwitchTime;
468 : XLogRecPtr lastSegSwitchLSN;
469 :
470 : /* These are accessed using atomics -- info_lck not needed */
471 : pg_atomic_uint64 logInsertResult; /* last byte + 1 inserted to buffers */
472 : pg_atomic_uint64 logWriteResult; /* last byte + 1 written out */
473 : pg_atomic_uint64 logFlushResult; /* last byte + 1 flushed */
474 :
475 : /*
476 : * Latest initialized page in the cache (last byte position + 1).
477 : *
478 : * To change the identity of a buffer (and InitializedUpTo), you need to
479 : * hold WALBufMappingLock. To change the identity of a buffer that's
480 : * still dirty, the old page needs to be written out first, and for that
481 : * you need WALWriteLock, and you need to ensure that there are no
482 : * in-progress insertions to the page by calling
483 : * WaitXLogInsertionsToFinish().
484 : */
485 : XLogRecPtr InitializedUpTo;
486 :
487 : /*
488 : * These values do not change after startup, although the pointed-to pages
489 : * and xlblocks values certainly do. xlblocks values are protected by
490 : * WALBufMappingLock.
491 : */
492 : char *pages; /* buffers for unwritten XLOG pages */
493 : pg_atomic_uint64 *xlblocks; /* 1st byte ptr-s + XLOG_BLCKSZ */
494 : int XLogCacheBlck; /* highest allocated xlog buffer index */
495 :
496 : /*
497 : * InsertTimeLineID is the timeline into which new WAL is being inserted
498 : * and flushed. It is zero during recovery, and does not change once set.
499 : *
500 : * If we create a new timeline when the system was started up,
501 : * PrevTimeLineID is the old timeline's ID that we forked off from.
502 : * Otherwise it's equal to InsertTimeLineID.
503 : */
504 : TimeLineID InsertTimeLineID;
505 : TimeLineID PrevTimeLineID;
506 :
507 : /*
508 : * SharedRecoveryState indicates if we're still in crash or archive
509 : * recovery. Protected by info_lck.
510 : */
511 : RecoveryState SharedRecoveryState;
512 :
513 : /*
514 : * InstallXLogFileSegmentActive indicates whether the checkpointer should
515 : * arrange for future segments by recycling and/or PreallocXlogFiles().
516 : * Protected by ControlFileLock. Only the startup process changes it. If
517 : * true, anyone can use InstallXLogFileSegment(). If false, the startup
518 : * process owns the exclusive right to install segments, by reading from
519 : * the archive and possibly replacing existing files.
520 : */
521 : bool InstallXLogFileSegmentActive;
522 :
523 : /*
524 : * WalWriterSleeping indicates whether the WAL writer is currently in
525 : * low-power mode (and hence should be nudged if an async commit occurs).
526 : * Protected by info_lck.
527 : */
528 : bool WalWriterSleeping;
529 :
530 : /*
531 : * During recovery, we keep a copy of the latest checkpoint record here.
532 : * lastCheckPointRecPtr points to start of checkpoint record and
533 : * lastCheckPointEndPtr points to end+1 of checkpoint record. Used by the
534 : * checkpointer when it wants to create a restartpoint.
535 : *
536 : * Protected by info_lck.
537 : */
538 : XLogRecPtr lastCheckPointRecPtr;
539 : XLogRecPtr lastCheckPointEndPtr;
540 : CheckPoint lastCheckPoint;
541 :
542 : /*
543 : * lastFpwDisableRecPtr points to the start of the last replayed
544 : * XLOG_FPW_CHANGE record that instructs full_page_writes is disabled.
545 : */
546 : XLogRecPtr lastFpwDisableRecPtr;
547 :
548 : slock_t info_lck; /* locks shared variables shown above */
549 : } XLogCtlData;
550 :
551 : /*
552 : * Classification of XLogRecordInsert operations.
553 : */
554 : typedef enum
555 : {
556 : WALINSERT_NORMAL,
557 : WALINSERT_SPECIAL_SWITCH,
558 : WALINSERT_SPECIAL_CHECKPOINT
559 : } WalInsertClass;
560 :
561 : static XLogCtlData *XLogCtl = NULL;
562 :
563 : /* a private copy of XLogCtl->Insert.WALInsertLocks, for convenience */
564 : static WALInsertLockPadded *WALInsertLocks = NULL;
565 :
566 : /*
567 : * We maintain an image of pg_control in shared memory.
568 : */
569 : static ControlFileData *ControlFile = NULL;
570 :
571 : /*
572 : * Calculate the amount of space left on the page after 'endptr'. Beware
573 : * multiple evaluation!
574 : */
575 : #define INSERT_FREESPACE(endptr) \
576 : (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))
577 :
578 : /* Macro to advance to next buffer index. */
579 : #define NextBufIdx(idx) \
580 : (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
581 :
582 : /*
583 : * XLogRecPtrToBufIdx returns the index of the WAL buffer that holds, or
584 : * would hold if it was in cache, the page containing 'recptr'.
585 : */
586 : #define XLogRecPtrToBufIdx(recptr) \
587 : (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))
588 :
589 : /*
590 : * These are the number of bytes in a WAL page usable for WAL data.
591 : */
592 : #define UsableBytesInPage (XLOG_BLCKSZ - SizeOfXLogShortPHD)
593 :
594 : /*
595 : * Convert values of GUCs measured in megabytes to equiv. segment count.
596 : * Rounds down.
597 : */
598 : #define ConvertToXSegs(x, segsize) XLogMBVarToSegs((x), (segsize))
599 :
600 : /* The number of bytes in a WAL segment usable for WAL data. */
601 : static int UsableBytesInSegment;
602 :
603 : /*
604 : * Private, possibly out-of-date copy of shared LogwrtResult.
605 : * See discussion above.
606 : */
607 : static XLogwrtResult LogwrtResult = {0, 0};
608 :
609 : /*
610 : * Update local copy of shared XLogCtl->log{Write,Flush}Result
611 : *
612 : * It's critical that Flush always trails Write, so the order of the reads is
613 : * important, as is the barrier. See also XLogWrite.
614 : */
615 : #define RefreshXLogWriteResult(_target) \
616 : do { \
617 : _target.Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult); \
618 : pg_read_barrier(); \
619 : _target.Write = pg_atomic_read_u64(&XLogCtl->logWriteResult); \
620 : } while (0)
621 :
622 : /*
623 : * openLogFile is -1 or a kernel FD for an open log file segment.
624 : * openLogSegNo identifies the segment, and openLogTLI the corresponding TLI.
625 : * These variables are only used to write the XLOG, and so will normally refer
626 : * to the active segment.
627 : *
628 : * Note: call Reserve/ReleaseExternalFD to track consumption of this FD.
629 : */
630 : static int openLogFile = -1;
631 : static XLogSegNo openLogSegNo = 0;
632 : static TimeLineID openLogTLI = 0;
633 :
634 : /*
635 : * Local copies of equivalent fields in the control file. When running
636 : * crash recovery, LocalMinRecoveryPoint is set to InvalidXLogRecPtr as we
637 : * expect to replay all the WAL available, and updateMinRecoveryPoint is
638 : * switched to false to prevent any updates while replaying records.
639 : * Those values are kept consistent as long as crash recovery runs.
640 : */
641 : static XLogRecPtr LocalMinRecoveryPoint;
642 : static TimeLineID LocalMinRecoveryPointTLI;
643 : static bool updateMinRecoveryPoint = true;
644 :
645 : /* For WALInsertLockAcquire/Release functions */
646 : static int MyLockNo = 0;
647 : static bool holdingAllLocks = false;
648 :
649 : #ifdef WAL_DEBUG
650 : static MemoryContext walDebugCxt = NULL;
651 : #endif
652 :
653 : static void CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI,
654 : XLogRecPtr EndOfLog,
655 : TimeLineID newTLI);
656 : static void CheckRequiredParameterValues(void);
657 : static void XLogReportParameters(void);
658 : static int LocalSetXLogInsertAllowed(void);
659 : static void CreateEndOfRecoveryRecord(void);
660 : static XLogRecPtr CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn,
661 : XLogRecPtr pagePtr,
662 : TimeLineID newTLI);
663 : static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
664 : static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo);
665 : static XLogRecPtr XLogGetReplicationSlotMinimumLSN(void);
666 :
667 : static void AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli,
668 : bool opportunistic);
669 : static void XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible);
670 : static bool InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
671 : bool find_free, XLogSegNo max_segno,
672 : TimeLineID tli);
673 : static void XLogFileClose(void);
674 : static void PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli);
675 : static void RemoveTempXlogFiles(void);
676 : static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr,
677 : XLogRecPtr endptr, TimeLineID insertTLI);
678 : static void RemoveXlogFile(const struct dirent *segment_de,
679 : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
680 : TimeLineID insertTLI);
681 : static void UpdateLastRemovedPtr(char *filename);
682 : static void ValidateXLOGDirectoryStructure(void);
683 : static void CleanupBackupHistory(void);
684 : static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force);
685 : static bool PerformRecoveryXLogAction(void);
686 : static void InitControlFile(uint64 sysidentifier);
687 : static void WriteControlFile(void);
688 : static void ReadControlFile(void);
689 : static void UpdateControlFile(void);
690 : static char *str_time(pg_time_t tnow);
691 :
692 : static int get_sync_bit(int method);
693 :
694 : static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch,
695 : XLogRecData *rdata,
696 : XLogRecPtr StartPos, XLogRecPtr EndPos,
697 : TimeLineID tli);
698 : static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos,
699 : XLogRecPtr *EndPos, XLogRecPtr *PrevPtr);
700 : static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos,
701 : XLogRecPtr *PrevPtr);
702 : static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto);
703 : static char *GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli);
704 : static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos);
705 : static XLogRecPtr XLogBytePosToEndRecPtr(uint64 bytepos);
706 : static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr);
707 :
708 : static void WALInsertLockAcquire(void);
709 : static void WALInsertLockAcquireExclusive(void);
710 : static void WALInsertLockRelease(void);
711 : static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
712 :
713 : /*
714 : * Insert an XLOG record represented by an already-constructed chain of data
715 : * chunks. This is a low-level routine; to construct the WAL record header
716 : * and data, use the higher-level routines in xloginsert.c.
717 : *
718 : * If 'fpw_lsn' is valid, it is the oldest LSN among the pages that this
719 : * WAL record applies to, that were not included in the record as full page
720 : * images. If fpw_lsn <= RedoRecPtr, the function does not perform the
721 : * insertion and returns InvalidXLogRecPtr. The caller can then recalculate
722 : * which pages need a full-page image, and retry. If fpw_lsn is invalid, the
723 : * record is always inserted.
724 : *
725 : * 'flags' gives more in-depth control on the record being inserted. See
726 : * XLogSetRecordFlags() for details.
727 : *
728 : * 'topxid_included' tells whether the top-transaction id is logged along with
729 : * current subtransaction. See XLogRecordAssemble().
730 : *
731 : * The first XLogRecData in the chain must be for the record header, and its
732 : * data must be MAXALIGNed. XLogInsertRecord fills in the xl_prev and
733 : * xl_crc fields in the header, the rest of the header must already be filled
734 : * by the caller.
735 : *
736 : * Returns XLOG pointer to end of record (beginning of next record).
737 : * This can be used as LSN for data pages affected by the logged action.
738 : * (LSN is the XLOG point up to which the XLOG must be flushed to disk
739 : * before the data page can be written out. This implements the basic
740 : * WAL rule "write the log before the data".)
741 : */
742 : XLogRecPtr
743 25544364 : XLogInsertRecord(XLogRecData *rdata,
744 : XLogRecPtr fpw_lsn,
745 : uint8 flags,
746 : int num_fpi,
747 : bool topxid_included)
748 : {
749 25544364 : XLogCtlInsert *Insert = &XLogCtl->Insert;
750 : pg_crc32c rdata_crc;
751 : bool inserted;
752 25544364 : XLogRecord *rechdr = (XLogRecord *) rdata->data;
753 25544364 : uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
754 25544364 : WalInsertClass class = WALINSERT_NORMAL;
755 : XLogRecPtr StartPos;
756 : XLogRecPtr EndPos;
757 25544364 : bool prevDoPageWrites = doPageWrites;
758 : TimeLineID insertTLI;
759 :
760 : /* Does this record type require special handling? */
761 25544364 : if (unlikely(rechdr->xl_rmid == RM_XLOG_ID))
762 : {
763 340720 : if (info == XLOG_SWITCH)
764 666 : class = WALINSERT_SPECIAL_SWITCH;
765 340054 : else if (info == XLOG_CHECKPOINT_REDO)
766 666 : class = WALINSERT_SPECIAL_CHECKPOINT;
767 : }
768 :
769 : /* we assume that all of the record header is in the first chunk */
770 : Assert(rdata->len >= SizeOfXLogRecord);
771 :
772 : /* cross-check on whether we should be here or not */
773 25544364 : if (!XLogInsertAllowed())
774 0 : elog(ERROR, "cannot make new WAL entries during recovery");
775 :
776 : /*
777 : * Given that we're not in recovery, InsertTimeLineID is set and can't
778 : * change, so we can read it without a lock.
779 : */
780 25544364 : insertTLI = XLogCtl->InsertTimeLineID;
781 :
782 : /*----------
783 : *
784 : * We have now done all the preparatory work we can without holding a
785 : * lock or modifying shared state. From here on, inserting the new WAL
786 : * record to the shared WAL buffer cache is a two-step process:
787 : *
788 : * 1. Reserve the right amount of space from the WAL. The current head of
789 : * reserved space is kept in Insert->CurrBytePos, and is protected by
790 : * insertpos_lck.
791 : *
792 : * 2. Copy the record to the reserved WAL space. This involves finding the
793 : * correct WAL buffer containing the reserved space, and copying the
794 : * record in place. This can be done concurrently in multiple processes.
795 : *
796 : * To keep track of which insertions are still in-progress, each concurrent
797 : * inserter acquires an insertion lock. In addition to just indicating that
798 : * an insertion is in progress, the lock tells others how far the inserter
799 : * has progressed. There is a small fixed number of insertion locks,
800 : * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
801 : * boundary, it updates the value stored in the lock to the how far it has
802 : * inserted, to allow the previous buffer to be flushed.
803 : *
804 : * Holding onto an insertion lock also protects RedoRecPtr and
805 : * fullPageWrites from changing until the insertion is finished.
806 : *
807 : * Step 2 can usually be done completely in parallel. If the required WAL
808 : * page is not initialized yet, you have to grab WALBufMappingLock to
809 : * initialize it, but the WAL writer tries to do that ahead of insertions
810 : * to avoid that from happening in the critical path.
811 : *
812 : *----------
813 : */
814 25544364 : START_CRIT_SECTION();
815 :
816 25544364 : if (likely(class == WALINSERT_NORMAL))
817 : {
818 25543032 : WALInsertLockAcquire();
819 :
820 : /*
821 : * Check to see if my copy of RedoRecPtr is out of date. If so, may
822 : * have to go back and have the caller recompute everything. This can
823 : * only happen just after a checkpoint, so it's better to be slow in
824 : * this case and fast otherwise.
825 : *
826 : * Also check to see if fullPageWrites was just turned on or there's a
827 : * running backup (which forces full-page writes); if we weren't
828 : * already doing full-page writes then go back and recompute.
829 : *
830 : * If we aren't doing full-page writes then RedoRecPtr doesn't
831 : * actually affect the contents of the XLOG record, so we'll update
832 : * our local copy but not force a recomputation. (If doPageWrites was
833 : * just turned off, we could recompute the record without full pages,
834 : * but we choose not to bother.)
835 : */
836 25543032 : if (RedoRecPtr != Insert->RedoRecPtr)
837 : {
838 : Assert(RedoRecPtr < Insert->RedoRecPtr);
839 10050 : RedoRecPtr = Insert->RedoRecPtr;
840 : }
841 25543032 : doPageWrites = (Insert->fullPageWrites || Insert->runningBackups > 0);
842 :
843 25543032 : if (doPageWrites &&
844 25121604 : (!prevDoPageWrites ||
845 23904742 : (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr)))
846 : {
847 : /*
848 : * Oops, some buffer now needs to be backed up that the caller
849 : * didn't back up. Start over.
850 : */
851 11318 : WALInsertLockRelease();
852 11318 : END_CRIT_SECTION();
853 11318 : return InvalidXLogRecPtr;
854 : }
855 :
856 : /*
857 : * Reserve space for the record in the WAL. This also sets the xl_prev
858 : * pointer.
859 : */
860 25531714 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
861 : &rechdr->xl_prev);
862 :
863 : /* Normal records are always inserted. */
864 25531714 : inserted = true;
865 : }
866 1332 : else if (class == WALINSERT_SPECIAL_SWITCH)
867 : {
868 : /*
869 : * In order to insert an XLOG_SWITCH record, we need to hold all of
870 : * the WAL insertion locks, not just one, so that no one else can
871 : * begin inserting a record until we've figured out how much space
872 : * remains in the current WAL segment and claimed all of it.
873 : *
874 : * Nonetheless, this case is simpler than the normal cases handled
875 : * below, which must check for changes in doPageWrites and RedoRecPtr.
876 : * Those checks are only needed for records that can contain buffer
877 : * references, and an XLOG_SWITCH record never does.
878 : */
879 : Assert(fpw_lsn == InvalidXLogRecPtr);
880 666 : WALInsertLockAcquireExclusive();
881 666 : inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
882 : }
883 : else
884 : {
885 : Assert(class == WALINSERT_SPECIAL_CHECKPOINT);
886 :
887 : /*
888 : * We need to update both the local and shared copies of RedoRecPtr,
889 : * which means that we need to hold all the WAL insertion locks.
890 : * However, there can't be any buffer references, so as above, we need
891 : * not check RedoRecPtr before inserting the record; we just need to
892 : * update it afterwards.
893 : */
894 : Assert(fpw_lsn == InvalidXLogRecPtr);
895 666 : WALInsertLockAcquireExclusive();
896 666 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
897 : &rechdr->xl_prev);
898 666 : RedoRecPtr = Insert->RedoRecPtr = StartPos;
899 666 : inserted = true;
900 : }
901 :
902 25533046 : if (inserted)
903 : {
904 : /*
905 : * Now that xl_prev has been filled in, calculate CRC of the record
906 : * header.
907 : */
908 25532950 : rdata_crc = rechdr->xl_crc;
909 25532950 : COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
910 25532950 : FIN_CRC32C(rdata_crc);
911 25532950 : rechdr->xl_crc = rdata_crc;
912 :
913 : /*
914 : * All the record data, including the header, is now ready to be
915 : * inserted. Copy the record in the space reserved.
916 : */
917 25532950 : CopyXLogRecordToWAL(rechdr->xl_tot_len,
918 : class == WALINSERT_SPECIAL_SWITCH, rdata,
919 : StartPos, EndPos, insertTLI);
920 :
921 : /*
922 : * Unless record is flagged as not important, update LSN of last
923 : * important record in the current slot. When holding all locks, just
924 : * update the first one.
925 : */
926 25532950 : if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
927 : {
928 25383164 : int lockno = holdingAllLocks ? 0 : MyLockNo;
929 :
930 25383164 : WALInsertLocks[lockno].l.lastImportantAt = StartPos;
931 : }
932 : }
933 : else
934 : {
935 : /*
936 : * This was an xlog-switch record, but the current insert location was
937 : * already exactly at the beginning of a segment, so there was no need
938 : * to do anything.
939 : */
940 : }
941 :
942 : /*
943 : * Done! Let others know that we're finished.
944 : */
945 25533046 : WALInsertLockRelease();
946 :
947 25533046 : END_CRIT_SECTION();
948 :
949 25533046 : MarkCurrentTransactionIdLoggedIfAny();
950 :
951 : /*
952 : * Mark top transaction id is logged (if needed) so that we should not try
953 : * to log it again with the next WAL record in the current subtransaction.
954 : */
955 25533046 : if (topxid_included)
956 434 : MarkSubxactTopXidLogged();
957 :
958 : /*
959 : * Update shared LogwrtRqst.Write, if we crossed page boundary.
960 : */
961 25533046 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
962 : {
963 545620 : SpinLockAcquire(&XLogCtl->info_lck);
964 : /* advance global request to include new block(s) */
965 545620 : if (XLogCtl->LogwrtRqst.Write < EndPos)
966 545244 : XLogCtl->LogwrtRqst.Write = EndPos;
967 545620 : SpinLockRelease(&XLogCtl->info_lck);
968 545620 : RefreshXLogWriteResult(LogwrtResult);
969 : }
970 :
971 : /*
972 : * If this was an XLOG_SWITCH record, flush the record and the empty
973 : * padding space that fills the rest of the segment, and perform
974 : * end-of-segment actions (eg, notifying archiver).
975 : */
976 25533046 : if (class == WALINSERT_SPECIAL_SWITCH)
977 : {
978 : TRACE_POSTGRESQL_WAL_SWITCH();
979 666 : XLogFlush(EndPos);
980 :
981 : /*
982 : * Even though we reserved the rest of the segment for us, which is
983 : * reflected in EndPos, we return a pointer to just the end of the
984 : * xlog-switch record.
985 : */
986 666 : if (inserted)
987 : {
988 570 : EndPos = StartPos + SizeOfXLogRecord;
989 570 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
990 : {
991 0 : uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
992 :
993 0 : if (offset == EndPos % XLOG_BLCKSZ)
994 0 : EndPos += SizeOfXLogLongPHD;
995 : else
996 0 : EndPos += SizeOfXLogShortPHD;
997 : }
998 : }
999 : }
1000 :
1001 : #ifdef WAL_DEBUG
1002 : if (XLOG_DEBUG)
1003 : {
1004 : static XLogReaderState *debug_reader = NULL;
1005 : XLogRecord *record;
1006 : DecodedXLogRecord *decoded;
1007 : StringInfoData buf;
1008 : StringInfoData recordBuf;
1009 : char *errormsg = NULL;
1010 : MemoryContext oldCxt;
1011 :
1012 : oldCxt = MemoryContextSwitchTo(walDebugCxt);
1013 :
1014 : initStringInfo(&buf);
1015 : appendStringInfo(&buf, "INSERT @ %X/%X: ", LSN_FORMAT_ARGS(EndPos));
1016 :
1017 : /*
1018 : * We have to piece together the WAL record data from the XLogRecData
1019 : * entries, so that we can pass it to the rm_desc function as one
1020 : * contiguous chunk.
1021 : */
1022 : initStringInfo(&recordBuf);
1023 : for (; rdata != NULL; rdata = rdata->next)
1024 : appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1025 :
1026 : /* We also need temporary space to decode the record. */
1027 : record = (XLogRecord *) recordBuf.data;
1028 : decoded = (DecodedXLogRecord *)
1029 : palloc(DecodeXLogRecordRequiredSpace(record->xl_tot_len));
1030 :
1031 : if (!debug_reader)
1032 : debug_reader = XLogReaderAllocate(wal_segment_size, NULL,
1033 : XL_ROUTINE(.page_read = NULL,
1034 : .segment_open = NULL,
1035 : .segment_close = NULL),
1036 : NULL);
1037 : if (!debug_reader)
1038 : {
1039 : appendStringInfoString(&buf, "error decoding record: out of memory while allocating a WAL reading processor");
1040 : }
1041 : else if (!DecodeXLogRecord(debug_reader,
1042 : decoded,
1043 : record,
1044 : EndPos,
1045 : &errormsg))
1046 : {
1047 : appendStringInfo(&buf, "error decoding record: %s",
1048 : errormsg ? errormsg : "no error message");
1049 : }
1050 : else
1051 : {
1052 : appendStringInfoString(&buf, " - ");
1053 :
1054 : debug_reader->record = decoded;
1055 : xlog_outdesc(&buf, debug_reader);
1056 : debug_reader->record = NULL;
1057 : }
1058 : elog(LOG, "%s", buf.data);
1059 :
1060 : pfree(decoded);
1061 : pfree(buf.data);
1062 : pfree(recordBuf.data);
1063 : MemoryContextSwitchTo(oldCxt);
1064 : }
1065 : #endif
1066 :
1067 : /*
1068 : * Update our global variables
1069 : */
1070 25533046 : ProcLastRecPtr = StartPos;
1071 25533046 : XactLastRecEnd = EndPos;
1072 :
1073 : /* Report WAL traffic to the instrumentation. */
1074 25533046 : if (inserted)
1075 : {
1076 25532950 : pgWalUsage.wal_bytes += rechdr->xl_tot_len;
1077 25532950 : pgWalUsage.wal_records++;
1078 25532950 : pgWalUsage.wal_fpi += num_fpi;
1079 : }
1080 :
1081 25533046 : return EndPos;
1082 : }
1083 :
1084 : /*
1085 : * Reserves the right amount of space for a record of given size from the WAL.
1086 : * *StartPos is set to the beginning of the reserved section, *EndPos to
1087 : * its end+1. *PrevPtr is set to the beginning of the previous record; it is
1088 : * used to set the xl_prev of this record.
1089 : *
1090 : * This is the performance critical part of XLogInsert that must be serialized
1091 : * across backends. The rest can happen mostly in parallel. Try to keep this
1092 : * section as short as possible, insertpos_lck can be heavily contended on a
1093 : * busy system.
1094 : *
1095 : * NB: The space calculation here must match the code in CopyXLogRecordToWAL,
1096 : * where we actually copy the record to the reserved space.
1097 : *
1098 : * NB: Testing shows that XLogInsertRecord runs faster if this code is inlined;
1099 : * however, because there are two call sites, the compiler is reluctant to
1100 : * inline. We use pg_attribute_always_inline here to try to convince it.
1101 : */
1102 : static pg_attribute_always_inline void
1103 25532380 : ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos,
1104 : XLogRecPtr *PrevPtr)
1105 : {
1106 25532380 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1107 : uint64 startbytepos;
1108 : uint64 endbytepos;
1109 : uint64 prevbytepos;
1110 :
1111 25532380 : size = MAXALIGN(size);
1112 :
1113 : /* All (non xlog-switch) records should contain data. */
1114 : Assert(size > SizeOfXLogRecord);
1115 :
1116 : /*
1117 : * The duration the spinlock needs to be held is minimized by minimizing
1118 : * the calculations that have to be done while holding the lock. The
1119 : * current tip of reserved WAL is kept in CurrBytePos, as a byte position
1120 : * that only counts "usable" bytes in WAL, that is, it excludes all WAL
1121 : * page headers. The mapping between "usable" byte positions and physical
1122 : * positions (XLogRecPtrs) can be done outside the locked region, and
1123 : * because the usable byte position doesn't include any headers, reserving
1124 : * X bytes from WAL is almost as simple as "CurrBytePos += X".
1125 : */
1126 25532380 : SpinLockAcquire(&Insert->insertpos_lck);
1127 :
1128 25532380 : startbytepos = Insert->CurrBytePos;
1129 25532380 : endbytepos = startbytepos + size;
1130 25532380 : prevbytepos = Insert->PrevBytePos;
1131 25532380 : Insert->CurrBytePos = endbytepos;
1132 25532380 : Insert->PrevBytePos = startbytepos;
1133 :
1134 25532380 : SpinLockRelease(&Insert->insertpos_lck);
1135 :
1136 25532380 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1137 25532380 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1138 25532380 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1139 :
1140 : /*
1141 : * Check that the conversions between "usable byte positions" and
1142 : * XLogRecPtrs work consistently in both directions.
1143 : */
1144 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1145 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1146 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1147 25532380 : }
1148 :
1149 : /*
1150 : * Like ReserveXLogInsertLocation(), but for an xlog-switch record.
1151 : *
1152 : * A log-switch record is handled slightly differently. The rest of the
1153 : * segment will be reserved for this insertion, as indicated by the returned
1154 : * *EndPos value. However, if we are already at the beginning of the current
1155 : * segment, *StartPos and *EndPos are set to the current location without
1156 : * reserving any space, and the function returns false.
1157 : */
1158 : static bool
1159 666 : ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
1160 : {
1161 666 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1162 : uint64 startbytepos;
1163 : uint64 endbytepos;
1164 : uint64 prevbytepos;
1165 666 : uint32 size = MAXALIGN(SizeOfXLogRecord);
1166 : XLogRecPtr ptr;
1167 : uint32 segleft;
1168 :
1169 : /*
1170 : * These calculations are a bit heavy-weight to be done while holding a
1171 : * spinlock, but since we're holding all the WAL insertion locks, there
1172 : * are no other inserters competing for it. GetXLogInsertRecPtr() does
1173 : * compete for it, but that's not called very frequently.
1174 : */
1175 666 : SpinLockAcquire(&Insert->insertpos_lck);
1176 :
1177 666 : startbytepos = Insert->CurrBytePos;
1178 :
1179 666 : ptr = XLogBytePosToEndRecPtr(startbytepos);
1180 666 : if (XLogSegmentOffset(ptr, wal_segment_size) == 0)
1181 : {
1182 96 : SpinLockRelease(&Insert->insertpos_lck);
1183 96 : *EndPos = *StartPos = ptr;
1184 96 : return false;
1185 : }
1186 :
1187 570 : endbytepos = startbytepos + size;
1188 570 : prevbytepos = Insert->PrevBytePos;
1189 :
1190 570 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1191 570 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1192 :
1193 570 : segleft = wal_segment_size - XLogSegmentOffset(*EndPos, wal_segment_size);
1194 570 : if (segleft != wal_segment_size)
1195 : {
1196 : /* consume the rest of the segment */
1197 570 : *EndPos += segleft;
1198 570 : endbytepos = XLogRecPtrToBytePos(*EndPos);
1199 : }
1200 570 : Insert->CurrBytePos = endbytepos;
1201 570 : Insert->PrevBytePos = startbytepos;
1202 :
1203 570 : SpinLockRelease(&Insert->insertpos_lck);
1204 :
1205 570 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1206 :
1207 : Assert(XLogSegmentOffset(*EndPos, wal_segment_size) == 0);
1208 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1209 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1210 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1211 :
1212 570 : return true;
1213 : }
1214 :
1215 : /*
1216 : * Subroutine of XLogInsertRecord. Copies a WAL record to an already-reserved
1217 : * area in the WAL.
1218 : */
1219 : static void
1220 25532950 : CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata,
1221 : XLogRecPtr StartPos, XLogRecPtr EndPos, TimeLineID tli)
1222 : {
1223 : char *currpos;
1224 : int freespace;
1225 : int written;
1226 : XLogRecPtr CurrPos;
1227 : XLogPageHeader pagehdr;
1228 :
1229 : /*
1230 : * Get a pointer to the right place in the right WAL buffer to start
1231 : * inserting to.
1232 : */
1233 25532950 : CurrPos = StartPos;
1234 25532950 : currpos = GetXLogBuffer(CurrPos, tli);
1235 25532950 : freespace = INSERT_FREESPACE(CurrPos);
1236 :
1237 : /*
1238 : * there should be enough space for at least the first field (xl_tot_len)
1239 : * on this page.
1240 : */
1241 : Assert(freespace >= sizeof(uint32));
1242 :
1243 : /* Copy record data */
1244 25532950 : written = 0;
1245 113600412 : while (rdata != NULL)
1246 : {
1247 88067462 : char *rdata_data = rdata->data;
1248 88067462 : int rdata_len = rdata->len;
1249 :
1250 88768322 : while (rdata_len > freespace)
1251 : {
1252 : /*
1253 : * Write what fits on this page, and continue on the next page.
1254 : */
1255 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1256 700860 : memcpy(currpos, rdata_data, freespace);
1257 700860 : rdata_data += freespace;
1258 700860 : rdata_len -= freespace;
1259 700860 : written += freespace;
1260 700860 : CurrPos += freespace;
1261 :
1262 : /*
1263 : * Get pointer to beginning of next page, and set the xlp_rem_len
1264 : * in the page header. Set XLP_FIRST_IS_CONTRECORD.
1265 : *
1266 : * It's safe to set the contrecord flag and xlp_rem_len without a
1267 : * lock on the page. All the other flags were already set when the
1268 : * page was initialized, in AdvanceXLInsertBuffer, and we're the
1269 : * only backend that needs to set the contrecord flag.
1270 : */
1271 700860 : currpos = GetXLogBuffer(CurrPos, tli);
1272 700860 : pagehdr = (XLogPageHeader) currpos;
1273 700860 : pagehdr->xlp_rem_len = write_len - written;
1274 700860 : pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1275 :
1276 : /* skip over the page header */
1277 700860 : if (XLogSegmentOffset(CurrPos, wal_segment_size) == 0)
1278 : {
1279 916 : CurrPos += SizeOfXLogLongPHD;
1280 916 : currpos += SizeOfXLogLongPHD;
1281 : }
1282 : else
1283 : {
1284 699944 : CurrPos += SizeOfXLogShortPHD;
1285 699944 : currpos += SizeOfXLogShortPHD;
1286 : }
1287 700860 : freespace = INSERT_FREESPACE(CurrPos);
1288 : }
1289 :
1290 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1291 88067462 : memcpy(currpos, rdata_data, rdata_len);
1292 88067462 : currpos += rdata_len;
1293 88067462 : CurrPos += rdata_len;
1294 88067462 : freespace -= rdata_len;
1295 88067462 : written += rdata_len;
1296 :
1297 88067462 : rdata = rdata->next;
1298 : }
1299 : Assert(written == write_len);
1300 :
1301 : /*
1302 : * If this was an xlog-switch, it's not enough to write the switch record,
1303 : * we also have to consume all the remaining space in the WAL segment. We
1304 : * have already reserved that space, but we need to actually fill it.
1305 : */
1306 25532950 : if (isLogSwitch && XLogSegmentOffset(CurrPos, wal_segment_size) != 0)
1307 : {
1308 : /* An xlog-switch record doesn't contain any data besides the header */
1309 : Assert(write_len == SizeOfXLogRecord);
1310 :
1311 : /* Assert that we did reserve the right amount of space */
1312 : Assert(XLogSegmentOffset(EndPos, wal_segment_size) == 0);
1313 :
1314 : /* Use up all the remaining space on the current page */
1315 570 : CurrPos += freespace;
1316 :
1317 : /*
1318 : * Cause all remaining pages in the segment to be flushed, leaving the
1319 : * XLog position where it should be, at the start of the next segment.
1320 : * We do this one page at a time, to make sure we don't deadlock
1321 : * against ourselves if wal_buffers < wal_segment_size.
1322 : */
1323 899808 : while (CurrPos < EndPos)
1324 : {
1325 : /*
1326 : * The minimal action to flush the page would be to call
1327 : * WALInsertLockUpdateInsertingAt(CurrPos) followed by
1328 : * AdvanceXLInsertBuffer(...). The page would be left initialized
1329 : * mostly to zeros, except for the page header (always the short
1330 : * variant, as this is never a segment's first page).
1331 : *
1332 : * The large vistas of zeros are good for compressibility, but the
1333 : * headers interrupting them every XLOG_BLCKSZ (with values that
1334 : * differ from page to page) are not. The effect varies with
1335 : * compression tool, but bzip2 for instance compresses about an
1336 : * order of magnitude worse if those headers are left in place.
1337 : *
1338 : * Rather than complicating AdvanceXLInsertBuffer itself (which is
1339 : * called in heavily-loaded circumstances as well as this lightly-
1340 : * loaded one) with variant behavior, we just use GetXLogBuffer
1341 : * (which itself calls the two methods we need) to get the pointer
1342 : * and zero most of the page. Then we just zero the page header.
1343 : */
1344 899238 : currpos = GetXLogBuffer(CurrPos, tli);
1345 3596952 : MemSet(currpos, 0, SizeOfXLogShortPHD);
1346 :
1347 899238 : CurrPos += XLOG_BLCKSZ;
1348 : }
1349 : }
1350 : else
1351 : {
1352 : /* Align the end position, so that the next record starts aligned */
1353 25532380 : CurrPos = MAXALIGN64(CurrPos);
1354 : }
1355 :
1356 25532950 : if (CurrPos != EndPos)
1357 0 : ereport(PANIC,
1358 : errcode(ERRCODE_DATA_CORRUPTED),
1359 : errmsg_internal("space reserved for WAL record does not match what was written"));
1360 25532950 : }
1361 :
1362 : /*
1363 : * Acquire a WAL insertion lock, for inserting to WAL.
1364 : */
1365 : static void
1366 25543052 : WALInsertLockAcquire(void)
1367 : {
1368 : bool immed;
1369 :
1370 : /*
1371 : * It doesn't matter which of the WAL insertion locks we acquire, so try
1372 : * the one we used last time. If the system isn't particularly busy, it's
1373 : * a good bet that it's still available, and it's good to have some
1374 : * affinity to a particular lock so that you don't unnecessarily bounce
1375 : * cache lines between processes when there's no contention.
1376 : *
1377 : * If this is the first time through in this backend, pick a lock
1378 : * (semi-)randomly. This allows the locks to be used evenly if you have a
1379 : * lot of very short connections.
1380 : */
1381 : static int lockToTry = -1;
1382 :
1383 25543052 : if (lockToTry == -1)
1384 12354 : lockToTry = MyProcNumber % NUM_XLOGINSERT_LOCKS;
1385 25543052 : MyLockNo = lockToTry;
1386 :
1387 : /*
1388 : * The insertingAt value is initially set to 0, as we don't know our
1389 : * insert location yet.
1390 : */
1391 25543052 : immed = LWLockAcquire(&WALInsertLocks[MyLockNo].l.lock, LW_EXCLUSIVE);
1392 25543052 : if (!immed)
1393 : {
1394 : /*
1395 : * If we couldn't get the lock immediately, try another lock next
1396 : * time. On a system with more insertion locks than concurrent
1397 : * inserters, this causes all the inserters to eventually migrate to a
1398 : * lock that no-one else is using. On a system with more inserters
1399 : * than locks, it still helps to distribute the inserters evenly
1400 : * across the locks.
1401 : */
1402 874 : lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1403 : }
1404 25543052 : }
1405 :
1406 : /*
1407 : * Acquire all WAL insertion locks, to prevent other backends from inserting
1408 : * to WAL.
1409 : */
1410 : static void
1411 4784 : WALInsertLockAcquireExclusive(void)
1412 : {
1413 : int i;
1414 :
1415 : /*
1416 : * When holding all the locks, all but the last lock's insertingAt
1417 : * indicator is set to 0xFFFFFFFFFFFFFFFF, which is higher than any real
1418 : * XLogRecPtr value, to make sure that no-one blocks waiting on those.
1419 : */
1420 38272 : for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1421 : {
1422 33488 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1423 33488 : LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1424 33488 : &WALInsertLocks[i].l.insertingAt,
1425 : PG_UINT64_MAX);
1426 : }
1427 : /* Variable value reset to 0 at release */
1428 4784 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1429 :
1430 4784 : holdingAllLocks = true;
1431 4784 : }
1432 :
1433 : /*
1434 : * Release our insertion lock (or locks, if we're holding them all).
1435 : *
1436 : * NB: Reset all variables to 0, so they cause LWLockWaitForVar to block the
1437 : * next time the lock is acquired.
1438 : */
1439 : static void
1440 25547836 : WALInsertLockRelease(void)
1441 : {
1442 25547836 : if (holdingAllLocks)
1443 : {
1444 : int i;
1445 :
1446 43056 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1447 38272 : LWLockReleaseClearVar(&WALInsertLocks[i].l.lock,
1448 38272 : &WALInsertLocks[i].l.insertingAt,
1449 : 0);
1450 :
1451 4784 : holdingAllLocks = false;
1452 : }
1453 : else
1454 : {
1455 25543052 : LWLockReleaseClearVar(&WALInsertLocks[MyLockNo].l.lock,
1456 25543052 : &WALInsertLocks[MyLockNo].l.insertingAt,
1457 : 0);
1458 : }
1459 25547836 : }
1460 :
1461 : /*
1462 : * Update our insertingAt value, to let others know that we've finished
1463 : * inserting up to that point.
1464 : */
1465 : static void
1466 1383174 : WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
1467 : {
1468 1383174 : if (holdingAllLocks)
1469 : {
1470 : /*
1471 : * We use the last lock to mark our actual position, see comments in
1472 : * WALInsertLockAcquireExclusive.
1473 : */
1474 891522 : LWLockUpdateVar(&WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.lock,
1475 891522 : &WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.insertingAt,
1476 : insertingAt);
1477 : }
1478 : else
1479 491652 : LWLockUpdateVar(&WALInsertLocks[MyLockNo].l.lock,
1480 491652 : &WALInsertLocks[MyLockNo].l.insertingAt,
1481 : insertingAt);
1482 1383174 : }
1483 :
1484 : /*
1485 : * Wait for any WAL insertions < upto to finish.
1486 : *
1487 : * Returns the location of the oldest insertion that is still in-progress.
1488 : * Any WAL prior to that point has been fully copied into WAL buffers, and
1489 : * can be flushed out to disk. Because this waits for any insertions older
1490 : * than 'upto' to finish, the return value is always >= 'upto'.
1491 : *
1492 : * Note: When you are about to write out WAL, you must call this function
1493 : * *before* acquiring WALWriteLock, to avoid deadlocks. This function might
1494 : * need to wait for an insertion to finish (or at least advance to next
1495 : * uninitialized page), and the inserter might need to evict an old WAL buffer
1496 : * to make room for a new one, which in turn requires WALWriteLock.
1497 : */
1498 : static XLogRecPtr
1499 1250108 : WaitXLogInsertionsToFinish(XLogRecPtr upto)
1500 : {
1501 : uint64 bytepos;
1502 : XLogRecPtr inserted;
1503 : XLogRecPtr reservedUpto;
1504 : XLogRecPtr finishedUpto;
1505 1250108 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1506 : int i;
1507 :
1508 1250108 : if (MyProc == NULL)
1509 0 : elog(PANIC, "cannot wait without a PGPROC structure");
1510 :
1511 : /*
1512 : * Check if there's any work to do. Use a barrier to ensure we get the
1513 : * freshest value.
1514 : */
1515 1250108 : inserted = pg_atomic_read_membarrier_u64(&XLogCtl->logInsertResult);
1516 1250108 : if (upto <= inserted)
1517 922632 : return inserted;
1518 :
1519 : /* Read the current insert position */
1520 327476 : SpinLockAcquire(&Insert->insertpos_lck);
1521 327476 : bytepos = Insert->CurrBytePos;
1522 327476 : SpinLockRelease(&Insert->insertpos_lck);
1523 327476 : reservedUpto = XLogBytePosToEndRecPtr(bytepos);
1524 :
1525 : /*
1526 : * No-one should request to flush a piece of WAL that hasn't even been
1527 : * reserved yet. However, it can happen if there is a block with a bogus
1528 : * LSN on disk, for example. XLogFlush checks for that situation and
1529 : * complains, but only after the flush. Here we just assume that to mean
1530 : * that all WAL that has been reserved needs to be finished. In this
1531 : * corner-case, the return value can be smaller than 'upto' argument.
1532 : */
1533 327476 : if (upto > reservedUpto)
1534 : {
1535 0 : ereport(LOG,
1536 : (errmsg("request to flush past end of generated WAL; request %X/%X, current position %X/%X",
1537 : LSN_FORMAT_ARGS(upto), LSN_FORMAT_ARGS(reservedUpto))));
1538 0 : upto = reservedUpto;
1539 : }
1540 :
1541 : /*
1542 : * Loop through all the locks, sleeping on any in-progress insert older
1543 : * than 'upto'.
1544 : *
1545 : * finishedUpto is our return value, indicating the point upto which all
1546 : * the WAL insertions have been finished. Initialize it to the head of
1547 : * reserved WAL, and as we iterate through the insertion locks, back it
1548 : * out for any insertion that's still in progress.
1549 : */
1550 327476 : finishedUpto = reservedUpto;
1551 2947284 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1552 : {
1553 2619808 : XLogRecPtr insertingat = InvalidXLogRecPtr;
1554 :
1555 : do
1556 : {
1557 : /*
1558 : * See if this insertion is in progress. LWLockWaitForVar will
1559 : * wait for the lock to be released, or for the 'value' to be set
1560 : * by a LWLockUpdateVar call. When a lock is initially acquired,
1561 : * its value is 0 (InvalidXLogRecPtr), which means that we don't
1562 : * know where it's inserting yet. We will have to wait for it. If
1563 : * it's a small insertion, the record will most likely fit on the
1564 : * same page and the inserter will release the lock without ever
1565 : * calling LWLockUpdateVar. But if it has to sleep, it will
1566 : * advertise the insertion point with LWLockUpdateVar before
1567 : * sleeping.
1568 : *
1569 : * In this loop we are only waiting for insertions that started
1570 : * before WaitXLogInsertionsToFinish was called. The lack of
1571 : * memory barriers in the loop means that we might see locks as
1572 : * "unused" that have since become used. This is fine because
1573 : * they only can be used for later insertions that we would not
1574 : * want to wait on anyway. Not taking a lock to acquire the
1575 : * current insertingAt value means that we might see older
1576 : * insertingAt values. This is also fine, because if we read a
1577 : * value too old, we will add ourselves to the wait queue, which
1578 : * contains atomic operations.
1579 : */
1580 2623076 : if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1581 2623076 : &WALInsertLocks[i].l.insertingAt,
1582 : insertingat, &insertingat))
1583 : {
1584 : /* the lock was free, so no insertion in progress */
1585 1905138 : insertingat = InvalidXLogRecPtr;
1586 1905138 : break;
1587 : }
1588 :
1589 : /*
1590 : * This insertion is still in progress. Have to wait, unless the
1591 : * inserter has proceeded past 'upto'.
1592 : */
1593 717938 : } while (insertingat < upto);
1594 :
1595 2619808 : if (insertingat != InvalidXLogRecPtr && insertingat < finishedUpto)
1596 101702 : finishedUpto = insertingat;
1597 : }
1598 :
1599 : /*
1600 : * Advance the limit we know to have been inserted and return the freshest
1601 : * value we know of, which might be beyond what we requested if somebody
1602 : * is concurrently doing this with an 'upto' pointer ahead of us.
1603 : */
1604 327476 : finishedUpto = pg_atomic_monotonic_advance_u64(&XLogCtl->logInsertResult,
1605 : finishedUpto);
1606 :
1607 327476 : return finishedUpto;
1608 : }
1609 :
1610 : /*
1611 : * Get a pointer to the right location in the WAL buffer containing the
1612 : * given XLogRecPtr.
1613 : *
1614 : * If the page is not initialized yet, it is initialized. That might require
1615 : * evicting an old dirty buffer from the buffer cache, which means I/O.
1616 : *
1617 : * The caller must ensure that the page containing the requested location
1618 : * isn't evicted yet, and won't be evicted. The way to ensure that is to
1619 : * hold onto a WAL insertion lock with the insertingAt position set to
1620 : * something <= ptr. GetXLogBuffer() will update insertingAt if it needs
1621 : * to evict an old page from the buffer. (This means that once you call
1622 : * GetXLogBuffer() with a given 'ptr', you must not access anything before
1623 : * that point anymore, and must not call GetXLogBuffer() with an older 'ptr'
1624 : * later, because older buffers might be recycled already)
1625 : */
1626 : static char *
1627 27133068 : GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli)
1628 : {
1629 : int idx;
1630 : XLogRecPtr endptr;
1631 : static uint64 cachedPage = 0;
1632 : static char *cachedPos = NULL;
1633 : XLogRecPtr expectedEndPtr;
1634 :
1635 : /*
1636 : * Fast path for the common case that we need to access again the same
1637 : * page as last time.
1638 : */
1639 27133068 : if (ptr / XLOG_BLCKSZ == cachedPage)
1640 : {
1641 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1642 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1643 25362100 : return cachedPos + ptr % XLOG_BLCKSZ;
1644 : }
1645 :
1646 : /*
1647 : * The XLog buffer cache is organized so that a page is always loaded to a
1648 : * particular buffer. That way we can easily calculate the buffer a given
1649 : * page must be loaded into, from the XLogRecPtr alone.
1650 : */
1651 1770968 : idx = XLogRecPtrToBufIdx(ptr);
1652 :
1653 : /*
1654 : * See what page is loaded in the buffer at the moment. It could be the
1655 : * page we're looking for, or something older. It can't be anything newer
1656 : * - that would imply the page we're looking for has already been written
1657 : * out to disk and evicted, and the caller is responsible for making sure
1658 : * that doesn't happen.
1659 : *
1660 : * We don't hold a lock while we read the value. If someone is just about
1661 : * to initialize or has just initialized the page, it's possible that we
1662 : * get InvalidXLogRecPtr. That's ok, we'll grab the mapping lock (in
1663 : * AdvanceXLInsertBuffer) and retry if we see anything other than the page
1664 : * we're looking for.
1665 : */
1666 1770968 : expectedEndPtr = ptr;
1667 1770968 : expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1668 :
1669 1770968 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1670 1770968 : if (expectedEndPtr != endptr)
1671 : {
1672 : XLogRecPtr initializedUpto;
1673 :
1674 : /*
1675 : * Before calling AdvanceXLInsertBuffer(), which can block, let others
1676 : * know how far we're finished with inserting the record.
1677 : *
1678 : * NB: If 'ptr' points to just after the page header, advertise a
1679 : * position at the beginning of the page rather than 'ptr' itself. If
1680 : * there are no other insertions running, someone might try to flush
1681 : * up to our advertised location. If we advertised a position after
1682 : * the page header, someone might try to flush the page header, even
1683 : * though page might actually not be initialized yet. As the first
1684 : * inserter on the page, we are effectively responsible for making
1685 : * sure that it's initialized, before we let insertingAt to move past
1686 : * the page header.
1687 : */
1688 1383174 : if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
1689 11112 : XLogSegmentOffset(ptr, wal_segment_size) > XLOG_BLCKSZ)
1690 11112 : initializedUpto = ptr - SizeOfXLogShortPHD;
1691 1372062 : else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
1692 854 : XLogSegmentOffset(ptr, wal_segment_size) < XLOG_BLCKSZ)
1693 378 : initializedUpto = ptr - SizeOfXLogLongPHD;
1694 : else
1695 1371684 : initializedUpto = ptr;
1696 :
1697 1383174 : WALInsertLockUpdateInsertingAt(initializedUpto);
1698 :
1699 1383174 : AdvanceXLInsertBuffer(ptr, tli, false);
1700 1383174 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1701 :
1702 1383174 : if (expectedEndPtr != endptr)
1703 0 : elog(PANIC, "could not find WAL buffer for %X/%X",
1704 : LSN_FORMAT_ARGS(ptr));
1705 : }
1706 : else
1707 : {
1708 : /*
1709 : * Make sure the initialization of the page is visible to us, and
1710 : * won't arrive later to overwrite the WAL data we write on the page.
1711 : */
1712 387794 : pg_memory_barrier();
1713 : }
1714 :
1715 : /*
1716 : * Found the buffer holding this page. Return a pointer to the right
1717 : * offset within the page.
1718 : */
1719 1770968 : cachedPage = ptr / XLOG_BLCKSZ;
1720 1770968 : cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1721 :
1722 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1723 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1724 :
1725 1770968 : return cachedPos + ptr % XLOG_BLCKSZ;
1726 : }
1727 :
1728 : /*
1729 : * Read WAL data directly from WAL buffers, if available. Returns the number
1730 : * of bytes read successfully.
1731 : *
1732 : * Fewer than 'count' bytes may be read if some of the requested WAL data has
1733 : * already been evicted.
1734 : *
1735 : * No locks are taken.
1736 : *
1737 : * Caller should ensure that it reads no further than LogwrtResult.Write
1738 : * (which should have been updated by the caller when determining how far to
1739 : * read). The 'tli' argument is only used as a convenient safety check so that
1740 : * callers do not read from WAL buffers on a historical timeline.
1741 : */
1742 : Size
1743 52542 : WALReadFromBuffers(char *dstbuf, XLogRecPtr startptr, Size count,
1744 : TimeLineID tli)
1745 : {
1746 52542 : char *pdst = dstbuf;
1747 52542 : XLogRecPtr recptr = startptr;
1748 : XLogRecPtr inserted;
1749 52542 : Size nbytes = count;
1750 :
1751 52542 : if (RecoveryInProgress() || tli != GetWALInsertionTimeLine())
1752 1774 : return 0;
1753 :
1754 : Assert(!XLogRecPtrIsInvalid(startptr));
1755 :
1756 : /*
1757 : * Caller should ensure that the requested data has been inserted into WAL
1758 : * buffers before we try to read it.
1759 : */
1760 50768 : inserted = pg_atomic_read_u64(&XLogCtl->logInsertResult);
1761 50768 : if (startptr + count > inserted)
1762 0 : ereport(ERROR,
1763 : errmsg("cannot read past end of generated WAL: requested %X/%X, current position %X/%X",
1764 : LSN_FORMAT_ARGS(startptr + count),
1765 : LSN_FORMAT_ARGS(inserted)));
1766 :
1767 : /*
1768 : * Loop through the buffers without a lock. For each buffer, atomically
1769 : * read and verify the end pointer, then copy the data out, and finally
1770 : * re-read and re-verify the end pointer.
1771 : *
1772 : * Once a page is evicted, it never returns to the WAL buffers, so if the
1773 : * end pointer matches the expected end pointer before and after we copy
1774 : * the data, then the right page must have been present during the data
1775 : * copy. Read barriers are necessary to ensure that the data copy actually
1776 : * happens between the two verification steps.
1777 : *
1778 : * If either verification fails, we simply terminate the loop and return
1779 : * with the data that had been already copied out successfully.
1780 : */
1781 120830 : while (nbytes > 0)
1782 : {
1783 84580 : uint32 offset = recptr % XLOG_BLCKSZ;
1784 84580 : int idx = XLogRecPtrToBufIdx(recptr);
1785 : XLogRecPtr expectedEndPtr;
1786 : XLogRecPtr endptr;
1787 : const char *page;
1788 : const char *psrc;
1789 : Size npagebytes;
1790 :
1791 : /*
1792 : * Calculate the end pointer we expect in the xlblocks array if the
1793 : * correct page is present.
1794 : */
1795 84580 : expectedEndPtr = recptr + (XLOG_BLCKSZ - offset);
1796 :
1797 : /*
1798 : * First verification step: check that the correct page is present in
1799 : * the WAL buffers.
1800 : */
1801 84580 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1802 84580 : if (expectedEndPtr != endptr)
1803 14510 : break;
1804 :
1805 : /*
1806 : * The correct page is present (or was at the time the endptr was
1807 : * read; must re-verify later). Calculate pointer to source data and
1808 : * determine how much data to read from this page.
1809 : */
1810 70070 : page = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1811 70070 : psrc = page + offset;
1812 70070 : npagebytes = Min(nbytes, XLOG_BLCKSZ - offset);
1813 :
1814 : /*
1815 : * Ensure that the data copy and the first verification step are not
1816 : * reordered.
1817 : */
1818 70070 : pg_read_barrier();
1819 :
1820 : /* data copy */
1821 70070 : memcpy(pdst, psrc, npagebytes);
1822 :
1823 : /*
1824 : * Ensure that the data copy and the second verification step are not
1825 : * reordered.
1826 : */
1827 70070 : pg_read_barrier();
1828 :
1829 : /*
1830 : * Second verification step: check that the page we read from wasn't
1831 : * evicted while we were copying the data.
1832 : */
1833 70070 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1834 70070 : if (expectedEndPtr != endptr)
1835 8 : break;
1836 :
1837 70062 : pdst += npagebytes;
1838 70062 : recptr += npagebytes;
1839 70062 : nbytes -= npagebytes;
1840 : }
1841 :
1842 : Assert(pdst - dstbuf <= count);
1843 :
1844 50768 : return pdst - dstbuf;
1845 : }
1846 :
1847 : /*
1848 : * Converts a "usable byte position" to XLogRecPtr. A usable byte position
1849 : * is the position starting from the beginning of WAL, excluding all WAL
1850 : * page headers.
1851 : */
1852 : static XLogRecPtr
1853 51071568 : XLogBytePosToRecPtr(uint64 bytepos)
1854 : {
1855 : uint64 fullsegs;
1856 : uint64 fullpages;
1857 : uint64 bytesleft;
1858 : uint32 seg_offset;
1859 : XLogRecPtr result;
1860 :
1861 51071568 : fullsegs = bytepos / UsableBytesInSegment;
1862 51071568 : bytesleft = bytepos % UsableBytesInSegment;
1863 :
1864 51071568 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1865 : {
1866 : /* fits on first page of segment */
1867 87866 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1868 : }
1869 : else
1870 : {
1871 : /* account for the first page on segment with long header */
1872 50983702 : seg_offset = XLOG_BLCKSZ;
1873 50983702 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1874 :
1875 50983702 : fullpages = bytesleft / UsableBytesInPage;
1876 50983702 : bytesleft = bytesleft % UsableBytesInPage;
1877 :
1878 50983702 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1879 : }
1880 :
1881 51071568 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1882 :
1883 51071568 : return result;
1884 : }
1885 :
1886 : /*
1887 : * Like XLogBytePosToRecPtr, but if the position is at a page boundary,
1888 : * returns a pointer to the beginning of the page (ie. before page header),
1889 : * not to where the first xlog record on that page would go to. This is used
1890 : * when converting a pointer to the end of a record.
1891 : */
1892 : static XLogRecPtr
1893 25861092 : XLogBytePosToEndRecPtr(uint64 bytepos)
1894 : {
1895 : uint64 fullsegs;
1896 : uint64 fullpages;
1897 : uint64 bytesleft;
1898 : uint32 seg_offset;
1899 : XLogRecPtr result;
1900 :
1901 25861092 : fullsegs = bytepos / UsableBytesInSegment;
1902 25861092 : bytesleft = bytepos % UsableBytesInSegment;
1903 :
1904 25861092 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1905 : {
1906 : /* fits on first page of segment */
1907 132562 : if (bytesleft == 0)
1908 87766 : seg_offset = 0;
1909 : else
1910 44796 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1911 : }
1912 : else
1913 : {
1914 : /* account for the first page on segment with long header */
1915 25728530 : seg_offset = XLOG_BLCKSZ;
1916 25728530 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1917 :
1918 25728530 : fullpages = bytesleft / UsableBytesInPage;
1919 25728530 : bytesleft = bytesleft % UsableBytesInPage;
1920 :
1921 25728530 : if (bytesleft == 0)
1922 25072 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
1923 : else
1924 25703458 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1925 : }
1926 :
1927 25861092 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1928 :
1929 25861092 : return result;
1930 : }
1931 :
1932 : /*
1933 : * Convert an XLogRecPtr to a "usable byte position".
1934 : */
1935 : static uint64
1936 3410 : XLogRecPtrToBytePos(XLogRecPtr ptr)
1937 : {
1938 : uint64 fullsegs;
1939 : uint32 fullpages;
1940 : uint32 offset;
1941 : uint64 result;
1942 :
1943 3410 : XLByteToSeg(ptr, fullsegs, wal_segment_size);
1944 :
1945 3410 : fullpages = (XLogSegmentOffset(ptr, wal_segment_size)) / XLOG_BLCKSZ;
1946 3410 : offset = ptr % XLOG_BLCKSZ;
1947 :
1948 3410 : if (fullpages == 0)
1949 : {
1950 966 : result = fullsegs * UsableBytesInSegment;
1951 966 : if (offset > 0)
1952 : {
1953 : Assert(offset >= SizeOfXLogLongPHD);
1954 366 : result += offset - SizeOfXLogLongPHD;
1955 : }
1956 : }
1957 : else
1958 : {
1959 2444 : result = fullsegs * UsableBytesInSegment +
1960 2444 : (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
1961 2444 : (fullpages - 1) * UsableBytesInPage; /* full pages */
1962 2444 : if (offset > 0)
1963 : {
1964 : Assert(offset >= SizeOfXLogShortPHD);
1965 2426 : result += offset - SizeOfXLogShortPHD;
1966 : }
1967 : }
1968 :
1969 3410 : return result;
1970 : }
1971 :
1972 : /*
1973 : * Initialize XLOG buffers, writing out old buffers if they still contain
1974 : * unwritten data, upto the page containing 'upto'. Or if 'opportunistic' is
1975 : * true, initialize as many pages as we can without having to write out
1976 : * unwritten data. Any new pages are initialized to zeros, with pages headers
1977 : * initialized properly.
1978 : */
1979 : static void
1980 1385610 : AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
1981 : {
1982 1385610 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1983 : int nextidx;
1984 : XLogRecPtr OldPageRqstPtr;
1985 : XLogwrtRqst WriteRqst;
1986 1385610 : XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
1987 : XLogRecPtr NewPageBeginPtr;
1988 : XLogPageHeader NewPage;
1989 1385610 : int npages pg_attribute_unused() = 0;
1990 :
1991 1385610 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
1992 :
1993 : /*
1994 : * Now that we have the lock, check if someone initialized the page
1995 : * already.
1996 : */
1997 4094350 : while (upto >= XLogCtl->InitializedUpTo || opportunistic)
1998 : {
1999 2711176 : nextidx = XLogRecPtrToBufIdx(XLogCtl->InitializedUpTo);
2000 :
2001 : /*
2002 : * Get ending-offset of the buffer page we need to replace (this may
2003 : * be zero if the buffer hasn't been used yet). Fall through if it's
2004 : * already written out.
2005 : */
2006 2711176 : OldPageRqstPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[nextidx]);
2007 2711176 : if (LogwrtResult.Write < OldPageRqstPtr)
2008 : {
2009 : /*
2010 : * Nope, got work to do. If we just want to pre-initialize as much
2011 : * as we can without flushing, give up now.
2012 : */
2013 1027314 : if (opportunistic)
2014 2436 : break;
2015 :
2016 : /* Advance shared memory write request position */
2017 1024878 : SpinLockAcquire(&XLogCtl->info_lck);
2018 1024878 : if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2019 846986 : XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2020 1024878 : SpinLockRelease(&XLogCtl->info_lck);
2021 :
2022 : /*
2023 : * Acquire an up-to-date LogwrtResult value and see if we still
2024 : * need to write it or if someone else already did.
2025 : */
2026 1024878 : RefreshXLogWriteResult(LogwrtResult);
2027 1024878 : if (LogwrtResult.Write < OldPageRqstPtr)
2028 : {
2029 : /*
2030 : * Must acquire write lock. Release WALBufMappingLock first,
2031 : * to make sure that all insertions that we need to wait for
2032 : * can finish (up to this same position). Otherwise we risk
2033 : * deadlock.
2034 : */
2035 1020988 : LWLockRelease(WALBufMappingLock);
2036 :
2037 1020988 : WaitXLogInsertionsToFinish(OldPageRqstPtr);
2038 :
2039 1020988 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2040 :
2041 1020988 : RefreshXLogWriteResult(LogwrtResult);
2042 1020988 : if (LogwrtResult.Write >= OldPageRqstPtr)
2043 : {
2044 : /* OK, someone wrote it already */
2045 4038 : LWLockRelease(WALWriteLock);
2046 : }
2047 : else
2048 : {
2049 : /* Have to write it ourselves */
2050 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2051 1016950 : WriteRqst.Write = OldPageRqstPtr;
2052 1016950 : WriteRqst.Flush = 0;
2053 1016950 : XLogWrite(WriteRqst, tli, false);
2054 1016950 : LWLockRelease(WALWriteLock);
2055 1016950 : PendingWalStats.wal_buffers_full++;
2056 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE();
2057 : }
2058 : /* Re-acquire WALBufMappingLock and retry */
2059 1020988 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2060 1020988 : continue;
2061 : }
2062 : }
2063 :
2064 : /*
2065 : * Now the next buffer slot is free and we can set it up to be the
2066 : * next output page.
2067 : */
2068 1687752 : NewPageBeginPtr = XLogCtl->InitializedUpTo;
2069 1687752 : NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2070 :
2071 : Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2072 :
2073 1687752 : NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
2074 :
2075 : /*
2076 : * Mark the xlblock with InvalidXLogRecPtr and issue a write barrier
2077 : * before initializing. Otherwise, the old page may be partially
2078 : * zeroed but look valid.
2079 : */
2080 1687752 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], InvalidXLogRecPtr);
2081 1687752 : pg_write_barrier();
2082 :
2083 : /*
2084 : * Be sure to re-zero the buffer so that bytes beyond what we've
2085 : * written will look like zeroes and not valid XLOG records...
2086 : */
2087 1687752 : MemSet((char *) NewPage, 0, XLOG_BLCKSZ);
2088 :
2089 : /*
2090 : * Fill the new page's header
2091 : */
2092 1687752 : NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2093 :
2094 : /* NewPage->xlp_info = 0; */ /* done by memset */
2095 1687752 : NewPage->xlp_tli = tli;
2096 1687752 : NewPage->xlp_pageaddr = NewPageBeginPtr;
2097 :
2098 : /* NewPage->xlp_rem_len = 0; */ /* done by memset */
2099 :
2100 : /*
2101 : * If online backup is not in progress, mark the header to indicate
2102 : * that WAL records beginning in this page have removable backup
2103 : * blocks. This allows the WAL archiver to know whether it is safe to
2104 : * compress archived WAL data by transforming full-block records into
2105 : * the non-full-block format. It is sufficient to record this at the
2106 : * page level because we force a page switch (in fact a segment
2107 : * switch) when starting a backup, so the flag will be off before any
2108 : * records can be written during the backup. At the end of a backup,
2109 : * the last page will be marked as all unsafe when perhaps only part
2110 : * is unsafe, but at worst the archiver would miss the opportunity to
2111 : * compress a few records.
2112 : */
2113 1687752 : if (Insert->runningBackups == 0)
2114 1411488 : NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2115 :
2116 : /*
2117 : * If first page of an XLOG segment file, make it a long header.
2118 : */
2119 1687752 : if ((XLogSegmentOffset(NewPage->xlp_pageaddr, wal_segment_size)) == 0)
2120 : {
2121 1328 : XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2122 :
2123 1328 : NewLongPage->xlp_sysid = ControlFile->system_identifier;
2124 1328 : NewLongPage->xlp_seg_size = wal_segment_size;
2125 1328 : NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
2126 1328 : NewPage->xlp_info |= XLP_LONG_HEADER;
2127 : }
2128 :
2129 : /*
2130 : * Make sure the initialization of the page becomes visible to others
2131 : * before the xlblocks update. GetXLogBuffer() reads xlblocks without
2132 : * holding a lock.
2133 : */
2134 1687752 : pg_write_barrier();
2135 :
2136 1687752 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], NewPageEndPtr);
2137 1687752 : XLogCtl->InitializedUpTo = NewPageEndPtr;
2138 :
2139 1687752 : npages++;
2140 : }
2141 1385610 : LWLockRelease(WALBufMappingLock);
2142 :
2143 : #ifdef WAL_DEBUG
2144 : if (XLOG_DEBUG && npages > 0)
2145 : {
2146 : elog(DEBUG1, "initialized %d pages, up to %X/%X",
2147 : npages, LSN_FORMAT_ARGS(NewPageEndPtr));
2148 : }
2149 : #endif
2150 1385610 : }
2151 :
2152 : /*
2153 : * Calculate CheckPointSegments based on max_wal_size_mb and
2154 : * checkpoint_completion_target.
2155 : */
2156 : static void
2157 12972 : CalculateCheckpointSegments(void)
2158 : {
2159 : double target;
2160 :
2161 : /*-------
2162 : * Calculate the distance at which to trigger a checkpoint, to avoid
2163 : * exceeding max_wal_size_mb. This is based on two assumptions:
2164 : *
2165 : * a) we keep WAL for only one checkpoint cycle (prior to PG11 we kept
2166 : * WAL for two checkpoint cycles to allow us to recover from the
2167 : * secondary checkpoint if the first checkpoint failed, though we
2168 : * only did this on the primary anyway, not on standby. Keeping just
2169 : * one checkpoint simplifies processing and reduces disk space in
2170 : * many smaller databases.)
2171 : * b) during checkpoint, we consume checkpoint_completion_target *
2172 : * number of segments consumed between checkpoints.
2173 : *-------
2174 : */
2175 12972 : target = (double) ConvertToXSegs(max_wal_size_mb, wal_segment_size) /
2176 12972 : (1.0 + CheckPointCompletionTarget);
2177 :
2178 : /* round down */
2179 12972 : CheckPointSegments = (int) target;
2180 :
2181 12972 : if (CheckPointSegments < 1)
2182 18 : CheckPointSegments = 1;
2183 12972 : }
2184 :
2185 : void
2186 9522 : assign_max_wal_size(int newval, void *extra)
2187 : {
2188 9522 : max_wal_size_mb = newval;
2189 9522 : CalculateCheckpointSegments();
2190 9522 : }
2191 :
2192 : void
2193 1830 : assign_checkpoint_completion_target(double newval, void *extra)
2194 : {
2195 1830 : CheckPointCompletionTarget = newval;
2196 1830 : CalculateCheckpointSegments();
2197 1830 : }
2198 :
2199 : bool
2200 3532 : check_wal_segment_size(int *newval, void **extra, GucSource source)
2201 : {
2202 3532 : if (!IsValidWalSegSize(*newval))
2203 : {
2204 0 : GUC_check_errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.");
2205 0 : return false;
2206 : }
2207 :
2208 3532 : return true;
2209 : }
2210 :
2211 : /*
2212 : * GUC check_hook for max_slot_wal_keep_size
2213 : *
2214 : * We don't allow the value of max_slot_wal_keep_size other than -1 during the
2215 : * binary upgrade. See start_postmaster() in pg_upgrade for more details.
2216 : */
2217 : bool
2218 2038 : check_max_slot_wal_keep_size(int *newval, void **extra, GucSource source)
2219 : {
2220 2038 : if (IsBinaryUpgrade && *newval != -1)
2221 : {
2222 0 : GUC_check_errdetail("\"%s\" must be set to -1 during binary upgrade mode.",
2223 : "max_slot_wal_keep_size");
2224 0 : return false;
2225 : }
2226 :
2227 2038 : return true;
2228 : }
2229 :
2230 : /*
2231 : * At a checkpoint, how many WAL segments to recycle as preallocated future
2232 : * XLOG segments? Returns the highest segment that should be preallocated.
2233 : */
2234 : static XLogSegNo
2235 1704 : XLOGfileslop(XLogRecPtr lastredoptr)
2236 : {
2237 : XLogSegNo minSegNo;
2238 : XLogSegNo maxSegNo;
2239 : double distance;
2240 : XLogSegNo recycleSegNo;
2241 :
2242 : /*
2243 : * Calculate the segment numbers that min_wal_size_mb and max_wal_size_mb
2244 : * correspond to. Always recycle enough segments to meet the minimum, and
2245 : * remove enough segments to stay below the maximum.
2246 : */
2247 1704 : minSegNo = lastredoptr / wal_segment_size +
2248 1704 : ConvertToXSegs(min_wal_size_mb, wal_segment_size) - 1;
2249 1704 : maxSegNo = lastredoptr / wal_segment_size +
2250 1704 : ConvertToXSegs(max_wal_size_mb, wal_segment_size) - 1;
2251 :
2252 : /*
2253 : * Between those limits, recycle enough segments to get us through to the
2254 : * estimated end of next checkpoint.
2255 : *
2256 : * To estimate where the next checkpoint will finish, assume that the
2257 : * system runs steadily consuming CheckPointDistanceEstimate bytes between
2258 : * every checkpoint.
2259 : */
2260 1704 : distance = (1.0 + CheckPointCompletionTarget) * CheckPointDistanceEstimate;
2261 : /* add 10% for good measure. */
2262 1704 : distance *= 1.10;
2263 :
2264 1704 : recycleSegNo = (XLogSegNo) ceil(((double) lastredoptr + distance) /
2265 : wal_segment_size);
2266 :
2267 1704 : if (recycleSegNo < minSegNo)
2268 1378 : recycleSegNo = minSegNo;
2269 1704 : if (recycleSegNo > maxSegNo)
2270 128 : recycleSegNo = maxSegNo;
2271 :
2272 1704 : return recycleSegNo;
2273 : }
2274 :
2275 : /*
2276 : * Check whether we've consumed enough xlog space that a checkpoint is needed.
2277 : *
2278 : * new_segno indicates a log file that has just been filled up (or read
2279 : * during recovery). We measure the distance from RedoRecPtr to new_segno
2280 : * and see if that exceeds CheckPointSegments.
2281 : *
2282 : * Note: it is caller's responsibility that RedoRecPtr is up-to-date.
2283 : */
2284 : bool
2285 1912 : XLogCheckpointNeeded(XLogSegNo new_segno)
2286 : {
2287 : XLogSegNo old_segno;
2288 :
2289 1912 : XLByteToSeg(RedoRecPtr, old_segno, wal_segment_size);
2290 :
2291 1912 : if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
2292 302 : return true;
2293 1610 : return false;
2294 : }
2295 :
2296 : /*
2297 : * Write and/or fsync the log at least as far as WriteRqst indicates.
2298 : *
2299 : * If flexible == true, we don't have to write as far as WriteRqst, but
2300 : * may stop at any convenient boundary (such as a cache or logfile boundary).
2301 : * This option allows us to avoid uselessly issuing multiple writes when a
2302 : * single one would do.
2303 : *
2304 : * Must be called with WALWriteLock held. WaitXLogInsertionsToFinish(WriteRqst)
2305 : * must be called before grabbing the lock, to make sure the data is ready to
2306 : * write.
2307 : */
2308 : static void
2309 1240536 : XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
2310 : {
2311 : bool ispartialpage;
2312 : bool last_iteration;
2313 : bool finishing_seg;
2314 : int curridx;
2315 : int npages;
2316 : int startidx;
2317 : uint32 startoffset;
2318 :
2319 : /* We should always be inside a critical section here */
2320 : Assert(CritSectionCount > 0);
2321 :
2322 : /*
2323 : * Update local LogwrtResult (caller probably did this already, but...)
2324 : */
2325 1240536 : RefreshXLogWriteResult(LogwrtResult);
2326 :
2327 : /*
2328 : * Since successive pages in the xlog cache are consecutively allocated,
2329 : * we can usually gather multiple pages together and issue just one
2330 : * write() call. npages is the number of pages we have determined can be
2331 : * written together; startidx is the cache block index of the first one,
2332 : * and startoffset is the file offset at which it should go. The latter
2333 : * two variables are only valid when npages > 0, but we must initialize
2334 : * all of them to keep the compiler quiet.
2335 : */
2336 1240536 : npages = 0;
2337 1240536 : startidx = 0;
2338 1240536 : startoffset = 0;
2339 :
2340 : /*
2341 : * Within the loop, curridx is the cache block index of the page to
2342 : * consider writing. Begin at the buffer containing the next unwritten
2343 : * page, or last partially written page.
2344 : */
2345 1240536 : curridx = XLogRecPtrToBufIdx(LogwrtResult.Write);
2346 :
2347 2858038 : while (LogwrtResult.Write < WriteRqst.Write)
2348 : {
2349 : /*
2350 : * Make sure we're not ahead of the insert process. This could happen
2351 : * if we're passed a bogus WriteRqst.Write that is past the end of the
2352 : * last page that's been initialized by AdvanceXLInsertBuffer.
2353 : */
2354 1840028 : XLogRecPtr EndPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[curridx]);
2355 :
2356 1840028 : if (LogwrtResult.Write >= EndPtr)
2357 0 : elog(PANIC, "xlog write request %X/%X is past end of log %X/%X",
2358 : LSN_FORMAT_ARGS(LogwrtResult.Write),
2359 : LSN_FORMAT_ARGS(EndPtr));
2360 :
2361 : /* Advance LogwrtResult.Write to end of current buffer page */
2362 1840028 : LogwrtResult.Write = EndPtr;
2363 1840028 : ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2364 :
2365 1840028 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2366 : wal_segment_size))
2367 : {
2368 : /*
2369 : * Switch to new logfile segment. We cannot have any pending
2370 : * pages here (since we dump what we have at segment end).
2371 : */
2372 : Assert(npages == 0);
2373 14226 : if (openLogFile >= 0)
2374 2942 : XLogFileClose();
2375 14226 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2376 : wal_segment_size);
2377 14226 : openLogTLI = tli;
2378 :
2379 : /* create/use new log file */
2380 14226 : openLogFile = XLogFileInit(openLogSegNo, tli);
2381 14226 : ReserveExternalFD();
2382 : }
2383 :
2384 : /* Make sure we have the current logfile open */
2385 1840028 : if (openLogFile < 0)
2386 : {
2387 0 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2388 : wal_segment_size);
2389 0 : openLogTLI = tli;
2390 0 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2391 0 : ReserveExternalFD();
2392 : }
2393 :
2394 : /* Add current page to the set of pending pages-to-dump */
2395 1840028 : if (npages == 0)
2396 : {
2397 : /* first of group */
2398 1249796 : startidx = curridx;
2399 1249796 : startoffset = XLogSegmentOffset(LogwrtResult.Write - XLOG_BLCKSZ,
2400 : wal_segment_size);
2401 : }
2402 1840028 : npages++;
2403 :
2404 : /*
2405 : * Dump the set if this will be the last loop iteration, or if we are
2406 : * at the last page of the cache area (since the next page won't be
2407 : * contiguous in memory), or if we are at the end of the logfile
2408 : * segment.
2409 : */
2410 1840028 : last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2411 :
2412 3459076 : finishing_seg = !ispartialpage &&
2413 1619048 : (startoffset + npages * XLOG_BLCKSZ) >= wal_segment_size;
2414 :
2415 1840028 : if (last_iteration ||
2416 599734 : curridx == XLogCtl->XLogCacheBlck ||
2417 : finishing_seg)
2418 : {
2419 : char *from;
2420 : Size nbytes;
2421 : Size nleft;
2422 : ssize_t written;
2423 : instr_time start;
2424 :
2425 : /* OK to write the page(s) */
2426 1249796 : from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2427 1249796 : nbytes = npages * (Size) XLOG_BLCKSZ;
2428 1249796 : nleft = nbytes;
2429 : do
2430 : {
2431 1249796 : errno = 0;
2432 :
2433 : /* Measure I/O timing to write WAL data */
2434 1249796 : if (track_wal_io_timing)
2435 0 : INSTR_TIME_SET_CURRENT(start);
2436 : else
2437 1249796 : INSTR_TIME_SET_ZERO(start);
2438 :
2439 1249796 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
2440 1249796 : written = pg_pwrite(openLogFile, from, nleft, startoffset);
2441 1249796 : pgstat_report_wait_end();
2442 :
2443 : /*
2444 : * Increment the I/O timing and the number of times WAL data
2445 : * were written out to disk.
2446 : */
2447 1249796 : if (track_wal_io_timing)
2448 : {
2449 : instr_time end;
2450 :
2451 0 : INSTR_TIME_SET_CURRENT(end);
2452 0 : INSTR_TIME_ACCUM_DIFF(PendingWalStats.wal_write_time, end, start);
2453 : }
2454 :
2455 1249796 : PendingWalStats.wal_write++;
2456 :
2457 1249796 : if (written <= 0)
2458 : {
2459 : char xlogfname[MAXFNAMELEN];
2460 : int save_errno;
2461 :
2462 0 : if (errno == EINTR)
2463 0 : continue;
2464 :
2465 0 : save_errno = errno;
2466 0 : XLogFileName(xlogfname, tli, openLogSegNo,
2467 : wal_segment_size);
2468 0 : errno = save_errno;
2469 0 : ereport(PANIC,
2470 : (errcode_for_file_access(),
2471 : errmsg("could not write to log file \"%s\" at offset %u, length %zu: %m",
2472 : xlogfname, startoffset, nleft)));
2473 : }
2474 1249796 : nleft -= written;
2475 1249796 : from += written;
2476 1249796 : startoffset += written;
2477 1249796 : } while (nleft > 0);
2478 :
2479 1249796 : npages = 0;
2480 :
2481 : /*
2482 : * If we just wrote the whole last page of a logfile segment,
2483 : * fsync the segment immediately. This avoids having to go back
2484 : * and re-open prior segments when an fsync request comes along
2485 : * later. Doing it here ensures that one and only one backend will
2486 : * perform this fsync.
2487 : *
2488 : * This is also the right place to notify the Archiver that the
2489 : * segment is ready to copy to archival storage, and to update the
2490 : * timer for archive_timeout, and to signal for a checkpoint if
2491 : * too many logfile segments have been used since the last
2492 : * checkpoint.
2493 : */
2494 1249796 : if (finishing_seg)
2495 : {
2496 1510 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2497 :
2498 : /* signal that we need to wakeup walsenders later */
2499 1510 : WalSndWakeupRequest();
2500 :
2501 1510 : LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2502 :
2503 1510 : if (XLogArchivingActive())
2504 130 : XLogArchiveNotifySeg(openLogSegNo, tli);
2505 :
2506 1510 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
2507 1510 : XLogCtl->lastSegSwitchLSN = LogwrtResult.Flush;
2508 :
2509 : /*
2510 : * Request a checkpoint if we've consumed too much xlog since
2511 : * the last one. For speed, we first check using the local
2512 : * copy of RedoRecPtr, which might be out of date; if it looks
2513 : * like a checkpoint is needed, forcibly update RedoRecPtr and
2514 : * recheck.
2515 : */
2516 1510 : if (IsUnderPostmaster && XLogCheckpointNeeded(openLogSegNo))
2517 : {
2518 116 : (void) GetRedoRecPtr();
2519 116 : if (XLogCheckpointNeeded(openLogSegNo))
2520 56 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
2521 : }
2522 : }
2523 : }
2524 :
2525 1840028 : if (ispartialpage)
2526 : {
2527 : /* Only asked to write a partial page */
2528 220980 : LogwrtResult.Write = WriteRqst.Write;
2529 220980 : break;
2530 : }
2531 1619048 : curridx = NextBufIdx(curridx);
2532 :
2533 : /* If flexible, break out of loop as soon as we wrote something */
2534 1619048 : if (flexible && npages == 0)
2535 1546 : break;
2536 : }
2537 :
2538 : Assert(npages == 0);
2539 :
2540 : /*
2541 : * If asked to flush, do so
2542 : */
2543 1240536 : if (LogwrtResult.Flush < WriteRqst.Flush &&
2544 223002 : LogwrtResult.Flush < LogwrtResult.Write)
2545 : {
2546 : /*
2547 : * Could get here without iterating above loop, in which case we might
2548 : * have no open file or the wrong one. However, we do not need to
2549 : * fsync more than one file.
2550 : */
2551 222948 : if (wal_sync_method != WAL_SYNC_METHOD_OPEN &&
2552 222948 : wal_sync_method != WAL_SYNC_METHOD_OPEN_DSYNC)
2553 : {
2554 222948 : if (openLogFile >= 0 &&
2555 222938 : !XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2556 : wal_segment_size))
2557 0 : XLogFileClose();
2558 222948 : if (openLogFile < 0)
2559 : {
2560 10 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2561 : wal_segment_size);
2562 10 : openLogTLI = tli;
2563 10 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2564 10 : ReserveExternalFD();
2565 : }
2566 :
2567 222948 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2568 : }
2569 :
2570 : /* signal that we need to wakeup walsenders later */
2571 222948 : WalSndWakeupRequest();
2572 :
2573 222948 : LogwrtResult.Flush = LogwrtResult.Write;
2574 : }
2575 :
2576 : /*
2577 : * Update shared-memory status
2578 : *
2579 : * We make sure that the shared 'request' values do not fall behind the
2580 : * 'result' values. This is not absolutely essential, but it saves some
2581 : * code in a couple of places.
2582 : */
2583 1240536 : SpinLockAcquire(&XLogCtl->info_lck);
2584 1240536 : if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
2585 209856 : XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
2586 1240536 : if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
2587 223786 : XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
2588 1240536 : SpinLockRelease(&XLogCtl->info_lck);
2589 :
2590 : /*
2591 : * We write Write first, bar, then Flush. When reading, the opposite must
2592 : * be done (with a matching barrier in between), so that we always see a
2593 : * Flush value that trails behind the Write value seen.
2594 : */
2595 1240536 : pg_atomic_write_u64(&XLogCtl->logWriteResult, LogwrtResult.Write);
2596 1240536 : pg_write_barrier();
2597 1240536 : pg_atomic_write_u64(&XLogCtl->logFlushResult, LogwrtResult.Flush);
2598 :
2599 : #ifdef USE_ASSERT_CHECKING
2600 : {
2601 : XLogRecPtr Flush;
2602 : XLogRecPtr Write;
2603 : XLogRecPtr Insert;
2604 :
2605 : Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult);
2606 : pg_read_barrier();
2607 : Write = pg_atomic_read_u64(&XLogCtl->logWriteResult);
2608 : pg_read_barrier();
2609 : Insert = pg_atomic_read_u64(&XLogCtl->logInsertResult);
2610 :
2611 : /* WAL written to disk is always ahead of WAL flushed */
2612 : Assert(Write >= Flush);
2613 :
2614 : /* WAL inserted to buffers is always ahead of WAL written */
2615 : Assert(Insert >= Write);
2616 : }
2617 : #endif
2618 1240536 : }
2619 :
2620 : /*
2621 : * Record the LSN for an asynchronous transaction commit/abort
2622 : * and nudge the WALWriter if there is work for it to do.
2623 : * (This should not be called for synchronous commits.)
2624 : */
2625 : void
2626 46156 : XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
2627 : {
2628 46156 : XLogRecPtr WriteRqstPtr = asyncXactLSN;
2629 : bool sleeping;
2630 46156 : bool wakeup = false;
2631 : XLogRecPtr prevAsyncXactLSN;
2632 :
2633 46156 : SpinLockAcquire(&XLogCtl->info_lck);
2634 46156 : sleeping = XLogCtl->WalWriterSleeping;
2635 46156 : prevAsyncXactLSN = XLogCtl->asyncXactLSN;
2636 46156 : if (XLogCtl->asyncXactLSN < asyncXactLSN)
2637 45450 : XLogCtl->asyncXactLSN = asyncXactLSN;
2638 46156 : SpinLockRelease(&XLogCtl->info_lck);
2639 :
2640 : /*
2641 : * If somebody else already called this function with a more aggressive
2642 : * LSN, they will have done what we needed (and perhaps more).
2643 : */
2644 46156 : if (asyncXactLSN <= prevAsyncXactLSN)
2645 706 : return;
2646 :
2647 : /*
2648 : * If the WALWriter is sleeping, kick it to make it come out of low-power
2649 : * mode, so that this async commit will reach disk within the expected
2650 : * amount of time. Otherwise, determine whether it has enough WAL
2651 : * available to flush, the same way that XLogBackgroundFlush() does.
2652 : */
2653 45450 : if (sleeping)
2654 20 : wakeup = true;
2655 : else
2656 : {
2657 : int flushblocks;
2658 :
2659 45430 : RefreshXLogWriteResult(LogwrtResult);
2660 :
2661 45430 : flushblocks =
2662 45430 : WriteRqstPtr / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
2663 :
2664 45430 : if (WalWriterFlushAfter == 0 || flushblocks >= WalWriterFlushAfter)
2665 184 : wakeup = true;
2666 : }
2667 :
2668 45450 : if (wakeup && ProcGlobal->walwriterLatch)
2669 34 : SetLatch(ProcGlobal->walwriterLatch);
2670 : }
2671 :
2672 : /*
2673 : * Record the LSN up to which we can remove WAL because it's not required by
2674 : * any replication slot.
2675 : */
2676 : void
2677 41566 : XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
2678 : {
2679 41566 : SpinLockAcquire(&XLogCtl->info_lck);
2680 41566 : XLogCtl->replicationSlotMinLSN = lsn;
2681 41566 : SpinLockRelease(&XLogCtl->info_lck);
2682 41566 : }
2683 :
2684 :
2685 : /*
2686 : * Return the oldest LSN we must retain to satisfy the needs of some
2687 : * replication slot.
2688 : */
2689 : static XLogRecPtr
2690 2522 : XLogGetReplicationSlotMinimumLSN(void)
2691 : {
2692 : XLogRecPtr retval;
2693 :
2694 2522 : SpinLockAcquire(&XLogCtl->info_lck);
2695 2522 : retval = XLogCtl->replicationSlotMinLSN;
2696 2522 : SpinLockRelease(&XLogCtl->info_lck);
2697 :
2698 2522 : return retval;
2699 : }
2700 :
2701 : /*
2702 : * Advance minRecoveryPoint in control file.
2703 : *
2704 : * If we crash during recovery, we must reach this point again before the
2705 : * database is consistent.
2706 : *
2707 : * If 'force' is true, 'lsn' argument is ignored. Otherwise, minRecoveryPoint
2708 : * is only updated if it's not already greater than or equal to 'lsn'.
2709 : */
2710 : static void
2711 178654 : UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
2712 : {
2713 : /* Quick check using our local copy of the variable */
2714 178654 : if (!updateMinRecoveryPoint || (!force && lsn <= LocalMinRecoveryPoint))
2715 165594 : return;
2716 :
2717 : /*
2718 : * An invalid minRecoveryPoint means that we need to recover all the WAL,
2719 : * i.e., we're doing crash recovery. We never modify the control file's
2720 : * value in that case, so we can short-circuit future checks here too. The
2721 : * local values of minRecoveryPoint and minRecoveryPointTLI should not be
2722 : * updated until crash recovery finishes. We only do this for the startup
2723 : * process as it should not update its own reference of minRecoveryPoint
2724 : * until it has finished crash recovery to make sure that all WAL
2725 : * available is replayed in this case. This also saves from extra locks
2726 : * taken on the control file from the startup process.
2727 : */
2728 13060 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint) && InRecovery)
2729 : {
2730 62 : updateMinRecoveryPoint = false;
2731 62 : return;
2732 : }
2733 :
2734 12998 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2735 :
2736 : /* update local copy */
2737 12998 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
2738 12998 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
2739 :
2740 12998 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint))
2741 2 : updateMinRecoveryPoint = false;
2742 12996 : else if (force || LocalMinRecoveryPoint < lsn)
2743 : {
2744 : XLogRecPtr newMinRecoveryPoint;
2745 : TimeLineID newMinRecoveryPointTLI;
2746 :
2747 : /*
2748 : * To avoid having to update the control file too often, we update it
2749 : * all the way to the last record being replayed, even though 'lsn'
2750 : * would suffice for correctness. This also allows the 'force' case
2751 : * to not need a valid 'lsn' value.
2752 : *
2753 : * Another important reason for doing it this way is that the passed
2754 : * 'lsn' value could be bogus, i.e., past the end of available WAL, if
2755 : * the caller got it from a corrupted heap page. Accepting such a
2756 : * value as the min recovery point would prevent us from coming up at
2757 : * all. Instead, we just log a warning and continue with recovery.
2758 : * (See also the comments about corrupt LSNs in XLogFlush.)
2759 : */
2760 12464 : newMinRecoveryPoint = GetCurrentReplayRecPtr(&newMinRecoveryPointTLI);
2761 12464 : if (!force && newMinRecoveryPoint < lsn)
2762 0 : elog(WARNING,
2763 : "xlog min recovery request %X/%X is past current point %X/%X",
2764 : LSN_FORMAT_ARGS(lsn), LSN_FORMAT_ARGS(newMinRecoveryPoint));
2765 :
2766 : /* update control file */
2767 12464 : if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2768 : {
2769 12416 : ControlFile->minRecoveryPoint = newMinRecoveryPoint;
2770 12416 : ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
2771 12416 : UpdateControlFile();
2772 12416 : LocalMinRecoveryPoint = newMinRecoveryPoint;
2773 12416 : LocalMinRecoveryPointTLI = newMinRecoveryPointTLI;
2774 :
2775 12416 : ereport(DEBUG2,
2776 : (errmsg_internal("updated min recovery point to %X/%X on timeline %u",
2777 : LSN_FORMAT_ARGS(newMinRecoveryPoint),
2778 : newMinRecoveryPointTLI)));
2779 : }
2780 : }
2781 12998 : LWLockRelease(ControlFileLock);
2782 : }
2783 :
2784 : /*
2785 : * Ensure that all XLOG data through the given position is flushed to disk.
2786 : *
2787 : * NOTE: this differs from XLogWrite mainly in that the WALWriteLock is not
2788 : * already held, and we try to avoid acquiring it if possible.
2789 : */
2790 : void
2791 1123272 : XLogFlush(XLogRecPtr record)
2792 : {
2793 : XLogRecPtr WriteRqstPtr;
2794 : XLogwrtRqst WriteRqst;
2795 1123272 : TimeLineID insertTLI = XLogCtl->InsertTimeLineID;
2796 :
2797 : /*
2798 : * During REDO, we are reading not writing WAL. Therefore, instead of
2799 : * trying to flush the WAL, we should update minRecoveryPoint instead. We
2800 : * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2801 : * to act this way too, and because when it tries to write the
2802 : * end-of-recovery checkpoint, it should indeed flush.
2803 : */
2804 1123272 : if (!XLogInsertAllowed())
2805 : {
2806 178470 : UpdateMinRecoveryPoint(record, false);
2807 884580 : return;
2808 : }
2809 :
2810 : /* Quick exit if already known flushed */
2811 944802 : if (record <= LogwrtResult.Flush)
2812 706110 : return;
2813 :
2814 : #ifdef WAL_DEBUG
2815 : if (XLOG_DEBUG)
2816 : elog(LOG, "xlog flush request %X/%X; write %X/%X; flush %X/%X",
2817 : LSN_FORMAT_ARGS(record),
2818 : LSN_FORMAT_ARGS(LogwrtResult.Write),
2819 : LSN_FORMAT_ARGS(LogwrtResult.Flush));
2820 : #endif
2821 :
2822 238692 : START_CRIT_SECTION();
2823 :
2824 : /*
2825 : * Since fsync is usually a horribly expensive operation, we try to
2826 : * piggyback as much data as we can on each fsync: if we see any more data
2827 : * entered into the xlog buffer, we'll write and fsync that too, so that
2828 : * the final value of LogwrtResult.Flush is as large as possible. This
2829 : * gives us some chance of avoiding another fsync immediately after.
2830 : */
2831 :
2832 : /* initialize to given target; may increase below */
2833 238692 : WriteRqstPtr = record;
2834 :
2835 : /*
2836 : * Now wait until we get the write lock, or someone else does the flush
2837 : * for us.
2838 : */
2839 : for (;;)
2840 4222 : {
2841 : XLogRecPtr insertpos;
2842 :
2843 : /* done already? */
2844 242914 : RefreshXLogWriteResult(LogwrtResult);
2845 242914 : if (record <= LogwrtResult.Flush)
2846 16230 : break;
2847 :
2848 : /*
2849 : * Before actually performing the write, wait for all in-flight
2850 : * insertions to the pages we're about to write to finish.
2851 : */
2852 226684 : SpinLockAcquire(&XLogCtl->info_lck);
2853 226684 : if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2854 6732 : WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2855 226684 : SpinLockRelease(&XLogCtl->info_lck);
2856 226684 : insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2857 :
2858 : /*
2859 : * Try to get the write lock. If we can't get it immediately, wait
2860 : * until it's released, and recheck if we still need to do the flush
2861 : * or if the backend that held the lock did it for us already. This
2862 : * helps to maintain a good rate of group committing when the system
2863 : * is bottlenecked by the speed of fsyncing.
2864 : */
2865 226684 : if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2866 : {
2867 : /*
2868 : * The lock is now free, but we didn't acquire it yet. Before we
2869 : * do, loop back to check if someone else flushed the record for
2870 : * us already.
2871 : */
2872 4222 : continue;
2873 : }
2874 :
2875 : /* Got the lock; recheck whether request is satisfied */
2876 222462 : RefreshXLogWriteResult(LogwrtResult);
2877 222462 : if (record <= LogwrtResult.Flush)
2878 : {
2879 1234 : LWLockRelease(WALWriteLock);
2880 1234 : break;
2881 : }
2882 :
2883 : /*
2884 : * Sleep before flush! By adding a delay here, we may give further
2885 : * backends the opportunity to join the backlog of group commit
2886 : * followers; this can significantly improve transaction throughput,
2887 : * at the risk of increasing transaction latency.
2888 : *
2889 : * We do not sleep if enableFsync is not turned on, nor if there are
2890 : * fewer than CommitSiblings other backends with active transactions.
2891 : */
2892 221228 : if (CommitDelay > 0 && enableFsync &&
2893 0 : MinimumActiveBackends(CommitSiblings))
2894 : {
2895 0 : pg_usleep(CommitDelay);
2896 :
2897 : /*
2898 : * Re-check how far we can now flush the WAL. It's generally not
2899 : * safe to call WaitXLogInsertionsToFinish while holding
2900 : * WALWriteLock, because an in-progress insertion might need to
2901 : * also grab WALWriteLock to make progress. But we know that all
2902 : * the insertions up to insertpos have already finished, because
2903 : * that's what the earlier WaitXLogInsertionsToFinish() returned.
2904 : * We're only calling it again to allow insertpos to be moved
2905 : * further forward, not to actually wait for anyone.
2906 : */
2907 0 : insertpos = WaitXLogInsertionsToFinish(insertpos);
2908 : }
2909 :
2910 : /* try to write/flush later additions to XLOG as well */
2911 221228 : WriteRqst.Write = insertpos;
2912 221228 : WriteRqst.Flush = insertpos;
2913 :
2914 221228 : XLogWrite(WriteRqst, insertTLI, false);
2915 :
2916 221228 : LWLockRelease(WALWriteLock);
2917 : /* done */
2918 221228 : break;
2919 : }
2920 :
2921 238692 : END_CRIT_SECTION();
2922 :
2923 : /* wake up walsenders now that we've released heavily contended locks */
2924 238692 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
2925 :
2926 : /*
2927 : * If we still haven't flushed to the request point then we have a
2928 : * problem; most likely, the requested flush point is past end of XLOG.
2929 : * This has been seen to occur when a disk page has a corrupted LSN.
2930 : *
2931 : * Formerly we treated this as a PANIC condition, but that hurts the
2932 : * system's robustness rather than helping it: we do not want to take down
2933 : * the whole system due to corruption on one data page. In particular, if
2934 : * the bad page is encountered again during recovery then we would be
2935 : * unable to restart the database at all! (This scenario actually
2936 : * happened in the field several times with 7.1 releases.) As of 8.4, bad
2937 : * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2938 : * the only time we can reach here during recovery is while flushing the
2939 : * end-of-recovery checkpoint record, and we don't expect that to have a
2940 : * bad LSN.
2941 : *
2942 : * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2943 : * since xact.c calls this routine inside a critical section. However,
2944 : * calls from bufmgr.c are not within critical sections and so we will not
2945 : * force a restart for a bad LSN on a data page.
2946 : */
2947 238692 : if (LogwrtResult.Flush < record)
2948 0 : elog(ERROR,
2949 : "xlog flush request %X/%X is not satisfied --- flushed only to %X/%X",
2950 : LSN_FORMAT_ARGS(record),
2951 : LSN_FORMAT_ARGS(LogwrtResult.Flush));
2952 : }
2953 :
2954 : /*
2955 : * Write & flush xlog, but without specifying exactly where to.
2956 : *
2957 : * We normally write only completed blocks; but if there is nothing to do on
2958 : * that basis, we check for unwritten async commits in the current incomplete
2959 : * block, and write through the latest one of those. Thus, if async commits
2960 : * are not being used, we will write complete blocks only.
2961 : *
2962 : * If, based on the above, there's anything to write we do so immediately. But
2963 : * to avoid calling fsync, fdatasync et. al. at a rate that'd impact
2964 : * concurrent IO, we only flush WAL every wal_writer_delay ms, or if there's
2965 : * more than wal_writer_flush_after unflushed blocks.
2966 : *
2967 : * We can guarantee that async commits reach disk after at most three
2968 : * wal_writer_delay cycles. (When flushing complete blocks, we allow XLogWrite
2969 : * to write "flexibly", meaning it can stop at the end of the buffer ring;
2970 : * this makes a difference only with very high load or long wal_writer_delay,
2971 : * but imposes one extra cycle for the worst case for async commits.)
2972 : *
2973 : * This routine is invoked periodically by the background walwriter process.
2974 : *
2975 : * Returns true if there was any work to do, even if we skipped flushing due
2976 : * to wal_writer_delay/wal_writer_flush_after.
2977 : */
2978 : bool
2979 27784 : XLogBackgroundFlush(void)
2980 : {
2981 : XLogwrtRqst WriteRqst;
2982 27784 : bool flexible = true;
2983 : static TimestampTz lastflush;
2984 : TimestampTz now;
2985 : int flushblocks;
2986 : TimeLineID insertTLI;
2987 :
2988 : /* XLOG doesn't need flushing during recovery */
2989 27784 : if (RecoveryInProgress())
2990 16 : return false;
2991 :
2992 : /*
2993 : * Since we're not in recovery, InsertTimeLineID is set and can't change,
2994 : * so we can read it without a lock.
2995 : */
2996 27768 : insertTLI = XLogCtl->InsertTimeLineID;
2997 :
2998 : /* read updated LogwrtRqst */
2999 27768 : SpinLockAcquire(&XLogCtl->info_lck);
3000 27768 : WriteRqst = XLogCtl->LogwrtRqst;
3001 27768 : SpinLockRelease(&XLogCtl->info_lck);
3002 :
3003 : /* back off to last completed page boundary */
3004 27768 : WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
3005 :
3006 : /* if we have already flushed that far, consider async commit records */
3007 27768 : RefreshXLogWriteResult(LogwrtResult);
3008 27768 : if (WriteRqst.Write <= LogwrtResult.Flush)
3009 : {
3010 26138 : SpinLockAcquire(&XLogCtl->info_lck);
3011 26138 : WriteRqst.Write = XLogCtl->asyncXactLSN;
3012 26138 : SpinLockRelease(&XLogCtl->info_lck);
3013 26138 : flexible = false; /* ensure it all gets written */
3014 : }
3015 :
3016 : /*
3017 : * If already known flushed, we're done. Just need to check if we are
3018 : * holding an open file handle to a logfile that's no longer in use,
3019 : * preventing the file from being deleted.
3020 : */
3021 27768 : if (WriteRqst.Write <= LogwrtResult.Flush)
3022 : {
3023 25332 : if (openLogFile >= 0)
3024 : {
3025 5288 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
3026 : wal_segment_size))
3027 : {
3028 174 : XLogFileClose();
3029 : }
3030 : }
3031 25332 : return false;
3032 : }
3033 :
3034 : /*
3035 : * Determine how far to flush WAL, based on the wal_writer_delay and
3036 : * wal_writer_flush_after GUCs.
3037 : *
3038 : * Note that XLogSetAsyncXactLSN() performs similar calculation based on
3039 : * wal_writer_flush_after, to decide when to wake us up. Make sure the
3040 : * logic is the same in both places if you change this.
3041 : */
3042 2436 : now = GetCurrentTimestamp();
3043 2436 : flushblocks =
3044 2436 : WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3045 :
3046 2436 : if (WalWriterFlushAfter == 0 || lastflush == 0)
3047 : {
3048 : /* first call, or block based limits disabled */
3049 308 : WriteRqst.Flush = WriteRqst.Write;
3050 308 : lastflush = now;
3051 : }
3052 2128 : else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3053 : {
3054 : /*
3055 : * Flush the writes at least every WalWriterDelay ms. This is
3056 : * important to bound the amount of time it takes for an asynchronous
3057 : * commit to hit disk.
3058 : */
3059 2108 : WriteRqst.Flush = WriteRqst.Write;
3060 2108 : lastflush = now;
3061 : }
3062 20 : else if (flushblocks >= WalWriterFlushAfter)
3063 : {
3064 : /* exceeded wal_writer_flush_after blocks, flush */
3065 10 : WriteRqst.Flush = WriteRqst.Write;
3066 10 : lastflush = now;
3067 : }
3068 : else
3069 : {
3070 : /* no flushing, this time round */
3071 10 : WriteRqst.Flush = 0;
3072 : }
3073 :
3074 : #ifdef WAL_DEBUG
3075 : if (XLOG_DEBUG)
3076 : elog(LOG, "xlog bg flush request write %X/%X; flush: %X/%X, current is write %X/%X; flush %X/%X",
3077 : LSN_FORMAT_ARGS(WriteRqst.Write),
3078 : LSN_FORMAT_ARGS(WriteRqst.Flush),
3079 : LSN_FORMAT_ARGS(LogwrtResult.Write),
3080 : LSN_FORMAT_ARGS(LogwrtResult.Flush));
3081 : #endif
3082 :
3083 2436 : START_CRIT_SECTION();
3084 :
3085 : /* now wait for any in-progress insertions to finish and get write lock */
3086 2436 : WaitXLogInsertionsToFinish(WriteRqst.Write);
3087 2436 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
3088 2436 : RefreshXLogWriteResult(LogwrtResult);
3089 2436 : if (WriteRqst.Write > LogwrtResult.Write ||
3090 90 : WriteRqst.Flush > LogwrtResult.Flush)
3091 : {
3092 2358 : XLogWrite(WriteRqst, insertTLI, flexible);
3093 : }
3094 2436 : LWLockRelease(WALWriteLock);
3095 :
3096 2436 : END_CRIT_SECTION();
3097 :
3098 : /* wake up walsenders now that we've released heavily contended locks */
3099 2436 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
3100 :
3101 : /*
3102 : * Great, done. To take some work off the critical path, try to initialize
3103 : * as many of the no-longer-needed WAL buffers for future use as we can.
3104 : */
3105 2436 : AdvanceXLInsertBuffer(InvalidXLogRecPtr, insertTLI, true);
3106 :
3107 : /*
3108 : * If we determined that we need to write data, but somebody else
3109 : * wrote/flushed already, it should be considered as being active, to
3110 : * avoid hibernating too early.
3111 : */
3112 2436 : return true;
3113 : }
3114 :
3115 : /*
3116 : * Test whether XLOG data has been flushed up to (at least) the given position.
3117 : *
3118 : * Returns true if a flush is still needed. (It may be that someone else
3119 : * is already in process of flushing that far, however.)
3120 : */
3121 : bool
3122 16296592 : XLogNeedsFlush(XLogRecPtr record)
3123 : {
3124 : /*
3125 : * During recovery, we don't flush WAL but update minRecoveryPoint
3126 : * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3127 : * would need to be updated.
3128 : */
3129 16296592 : if (RecoveryInProgress())
3130 : {
3131 : /*
3132 : * An invalid minRecoveryPoint means that we need to recover all the
3133 : * WAL, i.e., we're doing crash recovery. We never modify the control
3134 : * file's value in that case, so we can short-circuit future checks
3135 : * here too. This triggers a quick exit path for the startup process,
3136 : * which cannot update its local copy of minRecoveryPoint as long as
3137 : * it has not replayed all WAL available when doing crash recovery.
3138 : */
3139 1557394 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint) && InRecovery)
3140 0 : updateMinRecoveryPoint = false;
3141 :
3142 : /* Quick exit if already known to be updated or cannot be updated */
3143 1557394 : if (record <= LocalMinRecoveryPoint || !updateMinRecoveryPoint)
3144 1520668 : return false;
3145 :
3146 : /*
3147 : * Update local copy of minRecoveryPoint. But if the lock is busy,
3148 : * just return a conservative guess.
3149 : */
3150 36726 : if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
3151 0 : return true;
3152 36726 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
3153 36726 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
3154 36726 : LWLockRelease(ControlFileLock);
3155 :
3156 : /*
3157 : * Check minRecoveryPoint for any other process than the startup
3158 : * process doing crash recovery, which should not update the control
3159 : * file value if crash recovery is still running.
3160 : */
3161 36726 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint))
3162 0 : updateMinRecoveryPoint = false;
3163 :
3164 : /* check again */
3165 36726 : if (record <= LocalMinRecoveryPoint || !updateMinRecoveryPoint)
3166 136 : return false;
3167 : else
3168 36590 : return true;
3169 : }
3170 :
3171 : /* Quick exit if already known flushed */
3172 14739198 : if (record <= LogwrtResult.Flush)
3173 14318192 : return false;
3174 :
3175 : /* read LogwrtResult and update local state */
3176 421006 : RefreshXLogWriteResult(LogwrtResult);
3177 :
3178 : /* check again */
3179 421006 : if (record <= LogwrtResult.Flush)
3180 3620 : return false;
3181 :
3182 417386 : return true;
3183 : }
3184 :
3185 : /*
3186 : * Try to make a given XLOG file segment exist.
3187 : *
3188 : * logsegno: identify segment.
3189 : *
3190 : * *added: on return, true if this call raised the number of extant segments.
3191 : *
3192 : * path: on return, this char[MAXPGPATH] has the path to the logsegno file.
3193 : *
3194 : * Returns -1 or FD of opened file. A -1 here is not an error; a caller
3195 : * wanting an open segment should attempt to open "path", which usually will
3196 : * succeed. (This is weird, but it's efficient for the callers.)
3197 : */
3198 : static int
3199 14782 : XLogFileInitInternal(XLogSegNo logsegno, TimeLineID logtli,
3200 : bool *added, char *path)
3201 : {
3202 : char tmppath[MAXPGPATH];
3203 : XLogSegNo installed_segno;
3204 : XLogSegNo max_segno;
3205 : int fd;
3206 : int save_errno;
3207 14782 : int open_flags = O_RDWR | O_CREAT | O_EXCL | PG_BINARY;
3208 :
3209 : Assert(logtli != 0);
3210 :
3211 14782 : XLogFilePath(path, logtli, logsegno, wal_segment_size);
3212 :
3213 : /*
3214 : * Try to use existent file (checkpoint maker may have created it already)
3215 : */
3216 14782 : *added = false;
3217 14782 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3218 14782 : get_sync_bit(wal_sync_method));
3219 14782 : if (fd < 0)
3220 : {
3221 840 : if (errno != ENOENT)
3222 0 : ereport(ERROR,
3223 : (errcode_for_file_access(),
3224 : errmsg("could not open file \"%s\": %m", path)));
3225 : }
3226 : else
3227 13942 : return fd;
3228 :
3229 : /*
3230 : * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3231 : * another process is doing the same thing. If so, we will end up
3232 : * pre-creating an extra log segment. That seems OK, and better than
3233 : * holding the lock throughout this lengthy process.
3234 : */
3235 840 : elog(DEBUG2, "creating and filling new WAL file");
3236 :
3237 840 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3238 :
3239 840 : unlink(tmppath);
3240 :
3241 840 : if (io_direct_flags & IO_DIRECT_WAL_INIT)
3242 0 : open_flags |= PG_O_DIRECT;
3243 :
3244 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3245 840 : fd = BasicOpenFile(tmppath, open_flags);
3246 840 : if (fd < 0)
3247 0 : ereport(ERROR,
3248 : (errcode_for_file_access(),
3249 : errmsg("could not create file \"%s\": %m", tmppath)));
3250 :
3251 840 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3252 840 : save_errno = 0;
3253 840 : if (wal_init_zero)
3254 : {
3255 : ssize_t rc;
3256 :
3257 : /*
3258 : * Zero-fill the file. With this setting, we do this the hard way to
3259 : * ensure that all the file space has really been allocated. On
3260 : * platforms that allow "holes" in files, just seeking to the end
3261 : * doesn't allocate intermediate space. This way, we know that we
3262 : * have all the space and (after the fsync below) that all the
3263 : * indirect blocks are down on disk. Therefore, fdatasync(2) or
3264 : * O_DSYNC will be sufficient to sync future writes to the log file.
3265 : */
3266 840 : rc = pg_pwrite_zeros(fd, wal_segment_size, 0);
3267 :
3268 840 : if (rc < 0)
3269 0 : save_errno = errno;
3270 : }
3271 : else
3272 : {
3273 : /*
3274 : * Otherwise, seeking to the end and writing a solitary byte is
3275 : * enough.
3276 : */
3277 0 : errno = 0;
3278 0 : if (pg_pwrite(fd, "\0", 1, wal_segment_size - 1) != 1)
3279 : {
3280 : /* if write didn't set errno, assume no disk space */
3281 0 : save_errno = errno ? errno : ENOSPC;
3282 : }
3283 : }
3284 840 : pgstat_report_wait_end();
3285 :
3286 840 : if (save_errno)
3287 : {
3288 : /*
3289 : * If we fail to make the file, delete it to release disk space
3290 : */
3291 0 : unlink(tmppath);
3292 :
3293 0 : close(fd);
3294 :
3295 0 : errno = save_errno;
3296 :
3297 0 : ereport(ERROR,
3298 : (errcode_for_file_access(),
3299 : errmsg("could not write to file \"%s\": %m", tmppath)));
3300 : }
3301 :
3302 840 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
3303 840 : if (pg_fsync(fd) != 0)
3304 : {
3305 0 : save_errno = errno;
3306 0 : close(fd);
3307 0 : errno = save_errno;
3308 0 : ereport(ERROR,
3309 : (errcode_for_file_access(),
3310 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3311 : }
3312 840 : pgstat_report_wait_end();
3313 :
3314 840 : if (close(fd) != 0)
3315 0 : ereport(ERROR,
3316 : (errcode_for_file_access(),
3317 : errmsg("could not close file \"%s\": %m", tmppath)));
3318 :
3319 : /*
3320 : * Now move the segment into place with its final name. Cope with
3321 : * possibility that someone else has created the file while we were
3322 : * filling ours: if so, use ours to pre-create a future log segment.
3323 : */
3324 840 : installed_segno = logsegno;
3325 :
3326 : /*
3327 : * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3328 : * that was a constant, but that was always a bit dubious: normally, at a
3329 : * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3330 : * here, it was the offset from the insert location. We can't do the
3331 : * normal XLOGfileslop calculation here because we don't have access to
3332 : * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3333 : * CheckPointSegments.
3334 : */
3335 840 : max_segno = logsegno + CheckPointSegments;
3336 840 : if (InstallXLogFileSegment(&installed_segno, tmppath, true, max_segno,
3337 : logtli))
3338 : {
3339 840 : *added = true;
3340 840 : elog(DEBUG2, "done creating and filling new WAL file");
3341 : }
3342 : else
3343 : {
3344 : /*
3345 : * No need for any more future segments, or InstallXLogFileSegment()
3346 : * failed to rename the file into place. If the rename failed, a
3347 : * caller opening the file may fail.
3348 : */
3349 0 : unlink(tmppath);
3350 0 : elog(DEBUG2, "abandoned new WAL file");
3351 : }
3352 :
3353 840 : return -1;
3354 : }
3355 :
3356 : /*
3357 : * Create a new XLOG file segment, or open a pre-existing one.
3358 : *
3359 : * logsegno: identify segment to be created/opened.
3360 : *
3361 : * Returns FD of opened file.
3362 : *
3363 : * Note: errors here are ERROR not PANIC because we might or might not be
3364 : * inside a critical section (eg, during checkpoint there is no reason to
3365 : * take down the system on failure). They will promote to PANIC if we are
3366 : * in a critical section.
3367 : */
3368 : int
3369 14632 : XLogFileInit(XLogSegNo logsegno, TimeLineID logtli)
3370 : {
3371 : bool ignore_added;
3372 : char path[MAXPGPATH];
3373 : int fd;
3374 :
3375 : Assert(logtli != 0);
3376 :
3377 14632 : fd = XLogFileInitInternal(logsegno, logtli, &ignore_added, path);
3378 14632 : if (fd >= 0)
3379 13816 : return fd;
3380 :
3381 : /* Now open original target segment (might not be file I just made) */
3382 816 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3383 816 : get_sync_bit(wal_sync_method));
3384 816 : if (fd < 0)
3385 0 : ereport(ERROR,
3386 : (errcode_for_file_access(),
3387 : errmsg("could not open file \"%s\": %m", path)));
3388 816 : return fd;
3389 : }
3390 :
3391 : /*
3392 : * Create a new XLOG file segment by copying a pre-existing one.
3393 : *
3394 : * destsegno: identify segment to be created.
3395 : *
3396 : * srcTLI, srcsegno: identify segment to be copied (could be from
3397 : * a different timeline)
3398 : *
3399 : * upto: how much of the source file to copy (the rest is filled with
3400 : * zeros)
3401 : *
3402 : * Currently this is only used during recovery, and so there are no locking
3403 : * considerations. But we should be just as tense as XLogFileInit to avoid
3404 : * emplacing a bogus file.
3405 : */
3406 : static void
3407 74 : XLogFileCopy(TimeLineID destTLI, XLogSegNo destsegno,
3408 : TimeLineID srcTLI, XLogSegNo srcsegno,
3409 : int upto)
3410 : {
3411 : char path[MAXPGPATH];
3412 : char tmppath[MAXPGPATH];
3413 : PGAlignedXLogBlock buffer;
3414 : int srcfd;
3415 : int fd;
3416 : int nbytes;
3417 :
3418 : /*
3419 : * Open the source file
3420 : */
3421 74 : XLogFilePath(path, srcTLI, srcsegno, wal_segment_size);
3422 74 : srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
3423 74 : if (srcfd < 0)
3424 0 : ereport(ERROR,
3425 : (errcode_for_file_access(),
3426 : errmsg("could not open file \"%s\": %m", path)));
3427 :
3428 : /*
3429 : * Copy into a temp file name.
3430 : */
3431 74 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3432 :
3433 74 : unlink(tmppath);
3434 :
3435 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3436 74 : fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
3437 74 : if (fd < 0)
3438 0 : ereport(ERROR,
3439 : (errcode_for_file_access(),
3440 : errmsg("could not create file \"%s\": %m", tmppath)));
3441 :
3442 : /*
3443 : * Do the data copying.
3444 : */
3445 151626 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += sizeof(buffer))
3446 : {
3447 : int nread;
3448 :
3449 151552 : nread = upto - nbytes;
3450 :
3451 : /*
3452 : * The part that is not read from the source file is filled with
3453 : * zeros.
3454 : */
3455 151552 : if (nread < sizeof(buffer))
3456 74 : memset(buffer.data, 0, sizeof(buffer));
3457 :
3458 151552 : if (nread > 0)
3459 : {
3460 : int r;
3461 :
3462 5180 : if (nread > sizeof(buffer))
3463 5106 : nread = sizeof(buffer);
3464 5180 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
3465 5180 : r = read(srcfd, buffer.data, nread);
3466 5180 : if (r != nread)
3467 : {
3468 0 : if (r < 0)
3469 0 : ereport(ERROR,
3470 : (errcode_for_file_access(),
3471 : errmsg("could not read file \"%s\": %m",
3472 : path)));
3473 : else
3474 0 : ereport(ERROR,
3475 : (errcode(ERRCODE_DATA_CORRUPTED),
3476 : errmsg("could not read file \"%s\": read %d of %zu",
3477 : path, r, (Size) nread)));
3478 : }
3479 5180 : pgstat_report_wait_end();
3480 : }
3481 151552 : errno = 0;
3482 151552 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
3483 151552 : if ((int) write(fd, buffer.data, sizeof(buffer)) != (int) sizeof(buffer))
3484 : {
3485 0 : int save_errno = errno;
3486 :
3487 : /*
3488 : * If we fail to make the file, delete it to release disk space
3489 : */
3490 0 : unlink(tmppath);
3491 : /* if write didn't set errno, assume problem is no disk space */
3492 0 : errno = save_errno ? save_errno : ENOSPC;
3493 :
3494 0 : ereport(ERROR,
3495 : (errcode_for_file_access(),
3496 : errmsg("could not write to file \"%s\": %m", tmppath)));
3497 : }
3498 151552 : pgstat_report_wait_end();
3499 : }
3500 :
3501 74 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
3502 74 : if (pg_fsync(fd) != 0)
3503 0 : ereport(data_sync_elevel(ERROR),
3504 : (errcode_for_file_access(),
3505 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3506 74 : pgstat_report_wait_end();
3507 :
3508 74 : if (CloseTransientFile(fd) != 0)
3509 0 : ereport(ERROR,
3510 : (errcode_for_file_access(),
3511 : errmsg("could not close file \"%s\": %m", tmppath)));
3512 :
3513 74 : if (CloseTransientFile(srcfd) != 0)
3514 0 : ereport(ERROR,
3515 : (errcode_for_file_access(),
3516 : errmsg("could not close file \"%s\": %m", path)));
3517 :
3518 : /*
3519 : * Now move the segment into place with its final name.
3520 : */
3521 74 : if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, destTLI))
3522 0 : elog(ERROR, "InstallXLogFileSegment should not have failed");
3523 74 : }
3524 :
3525 : /*
3526 : * Install a new XLOG segment file as a current or future log segment.
3527 : *
3528 : * This is used both to install a newly-created segment (which has a temp
3529 : * filename while it's being created) and to recycle an old segment.
3530 : *
3531 : * *segno: identify segment to install as (or first possible target).
3532 : * When find_free is true, this is modified on return to indicate the
3533 : * actual installation location or last segment searched.
3534 : *
3535 : * tmppath: initial name of file to install. It will be renamed into place.
3536 : *
3537 : * find_free: if true, install the new segment at the first empty segno
3538 : * number at or after the passed numbers. If false, install the new segment
3539 : * exactly where specified, deleting any existing segment file there.
3540 : *
3541 : * max_segno: maximum segment number to install the new file as. Fail if no
3542 : * free slot is found between *segno and max_segno. (Ignored when find_free
3543 : * is false.)
3544 : *
3545 : * tli: The timeline on which the new segment should be installed.
3546 : *
3547 : * Returns true if the file was installed successfully. false indicates that
3548 : * max_segno limit was exceeded, the startup process has disabled this
3549 : * function for now, or an error occurred while renaming the file into place.
3550 : */
3551 : static bool
3552 2100 : InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
3553 : bool find_free, XLogSegNo max_segno, TimeLineID tli)
3554 : {
3555 : char path[MAXPGPATH];
3556 : struct stat stat_buf;
3557 :
3558 : Assert(tli != 0);
3559 :
3560 2100 : XLogFilePath(path, tli, *segno, wal_segment_size);
3561 :
3562 2100 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3563 2100 : if (!XLogCtl->InstallXLogFileSegmentActive)
3564 : {
3565 0 : LWLockRelease(ControlFileLock);
3566 0 : return false;
3567 : }
3568 :
3569 2100 : if (!find_free)
3570 : {
3571 : /* Force installation: get rid of any pre-existing segment file */
3572 74 : durable_unlink(path, DEBUG1);
3573 : }
3574 : else
3575 : {
3576 : /* Find a free slot to put it in */
3577 3278 : while (stat(path, &stat_buf) == 0)
3578 : {
3579 1284 : if ((*segno) >= max_segno)
3580 : {
3581 : /* Failed to find a free slot within specified range */
3582 32 : LWLockRelease(ControlFileLock);
3583 32 : return false;
3584 : }
3585 1252 : (*segno)++;
3586 1252 : XLogFilePath(path, tli, *segno, wal_segment_size);
3587 : }
3588 : }
3589 :
3590 : Assert(access(path, F_OK) != 0 && errno == ENOENT);
3591 2068 : if (durable_rename(tmppath, path, LOG) != 0)
3592 : {
3593 0 : LWLockRelease(ControlFileLock);
3594 : /* durable_rename already emitted log message */
3595 0 : return false;
3596 : }
3597 :
3598 2068 : LWLockRelease(ControlFileLock);
3599 :
3600 2068 : return true;
3601 : }
3602 :
3603 : /*
3604 : * Open a pre-existing logfile segment for writing.
3605 : */
3606 : int
3607 10 : XLogFileOpen(XLogSegNo segno, TimeLineID tli)
3608 : {
3609 : char path[MAXPGPATH];
3610 : int fd;
3611 :
3612 10 : XLogFilePath(path, tli, segno, wal_segment_size);
3613 :
3614 10 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3615 10 : get_sync_bit(wal_sync_method));
3616 10 : if (fd < 0)
3617 0 : ereport(PANIC,
3618 : (errcode_for_file_access(),
3619 : errmsg("could not open file \"%s\": %m", path)));
3620 :
3621 10 : return fd;
3622 : }
3623 :
3624 : /*
3625 : * Close the current logfile segment for writing.
3626 : */
3627 : static void
3628 3116 : XLogFileClose(void)
3629 : {
3630 : Assert(openLogFile >= 0);
3631 :
3632 : /*
3633 : * WAL segment files will not be re-read in normal operation, so we advise
3634 : * the OS to release any cached pages. But do not do so if WAL archiving
3635 : * or streaming is active, because archiver and walsender process could
3636 : * use the cache to read the WAL segment.
3637 : */
3638 : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
3639 3116 : if (!XLogIsNeeded() && (io_direct_flags & IO_DIRECT_WAL) == 0)
3640 2144 : (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3641 : #endif
3642 :
3643 3116 : if (close(openLogFile) != 0)
3644 : {
3645 : char xlogfname[MAXFNAMELEN];
3646 0 : int save_errno = errno;
3647 :
3648 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo, wal_segment_size);
3649 0 : errno = save_errno;
3650 0 : ereport(PANIC,
3651 : (errcode_for_file_access(),
3652 : errmsg("could not close file \"%s\": %m", xlogfname)));
3653 : }
3654 :
3655 3116 : openLogFile = -1;
3656 3116 : ReleaseExternalFD();
3657 3116 : }
3658 :
3659 : /*
3660 : * Preallocate log files beyond the specified log endpoint.
3661 : *
3662 : * XXX this is currently extremely conservative, since it forces only one
3663 : * future log segment to exist, and even that only if we are 75% done with
3664 : * the current one. This is only appropriate for very low-WAL-volume systems.
3665 : * High-volume systems will be OK once they've built up a sufficient set of
3666 : * recycled log segments, but the startup transient is likely to include
3667 : * a lot of segment creations by foreground processes, which is not so good.
3668 : *
3669 : * XLogFileInitInternal() can ereport(ERROR). All known causes indicate big
3670 : * trouble; for example, a full filesystem is one cause. The checkpoint WAL
3671 : * and/or ControlFile updates already completed. If a RequestCheckpoint()
3672 : * initiated the present checkpoint and an ERROR ends this function, the
3673 : * command that called RequestCheckpoint() fails. That's not ideal, but it's
3674 : * not worth contorting more functions to use caller-specified elevel values.
3675 : * (With or without RequestCheckpoint(), an ERROR forestalls some inessential
3676 : * reporting and resource reclamation.)
3677 : */
3678 : static void
3679 2162 : PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli)
3680 : {
3681 : XLogSegNo _logSegNo;
3682 : int lf;
3683 : bool added;
3684 : char path[MAXPGPATH];
3685 : uint64 offset;
3686 :
3687 2162 : if (!XLogCtl->InstallXLogFileSegmentActive)
3688 8 : return; /* unlocked check says no */
3689 :
3690 2154 : XLByteToPrevSeg(endptr, _logSegNo, wal_segment_size);
3691 2154 : offset = XLogSegmentOffset(endptr - 1, wal_segment_size);
3692 2154 : if (offset >= (uint32) (0.75 * wal_segment_size))
3693 : {
3694 150 : _logSegNo++;
3695 150 : lf = XLogFileInitInternal(_logSegNo, tli, &added, path);
3696 150 : if (lf >= 0)
3697 126 : close(lf);
3698 150 : if (added)
3699 24 : CheckpointStats.ckpt_segs_added++;
3700 : }
3701 : }
3702 :
3703 : /*
3704 : * Throws an error if the given log segment has already been removed or
3705 : * recycled. The caller should only pass a segment that it knows to have
3706 : * existed while the server has been running, as this function always
3707 : * succeeds if no WAL segments have been removed since startup.
3708 : * 'tli' is only used in the error message.
3709 : *
3710 : * Note: this function guarantees to keep errno unchanged on return.
3711 : * This supports callers that use this to possibly deliver a better
3712 : * error message about a missing file, while still being able to throw
3713 : * a normal file-access error afterwards, if this does return.
3714 : */
3715 : void
3716 102062 : CheckXLogRemoved(XLogSegNo segno, TimeLineID tli)
3717 : {
3718 102062 : int save_errno = errno;
3719 : XLogSegNo lastRemovedSegNo;
3720 :
3721 102062 : SpinLockAcquire(&XLogCtl->info_lck);
3722 102062 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3723 102062 : SpinLockRelease(&XLogCtl->info_lck);
3724 :
3725 102062 : if (segno <= lastRemovedSegNo)
3726 : {
3727 : char filename[MAXFNAMELEN];
3728 :
3729 0 : XLogFileName(filename, tli, segno, wal_segment_size);
3730 0 : errno = save_errno;
3731 0 : ereport(ERROR,
3732 : (errcode_for_file_access(),
3733 : errmsg("requested WAL segment %s has already been removed",
3734 : filename)));
3735 : }
3736 102062 : errno = save_errno;
3737 102062 : }
3738 :
3739 : /*
3740 : * Return the last WAL segment removed, or 0 if no segment has been removed
3741 : * since startup.
3742 : *
3743 : * NB: the result can be out of date arbitrarily fast, the caller has to deal
3744 : * with that.
3745 : */
3746 : XLogSegNo
3747 1860 : XLogGetLastRemovedSegno(void)
3748 : {
3749 : XLogSegNo lastRemovedSegNo;
3750 :
3751 1860 : SpinLockAcquire(&XLogCtl->info_lck);
3752 1860 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3753 1860 : SpinLockRelease(&XLogCtl->info_lck);
3754 :
3755 1860 : return lastRemovedSegNo;
3756 : }
3757 :
3758 : /*
3759 : * Return the oldest WAL segment on the given TLI that still exists in
3760 : * XLOGDIR, or 0 if none.
3761 : */
3762 : XLogSegNo
3763 4 : XLogGetOldestSegno(TimeLineID tli)
3764 : {
3765 : DIR *xldir;
3766 : struct dirent *xlde;
3767 4 : XLogSegNo oldest_segno = 0;
3768 :
3769 4 : xldir = AllocateDir(XLOGDIR);
3770 28 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3771 : {
3772 : TimeLineID file_tli;
3773 : XLogSegNo file_segno;
3774 :
3775 : /* Ignore files that are not XLOG segments. */
3776 24 : if (!IsXLogFileName(xlde->d_name))
3777 16 : continue;
3778 :
3779 : /* Parse filename to get TLI and segno. */
3780 8 : XLogFromFileName(xlde->d_name, &file_tli, &file_segno,
3781 : wal_segment_size);
3782 :
3783 : /* Ignore anything that's not from the TLI of interest. */
3784 8 : if (tli != file_tli)
3785 0 : continue;
3786 :
3787 : /* If it's the oldest so far, update oldest_segno. */
3788 8 : if (oldest_segno == 0 || file_segno < oldest_segno)
3789 8 : oldest_segno = file_segno;
3790 : }
3791 :
3792 4 : FreeDir(xldir);
3793 4 : return oldest_segno;
3794 : }
3795 :
3796 : /*
3797 : * Update the last removed segno pointer in shared memory, to reflect that the
3798 : * given XLOG file has been removed.
3799 : */
3800 : static void
3801 1212 : UpdateLastRemovedPtr(char *filename)
3802 : {
3803 : uint32 tli;
3804 : XLogSegNo segno;
3805 :
3806 1212 : XLogFromFileName(filename, &tli, &segno, wal_segment_size);
3807 :
3808 1212 : SpinLockAcquire(&XLogCtl->info_lck);
3809 1212 : if (segno > XLogCtl->lastRemovedSegNo)
3810 450 : XLogCtl->lastRemovedSegNo = segno;
3811 1212 : SpinLockRelease(&XLogCtl->info_lck);
3812 1212 : }
3813 :
3814 : /*
3815 : * Remove all temporary log files in pg_wal
3816 : *
3817 : * This is called at the beginning of recovery after a previous crash,
3818 : * at a point where no other processes write fresh WAL data.
3819 : */
3820 : static void
3821 326 : RemoveTempXlogFiles(void)
3822 : {
3823 : DIR *xldir;
3824 : struct dirent *xlde;
3825 :
3826 326 : elog(DEBUG2, "removing all temporary WAL segments");
3827 :
3828 326 : xldir = AllocateDir(XLOGDIR);
3829 2068 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3830 : {
3831 : char path[MAXPGPATH];
3832 :
3833 1742 : if (strncmp(xlde->d_name, "xlogtemp.", 9) != 0)
3834 1742 : continue;
3835 :
3836 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
3837 0 : unlink(path);
3838 0 : elog(DEBUG2, "removed temporary WAL segment \"%s\"", path);
3839 : }
3840 326 : FreeDir(xldir);
3841 326 : }
3842 :
3843 : /*
3844 : * Recycle or remove all log files older or equal to passed segno.
3845 : *
3846 : * endptr is current (or recent) end of xlog, and lastredoptr is the
3847 : * redo pointer of the last checkpoint. These are used to determine
3848 : * whether we want to recycle rather than delete no-longer-wanted log files.
3849 : *
3850 : * insertTLI is the current timeline for XLOG insertion. Any recycled
3851 : * segments should be reused for this timeline.
3852 : */
3853 : static void
3854 1704 : RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr,
3855 : TimeLineID insertTLI)
3856 : {
3857 : DIR *xldir;
3858 : struct dirent *xlde;
3859 : char lastoff[MAXFNAMELEN];
3860 : XLogSegNo endlogSegNo;
3861 : XLogSegNo recycleSegNo;
3862 :
3863 : /* Initialize info about where to try to recycle to */
3864 1704 : XLByteToSeg(endptr, endlogSegNo, wal_segment_size);
3865 1704 : recycleSegNo = XLOGfileslop(lastredoptr);
3866 :
3867 : /*
3868 : * Construct a filename of the last segment to be kept. The timeline ID
3869 : * doesn't matter, we ignore that in the comparison. (During recovery,
3870 : * InsertTimeLineID isn't set, so we can't use that.)
3871 : */
3872 1704 : XLogFileName(lastoff, 0, segno, wal_segment_size);
3873 :
3874 1704 : elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
3875 : lastoff);
3876 :
3877 1704 : xldir = AllocateDir(XLOGDIR);
3878 :
3879 14874 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3880 : {
3881 : /* Ignore files that are not XLOG segments */
3882 13170 : if (!IsXLogFileName(xlde->d_name) &&
3883 6964 : !IsPartialXLogFileName(xlde->d_name))
3884 6960 : continue;
3885 :
3886 : /*
3887 : * We ignore the timeline part of the XLOG segment identifiers in
3888 : * deciding whether a segment is still needed. This ensures that we
3889 : * won't prematurely remove a segment from a parent timeline. We could
3890 : * probably be a little more proactive about removing segments of
3891 : * non-parent timelines, but that would be a whole lot more
3892 : * complicated.
3893 : *
3894 : * We use the alphanumeric sorting property of the filenames to decide
3895 : * which ones are earlier than the lastoff segment.
3896 : */
3897 6210 : if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
3898 : {
3899 1228 : if (XLogArchiveCheckDone(xlde->d_name))
3900 : {
3901 : /* Update the last removed location in shared memory first */
3902 1212 : UpdateLastRemovedPtr(xlde->d_name);
3903 :
3904 1212 : RemoveXlogFile(xlde, recycleSegNo, &endlogSegNo, insertTLI);
3905 : }
3906 : }
3907 : }
3908 :
3909 1704 : FreeDir(xldir);
3910 1704 : }
3911 :
3912 : /*
3913 : * Recycle or remove WAL files that are not part of the given timeline's
3914 : * history.
3915 : *
3916 : * This is called during recovery, whenever we switch to follow a new
3917 : * timeline, and at the end of recovery when we create a new timeline. We
3918 : * wouldn't otherwise care about extra WAL files lying in pg_wal, but they
3919 : * might be leftover pre-allocated or recycled WAL segments on the old timeline
3920 : * that we haven't used yet, and contain garbage. If we just leave them in
3921 : * pg_wal, they will eventually be archived, and we can't let that happen.
3922 : * Files that belong to our timeline history are valid, because we have
3923 : * successfully replayed them, but from others we can't be sure.
3924 : *
3925 : * 'switchpoint' is the current point in WAL where we switch to new timeline,
3926 : * and 'newTLI' is the new timeline we switch to.
3927 : */
3928 : void
3929 110 : RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
3930 : {
3931 : DIR *xldir;
3932 : struct dirent *xlde;
3933 : char switchseg[MAXFNAMELEN];
3934 : XLogSegNo endLogSegNo;
3935 : XLogSegNo switchLogSegNo;
3936 : XLogSegNo recycleSegNo;
3937 :
3938 : /*
3939 : * Initialize info about where to begin the work. This will recycle,
3940 : * somewhat arbitrarily, 10 future segments.
3941 : */
3942 110 : XLByteToPrevSeg(switchpoint, switchLogSegNo, wal_segment_size);
3943 110 : XLByteToSeg(switchpoint, endLogSegNo, wal_segment_size);
3944 110 : recycleSegNo = endLogSegNo + 10;
3945 :
3946 : /*
3947 : * Construct a filename of the last segment to be kept.
3948 : */
3949 110 : XLogFileName(switchseg, newTLI, switchLogSegNo, wal_segment_size);
3950 :
3951 110 : elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
3952 : switchseg);
3953 :
3954 110 : xldir = AllocateDir(XLOGDIR);
3955 :
3956 1052 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3957 : {
3958 : /* Ignore files that are not XLOG segments */
3959 942 : if (!IsXLogFileName(xlde->d_name))
3960 586 : continue;
3961 :
3962 : /*
3963 : * Remove files that are on a timeline older than the new one we're
3964 : * switching to, but with a segment number >= the first segment on the
3965 : * new timeline.
3966 : */
3967 356 : if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
3968 232 : strcmp(xlde->d_name + 8, switchseg + 8) > 0)
3969 : {
3970 : /*
3971 : * If the file has already been marked as .ready, however, don't
3972 : * remove it yet. It should be OK to remove it - files that are
3973 : * not part of our timeline history are not required for recovery
3974 : * - but seems safer to let them be archived and removed later.
3975 : */
3976 34 : if (!XLogArchiveIsReady(xlde->d_name))
3977 34 : RemoveXlogFile(xlde, recycleSegNo, &endLogSegNo, newTLI);
3978 : }
3979 : }
3980 :
3981 110 : FreeDir(xldir);
3982 110 : }
3983 :
3984 : /*
3985 : * Recycle or remove a log file that's no longer needed.
3986 : *
3987 : * segment_de is the dirent structure of the segment to recycle or remove.
3988 : * recycleSegNo is the segment number to recycle up to. endlogSegNo is
3989 : * the segment number of the current (or recent) end of WAL.
3990 : *
3991 : * endlogSegNo gets incremented if the segment is recycled so as it is not
3992 : * checked again with future callers of this function.
3993 : *
3994 : * insertTLI is the current timeline for XLOG insertion. Any recycled segments
3995 : * should be used for this timeline.
3996 : */
3997 : static void
3998 1246 : RemoveXlogFile(const struct dirent *segment_de,
3999 : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
4000 : TimeLineID insertTLI)
4001 : {
4002 : char path[MAXPGPATH];
4003 : #ifdef WIN32
4004 : char newpath[MAXPGPATH];
4005 : #endif
4006 1246 : const char *segname = segment_de->d_name;
4007 :
4008 1246 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
4009 :
4010 : /*
4011 : * Before deleting the file, see if it can be recycled as a future log
4012 : * segment. Only recycle normal files, because we don't want to recycle
4013 : * symbolic links pointing to a separate archive directory.
4014 : */
4015 1246 : if (wal_recycle &&
4016 1246 : *endlogSegNo <= recycleSegNo &&
4017 2380 : XLogCtl->InstallXLogFileSegmentActive && /* callee rechecks this */
4018 2372 : get_dirent_type(path, segment_de, false, DEBUG2) == PGFILETYPE_REG &&
4019 1186 : InstallXLogFileSegment(endlogSegNo, path,
4020 : true, recycleSegNo, insertTLI))
4021 : {
4022 1154 : ereport(DEBUG2,
4023 : (errmsg_internal("recycled write-ahead log file \"%s\"",
4024 : segname)));
4025 1154 : CheckpointStats.ckpt_segs_recycled++;
4026 : /* Needn't recheck that slot on future iterations */
4027 1154 : (*endlogSegNo)++;
4028 : }
4029 : else
4030 : {
4031 : /* No need for any more future segments, or recycling failed ... */
4032 : int rc;
4033 :
4034 92 : ereport(DEBUG2,
4035 : (errmsg_internal("removing write-ahead log file \"%s\"",
4036 : segname)));
4037 :
4038 : #ifdef WIN32
4039 :
4040 : /*
4041 : * On Windows, if another process (e.g another backend) holds the file
4042 : * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4043 : * will still show up in directory listing until the last handle is
4044 : * closed. To avoid confusing the lingering deleted file for a live
4045 : * WAL file that needs to be archived, rename it before deleting it.
4046 : *
4047 : * If another process holds the file open without FILE_SHARE_DELETE
4048 : * flag, rename will fail. We'll try again at the next checkpoint.
4049 : */
4050 : snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4051 : if (rename(path, newpath) != 0)
4052 : {
4053 : ereport(LOG,
4054 : (errcode_for_file_access(),
4055 : errmsg("could not rename file \"%s\": %m",
4056 : path)));
4057 : return;
4058 : }
4059 : rc = durable_unlink(newpath, LOG);
4060 : #else
4061 92 : rc = durable_unlink(path, LOG);
4062 : #endif
4063 92 : if (rc != 0)
4064 : {
4065 : /* Message already logged by durable_unlink() */
4066 0 : return;
4067 : }
4068 92 : CheckpointStats.ckpt_segs_removed++;
4069 : }
4070 :
4071 1246 : XLogArchiveCleanup(segname);
4072 : }
4073 :
4074 : /*
4075 : * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
4076 : * If the latter do not exist, recreate them.
4077 : *
4078 : * It is not the goal of this function to verify the contents of these
4079 : * directories, but to help in cases where someone has performed a cluster
4080 : * copy for PITR purposes but omitted pg_wal from the copy.
4081 : *
4082 : * We could also recreate pg_wal if it doesn't exist, but a deliberate
4083 : * policy decision was made not to. It is fairly common for pg_wal to be
4084 : * a symlink, and if that was the DBA's intent then automatically making a
4085 : * plain directory would result in degraded performance with no notice.
4086 : */
4087 : static void
4088 1520 : ValidateXLOGDirectoryStructure(void)
4089 : {
4090 : char path[MAXPGPATH];
4091 : struct stat stat_buf;
4092 :
4093 : /* Check for pg_wal; if it doesn't exist, error out */
4094 1520 : if (stat(XLOGDIR, &stat_buf) != 0 ||
4095 1520 : !S_ISDIR(stat_buf.st_mode))
4096 0 : ereport(FATAL,
4097 : (errcode_for_file_access(),
4098 : errmsg("required WAL directory \"%s\" does not exist",
4099 : XLOGDIR)));
4100 :
4101 : /* Check for archive_status */
4102 1520 : snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4103 1520 : if (stat(path, &stat_buf) == 0)
4104 : {
4105 : /* Check for weird cases where it exists but isn't a directory */
4106 1518 : if (!S_ISDIR(stat_buf.st_mode))
4107 0 : ereport(FATAL,
4108 : (errcode_for_file_access(),
4109 : errmsg("required WAL directory \"%s\" does not exist",
4110 : path)));
4111 : }
4112 : else
4113 : {
4114 2 : ereport(LOG,
4115 : (errmsg("creating missing WAL directory \"%s\"", path)));
4116 2 : if (MakePGDirectory(path) < 0)
4117 0 : ereport(FATAL,
4118 : (errcode_for_file_access(),
4119 : errmsg("could not create missing directory \"%s\": %m",
4120 : path)));
4121 : }
4122 :
4123 : /* Check for summaries */
4124 1520 : snprintf(path, MAXPGPATH, XLOGDIR "/summaries");
4125 1520 : if (stat(path, &stat_buf) == 0)
4126 : {
4127 : /* Check for weird cases where it exists but isn't a directory */
4128 1518 : if (!S_ISDIR(stat_buf.st_mode))
4129 0 : ereport(FATAL,
4130 : (errmsg("required WAL directory \"%s\" does not exist",
4131 : path)));
4132 : }
4133 : else
4134 : {
4135 2 : ereport(LOG,
4136 : (errmsg("creating missing WAL directory \"%s\"", path)));
4137 2 : if (MakePGDirectory(path) < 0)
4138 0 : ereport(FATAL,
4139 : (errmsg("could not create missing directory \"%s\": %m",
4140 : path)));
4141 : }
4142 1520 : }
4143 :
4144 : /*
4145 : * Remove previous backup history files. This also retries creation of
4146 : * .ready files for any backup history files for which XLogArchiveNotify
4147 : * failed earlier.
4148 : */
4149 : static void
4150 260 : CleanupBackupHistory(void)
4151 : {
4152 : DIR *xldir;
4153 : struct dirent *xlde;
4154 : char path[MAXPGPATH + sizeof(XLOGDIR)];
4155 :
4156 260 : xldir = AllocateDir(XLOGDIR);
4157 :
4158 2354 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4159 : {
4160 2094 : if (IsBackupHistoryFileName(xlde->d_name))
4161 : {
4162 274 : if (XLogArchiveCheckDone(xlde->d_name))
4163 : {
4164 224 : elog(DEBUG2, "removing WAL backup history file \"%s\"",
4165 : xlde->d_name);
4166 224 : snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
4167 224 : unlink(path);
4168 224 : XLogArchiveCleanup(xlde->d_name);
4169 : }
4170 : }
4171 : }
4172 :
4173 260 : FreeDir(xldir);
4174 260 : }
4175 :
4176 : /*
4177 : * I/O routines for pg_control
4178 : *
4179 : * *ControlFile is a buffer in shared memory that holds an image of the
4180 : * contents of pg_control. WriteControlFile() initializes pg_control
4181 : * given a preloaded buffer, ReadControlFile() loads the buffer from
4182 : * the pg_control file (during postmaster or standalone-backend startup),
4183 : * and UpdateControlFile() rewrites pg_control after we modify xlog state.
4184 : * InitControlFile() fills the buffer with initial values.
4185 : *
4186 : * For simplicity, WriteControlFile() initializes the fields of pg_control
4187 : * that are related to checking backend/database compatibility, and
4188 : * ReadControlFile() verifies they are correct. We could split out the
4189 : * I/O and compatibility-check functions, but there seems no need currently.
4190 : */
4191 :
4192 : static void
4193 80 : InitControlFile(uint64 sysidentifier)
4194 : {
4195 : char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
4196 :
4197 : /*
4198 : * Generate a random nonce. This is used for authentication requests that
4199 : * will fail because the user does not exist. The nonce is used to create
4200 : * a genuine-looking password challenge for the non-existent user, in lieu
4201 : * of an actual stored password.
4202 : */
4203 80 : if (!pg_strong_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
4204 0 : ereport(PANIC,
4205 : (errcode(ERRCODE_INTERNAL_ERROR),
4206 : errmsg("could not generate secret authorization token")));
4207 :
4208 80 : memset(ControlFile, 0, sizeof(ControlFileData));
4209 : /* Initialize pg_control status fields */
4210 80 : ControlFile->system_identifier = sysidentifier;
4211 80 : memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
4212 80 : ControlFile->state = DB_SHUTDOWNED;
4213 80 : ControlFile->unloggedLSN = FirstNormalUnloggedLSN;
4214 :
4215 : /* Set important parameter values for use when replaying WAL */
4216 80 : ControlFile->MaxConnections = MaxConnections;
4217 80 : ControlFile->max_worker_processes = max_worker_processes;
4218 80 : ControlFile->max_wal_senders = max_wal_senders;
4219 80 : ControlFile->max_prepared_xacts = max_prepared_xacts;
4220 80 : ControlFile->max_locks_per_xact = max_locks_per_xact;
4221 80 : ControlFile->wal_level = wal_level;
4222 80 : ControlFile->wal_log_hints = wal_log_hints;
4223 80 : ControlFile->track_commit_timestamp = track_commit_timestamp;
4224 80 : ControlFile->data_checksum_version = bootstrap_data_checksum_version;
4225 80 : }
4226 :
4227 : static void
4228 80 : WriteControlFile(void)
4229 : {
4230 : int fd;
4231 : char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4232 :
4233 : /*
4234 : * Initialize version and compatibility-check fields
4235 : */
4236 80 : ControlFile->pg_control_version = PG_CONTROL_VERSION;
4237 80 : ControlFile->catalog_version_no = CATALOG_VERSION_NO;
4238 :
4239 80 : ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4240 80 : ControlFile->floatFormat = FLOATFORMAT_VALUE;
4241 :
4242 80 : ControlFile->blcksz = BLCKSZ;
4243 80 : ControlFile->relseg_size = RELSEG_SIZE;
4244 80 : ControlFile->xlog_blcksz = XLOG_BLCKSZ;
4245 80 : ControlFile->xlog_seg_size = wal_segment_size;
4246 :
4247 80 : ControlFile->nameDataLen = NAMEDATALEN;
4248 80 : ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
4249 :
4250 80 : ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
4251 80 : ControlFile->loblksize = LOBLKSIZE;
4252 :
4253 80 : ControlFile->float8ByVal = FLOAT8PASSBYVAL;
4254 :
4255 : /* Contents are protected with a CRC */
4256 80 : INIT_CRC32C(ControlFile->crc);
4257 80 : COMP_CRC32C(ControlFile->crc,
4258 : (char *) ControlFile,
4259 : offsetof(ControlFileData, crc));
4260 80 : FIN_CRC32C(ControlFile->crc);
4261 :
4262 : /*
4263 : * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4264 : * the excess over sizeof(ControlFileData). This reduces the odds of
4265 : * premature-EOF errors when reading pg_control. We'll still fail when we
4266 : * check the contents of the file, but hopefully with a more specific
4267 : * error than "couldn't read pg_control".
4268 : */
4269 80 : memset(buffer, 0, PG_CONTROL_FILE_SIZE);
4270 80 : memcpy(buffer, ControlFile, sizeof(ControlFileData));
4271 :
4272 80 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4273 : O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
4274 80 : if (fd < 0)
4275 0 : ereport(PANIC,
4276 : (errcode_for_file_access(),
4277 : errmsg("could not create file \"%s\": %m",
4278 : XLOG_CONTROL_FILE)));
4279 :
4280 80 : errno = 0;
4281 80 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
4282 80 : if (write(fd, buffer, PG_CONTROL_FILE_SIZE) != PG_CONTROL_FILE_SIZE)
4283 : {
4284 : /* if write didn't set errno, assume problem is no disk space */
4285 0 : if (errno == 0)
4286 0 : errno = ENOSPC;
4287 0 : ereport(PANIC,
4288 : (errcode_for_file_access(),
4289 : errmsg("could not write to file \"%s\": %m",
4290 : XLOG_CONTROL_FILE)));
4291 : }
4292 80 : pgstat_report_wait_end();
4293 :
4294 80 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
4295 80 : if (pg_fsync(fd) != 0)
4296 0 : ereport(PANIC,
4297 : (errcode_for_file_access(),
4298 : errmsg("could not fsync file \"%s\": %m",
4299 : XLOG_CONTROL_FILE)));
4300 80 : pgstat_report_wait_end();
4301 :
4302 80 : if (close(fd) != 0)
4303 0 : ereport(PANIC,
4304 : (errcode_for_file_access(),
4305 : errmsg("could not close file \"%s\": %m",
4306 : XLOG_CONTROL_FILE)));
4307 80 : }
4308 :
4309 : static void
4310 1620 : ReadControlFile(void)
4311 : {
4312 : pg_crc32c crc;
4313 : int fd;
4314 : static char wal_segsz_str[20];
4315 : int r;
4316 :
4317 : /*
4318 : * Read data...
4319 : */
4320 1620 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4321 : O_RDWR | PG_BINARY);
4322 1620 : if (fd < 0)
4323 0 : ereport(PANIC,
4324 : (errcode_for_file_access(),
4325 : errmsg("could not open file \"%s\": %m",
4326 : XLOG_CONTROL_FILE)));
4327 :
4328 1620 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
4329 1620 : r = read(fd, ControlFile, sizeof(ControlFileData));
4330 1620 : if (r != sizeof(ControlFileData))
4331 : {
4332 0 : if (r < 0)
4333 0 : ereport(PANIC,
4334 : (errcode_for_file_access(),
4335 : errmsg("could not read file \"%s\": %m",
4336 : XLOG_CONTROL_FILE)));
4337 : else
4338 0 : ereport(PANIC,
4339 : (errcode(ERRCODE_DATA_CORRUPTED),
4340 : errmsg("could not read file \"%s\": read %d of %zu",
4341 : XLOG_CONTROL_FILE, r, sizeof(ControlFileData))));
4342 : }
4343 1620 : pgstat_report_wait_end();
4344 :
4345 1620 : close(fd);
4346 :
4347 : /*
4348 : * Check for expected pg_control format version. If this is wrong, the
4349 : * CRC check will likely fail because we'll be checking the wrong number
4350 : * of bytes. Complaining about wrong version will probably be more
4351 : * enlightening than complaining about wrong CRC.
4352 : */
4353 :
4354 1620 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
4355 0 : ereport(FATAL,
4356 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4357 : errmsg("database files are incompatible with server"),
4358 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4359 : " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4360 : ControlFile->pg_control_version, ControlFile->pg_control_version,
4361 : PG_CONTROL_VERSION, PG_CONTROL_VERSION),
4362 : errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4363 :
4364 1620 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
4365 0 : ereport(FATAL,
4366 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4367 : errmsg("database files are incompatible with server"),
4368 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4369 : " but the server was compiled with PG_CONTROL_VERSION %d.",
4370 : ControlFile->pg_control_version, PG_CONTROL_VERSION),
4371 : errhint("It looks like you need to initdb.")));
4372 :
4373 : /* Now check the CRC. */
4374 1620 : INIT_CRC32C(crc);
4375 1620 : COMP_CRC32C(crc,
4376 : (char *) ControlFile,
4377 : offsetof(ControlFileData, crc));
4378 1620 : FIN_CRC32C(crc);
4379 :
4380 1620 : if (!EQ_CRC32C(crc, ControlFile->crc))
4381 0 : ereport(FATAL,
4382 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4383 : errmsg("incorrect checksum in control file")));
4384 :
4385 : /*
4386 : * Do compatibility checking immediately. If the database isn't
4387 : * compatible with the backend executable, we want to abort before we can
4388 : * possibly do any damage.
4389 : */
4390 1620 : if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
4391 0 : ereport(FATAL,
4392 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4393 : errmsg("database files are incompatible with server"),
4394 : errdetail("The database cluster was initialized with CATALOG_VERSION_NO %d,"
4395 : " but the server was compiled with CATALOG_VERSION_NO %d.",
4396 : ControlFile->catalog_version_no, CATALOG_VERSION_NO),
4397 : errhint("It looks like you need to initdb.")));
4398 1620 : if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
4399 0 : ereport(FATAL,
4400 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4401 : errmsg("database files are incompatible with server"),
4402 : errdetail("The database cluster was initialized with MAXALIGN %d,"
4403 : " but the server was compiled with MAXALIGN %d.",
4404 : ControlFile->maxAlign, MAXIMUM_ALIGNOF),
4405 : errhint("It looks like you need to initdb.")));
4406 1620 : if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
4407 0 : ereport(FATAL,
4408 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4409 : errmsg("database files are incompatible with server"),
4410 : errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4411 : errhint("It looks like you need to initdb.")));
4412 1620 : if (ControlFile->blcksz != BLCKSZ)
4413 0 : ereport(FATAL,
4414 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4415 : errmsg("database files are incompatible with server"),
4416 : errdetail("The database cluster was initialized with BLCKSZ %d,"
4417 : " but the server was compiled with BLCKSZ %d.",
4418 : ControlFile->blcksz, BLCKSZ),
4419 : errhint("It looks like you need to recompile or initdb.")));
4420 1620 : if (ControlFile->relseg_size != RELSEG_SIZE)
4421 0 : ereport(FATAL,
4422 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4423 : errmsg("database files are incompatible with server"),
4424 : errdetail("The database cluster was initialized with RELSEG_SIZE %d,"
4425 : " but the server was compiled with RELSEG_SIZE %d.",
4426 : ControlFile->relseg_size, RELSEG_SIZE),
4427 : errhint("It looks like you need to recompile or initdb.")));
4428 1620 : if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
4429 0 : ereport(FATAL,
4430 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4431 : errmsg("database files are incompatible with server"),
4432 : errdetail("The database cluster was initialized with XLOG_BLCKSZ %d,"
4433 : " but the server was compiled with XLOG_BLCKSZ %d.",
4434 : ControlFile->xlog_blcksz, XLOG_BLCKSZ),
4435 : errhint("It looks like you need to recompile or initdb.")));
4436 1620 : if (ControlFile->nameDataLen != NAMEDATALEN)
4437 0 : ereport(FATAL,
4438 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4439 : errmsg("database files are incompatible with server"),
4440 : errdetail("The database cluster was initialized with NAMEDATALEN %d,"
4441 : " but the server was compiled with NAMEDATALEN %d.",
4442 : ControlFile->nameDataLen, NAMEDATALEN),
4443 : errhint("It looks like you need to recompile or initdb.")));
4444 1620 : if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
4445 0 : ereport(FATAL,
4446 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4447 : errmsg("database files are incompatible with server"),
4448 : errdetail("The database cluster was initialized with INDEX_MAX_KEYS %d,"
4449 : " but the server was compiled with INDEX_MAX_KEYS %d.",
4450 : ControlFile->indexMaxKeys, INDEX_MAX_KEYS),
4451 : errhint("It looks like you need to recompile or initdb.")));
4452 1620 : if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
4453 0 : ereport(FATAL,
4454 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4455 : errmsg("database files are incompatible with server"),
4456 : errdetail("The database cluster was initialized with TOAST_MAX_CHUNK_SIZE %d,"
4457 : " but the server was compiled with TOAST_MAX_CHUNK_SIZE %d.",
4458 : ControlFile->toast_max_chunk_size, (int) TOAST_MAX_CHUNK_SIZE),
4459 : errhint("It looks like you need to recompile or initdb.")));
4460 1620 : if (ControlFile->loblksize != LOBLKSIZE)
4461 0 : ereport(FATAL,
4462 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4463 : errmsg("database files are incompatible with server"),
4464 : errdetail("The database cluster was initialized with LOBLKSIZE %d,"
4465 : " but the server was compiled with LOBLKSIZE %d.",
4466 : ControlFile->loblksize, (int) LOBLKSIZE),
4467 : errhint("It looks like you need to recompile or initdb.")));
4468 :
4469 : #ifdef USE_FLOAT8_BYVAL
4470 1620 : if (ControlFile->float8ByVal != true)
4471 0 : ereport(FATAL,
4472 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4473 : errmsg("database files are incompatible with server"),
4474 : errdetail("The database cluster was initialized without USE_FLOAT8_BYVAL"
4475 : " but the server was compiled with USE_FLOAT8_BYVAL."),
4476 : errhint("It looks like you need to recompile or initdb.")));
4477 : #else
4478 : if (ControlFile->float8ByVal != false)
4479 : ereport(FATAL,
4480 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4481 : errmsg("database files are incompatible with server"),
4482 : errdetail("The database cluster was initialized with USE_FLOAT8_BYVAL"
4483 : " but the server was compiled without USE_FLOAT8_BYVAL."),
4484 : errhint("It looks like you need to recompile or initdb.")));
4485 : #endif
4486 :
4487 1620 : wal_segment_size = ControlFile->xlog_seg_size;
4488 :
4489 1620 : if (!IsValidWalSegSize(wal_segment_size))
4490 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4491 : errmsg_plural("invalid WAL segment size in control file (%d byte)",
4492 : "invalid WAL segment size in control file (%d bytes)",
4493 : wal_segment_size,
4494 : wal_segment_size),
4495 : errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.")));
4496 :
4497 1620 : snprintf(wal_segsz_str, sizeof(wal_segsz_str), "%d", wal_segment_size);
4498 1620 : SetConfigOption("wal_segment_size", wal_segsz_str, PGC_INTERNAL,
4499 : PGC_S_DYNAMIC_DEFAULT);
4500 :
4501 : /* check and update variables dependent on wal_segment_size */
4502 1620 : if (ConvertToXSegs(min_wal_size_mb, wal_segment_size) < 2)
4503 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4504 : errmsg("min_wal_size must be at least twice wal_segment_size")));
4505 :
4506 1620 : if (ConvertToXSegs(max_wal_size_mb, wal_segment_size) < 2)
4507 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4508 : errmsg("max_wal_size must be at least twice wal_segment_size")));
4509 :
4510 1620 : UsableBytesInSegment =
4511 1620 : (wal_segment_size / XLOG_BLCKSZ * UsableBytesInPage) -
4512 : (SizeOfXLogLongPHD - SizeOfXLogShortPHD);
4513 :
4514 1620 : CalculateCheckpointSegments();
4515 :
4516 : /* Make the initdb settings visible as GUC variables, too */
4517 1620 : SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4518 : PGC_INTERNAL, PGC_S_DYNAMIC_DEFAULT);
4519 1620 : }
4520 :
4521 : /*
4522 : * Utility wrapper to update the control file. Note that the control
4523 : * file gets flushed.
4524 : */
4525 : static void
4526 17934 : UpdateControlFile(void)
4527 : {
4528 17934 : update_controlfile(DataDir, ControlFile, true);
4529 17934 : }
4530 :
4531 : /*
4532 : * Returns the unique system identifier from control file.
4533 : */
4534 : uint64
4535 2402 : GetSystemIdentifier(void)
4536 : {
4537 : Assert(ControlFile != NULL);
4538 2402 : return ControlFile->system_identifier;
4539 : }
4540 :
4541 : /*
4542 : * Returns the random nonce from control file.
4543 : */
4544 : char *
4545 2 : GetMockAuthenticationNonce(void)
4546 : {
4547 : Assert(ControlFile != NULL);
4548 2 : return ControlFile->mock_authentication_nonce;
4549 : }
4550 :
4551 : /*
4552 : * Are checksums enabled for data pages?
4553 : */
4554 : bool
4555 19439700 : DataChecksumsEnabled(void)
4556 : {
4557 : Assert(ControlFile != NULL);
4558 19439700 : return (ControlFile->data_checksum_version > 0);
4559 : }
4560 :
4561 : /*
4562 : * Returns a fake LSN for unlogged relations.
4563 : *
4564 : * Each call generates an LSN that is greater than any previous value
4565 : * returned. The current counter value is saved and restored across clean
4566 : * shutdowns, but like unlogged relations, does not survive a crash. This can
4567 : * be used in lieu of real LSN values returned by XLogInsert, if you need an
4568 : * LSN-like increasing sequence of numbers without writing any WAL.
4569 : */
4570 : XLogRecPtr
4571 66 : GetFakeLSNForUnloggedRel(void)
4572 : {
4573 66 : return pg_atomic_fetch_add_u64(&XLogCtl->unloggedLSN, 1);
4574 : }
4575 :
4576 : /*
4577 : * Auto-tune the number of XLOG buffers.
4578 : *
4579 : * The preferred setting for wal_buffers is about 3% of shared_buffers, with
4580 : * a maximum of one XLOG segment (there is little reason to think that more
4581 : * is helpful, at least so long as we force an fsync when switching log files)
4582 : * and a minimum of 8 blocks (which was the default value prior to PostgreSQL
4583 : * 9.1, when auto-tuning was added).
4584 : *
4585 : * This should not be called until NBuffers has received its final value.
4586 : */
4587 : static int
4588 1766 : XLOGChooseNumBuffers(void)
4589 : {
4590 : int xbuffers;
4591 :
4592 1766 : xbuffers = NBuffers / 32;
4593 1766 : if (xbuffers > (wal_segment_size / XLOG_BLCKSZ))
4594 42 : xbuffers = (wal_segment_size / XLOG_BLCKSZ);
4595 1766 : if (xbuffers < 8)
4596 708 : xbuffers = 8;
4597 1766 : return xbuffers;
4598 : }
4599 :
4600 : /*
4601 : * GUC check_hook for wal_buffers
4602 : */
4603 : bool
4604 3596 : check_wal_buffers(int *newval, void **extra, GucSource source)
4605 : {
4606 : /*
4607 : * -1 indicates a request for auto-tune.
4608 : */
4609 3596 : if (*newval == -1)
4610 : {
4611 : /*
4612 : * If we haven't yet changed the boot_val default of -1, just let it
4613 : * be. We'll fix it when XLOGShmemSize is called.
4614 : */
4615 1830 : if (XLOGbuffers == -1)
4616 1830 : return true;
4617 :
4618 : /* Otherwise, substitute the auto-tune value */
4619 0 : *newval = XLOGChooseNumBuffers();
4620 : }
4621 :
4622 : /*
4623 : * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
4624 : * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
4625 : * the case, we just silently treat such values as a request for the
4626 : * minimum. (We could throw an error instead, but that doesn't seem very
4627 : * helpful.)
4628 : */
4629 1766 : if (*newval < 4)
4630 0 : *newval = 4;
4631 :
4632 1766 : return true;
4633 : }
4634 :
4635 : /*
4636 : * GUC check_hook for wal_consistency_checking
4637 : */
4638 : bool
4639 1834 : check_wal_consistency_checking(char **newval, void **extra, GucSource source)
4640 : {
4641 : char *rawstring;
4642 : List *elemlist;
4643 : ListCell *l;
4644 : bool newwalconsistency[RM_MAX_ID + 1];
4645 :
4646 : /* Initialize the array */
4647 60522 : MemSet(newwalconsistency, 0, (RM_MAX_ID + 1) * sizeof(bool));
4648 :
4649 : /* Need a modifiable copy of string */
4650 1834 : rawstring = pstrdup(*newval);
4651 :
4652 : /* Parse string into list of identifiers */
4653 1834 : if (!SplitIdentifierString(rawstring, ',', &elemlist))
4654 : {
4655 : /* syntax error in list */
4656 0 : GUC_check_errdetail("List syntax is invalid.");
4657 0 : pfree(rawstring);
4658 0 : list_free(elemlist);
4659 0 : return false;
4660 : }
4661 :
4662 1838 : foreach(l, elemlist)
4663 : {
4664 4 : char *tok = (char *) lfirst(l);
4665 : int rmid;
4666 :
4667 : /* Check for 'all'. */
4668 4 : if (pg_strcasecmp(tok, "all") == 0)
4669 : {
4670 0 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4671 0 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL)
4672 0 : newwalconsistency[rmid] = true;
4673 : }
4674 : else
4675 : {
4676 : /* Check if the token matches any known resource manager. */
4677 4 : bool found = false;
4678 :
4679 72 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4680 : {
4681 108 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL &&
4682 36 : pg_strcasecmp(tok, GetRmgr(rmid).rm_name) == 0)
4683 : {
4684 4 : newwalconsistency[rmid] = true;
4685 4 : found = true;
4686 4 : break;
4687 : }
4688 : }
4689 4 : if (!found)
4690 : {
4691 : /*
4692 : * During startup, it might be a not-yet-loaded custom
4693 : * resource manager. Defer checking until
4694 : * InitializeWalConsistencyChecking().
4695 : */
4696 0 : if (!process_shared_preload_libraries_done)
4697 : {
4698 0 : check_wal_consistency_checking_deferred = true;
4699 : }
4700 : else
4701 : {
4702 0 : GUC_check_errdetail("Unrecognized key word: \"%s\".", tok);
4703 0 : pfree(rawstring);
4704 0 : list_free(elemlist);
4705 0 : return false;
4706 : }
4707 : }
4708 : }
4709 : }
4710 :
4711 1834 : pfree(rawstring);
4712 1834 : list_free(elemlist);
4713 :
4714 : /* assign new value */
4715 1834 : *extra = guc_malloc(ERROR, (RM_MAX_ID + 1) * sizeof(bool));
4716 1834 : memcpy(*extra, newwalconsistency, (RM_MAX_ID + 1) * sizeof(bool));
4717 1834 : return true;
4718 : }
4719 :
4720 : /*
4721 : * GUC assign_hook for wal_consistency_checking
4722 : */
4723 : void
4724 1834 : assign_wal_consistency_checking(const char *newval, void *extra)
4725 : {
4726 : /*
4727 : * If some checks were deferred, it's possible that the checks will fail
4728 : * later during InitializeWalConsistencyChecking(). But in that case, the
4729 : * postmaster will exit anyway, so it's safe to proceed with the
4730 : * assignment.
4731 : *
4732 : * Any built-in resource managers specified are assigned immediately,
4733 : * which affects WAL created before shared_preload_libraries are
4734 : * processed. Any custom resource managers specified won't be assigned
4735 : * until after shared_preload_libraries are processed, but that's OK
4736 : * because WAL for a custom resource manager can't be written before the
4737 : * module is loaded anyway.
4738 : */
4739 1834 : wal_consistency_checking = extra;
4740 1834 : }
4741 :
4742 : /*
4743 : * InitializeWalConsistencyChecking: run after loading custom resource managers
4744 : *
4745 : * If any unknown resource managers were specified in the
4746 : * wal_consistency_checking GUC, processing was deferred. Now that
4747 : * shared_preload_libraries have been loaded, process wal_consistency_checking
4748 : * again.
4749 : */
4750 : void
4751 1526 : InitializeWalConsistencyChecking(void)
4752 : {
4753 : Assert(process_shared_preload_libraries_done);
4754 :
4755 1526 : if (check_wal_consistency_checking_deferred)
4756 : {
4757 : struct config_generic *guc;
4758 :
4759 0 : guc = find_option("wal_consistency_checking", false, false, ERROR);
4760 :
4761 0 : check_wal_consistency_checking_deferred = false;
4762 :
4763 0 : set_config_option_ext("wal_consistency_checking",
4764 : wal_consistency_checking_string,
4765 : guc->scontext, guc->source, guc->srole,
4766 : GUC_ACTION_SET, true, ERROR, false);
4767 :
4768 : /* checking should not be deferred again */
4769 : Assert(!check_wal_consistency_checking_deferred);
4770 : }
4771 1526 : }
4772 :
4773 : /*
4774 : * GUC show_hook for archive_command
4775 : */
4776 : const char *
4777 3888 : show_archive_command(void)
4778 : {
4779 3888 : if (XLogArchivingActive())
4780 0 : return XLogArchiveCommand;
4781 : else
4782 3888 : return "(disabled)";
4783 : }
4784 :
4785 : /*
4786 : * GUC show_hook for in_hot_standby
4787 : */
4788 : const char *
4789 25432 : show_in_hot_standby(void)
4790 : {
4791 : /*
4792 : * We display the actual state based on shared memory, so that this GUC
4793 : * reports up-to-date state if examined intra-query. The underlying
4794 : * variable (in_hot_standby_guc) changes only when we transmit a new value
4795 : * to the client.
4796 : */
4797 25432 : return RecoveryInProgress() ? "on" : "off";
4798 : }
4799 :
4800 : /*
4801 : * Read the control file, set respective GUCs.
4802 : *
4803 : * This is to be called during startup, including a crash recovery cycle,
4804 : * unless in bootstrap mode, where no control file yet exists. As there's no
4805 : * usable shared memory yet (its sizing can depend on the contents of the
4806 : * control file!), first store the contents in local memory. XLOGShmemInit()
4807 : * will then copy it to shared memory later.
4808 : *
4809 : * reset just controls whether previous contents are to be expected (in the
4810 : * reset case, there's a dangling pointer into old shared memory), or not.
4811 : */
4812 : void
4813 1540 : LocalProcessControlFile(bool reset)
4814 : {
4815 : Assert(reset || ControlFile == NULL);
4816 1540 : ControlFile = palloc(sizeof(ControlFileData));
4817 1540 : ReadControlFile();
4818 1540 : }
4819 :
4820 : /*
4821 : * Get the wal_level from the control file. For a standby, this value should be
4822 : * considered as its active wal_level, because it may be different from what
4823 : * was originally configured on standby.
4824 : */
4825 : WalLevel
4826 132 : GetActiveWalLevelOnStandby(void)
4827 : {
4828 132 : return ControlFile->wal_level;
4829 : }
4830 :
4831 : /*
4832 : * Initialization of shared memory for XLOG
4833 : */
4834 : Size
4835 5066 : XLOGShmemSize(void)
4836 : {
4837 : Size size;
4838 :
4839 : /*
4840 : * If the value of wal_buffers is -1, use the preferred auto-tune value.
4841 : * This isn't an amazingly clean place to do this, but we must wait till
4842 : * NBuffers has received its final value, and must do it before using the
4843 : * value of XLOGbuffers to do anything important.
4844 : *
4845 : * We prefer to report this value's source as PGC_S_DYNAMIC_DEFAULT.
4846 : * However, if the DBA explicitly set wal_buffers = -1 in the config file,
4847 : * then PGC_S_DYNAMIC_DEFAULT will fail to override that and we must force
4848 : * the matter with PGC_S_OVERRIDE.
4849 : */
4850 5066 : if (XLOGbuffers == -1)
4851 : {
4852 : char buf[32];
4853 :
4854 1766 : snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
4855 1766 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4856 : PGC_S_DYNAMIC_DEFAULT);
4857 1766 : if (XLOGbuffers == -1) /* failed to apply it? */
4858 0 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4859 : PGC_S_OVERRIDE);
4860 : }
4861 : Assert(XLOGbuffers > 0);
4862 :
4863 : /* XLogCtl */
4864 5066 : size = sizeof(XLogCtlData);
4865 :
4866 : /* WAL insertion locks, plus alignment */
4867 5066 : size = add_size(size, mul_size(sizeof(WALInsertLockPadded), NUM_XLOGINSERT_LOCKS + 1));
4868 : /* xlblocks array */
4869 5066 : size = add_size(size, mul_size(sizeof(pg_atomic_uint64), XLOGbuffers));
4870 : /* extra alignment padding for XLOG I/O buffers */
4871 5066 : size = add_size(size, Max(XLOG_BLCKSZ, PG_IO_ALIGN_SIZE));
4872 : /* and the buffers themselves */
4873 5066 : size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4874 :
4875 : /*
4876 : * Note: we don't count ControlFileData, it comes out of the "slop factor"
4877 : * added by CreateSharedMemoryAndSemaphores. This lets us use this
4878 : * routine again below to compute the actual allocation size.
4879 : */
4880 :
4881 5066 : return size;
4882 : }
4883 :
4884 : void
4885 1768 : XLOGShmemInit(void)
4886 : {
4887 : bool foundCFile,
4888 : foundXLog;
4889 : char *allocptr;
4890 : int i;
4891 : ControlFileData *localControlFile;
4892 :
4893 : #ifdef WAL_DEBUG
4894 :
4895 : /*
4896 : * Create a memory context for WAL debugging that's exempt from the normal
4897 : * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4898 : * an allocation fails, but wal_debug is not for production use anyway.
4899 : */
4900 : if (walDebugCxt == NULL)
4901 : {
4902 : walDebugCxt = AllocSetContextCreate(TopMemoryContext,
4903 : "WAL Debug",
4904 : ALLOCSET_DEFAULT_SIZES);
4905 : MemoryContextAllowInCriticalSection(walDebugCxt, true);
4906 : }
4907 : #endif
4908 :
4909 :
4910 1768 : XLogCtl = (XLogCtlData *)
4911 1768 : ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
4912 :
4913 1768 : localControlFile = ControlFile;
4914 1768 : ControlFile = (ControlFileData *)
4915 1768 : ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
4916 :
4917 1768 : if (foundCFile || foundXLog)
4918 : {
4919 : /* both should be present or neither */
4920 : Assert(foundCFile && foundXLog);
4921 :
4922 : /* Initialize local copy of WALInsertLocks */
4923 0 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks;
4924 :
4925 0 : if (localControlFile)
4926 0 : pfree(localControlFile);
4927 0 : return;
4928 : }
4929 1768 : memset(XLogCtl, 0, sizeof(XLogCtlData));
4930 :
4931 : /*
4932 : * Already have read control file locally, unless in bootstrap mode. Move
4933 : * contents into shared memory.
4934 : */
4935 1768 : if (localControlFile)
4936 : {
4937 1528 : memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
4938 1528 : pfree(localControlFile);
4939 : }
4940 :
4941 : /*
4942 : * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
4943 : * multiple of the alignment for same, so no extra alignment padding is
4944 : * needed here.
4945 : */
4946 1768 : allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
4947 1768 : XLogCtl->xlblocks = (pg_atomic_uint64 *) allocptr;
4948 1768 : allocptr += sizeof(pg_atomic_uint64) * XLOGbuffers;
4949 :
4950 493528 : for (i = 0; i < XLOGbuffers; i++)
4951 : {
4952 491760 : pg_atomic_init_u64(&XLogCtl->xlblocks[i], InvalidXLogRecPtr);
4953 : }
4954 :
4955 : /* WAL insertion locks. Ensure they're aligned to the full padded size */
4956 1768 : allocptr += sizeof(WALInsertLockPadded) -
4957 1768 : ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
4958 1768 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks =
4959 : (WALInsertLockPadded *) allocptr;
4960 1768 : allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
4961 :
4962 15912 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
4963 : {
4964 14144 : LWLockInitialize(&WALInsertLocks[i].l.lock, LWTRANCHE_WAL_INSERT);
4965 14144 : pg_atomic_init_u64(&WALInsertLocks[i].l.insertingAt, InvalidXLogRecPtr);
4966 14144 : WALInsertLocks[i].l.lastImportantAt = InvalidXLogRecPtr;
4967 : }
4968 :
4969 : /*
4970 : * Align the start of the page buffers to a full xlog block size boundary.
4971 : * This simplifies some calculations in XLOG insertion. It is also
4972 : * required for O_DIRECT.
4973 : */
4974 1768 : allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
4975 1768 : XLogCtl->pages = allocptr;
4976 1768 : memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
4977 :
4978 : /*
4979 : * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
4980 : * in additional info.)
4981 : */
4982 1768 : XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
4983 1768 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
4984 1768 : XLogCtl->InstallXLogFileSegmentActive = false;
4985 1768 : XLogCtl->WalWriterSleeping = false;
4986 :
4987 1768 : SpinLockInit(&XLogCtl->Insert.insertpos_lck);
4988 1768 : SpinLockInit(&XLogCtl->info_lck);
4989 1768 : pg_atomic_init_u64(&XLogCtl->logInsertResult, InvalidXLogRecPtr);
4990 1768 : pg_atomic_init_u64(&XLogCtl->logWriteResult, InvalidXLogRecPtr);
4991 1768 : pg_atomic_init_u64(&XLogCtl->logFlushResult, InvalidXLogRecPtr);
4992 1768 : pg_atomic_init_u64(&XLogCtl->unloggedLSN, InvalidXLogRecPtr);
4993 : }
4994 :
4995 : /*
4996 : * This func must be called ONCE on system install. It creates pg_control
4997 : * and the initial XLOG segment.
4998 : */
4999 : void
5000 80 : BootStrapXLOG(void)
5001 : {
5002 : CheckPoint checkPoint;
5003 : char *buffer;
5004 : XLogPageHeader page;
5005 : XLogLongPageHeader longpage;
5006 : XLogRecord *record;
5007 : char *recptr;
5008 : uint64 sysidentifier;
5009 : struct timeval tv;
5010 : pg_crc32c crc;
5011 :
5012 : /* allow ordinary WAL segment creation, like StartupXLOG() would */
5013 80 : SetInstallXLogFileSegmentActive();
5014 :
5015 : /*
5016 : * Select a hopefully-unique system identifier code for this installation.
5017 : * We use the result of gettimeofday(), including the fractional seconds
5018 : * field, as being about as unique as we can easily get. (Think not to
5019 : * use random(), since it hasn't been seeded and there's no portable way
5020 : * to seed it other than the system clock value...) The upper half of the
5021 : * uint64 value is just the tv_sec part, while the lower half contains the
5022 : * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
5023 : * PID for a little extra uniqueness. A person knowing this encoding can
5024 : * determine the initialization time of the installation, which could
5025 : * perhaps be useful sometimes.
5026 : */
5027 80 : gettimeofday(&tv, NULL);
5028 80 : sysidentifier = ((uint64) tv.tv_sec) << 32;
5029 80 : sysidentifier |= ((uint64) tv.tv_usec) << 12;
5030 80 : sysidentifier |= getpid() & 0xFFF;
5031 :
5032 : /* page buffer must be aligned suitably for O_DIRECT */
5033 80 : buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
5034 80 : page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
5035 80 : memset(page, 0, XLOG_BLCKSZ);
5036 :
5037 : /*
5038 : * Set up information for the initial checkpoint record
5039 : *
5040 : * The initial checkpoint record is written to the beginning of the WAL
5041 : * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5042 : * used, so that we can use 0/0 to mean "before any valid WAL segment".
5043 : */
5044 80 : checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
5045 80 : checkPoint.ThisTimeLineID = BootstrapTimeLineID;
5046 80 : checkPoint.PrevTimeLineID = BootstrapTimeLineID;
5047 80 : checkPoint.fullPageWrites = fullPageWrites;
5048 : checkPoint.nextXid =
5049 80 : FullTransactionIdFromEpochAndXid(0, FirstNormalTransactionId);
5050 80 : checkPoint.nextOid = FirstGenbkiObjectId;
5051 80 : checkPoint.nextMulti = FirstMultiXactId;
5052 80 : checkPoint.nextMultiOffset = 0;
5053 80 : checkPoint.oldestXid = FirstNormalTransactionId;
5054 80 : checkPoint.oldestXidDB = Template1DbOid;
5055 80 : checkPoint.oldestMulti = FirstMultiXactId;
5056 80 : checkPoint.oldestMultiDB = Template1DbOid;
5057 80 : checkPoint.oldestCommitTsXid = InvalidTransactionId;
5058 80 : checkPoint.newestCommitTsXid = InvalidTransactionId;
5059 80 : checkPoint.time = (pg_time_t) time(NULL);
5060 80 : checkPoint.oldestActiveXid = InvalidTransactionId;
5061 :
5062 80 : TransamVariables->nextXid = checkPoint.nextXid;
5063 80 : TransamVariables->nextOid = checkPoint.nextOid;
5064 80 : TransamVariables->oidCount = 0;
5065 80 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5066 80 : AdvanceOldestClogXid(checkPoint.oldestXid);
5067 80 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5068 80 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5069 80 : SetCommitTsLimit(InvalidTransactionId, InvalidTransactionId);
5070 :
5071 : /* Set up the XLOG page header */
5072 80 : page->xlp_magic = XLOG_PAGE_MAGIC;
5073 80 : page->xlp_info = XLP_LONG_HEADER;
5074 80 : page->xlp_tli = BootstrapTimeLineID;
5075 80 : page->xlp_pageaddr = wal_segment_size;
5076 80 : longpage = (XLogLongPageHeader) page;
5077 80 : longpage->xlp_sysid = sysidentifier;
5078 80 : longpage->xlp_seg_size = wal_segment_size;
5079 80 : longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5080 :
5081 : /* Insert the initial checkpoint record */
5082 80 : recptr = ((char *) page + SizeOfXLogLongPHD);
5083 80 : record = (XLogRecord *) recptr;
5084 80 : record->xl_prev = 0;
5085 80 : record->xl_xid = InvalidTransactionId;
5086 80 : record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
5087 80 : record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
5088 80 : record->xl_rmid = RM_XLOG_ID;
5089 80 : recptr += SizeOfXLogRecord;
5090 : /* fill the XLogRecordDataHeaderShort struct */
5091 80 : *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
5092 80 : *(recptr++) = sizeof(checkPoint);
5093 80 : memcpy(recptr, &checkPoint, sizeof(checkPoint));
5094 80 : recptr += sizeof(checkPoint);
5095 : Assert(recptr - (char *) record == record->xl_tot_len);
5096 :
5097 80 : INIT_CRC32C(crc);
5098 80 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
5099 80 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5100 80 : FIN_CRC32C(crc);
5101 80 : record->xl_crc = crc;
5102 :
5103 : /* Create first XLOG segment file */
5104 80 : openLogTLI = BootstrapTimeLineID;
5105 80 : openLogFile = XLogFileInit(1, BootstrapTimeLineID);
5106 :
5107 : /*
5108 : * We needn't bother with Reserve/ReleaseExternalFD here, since we'll
5109 : * close the file again in a moment.
5110 : */
5111 :
5112 : /* Write the first page with the initial record */
5113 80 : errno = 0;
5114 80 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
5115 80 : if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5116 : {
5117 : /* if write didn't set errno, assume problem is no disk space */
5118 0 : if (errno == 0)
5119 0 : errno = ENOSPC;
5120 0 : ereport(PANIC,
5121 : (errcode_for_file_access(),
5122 : errmsg("could not write bootstrap write-ahead log file: %m")));
5123 : }
5124 80 : pgstat_report_wait_end();
5125 :
5126 80 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
5127 80 : if (pg_fsync(openLogFile) != 0)
5128 0 : ereport(PANIC,
5129 : (errcode_for_file_access(),
5130 : errmsg("could not fsync bootstrap write-ahead log file: %m")));
5131 80 : pgstat_report_wait_end();
5132 :
5133 80 : if (close(openLogFile) != 0)
5134 0 : ereport(PANIC,
5135 : (errcode_for_file_access(),
5136 : errmsg("could not close bootstrap write-ahead log file: %m")));
5137 :
5138 80 : openLogFile = -1;
5139 :
5140 : /* Now create pg_control */
5141 80 : InitControlFile(sysidentifier);
5142 80 : ControlFile->time = checkPoint.time;
5143 80 : ControlFile->checkPoint = checkPoint.redo;
5144 80 : ControlFile->checkPointCopy = checkPoint;
5145 :
5146 : /* some additional ControlFile fields are set in WriteControlFile() */
5147 80 : WriteControlFile();
5148 :
5149 : /* Bootstrap the commit log, too */
5150 80 : BootStrapCLOG();
5151 80 : BootStrapCommitTs();
5152 80 : BootStrapSUBTRANS();
5153 80 : BootStrapMultiXact();
5154 :
5155 80 : pfree(buffer);
5156 :
5157 : /*
5158 : * Force control file to be read - in contrast to normal processing we'd
5159 : * otherwise never run the checks and GUC related initializations therein.
5160 : */
5161 80 : ReadControlFile();
5162 80 : }
5163 :
5164 : static char *
5165 1360 : str_time(pg_time_t tnow)
5166 : {
5167 : static char buf[128];
5168 :
5169 1360 : pg_strftime(buf, sizeof(buf),
5170 : "%Y-%m-%d %H:%M:%S %Z",
5171 1360 : pg_localtime(&tnow, log_timezone));
5172 :
5173 1360 : return buf;
5174 : }
5175 :
5176 : /*
5177 : * Initialize the first WAL segment on new timeline.
5178 : */
5179 : static void
5180 90 : XLogInitNewTimeline(TimeLineID endTLI, XLogRecPtr endOfLog, TimeLineID newTLI)
5181 : {
5182 : char xlogfname[MAXFNAMELEN];
5183 : XLogSegNo endLogSegNo;
5184 : XLogSegNo startLogSegNo;
5185 :
5186 : /* we always switch to a new timeline after archive recovery */
5187 : Assert(endTLI != newTLI);
5188 :
5189 : /*
5190 : * Update min recovery point one last time.
5191 : */
5192 90 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5193 :
5194 : /*
5195 : * Calculate the last segment on the old timeline, and the first segment
5196 : * on the new timeline. If the switch happens in the middle of a segment,
5197 : * they are the same, but if the switch happens exactly at a segment
5198 : * boundary, startLogSegNo will be endLogSegNo + 1.
5199 : */
5200 90 : XLByteToPrevSeg(endOfLog, endLogSegNo, wal_segment_size);
5201 90 : XLByteToSeg(endOfLog, startLogSegNo, wal_segment_size);
5202 :
5203 : /*
5204 : * Initialize the starting WAL segment for the new timeline. If the switch
5205 : * happens in the middle of a segment, copy data from the last WAL segment
5206 : * of the old timeline up to the switch point, to the starting WAL segment
5207 : * on the new timeline.
5208 : */
5209 90 : if (endLogSegNo == startLogSegNo)
5210 : {
5211 : /*
5212 : * Make a copy of the file on the new timeline.
5213 : *
5214 : * Writing WAL isn't allowed yet, so there are no locking
5215 : * considerations. But we should be just as tense as XLogFileInit to
5216 : * avoid emplacing a bogus file.
5217 : */
5218 74 : XLogFileCopy(newTLI, endLogSegNo, endTLI, endLogSegNo,
5219 74 : XLogSegmentOffset(endOfLog, wal_segment_size));
5220 : }
5221 : else
5222 : {
5223 : /*
5224 : * The switch happened at a segment boundary, so just create the next
5225 : * segment on the new timeline.
5226 : */
5227 : int fd;
5228 :
5229 16 : fd = XLogFileInit(startLogSegNo, newTLI);
5230 :
5231 16 : if (close(fd) != 0)
5232 : {
5233 0 : int save_errno = errno;
5234 :
5235 0 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
5236 0 : errno = save_errno;
5237 0 : ereport(ERROR,
5238 : (errcode_for_file_access(),
5239 : errmsg("could not close file \"%s\": %m", xlogfname)));
5240 : }
5241 : }
5242 :
5243 : /*
5244 : * Let's just make real sure there are not .ready or .done flags posted
5245 : * for the new segment.
5246 : */
5247 90 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
5248 90 : XLogArchiveCleanup(xlogfname);
5249 90 : }
5250 :
5251 : /*
5252 : * Perform cleanup actions at the conclusion of archive recovery.
5253 : */
5254 : static void
5255 90 : CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog,
5256 : TimeLineID newTLI)
5257 : {
5258 : /*
5259 : * Execute the recovery_end_command, if any.
5260 : */
5261 90 : if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
5262 4 : ExecuteRecoveryCommand(recoveryEndCommand,
5263 : "recovery_end_command",
5264 : true,
5265 : WAIT_EVENT_RECOVERY_END_COMMAND);
5266 :
5267 : /*
5268 : * We switched to a new timeline. Clean up segments on the old timeline.
5269 : *
5270 : * If there are any higher-numbered segments on the old timeline, remove
5271 : * them. They might contain valid WAL, but they might also be
5272 : * pre-allocated files containing garbage. In any case, they are not part
5273 : * of the new timeline's history so we don't need them.
5274 : */
5275 90 : RemoveNonParentXlogFiles(EndOfLog, newTLI);
5276 :
5277 : /*
5278 : * If the switch happened in the middle of a segment, what to do with the
5279 : * last, partial segment on the old timeline? If we don't archive it, and
5280 : * the server that created the WAL never archives it either (e.g. because
5281 : * it was hit by a meteor), it will never make it to the archive. That's
5282 : * OK from our point of view, because the new segment that we created with
5283 : * the new TLI contains all the WAL from the old timeline up to the switch
5284 : * point. But if you later try to do PITR to the "missing" WAL on the old
5285 : * timeline, recovery won't find it in the archive. It's physically
5286 : * present in the new file with new TLI, but recovery won't look there
5287 : * when it's recovering to the older timeline. On the other hand, if we
5288 : * archive the partial segment, and the original server on that timeline
5289 : * is still running and archives the completed version of the same segment
5290 : * later, it will fail. (We used to do that in 9.4 and below, and it
5291 : * caused such problems).
5292 : *
5293 : * As a compromise, we rename the last segment with the .partial suffix,
5294 : * and archive it. Archive recovery will never try to read .partial
5295 : * segments, so they will normally go unused. But in the odd PITR case,
5296 : * the administrator can copy them manually to the pg_wal directory
5297 : * (removing the suffix). They can be useful in debugging, too.
5298 : *
5299 : * If a .done or .ready file already exists for the old timeline, however,
5300 : * we had already determined that the segment is complete, so we can let
5301 : * it be archived normally. (In particular, if it was restored from the
5302 : * archive to begin with, it's expected to have a .done file).
5303 : */
5304 90 : if (XLogSegmentOffset(EndOfLog, wal_segment_size) != 0 &&
5305 : XLogArchivingActive())
5306 : {
5307 : char origfname[MAXFNAMELEN];
5308 : XLogSegNo endLogSegNo;
5309 :
5310 14 : XLByteToPrevSeg(EndOfLog, endLogSegNo, wal_segment_size);
5311 14 : XLogFileName(origfname, EndOfLogTLI, endLogSegNo, wal_segment_size);
5312 :
5313 14 : if (!XLogArchiveIsReadyOrDone(origfname))
5314 : {
5315 : char origpath[MAXPGPATH];
5316 : char partialfname[MAXFNAMELEN];
5317 : char partialpath[MAXPGPATH];
5318 :
5319 8 : XLogFilePath(origpath, EndOfLogTLI, endLogSegNo, wal_segment_size);
5320 8 : snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
5321 8 : snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
5322 :
5323 : /*
5324 : * Make sure there's no .done or .ready file for the .partial
5325 : * file.
5326 : */
5327 8 : XLogArchiveCleanup(partialfname);
5328 :
5329 8 : durable_rename(origpath, partialpath, ERROR);
5330 8 : XLogArchiveNotify(partialfname);
5331 : }
5332 : }
5333 90 : }
5334 :
5335 : /*
5336 : * Check to see if required parameters are set high enough on this server
5337 : * for various aspects of recovery operation.
5338 : *
5339 : * Note that all the parameters which this function tests need to be
5340 : * listed in Administrator's Overview section in high-availability.sgml.
5341 : * If you change them, don't forget to update the list.
5342 : */
5343 : static void
5344 452 : CheckRequiredParameterValues(void)
5345 : {
5346 : /*
5347 : * For archive recovery, the WAL must be generated with at least 'replica'
5348 : * wal_level.
5349 : */
5350 452 : if (ArchiveRecoveryRequested && ControlFile->wal_level == WAL_LEVEL_MINIMAL)
5351 : {
5352 4 : ereport(FATAL,
5353 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5354 : errmsg("WAL was generated with wal_level=minimal, cannot continue recovering"),
5355 : errdetail("This happens if you temporarily set wal_level=minimal on the server."),
5356 : errhint("Use a backup taken after setting wal_level to higher than minimal.")));
5357 : }
5358 :
5359 : /*
5360 : * For Hot Standby, the WAL must be generated with 'replica' mode, and we
5361 : * must have at least as many backend slots as the primary.
5362 : */
5363 448 : if (ArchiveRecoveryRequested && EnableHotStandby)
5364 : {
5365 : /* We ignore autovacuum_max_workers when we make this test. */
5366 214 : RecoveryRequiresIntParameter("max_connections",
5367 : MaxConnections,
5368 214 : ControlFile->MaxConnections);
5369 214 : RecoveryRequiresIntParameter("max_worker_processes",
5370 : max_worker_processes,
5371 214 : ControlFile->max_worker_processes);
5372 214 : RecoveryRequiresIntParameter("max_wal_senders",
5373 : max_wal_senders,
5374 214 : ControlFile->max_wal_senders);
5375 214 : RecoveryRequiresIntParameter("max_prepared_transactions",
5376 : max_prepared_xacts,
5377 214 : ControlFile->max_prepared_xacts);
5378 214 : RecoveryRequiresIntParameter("max_locks_per_transaction",
5379 : max_locks_per_xact,
5380 214 : ControlFile->max_locks_per_xact);
5381 : }
5382 448 : }
5383 :
5384 : /*
5385 : * This must be called ONCE during postmaster or standalone-backend startup
5386 : */
5387 : void
5388 1520 : StartupXLOG(void)
5389 : {
5390 : XLogCtlInsert *Insert;
5391 : CheckPoint checkPoint;
5392 : bool wasShutdown;
5393 : bool didCrash;
5394 : bool haveTblspcMap;
5395 : bool haveBackupLabel;
5396 : XLogRecPtr EndOfLog;
5397 : TimeLineID EndOfLogTLI;
5398 : TimeLineID newTLI;
5399 : bool performedWalRecovery;
5400 : EndOfWalRecoveryInfo *endOfRecoveryInfo;
5401 : XLogRecPtr abortedRecPtr;
5402 : XLogRecPtr missingContrecPtr;
5403 : TransactionId oldestActiveXID;
5404 1520 : bool promoted = false;
5405 :
5406 : /*
5407 : * We should have an aux process resource owner to use, and we should not
5408 : * be in a transaction that's installed some other resowner.
5409 : */
5410 : Assert(AuxProcessResourceOwner != NULL);
5411 : Assert(CurrentResourceOwner == NULL ||
5412 : CurrentResourceOwner == AuxProcessResourceOwner);
5413 1520 : CurrentResourceOwner = AuxProcessResourceOwner;
5414 :
5415 : /*
5416 : * Check that contents look valid.
5417 : */
5418 1520 : if (!XRecOffIsValid(ControlFile->checkPoint))
5419 0 : ereport(FATAL,
5420 : (errcode(ERRCODE_DATA_CORRUPTED),
5421 : errmsg("control file contains invalid checkpoint location")));
5422 :
5423 1520 : switch (ControlFile->state)
5424 : {
5425 1142 : case DB_SHUTDOWNED:
5426 :
5427 : /*
5428 : * This is the expected case, so don't be chatty in standalone
5429 : * mode
5430 : */
5431 1142 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
5432 : (errmsg("database system was shut down at %s",
5433 : str_time(ControlFile->time))));
5434 1142 : break;
5435 :
5436 52 : case DB_SHUTDOWNED_IN_RECOVERY:
5437 52 : ereport(LOG,
5438 : (errmsg("database system was shut down in recovery at %s",
5439 : str_time(ControlFile->time))));
5440 52 : break;
5441 :
5442 0 : case DB_SHUTDOWNING:
5443 0 : ereport(LOG,
5444 : (errmsg("database system shutdown was interrupted; last known up at %s",
5445 : str_time(ControlFile->time))));
5446 0 : break;
5447 :
5448 0 : case DB_IN_CRASH_RECOVERY:
5449 0 : ereport(LOG,
5450 : (errmsg("database system was interrupted while in recovery at %s",
5451 : str_time(ControlFile->time)),
5452 : errhint("This probably means that some data is corrupted and"
5453 : " you will have to use the last backup for recovery.")));
5454 0 : break;
5455 :
5456 12 : case DB_IN_ARCHIVE_RECOVERY:
5457 12 : ereport(LOG,
5458 : (errmsg("database system was interrupted while in recovery at log time %s",
5459 : str_time(ControlFile->checkPointCopy.time)),
5460 : errhint("If this has occurred more than once some data might be corrupted"
5461 : " and you might need to choose an earlier recovery target.")));
5462 12 : break;
5463 :
5464 314 : case DB_IN_PRODUCTION:
5465 314 : ereport(LOG,
5466 : (errmsg("database system was interrupted; last known up at %s",
5467 : str_time(ControlFile->time))));
5468 314 : break;
5469 :
5470 0 : default:
5471 0 : ereport(FATAL,
5472 : (errcode(ERRCODE_DATA_CORRUPTED),
5473 : errmsg("control file contains invalid database cluster state")));
5474 : }
5475 :
5476 : /* This is just to allow attaching to startup process with a debugger */
5477 : #ifdef XLOG_REPLAY_DELAY
5478 : if (ControlFile->state != DB_SHUTDOWNED)
5479 : pg_usleep(60000000L);
5480 : #endif
5481 :
5482 : /*
5483 : * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
5484 : * In cases where someone has performed a copy for PITR, these directories
5485 : * may have been excluded and need to be re-created.
5486 : */
5487 1520 : ValidateXLOGDirectoryStructure();
5488 :
5489 : /* Set up timeout handler needed to report startup progress. */
5490 1520 : if (!IsBootstrapProcessingMode())
5491 1440 : RegisterTimeout(STARTUP_PROGRESS_TIMEOUT,
5492 : startup_progress_timeout_handler);
5493 :
5494 : /*----------
5495 : * If we previously crashed, perform a couple of actions:
5496 : *
5497 : * - The pg_wal directory may still include some temporary WAL segments
5498 : * used when creating a new segment, so perform some clean up to not
5499 : * bloat this path. This is done first as there is no point to sync
5500 : * this temporary data.
5501 : *
5502 : * - There might be data which we had written, intending to fsync it, but
5503 : * which we had not actually fsync'd yet. Therefore, a power failure in
5504 : * the near future might cause earlier unflushed writes to be lost, even
5505 : * though more recent data written to disk from here on would be
5506 : * persisted. To avoid that, fsync the entire data directory.
5507 : */
5508 1520 : if (ControlFile->state != DB_SHUTDOWNED &&
5509 378 : ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
5510 : {
5511 326 : RemoveTempXlogFiles();
5512 326 : SyncDataDirectory();
5513 326 : didCrash = true;
5514 : }
5515 : else
5516 1194 : didCrash = false;
5517 :
5518 : /*
5519 : * Prepare for WAL recovery if needed.
5520 : *
5521 : * InitWalRecovery analyzes the control file and the backup label file, if
5522 : * any. It updates the in-memory ControlFile buffer according to the
5523 : * starting checkpoint, and sets InRecovery and ArchiveRecoveryRequested.
5524 : * It also applies the tablespace map file, if any.
5525 : */
5526 1520 : InitWalRecovery(ControlFile, &wasShutdown,
5527 : &haveBackupLabel, &haveTblspcMap);
5528 1520 : checkPoint = ControlFile->checkPointCopy;
5529 :
5530 : /* initialize shared memory variables from the checkpoint record */
5531 1520 : TransamVariables->nextXid = checkPoint.nextXid;
5532 1520 : TransamVariables->nextOid = checkPoint.nextOid;
5533 1520 : TransamVariables->oidCount = 0;
5534 1520 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5535 1520 : AdvanceOldestClogXid(checkPoint.oldestXid);
5536 1520 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5537 1520 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5538 1520 : SetCommitTsLimit(checkPoint.oldestCommitTsXid,
5539 : checkPoint.newestCommitTsXid);
5540 1520 : XLogCtl->ckptFullXid = checkPoint.nextXid;
5541 :
5542 : /*
5543 : * Clear out any old relcache cache files. This is *necessary* if we do
5544 : * any WAL replay, since that would probably result in the cache files
5545 : * being out of sync with database reality. In theory we could leave them
5546 : * in place if the database had been cleanly shut down, but it seems
5547 : * safest to just remove them always and let them be rebuilt during the
5548 : * first backend startup. These files needs to be removed from all
5549 : * directories including pg_tblspc, however the symlinks are created only
5550 : * after reading tablespace_map file in case of archive recovery from
5551 : * backup, so needs to clear old relcache files here after creating
5552 : * symlinks.
5553 : */
5554 1520 : RelationCacheInitFileRemove();
5555 :
5556 : /*
5557 : * Initialize replication slots, before there's a chance to remove
5558 : * required resources.
5559 : */
5560 1520 : StartupReplicationSlots();
5561 :
5562 : /*
5563 : * Startup logical state, needs to be setup now so we have proper data
5564 : * during crash recovery.
5565 : */
5566 1520 : StartupReorderBuffer();
5567 :
5568 : /*
5569 : * Startup CLOG. This must be done after TransamVariables->nextXid has
5570 : * been initialized and before we accept connections or begin WAL replay.
5571 : */
5572 1520 : StartupCLOG();
5573 :
5574 : /*
5575 : * Startup MultiXact. We need to do this early to be able to replay
5576 : * truncations.
5577 : */
5578 1520 : StartupMultiXact();
5579 :
5580 : /*
5581 : * Ditto for commit timestamps. Activate the facility if the setting is
5582 : * enabled in the control file, as there should be no tracking of commit
5583 : * timestamps done when the setting was disabled. This facility can be
5584 : * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
5585 : */
5586 1520 : if (ControlFile->track_commit_timestamp)
5587 16 : StartupCommitTs();
5588 :
5589 : /*
5590 : * Recover knowledge about replay progress of known replication partners.
5591 : */
5592 1520 : StartupReplicationOrigin();
5593 :
5594 : /*
5595 : * Initialize unlogged LSN. On a clean shutdown, it's restored from the
5596 : * control file. On recovery, all unlogged relations are blown away, so
5597 : * the unlogged LSN counter can be reset too.
5598 : */
5599 1520 : if (ControlFile->state == DB_SHUTDOWNED)
5600 1128 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5601 1128 : ControlFile->unloggedLSN);
5602 : else
5603 392 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5604 : FirstNormalUnloggedLSN);
5605 :
5606 : /*
5607 : * Copy any missing timeline history files between 'now' and the recovery
5608 : * target timeline from archive to pg_wal. While we don't need those files
5609 : * ourselves - the history file of the recovery target timeline covers all
5610 : * the previous timelines in the history too - a cascading standby server
5611 : * might be interested in them. Or, if you archive the WAL from this
5612 : * server to a different archive than the primary, it'd be good for all
5613 : * the history files to get archived there after failover, so that you can
5614 : * use one of the old timelines as a PITR target. Timeline history files
5615 : * are small, so it's better to copy them unnecessarily than not copy them
5616 : * and regret later.
5617 : */
5618 1520 : restoreTimeLineHistoryFiles(checkPoint.ThisTimeLineID, recoveryTargetTLI);
5619 :
5620 : /*
5621 : * Before running in recovery, scan pg_twophase and fill in its status to
5622 : * be able to work on entries generated by redo. Doing a scan before
5623 : * taking any recovery action has the merit to discard any 2PC files that
5624 : * are newer than the first record to replay, saving from any conflicts at
5625 : * replay. This avoids as well any subsequent scans when doing recovery
5626 : * of the on-disk two-phase data.
5627 : */
5628 1520 : restoreTwoPhaseData();
5629 :
5630 : /*
5631 : * When starting with crash recovery, reset pgstat data - it might not be
5632 : * valid. Otherwise restore pgstat data. It's safe to do this here,
5633 : * because postmaster will not yet have started any other processes.
5634 : *
5635 : * NB: Restoring replication slot stats relies on slot state to have
5636 : * already been restored from disk.
5637 : *
5638 : * TODO: With a bit of extra work we could just start with a pgstat file
5639 : * associated with the checkpoint redo location we're starting from.
5640 : */
5641 1520 : if (didCrash)
5642 326 : pgstat_discard_stats();
5643 : else
5644 1194 : pgstat_restore_stats();
5645 :
5646 1520 : lastFullPageWrites = checkPoint.fullPageWrites;
5647 :
5648 1520 : RedoRecPtr = XLogCtl->RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
5649 1520 : doPageWrites = lastFullPageWrites;
5650 :
5651 : /* REDO */
5652 1520 : if (InRecovery)
5653 : {
5654 : /* Initialize state for RecoveryInProgress() */
5655 392 : SpinLockAcquire(&XLogCtl->info_lck);
5656 392 : if (InArchiveRecovery)
5657 196 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
5658 : else
5659 196 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
5660 392 : SpinLockRelease(&XLogCtl->info_lck);
5661 :
5662 : /*
5663 : * Update pg_control to show that we are recovering and to show the
5664 : * selected checkpoint as the place we are starting from. We also mark
5665 : * pg_control with any minimum recovery stop point obtained from a
5666 : * backup history file.
5667 : *
5668 : * No need to hold ControlFileLock yet, we aren't up far enough.
5669 : */
5670 392 : UpdateControlFile();
5671 :
5672 : /*
5673 : * If there was a backup label file, it's done its job and the info
5674 : * has now been propagated into pg_control. We must get rid of the
5675 : * label file so that if we crash during recovery, we'll pick up at
5676 : * the latest recovery restartpoint instead of going all the way back
5677 : * to the backup start point. It seems prudent though to just rename
5678 : * the file out of the way rather than delete it completely.
5679 : */
5680 392 : if (haveBackupLabel)
5681 : {
5682 126 : unlink(BACKUP_LABEL_OLD);
5683 126 : durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, FATAL);
5684 : }
5685 :
5686 : /*
5687 : * If there was a tablespace_map file, it's done its job and the
5688 : * symlinks have been created. We must get rid of the map file so
5689 : * that if we crash during recovery, we don't create symlinks again.
5690 : * It seems prudent though to just rename the file out of the way
5691 : * rather than delete it completely.
5692 : */
5693 392 : if (haveTblspcMap)
5694 : {
5695 2 : unlink(TABLESPACE_MAP_OLD);
5696 2 : durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, FATAL);
5697 : }
5698 :
5699 : /*
5700 : * Initialize our local copy of minRecoveryPoint. When doing crash
5701 : * recovery we want to replay up to the end of WAL. Particularly, in
5702 : * the case of a promoted standby minRecoveryPoint value in the
5703 : * control file is only updated after the first checkpoint. However,
5704 : * if the instance crashes before the first post-recovery checkpoint
5705 : * is completed then recovery will use a stale location causing the
5706 : * startup process to think that there are still invalid page
5707 : * references when checking for data consistency.
5708 : */
5709 392 : if (InArchiveRecovery)
5710 : {
5711 196 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
5712 196 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
5713 : }
5714 : else
5715 : {
5716 196 : LocalMinRecoveryPoint = InvalidXLogRecPtr;
5717 196 : LocalMinRecoveryPointTLI = 0;
5718 : }
5719 :
5720 : /* Check that the GUCs used to generate the WAL allow recovery */
5721 392 : CheckRequiredParameterValues();
5722 :
5723 : /*
5724 : * We're in recovery, so unlogged relations may be trashed and must be
5725 : * reset. This should be done BEFORE allowing Hot Standby
5726 : * connections, so that read-only backends don't try to read whatever
5727 : * garbage is left over from before.
5728 : */
5729 392 : ResetUnloggedRelations(UNLOGGED_RELATION_CLEANUP);
5730 :
5731 : /*
5732 : * Likewise, delete any saved transaction snapshot files that got left
5733 : * behind by crashed backends.
5734 : */
5735 392 : DeleteAllExportedSnapshotFiles();
5736 :
5737 : /*
5738 : * Initialize for Hot Standby, if enabled. We won't let backends in
5739 : * yet, not until we've reached the min recovery point specified in
5740 : * control file and we've established a recovery snapshot from a
5741 : * running-xacts WAL record.
5742 : */
5743 392 : if (ArchiveRecoveryRequested && EnableHotStandby)
5744 : {
5745 : TransactionId *xids;
5746 : int nxids;
5747 :
5748 186 : ereport(DEBUG1,
5749 : (errmsg_internal("initializing for hot standby")));
5750 :
5751 186 : InitRecoveryTransactionEnvironment();
5752 :
5753 186 : if (wasShutdown)
5754 44 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
5755 : else
5756 142 : oldestActiveXID = checkPoint.oldestActiveXid;
5757 : Assert(TransactionIdIsValid(oldestActiveXID));
5758 :
5759 : /* Tell procarray about the range of xids it has to deal with */
5760 186 : ProcArrayInitRecovery(XidFromFullTransactionId(TransamVariables->nextXid));
5761 :
5762 : /*
5763 : * Startup subtrans only. CLOG, MultiXact and commit timestamp
5764 : * have already been started up and other SLRUs are not maintained
5765 : * during recovery and need not be started yet.
5766 : */
5767 186 : StartupSUBTRANS(oldestActiveXID);
5768 :
5769 : /*
5770 : * If we're beginning at a shutdown checkpoint, we know that
5771 : * nothing was running on the primary at this point. So fake-up an
5772 : * empty running-xacts record and use that here and now. Recover
5773 : * additional standby state for prepared transactions.
5774 : */
5775 186 : if (wasShutdown)
5776 : {
5777 : RunningTransactionsData running;
5778 : TransactionId latestCompletedXid;
5779 :
5780 : /*
5781 : * Construct a RunningTransactions snapshot representing a
5782 : * shut down server, with only prepared transactions still
5783 : * alive. We're never overflowed at this point because all
5784 : * subxids are listed with their parent prepared transactions.
5785 : */
5786 44 : running.xcnt = nxids;
5787 44 : running.subxcnt = 0;
5788 44 : running.subxid_overflow = false;
5789 44 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5790 44 : running.oldestRunningXid = oldestActiveXID;
5791 44 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5792 44 : TransactionIdRetreat(latestCompletedXid);
5793 : Assert(TransactionIdIsNormal(latestCompletedXid));
5794 44 : running.latestCompletedXid = latestCompletedXid;
5795 44 : running.xids = xids;
5796 :
5797 44 : ProcArrayApplyRecoveryInfo(&running);
5798 :
5799 44 : StandbyRecoverPreparedTransactions();
5800 : }
5801 : }
5802 :
5803 : /*
5804 : * We're all set for replaying the WAL now. Do it.
5805 : */
5806 392 : PerformWalRecovery();
5807 292 : performedWalRecovery = true;
5808 : }
5809 : else
5810 1128 : performedWalRecovery = false;
5811 :
5812 : /*
5813 : * Finish WAL recovery.
5814 : */
5815 1420 : endOfRecoveryInfo = FinishWalRecovery();
5816 1420 : EndOfLog = endOfRecoveryInfo->endOfLog;
5817 1420 : EndOfLogTLI = endOfRecoveryInfo->endOfLogTLI;
5818 1420 : abortedRecPtr = endOfRecoveryInfo->abortedRecPtr;
5819 1420 : missingContrecPtr = endOfRecoveryInfo->missingContrecPtr;
5820 :
5821 : /*
5822 : * Reset ps status display, so as no information related to recovery shows
5823 : * up.
5824 : */
5825 1420 : set_ps_display("");
5826 :
5827 : /*
5828 : * When recovering from a backup (we are in recovery, and archive recovery
5829 : * was requested), complain if we did not roll forward far enough to reach
5830 : * the point where the database is consistent. For regular online
5831 : * backup-from-primary, that means reaching the end-of-backup WAL record
5832 : * (at which point we reset backupStartPoint to be Invalid), for
5833 : * backup-from-replica (which can't inject records into the WAL stream),
5834 : * that point is when we reach the minRecoveryPoint in pg_control (which
5835 : * we purposefully copy last when backing up from a replica). For
5836 : * pg_rewind (which creates a backup_label with a method of "pg_rewind")
5837 : * or snapshot-style backups (which don't), backupEndRequired will be set
5838 : * to false.
5839 : *
5840 : * Note: it is indeed okay to look at the local variable
5841 : * LocalMinRecoveryPoint here, even though ControlFile->minRecoveryPoint
5842 : * might be further ahead --- ControlFile->minRecoveryPoint cannot have
5843 : * been advanced beyond the WAL we processed.
5844 : */
5845 1420 : if (InRecovery &&
5846 292 : (EndOfLog < LocalMinRecoveryPoint ||
5847 292 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint)))
5848 : {
5849 : /*
5850 : * Ran off end of WAL before reaching end-of-backup WAL record, or
5851 : * minRecoveryPoint. That's a bad sign, indicating that you tried to
5852 : * recover from an online backup but never called pg_backup_stop(), or
5853 : * you didn't archive all the WAL needed.
5854 : */
5855 0 : if (ArchiveRecoveryRequested || ControlFile->backupEndRequired)
5856 : {
5857 0 : if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint) || ControlFile->backupEndRequired)
5858 0 : ereport(FATAL,
5859 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5860 : errmsg("WAL ends before end of online backup"),
5861 : errhint("All WAL generated while online backup was taken must be available at recovery.")));
5862 : else
5863 0 : ereport(FATAL,
5864 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5865 : errmsg("WAL ends before consistent recovery point")));
5866 : }
5867 : }
5868 :
5869 : /*
5870 : * Reset unlogged relations to the contents of their INIT fork. This is
5871 : * done AFTER recovery is complete so as to include any unlogged relations
5872 : * created during recovery, but BEFORE recovery is marked as having
5873 : * completed successfully. Otherwise we'd not retry if any of the post
5874 : * end-of-recovery steps fail.
5875 : */
5876 1420 : if (InRecovery)
5877 292 : ResetUnloggedRelations(UNLOGGED_RELATION_INIT);
5878 :
5879 : /*
5880 : * Pre-scan prepared transactions to find out the range of XIDs present.
5881 : * This information is not quite needed yet, but it is positioned here so
5882 : * as potential problems are detected before any on-disk change is done.
5883 : */
5884 1420 : oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
5885 :
5886 : /*
5887 : * Allow ordinary WAL segment creation before possibly switching to a new
5888 : * timeline, which creates a new segment, and after the last ReadRecord().
5889 : */
5890 1420 : SetInstallXLogFileSegmentActive();
5891 :
5892 : /*
5893 : * Consider whether we need to assign a new timeline ID.
5894 : *
5895 : * If we did archive recovery, we always assign a new ID. This handles a
5896 : * couple of issues. If we stopped short of the end of WAL during
5897 : * recovery, then we are clearly generating a new timeline and must assign
5898 : * it a unique new ID. Even if we ran to the end, modifying the current
5899 : * last segment is problematic because it may result in trying to
5900 : * overwrite an already-archived copy of that segment, and we encourage
5901 : * DBAs to make their archive_commands reject that. We can dodge the
5902 : * problem by making the new active segment have a new timeline ID.
5903 : *
5904 : * In a normal crash recovery, we can just extend the timeline we were in.
5905 : */
5906 1420 : newTLI = endOfRecoveryInfo->lastRecTLI;
5907 1420 : if (ArchiveRecoveryRequested)
5908 : {
5909 90 : newTLI = findNewestTimeLine(recoveryTargetTLI) + 1;
5910 90 : ereport(LOG,
5911 : (errmsg("selected new timeline ID: %u", newTLI)));
5912 :
5913 : /*
5914 : * Make a writable copy of the last WAL segment. (Note that we also
5915 : * have a copy of the last block of the old WAL in
5916 : * endOfRecovery->lastPage; we will use that below.)
5917 : */
5918 90 : XLogInitNewTimeline(EndOfLogTLI, EndOfLog, newTLI);
5919 :
5920 : /*
5921 : * Remove the signal files out of the way, so that we don't
5922 : * accidentally re-enter archive recovery mode in a subsequent crash.
5923 : */
5924 90 : if (endOfRecoveryInfo->standby_signal_file_found)
5925 84 : durable_unlink(STANDBY_SIGNAL_FILE, FATAL);
5926 :
5927 90 : if (endOfRecoveryInfo->recovery_signal_file_found)
5928 6 : durable_unlink(RECOVERY_SIGNAL_FILE, FATAL);
5929 :
5930 : /*
5931 : * Write the timeline history file, and have it archived. After this
5932 : * point (or rather, as soon as the file is archived), the timeline
5933 : * will appear as "taken" in the WAL archive and to any standby
5934 : * servers. If we crash before actually switching to the new
5935 : * timeline, standby servers will nevertheless think that we switched
5936 : * to the new timeline, and will try to connect to the new timeline.
5937 : * To minimize the window for that, try to do as little as possible
5938 : * between here and writing the end-of-recovery record.
5939 : */
5940 90 : writeTimeLineHistory(newTLI, recoveryTargetTLI,
5941 : EndOfLog, endOfRecoveryInfo->recoveryStopReason);
5942 :
5943 90 : ereport(LOG,
5944 : (errmsg("archive recovery complete")));
5945 : }
5946 :
5947 : /* Save the selected TimeLineID in shared memory, too */
5948 1420 : XLogCtl->InsertTimeLineID = newTLI;
5949 1420 : XLogCtl->PrevTimeLineID = endOfRecoveryInfo->lastRecTLI;
5950 :
5951 : /*
5952 : * Actually, if WAL ended in an incomplete record, skip the parts that
5953 : * made it through and start writing after the portion that persisted.
5954 : * (It's critical to first write an OVERWRITE_CONTRECORD message, which
5955 : * we'll do as soon as we're open for writing new WAL.)
5956 : */
5957 1420 : if (!XLogRecPtrIsInvalid(missingContrecPtr))
5958 : {
5959 : /*
5960 : * We should only have a missingContrecPtr if we're not switching to a
5961 : * new timeline. When a timeline switch occurs, WAL is copied from the
5962 : * old timeline to the new only up to the end of the last complete
5963 : * record, so there can't be an incomplete WAL record that we need to
5964 : * disregard.
5965 : */
5966 : Assert(newTLI == endOfRecoveryInfo->lastRecTLI);
5967 : Assert(!XLogRecPtrIsInvalid(abortedRecPtr));
5968 20 : EndOfLog = missingContrecPtr;
5969 : }
5970 :
5971 : /*
5972 : * Prepare to write WAL starting at EndOfLog location, and init xlog
5973 : * buffer cache using the block containing the last record from the
5974 : * previous incarnation.
5975 : */
5976 1420 : Insert = &XLogCtl->Insert;
5977 1420 : Insert->PrevBytePos = XLogRecPtrToBytePos(endOfRecoveryInfo->lastRec);
5978 1420 : Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
5979 :
5980 : /*
5981 : * Tricky point here: lastPage contains the *last* block that the LastRec
5982 : * record spans, not the one it starts in. The last block is indeed the
5983 : * one we want to use.
5984 : */
5985 1420 : if (EndOfLog % XLOG_BLCKSZ != 0)
5986 : {
5987 : char *page;
5988 : int len;
5989 : int firstIdx;
5990 :
5991 1372 : firstIdx = XLogRecPtrToBufIdx(EndOfLog);
5992 1372 : len = EndOfLog - endOfRecoveryInfo->lastPageBeginPtr;
5993 : Assert(len < XLOG_BLCKSZ);
5994 :
5995 : /* Copy the valid part of the last block, and zero the rest */
5996 1372 : page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
5997 1372 : memcpy(page, endOfRecoveryInfo->lastPage, len);
5998 1372 : memset(page + len, 0, XLOG_BLCKSZ - len);
5999 :
6000 1372 : pg_atomic_write_u64(&XLogCtl->xlblocks[firstIdx], endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ);
6001 1372 : XLogCtl->InitializedUpTo = endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ;
6002 : }
6003 : else
6004 : {
6005 : /*
6006 : * There is no partial block to copy. Just set InitializedUpTo, and
6007 : * let the first attempt to insert a log record to initialize the next
6008 : * buffer.
6009 : */
6010 48 : XLogCtl->InitializedUpTo = EndOfLog;
6011 : }
6012 :
6013 : /*
6014 : * Update local and shared status. This is OK to do without any locks
6015 : * because no other process can be reading or writing WAL yet.
6016 : */
6017 1420 : LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
6018 1420 : pg_atomic_write_u64(&XLogCtl->logInsertResult, EndOfLog);
6019 1420 : pg_atomic_write_u64(&XLogCtl->logWriteResult, EndOfLog);
6020 1420 : pg_atomic_write_u64(&XLogCtl->logFlushResult, EndOfLog);
6021 1420 : XLogCtl->LogwrtRqst.Write = EndOfLog;
6022 1420 : XLogCtl->LogwrtRqst.Flush = EndOfLog;
6023 :
6024 : /*
6025 : * Preallocate additional log files, if wanted.
6026 : */
6027 1420 : PreallocXlogFiles(EndOfLog, newTLI);
6028 :
6029 : /*
6030 : * Okay, we're officially UP.
6031 : */
6032 1420 : InRecovery = false;
6033 :
6034 : /* start the archive_timeout timer and LSN running */
6035 1420 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
6036 1420 : XLogCtl->lastSegSwitchLSN = EndOfLog;
6037 :
6038 : /* also initialize latestCompletedXid, to nextXid - 1 */
6039 1420 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
6040 1420 : TransamVariables->latestCompletedXid = TransamVariables->nextXid;
6041 1420 : FullTransactionIdRetreat(&TransamVariables->latestCompletedXid);
6042 1420 : LWLockRelease(ProcArrayLock);
6043 :
6044 : /*
6045 : * Start up subtrans, if not already done for hot standby. (commit
6046 : * timestamps are started below, if necessary.)
6047 : */
6048 1420 : if (standbyState == STANDBY_DISABLED)
6049 1330 : StartupSUBTRANS(oldestActiveXID);
6050 :
6051 : /*
6052 : * Perform end of recovery actions for any SLRUs that need it.
6053 : */
6054 1420 : TrimCLOG();
6055 1420 : TrimMultiXact();
6056 :
6057 : /*
6058 : * Reload shared-memory state for prepared transactions. This needs to
6059 : * happen before renaming the last partial segment of the old timeline as
6060 : * it may be possible that we have to recovery some transactions from it.
6061 : */
6062 1420 : RecoverPreparedTransactions();
6063 :
6064 : /* Shut down xlogreader */
6065 1420 : ShutdownWalRecovery();
6066 :
6067 : /* Enable WAL writes for this backend only. */
6068 1420 : LocalSetXLogInsertAllowed();
6069 :
6070 : /* If necessary, write overwrite-contrecord before doing anything else */
6071 1420 : if (!XLogRecPtrIsInvalid(abortedRecPtr))
6072 : {
6073 : Assert(!XLogRecPtrIsInvalid(missingContrecPtr));
6074 20 : CreateOverwriteContrecordRecord(abortedRecPtr, missingContrecPtr, newTLI);
6075 : }
6076 :
6077 : /*
6078 : * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
6079 : * record before resource manager writes cleanup WAL records or checkpoint
6080 : * record is written.
6081 : */
6082 1420 : Insert->fullPageWrites = lastFullPageWrites;
6083 1420 : UpdateFullPageWrites();
6084 :
6085 : /*
6086 : * Emit checkpoint or end-of-recovery record in XLOG, if required.
6087 : */
6088 1420 : if (performedWalRecovery)
6089 292 : promoted = PerformRecoveryXLogAction();
6090 :
6091 : /*
6092 : * If any of the critical GUCs have changed, log them before we allow
6093 : * backends to write WAL.
6094 : */
6095 1420 : XLogReportParameters();
6096 :
6097 : /* If this is archive recovery, perform post-recovery cleanup actions. */
6098 1420 : if (ArchiveRecoveryRequested)
6099 90 : CleanupAfterArchiveRecovery(EndOfLogTLI, EndOfLog, newTLI);
6100 :
6101 : /*
6102 : * Local WAL inserts enabled, so it's time to finish initialization of
6103 : * commit timestamp.
6104 : */
6105 1420 : CompleteCommitTsInitialization();
6106 :
6107 : /*
6108 : * All done with end-of-recovery actions.
6109 : *
6110 : * Now allow backends to write WAL and update the control file status in
6111 : * consequence. SharedRecoveryState, that controls if backends can write
6112 : * WAL, is updated while holding ControlFileLock to prevent other backends
6113 : * to look at an inconsistent state of the control file in shared memory.
6114 : * There is still a small window during which backends can write WAL and
6115 : * the control file is still referring to a system not in DB_IN_PRODUCTION
6116 : * state while looking at the on-disk control file.
6117 : *
6118 : * Also, we use info_lck to update SharedRecoveryState to ensure that
6119 : * there are no race conditions concerning visibility of other recent
6120 : * updates to shared memory.
6121 : */
6122 1420 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6123 1420 : ControlFile->state = DB_IN_PRODUCTION;
6124 :
6125 1420 : SpinLockAcquire(&XLogCtl->info_lck);
6126 1420 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_DONE;
6127 1420 : SpinLockRelease(&XLogCtl->info_lck);
6128 :
6129 1420 : UpdateControlFile();
6130 1420 : LWLockRelease(ControlFileLock);
6131 :
6132 : /*
6133 : * Shutdown the recovery environment. This must occur after
6134 : * RecoverPreparedTransactions() (see notes in lock_twophase_recover())
6135 : * and after switching SharedRecoveryState to RECOVERY_STATE_DONE so as
6136 : * any session building a snapshot will not rely on KnownAssignedXids as
6137 : * RecoveryInProgress() would return false at this stage. This is
6138 : * particularly critical for prepared 2PC transactions, that would still
6139 : * need to be included in snapshots once recovery has ended.
6140 : */
6141 1420 : if (standbyState != STANDBY_DISABLED)
6142 90 : ShutdownRecoveryTransactionEnvironment();
6143 :
6144 : /*
6145 : * If there were cascading standby servers connected to us, nudge any wal
6146 : * sender processes to notice that we've been promoted.
6147 : */
6148 1420 : WalSndWakeup(true, true);
6149 :
6150 : /*
6151 : * If this was a promotion, request an (online) checkpoint now. This isn't
6152 : * required for consistency, but the last restartpoint might be far back,
6153 : * and in case of a crash, recovering from it might take a longer than is
6154 : * appropriate now that we're not in standby mode anymore.
6155 : */
6156 1420 : if (promoted)
6157 76 : RequestCheckpoint(CHECKPOINT_FORCE);
6158 1420 : }
6159 :
6160 : /*
6161 : * Callback from PerformWalRecovery(), called when we switch from crash
6162 : * recovery to archive recovery mode. Updates the control file accordingly.
6163 : */
6164 : void
6165 4 : SwitchIntoArchiveRecovery(XLogRecPtr EndRecPtr, TimeLineID replayTLI)
6166 : {
6167 : /* initialize minRecoveryPoint to this record */
6168 4 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6169 4 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
6170 4 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6171 : {
6172 4 : ControlFile->minRecoveryPoint = EndRecPtr;
6173 4 : ControlFile->minRecoveryPointTLI = replayTLI;
6174 : }
6175 : /* update local copy */
6176 4 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
6177 4 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
6178 :
6179 : /*
6180 : * The startup process can update its local copy of minRecoveryPoint from
6181 : * this point.
6182 : */
6183 4 : updateMinRecoveryPoint = true;
6184 :
6185 4 : UpdateControlFile();
6186 :
6187 : /*
6188 : * We update SharedRecoveryState while holding the lock on ControlFileLock
6189 : * so both states are consistent in shared memory.
6190 : */
6191 4 : SpinLockAcquire(&XLogCtl->info_lck);
6192 4 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
6193 4 : SpinLockRelease(&XLogCtl->info_lck);
6194 :
6195 4 : LWLockRelease(ControlFileLock);
6196 4 : }
6197 :
6198 : /*
6199 : * Callback from PerformWalRecovery(), called when we reach the end of backup.
6200 : * Updates the control file accordingly.
6201 : */
6202 : void
6203 126 : ReachedEndOfBackup(XLogRecPtr EndRecPtr, TimeLineID tli)
6204 : {
6205 : /*
6206 : * We have reached the end of base backup, as indicated by pg_control. The
6207 : * data on disk is now consistent (unless minRecoveryPoint is further
6208 : * ahead, which can happen if we crashed during previous recovery). Reset
6209 : * backupStartPoint and backupEndPoint, and update minRecoveryPoint to
6210 : * make sure we don't allow starting up at an earlier point even if
6211 : * recovery is stopped and restarted soon after this.
6212 : */
6213 126 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6214 :
6215 126 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6216 : {
6217 118 : ControlFile->minRecoveryPoint = EndRecPtr;
6218 118 : ControlFile->minRecoveryPointTLI = tli;
6219 : }
6220 :
6221 126 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
6222 126 : ControlFile->backupEndPoint = InvalidXLogRecPtr;
6223 126 : ControlFile->backupEndRequired = false;
6224 126 : UpdateControlFile();
6225 :
6226 126 : LWLockRelease(ControlFileLock);
6227 126 : }
6228 :
6229 : /*
6230 : * Perform whatever XLOG actions are necessary at end of REDO.
6231 : *
6232 : * The goal here is to make sure that we'll be able to recover properly if
6233 : * we crash again. If we choose to write a checkpoint, we'll write a shutdown
6234 : * checkpoint rather than an on-line one. This is not particularly critical,
6235 : * but since we may be assigning a new TLI, using a shutdown checkpoint allows
6236 : * us to have the rule that TLI only changes in shutdown checkpoints, which
6237 : * allows some extra error checking in xlog_redo.
6238 : */
6239 : static bool
6240 292 : PerformRecoveryXLogAction(void)
6241 : {
6242 292 : bool promoted = false;
6243 :
6244 : /*
6245 : * Perform a checkpoint to update all our recovery activity to disk.
6246 : *
6247 : * Note that we write a shutdown checkpoint rather than an on-line one.
6248 : * This is not particularly critical, but since we may be assigning a new
6249 : * TLI, using a shutdown checkpoint allows us to have the rule that TLI
6250 : * only changes in shutdown checkpoints, which allows some extra error
6251 : * checking in xlog_redo.
6252 : *
6253 : * In promotion, only create a lightweight end-of-recovery record instead
6254 : * of a full checkpoint. A checkpoint is requested later, after we're
6255 : * fully out of recovery mode and already accepting queries.
6256 : */
6257 382 : if (ArchiveRecoveryRequested && IsUnderPostmaster &&
6258 90 : PromoteIsTriggered())
6259 : {
6260 76 : promoted = true;
6261 :
6262 : /*
6263 : * Insert a special WAL record to mark the end of recovery, since we
6264 : * aren't doing a checkpoint. That means that the checkpointer process
6265 : * may likely be in the middle of a time-smoothed restartpoint and
6266 : * could continue to be for minutes after this. That sounds strange,
6267 : * but the effect is roughly the same and it would be stranger to try
6268 : * to come out of the restartpoint and then checkpoint. We request a
6269 : * checkpoint later anyway, just for safety.
6270 : */
6271 76 : CreateEndOfRecoveryRecord();
6272 : }
6273 : else
6274 : {
6275 216 : RequestCheckpoint(CHECKPOINT_END_OF_RECOVERY |
6276 : CHECKPOINT_IMMEDIATE |
6277 : CHECKPOINT_WAIT);
6278 : }
6279 :
6280 292 : return promoted;
6281 : }
6282 :
6283 : /*
6284 : * Is the system still in recovery?
6285 : *
6286 : * Unlike testing InRecovery, this works in any process that's connected to
6287 : * shared memory.
6288 : */
6289 : bool
6290 106193714 : RecoveryInProgress(void)
6291 : {
6292 : /*
6293 : * We check shared state each time only until we leave recovery mode. We
6294 : * can't re-enter recovery, so there's no need to keep checking after the
6295 : * shared variable has once been seen false.
6296 : */
6297 106193714 : if (!LocalRecoveryInProgress)
6298 101513250 : return false;
6299 : else
6300 : {
6301 : /*
6302 : * use volatile pointer to make sure we make a fresh read of the
6303 : * shared variable.
6304 : */
6305 4680464 : volatile XLogCtlData *xlogctl = XLogCtl;
6306 :
6307 4680464 : LocalRecoveryInProgress = (xlogctl->SharedRecoveryState != RECOVERY_STATE_DONE);
6308 :
6309 : /*
6310 : * Note: We don't need a memory barrier when we're still in recovery.
6311 : * We might exit recovery immediately after return, so the caller
6312 : * can't rely on 'true' meaning that we're still in recovery anyway.
6313 : */
6314 :
6315 4680464 : return LocalRecoveryInProgress;
6316 : }
6317 : }
6318 :
6319 : /*
6320 : * Returns current recovery state from shared memory.
6321 : *
6322 : * This returned state is kept consistent with the contents of the control
6323 : * file. See details about the possible values of RecoveryState in xlog.h.
6324 : */
6325 : RecoveryState
6326 102 : GetRecoveryState(void)
6327 : {
6328 : RecoveryState retval;
6329 :
6330 102 : SpinLockAcquire(&XLogCtl->info_lck);
6331 102 : retval = XLogCtl->SharedRecoveryState;
6332 102 : SpinLockRelease(&XLogCtl->info_lck);
6333 :
6334 102 : return retval;
6335 : }
6336 :
6337 : /*
6338 : * Is this process allowed to insert new WAL records?
6339 : *
6340 : * Ordinarily this is essentially equivalent to !RecoveryInProgress().
6341 : * But we also have provisions for forcing the result "true" or "false"
6342 : * within specific processes regardless of the global state.
6343 : */
6344 : bool
6345 53181802 : XLogInsertAllowed(void)
6346 : {
6347 : /*
6348 : * If value is "unconditionally true" or "unconditionally false", just
6349 : * return it. This provides the normal fast path once recovery is known
6350 : * done.
6351 : */
6352 53181802 : if (LocalXLogInsertAllowed >= 0)
6353 52990486 : return (bool) LocalXLogInsertAllowed;
6354 :
6355 : /*
6356 : * Else, must check to see if we're still in recovery.
6357 : */
6358 191316 : if (RecoveryInProgress())
6359 178470 : return false;
6360 :
6361 : /*
6362 : * On exit from recovery, reset to "unconditionally true", since there is
6363 : * no need to keep checking.
6364 : */
6365 12846 : LocalXLogInsertAllowed = 1;
6366 12846 : return true;
6367 : }
6368 :
6369 : /*
6370 : * Make XLogInsertAllowed() return true in the current process only.
6371 : *
6372 : * Note: it is allowed to switch LocalXLogInsertAllowed back to -1 later,
6373 : * and even call LocalSetXLogInsertAllowed() again after that.
6374 : *
6375 : * Returns the previous value of LocalXLogInsertAllowed.
6376 : */
6377 : static int
6378 1476 : LocalSetXLogInsertAllowed(void)
6379 : {
6380 1476 : int oldXLogAllowed = LocalXLogInsertAllowed;
6381 :
6382 1476 : LocalXLogInsertAllowed = 1;
6383 :
6384 1476 : return oldXLogAllowed;
6385 : }
6386 :
6387 : /*
6388 : * Return the current Redo pointer from shared memory.
6389 : *
6390 : * As a side-effect, the local RedoRecPtr copy is updated.
6391 : */
6392 : XLogRecPtr
6393 365442 : GetRedoRecPtr(void)
6394 : {
6395 : XLogRecPtr ptr;
6396 :
6397 : /*
6398 : * The possibly not up-to-date copy in XlogCtl is enough. Even if we
6399 : * grabbed a WAL insertion lock to read the authoritative value in
6400 : * Insert->RedoRecPtr, someone might update it just after we've released
6401 : * the lock.
6402 : */
6403 365442 : SpinLockAcquire(&XLogCtl->info_lck);
6404 365442 : ptr = XLogCtl->RedoRecPtr;
6405 365442 : SpinLockRelease(&XLogCtl->info_lck);
6406 :
6407 365442 : if (RedoRecPtr < ptr)
6408 1570 : RedoRecPtr = ptr;
6409 :
6410 365442 : return RedoRecPtr;
6411 : }
6412 :
6413 : /*
6414 : * Return information needed to decide whether a modified block needs a
6415 : * full-page image to be included in the WAL record.
6416 : *
6417 : * The returned values are cached copies from backend-private memory, and
6418 : * possibly out-of-date or, indeed, uninitialized, in which case they will
6419 : * be InvalidXLogRecPtr and false, respectively. XLogInsertRecord will
6420 : * re-check them against up-to-date values, while holding the WAL insert lock.
6421 : */
6422 : void
6423 25804650 : GetFullPageWriteInfo(XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
6424 : {
6425 25804650 : *RedoRecPtr_p = RedoRecPtr;
6426 25804650 : *doPageWrites_p = doPageWrites;
6427 25804650 : }
6428 :
6429 : /*
6430 : * GetInsertRecPtr -- Returns the current insert position.
6431 : *
6432 : * NOTE: The value *actually* returned is the position of the last full
6433 : * xlog page. It lags behind the real insert position by at most 1 page.
6434 : * For that, we don't need to scan through WAL insertion locks, and an
6435 : * approximation is enough for the current usage of this function.
6436 : */
6437 : XLogRecPtr
6438 3898 : GetInsertRecPtr(void)
6439 : {
6440 : XLogRecPtr recptr;
6441 :
6442 3898 : SpinLockAcquire(&XLogCtl->info_lck);
6443 3898 : recptr = XLogCtl->LogwrtRqst.Write;
6444 3898 : SpinLockRelease(&XLogCtl->info_lck);
6445 :
6446 3898 : return recptr;
6447 : }
6448 :
6449 : /*
6450 : * GetFlushRecPtr -- Returns the current flush position, ie, the last WAL
6451 : * position known to be fsync'd to disk. This should only be used on a
6452 : * system that is known not to be in recovery.
6453 : */
6454 : XLogRecPtr
6455 299414 : GetFlushRecPtr(TimeLineID *insertTLI)
6456 : {
6457 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6458 :
6459 299414 : RefreshXLogWriteResult(LogwrtResult);
6460 :
6461 : /*
6462 : * If we're writing and flushing WAL, the time line can't be changing, so
6463 : * no lock is required.
6464 : */
6465 299414 : if (insertTLI)
6466 47642 : *insertTLI = XLogCtl->InsertTimeLineID;
6467 :
6468 299414 : return LogwrtResult.Flush;
6469 : }
6470 :
6471 : /*
6472 : * GetWALInsertionTimeLine -- Returns the current timeline of a system that
6473 : * is not in recovery.
6474 : */
6475 : TimeLineID
6476 105402 : GetWALInsertionTimeLine(void)
6477 : {
6478 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6479 :
6480 : /* Since the value can't be changing, no lock is required. */
6481 105402 : return XLogCtl->InsertTimeLineID;
6482 : }
6483 :
6484 : /*
6485 : * GetLastImportantRecPtr -- Returns the LSN of the last important record
6486 : * inserted. All records not explicitly marked as unimportant are considered
6487 : * important.
6488 : *
6489 : * The LSN is determined by computing the maximum of
6490 : * WALInsertLocks[i].lastImportantAt.
6491 : */
6492 : XLogRecPtr
6493 1670 : GetLastImportantRecPtr(void)
6494 : {
6495 1670 : XLogRecPtr res = InvalidXLogRecPtr;
6496 : int i;
6497 :
6498 15030 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
6499 : {
6500 : XLogRecPtr last_important;
6501 :
6502 : /*
6503 : * Need to take a lock to prevent torn reads of the LSN, which are
6504 : * possible on some of the supported platforms. WAL insert locks only
6505 : * support exclusive mode, so we have to use that.
6506 : */
6507 13360 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
6508 13360 : last_important = WALInsertLocks[i].l.lastImportantAt;
6509 13360 : LWLockRelease(&WALInsertLocks[i].l.lock);
6510 :
6511 13360 : if (res < last_important)
6512 2668 : res = last_important;
6513 : }
6514 :
6515 1670 : return res;
6516 : }
6517 :
6518 : /*
6519 : * Get the time and LSN of the last xlog segment switch
6520 : */
6521 : pg_time_t
6522 0 : GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
6523 : {
6524 : pg_time_t result;
6525 :
6526 : /* Need WALWriteLock, but shared lock is sufficient */
6527 0 : LWLockAcquire(WALWriteLock, LW_SHARED);
6528 0 : result = XLogCtl->lastSegSwitchTime;
6529 0 : *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
6530 0 : LWLockRelease(WALWriteLock);
6531 :
6532 0 : return result;
6533 : }
6534 :
6535 : /*
6536 : * This must be called ONCE during postmaster or standalone-backend shutdown
6537 : */
6538 : void
6539 1000 : ShutdownXLOG(int code, Datum arg)
6540 : {
6541 : /*
6542 : * We should have an aux process resource owner to use, and we should not
6543 : * be in a transaction that's installed some other resowner.
6544 : */
6545 : Assert(AuxProcessResourceOwner != NULL);
6546 : Assert(CurrentResourceOwner == NULL ||
6547 : CurrentResourceOwner == AuxProcessResourceOwner);
6548 1000 : CurrentResourceOwner = AuxProcessResourceOwner;
6549 :
6550 : /* Don't be chatty in standalone mode */
6551 1000 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
6552 : (errmsg("shutting down")));
6553 :
6554 : /*
6555 : * Signal walsenders to move to stopping state.
6556 : */
6557 1000 : WalSndInitStopping();
6558 :
6559 : /*
6560 : * Wait for WAL senders to be in stopping state. This prevents commands
6561 : * from writing new WAL.
6562 : */
6563 1000 : WalSndWaitStopping();
6564 :
6565 1000 : if (RecoveryInProgress())
6566 94 : CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
6567 : else
6568 : {
6569 : /*
6570 : * If archiving is enabled, rotate the last XLOG file so that all the
6571 : * remaining records are archived (postmaster wakes up the archiver
6572 : * process one more time at the end of shutdown). The checkpoint
6573 : * record will go to the next XLOG file and won't be archived (yet).
6574 : */
6575 906 : if (XLogArchivingActive())
6576 22 : RequestXLogSwitch(false);
6577 :
6578 906 : CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
6579 : }
6580 1000 : }
6581 :
6582 : /*
6583 : * Log start of a checkpoint.
6584 : */
6585 : static void
6586 1240 : LogCheckpointStart(int flags, bool restartpoint)
6587 : {
6588 1240 : if (restartpoint)
6589 76 : ereport(LOG,
6590 : /* translator: the placeholders show checkpoint options */
6591 : (errmsg("restartpoint starting:%s%s%s%s%s%s%s%s",
6592 : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6593 : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6594 : (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
6595 : (flags & CHECKPOINT_FORCE) ? " force" : "",
6596 : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6597 : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6598 : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6599 : (flags & CHECKPOINT_FLUSH_ALL) ? " flush-all" : "")));
6600 : else
6601 1164 : ereport(LOG,
6602 : /* translator: the placeholders show checkpoint options */
6603 : (errmsg("checkpoint starting:%s%s%s%s%s%s%s%s",
6604 : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6605 : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6606 : (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
6607 : (flags & CHECKPOINT_FORCE) ? " force" : "",
6608 : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6609 : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6610 : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6611 : (flags & CHECKPOINT_FLUSH_ALL) ? " flush-all" : "")));
6612 1240 : }
6613 :
6614 : /*
6615 : * Log end of a checkpoint.
6616 : */
6617 : static void
6618 1704 : LogCheckpointEnd(bool restartpoint)
6619 : {
6620 : long write_msecs,
6621 : sync_msecs,
6622 : total_msecs,
6623 : longest_msecs,
6624 : average_msecs;
6625 : uint64 average_sync_time;
6626 :
6627 1704 : CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
6628 :
6629 1704 : write_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_write_t,
6630 : CheckpointStats.ckpt_sync_t);
6631 :
6632 1704 : sync_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_sync_t,
6633 : CheckpointStats.ckpt_sync_end_t);
6634 :
6635 : /* Accumulate checkpoint timing summary data, in milliseconds. */
6636 1704 : PendingCheckpointerStats.write_time += write_msecs;
6637 1704 : PendingCheckpointerStats.sync_time += sync_msecs;
6638 :
6639 : /*
6640 : * All of the published timing statistics are accounted for. Only
6641 : * continue if a log message is to be written.
6642 : */
6643 1704 : if (!log_checkpoints)
6644 464 : return;
6645 :
6646 1240 : total_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_start_t,
6647 : CheckpointStats.ckpt_end_t);
6648 :
6649 : /*
6650 : * Timing values returned from CheckpointStats are in microseconds.
6651 : * Convert to milliseconds for consistent printing.
6652 : */
6653 1240 : longest_msecs = (long) ((CheckpointStats.ckpt_longest_sync + 999) / 1000);
6654 :
6655 1240 : average_sync_time = 0;
6656 1240 : if (CheckpointStats.ckpt_sync_rels > 0)
6657 0 : average_sync_time = CheckpointStats.ckpt_agg_sync_time /
6658 0 : CheckpointStats.ckpt_sync_rels;
6659 1240 : average_msecs = (long) ((average_sync_time + 999) / 1000);
6660 :
6661 : /*
6662 : * ControlFileLock is not required to see ControlFile->checkPoint and
6663 : * ->checkPointCopy here as we are the only updator of those variables at
6664 : * this moment.
6665 : */
6666 1240 : if (restartpoint)
6667 76 : ereport(LOG,
6668 : (errmsg("restartpoint complete: wrote %d buffers (%.1f%%); "
6669 : "%d WAL file(s) added, %d removed, %d recycled; "
6670 : "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s; "
6671 : "sync files=%d, longest=%ld.%03d s, average=%ld.%03d s; "
6672 : "distance=%d kB, estimate=%d kB; "
6673 : "lsn=%X/%X, redo lsn=%X/%X",
6674 : CheckpointStats.ckpt_bufs_written,
6675 : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6676 : CheckpointStats.ckpt_segs_added,
6677 : CheckpointStats.ckpt_segs_removed,
6678 : CheckpointStats.ckpt_segs_recycled,
6679 : write_msecs / 1000, (int) (write_msecs % 1000),
6680 : sync_msecs / 1000, (int) (sync_msecs % 1000),
6681 : total_msecs / 1000, (int) (total_msecs % 1000),
6682 : CheckpointStats.ckpt_sync_rels,
6683 : longest_msecs / 1000, (int) (longest_msecs % 1000),
6684 : average_msecs / 1000, (int) (average_msecs % 1000),
6685 : (int) (PrevCheckPointDistance / 1024.0),
6686 : (int) (CheckPointDistanceEstimate / 1024.0),
6687 : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6688 : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6689 : else
6690 1164 : ereport(LOG,
6691 : (errmsg("checkpoint complete: wrote %d buffers (%.1f%%); "
6692 : "%d WAL file(s) added, %d removed, %d recycled; "
6693 : "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s; "
6694 : "sync files=%d, longest=%ld.%03d s, average=%ld.%03d s; "
6695 : "distance=%d kB, estimate=%d kB; "
6696 : "lsn=%X/%X, redo lsn=%X/%X",
6697 : CheckpointStats.ckpt_bufs_written,
6698 : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6699 : CheckpointStats.ckpt_segs_added,
6700 : CheckpointStats.ckpt_segs_removed,
6701 : CheckpointStats.ckpt_segs_recycled,
6702 : write_msecs / 1000, (int) (write_msecs % 1000),
6703 : sync_msecs / 1000, (int) (sync_msecs % 1000),
6704 : total_msecs / 1000, (int) (total_msecs % 1000),
6705 : CheckpointStats.ckpt_sync_rels,
6706 : longest_msecs / 1000, (int) (longest_msecs % 1000),
6707 : average_msecs / 1000, (int) (average_msecs % 1000),
6708 : (int) (PrevCheckPointDistance / 1024.0),
6709 : (int) (CheckPointDistanceEstimate / 1024.0),
6710 : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6711 : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6712 : }
6713 :
6714 : /*
6715 : * Update the estimate of distance between checkpoints.
6716 : *
6717 : * The estimate is used to calculate the number of WAL segments to keep
6718 : * preallocated, see XLOGfileslop().
6719 : */
6720 : static void
6721 1704 : UpdateCheckPointDistanceEstimate(uint64 nbytes)
6722 : {
6723 : /*
6724 : * To estimate the number of segments consumed between checkpoints, keep a
6725 : * moving average of the amount of WAL generated in previous checkpoint
6726 : * cycles. However, if the load is bursty, with quiet periods and busy
6727 : * periods, we want to cater for the peak load. So instead of a plain
6728 : * moving average, let the average decline slowly if the previous cycle
6729 : * used less WAL than estimated, but bump it up immediately if it used
6730 : * more.
6731 : *
6732 : * When checkpoints are triggered by max_wal_size, this should converge to
6733 : * CheckpointSegments * wal_segment_size,
6734 : *
6735 : * Note: This doesn't pay any attention to what caused the checkpoint.
6736 : * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
6737 : * starting a base backup, are counted the same as those created
6738 : * automatically. The slow-decline will largely mask them out, if they are
6739 : * not frequent. If they are frequent, it seems reasonable to count them
6740 : * in as any others; if you issue a manual checkpoint every 5 minutes and
6741 : * never let a timed checkpoint happen, it makes sense to base the
6742 : * preallocation on that 5 minute interval rather than whatever
6743 : * checkpoint_timeout is set to.
6744 : */
6745 1704 : PrevCheckPointDistance = nbytes;
6746 1704 : if (CheckPointDistanceEstimate < nbytes)
6747 1084 : CheckPointDistanceEstimate = nbytes;
6748 : else
6749 620 : CheckPointDistanceEstimate =
6750 620 : (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
6751 1704 : }
6752 :
6753 : /*
6754 : * Update the ps display for a process running a checkpoint. Note that
6755 : * this routine should not do any allocations so as it can be called
6756 : * from a critical section.
6757 : */
6758 : static void
6759 3408 : update_checkpoint_display(int flags, bool restartpoint, bool reset)
6760 : {
6761 : /*
6762 : * The status is reported only for end-of-recovery and shutdown
6763 : * checkpoints or shutdown restartpoints. Updating the ps display is
6764 : * useful in those situations as it may not be possible to rely on
6765 : * pg_stat_activity to see the status of the checkpointer or the startup
6766 : * process.
6767 : */
6768 3408 : if ((flags & (CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IS_SHUTDOWN)) == 0)
6769 1412 : return;
6770 :
6771 1996 : if (reset)
6772 998 : set_ps_display("");
6773 : else
6774 : {
6775 : char activitymsg[128];
6776 :
6777 2994 : snprintf(activitymsg, sizeof(activitymsg), "performing %s%s%s",
6778 998 : (flags & CHECKPOINT_END_OF_RECOVERY) ? "end-of-recovery " : "",
6779 998 : (flags & CHECKPOINT_IS_SHUTDOWN) ? "shutdown " : "",
6780 : restartpoint ? "restartpoint" : "checkpoint");
6781 998 : set_ps_display(activitymsg);
6782 : }
6783 : }
6784 :
6785 :
6786 : /*
6787 : * Perform a checkpoint --- either during shutdown, or on-the-fly
6788 : *
6789 : * flags is a bitwise OR of the following:
6790 : * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
6791 : * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
6792 : * CHECKPOINT_IMMEDIATE: finish the checkpoint ASAP,
6793 : * ignoring checkpoint_completion_target parameter.
6794 : * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
6795 : * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
6796 : * CHECKPOINT_END_OF_RECOVERY).
6797 : * CHECKPOINT_FLUSH_ALL: also flush buffers of unlogged tables.
6798 : *
6799 : * Note: flags contains other bits, of interest here only for logging purposes.
6800 : * In particular note that this routine is synchronous and does not pay
6801 : * attention to CHECKPOINT_WAIT.
6802 : *
6803 : * If !shutdown then we are writing an online checkpoint. An XLOG_CHECKPOINT_REDO
6804 : * record is inserted into WAL at the logical location of the checkpoint, before
6805 : * flushing anything to disk, and when the checkpoint is eventually completed,
6806 : * and it is from this point that WAL replay will begin in the case of a recovery
6807 : * from this checkpoint. Once everything is written to disk, an
6808 : * XLOG_CHECKPOINT_ONLINE record is written to complete the checkpoint, and
6809 : * points back to the earlier XLOG_CHECKPOINT_REDO record. This mechanism allows
6810 : * other write-ahead log records to be written while the checkpoint is in
6811 : * progress, but we must be very careful about order of operations. This function
6812 : * may take many minutes to execute on a busy system.
6813 : *
6814 : * On the other hand, when shutdown is true, concurrent insertion into the
6815 : * write-ahead log is impossible, so there is no need for two separate records.
6816 : * In this case, we only insert an XLOG_CHECKPOINT_SHUTDOWN record, and it's
6817 : * both the record marking the completion of the checkpoint and the location
6818 : * from which WAL replay would begin if needed.
6819 : */
6820 : void
6821 1628 : CreateCheckPoint(int flags)
6822 : {
6823 : bool shutdown;
6824 : CheckPoint checkPoint;
6825 : XLogRecPtr recptr;
6826 : XLogSegNo _logSegNo;
6827 1628 : XLogCtlInsert *Insert = &XLogCtl->Insert;
6828 : uint32 freespace;
6829 : XLogRecPtr PriorRedoPtr;
6830 : XLogRecPtr last_important_lsn;
6831 : VirtualTransactionId *vxids;
6832 : int nvxids;
6833 1628 : int oldXLogAllowed = 0;
6834 :
6835 : /*
6836 : * An end-of-recovery checkpoint is really a shutdown checkpoint, just
6837 : * issued at a different time.
6838 : */
6839 1628 : if (flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY))
6840 962 : shutdown = true;
6841 : else
6842 666 : shutdown = false;
6843 :
6844 : /* sanity check */
6845 1628 : if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
6846 0 : elog(ERROR, "can't create a checkpoint during recovery");
6847 :
6848 : /*
6849 : * Prepare to accumulate statistics.
6850 : *
6851 : * Note: because it is possible for log_checkpoints to change while a
6852 : * checkpoint proceeds, we always accumulate stats, even if
6853 : * log_checkpoints is currently off.
6854 : */
6855 17908 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
6856 1628 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
6857 :
6858 : /*
6859 : * Let smgr prepare for checkpoint; this has to happen outside the
6860 : * critical section and before we determine the REDO pointer. Note that
6861 : * smgr must not do anything that'd have to be undone if we decide no
6862 : * checkpoint is needed.
6863 : */
6864 1628 : SyncPreCheckpoint();
6865 :
6866 : /*
6867 : * Use a critical section to force system panic if we have trouble.
6868 : */
6869 1628 : START_CRIT_SECTION();
6870 :
6871 1628 : if (shutdown)
6872 : {
6873 962 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6874 962 : ControlFile->state = DB_SHUTDOWNING;
6875 962 : UpdateControlFile();
6876 962 : LWLockRelease(ControlFileLock);
6877 : }
6878 :
6879 : /* Begin filling in the checkpoint WAL record */
6880 19536 : MemSet(&checkPoint, 0, sizeof(checkPoint));
6881 1628 : checkPoint.time = (pg_time_t) time(NULL);
6882 :
6883 : /*
6884 : * For Hot Standby, derive the oldestActiveXid before we fix the redo
6885 : * pointer. This allows us to begin accumulating changes to assemble our
6886 : * starting snapshot of locks and transactions.
6887 : */
6888 1628 : if (!shutdown && XLogStandbyInfoActive())
6889 614 : checkPoint.oldestActiveXid = GetOldestActiveTransactionId();
6890 : else
6891 1014 : checkPoint.oldestActiveXid = InvalidTransactionId;
6892 :
6893 : /*
6894 : * Get location of last important record before acquiring insert locks (as
6895 : * GetLastImportantRecPtr() also locks WAL locks).
6896 : */
6897 1628 : last_important_lsn = GetLastImportantRecPtr();
6898 :
6899 : /*
6900 : * If this isn't a shutdown or forced checkpoint, and if there has been no
6901 : * WAL activity requiring a checkpoint, skip it. The idea here is to
6902 : * avoid inserting duplicate checkpoints when the system is idle.
6903 : */
6904 1628 : if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY |
6905 : CHECKPOINT_FORCE)) == 0)
6906 : {
6907 52 : if (last_important_lsn == ControlFile->checkPoint)
6908 : {
6909 0 : END_CRIT_SECTION();
6910 0 : ereport(DEBUG1,
6911 : (errmsg_internal("checkpoint skipped because system is idle")));
6912 0 : return;
6913 : }
6914 : }
6915 :
6916 : /*
6917 : * An end-of-recovery checkpoint is created before anyone is allowed to
6918 : * write WAL. To allow us to write the checkpoint record, temporarily
6919 : * enable XLogInsertAllowed.
6920 : */
6921 1628 : if (flags & CHECKPOINT_END_OF_RECOVERY)
6922 56 : oldXLogAllowed = LocalSetXLogInsertAllowed();
6923 :
6924 1628 : checkPoint.ThisTimeLineID = XLogCtl->InsertTimeLineID;
6925 1628 : if (flags & CHECKPOINT_END_OF_RECOVERY)
6926 56 : checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
6927 : else
6928 1572 : checkPoint.PrevTimeLineID = checkPoint.ThisTimeLineID;
6929 :
6930 : /*
6931 : * We must block concurrent insertions while examining insert state.
6932 : */
6933 1628 : WALInsertLockAcquireExclusive();
6934 :
6935 1628 : checkPoint.fullPageWrites = Insert->fullPageWrites;
6936 :
6937 1628 : if (shutdown)
6938 : {
6939 962 : XLogRecPtr curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
6940 :
6941 : /*
6942 : * Compute new REDO record ptr = location of next XLOG record.
6943 : *
6944 : * Since this is a shutdown checkpoint, there can't be any concurrent
6945 : * WAL insertion.
6946 : */
6947 962 : freespace = INSERT_FREESPACE(curInsert);
6948 962 : if (freespace == 0)
6949 : {
6950 0 : if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
6951 0 : curInsert += SizeOfXLogLongPHD;
6952 : else
6953 0 : curInsert += SizeOfXLogShortPHD;
6954 : }
6955 962 : checkPoint.redo = curInsert;
6956 :
6957 : /*
6958 : * Here we update the shared RedoRecPtr for future XLogInsert calls;
6959 : * this must be done while holding all the insertion locks.
6960 : *
6961 : * Note: if we fail to complete the checkpoint, RedoRecPtr will be
6962 : * left pointing past where it really needs to point. This is okay;
6963 : * the only consequence is that XLogInsert might back up whole buffers
6964 : * that it didn't really need to. We can't postpone advancing
6965 : * RedoRecPtr because XLogInserts that happen while we are dumping
6966 : * buffers must assume that their buffer changes are not included in
6967 : * the checkpoint.
6968 : */
6969 962 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
6970 : }
6971 :
6972 : /*
6973 : * Now we can release the WAL insertion locks, allowing other xacts to
6974 : * proceed while we are flushing disk buffers.
6975 : */
6976 1628 : WALInsertLockRelease();
6977 :
6978 : /*
6979 : * If this is an online checkpoint, we have not yet determined the redo
6980 : * point. We do so now by inserting the special XLOG_CHECKPOINT_REDO
6981 : * record; the LSN at which it starts becomes the new redo pointer. We
6982 : * don't do this for a shutdown checkpoint, because in that case no WAL
6983 : * can be written between the redo point and the insertion of the
6984 : * checkpoint record itself, so the checkpoint record itself serves to
6985 : * mark the redo point.
6986 : */
6987 1628 : if (!shutdown)
6988 : {
6989 666 : int dummy = 0;
6990 :
6991 : /* Record must have payload to avoid assertion failure. */
6992 666 : XLogBeginInsert();
6993 666 : XLogRegisterData((char *) &dummy, sizeof(dummy));
6994 666 : (void) XLogInsert(RM_XLOG_ID, XLOG_CHECKPOINT_REDO);
6995 :
6996 : /*
6997 : * XLogInsertRecord will have updated XLogCtl->Insert.RedoRecPtr in
6998 : * shared memory and RedoRecPtr in backend-local memory, but we need
6999 : * to copy that into the record that will be inserted when the
7000 : * checkpoint is complete.
7001 : */
7002 666 : checkPoint.redo = RedoRecPtr;
7003 : }
7004 :
7005 : /* Update the info_lck-protected copy of RedoRecPtr as well */
7006 1628 : SpinLockAcquire(&XLogCtl->info_lck);
7007 1628 : XLogCtl->RedoRecPtr = checkPoint.redo;
7008 1628 : SpinLockRelease(&XLogCtl->info_lck);
7009 :
7010 : /*
7011 : * If enabled, log checkpoint start. We postpone this until now so as not
7012 : * to log anything if we decided to skip the checkpoint.
7013 : */
7014 1628 : if (log_checkpoints)
7015 1164 : LogCheckpointStart(flags, false);
7016 :
7017 : /* Update the process title */
7018 1628 : update_checkpoint_display(flags, false, false);
7019 :
7020 : TRACE_POSTGRESQL_CHECKPOINT_START(flags);
7021 :
7022 : /*
7023 : * Get the other info we need for the checkpoint record.
7024 : *
7025 : * We don't need to save oldestClogXid in the checkpoint, it only matters
7026 : * for the short period in which clog is being truncated, and if we crash
7027 : * during that we'll redo the clog truncation and fix up oldestClogXid
7028 : * there.
7029 : */
7030 1628 : LWLockAcquire(XidGenLock, LW_SHARED);
7031 1628 : checkPoint.nextXid = TransamVariables->nextXid;
7032 1628 : checkPoint.oldestXid = TransamVariables->oldestXid;
7033 1628 : checkPoint.oldestXidDB = TransamVariables->oldestXidDB;
7034 1628 : LWLockRelease(XidGenLock);
7035 :
7036 1628 : LWLockAcquire(CommitTsLock, LW_SHARED);
7037 1628 : checkPoint.oldestCommitTsXid = TransamVariables->oldestCommitTsXid;
7038 1628 : checkPoint.newestCommitTsXid = TransamVariables->newestCommitTsXid;
7039 1628 : LWLockRelease(CommitTsLock);
7040 :
7041 1628 : LWLockAcquire(OidGenLock, LW_SHARED);
7042 1628 : checkPoint.nextOid = TransamVariables->nextOid;
7043 1628 : if (!shutdown)
7044 666 : checkPoint.nextOid += TransamVariables->oidCount;
7045 1628 : LWLockRelease(OidGenLock);
7046 :
7047 1628 : MultiXactGetCheckptMulti(shutdown,
7048 : &checkPoint.nextMulti,
7049 : &checkPoint.nextMultiOffset,
7050 : &checkPoint.oldestMulti,
7051 : &checkPoint.oldestMultiDB);
7052 :
7053 : /*
7054 : * Having constructed the checkpoint record, ensure all shmem disk buffers
7055 : * and commit-log buffers are flushed to disk.
7056 : *
7057 : * This I/O could fail for various reasons. If so, we will fail to
7058 : * complete the checkpoint, but there is no reason to force a system
7059 : * panic. Accordingly, exit critical section while doing it.
7060 : */
7061 1628 : END_CRIT_SECTION();
7062 :
7063 : /*
7064 : * In some cases there are groups of actions that must all occur on one
7065 : * side or the other of a checkpoint record. Before flushing the
7066 : * checkpoint record we must explicitly wait for any backend currently
7067 : * performing those groups of actions.
7068 : *
7069 : * One example is end of transaction, so we must wait for any transactions
7070 : * that are currently in commit critical sections. If an xact inserted
7071 : * its commit record into XLOG just before the REDO point, then a crash
7072 : * restart from the REDO point would not replay that record, which means
7073 : * that our flushing had better include the xact's update of pg_xact. So
7074 : * we wait till he's out of his commit critical section before proceeding.
7075 : * See notes in RecordTransactionCommit().
7076 : *
7077 : * Because we've already released the insertion locks, this test is a bit
7078 : * fuzzy: it is possible that we will wait for xacts we didn't really need
7079 : * to wait for. But the delay should be short and it seems better to make
7080 : * checkpoint take a bit longer than to hold off insertions longer than
7081 : * necessary. (In fact, the whole reason we have this issue is that xact.c
7082 : * does commit record XLOG insertion and clog update as two separate steps
7083 : * protected by different locks, but again that seems best on grounds of
7084 : * minimizing lock contention.)
7085 : *
7086 : * A transaction that has not yet set delayChkptFlags when we look cannot
7087 : * be at risk, since it has not inserted its commit record yet; and one
7088 : * that's already cleared it is not at risk either, since it's done fixing
7089 : * clog and we will correctly flush the update below. So we cannot miss
7090 : * any xacts we need to wait for.
7091 : */
7092 1628 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_START);
7093 1628 : if (nvxids > 0)
7094 : {
7095 : do
7096 : {
7097 22 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_START);
7098 22 : pg_usleep(10000L); /* wait for 10 msec */
7099 22 : pgstat_report_wait_end();
7100 22 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7101 22 : DELAY_CHKPT_START));
7102 : }
7103 1628 : pfree(vxids);
7104 :
7105 1628 : CheckPointGuts(checkPoint.redo, flags);
7106 :
7107 1628 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_COMPLETE);
7108 1628 : if (nvxids > 0)
7109 : {
7110 : do
7111 : {
7112 0 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_COMPLETE);
7113 0 : pg_usleep(10000L); /* wait for 10 msec */
7114 0 : pgstat_report_wait_end();
7115 0 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7116 0 : DELAY_CHKPT_COMPLETE));
7117 : }
7118 1628 : pfree(vxids);
7119 :
7120 : /*
7121 : * Take a snapshot of running transactions and write this to WAL. This
7122 : * allows us to reconstruct the state of running transactions during
7123 : * archive recovery, if required. Skip, if this info disabled.
7124 : *
7125 : * If we are shutting down, or Startup process is completing crash
7126 : * recovery we don't need to write running xact data.
7127 : */
7128 1628 : if (!shutdown && XLogStandbyInfoActive())
7129 614 : LogStandbySnapshot();
7130 :
7131 1628 : START_CRIT_SECTION();
7132 :
7133 : /*
7134 : * Now insert the checkpoint record into XLOG.
7135 : */
7136 1628 : XLogBeginInsert();
7137 1628 : XLogRegisterData((char *) (&checkPoint), sizeof(checkPoint));
7138 1628 : recptr = XLogInsert(RM_XLOG_ID,
7139 : shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
7140 : XLOG_CHECKPOINT_ONLINE);
7141 :
7142 1628 : XLogFlush(recptr);
7143 :
7144 : /*
7145 : * We mustn't write any new WAL after a shutdown checkpoint, or it will be
7146 : * overwritten at next startup. No-one should even try, this just allows
7147 : * sanity-checking. In the case of an end-of-recovery checkpoint, we want
7148 : * to just temporarily disable writing until the system has exited
7149 : * recovery.
7150 : */
7151 1628 : if (shutdown)
7152 : {
7153 962 : if (flags & CHECKPOINT_END_OF_RECOVERY)
7154 56 : LocalXLogInsertAllowed = oldXLogAllowed;
7155 : else
7156 906 : LocalXLogInsertAllowed = 0; /* never again write WAL */
7157 : }
7158 :
7159 : /*
7160 : * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
7161 : * = end of actual checkpoint record.
7162 : */
7163 1628 : if (shutdown && checkPoint.redo != ProcLastRecPtr)
7164 0 : ereport(PANIC,
7165 : (errmsg("concurrent write-ahead log activity while database system is shutting down")));
7166 :
7167 : /*
7168 : * Remember the prior checkpoint's redo ptr for
7169 : * UpdateCheckPointDistanceEstimate()
7170 : */
7171 1628 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7172 :
7173 : /*
7174 : * Update the control file.
7175 : */
7176 1628 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7177 1628 : if (shutdown)
7178 962 : ControlFile->state = DB_SHUTDOWNED;
7179 1628 : ControlFile->checkPoint = ProcLastRecPtr;
7180 1628 : ControlFile->checkPointCopy = checkPoint;
7181 : /* crash recovery should always recover to the end of WAL */
7182 1628 : ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
7183 1628 : ControlFile->minRecoveryPointTLI = 0;
7184 :
7185 : /*
7186 : * Persist unloggedLSN value. It's reset on crash recovery, so this goes
7187 : * unused on non-shutdown checkpoints, but seems useful to store it always
7188 : * for debugging purposes.
7189 : */
7190 1628 : ControlFile->unloggedLSN = pg_atomic_read_membarrier_u64(&XLogCtl->unloggedLSN);
7191 :
7192 1628 : UpdateControlFile();
7193 1628 : LWLockRelease(ControlFileLock);
7194 :
7195 : /* Update shared-memory copy of checkpoint XID/epoch */
7196 1628 : SpinLockAcquire(&XLogCtl->info_lck);
7197 1628 : XLogCtl->ckptFullXid = checkPoint.nextXid;
7198 1628 : SpinLockRelease(&XLogCtl->info_lck);
7199 :
7200 : /*
7201 : * We are now done with critical updates; no need for system panic if we
7202 : * have trouble while fooling with old log segments.
7203 : */
7204 1628 : END_CRIT_SECTION();
7205 :
7206 : /*
7207 : * WAL summaries end when the next XLOG_CHECKPOINT_REDO or
7208 : * XLOG_CHECKPOINT_SHUTDOWN record is reached. This is the first point
7209 : * where (a) we're not inside of a critical section and (b) we can be
7210 : * certain that the relevant record has been flushed to disk, which must
7211 : * happen before it can be summarized.
7212 : *
7213 : * If this is a shutdown checkpoint, then this happens reasonably
7214 : * promptly: we've only just inserted and flushed the
7215 : * XLOG_CHECKPOINT_SHUTDOWN record. If this is not a shutdown checkpoint,
7216 : * then this might not be very prompt at all: the XLOG_CHECKPOINT_REDO
7217 : * record was written before we began flushing data to disk, and that
7218 : * could be many minutes ago at this point. However, we don't XLogFlush()
7219 : * after inserting that record, so we're not guaranteed that it's on disk
7220 : * until after the above call that flushes the XLOG_CHECKPOINT_ONLINE
7221 : * record.
7222 : */
7223 1628 : SetWalSummarizerLatch();
7224 :
7225 : /*
7226 : * Let smgr do post-checkpoint cleanup (eg, deleting old files).
7227 : */
7228 1628 : SyncPostCheckpoint();
7229 :
7230 : /*
7231 : * Update the average distance between checkpoints if the prior checkpoint
7232 : * exists.
7233 : */
7234 1628 : if (PriorRedoPtr != InvalidXLogRecPtr)
7235 1628 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7236 :
7237 : /*
7238 : * Delete old log files, those no longer needed for last checkpoint to
7239 : * prevent the disk holding the xlog from growing full.
7240 : */
7241 1628 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7242 1628 : KeepLogSeg(recptr, &_logSegNo);
7243 1628 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED,
7244 : _logSegNo, InvalidOid,
7245 : InvalidTransactionId))
7246 : {
7247 : /*
7248 : * Some slots have been invalidated; recalculate the old-segment
7249 : * horizon, starting again from RedoRecPtr.
7250 : */
7251 6 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7252 6 : KeepLogSeg(recptr, &_logSegNo);
7253 : }
7254 1628 : _logSegNo--;
7255 1628 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr,
7256 : checkPoint.ThisTimeLineID);
7257 :
7258 : /*
7259 : * Make more log segments if needed. (Do this after recycling old log
7260 : * segments, since that may supply some of the needed files.)
7261 : */
7262 1628 : if (!shutdown)
7263 666 : PreallocXlogFiles(recptr, checkPoint.ThisTimeLineID);
7264 :
7265 : /*
7266 : * Truncate pg_subtrans if possible. We can throw away all data before
7267 : * the oldest XMIN of any running transaction. No future transaction will
7268 : * attempt to reference any pg_subtrans entry older than that (see Asserts
7269 : * in subtrans.c). During recovery, though, we mustn't do this because
7270 : * StartupSUBTRANS hasn't been called yet.
7271 : */
7272 1628 : if (!RecoveryInProgress())
7273 1572 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7274 :
7275 : /* Real work is done; log and update stats. */
7276 1628 : LogCheckpointEnd(false);
7277 :
7278 : /* Reset the process title */
7279 1628 : update_checkpoint_display(flags, false, true);
7280 :
7281 : TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
7282 : NBuffers,
7283 : CheckpointStats.ckpt_segs_added,
7284 : CheckpointStats.ckpt_segs_removed,
7285 : CheckpointStats.ckpt_segs_recycled);
7286 : }
7287 :
7288 : /*
7289 : * Mark the end of recovery in WAL though without running a full checkpoint.
7290 : * We can expect that a restartpoint is likely to be in progress as we
7291 : * do this, though we are unwilling to wait for it to complete.
7292 : *
7293 : * CreateRestartPoint() allows for the case where recovery may end before
7294 : * the restartpoint completes so there is no concern of concurrent behaviour.
7295 : */
7296 : static void
7297 76 : CreateEndOfRecoveryRecord(void)
7298 : {
7299 : xl_end_of_recovery xlrec;
7300 : XLogRecPtr recptr;
7301 :
7302 : /* sanity check */
7303 76 : if (!RecoveryInProgress())
7304 0 : elog(ERROR, "can only be used to end recovery");
7305 :
7306 76 : xlrec.end_time = GetCurrentTimestamp();
7307 :
7308 76 : WALInsertLockAcquireExclusive();
7309 76 : xlrec.ThisTimeLineID = XLogCtl->InsertTimeLineID;
7310 76 : xlrec.PrevTimeLineID = XLogCtl->PrevTimeLineID;
7311 76 : WALInsertLockRelease();
7312 :
7313 76 : START_CRIT_SECTION();
7314 :
7315 76 : XLogBeginInsert();
7316 76 : XLogRegisterData((char *) &xlrec, sizeof(xl_end_of_recovery));
7317 76 : recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
7318 :
7319 76 : XLogFlush(recptr);
7320 :
7321 : /*
7322 : * Update the control file so that crash recovery can follow the timeline
7323 : * changes to this point.
7324 : */
7325 76 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7326 76 : ControlFile->minRecoveryPoint = recptr;
7327 76 : ControlFile->minRecoveryPointTLI = xlrec.ThisTimeLineID;
7328 76 : UpdateControlFile();
7329 76 : LWLockRelease(ControlFileLock);
7330 :
7331 76 : END_CRIT_SECTION();
7332 76 : }
7333 :
7334 : /*
7335 : * Write an OVERWRITE_CONTRECORD message.
7336 : *
7337 : * When on WAL replay we expect a continuation record at the start of a page
7338 : * that is not there, recovery ends and WAL writing resumes at that point.
7339 : * But it's wrong to resume writing new WAL back at the start of the record
7340 : * that was broken, because downstream consumers of that WAL (physical
7341 : * replicas) are not prepared to "rewind". So the first action after
7342 : * finishing replay of all valid WAL must be to write a record of this type
7343 : * at the point where the contrecord was missing; to support xlogreader
7344 : * detecting the special case, XLP_FIRST_IS_OVERWRITE_CONTRECORD is also added
7345 : * to the page header where the record occurs. xlogreader has an ad-hoc
7346 : * mechanism to report metadata about the broken record, which is what we
7347 : * use here.
7348 : *
7349 : * At replay time, XLP_FIRST_IS_OVERWRITE_CONTRECORD instructs xlogreader to
7350 : * skip the record it was reading, and pass back the LSN of the skipped
7351 : * record, so that its caller can verify (on "replay" of that record) that the
7352 : * XLOG_OVERWRITE_CONTRECORD matches what was effectively overwritten.
7353 : *
7354 : * 'aborted_lsn' is the beginning position of the record that was incomplete.
7355 : * It is included in the WAL record. 'pagePtr' and 'newTLI' point to the
7356 : * beginning of the XLOG page where the record is to be inserted. They must
7357 : * match the current WAL insert position, they're passed here just so that we
7358 : * can verify that.
7359 : */
7360 : static XLogRecPtr
7361 20 : CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn, XLogRecPtr pagePtr,
7362 : TimeLineID newTLI)
7363 : {
7364 : xl_overwrite_contrecord xlrec;
7365 : XLogRecPtr recptr;
7366 : XLogPageHeader pagehdr;
7367 : XLogRecPtr startPos;
7368 :
7369 : /* sanity checks */
7370 20 : if (!RecoveryInProgress())
7371 0 : elog(ERROR, "can only be used at end of recovery");
7372 20 : if (pagePtr % XLOG_BLCKSZ != 0)
7373 0 : elog(ERROR, "invalid position for missing continuation record %X/%X",
7374 : LSN_FORMAT_ARGS(pagePtr));
7375 :
7376 : /* The current WAL insert position should be right after the page header */
7377 20 : startPos = pagePtr;
7378 20 : if (XLogSegmentOffset(startPos, wal_segment_size) == 0)
7379 2 : startPos += SizeOfXLogLongPHD;
7380 : else
7381 18 : startPos += SizeOfXLogShortPHD;
7382 20 : recptr = GetXLogInsertRecPtr();
7383 20 : if (recptr != startPos)
7384 0 : elog(ERROR, "invalid WAL insert position %X/%X for OVERWRITE_CONTRECORD",
7385 : LSN_FORMAT_ARGS(recptr));
7386 :
7387 20 : START_CRIT_SECTION();
7388 :
7389 : /*
7390 : * Initialize the XLOG page header (by GetXLogBuffer), and set the
7391 : * XLP_FIRST_IS_OVERWRITE_CONTRECORD flag.
7392 : *
7393 : * No other backend is allowed to write WAL yet, so acquiring the WAL
7394 : * insertion lock is just pro forma.
7395 : */
7396 20 : WALInsertLockAcquire();
7397 20 : pagehdr = (XLogPageHeader) GetXLogBuffer(pagePtr, newTLI);
7398 20 : pagehdr->xlp_info |= XLP_FIRST_IS_OVERWRITE_CONTRECORD;
7399 20 : WALInsertLockRelease();
7400 :
7401 : /*
7402 : * Insert the XLOG_OVERWRITE_CONTRECORD record as the first record on the
7403 : * page. We know it becomes the first record, because no other backend is
7404 : * allowed to write WAL yet.
7405 : */
7406 20 : XLogBeginInsert();
7407 20 : xlrec.overwritten_lsn = aborted_lsn;
7408 20 : xlrec.overwrite_time = GetCurrentTimestamp();
7409 20 : XLogRegisterData((char *) &xlrec, sizeof(xl_overwrite_contrecord));
7410 20 : recptr = XLogInsert(RM_XLOG_ID, XLOG_OVERWRITE_CONTRECORD);
7411 :
7412 : /* check that the record was inserted to the right place */
7413 20 : if (ProcLastRecPtr != startPos)
7414 0 : elog(ERROR, "OVERWRITE_CONTRECORD was inserted to unexpected position %X/%X",
7415 : LSN_FORMAT_ARGS(ProcLastRecPtr));
7416 :
7417 20 : XLogFlush(recptr);
7418 :
7419 20 : END_CRIT_SECTION();
7420 :
7421 20 : return recptr;
7422 : }
7423 :
7424 : /*
7425 : * Flush all data in shared memory to disk, and fsync
7426 : *
7427 : * This is the common code shared between regular checkpoints and
7428 : * recovery restartpoints.
7429 : */
7430 : static void
7431 1704 : CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
7432 : {
7433 1704 : CheckPointRelationMap();
7434 1704 : CheckPointReplicationSlots(flags & CHECKPOINT_IS_SHUTDOWN);
7435 1704 : CheckPointSnapBuild();
7436 1704 : CheckPointLogicalRewriteHeap();
7437 1704 : CheckPointReplicationOrigin();
7438 :
7439 : /* Write out all dirty data in SLRUs and the main buffer pool */
7440 : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_START(flags);
7441 1704 : CheckpointStats.ckpt_write_t = GetCurrentTimestamp();
7442 1704 : CheckPointCLOG();
7443 1704 : CheckPointCommitTs();
7444 1704 : CheckPointSUBTRANS();
7445 1704 : CheckPointMultiXact();
7446 1704 : CheckPointPredicate();
7447 1704 : CheckPointBuffers(flags);
7448 :
7449 : /* Perform all queued up fsyncs */
7450 : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_SYNC_START();
7451 1704 : CheckpointStats.ckpt_sync_t = GetCurrentTimestamp();
7452 1704 : ProcessSyncRequests();
7453 1704 : CheckpointStats.ckpt_sync_end_t = GetCurrentTimestamp();
7454 : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_DONE();
7455 :
7456 : /* We deliberately delay 2PC checkpointing as long as possible */
7457 1704 : CheckPointTwoPhase(checkPointRedo);
7458 1704 : }
7459 :
7460 : /*
7461 : * Save a checkpoint for recovery restart if appropriate
7462 : *
7463 : * This function is called each time a checkpoint record is read from XLOG.
7464 : * It must determine whether the checkpoint represents a safe restartpoint or
7465 : * not. If so, the checkpoint record is stashed in shared memory so that
7466 : * CreateRestartPoint can consult it. (Note that the latter function is
7467 : * executed by the checkpointer, while this one will be executed by the
7468 : * startup process.)
7469 : */
7470 : static void
7471 408 : RecoveryRestartPoint(const CheckPoint *checkPoint, XLogReaderState *record)
7472 : {
7473 : /*
7474 : * Also refrain from creating a restartpoint if we have seen any
7475 : * references to non-existent pages. Restarting recovery from the
7476 : * restartpoint would not see the references, so we would lose the
7477 : * cross-check that the pages belonged to a relation that was dropped
7478 : * later.
7479 : */
7480 408 : if (XLogHaveInvalidPages())
7481 : {
7482 0 : elog(DEBUG2,
7483 : "could not record restart point at %X/%X because there "
7484 : "are unresolved references to invalid pages",
7485 : LSN_FORMAT_ARGS(checkPoint->redo));
7486 0 : return;
7487 : }
7488 :
7489 : /*
7490 : * Copy the checkpoint record to shared memory, so that checkpointer can
7491 : * work out the next time it wants to perform a restartpoint.
7492 : */
7493 408 : SpinLockAcquire(&XLogCtl->info_lck);
7494 408 : XLogCtl->lastCheckPointRecPtr = record->ReadRecPtr;
7495 408 : XLogCtl->lastCheckPointEndPtr = record->EndRecPtr;
7496 408 : XLogCtl->lastCheckPoint = *checkPoint;
7497 408 : SpinLockRelease(&XLogCtl->info_lck);
7498 : }
7499 :
7500 : /*
7501 : * Establish a restartpoint if possible.
7502 : *
7503 : * This is similar to CreateCheckPoint, but is used during WAL recovery
7504 : * to establish a point from which recovery can roll forward without
7505 : * replaying the entire recovery log.
7506 : *
7507 : * Returns true if a new restartpoint was established. We can only establish
7508 : * a restartpoint if we have replayed a safe checkpoint record since last
7509 : * restartpoint.
7510 : */
7511 : bool
7512 170 : CreateRestartPoint(int flags)
7513 : {
7514 : XLogRecPtr lastCheckPointRecPtr;
7515 : XLogRecPtr lastCheckPointEndPtr;
7516 : CheckPoint lastCheckPoint;
7517 : XLogRecPtr PriorRedoPtr;
7518 : XLogRecPtr receivePtr;
7519 : XLogRecPtr replayPtr;
7520 : TimeLineID replayTLI;
7521 : XLogRecPtr endptr;
7522 : XLogSegNo _logSegNo;
7523 : TimestampTz xtime;
7524 :
7525 : /* Concurrent checkpoint/restartpoint cannot happen */
7526 : Assert(!IsUnderPostmaster || MyBackendType == B_CHECKPOINTER);
7527 :
7528 : /* Get a local copy of the last safe checkpoint record. */
7529 170 : SpinLockAcquire(&XLogCtl->info_lck);
7530 170 : lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
7531 170 : lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
7532 170 : lastCheckPoint = XLogCtl->lastCheckPoint;
7533 170 : SpinLockRelease(&XLogCtl->info_lck);
7534 :
7535 : /*
7536 : * Check that we're still in recovery mode. It's ok if we exit recovery
7537 : * mode after this check, the restart point is valid anyway.
7538 : */
7539 170 : if (!RecoveryInProgress())
7540 : {
7541 0 : ereport(DEBUG2,
7542 : (errmsg_internal("skipping restartpoint, recovery has already ended")));
7543 0 : return false;
7544 : }
7545 :
7546 : /*
7547 : * If the last checkpoint record we've replayed is already our last
7548 : * restartpoint, we can't perform a new restart point. We still update
7549 : * minRecoveryPoint in that case, so that if this is a shutdown restart
7550 : * point, we won't start up earlier than before. That's not strictly
7551 : * necessary, but when hot standby is enabled, it would be rather weird if
7552 : * the database opened up for read-only connections at a point-in-time
7553 : * before the last shutdown. Such time travel is still possible in case of
7554 : * immediate shutdown, though.
7555 : *
7556 : * We don't explicitly advance minRecoveryPoint when we do create a
7557 : * restartpoint. It's assumed that flushing the buffers will do that as a
7558 : * side-effect.
7559 : */
7560 170 : if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
7561 156 : lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
7562 : {
7563 94 : ereport(DEBUG2,
7564 : (errmsg_internal("skipping restartpoint, already performed at %X/%X",
7565 : LSN_FORMAT_ARGS(lastCheckPoint.redo))));
7566 :
7567 94 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
7568 94 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7569 : {
7570 58 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7571 58 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7572 58 : UpdateControlFile();
7573 58 : LWLockRelease(ControlFileLock);
7574 : }
7575 94 : return false;
7576 : }
7577 :
7578 : /*
7579 : * Update the shared RedoRecPtr so that the startup process can calculate
7580 : * the number of segments replayed since last restartpoint, and request a
7581 : * restartpoint if it exceeds CheckPointSegments.
7582 : *
7583 : * Like in CreateCheckPoint(), hold off insertions to update it, although
7584 : * during recovery this is just pro forma, because no WAL insertions are
7585 : * happening.
7586 : */
7587 76 : WALInsertLockAcquireExclusive();
7588 76 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
7589 76 : WALInsertLockRelease();
7590 :
7591 : /* Also update the info_lck-protected copy */
7592 76 : SpinLockAcquire(&XLogCtl->info_lck);
7593 76 : XLogCtl->RedoRecPtr = lastCheckPoint.redo;
7594 76 : SpinLockRelease(&XLogCtl->info_lck);
7595 :
7596 : /*
7597 : * Prepare to accumulate statistics.
7598 : *
7599 : * Note: because it is possible for log_checkpoints to change while a
7600 : * checkpoint proceeds, we always accumulate stats, even if
7601 : * log_checkpoints is currently off.
7602 : */
7603 836 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
7604 76 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
7605 :
7606 76 : if (log_checkpoints)
7607 76 : LogCheckpointStart(flags, true);
7608 :
7609 : /* Update the process title */
7610 76 : update_checkpoint_display(flags, true, false);
7611 :
7612 76 : CheckPointGuts(lastCheckPoint.redo, flags);
7613 :
7614 : /*
7615 : * This location needs to be after CheckPointGuts() to ensure that some
7616 : * work has already happened during this checkpoint.
7617 : */
7618 : INJECTION_POINT("create-restart-point");
7619 :
7620 : /*
7621 : * Remember the prior checkpoint's redo ptr for
7622 : * UpdateCheckPointDistanceEstimate()
7623 : */
7624 76 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7625 :
7626 : /*
7627 : * Update pg_control, using current time. Check that it still shows an
7628 : * older checkpoint, else do nothing; this is a quick hack to make sure
7629 : * nothing really bad happens if somehow we get here after the
7630 : * end-of-recovery checkpoint.
7631 : */
7632 76 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7633 76 : if (ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
7634 : {
7635 : /*
7636 : * Update the checkpoint information. We do this even if the cluster
7637 : * does not show DB_IN_ARCHIVE_RECOVERY to match with the set of WAL
7638 : * segments recycled below.
7639 : */
7640 76 : ControlFile->checkPoint = lastCheckPointRecPtr;
7641 76 : ControlFile->checkPointCopy = lastCheckPoint;
7642 :
7643 : /*
7644 : * Ensure minRecoveryPoint is past the checkpoint record and update it
7645 : * if the control file still shows DB_IN_ARCHIVE_RECOVERY. Normally,
7646 : * this will have happened already while writing out dirty buffers,
7647 : * but not necessarily - e.g. because no buffers were dirtied. We do
7648 : * this because a backup performed in recovery uses minRecoveryPoint
7649 : * to determine which WAL files must be included in the backup, and
7650 : * the file (or files) containing the checkpoint record must be
7651 : * included, at a minimum. Note that for an ordinary restart of
7652 : * recovery there's no value in having the minimum recovery point any
7653 : * earlier than this anyway, because redo will begin just after the
7654 : * checkpoint record.
7655 : */
7656 76 : if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY)
7657 : {
7658 76 : if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
7659 : {
7660 34 : ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
7661 34 : ControlFile->minRecoveryPointTLI = lastCheckPoint.ThisTimeLineID;
7662 :
7663 : /* update local copy */
7664 34 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
7665 34 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
7666 : }
7667 76 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7668 36 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7669 : }
7670 76 : UpdateControlFile();
7671 : }
7672 76 : LWLockRelease(ControlFileLock);
7673 :
7674 : /*
7675 : * Update the average distance between checkpoints/restartpoints if the
7676 : * prior checkpoint exists.
7677 : */
7678 76 : if (PriorRedoPtr != InvalidXLogRecPtr)
7679 76 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7680 :
7681 : /*
7682 : * Delete old log files, those no longer needed for last restartpoint to
7683 : * prevent the disk holding the xlog from growing full.
7684 : */
7685 76 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7686 :
7687 : /*
7688 : * Retreat _logSegNo using the current end of xlog replayed or received,
7689 : * whichever is later.
7690 : */
7691 76 : receivePtr = GetWalRcvFlushRecPtr(NULL, NULL);
7692 76 : replayPtr = GetXLogReplayRecPtr(&replayTLI);
7693 76 : endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
7694 76 : KeepLogSeg(endptr, &_logSegNo);
7695 76 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED,
7696 : _logSegNo, InvalidOid,
7697 : InvalidTransactionId))
7698 : {
7699 : /*
7700 : * Some slots have been invalidated; recalculate the old-segment
7701 : * horizon, starting again from RedoRecPtr.
7702 : */
7703 2 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7704 2 : KeepLogSeg(endptr, &_logSegNo);
7705 : }
7706 76 : _logSegNo--;
7707 :
7708 : /*
7709 : * Try to recycle segments on a useful timeline. If we've been promoted
7710 : * since the beginning of this restartpoint, use the new timeline chosen
7711 : * at end of recovery. If we're still in recovery, use the timeline we're
7712 : * currently replaying.
7713 : *
7714 : * There is no guarantee that the WAL segments will be useful on the
7715 : * current timeline; if recovery proceeds to a new timeline right after
7716 : * this, the pre-allocated WAL segments on this timeline will not be used,
7717 : * and will go wasted until recycled on the next restartpoint. We'll live
7718 : * with that.
7719 : */
7720 76 : if (!RecoveryInProgress())
7721 0 : replayTLI = XLogCtl->InsertTimeLineID;
7722 :
7723 76 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr, replayTLI);
7724 :
7725 : /*
7726 : * Make more log segments if needed. (Do this after recycling old log
7727 : * segments, since that may supply some of the needed files.)
7728 : */
7729 76 : PreallocXlogFiles(endptr, replayTLI);
7730 :
7731 : /*
7732 : * Truncate pg_subtrans if possible. We can throw away all data before
7733 : * the oldest XMIN of any running transaction. No future transaction will
7734 : * attempt to reference any pg_subtrans entry older than that (see Asserts
7735 : * in subtrans.c). When hot standby is disabled, though, we mustn't do
7736 : * this because StartupSUBTRANS hasn't been called yet.
7737 : */
7738 76 : if (EnableHotStandby)
7739 76 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7740 :
7741 : /* Real work is done; log and update stats. */
7742 76 : LogCheckpointEnd(true);
7743 :
7744 : /* Reset the process title */
7745 76 : update_checkpoint_display(flags, true, true);
7746 :
7747 76 : xtime = GetLatestXTime();
7748 76 : ereport((log_checkpoints ? LOG : DEBUG2),
7749 : (errmsg("recovery restart point at %X/%X",
7750 : LSN_FORMAT_ARGS(lastCheckPoint.redo)),
7751 : xtime ? errdetail("Last completed transaction was at log time %s.",
7752 : timestamptz_to_str(xtime)) : 0));
7753 :
7754 : /*
7755 : * Finally, execute archive_cleanup_command, if any.
7756 : */
7757 76 : if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
7758 0 : ExecuteRecoveryCommand(archiveCleanupCommand,
7759 : "archive_cleanup_command",
7760 : false,
7761 : WAIT_EVENT_ARCHIVE_CLEANUP_COMMAND);
7762 :
7763 76 : return true;
7764 : }
7765 :
7766 : /*
7767 : * Report availability of WAL for the given target LSN
7768 : * (typically a slot's restart_lsn)
7769 : *
7770 : * Returns one of the following enum values:
7771 : *
7772 : * * WALAVAIL_RESERVED means targetLSN is available and it is in the range of
7773 : * max_wal_size.
7774 : *
7775 : * * WALAVAIL_EXTENDED means it is still available by preserving extra
7776 : * segments beyond max_wal_size. If max_slot_wal_keep_size is smaller
7777 : * than max_wal_size, this state is not returned.
7778 : *
7779 : * * WALAVAIL_UNRESERVED means it is being lost and the next checkpoint will
7780 : * remove reserved segments. The walsender using this slot may return to the
7781 : * above.
7782 : *
7783 : * * WALAVAIL_REMOVED means it has been removed. A replication stream on
7784 : * a slot with this LSN cannot continue. (Any associated walsender
7785 : * processes should have been terminated already.)
7786 : *
7787 : * * WALAVAIL_INVALID_LSN means the slot hasn't been set to reserve WAL.
7788 : */
7789 : WALAvailability
7790 832 : GetWALAvailability(XLogRecPtr targetLSN)
7791 : {
7792 : XLogRecPtr currpos; /* current write LSN */
7793 : XLogSegNo currSeg; /* segid of currpos */
7794 : XLogSegNo targetSeg; /* segid of targetLSN */
7795 : XLogSegNo oldestSeg; /* actual oldest segid */
7796 : XLogSegNo oldestSegMaxWalSize; /* oldest segid kept by max_wal_size */
7797 : XLogSegNo oldestSlotSeg; /* oldest segid kept by slot */
7798 : uint64 keepSegs;
7799 :
7800 : /*
7801 : * slot does not reserve WAL. Either deactivated, or has never been active
7802 : */
7803 832 : if (XLogRecPtrIsInvalid(targetLSN))
7804 22 : return WALAVAIL_INVALID_LSN;
7805 :
7806 : /*
7807 : * Calculate the oldest segment currently reserved by all slots,
7808 : * considering wal_keep_size and max_slot_wal_keep_size. Initialize
7809 : * oldestSlotSeg to the current segment.
7810 : */
7811 810 : currpos = GetXLogWriteRecPtr();
7812 810 : XLByteToSeg(currpos, oldestSlotSeg, wal_segment_size);
7813 810 : KeepLogSeg(currpos, &oldestSlotSeg);
7814 :
7815 : /*
7816 : * Find the oldest extant segment file. We get 1 until checkpoint removes
7817 : * the first WAL segment file since startup, which causes the status being
7818 : * wrong under certain abnormal conditions but that doesn't actually harm.
7819 : */
7820 810 : oldestSeg = XLogGetLastRemovedSegno() + 1;
7821 :
7822 : /* calculate oldest segment by max_wal_size */
7823 810 : XLByteToSeg(currpos, currSeg, wal_segment_size);
7824 810 : keepSegs = ConvertToXSegs(max_wal_size_mb, wal_segment_size) + 1;
7825 :
7826 810 : if (currSeg > keepSegs)
7827 16 : oldestSegMaxWalSize = currSeg - keepSegs;
7828 : else
7829 794 : oldestSegMaxWalSize = 1;
7830 :
7831 : /* the segment we care about */
7832 810 : XLByteToSeg(targetLSN, targetSeg, wal_segment_size);
7833 :
7834 : /*
7835 : * No point in returning reserved or extended status values if the
7836 : * targetSeg is known to be lost.
7837 : */
7838 810 : if (targetSeg >= oldestSlotSeg)
7839 : {
7840 : /* show "reserved" when targetSeg is within max_wal_size */
7841 808 : if (targetSeg >= oldestSegMaxWalSize)
7842 804 : return WALAVAIL_RESERVED;
7843 :
7844 : /* being retained by slots exceeding max_wal_size */
7845 4 : return WALAVAIL_EXTENDED;
7846 : }
7847 :
7848 : /* WAL segments are no longer retained but haven't been removed yet */
7849 2 : if (targetSeg >= oldestSeg)
7850 2 : return WALAVAIL_UNRESERVED;
7851 :
7852 : /* Definitely lost */
7853 0 : return WALAVAIL_REMOVED;
7854 : }
7855 :
7856 :
7857 : /*
7858 : * Retreat *logSegNo to the last segment that we need to retain because of
7859 : * either wal_keep_size or replication slots.
7860 : *
7861 : * This is calculated by subtracting wal_keep_size from the given xlog
7862 : * location, recptr and by making sure that that result is below the
7863 : * requirement of replication slots. For the latter criterion we do consider
7864 : * the effects of max_slot_wal_keep_size: reserve at most that much space back
7865 : * from recptr.
7866 : *
7867 : * Note about replication slots: if this function calculates a value
7868 : * that's further ahead than what slots need reserved, then affected
7869 : * slots need to be invalidated and this function invoked again.
7870 : * XXX it might be a good idea to rewrite this function so that
7871 : * invalidation is optionally done here, instead.
7872 : */
7873 : static void
7874 2522 : KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
7875 : {
7876 : XLogSegNo currSegNo;
7877 : XLogSegNo segno;
7878 : XLogRecPtr keep;
7879 :
7880 2522 : XLByteToSeg(recptr, currSegNo, wal_segment_size);
7881 2522 : segno = currSegNo;
7882 :
7883 : /*
7884 : * Calculate how many segments are kept by slots first, adjusting for
7885 : * max_slot_wal_keep_size.
7886 : */
7887 2522 : keep = XLogGetReplicationSlotMinimumLSN();
7888 2522 : if (keep != InvalidXLogRecPtr && keep < recptr)
7889 : {
7890 704 : XLByteToSeg(keep, segno, wal_segment_size);
7891 :
7892 : /* Cap by max_slot_wal_keep_size ... */
7893 704 : if (max_slot_wal_keep_size_mb >= 0)
7894 : {
7895 : uint64 slot_keep_segs;
7896 :
7897 40 : slot_keep_segs =
7898 40 : ConvertToXSegs(max_slot_wal_keep_size_mb, wal_segment_size);
7899 :
7900 40 : if (currSegNo - segno > slot_keep_segs)
7901 10 : segno = currSegNo - slot_keep_segs;
7902 : }
7903 : }
7904 :
7905 : /*
7906 : * If WAL summarization is in use, don't remove WAL that has yet to be
7907 : * summarized.
7908 : */
7909 2522 : keep = GetOldestUnsummarizedLSN(NULL, NULL, false);
7910 2522 : if (keep != InvalidXLogRecPtr)
7911 : {
7912 : XLogSegNo unsummarized_segno;
7913 :
7914 0 : XLByteToSeg(keep, unsummarized_segno, wal_segment_size);
7915 0 : if (unsummarized_segno < segno)
7916 0 : segno = unsummarized_segno;
7917 : }
7918 :
7919 : /* but, keep at least wal_keep_size if that's set */
7920 2522 : if (wal_keep_size_mb > 0)
7921 : {
7922 : uint64 keep_segs;
7923 :
7924 122 : keep_segs = ConvertToXSegs(wal_keep_size_mb, wal_segment_size);
7925 122 : if (currSegNo - segno < keep_segs)
7926 : {
7927 : /* avoid underflow, don't go below 1 */
7928 122 : if (currSegNo <= keep_segs)
7929 114 : segno = 1;
7930 : else
7931 8 : segno = currSegNo - keep_segs;
7932 : }
7933 : }
7934 :
7935 : /* don't delete WAL segments newer than the calculated segment */
7936 2522 : if (segno < *logSegNo)
7937 330 : *logSegNo = segno;
7938 2522 : }
7939 :
7940 : /*
7941 : * Write a NEXTOID log record
7942 : */
7943 : void
7944 992 : XLogPutNextOid(Oid nextOid)
7945 : {
7946 992 : XLogBeginInsert();
7947 992 : XLogRegisterData((char *) (&nextOid), sizeof(Oid));
7948 992 : (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
7949 :
7950 : /*
7951 : * We need not flush the NEXTOID record immediately, because any of the
7952 : * just-allocated OIDs could only reach disk as part of a tuple insert or
7953 : * update that would have its own XLOG record that must follow the NEXTOID
7954 : * record. Therefore, the standard buffer LSN interlock applied to those
7955 : * records will ensure no such OID reaches disk before the NEXTOID record
7956 : * does.
7957 : *
7958 : * Note, however, that the above statement only covers state "within" the
7959 : * database. When we use a generated OID as a file or directory name, we
7960 : * are in a sense violating the basic WAL rule, because that filesystem
7961 : * change may reach disk before the NEXTOID WAL record does. The impact
7962 : * of this is that if a database crash occurs immediately afterward, we
7963 : * might after restart re-generate the same OID and find that it conflicts
7964 : * with the leftover file or directory. But since for safety's sake we
7965 : * always loop until finding a nonconflicting filename, this poses no real
7966 : * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
7967 : */
7968 992 : }
7969 :
7970 : /*
7971 : * Write an XLOG SWITCH record.
7972 : *
7973 : * Here we just blindly issue an XLogInsert request for the record.
7974 : * All the magic happens inside XLogInsert.
7975 : *
7976 : * The return value is either the end+1 address of the switch record,
7977 : * or the end+1 address of the prior segment if we did not need to
7978 : * write a switch record because we are already at segment start.
7979 : */
7980 : XLogRecPtr
7981 666 : RequestXLogSwitch(bool mark_unimportant)
7982 : {
7983 : XLogRecPtr RecPtr;
7984 :
7985 : /* XLOG SWITCH has no data */
7986 666 : XLogBeginInsert();
7987 :
7988 666 : if (mark_unimportant)
7989 0 : XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT);
7990 666 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
7991 :
7992 666 : return RecPtr;
7993 : }
7994 :
7995 : /*
7996 : * Write a RESTORE POINT record
7997 : */
7998 : XLogRecPtr
7999 6 : XLogRestorePoint(const char *rpName)
8000 : {
8001 : XLogRecPtr RecPtr;
8002 : xl_restore_point xlrec;
8003 :
8004 6 : xlrec.rp_time = GetCurrentTimestamp();
8005 6 : strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
8006 :
8007 6 : XLogBeginInsert();
8008 6 : XLogRegisterData((char *) &xlrec, sizeof(xl_restore_point));
8009 :
8010 6 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
8011 :
8012 6 : ereport(LOG,
8013 : (errmsg("restore point \"%s\" created at %X/%X",
8014 : rpName, LSN_FORMAT_ARGS(RecPtr))));
8015 :
8016 6 : return RecPtr;
8017 : }
8018 :
8019 : /*
8020 : * Check if any of the GUC parameters that are critical for hot standby
8021 : * have changed, and update the value in pg_control file if necessary.
8022 : */
8023 : static void
8024 1420 : XLogReportParameters(void)
8025 : {
8026 1420 : if (wal_level != ControlFile->wal_level ||
8027 1048 : wal_log_hints != ControlFile->wal_log_hints ||
8028 906 : MaxConnections != ControlFile->MaxConnections ||
8029 904 : max_worker_processes != ControlFile->max_worker_processes ||
8030 902 : max_wal_senders != ControlFile->max_wal_senders ||
8031 880 : max_prepared_xacts != ControlFile->max_prepared_xacts ||
8032 718 : max_locks_per_xact != ControlFile->max_locks_per_xact ||
8033 718 : track_commit_timestamp != ControlFile->track_commit_timestamp)
8034 : {
8035 : /*
8036 : * The change in number of backend slots doesn't need to be WAL-logged
8037 : * if archiving is not enabled, as you can't start archive recovery
8038 : * with wal_level=minimal anyway. We don't really care about the
8039 : * values in pg_control either if wal_level=minimal, but seems better
8040 : * to keep them up-to-date to avoid confusion.
8041 : */
8042 716 : if (wal_level != ControlFile->wal_level || XLogIsNeeded())
8043 : {
8044 : xl_parameter_change xlrec;
8045 : XLogRecPtr recptr;
8046 :
8047 700 : xlrec.MaxConnections = MaxConnections;
8048 700 : xlrec.max_worker_processes = max_worker_processes;
8049 700 : xlrec.max_wal_senders = max_wal_senders;
8050 700 : xlrec.max_prepared_xacts = max_prepared_xacts;
8051 700 : xlrec.max_locks_per_xact = max_locks_per_xact;
8052 700 : xlrec.wal_level = wal_level;
8053 700 : xlrec.wal_log_hints = wal_log_hints;
8054 700 : xlrec.track_commit_timestamp = track_commit_timestamp;
8055 :
8056 700 : XLogBeginInsert();
8057 700 : XLogRegisterData((char *) &xlrec, sizeof(xlrec));
8058 :
8059 700 : recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
8060 700 : XLogFlush(recptr);
8061 : }
8062 :
8063 716 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8064 :
8065 716 : ControlFile->MaxConnections = MaxConnections;
8066 716 : ControlFile->max_worker_processes = max_worker_processes;
8067 716 : ControlFile->max_wal_senders = max_wal_senders;
8068 716 : ControlFile->max_prepared_xacts = max_prepared_xacts;
8069 716 : ControlFile->max_locks_per_xact = max_locks_per_xact;
8070 716 : ControlFile->wal_level = wal_level;
8071 716 : ControlFile->wal_log_hints = wal_log_hints;
8072 716 : ControlFile->track_commit_timestamp = track_commit_timestamp;
8073 716 : UpdateControlFile();
8074 :
8075 716 : LWLockRelease(ControlFileLock);
8076 : }
8077 1420 : }
8078 :
8079 : /*
8080 : * Update full_page_writes in shared memory, and write an
8081 : * XLOG_FPW_CHANGE record if necessary.
8082 : *
8083 : * Note: this function assumes there is no other process running
8084 : * concurrently that could update it.
8085 : */
8086 : void
8087 2344 : UpdateFullPageWrites(void)
8088 : {
8089 2344 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8090 : bool recoveryInProgress;
8091 :
8092 : /*
8093 : * Do nothing if full_page_writes has not been changed.
8094 : *
8095 : * It's safe to check the shared full_page_writes without the lock,
8096 : * because we assume that there is no concurrently running process which
8097 : * can update it.
8098 : */
8099 2344 : if (fullPageWrites == Insert->fullPageWrites)
8100 1536 : return;
8101 :
8102 : /*
8103 : * Perform this outside critical section so that the WAL insert
8104 : * initialization done by RecoveryInProgress() doesn't trigger an
8105 : * assertion failure.
8106 : */
8107 808 : recoveryInProgress = RecoveryInProgress();
8108 :
8109 808 : START_CRIT_SECTION();
8110 :
8111 : /*
8112 : * It's always safe to take full page images, even when not strictly
8113 : * required, but not the other round. So if we're setting full_page_writes
8114 : * to true, first set it true and then write the WAL record. If we're
8115 : * setting it to false, first write the WAL record and then set the global
8116 : * flag.
8117 : */
8118 808 : if (fullPageWrites)
8119 : {
8120 800 : WALInsertLockAcquireExclusive();
8121 800 : Insert->fullPageWrites = true;
8122 800 : WALInsertLockRelease();
8123 : }
8124 :
8125 : /*
8126 : * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
8127 : * full_page_writes during archive recovery, if required.
8128 : */
8129 808 : if (XLogStandbyInfoActive() && !recoveryInProgress)
8130 : {
8131 0 : XLogBeginInsert();
8132 0 : XLogRegisterData((char *) (&fullPageWrites), sizeof(bool));
8133 :
8134 0 : XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
8135 : }
8136 :
8137 808 : if (!fullPageWrites)
8138 : {
8139 8 : WALInsertLockAcquireExclusive();
8140 8 : Insert->fullPageWrites = false;
8141 8 : WALInsertLockRelease();
8142 : }
8143 808 : END_CRIT_SECTION();
8144 : }
8145 :
8146 : /*
8147 : * XLOG resource manager's routines
8148 : *
8149 : * Definitions of info values are in include/catalog/pg_control.h, though
8150 : * not all record types are related to control file updates.
8151 : *
8152 : * NOTE: Some XLOG record types that are directly related to WAL recovery
8153 : * are handled in xlogrecovery_redo().
8154 : */
8155 : void
8156 56652 : xlog_redo(XLogReaderState *record)
8157 : {
8158 56652 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
8159 56652 : XLogRecPtr lsn = record->EndRecPtr;
8160 :
8161 : /*
8162 : * In XLOG rmgr, backup blocks are only used by XLOG_FPI and
8163 : * XLOG_FPI_FOR_HINT records.
8164 : */
8165 : Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
8166 : !XLogRecHasAnyBlockRefs(record));
8167 :
8168 56652 : if (info == XLOG_NEXTOID)
8169 : {
8170 : Oid nextOid;
8171 :
8172 : /*
8173 : * We used to try to take the maximum of TransamVariables->nextOid and
8174 : * the recorded nextOid, but that fails if the OID counter wraps
8175 : * around. Since no OID allocation should be happening during replay
8176 : * anyway, better to just believe the record exactly. We still take
8177 : * OidGenLock while setting the variable, just in case.
8178 : */
8179 156 : memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
8180 156 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
8181 156 : TransamVariables->nextOid = nextOid;
8182 156 : TransamVariables->oidCount = 0;
8183 156 : LWLockRelease(OidGenLock);
8184 : }
8185 56496 : else if (info == XLOG_CHECKPOINT_SHUTDOWN)
8186 : {
8187 : CheckPoint checkPoint;
8188 : TimeLineID replayTLI;
8189 :
8190 58 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8191 : /* In a SHUTDOWN checkpoint, believe the counters exactly */
8192 58 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
8193 58 : TransamVariables->nextXid = checkPoint.nextXid;
8194 58 : LWLockRelease(XidGenLock);
8195 58 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
8196 58 : TransamVariables->nextOid = checkPoint.nextOid;
8197 58 : TransamVariables->oidCount = 0;
8198 58 : LWLockRelease(OidGenLock);
8199 58 : MultiXactSetNextMXact(checkPoint.nextMulti,
8200 : checkPoint.nextMultiOffset);
8201 :
8202 58 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8203 : checkPoint.oldestMultiDB);
8204 :
8205 : /*
8206 : * No need to set oldestClogXid here as well; it'll be set when we
8207 : * redo an xl_clog_truncate if it changed since initialization.
8208 : */
8209 58 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
8210 :
8211 : /*
8212 : * If we see a shutdown checkpoint while waiting for an end-of-backup
8213 : * record, the backup was canceled and the end-of-backup record will
8214 : * never arrive.
8215 : */
8216 58 : if (ArchiveRecoveryRequested &&
8217 58 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint) &&
8218 0 : XLogRecPtrIsInvalid(ControlFile->backupEndPoint))
8219 0 : ereport(PANIC,
8220 : (errmsg("online backup was canceled, recovery cannot continue")));
8221 :
8222 : /*
8223 : * If we see a shutdown checkpoint, we know that nothing was running
8224 : * on the primary at this point. So fake-up an empty running-xacts
8225 : * record and use that here and now. Recover additional standby state
8226 : * for prepared transactions.
8227 : */
8228 58 : if (standbyState >= STANDBY_INITIALIZED)
8229 : {
8230 : TransactionId *xids;
8231 : int nxids;
8232 : TransactionId oldestActiveXID;
8233 : TransactionId latestCompletedXid;
8234 : RunningTransactionsData running;
8235 :
8236 54 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
8237 :
8238 : /*
8239 : * Construct a RunningTransactions snapshot representing a shut
8240 : * down server, with only prepared transactions still alive. We're
8241 : * never overflowed at this point because all subxids are listed
8242 : * with their parent prepared transactions.
8243 : */
8244 54 : running.xcnt = nxids;
8245 54 : running.subxcnt = 0;
8246 54 : running.subxid_overflow = false;
8247 54 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
8248 54 : running.oldestRunningXid = oldestActiveXID;
8249 54 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
8250 54 : TransactionIdRetreat(latestCompletedXid);
8251 : Assert(TransactionIdIsNormal(latestCompletedXid));
8252 54 : running.latestCompletedXid = latestCompletedXid;
8253 54 : running.xids = xids;
8254 :
8255 54 : ProcArrayApplyRecoveryInfo(&running);
8256 :
8257 54 : StandbyRecoverPreparedTransactions();
8258 : }
8259 :
8260 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
8261 58 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8262 58 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
8263 58 : LWLockRelease(ControlFileLock);
8264 :
8265 : /* Update shared-memory copy of checkpoint XID/epoch */
8266 58 : SpinLockAcquire(&XLogCtl->info_lck);
8267 58 : XLogCtl->ckptFullXid = checkPoint.nextXid;
8268 58 : SpinLockRelease(&XLogCtl->info_lck);
8269 :
8270 : /*
8271 : * We should've already switched to the new TLI before replaying this
8272 : * record.
8273 : */
8274 58 : (void) GetCurrentReplayRecPtr(&replayTLI);
8275 58 : if (checkPoint.ThisTimeLineID != replayTLI)
8276 0 : ereport(PANIC,
8277 : (errmsg("unexpected timeline ID %u (should be %u) in shutdown checkpoint record",
8278 : checkPoint.ThisTimeLineID, replayTLI)));
8279 :
8280 58 : RecoveryRestartPoint(&checkPoint, record);
8281 : }
8282 56438 : else if (info == XLOG_CHECKPOINT_ONLINE)
8283 : {
8284 : CheckPoint checkPoint;
8285 : TimeLineID replayTLI;
8286 :
8287 350 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8288 : /* In an ONLINE checkpoint, treat the XID counter as a minimum */
8289 350 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
8290 350 : if (FullTransactionIdPrecedes(TransamVariables->nextXid,
8291 : checkPoint.nextXid))
8292 0 : TransamVariables->nextXid = checkPoint.nextXid;
8293 350 : LWLockRelease(XidGenLock);
8294 :
8295 : /*
8296 : * We ignore the nextOid counter in an ONLINE checkpoint, preferring
8297 : * to track OID assignment through XLOG_NEXTOID records. The nextOid
8298 : * counter is from the start of the checkpoint and might well be stale
8299 : * compared to later XLOG_NEXTOID records. We could try to take the
8300 : * maximum of the nextOid counter and our latest value, but since
8301 : * there's no particular guarantee about the speed with which the OID
8302 : * counter wraps around, that's a risky thing to do. In any case,
8303 : * users of the nextOid counter are required to avoid assignment of
8304 : * duplicates, so that a somewhat out-of-date value should be safe.
8305 : */
8306 :
8307 : /* Handle multixact */
8308 350 : MultiXactAdvanceNextMXact(checkPoint.nextMulti,
8309 : checkPoint.nextMultiOffset);
8310 :
8311 : /*
8312 : * NB: This may perform multixact truncation when replaying WAL
8313 : * generated by an older primary.
8314 : */
8315 350 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8316 : checkPoint.oldestMultiDB);
8317 350 : if (TransactionIdPrecedes(TransamVariables->oldestXid,
8318 : checkPoint.oldestXid))
8319 0 : SetTransactionIdLimit(checkPoint.oldestXid,
8320 : checkPoint.oldestXidDB);
8321 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
8322 350 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8323 350 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
8324 350 : LWLockRelease(ControlFileLock);
8325 :
8326 : /* Update shared-memory copy of checkpoint XID/epoch */
8327 350 : SpinLockAcquire(&XLogCtl->info_lck);
8328 350 : XLogCtl->ckptFullXid = checkPoint.nextXid;
8329 350 : SpinLockRelease(&XLogCtl->info_lck);
8330 :
8331 : /* TLI should not change in an on-line checkpoint */
8332 350 : (void) GetCurrentReplayRecPtr(&replayTLI);
8333 350 : if (checkPoint.ThisTimeLineID != replayTLI)
8334 0 : ereport(PANIC,
8335 : (errmsg("unexpected timeline ID %u (should be %u) in online checkpoint record",
8336 : checkPoint.ThisTimeLineID, replayTLI)));
8337 :
8338 350 : RecoveryRestartPoint(&checkPoint, record);
8339 : }
8340 56088 : else if (info == XLOG_OVERWRITE_CONTRECORD)
8341 : {
8342 : /* nothing to do here, handled in xlogrecovery_redo() */
8343 : }
8344 56086 : else if (info == XLOG_END_OF_RECOVERY)
8345 : {
8346 : xl_end_of_recovery xlrec;
8347 : TimeLineID replayTLI;
8348 :
8349 18 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
8350 :
8351 : /*
8352 : * For Hot Standby, we could treat this like a Shutdown Checkpoint,
8353 : * but this case is rarer and harder to test, so the benefit doesn't
8354 : * outweigh the potential extra cost of maintenance.
8355 : */
8356 :
8357 : /*
8358 : * We should've already switched to the new TLI before replaying this
8359 : * record.
8360 : */
8361 18 : (void) GetCurrentReplayRecPtr(&replayTLI);
8362 18 : if (xlrec.ThisTimeLineID != replayTLI)
8363 0 : ereport(PANIC,
8364 : (errmsg("unexpected timeline ID %u (should be %u) in end-of-recovery record",
8365 : xlrec.ThisTimeLineID, replayTLI)));
8366 : }
8367 56068 : else if (info == XLOG_NOOP)
8368 : {
8369 : /* nothing to do here */
8370 : }
8371 56068 : else if (info == XLOG_SWITCH)
8372 : {
8373 : /* nothing to do here */
8374 : }
8375 55850 : else if (info == XLOG_RESTORE_POINT)
8376 : {
8377 : /* nothing to do here, handled in xlogrecovery.c */
8378 : }
8379 55840 : else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
8380 : {
8381 : /*
8382 : * XLOG_FPI records contain nothing else but one or more block
8383 : * references. Every block reference must include a full-page image
8384 : * even if full_page_writes was disabled when the record was generated
8385 : * - otherwise there would be no point in this record.
8386 : *
8387 : * XLOG_FPI_FOR_HINT records are generated when a page needs to be
8388 : * WAL-logged because of a hint bit update. They are only generated
8389 : * when checksums and/or wal_log_hints are enabled. They may include
8390 : * no full-page images if full_page_writes was disabled when they were
8391 : * generated. In this case there is nothing to do here.
8392 : *
8393 : * No recovery conflicts are generated by these generic records - if a
8394 : * resource manager needs to generate conflicts, it has to define a
8395 : * separate WAL record type and redo routine.
8396 : */
8397 120146 : for (uint8 block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
8398 : {
8399 : Buffer buffer;
8400 :
8401 64870 : if (!XLogRecHasBlockImage(record, block_id))
8402 : {
8403 136 : if (info == XLOG_FPI)
8404 0 : elog(ERROR, "XLOG_FPI record did not contain a full-page image");
8405 136 : continue;
8406 : }
8407 :
8408 64734 : if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
8409 0 : elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
8410 64734 : UnlockReleaseBuffer(buffer);
8411 : }
8412 : }
8413 564 : else if (info == XLOG_BACKUP_END)
8414 : {
8415 : /* nothing to do here, handled in xlogrecovery_redo() */
8416 : }
8417 412 : else if (info == XLOG_PARAMETER_CHANGE)
8418 : {
8419 : xl_parameter_change xlrec;
8420 :
8421 : /* Update our copy of the parameters in pg_control */
8422 60 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
8423 :
8424 : /*
8425 : * Invalidate logical slots if we are in hot standby and the primary
8426 : * does not have a WAL level sufficient for logical decoding. No need
8427 : * to search for potentially conflicting logically slots if standby is
8428 : * running with wal_level lower than logical, because in that case, we
8429 : * would have either disallowed creation of logical slots or
8430 : * invalidated existing ones.
8431 : */
8432 60 : if (InRecovery && InHotStandby &&
8433 28 : xlrec.wal_level < WAL_LEVEL_LOGICAL &&
8434 12 : wal_level >= WAL_LEVEL_LOGICAL)
8435 8 : InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_LEVEL,
8436 : 0, InvalidOid,
8437 : InvalidTransactionId);
8438 :
8439 60 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8440 60 : ControlFile->MaxConnections = xlrec.MaxConnections;
8441 60 : ControlFile->max_worker_processes = xlrec.max_worker_processes;
8442 60 : ControlFile->max_wal_senders = xlrec.max_wal_senders;
8443 60 : ControlFile->max_prepared_xacts = xlrec.max_prepared_xacts;
8444 60 : ControlFile->max_locks_per_xact = xlrec.max_locks_per_xact;
8445 60 : ControlFile->wal_level = xlrec.wal_level;
8446 60 : ControlFile->wal_log_hints = xlrec.wal_log_hints;
8447 :
8448 : /*
8449 : * Update minRecoveryPoint to ensure that if recovery is aborted, we
8450 : * recover back up to this point before allowing hot standby again.
8451 : * This is important if the max_* settings are decreased, to ensure
8452 : * you don't run queries against the WAL preceding the change. The
8453 : * local copies cannot be updated as long as crash recovery is
8454 : * happening and we expect all the WAL to be replayed.
8455 : */
8456 60 : if (InArchiveRecovery)
8457 : {
8458 30 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
8459 30 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
8460 : }
8461 60 : if (LocalMinRecoveryPoint != InvalidXLogRecPtr && LocalMinRecoveryPoint < lsn)
8462 : {
8463 : TimeLineID replayTLI;
8464 :
8465 14 : (void) GetCurrentReplayRecPtr(&replayTLI);
8466 14 : ControlFile->minRecoveryPoint = lsn;
8467 14 : ControlFile->minRecoveryPointTLI = replayTLI;
8468 : }
8469 :
8470 60 : CommitTsParameterChange(xlrec.track_commit_timestamp,
8471 60 : ControlFile->track_commit_timestamp);
8472 60 : ControlFile->track_commit_timestamp = xlrec.track_commit_timestamp;
8473 :
8474 60 : UpdateControlFile();
8475 60 : LWLockRelease(ControlFileLock);
8476 :
8477 : /* Check to see if any parameter change gives a problem on recovery */
8478 60 : CheckRequiredParameterValues();
8479 : }
8480 352 : else if (info == XLOG_FPW_CHANGE)
8481 : {
8482 : bool fpw;
8483 :
8484 0 : memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
8485 :
8486 : /*
8487 : * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
8488 : * do_pg_backup_start() and do_pg_backup_stop() can check whether
8489 : * full_page_writes has been disabled during online backup.
8490 : */
8491 0 : if (!fpw)
8492 : {
8493 0 : SpinLockAcquire(&XLogCtl->info_lck);
8494 0 : if (XLogCtl->lastFpwDisableRecPtr < record->ReadRecPtr)
8495 0 : XLogCtl->lastFpwDisableRecPtr = record->ReadRecPtr;
8496 0 : SpinLockRelease(&XLogCtl->info_lck);
8497 : }
8498 :
8499 : /* Keep track of full_page_writes */
8500 0 : lastFullPageWrites = fpw;
8501 : }
8502 : else if (info == XLOG_CHECKPOINT_REDO)
8503 : {
8504 : /* nothing to do here, just for informational purposes */
8505 : }
8506 56648 : }
8507 :
8508 : /*
8509 : * Return the extra open flags used for opening a file, depending on the
8510 : * value of the GUCs wal_sync_method, fsync and io_direct.
8511 : */
8512 : static int
8513 15608 : get_sync_bit(int method)
8514 : {
8515 15608 : int o_direct_flag = 0;
8516 :
8517 : /*
8518 : * Use O_DIRECT if requested, except in walreceiver process. The WAL
8519 : * written by walreceiver is normally read by the startup process soon
8520 : * after it's written. Also, walreceiver performs unaligned writes, which
8521 : * don't work with O_DIRECT, so it is required for correctness too.
8522 : */
8523 15608 : if ((io_direct_flags & IO_DIRECT_WAL) && !AmWalReceiverProcess())
8524 14 : o_direct_flag = PG_O_DIRECT;
8525 :
8526 : /* If fsync is disabled, never open in sync mode */
8527 15608 : if (!enableFsync)
8528 15608 : return o_direct_flag;
8529 :
8530 0 : switch (method)
8531 : {
8532 : /*
8533 : * enum values for all sync options are defined even if they are
8534 : * not supported on the current platform. But if not, they are
8535 : * not included in the enum option array, and therefore will never
8536 : * be seen here.
8537 : */
8538 0 : case WAL_SYNC_METHOD_FSYNC:
8539 : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8540 : case WAL_SYNC_METHOD_FDATASYNC:
8541 0 : return o_direct_flag;
8542 : #ifdef O_SYNC
8543 0 : case WAL_SYNC_METHOD_OPEN:
8544 0 : return O_SYNC | o_direct_flag;
8545 : #endif
8546 : #ifdef O_DSYNC
8547 0 : case WAL_SYNC_METHOD_OPEN_DSYNC:
8548 0 : return O_DSYNC | o_direct_flag;
8549 : #endif
8550 0 : default:
8551 : /* can't happen (unless we are out of sync with option array) */
8552 0 : elog(ERROR, "unrecognized wal_sync_method: %d", method);
8553 : return 0; /* silence warning */
8554 : }
8555 : }
8556 :
8557 : /*
8558 : * GUC support
8559 : */
8560 : void
8561 1830 : assign_wal_sync_method(int new_wal_sync_method, void *extra)
8562 : {
8563 1830 : if (wal_sync_method != new_wal_sync_method)
8564 : {
8565 : /*
8566 : * To ensure that no blocks escape unsynced, force an fsync on the
8567 : * currently open log segment (if any). Also, if the open flag is
8568 : * changing, close the log file so it will be reopened (with new flag
8569 : * bit) at next use.
8570 : */
8571 0 : if (openLogFile >= 0)
8572 : {
8573 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
8574 0 : if (pg_fsync(openLogFile) != 0)
8575 : {
8576 : char xlogfname[MAXFNAMELEN];
8577 : int save_errno;
8578 :
8579 0 : save_errno = errno;
8580 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo,
8581 : wal_segment_size);
8582 0 : errno = save_errno;
8583 0 : ereport(PANIC,
8584 : (errcode_for_file_access(),
8585 : errmsg("could not fsync file \"%s\": %m", xlogfname)));
8586 : }
8587 :
8588 0 : pgstat_report_wait_end();
8589 0 : if (get_sync_bit(wal_sync_method) != get_sync_bit(new_wal_sync_method))
8590 0 : XLogFileClose();
8591 : }
8592 : }
8593 1830 : }
8594 :
8595 :
8596 : /*
8597 : * Issue appropriate kind of fsync (if any) for an XLOG output file.
8598 : *
8599 : * 'fd' is a file descriptor for the XLOG file to be fsync'd.
8600 : * 'segno' is for error reporting purposes.
8601 : */
8602 : void
8603 263892 : issue_xlog_fsync(int fd, XLogSegNo segno, TimeLineID tli)
8604 : {
8605 263892 : char *msg = NULL;
8606 : instr_time start;
8607 :
8608 : Assert(tli != 0);
8609 :
8610 : /*
8611 : * Quick exit if fsync is disabled or write() has already synced the WAL
8612 : * file.
8613 : */
8614 263892 : if (!enableFsync ||
8615 0 : wal_sync_method == WAL_SYNC_METHOD_OPEN ||
8616 0 : wal_sync_method == WAL_SYNC_METHOD_OPEN_DSYNC)
8617 263892 : return;
8618 :
8619 : /* Measure I/O timing to sync the WAL file */
8620 0 : if (track_wal_io_timing)
8621 0 : INSTR_TIME_SET_CURRENT(start);
8622 : else
8623 0 : INSTR_TIME_SET_ZERO(start);
8624 :
8625 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
8626 0 : switch (wal_sync_method)
8627 : {
8628 0 : case WAL_SYNC_METHOD_FSYNC:
8629 0 : if (pg_fsync_no_writethrough(fd) != 0)
8630 0 : msg = _("could not fsync file \"%s\": %m");
8631 0 : break;
8632 : #ifdef HAVE_FSYNC_WRITETHROUGH
8633 : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8634 : if (pg_fsync_writethrough(fd) != 0)
8635 : msg = _("could not fsync write-through file \"%s\": %m");
8636 : break;
8637 : #endif
8638 0 : case WAL_SYNC_METHOD_FDATASYNC:
8639 0 : if (pg_fdatasync(fd) != 0)
8640 0 : msg = _("could not fdatasync file \"%s\": %m");
8641 0 : break;
8642 0 : case WAL_SYNC_METHOD_OPEN:
8643 : case WAL_SYNC_METHOD_OPEN_DSYNC:
8644 : /* not reachable */
8645 : Assert(false);
8646 0 : break;
8647 0 : default:
8648 0 : ereport(PANIC,
8649 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8650 : errmsg_internal("unrecognized wal_sync_method: %d", wal_sync_method));
8651 : break;
8652 : }
8653 :
8654 : /* PANIC if failed to fsync */
8655 0 : if (msg)
8656 : {
8657 : char xlogfname[MAXFNAMELEN];
8658 0 : int save_errno = errno;
8659 :
8660 0 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
8661 0 : errno = save_errno;
8662 0 : ereport(PANIC,
8663 : (errcode_for_file_access(),
8664 : errmsg(msg, xlogfname)));
8665 : }
8666 :
8667 0 : pgstat_report_wait_end();
8668 :
8669 : /*
8670 : * Increment the I/O timing and the number of times WAL files were synced.
8671 : */
8672 0 : if (track_wal_io_timing)
8673 : {
8674 : instr_time end;
8675 :
8676 0 : INSTR_TIME_SET_CURRENT(end);
8677 0 : INSTR_TIME_ACCUM_DIFF(PendingWalStats.wal_sync_time, end, start);
8678 : }
8679 :
8680 0 : PendingWalStats.wal_sync++;
8681 : }
8682 :
8683 : /*
8684 : * do_pg_backup_start is the workhorse of the user-visible pg_backup_start()
8685 : * function. It creates the necessary starting checkpoint and constructs the
8686 : * backup state and tablespace map.
8687 : *
8688 : * Input parameters are "state" (the backup state), "fast" (if true, we do
8689 : * the checkpoint in immediate mode to make it faster), and "tablespaces"
8690 : * (if non-NULL, indicates a list of tablespaceinfo structs describing the
8691 : * cluster's tablespaces.).
8692 : *
8693 : * The tablespace map contents are appended to passed-in parameter
8694 : * tablespace_map and the caller is responsible for including it in the backup
8695 : * archive as 'tablespace_map'. The tablespace_map file is required mainly for
8696 : * tar format in windows as native windows utilities are not able to create
8697 : * symlinks while extracting files from tar. However for consistency and
8698 : * platform-independence, we do it the same way everywhere.
8699 : *
8700 : * It fills in "state" with the information required for the backup, such
8701 : * as the minimum WAL location that must be present to restore from this
8702 : * backup (starttli) and the corresponding timeline ID (starttli).
8703 : *
8704 : * Every successfully started backup must be stopped by calling
8705 : * do_pg_backup_stop() or do_pg_abort_backup(). There can be many
8706 : * backups active at the same time.
8707 : *
8708 : * It is the responsibility of the caller of this function to verify the
8709 : * permissions of the calling user!
8710 : */
8711 : void
8712 290 : do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces,
8713 : BackupState *state, StringInfo tblspcmapfile)
8714 : {
8715 : bool backup_started_in_recovery;
8716 :
8717 : Assert(state != NULL);
8718 290 : backup_started_in_recovery = RecoveryInProgress();
8719 :
8720 : /*
8721 : * During recovery, we don't need to check WAL level. Because, if WAL
8722 : * level is not sufficient, it's impossible to get here during recovery.
8723 : */
8724 290 : if (!backup_started_in_recovery && !XLogIsNeeded())
8725 0 : ereport(ERROR,
8726 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8727 : errmsg("WAL level not sufficient for making an online backup"),
8728 : errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
8729 :
8730 290 : if (strlen(backupidstr) > MAXPGPATH)
8731 2 : ereport(ERROR,
8732 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8733 : errmsg("backup label too long (max %d bytes)",
8734 : MAXPGPATH)));
8735 :
8736 288 : memcpy(state->name, backupidstr, strlen(backupidstr));
8737 :
8738 : /*
8739 : * Mark backup active in shared memory. We must do full-page WAL writes
8740 : * during an on-line backup even if not doing so at other times, because
8741 : * it's quite possible for the backup dump to obtain a "torn" (partially
8742 : * written) copy of a database page if it reads the page concurrently with
8743 : * our write to the same page. This can be fixed as long as the first
8744 : * write to the page in the WAL sequence is a full-page write. Hence, we
8745 : * increment runningBackups then force a CHECKPOINT, to ensure there are
8746 : * no dirty pages in shared memory that might get dumped while the backup
8747 : * is in progress without having a corresponding WAL record. (Once the
8748 : * backup is complete, we need not force full-page writes anymore, since
8749 : * we expect that any pages not modified during the backup interval must
8750 : * have been correctly captured by the backup.)
8751 : *
8752 : * Note that forcing full-page writes has no effect during an online
8753 : * backup from the standby.
8754 : *
8755 : * We must hold all the insertion locks to change the value of
8756 : * runningBackups, to ensure adequate interlocking against
8757 : * XLogInsertRecord().
8758 : */
8759 288 : WALInsertLockAcquireExclusive();
8760 288 : XLogCtl->Insert.runningBackups++;
8761 288 : WALInsertLockRelease();
8762 :
8763 : /*
8764 : * Ensure we decrement runningBackups if we fail below. NB -- for this to
8765 : * work correctly, it is critical that sessionBackupState is only updated
8766 : * after this block is over.
8767 : */
8768 288 : PG_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, DatumGetBool(true));
8769 : {
8770 288 : bool gotUniqueStartpoint = false;
8771 : DIR *tblspcdir;
8772 : struct dirent *de;
8773 : tablespaceinfo *ti;
8774 : int datadirpathlen;
8775 :
8776 : /*
8777 : * Force an XLOG file switch before the checkpoint, to ensure that the
8778 : * WAL segment the checkpoint is written to doesn't contain pages with
8779 : * old timeline IDs. That would otherwise happen if you called
8780 : * pg_backup_start() right after restoring from a PITR archive: the
8781 : * first WAL segment containing the startup checkpoint has pages in
8782 : * the beginning with the old timeline ID. That can cause trouble at
8783 : * recovery: we won't have a history file covering the old timeline if
8784 : * pg_wal directory was not included in the base backup and the WAL
8785 : * archive was cleared too before starting the backup.
8786 : *
8787 : * This also ensures that we have emitted a WAL page header that has
8788 : * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
8789 : * Therefore, if a WAL archiver (such as pglesslog) is trying to
8790 : * compress out removable backup blocks, it won't remove any that
8791 : * occur after this point.
8792 : *
8793 : * During recovery, we skip forcing XLOG file switch, which means that
8794 : * the backup taken during recovery is not available for the special
8795 : * recovery case described above.
8796 : */
8797 288 : if (!backup_started_in_recovery)
8798 274 : RequestXLogSwitch(false);
8799 :
8800 : do
8801 : {
8802 : bool checkpointfpw;
8803 :
8804 : /*
8805 : * Force a CHECKPOINT. Aside from being necessary to prevent torn
8806 : * page problems, this guarantees that two successive backup runs
8807 : * will have different checkpoint positions and hence different
8808 : * history file names, even if nothing happened in between.
8809 : *
8810 : * During recovery, establish a restartpoint if possible. We use
8811 : * the last restartpoint as the backup starting checkpoint. This
8812 : * means that two successive backup runs can have same checkpoint
8813 : * positions.
8814 : *
8815 : * Since the fact that we are executing do_pg_backup_start()
8816 : * during recovery means that checkpointer is running, we can use
8817 : * RequestCheckpoint() to establish a restartpoint.
8818 : *
8819 : * We use CHECKPOINT_IMMEDIATE only if requested by user (via
8820 : * passing fast = true). Otherwise this can take awhile.
8821 : */
8822 288 : RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT |
8823 : (fast ? CHECKPOINT_IMMEDIATE : 0));
8824 :
8825 : /*
8826 : * Now we need to fetch the checkpoint record location, and also
8827 : * its REDO pointer. The oldest point in WAL that would be needed
8828 : * to restore starting from the checkpoint is precisely the REDO
8829 : * pointer.
8830 : */
8831 288 : LWLockAcquire(ControlFileLock, LW_SHARED);
8832 288 : state->checkpointloc = ControlFile->checkPoint;
8833 288 : state->startpoint = ControlFile->checkPointCopy.redo;
8834 288 : state->starttli = ControlFile->checkPointCopy.ThisTimeLineID;
8835 288 : checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
8836 288 : LWLockRelease(ControlFileLock);
8837 :
8838 288 : if (backup_started_in_recovery)
8839 : {
8840 : XLogRecPtr recptr;
8841 :
8842 : /*
8843 : * Check to see if all WAL replayed during online backup
8844 : * (i.e., since last restartpoint used as backup starting
8845 : * checkpoint) contain full-page writes.
8846 : */
8847 14 : SpinLockAcquire(&XLogCtl->info_lck);
8848 14 : recptr = XLogCtl->lastFpwDisableRecPtr;
8849 14 : SpinLockRelease(&XLogCtl->info_lck);
8850 :
8851 14 : if (!checkpointfpw || state->startpoint <= recptr)
8852 0 : ereport(ERROR,
8853 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8854 : errmsg("WAL generated with full_page_writes=off was replayed "
8855 : "since last restartpoint"),
8856 : errhint("This means that the backup being taken on the standby "
8857 : "is corrupt and should not be used. "
8858 : "Enable full_page_writes and run CHECKPOINT on the primary, "
8859 : "and then try an online backup again.")));
8860 :
8861 : /*
8862 : * During recovery, since we don't use the end-of-backup WAL
8863 : * record and don't write the backup history file, the
8864 : * starting WAL location doesn't need to be unique. This means
8865 : * that two base backups started at the same time might use
8866 : * the same checkpoint as starting locations.
8867 : */
8868 14 : gotUniqueStartpoint = true;
8869 : }
8870 :
8871 : /*
8872 : * If two base backups are started at the same time (in WAL sender
8873 : * processes), we need to make sure that they use different
8874 : * checkpoints as starting locations, because we use the starting
8875 : * WAL location as a unique identifier for the base backup in the
8876 : * end-of-backup WAL record and when we write the backup history
8877 : * file. Perhaps it would be better generate a separate unique ID
8878 : * for each backup instead of forcing another checkpoint, but
8879 : * taking a checkpoint right after another is not that expensive
8880 : * either because only few buffers have been dirtied yet.
8881 : */
8882 288 : WALInsertLockAcquireExclusive();
8883 288 : if (XLogCtl->Insert.lastBackupStart < state->startpoint)
8884 : {
8885 288 : XLogCtl->Insert.lastBackupStart = state->startpoint;
8886 288 : gotUniqueStartpoint = true;
8887 : }
8888 288 : WALInsertLockRelease();
8889 288 : } while (!gotUniqueStartpoint);
8890 :
8891 : /*
8892 : * Construct tablespace_map file.
8893 : */
8894 288 : datadirpathlen = strlen(DataDir);
8895 :
8896 : /* Collect information about all tablespaces */
8897 288 : tblspcdir = AllocateDir("pg_tblspc");
8898 926 : while ((de = ReadDir(tblspcdir, "pg_tblspc")) != NULL)
8899 : {
8900 : char fullpath[MAXPGPATH + 10];
8901 : char linkpath[MAXPGPATH];
8902 638 : char *relpath = NULL;
8903 : char *s;
8904 : PGFileType de_type;
8905 : char *badp;
8906 : Oid tsoid;
8907 :
8908 : /*
8909 : * Try to parse the directory name as an unsigned integer.
8910 : *
8911 : * Tablespace directories should be positive integers that can be
8912 : * represented in 32 bits, with no leading zeroes or trailing
8913 : * garbage. If we come across a name that doesn't meet those
8914 : * criteria, skip it.
8915 : */
8916 638 : if (de->d_name[0] < '1' || de->d_name[1] > '9')
8917 576 : continue;
8918 62 : errno = 0;
8919 62 : tsoid = strtoul(de->d_name, &badp, 10);
8920 62 : if (*badp != '\0' || errno == EINVAL || errno == ERANGE)
8921 0 : continue;
8922 :
8923 62 : snprintf(fullpath, sizeof(fullpath), "pg_tblspc/%s", de->d_name);
8924 :
8925 62 : de_type = get_dirent_type(fullpath, de, false, ERROR);
8926 :
8927 62 : if (de_type == PGFILETYPE_LNK)
8928 : {
8929 : StringInfoData escapedpath;
8930 : int rllen;
8931 :
8932 34 : rllen = readlink(fullpath, linkpath, sizeof(linkpath));
8933 34 : if (rllen < 0)
8934 : {
8935 0 : ereport(WARNING,
8936 : (errmsg("could not read symbolic link \"%s\": %m",
8937 : fullpath)));
8938 0 : continue;
8939 : }
8940 34 : else if (rllen >= sizeof(linkpath))
8941 : {
8942 0 : ereport(WARNING,
8943 : (errmsg("symbolic link \"%s\" target is too long",
8944 : fullpath)));
8945 0 : continue;
8946 : }
8947 34 : linkpath[rllen] = '\0';
8948 :
8949 : /*
8950 : * Relpath holds the relative path of the tablespace directory
8951 : * when it's located within PGDATA, or NULL if it's located
8952 : * elsewhere.
8953 : */
8954 34 : if (rllen > datadirpathlen &&
8955 2 : strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
8956 0 : IS_DIR_SEP(linkpath[datadirpathlen]))
8957 0 : relpath = pstrdup(linkpath + datadirpathlen + 1);
8958 :
8959 : /*
8960 : * Add a backslash-escaped version of the link path to the
8961 : * tablespace map file.
8962 : */
8963 34 : initStringInfo(&escapedpath);
8964 920 : for (s = linkpath; *s; s++)
8965 : {
8966 886 : if (*s == '\n' || *s == '\r' || *s == '\\')
8967 0 : appendStringInfoChar(&escapedpath, '\\');
8968 886 : appendStringInfoChar(&escapedpath, *s);
8969 : }
8970 34 : appendStringInfo(tblspcmapfile, "%s %s\n",
8971 34 : de->d_name, escapedpath.data);
8972 34 : pfree(escapedpath.data);
8973 : }
8974 28 : else if (de_type == PGFILETYPE_DIR)
8975 : {
8976 : /*
8977 : * It's possible to use allow_in_place_tablespaces to create
8978 : * directories directly under pg_tblspc, for testing purposes
8979 : * only.
8980 : *
8981 : * In this case, we store a relative path rather than an
8982 : * absolute path into the tablespaceinfo.
8983 : */
8984 28 : snprintf(linkpath, sizeof(linkpath), "pg_tblspc/%s",
8985 28 : de->d_name);
8986 28 : relpath = pstrdup(linkpath);
8987 : }
8988 : else
8989 : {
8990 : /* Skip any other file type that appears here. */
8991 0 : continue;
8992 : }
8993 :
8994 62 : ti = palloc(sizeof(tablespaceinfo));
8995 62 : ti->oid = tsoid;
8996 62 : ti->path = pstrdup(linkpath);
8997 62 : ti->rpath = relpath;
8998 62 : ti->size = -1;
8999 :
9000 62 : if (tablespaces)
9001 62 : *tablespaces = lappend(*tablespaces, ti);
9002 : }
9003 288 : FreeDir(tblspcdir);
9004 :
9005 288 : state->starttime = (pg_time_t) time(NULL);
9006 : }
9007 288 : PG_END_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, DatumGetBool(true));
9008 :
9009 288 : state->started_in_recovery = backup_started_in_recovery;
9010 :
9011 : /*
9012 : * Mark that the start phase has correctly finished for the backup.
9013 : */
9014 288 : sessionBackupState = SESSION_BACKUP_RUNNING;
9015 288 : }
9016 :
9017 : /*
9018 : * Utility routine to fetch the session-level status of a backup running.
9019 : */
9020 : SessionBackupState
9021 330 : get_backup_status(void)
9022 : {
9023 330 : return sessionBackupState;
9024 : }
9025 :
9026 : /*
9027 : * do_pg_backup_stop
9028 : *
9029 : * Utility function called at the end of an online backup. It creates history
9030 : * file (if required), resets sessionBackupState and so on. It can optionally
9031 : * wait for WAL segments to be archived.
9032 : *
9033 : * "state" is filled with the information necessary to restore from this
9034 : * backup with its stop LSN (stoppoint), its timeline ID (stoptli), etc.
9035 : *
9036 : * It is the responsibility of the caller of this function to verify the
9037 : * permissions of the calling user!
9038 : */
9039 : void
9040 274 : do_pg_backup_stop(BackupState *state, bool waitforarchive)
9041 : {
9042 274 : bool backup_stopped_in_recovery = false;
9043 : char histfilepath[MAXPGPATH];
9044 : char lastxlogfilename[MAXFNAMELEN];
9045 : char histfilename[MAXFNAMELEN];
9046 : XLogSegNo _logSegNo;
9047 : FILE *fp;
9048 : int seconds_before_warning;
9049 274 : int waits = 0;
9050 274 : bool reported_waiting = false;
9051 :
9052 : Assert(state != NULL);
9053 :
9054 274 : backup_stopped_in_recovery = RecoveryInProgress();
9055 :
9056 : /*
9057 : * During recovery, we don't need to check WAL level. Because, if WAL
9058 : * level is not sufficient, it's impossible to get here during recovery.
9059 : */
9060 274 : if (!backup_stopped_in_recovery && !XLogIsNeeded())
9061 0 : ereport(ERROR,
9062 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9063 : errmsg("WAL level not sufficient for making an online backup"),
9064 : errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
9065 :
9066 : /*
9067 : * OK to update backup counter and session-level lock.
9068 : *
9069 : * Note that CHECK_FOR_INTERRUPTS() must not occur while updating them,
9070 : * otherwise they can be updated inconsistently, which might cause
9071 : * do_pg_abort_backup() to fail.
9072 : */
9073 274 : WALInsertLockAcquireExclusive();
9074 :
9075 : /*
9076 : * It is expected that each do_pg_backup_start() call is matched by
9077 : * exactly one do_pg_backup_stop() call.
9078 : */
9079 : Assert(XLogCtl->Insert.runningBackups > 0);
9080 274 : XLogCtl->Insert.runningBackups--;
9081 :
9082 : /*
9083 : * Clean up session-level lock.
9084 : *
9085 : * You might think that WALInsertLockRelease() can be called before
9086 : * cleaning up session-level lock because session-level lock doesn't need
9087 : * to be protected with WAL insertion lock. But since
9088 : * CHECK_FOR_INTERRUPTS() can occur in it, session-level lock must be
9089 : * cleaned up before it.
9090 : */
9091 274 : sessionBackupState = SESSION_BACKUP_NONE;
9092 :
9093 274 : WALInsertLockRelease();
9094 :
9095 : /*
9096 : * If we are taking an online backup from the standby, we confirm that the
9097 : * standby has not been promoted during the backup.
9098 : */
9099 274 : if (state->started_in_recovery && !backup_stopped_in_recovery)
9100 0 : ereport(ERROR,
9101 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9102 : errmsg("the standby was promoted during online backup"),
9103 : errhint("This means that the backup being taken is corrupt "
9104 : "and should not be used. "
9105 : "Try taking another online backup.")));
9106 :
9107 : /*
9108 : * During recovery, we don't write an end-of-backup record. We assume that
9109 : * pg_control was backed up last and its minimum recovery point can be
9110 : * available as the backup end location. Since we don't have an
9111 : * end-of-backup record, we use the pg_control value to check whether
9112 : * we've reached the end of backup when starting recovery from this
9113 : * backup. We have no way of checking if pg_control wasn't backed up last
9114 : * however.
9115 : *
9116 : * We don't force a switch to new WAL file but it is still possible to
9117 : * wait for all the required files to be archived if waitforarchive is
9118 : * true. This is okay if we use the backup to start a standby and fetch
9119 : * the missing WAL using streaming replication. But in the case of an
9120 : * archive recovery, a user should set waitforarchive to true and wait for
9121 : * them to be archived to ensure that all the required files are
9122 : * available.
9123 : *
9124 : * We return the current minimum recovery point as the backup end
9125 : * location. Note that it can be greater than the exact backup end
9126 : * location if the minimum recovery point is updated after the backup of
9127 : * pg_control. This is harmless for current uses.
9128 : *
9129 : * XXX currently a backup history file is for informational and debug
9130 : * purposes only. It's not essential for an online backup. Furthermore,
9131 : * even if it's created, it will not be archived during recovery because
9132 : * an archiver is not invoked. So it doesn't seem worthwhile to write a
9133 : * backup history file during recovery.
9134 : */
9135 274 : if (backup_stopped_in_recovery)
9136 : {
9137 : XLogRecPtr recptr;
9138 :
9139 : /*
9140 : * Check to see if all WAL replayed during online backup contain
9141 : * full-page writes.
9142 : */
9143 14 : SpinLockAcquire(&XLogCtl->info_lck);
9144 14 : recptr = XLogCtl->lastFpwDisableRecPtr;
9145 14 : SpinLockRelease(&XLogCtl->info_lck);
9146 :
9147 14 : if (state->startpoint <= recptr)
9148 0 : ereport(ERROR,
9149 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9150 : errmsg("WAL generated with full_page_writes=off was replayed "
9151 : "during online backup"),
9152 : errhint("This means that the backup being taken on the standby "
9153 : "is corrupt and should not be used. "
9154 : "Enable full_page_writes and run CHECKPOINT on the primary, "
9155 : "and then try an online backup again.")));
9156 :
9157 :
9158 14 : LWLockAcquire(ControlFileLock, LW_SHARED);
9159 14 : state->stoppoint = ControlFile->minRecoveryPoint;
9160 14 : state->stoptli = ControlFile->minRecoveryPointTLI;
9161 14 : LWLockRelease(ControlFileLock);
9162 : }
9163 : else
9164 : {
9165 : char *history_file;
9166 :
9167 : /*
9168 : * Write the backup-end xlog record
9169 : */
9170 260 : XLogBeginInsert();
9171 260 : XLogRegisterData((char *) (&state->startpoint),
9172 : sizeof(state->startpoint));
9173 260 : state->stoppoint = XLogInsert(RM_XLOG_ID, XLOG_BACKUP_END);
9174 :
9175 : /*
9176 : * Given that we're not in recovery, InsertTimeLineID is set and can't
9177 : * change, so we can read it without a lock.
9178 : */
9179 260 : state->stoptli = XLogCtl->InsertTimeLineID;
9180 :
9181 : /*
9182 : * Force a switch to a new xlog segment file, so that the backup is
9183 : * valid as soon as archiver moves out the current segment file.
9184 : */
9185 260 : RequestXLogSwitch(false);
9186 :
9187 260 : state->stoptime = (pg_time_t) time(NULL);
9188 :
9189 : /*
9190 : * Write the backup history file
9191 : */
9192 260 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9193 260 : BackupHistoryFilePath(histfilepath, state->stoptli, _logSegNo,
9194 : state->startpoint, wal_segment_size);
9195 260 : fp = AllocateFile(histfilepath, "w");
9196 260 : if (!fp)
9197 0 : ereport(ERROR,
9198 : (errcode_for_file_access(),
9199 : errmsg("could not create file \"%s\": %m",
9200 : histfilepath)));
9201 :
9202 : /* Build and save the contents of the backup history file */
9203 260 : history_file = build_backup_content(state, true);
9204 260 : fprintf(fp, "%s", history_file);
9205 260 : pfree(history_file);
9206 :
9207 260 : if (fflush(fp) || ferror(fp) || FreeFile(fp))
9208 0 : ereport(ERROR,
9209 : (errcode_for_file_access(),
9210 : errmsg("could not write file \"%s\": %m",
9211 : histfilepath)));
9212 :
9213 : /*
9214 : * Clean out any no-longer-needed history files. As a side effect,
9215 : * this will post a .ready file for the newly created history file,
9216 : * notifying the archiver that history file may be archived
9217 : * immediately.
9218 : */
9219 260 : CleanupBackupHistory();
9220 : }
9221 :
9222 : /*
9223 : * If archiving is enabled, wait for all the required WAL files to be
9224 : * archived before returning. If archiving isn't enabled, the required WAL
9225 : * needs to be transported via streaming replication (hopefully with
9226 : * wal_keep_size set high enough), or some more exotic mechanism like
9227 : * polling and copying files from pg_wal with script. We have no knowledge
9228 : * of those mechanisms, so it's up to the user to ensure that he gets all
9229 : * the required WAL.
9230 : *
9231 : * We wait until both the last WAL file filled during backup and the
9232 : * history file have been archived, and assume that the alphabetic sorting
9233 : * property of the WAL files ensures any earlier WAL files are safely
9234 : * archived as well.
9235 : *
9236 : * We wait forever, since archive_command is supposed to work and we
9237 : * assume the admin wanted his backup to work completely. If you don't
9238 : * wish to wait, then either waitforarchive should be passed in as false,
9239 : * or you can set statement_timeout. Also, some notices are issued to
9240 : * clue in anyone who might be doing this interactively.
9241 : */
9242 :
9243 274 : if (waitforarchive &&
9244 18 : ((!backup_stopped_in_recovery && XLogArchivingActive()) ||
9245 2 : (backup_stopped_in_recovery && XLogArchivingAlways())))
9246 : {
9247 6 : XLByteToPrevSeg(state->stoppoint, _logSegNo, wal_segment_size);
9248 6 : XLogFileName(lastxlogfilename, state->stoptli, _logSegNo,
9249 : wal_segment_size);
9250 :
9251 6 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9252 6 : BackupHistoryFileName(histfilename, state->stoptli, _logSegNo,
9253 : state->startpoint, wal_segment_size);
9254 :
9255 6 : seconds_before_warning = 60;
9256 6 : waits = 0;
9257 :
9258 18 : while (XLogArchiveIsBusy(lastxlogfilename) ||
9259 6 : XLogArchiveIsBusy(histfilename))
9260 : {
9261 6 : CHECK_FOR_INTERRUPTS();
9262 :
9263 6 : if (!reported_waiting && waits > 5)
9264 : {
9265 0 : ereport(NOTICE,
9266 : (errmsg("base backup done, waiting for required WAL segments to be archived")));
9267 0 : reported_waiting = true;
9268 : }
9269 :
9270 6 : (void) WaitLatch(MyLatch,
9271 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
9272 : 1000L,
9273 : WAIT_EVENT_BACKUP_WAIT_WAL_ARCHIVE);
9274 6 : ResetLatch(MyLatch);
9275 :
9276 6 : if (++waits >= seconds_before_warning)
9277 : {
9278 0 : seconds_before_warning *= 2; /* This wraps in >10 years... */
9279 0 : ereport(WARNING,
9280 : (errmsg("still waiting for all required WAL segments to be archived (%d seconds elapsed)",
9281 : waits),
9282 : errhint("Check that your archive_command is executing properly. "
9283 : "You can safely cancel this backup, "
9284 : "but the database backup will not be usable without all the WAL segments.")));
9285 : }
9286 : }
9287 :
9288 6 : ereport(NOTICE,
9289 : (errmsg("all required WAL segments have been archived")));
9290 : }
9291 268 : else if (waitforarchive)
9292 12 : ereport(NOTICE,
9293 : (errmsg("WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup")));
9294 274 : }
9295 :
9296 :
9297 : /*
9298 : * do_pg_abort_backup: abort a running backup
9299 : *
9300 : * This does just the most basic steps of do_pg_backup_stop(), by taking the
9301 : * system out of backup mode, thus making it a lot more safe to call from
9302 : * an error handler.
9303 : *
9304 : * 'arg' indicates that it's being called during backup setup; so
9305 : * sessionBackupState has not been modified yet, but runningBackups has
9306 : * already been incremented. When it's false, then it's invoked as a
9307 : * before_shmem_exit handler, and therefore we must not change state
9308 : * unless sessionBackupState indicates that a backup is actually running.
9309 : *
9310 : * NB: This gets used as a PG_ENSURE_ERROR_CLEANUP callback and
9311 : * before_shmem_exit handler, hence the odd-looking signature.
9312 : */
9313 : void
9314 20 : do_pg_abort_backup(int code, Datum arg)
9315 : {
9316 20 : bool during_backup_start = DatumGetBool(arg);
9317 :
9318 : /* If called during backup start, there shouldn't be one already running */
9319 : Assert(!during_backup_start || sessionBackupState == SESSION_BACKUP_NONE);
9320 :
9321 20 : if (during_backup_start || sessionBackupState != SESSION_BACKUP_NONE)
9322 : {
9323 14 : WALInsertLockAcquireExclusive();
9324 : Assert(XLogCtl->Insert.runningBackups > 0);
9325 14 : XLogCtl->Insert.runningBackups--;
9326 :
9327 14 : sessionBackupState = SESSION_BACKUP_NONE;
9328 14 : WALInsertLockRelease();
9329 :
9330 14 : if (!during_backup_start)
9331 14 : ereport(WARNING,
9332 : errmsg("aborting backup due to backend exiting before pg_backup_stop was called"));
9333 : }
9334 20 : }
9335 :
9336 : /*
9337 : * Register a handler that will warn about unterminated backups at end of
9338 : * session, unless this has already been done.
9339 : */
9340 : void
9341 10 : register_persistent_abort_backup_handler(void)
9342 : {
9343 : static bool already_done = false;
9344 :
9345 10 : if (already_done)
9346 2 : return;
9347 8 : before_shmem_exit(do_pg_abort_backup, DatumGetBool(false));
9348 8 : already_done = true;
9349 : }
9350 :
9351 : /*
9352 : * Get latest WAL insert pointer
9353 : */
9354 : XLogRecPtr
9355 4706 : GetXLogInsertRecPtr(void)
9356 : {
9357 4706 : XLogCtlInsert *Insert = &XLogCtl->Insert;
9358 : uint64 current_bytepos;
9359 :
9360 4706 : SpinLockAcquire(&Insert->insertpos_lck);
9361 4706 : current_bytepos = Insert->CurrBytePos;
9362 4706 : SpinLockRelease(&Insert->insertpos_lck);
9363 :
9364 4706 : return XLogBytePosToRecPtr(current_bytepos);
9365 : }
9366 :
9367 : /*
9368 : * Get latest WAL write pointer
9369 : */
9370 : XLogRecPtr
9371 2574 : GetXLogWriteRecPtr(void)
9372 : {
9373 2574 : RefreshXLogWriteResult(LogwrtResult);
9374 :
9375 2574 : return LogwrtResult.Write;
9376 : }
9377 :
9378 : /*
9379 : * Returns the redo pointer of the last checkpoint or restartpoint. This is
9380 : * the oldest point in WAL that we still need, if we have to restart recovery.
9381 : */
9382 : void
9383 118 : GetOldestRestartPoint(XLogRecPtr *oldrecptr, TimeLineID *oldtli)
9384 : {
9385 118 : LWLockAcquire(ControlFileLock, LW_SHARED);
9386 118 : *oldrecptr = ControlFile->checkPointCopy.redo;
9387 118 : *oldtli = ControlFile->checkPointCopy.ThisTimeLineID;
9388 118 : LWLockRelease(ControlFileLock);
9389 118 : }
9390 :
9391 : /* Thin wrapper around ShutdownWalRcv(). */
9392 : void
9393 1688 : XLogShutdownWalRcv(void)
9394 : {
9395 1688 : ShutdownWalRcv();
9396 :
9397 1688 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9398 1688 : XLogCtl->InstallXLogFileSegmentActive = false;
9399 1688 : LWLockRelease(ControlFileLock);
9400 1688 : }
9401 :
9402 : /* Enable WAL file recycling and preallocation. */
9403 : void
9404 1772 : SetInstallXLogFileSegmentActive(void)
9405 : {
9406 1772 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9407 1772 : XLogCtl->InstallXLogFileSegmentActive = true;
9408 1772 : LWLockRelease(ControlFileLock);
9409 1772 : }
9410 :
9411 : bool
9412 0 : IsInstallXLogFileSegmentActive(void)
9413 : {
9414 : bool result;
9415 :
9416 0 : LWLockAcquire(ControlFileLock, LW_SHARED);
9417 0 : result = XLogCtl->InstallXLogFileSegmentActive;
9418 0 : LWLockRelease(ControlFileLock);
9419 :
9420 0 : return result;
9421 : }
9422 :
9423 : /*
9424 : * Update the WalWriterSleeping flag.
9425 : */
9426 : void
9427 762 : SetWalWriterSleeping(bool sleeping)
9428 : {
9429 762 : SpinLockAcquire(&XLogCtl->info_lck);
9430 762 : XLogCtl->WalWriterSleeping = sleeping;
9431 762 : SpinLockRelease(&XLogCtl->info_lck);
9432 762 : }
|
