Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * xlogrecovery.c
4 : * Functions for WAL recovery, standby mode
5 : *
6 : * This source file contains functions controlling WAL recovery.
7 : * InitWalRecovery() initializes the system for crash or archive recovery,
8 : * or standby mode, depending on configuration options and the state of
9 : * the control file and possible backup label file. PerformWalRecovery()
10 : * performs the actual WAL replay, calling the rmgr-specific redo routines.
11 : * FinishWalRecovery() performs end-of-recovery checks and cleanup actions,
12 : * and prepares information needed to initialize the WAL for writes. In
13 : * addition to these three main functions, there are a bunch of functions
14 : * for interrogating recovery state and controlling the recovery process.
15 : *
16 : *
17 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
18 : * Portions Copyright (c) 1994, Regents of the University of California
19 : *
20 : * src/backend/access/transam/xlogrecovery.c
21 : *
22 : *-------------------------------------------------------------------------
23 : */
24 :
25 : #include "postgres.h"
26 :
27 : #include <ctype.h>
28 : #include <math.h>
29 : #include <time.h>
30 : #include <sys/stat.h>
31 : #include <sys/time.h>
32 : #include <unistd.h>
33 :
34 : #include "access/timeline.h"
35 : #include "access/transam.h"
36 : #include "access/xact.h"
37 : #include "access/xlog_internal.h"
38 : #include "access/xlogarchive.h"
39 : #include "access/xlogprefetcher.h"
40 : #include "access/xlogreader.h"
41 : #include "access/xlogrecovery.h"
42 : #include "access/xlogutils.h"
43 : #include "backup/basebackup.h"
44 : #include "catalog/pg_control.h"
45 : #include "commands/tablespace.h"
46 : #include "common/file_utils.h"
47 : #include "miscadmin.h"
48 : #include "pgstat.h"
49 : #include "postmaster/bgwriter.h"
50 : #include "postmaster/startup.h"
51 : #include "replication/slot.h"
52 : #include "replication/slotsync.h"
53 : #include "replication/walreceiver.h"
54 : #include "storage/fd.h"
55 : #include "storage/ipc.h"
56 : #include "storage/latch.h"
57 : #include "storage/pmsignal.h"
58 : #include "storage/procarray.h"
59 : #include "storage/spin.h"
60 : #include "utils/datetime.h"
61 : #include "utils/fmgrprotos.h"
62 : #include "utils/guc_hooks.h"
63 : #include "utils/pg_lsn.h"
64 : #include "utils/ps_status.h"
65 : #include "utils/pg_rusage.h"
66 :
67 : /* Unsupported old recovery command file names (relative to $PGDATA) */
68 : #define RECOVERY_COMMAND_FILE "recovery.conf"
69 : #define RECOVERY_COMMAND_DONE "recovery.done"
70 :
71 : /*
72 : * GUC support
73 : */
74 : const struct config_enum_entry recovery_target_action_options[] = {
75 : {"pause", RECOVERY_TARGET_ACTION_PAUSE, false},
76 : {"promote", RECOVERY_TARGET_ACTION_PROMOTE, false},
77 : {"shutdown", RECOVERY_TARGET_ACTION_SHUTDOWN, false},
78 : {NULL, 0, false}
79 : };
80 :
81 : /* options formerly taken from recovery.conf for archive recovery */
82 : char *recoveryRestoreCommand = NULL;
83 : char *recoveryEndCommand = NULL;
84 : char *archiveCleanupCommand = NULL;
85 : RecoveryTargetType recoveryTarget = RECOVERY_TARGET_UNSET;
86 : bool recoveryTargetInclusive = true;
87 : int recoveryTargetAction = RECOVERY_TARGET_ACTION_PAUSE;
88 : TransactionId recoveryTargetXid;
89 : char *recovery_target_time_string;
90 : TimestampTz recoveryTargetTime;
91 : const char *recoveryTargetName;
92 : XLogRecPtr recoveryTargetLSN;
93 : int recovery_min_apply_delay = 0;
94 :
95 : /* options formerly taken from recovery.conf for XLOG streaming */
96 : char *PrimaryConnInfo = NULL;
97 : char *PrimarySlotName = NULL;
98 : bool wal_receiver_create_temp_slot = false;
99 :
100 : /*
101 : * recoveryTargetTimeLineGoal: what the user requested, if any
102 : *
103 : * recoveryTargetTLIRequested: numeric value of requested timeline, if constant
104 : *
105 : * recoveryTargetTLI: the currently understood target timeline; changes
106 : *
107 : * expectedTLEs: a list of TimeLineHistoryEntries for recoveryTargetTLI and
108 : * the timelines of its known parents, newest first (so recoveryTargetTLI is
109 : * always the first list member). Only these TLIs are expected to be seen in
110 : * the WAL segments we read, and indeed only these TLIs will be considered as
111 : * candidate WAL files to open at all.
112 : *
113 : * curFileTLI: the TLI appearing in the name of the current input WAL file.
114 : * (This is not necessarily the same as the timeline from which we are
115 : * replaying WAL, which StartupXLOG calls replayTLI, because we could be
116 : * scanning data that was copied from an ancestor timeline when the current
117 : * file was created.) During a sequential scan we do not allow this value
118 : * to decrease.
119 : */
120 : RecoveryTargetTimeLineGoal recoveryTargetTimeLineGoal = RECOVERY_TARGET_TIMELINE_LATEST;
121 : TimeLineID recoveryTargetTLIRequested = 0;
122 : TimeLineID recoveryTargetTLI = 0;
123 : static List *expectedTLEs;
124 : static TimeLineID curFileTLI;
125 :
126 : /*
127 : * When ArchiveRecoveryRequested is set, archive recovery was requested,
128 : * ie. signal files were present. When InArchiveRecovery is set, we are
129 : * currently recovering using offline XLOG archives. These variables are only
130 : * valid in the startup process.
131 : *
132 : * When ArchiveRecoveryRequested is true, but InArchiveRecovery is false, we're
133 : * currently performing crash recovery using only XLOG files in pg_wal, but
134 : * will switch to using offline XLOG archives as soon as we reach the end of
135 : * WAL in pg_wal.
136 : */
137 : bool ArchiveRecoveryRequested = false;
138 : bool InArchiveRecovery = false;
139 :
140 : /*
141 : * When StandbyModeRequested is set, standby mode was requested, i.e.
142 : * standby.signal file was present. When StandbyMode is set, we are currently
143 : * in standby mode. These variables are only valid in the startup process.
144 : * They work similarly to ArchiveRecoveryRequested and InArchiveRecovery.
145 : */
146 : static bool StandbyModeRequested = false;
147 : bool StandbyMode = false;
148 :
149 : /* was a signal file present at startup? */
150 : static bool standby_signal_file_found = false;
151 : static bool recovery_signal_file_found = false;
152 :
153 : /*
154 : * CheckPointLoc is the position of the checkpoint record that determines
155 : * where to start the replay. It comes from the backup label file or the
156 : * control file.
157 : *
158 : * RedoStartLSN is the checkpoint's REDO location, also from the backup label
159 : * file or the control file. In standby mode, XLOG streaming usually starts
160 : * from the position where an invalid record was found. But if we fail to
161 : * read even the initial checkpoint record, we use the REDO location instead
162 : * of the checkpoint location as the start position of XLOG streaming.
163 : * Otherwise we would have to jump backwards to the REDO location after
164 : * reading the checkpoint record, because the REDO record can precede the
165 : * checkpoint record.
166 : */
167 : static XLogRecPtr CheckPointLoc = InvalidXLogRecPtr;
168 : static TimeLineID CheckPointTLI = 0;
169 : static XLogRecPtr RedoStartLSN = InvalidXLogRecPtr;
170 : static TimeLineID RedoStartTLI = 0;
171 :
172 : /*
173 : * Local copy of SharedHotStandbyActive variable. False actually means "not
174 : * known, need to check the shared state".
175 : */
176 : static bool LocalHotStandbyActive = false;
177 :
178 : /*
179 : * Local copy of SharedPromoteIsTriggered variable. False actually means "not
180 : * known, need to check the shared state".
181 : */
182 : static bool LocalPromoteIsTriggered = false;
183 :
184 : /* Has the recovery code requested a walreceiver wakeup? */
185 : static bool doRequestWalReceiverReply;
186 :
187 : /* XLogReader object used to parse the WAL records */
188 : static XLogReaderState *xlogreader = NULL;
189 :
190 : /* XLogPrefetcher object used to consume WAL records with read-ahead */
191 : static XLogPrefetcher *xlogprefetcher = NULL;
192 :
193 : /* Parameters passed down from ReadRecord to the XLogPageRead callback. */
194 : typedef struct XLogPageReadPrivate
195 : {
196 : int emode;
197 : bool fetching_ckpt; /* are we fetching a checkpoint record? */
198 : bool randAccess;
199 : TimeLineID replayTLI;
200 : } XLogPageReadPrivate;
201 :
202 : /* flag to tell XLogPageRead that we have started replaying */
203 : static bool InRedo = false;
204 :
205 : /*
206 : * Codes indicating where we got a WAL file from during recovery, or where
207 : * to attempt to get one.
208 : */
209 : typedef enum
210 : {
211 : XLOG_FROM_ANY = 0, /* request to read WAL from any source */
212 : XLOG_FROM_ARCHIVE, /* restored using restore_command */
213 : XLOG_FROM_PG_WAL, /* existing file in pg_wal */
214 : XLOG_FROM_STREAM, /* streamed from primary */
215 : } XLogSource;
216 :
217 : /* human-readable names for XLogSources, for debugging output */
218 : static const char *const xlogSourceNames[] = {"any", "archive", "pg_wal", "stream"};
219 :
220 : /*
221 : * readFile is -1 or a kernel FD for the log file segment that's currently
222 : * open for reading. readSegNo identifies the segment. readOff is the offset
223 : * of the page just read, readLen indicates how much of it has been read into
224 : * readBuf, and readSource indicates where we got the currently open file from.
225 : *
226 : * Note: we could use Reserve/ReleaseExternalFD to track consumption of this
227 : * FD too (like for openLogFile in xlog.c); but it doesn't currently seem
228 : * worthwhile, since the XLOG is not read by general-purpose sessions.
229 : */
230 : static int readFile = -1;
231 : static XLogSegNo readSegNo = 0;
232 : static uint32 readOff = 0;
233 : static uint32 readLen = 0;
234 : static XLogSource readSource = XLOG_FROM_ANY;
235 :
236 : /*
237 : * Keeps track of which source we're currently reading from. This is
238 : * different from readSource in that this is always set, even when we don't
239 : * currently have a WAL file open. If lastSourceFailed is set, our last
240 : * attempt to read from currentSource failed, and we should try another source
241 : * next.
242 : *
243 : * pendingWalRcvRestart is set when a config change occurs that requires a
244 : * walreceiver restart. This is only valid in XLOG_FROM_STREAM state.
245 : */
246 : static XLogSource currentSource = XLOG_FROM_ANY;
247 : static bool lastSourceFailed = false;
248 : static bool pendingWalRcvRestart = false;
249 :
250 : /*
251 : * These variables track when we last obtained some WAL data to process,
252 : * and where we got it from. (XLogReceiptSource is initially the same as
253 : * readSource, but readSource gets reset to zero when we don't have data
254 : * to process right now. It is also different from currentSource, which
255 : * also changes when we try to read from a source and fail, while
256 : * XLogReceiptSource tracks where we last successfully read some WAL.)
257 : */
258 : static TimestampTz XLogReceiptTime = 0;
259 : static XLogSource XLogReceiptSource = XLOG_FROM_ANY;
260 :
261 : /* Local copy of WalRcv->flushedUpto */
262 : static XLogRecPtr flushedUpto = 0;
263 : static TimeLineID receiveTLI = 0;
264 :
265 : /*
266 : * Copy of minRecoveryPoint and backupEndPoint from the control file.
267 : *
268 : * In order to reach consistency, we must replay the WAL up to
269 : * minRecoveryPoint. If backupEndRequired is true, we must also reach
270 : * backupEndPoint, or if it's invalid, an end-of-backup record corresponding
271 : * to backupStartPoint.
272 : *
273 : * Note: In archive recovery, after consistency has been reached, the
274 : * functions in xlog.c will start updating minRecoveryPoint in the control
275 : * file. But this copy of minRecoveryPoint variable reflects the value at the
276 : * beginning of recovery, and is *not* updated after consistency is reached.
277 : */
278 : static XLogRecPtr minRecoveryPoint;
279 : static TimeLineID minRecoveryPointTLI;
280 :
281 : static XLogRecPtr backupStartPoint;
282 : static XLogRecPtr backupEndPoint;
283 : static bool backupEndRequired = false;
284 :
285 : /*
286 : * Have we reached a consistent database state? In crash recovery, we have
287 : * to replay all the WAL, so reachedConsistency is never set. During archive
288 : * recovery, the database is consistent once minRecoveryPoint is reached.
289 : *
290 : * Consistent state means that the system is internally consistent, all
291 : * the WAL has been replayed up to a certain point, and importantly, there
292 : * is no trace of later actions on disk.
293 : */
294 : bool reachedConsistency = false;
295 :
296 : /* Buffers dedicated to consistency checks of size BLCKSZ */
297 : static char *replay_image_masked = NULL;
298 : static char *primary_image_masked = NULL;
299 :
300 :
301 : /*
302 : * Shared-memory state for WAL recovery.
303 : */
304 : typedef struct XLogRecoveryCtlData
305 : {
306 : /*
307 : * SharedHotStandbyActive indicates if we allow hot standby queries to be
308 : * run. Protected by info_lck.
309 : */
310 : bool SharedHotStandbyActive;
311 :
312 : /*
313 : * SharedPromoteIsTriggered indicates if a standby promotion has been
314 : * triggered. Protected by info_lck.
315 : */
316 : bool SharedPromoteIsTriggered;
317 :
318 : /*
319 : * recoveryWakeupLatch is used to wake up the startup process to continue
320 : * WAL replay, if it is waiting for WAL to arrive or promotion to be
321 : * requested.
322 : *
323 : * Note that the startup process also uses another latch, its procLatch,
324 : * to wait for recovery conflict. If we get rid of recoveryWakeupLatch for
325 : * signaling the startup process in favor of using its procLatch, which
326 : * comports better with possible generic signal handlers using that latch.
327 : * But we should not do that because the startup process doesn't assume
328 : * that it's waken up by walreceiver process or SIGHUP signal handler
329 : * while it's waiting for recovery conflict. The separate latches,
330 : * recoveryWakeupLatch and procLatch, should be used for inter-process
331 : * communication for WAL replay and recovery conflict, respectively.
332 : */
333 : Latch recoveryWakeupLatch;
334 :
335 : /*
336 : * Last record successfully replayed.
337 : */
338 : XLogRecPtr lastReplayedReadRecPtr; /* start position */
339 : XLogRecPtr lastReplayedEndRecPtr; /* end+1 position */
340 : TimeLineID lastReplayedTLI; /* timeline */
341 :
342 : /*
343 : * When we're currently replaying a record, ie. in a redo function,
344 : * replayEndRecPtr points to the end+1 of the record being replayed,
345 : * otherwise it's equal to lastReplayedEndRecPtr.
346 : */
347 : XLogRecPtr replayEndRecPtr;
348 : TimeLineID replayEndTLI;
349 : /* timestamp of last COMMIT/ABORT record replayed (or being replayed) */
350 : TimestampTz recoveryLastXTime;
351 :
352 : /*
353 : * timestamp of when we started replaying the current chunk of WAL data,
354 : * only relevant for replication or archive recovery
355 : */
356 : TimestampTz currentChunkStartTime;
357 : /* Recovery pause state */
358 : RecoveryPauseState recoveryPauseState;
359 : ConditionVariable recoveryNotPausedCV;
360 :
361 : slock_t info_lck; /* locks shared variables shown above */
362 : } XLogRecoveryCtlData;
363 :
364 : static XLogRecoveryCtlData *XLogRecoveryCtl = NULL;
365 :
366 : /*
367 : * abortedRecPtr is the start pointer of a broken record at end of WAL when
368 : * recovery completes; missingContrecPtr is the location of the first
369 : * contrecord that went missing. See CreateOverwriteContrecordRecord for
370 : * details.
371 : */
372 : static XLogRecPtr abortedRecPtr;
373 : static XLogRecPtr missingContrecPtr;
374 :
375 : /*
376 : * if recoveryStopsBefore/After returns true, it saves information of the stop
377 : * point here
378 : */
379 : static TransactionId recoveryStopXid;
380 : static TimestampTz recoveryStopTime;
381 : static XLogRecPtr recoveryStopLSN;
382 : static char recoveryStopName[MAXFNAMELEN];
383 : static bool recoveryStopAfter;
384 :
385 : /* prototypes for local functions */
386 : static void ApplyWalRecord(XLogReaderState *xlogreader, XLogRecord *record, TimeLineID *replayTLI);
387 :
388 : static void EnableStandbyMode(void);
389 : static void readRecoverySignalFile(void);
390 : static void validateRecoveryParameters(void);
391 : static bool read_backup_label(XLogRecPtr *checkPointLoc,
392 : TimeLineID *backupLabelTLI,
393 : bool *backupEndRequired, bool *backupFromStandby);
394 : static bool read_tablespace_map(List **tablespaces);
395 :
396 : static void xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI);
397 : static void CheckRecoveryConsistency(void);
398 : static void rm_redo_error_callback(void *arg);
399 : #ifdef WAL_DEBUG
400 : static void xlog_outrec(StringInfo buf, XLogReaderState *record);
401 : #endif
402 : static void xlog_block_info(StringInfo buf, XLogReaderState *record);
403 : static void checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI,
404 : TimeLineID prevTLI, TimeLineID replayTLI);
405 : static bool getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime);
406 : static void verifyBackupPageConsistency(XLogReaderState *record);
407 :
408 : static bool recoveryStopsBefore(XLogReaderState *record);
409 : static bool recoveryStopsAfter(XLogReaderState *record);
410 : static char *getRecoveryStopReason(void);
411 : static void recoveryPausesHere(bool endOfRecovery);
412 : static bool recoveryApplyDelay(XLogReaderState *record);
413 : static void ConfirmRecoveryPaused(void);
414 :
415 : static XLogRecord *ReadRecord(XLogPrefetcher *xlogprefetcher,
416 : int emode, bool fetching_ckpt,
417 : TimeLineID replayTLI);
418 :
419 : static int XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr,
420 : int reqLen, XLogRecPtr targetRecPtr, char *readBuf);
421 : static XLogPageReadResult WaitForWALToBecomeAvailable(XLogRecPtr RecPtr,
422 : bool randAccess,
423 : bool fetching_ckpt,
424 : XLogRecPtr tliRecPtr,
425 : TimeLineID replayTLI,
426 : XLogRecPtr replayLSN,
427 : bool nonblocking);
428 : static int emode_for_corrupt_record(int emode, XLogRecPtr RecPtr);
429 : static XLogRecord *ReadCheckpointRecord(XLogPrefetcher *xlogprefetcher,
430 : XLogRecPtr RecPtr, TimeLineID replayTLI);
431 : static bool rescanLatestTimeLine(TimeLineID replayTLI, XLogRecPtr replayLSN);
432 : static int XLogFileRead(XLogSegNo segno, int emode, TimeLineID tli,
433 : XLogSource source, bool notfoundOk);
434 : static int XLogFileReadAnyTLI(XLogSegNo segno, int emode, XLogSource source);
435 :
436 : static bool CheckForStandbyTrigger(void);
437 : static void SetPromoteIsTriggered(void);
438 : static bool HotStandbyActiveInReplay(void);
439 :
440 : static void SetCurrentChunkStartTime(TimestampTz xtime);
441 : static void SetLatestXTime(TimestampTz xtime);
442 :
443 : /*
444 : * Initialization of shared memory for WAL recovery
445 : */
446 : Size
447 5066 : XLogRecoveryShmemSize(void)
448 : {
449 : Size size;
450 :
451 : /* XLogRecoveryCtl */
452 5066 : size = sizeof(XLogRecoveryCtlData);
453 :
454 5066 : return size;
455 : }
456 :
457 : void
458 1768 : XLogRecoveryShmemInit(void)
459 : {
460 : bool found;
461 :
462 1768 : XLogRecoveryCtl = (XLogRecoveryCtlData *)
463 1768 : ShmemInitStruct("XLOG Recovery Ctl", XLogRecoveryShmemSize(), &found);
464 1768 : if (found)
465 0 : return;
466 1768 : memset(XLogRecoveryCtl, 0, sizeof(XLogRecoveryCtlData));
467 :
468 1768 : SpinLockInit(&XLogRecoveryCtl->info_lck);
469 1768 : InitSharedLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
470 1768 : ConditionVariableInit(&XLogRecoveryCtl->recoveryNotPausedCV);
471 : }
472 :
473 : /*
474 : * A thin wrapper to enable StandbyMode and do other preparatory work as
475 : * needed.
476 : */
477 : static void
478 180 : EnableStandbyMode(void)
479 : {
480 180 : StandbyMode = true;
481 :
482 : /*
483 : * To avoid server log bloat, we don't report recovery progress in a
484 : * standby as it will always be in recovery unless promoted. We disable
485 : * startup progress timeout in standby mode to avoid calling
486 : * startup_progress_timeout_handler() unnecessarily.
487 : */
488 180 : disable_startup_progress_timeout();
489 180 : }
490 :
491 : /*
492 : * Prepare the system for WAL recovery, if needed.
493 : *
494 : * This is called by StartupXLOG() which coordinates the server startup
495 : * sequence. This function analyzes the control file and the backup label
496 : * file, if any, and figures out whether we need to perform crash recovery or
497 : * archive recovery, and how far we need to replay the WAL to reach a
498 : * consistent state.
499 : *
500 : * This doesn't yet change the on-disk state, except for creating the symlinks
501 : * from table space map file if any, and for fetching WAL files needed to find
502 : * the checkpoint record. On entry, the caller has already read the control
503 : * file into memory, and passes it as argument. This function updates it to
504 : * reflect the recovery state, and the caller is expected to write it back to
505 : * disk does after initializing other subsystems, but before calling
506 : * PerformWalRecovery().
507 : *
508 : * This initializes some global variables like ArchiveRecoveryRequested, and
509 : * StandbyModeRequested and InRecovery.
510 : */
511 : void
512 1520 : InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr,
513 : bool *haveBackupLabel_ptr, bool *haveTblspcMap_ptr)
514 : {
515 : XLogPageReadPrivate *private;
516 : struct stat st;
517 : bool wasShutdown;
518 : XLogRecord *record;
519 : DBState dbstate_at_startup;
520 1520 : bool haveTblspcMap = false;
521 1520 : bool haveBackupLabel = false;
522 : CheckPoint checkPoint;
523 1520 : bool backupFromStandby = false;
524 :
525 1520 : dbstate_at_startup = ControlFile->state;
526 :
527 : /*
528 : * Initialize on the assumption we want to recover to the latest timeline
529 : * that's active according to pg_control.
530 : */
531 1520 : if (ControlFile->minRecoveryPointTLI >
532 1520 : ControlFile->checkPointCopy.ThisTimeLineID)
533 2 : recoveryTargetTLI = ControlFile->minRecoveryPointTLI;
534 : else
535 1518 : recoveryTargetTLI = ControlFile->checkPointCopy.ThisTimeLineID;
536 :
537 : /*
538 : * Check for signal files, and if so set up state for offline recovery
539 : */
540 1520 : readRecoverySignalFile();
541 1520 : validateRecoveryParameters();
542 :
543 : /*
544 : * Take ownership of the wakeup latch if we're going to sleep during
545 : * recovery, if required.
546 : */
547 1520 : if (ArchiveRecoveryRequested)
548 190 : OwnLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
549 :
550 : /*
551 : * Set the WAL reading processor now, as it will be needed when reading
552 : * the checkpoint record required (backup_label or not).
553 : */
554 1520 : private = palloc0(sizeof(XLogPageReadPrivate));
555 1520 : xlogreader =
556 1520 : XLogReaderAllocate(wal_segment_size, NULL,
557 1520 : XL_ROUTINE(.page_read = &XLogPageRead,
558 : .segment_open = NULL,
559 : .segment_close = wal_segment_close),
560 : private);
561 1520 : if (!xlogreader)
562 0 : ereport(ERROR,
563 : (errcode(ERRCODE_OUT_OF_MEMORY),
564 : errmsg("out of memory"),
565 : errdetail("Failed while allocating a WAL reading processor.")));
566 1520 : xlogreader->system_identifier = ControlFile->system_identifier;
567 :
568 : /*
569 : * Set the WAL decode buffer size. This limits how far ahead we can read
570 : * in the WAL.
571 : */
572 1520 : XLogReaderSetDecodeBuffer(xlogreader, NULL, wal_decode_buffer_size);
573 :
574 : /* Create a WAL prefetcher. */
575 1520 : xlogprefetcher = XLogPrefetcherAllocate(xlogreader);
576 :
577 : /*
578 : * Allocate two page buffers dedicated to WAL consistency checks. We do
579 : * it this way, rather than just making static arrays, for two reasons:
580 : * (1) no need to waste the storage in most instantiations of the backend;
581 : * (2) a static char array isn't guaranteed to have any particular
582 : * alignment, whereas palloc() will provide MAXALIGN'd storage.
583 : */
584 1520 : replay_image_masked = (char *) palloc(BLCKSZ);
585 1520 : primary_image_masked = (char *) palloc(BLCKSZ);
586 :
587 : /*
588 : * Read the backup_label file. We want to run this part of the recovery
589 : * process after checking for signal files and after performing validation
590 : * of the recovery parameters.
591 : */
592 1520 : if (read_backup_label(&CheckPointLoc, &CheckPointTLI, &backupEndRequired,
593 : &backupFromStandby))
594 : {
595 126 : List *tablespaces = NIL;
596 :
597 : /*
598 : * Archive recovery was requested, and thanks to the backup label
599 : * file, we know how far we need to replay to reach consistency. Enter
600 : * archive recovery directly.
601 : */
602 126 : InArchiveRecovery = true;
603 126 : if (StandbyModeRequested)
604 106 : EnableStandbyMode();
605 :
606 : /*
607 : * Omitting backup_label when creating a new replica, PITR node etc.
608 : * unfortunately is a common cause of corruption. Logging that
609 : * backup_label was used makes it a bit easier to exclude that as the
610 : * cause of observed corruption.
611 : *
612 : * Do so before we try to read the checkpoint record (which can fail),
613 : * as otherwise it can be hard to understand why a checkpoint other
614 : * than ControlFile->checkPoint is used.
615 : */
616 126 : ereport(LOG,
617 : (errmsg("starting backup recovery with redo LSN %X/%X, checkpoint LSN %X/%X, on timeline ID %u",
618 : LSN_FORMAT_ARGS(RedoStartLSN),
619 : LSN_FORMAT_ARGS(CheckPointLoc),
620 : CheckPointTLI)));
621 :
622 : /*
623 : * When a backup_label file is present, we want to roll forward from
624 : * the checkpoint it identifies, rather than using pg_control.
625 : */
626 126 : record = ReadCheckpointRecord(xlogprefetcher, CheckPointLoc,
627 : CheckPointTLI);
628 126 : if (record != NULL)
629 : {
630 126 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
631 126 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
632 126 : ereport(DEBUG1,
633 : (errmsg_internal("checkpoint record is at %X/%X",
634 : LSN_FORMAT_ARGS(CheckPointLoc))));
635 126 : InRecovery = true; /* force recovery even if SHUTDOWNED */
636 :
637 : /*
638 : * Make sure that REDO location exists. This may not be the case
639 : * if there was a crash during an online backup, which left a
640 : * backup_label around that references a WAL segment that's
641 : * already been archived.
642 : */
643 126 : if (checkPoint.redo < CheckPointLoc)
644 : {
645 126 : XLogPrefetcherBeginRead(xlogprefetcher, checkPoint.redo);
646 126 : if (!ReadRecord(xlogprefetcher, LOG, false,
647 : checkPoint.ThisTimeLineID))
648 0 : ereport(FATAL,
649 : (errmsg("could not find redo location %X/%X referenced by checkpoint record at %X/%X",
650 : LSN_FORMAT_ARGS(checkPoint.redo), LSN_FORMAT_ARGS(CheckPointLoc)),
651 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" or \"%s/standby.signal\" and add required recovery options.\n"
652 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
653 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
654 : DataDir, DataDir, DataDir, DataDir)));
655 : }
656 : }
657 : else
658 : {
659 0 : ereport(FATAL,
660 : (errmsg("could not locate required checkpoint record at %X/%X",
661 : LSN_FORMAT_ARGS(CheckPointLoc)),
662 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" or \"%s/standby.signal\" and add required recovery options.\n"
663 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
664 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
665 : DataDir, DataDir, DataDir, DataDir)));
666 : wasShutdown = false; /* keep compiler quiet */
667 : }
668 :
669 : /* Read the tablespace_map file if present and create symlinks. */
670 126 : if (read_tablespace_map(&tablespaces))
671 : {
672 : ListCell *lc;
673 :
674 4 : foreach(lc, tablespaces)
675 : {
676 2 : tablespaceinfo *ti = lfirst(lc);
677 : char *linkloc;
678 :
679 2 : linkloc = psprintf("pg_tblspc/%u", ti->oid);
680 :
681 : /*
682 : * Remove the existing symlink if any and Create the symlink
683 : * under PGDATA.
684 : */
685 2 : remove_tablespace_symlink(linkloc);
686 :
687 2 : if (symlink(ti->path, linkloc) < 0)
688 0 : ereport(ERROR,
689 : (errcode_for_file_access(),
690 : errmsg("could not create symbolic link \"%s\": %m",
691 : linkloc)));
692 :
693 2 : pfree(ti->path);
694 2 : pfree(ti);
695 : }
696 :
697 : /* tell the caller to delete it later */
698 2 : haveTblspcMap = true;
699 : }
700 :
701 : /* tell the caller to delete it later */
702 126 : haveBackupLabel = true;
703 : }
704 : else
705 : {
706 : /* No backup_label file has been found if we are here. */
707 :
708 : /*
709 : * If tablespace_map file is present without backup_label file, there
710 : * is no use of such file. There is no harm in retaining it, but it
711 : * is better to get rid of the map file so that we don't have any
712 : * redundant file in data directory and it will avoid any sort of
713 : * confusion. It seems prudent though to just rename the file out of
714 : * the way rather than delete it completely, also we ignore any error
715 : * that occurs in rename operation as even if map file is present
716 : * without backup_label file, it is harmless.
717 : */
718 1394 : if (stat(TABLESPACE_MAP, &st) == 0)
719 : {
720 2 : unlink(TABLESPACE_MAP_OLD);
721 2 : if (durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, DEBUG1) == 0)
722 2 : ereport(LOG,
723 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
724 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
725 : errdetail("File \"%s\" was renamed to \"%s\".",
726 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
727 : else
728 0 : ereport(LOG,
729 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
730 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
731 : errdetail("Could not rename file \"%s\" to \"%s\": %m.",
732 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
733 : }
734 :
735 : /*
736 : * It's possible that archive recovery was requested, but we don't
737 : * know how far we need to replay the WAL before we reach consistency.
738 : * This can happen for example if a base backup is taken from a
739 : * running server using an atomic filesystem snapshot, without calling
740 : * pg_backup_start/stop. Or if you just kill a running primary server
741 : * and put it into archive recovery by creating a recovery signal
742 : * file.
743 : *
744 : * Our strategy in that case is to perform crash recovery first,
745 : * replaying all the WAL present in pg_wal, and only enter archive
746 : * recovery after that.
747 : *
748 : * But usually we already know how far we need to replay the WAL (up
749 : * to minRecoveryPoint, up to backupEndPoint, or until we see an
750 : * end-of-backup record), and we can enter archive recovery directly.
751 : */
752 1394 : if (ArchiveRecoveryRequested &&
753 74 : (ControlFile->minRecoveryPoint != InvalidXLogRecPtr ||
754 18 : ControlFile->backupEndRequired ||
755 18 : ControlFile->backupEndPoint != InvalidXLogRecPtr ||
756 18 : ControlFile->state == DB_SHUTDOWNED))
757 : {
758 70 : InArchiveRecovery = true;
759 70 : if (StandbyModeRequested)
760 70 : EnableStandbyMode();
761 : }
762 :
763 : /*
764 : * For the same reason as when starting up with backup_label present,
765 : * emit a log message when we continue initializing from a base
766 : * backup.
767 : */
768 1394 : if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
769 0 : ereport(LOG,
770 : (errmsg("restarting backup recovery with redo LSN %X/%X",
771 : LSN_FORMAT_ARGS(ControlFile->backupStartPoint))));
772 :
773 : /* Get the last valid checkpoint record. */
774 1394 : CheckPointLoc = ControlFile->checkPoint;
775 1394 : CheckPointTLI = ControlFile->checkPointCopy.ThisTimeLineID;
776 1394 : RedoStartLSN = ControlFile->checkPointCopy.redo;
777 1394 : RedoStartTLI = ControlFile->checkPointCopy.ThisTimeLineID;
778 1394 : record = ReadCheckpointRecord(xlogprefetcher, CheckPointLoc,
779 : CheckPointTLI);
780 1394 : if (record != NULL)
781 : {
782 1394 : ereport(DEBUG1,
783 : (errmsg_internal("checkpoint record is at %X/%X",
784 : LSN_FORMAT_ARGS(CheckPointLoc))));
785 : }
786 : else
787 : {
788 : /*
789 : * We used to attempt to go back to a secondary checkpoint record
790 : * here, but only when not in standby mode. We now just fail if we
791 : * can't read the last checkpoint because this allows us to
792 : * simplify processing around checkpoints.
793 : */
794 0 : ereport(PANIC,
795 : (errmsg("could not locate a valid checkpoint record at %X/%X",
796 : LSN_FORMAT_ARGS(CheckPointLoc))));
797 : }
798 1394 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
799 1394 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
800 : }
801 :
802 1520 : if (ArchiveRecoveryRequested)
803 : {
804 190 : if (StandbyModeRequested)
805 180 : ereport(LOG,
806 : (errmsg("entering standby mode")));
807 10 : else if (recoveryTarget == RECOVERY_TARGET_XID)
808 0 : ereport(LOG,
809 : (errmsg("starting point-in-time recovery to XID %u",
810 : recoveryTargetXid)));
811 10 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
812 0 : ereport(LOG,
813 : (errmsg("starting point-in-time recovery to %s",
814 : timestamptz_to_str(recoveryTargetTime))));
815 10 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
816 6 : ereport(LOG,
817 : (errmsg("starting point-in-time recovery to \"%s\"",
818 : recoveryTargetName)));
819 4 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
820 0 : ereport(LOG,
821 : (errmsg("starting point-in-time recovery to WAL location (LSN) \"%X/%X\"",
822 : LSN_FORMAT_ARGS(recoveryTargetLSN))));
823 4 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
824 0 : ereport(LOG,
825 : (errmsg("starting point-in-time recovery to earliest consistent point")));
826 : else
827 4 : ereport(LOG,
828 : (errmsg("starting archive recovery")));
829 : }
830 :
831 : /*
832 : * If the location of the checkpoint record is not on the expected
833 : * timeline in the history of the requested timeline, we cannot proceed:
834 : * the backup is not part of the history of the requested timeline.
835 : */
836 : Assert(expectedTLEs); /* was initialized by reading checkpoint
837 : * record */
838 1520 : if (tliOfPointInHistory(CheckPointLoc, expectedTLEs) !=
839 : CheckPointTLI)
840 : {
841 : XLogRecPtr switchpoint;
842 :
843 : /*
844 : * tliSwitchPoint will throw an error if the checkpoint's timeline is
845 : * not in expectedTLEs at all.
846 : */
847 0 : switchpoint = tliSwitchPoint(ControlFile->checkPointCopy.ThisTimeLineID, expectedTLEs, NULL);
848 0 : ereport(FATAL,
849 : (errmsg("requested timeline %u is not a child of this server's history",
850 : recoveryTargetTLI),
851 : errdetail("Latest checkpoint is at %X/%X on timeline %u, but in the history of the requested timeline, the server forked off from that timeline at %X/%X.",
852 : LSN_FORMAT_ARGS(ControlFile->checkPoint),
853 : ControlFile->checkPointCopy.ThisTimeLineID,
854 : LSN_FORMAT_ARGS(switchpoint))));
855 : }
856 :
857 : /*
858 : * The min recovery point should be part of the requested timeline's
859 : * history, too.
860 : */
861 1520 : if (!XLogRecPtrIsInvalid(ControlFile->minRecoveryPoint) &&
862 66 : tliOfPointInHistory(ControlFile->minRecoveryPoint - 1, expectedTLEs) !=
863 66 : ControlFile->minRecoveryPointTLI)
864 0 : ereport(FATAL,
865 : (errmsg("requested timeline %u does not contain minimum recovery point %X/%X on timeline %u",
866 : recoveryTargetTLI,
867 : LSN_FORMAT_ARGS(ControlFile->minRecoveryPoint),
868 : ControlFile->minRecoveryPointTLI)));
869 :
870 1520 : ereport(DEBUG1,
871 : (errmsg_internal("redo record is at %X/%X; shutdown %s",
872 : LSN_FORMAT_ARGS(checkPoint.redo),
873 : wasShutdown ? "true" : "false")));
874 1520 : ereport(DEBUG1,
875 : (errmsg_internal("next transaction ID: " UINT64_FORMAT "; next OID: %u",
876 : U64FromFullTransactionId(checkPoint.nextXid),
877 : checkPoint.nextOid)));
878 1520 : ereport(DEBUG1,
879 : (errmsg_internal("next MultiXactId: %u; next MultiXactOffset: %u",
880 : checkPoint.nextMulti, checkPoint.nextMultiOffset)));
881 1520 : ereport(DEBUG1,
882 : (errmsg_internal("oldest unfrozen transaction ID: %u, in database %u",
883 : checkPoint.oldestXid, checkPoint.oldestXidDB)));
884 1520 : ereport(DEBUG1,
885 : (errmsg_internal("oldest MultiXactId: %u, in database %u",
886 : checkPoint.oldestMulti, checkPoint.oldestMultiDB)));
887 1520 : ereport(DEBUG1,
888 : (errmsg_internal("commit timestamp Xid oldest/newest: %u/%u",
889 : checkPoint.oldestCommitTsXid,
890 : checkPoint.newestCommitTsXid)));
891 1520 : if (!TransactionIdIsNormal(XidFromFullTransactionId(checkPoint.nextXid)))
892 0 : ereport(PANIC,
893 : (errmsg("invalid next transaction ID")));
894 :
895 : /* sanity check */
896 1520 : if (checkPoint.redo > CheckPointLoc)
897 0 : ereport(PANIC,
898 : (errmsg("invalid redo in checkpoint record")));
899 :
900 : /*
901 : * Check whether we need to force recovery from WAL. If it appears to
902 : * have been a clean shutdown and we did not have a recovery signal file,
903 : * then assume no recovery needed.
904 : */
905 1520 : if (checkPoint.redo < CheckPointLoc)
906 : {
907 200 : if (wasShutdown)
908 0 : ereport(PANIC,
909 : (errmsg("invalid redo record in shutdown checkpoint")));
910 200 : InRecovery = true;
911 : }
912 1320 : else if (ControlFile->state != DB_SHUTDOWNED)
913 178 : InRecovery = true;
914 1142 : else if (ArchiveRecoveryRequested)
915 : {
916 : /* force recovery due to presence of recovery signal file */
917 14 : InRecovery = true;
918 : }
919 :
920 : /*
921 : * If recovery is needed, update our in-memory copy of pg_control to show
922 : * that we are recovering and to show the selected checkpoint as the place
923 : * we are starting from. We also mark pg_control with any minimum recovery
924 : * stop point obtained from a backup history file.
925 : *
926 : * We don't write the changes to disk yet, though. Only do that after
927 : * initializing various subsystems.
928 : */
929 1520 : if (InRecovery)
930 : {
931 392 : if (InArchiveRecovery)
932 : {
933 196 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
934 : }
935 : else
936 : {
937 196 : ereport(LOG,
938 : (errmsg("database system was not properly shut down; "
939 : "automatic recovery in progress")));
940 196 : if (recoveryTargetTLI > ControlFile->checkPointCopy.ThisTimeLineID)
941 2 : ereport(LOG,
942 : (errmsg("crash recovery starts in timeline %u "
943 : "and has target timeline %u",
944 : ControlFile->checkPointCopy.ThisTimeLineID,
945 : recoveryTargetTLI)));
946 196 : ControlFile->state = DB_IN_CRASH_RECOVERY;
947 : }
948 392 : ControlFile->checkPoint = CheckPointLoc;
949 392 : ControlFile->checkPointCopy = checkPoint;
950 392 : if (InArchiveRecovery)
951 : {
952 : /* initialize minRecoveryPoint if not set yet */
953 196 : if (ControlFile->minRecoveryPoint < checkPoint.redo)
954 : {
955 132 : ControlFile->minRecoveryPoint = checkPoint.redo;
956 132 : ControlFile->minRecoveryPointTLI = checkPoint.ThisTimeLineID;
957 : }
958 : }
959 :
960 : /*
961 : * Set backupStartPoint if we're starting recovery from a base backup.
962 : *
963 : * Also set backupEndPoint and use minRecoveryPoint as the backup end
964 : * location if we're starting recovery from a base backup which was
965 : * taken from a standby. In this case, the database system status in
966 : * pg_control must indicate that the database was already in recovery.
967 : * Usually that will be DB_IN_ARCHIVE_RECOVERY but also can be
968 : * DB_SHUTDOWNED_IN_RECOVERY if recovery previously was interrupted
969 : * before reaching this point; e.g. because restore_command or
970 : * primary_conninfo were faulty.
971 : *
972 : * Any other state indicates that the backup somehow became corrupted
973 : * and we can't sensibly continue with recovery.
974 : */
975 392 : if (haveBackupLabel)
976 : {
977 126 : ControlFile->backupStartPoint = checkPoint.redo;
978 126 : ControlFile->backupEndRequired = backupEndRequired;
979 :
980 126 : if (backupFromStandby)
981 : {
982 8 : if (dbstate_at_startup != DB_IN_ARCHIVE_RECOVERY &&
983 : dbstate_at_startup != DB_SHUTDOWNED_IN_RECOVERY)
984 0 : ereport(FATAL,
985 : (errmsg("backup_label contains data inconsistent with control file"),
986 : errhint("This means that the backup is corrupted and you will "
987 : "have to use another backup for recovery.")));
988 8 : ControlFile->backupEndPoint = ControlFile->minRecoveryPoint;
989 : }
990 : }
991 : }
992 :
993 : /* remember these, so that we know when we have reached consistency */
994 1520 : backupStartPoint = ControlFile->backupStartPoint;
995 1520 : backupEndRequired = ControlFile->backupEndRequired;
996 1520 : backupEndPoint = ControlFile->backupEndPoint;
997 1520 : if (InArchiveRecovery)
998 : {
999 196 : minRecoveryPoint = ControlFile->minRecoveryPoint;
1000 196 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
1001 : }
1002 : else
1003 : {
1004 1324 : minRecoveryPoint = InvalidXLogRecPtr;
1005 1324 : minRecoveryPointTLI = 0;
1006 : }
1007 :
1008 : /*
1009 : * Start recovery assuming that the final record isn't lost.
1010 : */
1011 1520 : abortedRecPtr = InvalidXLogRecPtr;
1012 1520 : missingContrecPtr = InvalidXLogRecPtr;
1013 :
1014 1520 : *wasShutdown_ptr = wasShutdown;
1015 1520 : *haveBackupLabel_ptr = haveBackupLabel;
1016 1520 : *haveTblspcMap_ptr = haveTblspcMap;
1017 1520 : }
1018 :
1019 : /*
1020 : * See if there are any recovery signal files and if so, set state for
1021 : * recovery.
1022 : *
1023 : * See if there is a recovery command file (recovery.conf), and if so
1024 : * throw an ERROR since as of PG12 we no longer recognize that.
1025 : */
1026 : static void
1027 1520 : readRecoverySignalFile(void)
1028 : {
1029 : struct stat stat_buf;
1030 :
1031 1520 : if (IsBootstrapProcessingMode())
1032 1330 : return;
1033 :
1034 : /*
1035 : * Check for old recovery API file: recovery.conf
1036 : */
1037 1440 : if (stat(RECOVERY_COMMAND_FILE, &stat_buf) == 0)
1038 0 : ereport(FATAL,
1039 : (errcode_for_file_access(),
1040 : errmsg("using recovery command file \"%s\" is not supported",
1041 : RECOVERY_COMMAND_FILE)));
1042 :
1043 : /*
1044 : * Remove unused .done file, if present. Ignore if absent.
1045 : */
1046 1440 : unlink(RECOVERY_COMMAND_DONE);
1047 :
1048 : /*
1049 : * Check for recovery signal files and if found, fsync them since they
1050 : * represent server state information. We don't sweat too much about the
1051 : * possibility of fsync failure, however.
1052 : *
1053 : * If present, standby signal file takes precedence. If neither is present
1054 : * then we won't enter archive recovery.
1055 : */
1056 1440 : if (stat(STANDBY_SIGNAL_FILE, &stat_buf) == 0)
1057 : {
1058 : int fd;
1059 :
1060 180 : fd = BasicOpenFilePerm(STANDBY_SIGNAL_FILE, O_RDWR | PG_BINARY,
1061 : S_IRUSR | S_IWUSR);
1062 180 : if (fd >= 0)
1063 : {
1064 180 : (void) pg_fsync(fd);
1065 180 : close(fd);
1066 : }
1067 180 : standby_signal_file_found = true;
1068 : }
1069 1260 : else if (stat(RECOVERY_SIGNAL_FILE, &stat_buf) == 0)
1070 : {
1071 : int fd;
1072 :
1073 10 : fd = BasicOpenFilePerm(RECOVERY_SIGNAL_FILE, O_RDWR | PG_BINARY,
1074 : S_IRUSR | S_IWUSR);
1075 10 : if (fd >= 0)
1076 : {
1077 10 : (void) pg_fsync(fd);
1078 10 : close(fd);
1079 : }
1080 10 : recovery_signal_file_found = true;
1081 : }
1082 :
1083 1440 : StandbyModeRequested = false;
1084 1440 : ArchiveRecoveryRequested = false;
1085 1440 : if (standby_signal_file_found)
1086 : {
1087 180 : StandbyModeRequested = true;
1088 180 : ArchiveRecoveryRequested = true;
1089 : }
1090 1260 : else if (recovery_signal_file_found)
1091 : {
1092 10 : StandbyModeRequested = false;
1093 10 : ArchiveRecoveryRequested = true;
1094 : }
1095 : else
1096 1250 : return;
1097 :
1098 : /*
1099 : * We don't support standby mode in standalone backends; that requires
1100 : * other processes such as the WAL receiver to be alive.
1101 : */
1102 190 : if (StandbyModeRequested && !IsUnderPostmaster)
1103 0 : ereport(FATAL,
1104 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1105 : errmsg("standby mode is not supported by single-user servers")));
1106 : }
1107 :
1108 : static void
1109 1520 : validateRecoveryParameters(void)
1110 : {
1111 1520 : if (!ArchiveRecoveryRequested)
1112 1330 : return;
1113 :
1114 : /*
1115 : * Check for compulsory parameters
1116 : */
1117 190 : if (StandbyModeRequested)
1118 : {
1119 180 : if ((PrimaryConnInfo == NULL || strcmp(PrimaryConnInfo, "") == 0) &&
1120 18 : (recoveryRestoreCommand == NULL || strcmp(recoveryRestoreCommand, "") == 0))
1121 4 : ereport(WARNING,
1122 : (errmsg("specified neither primary_conninfo nor restore_command"),
1123 : errhint("The database server will regularly poll the pg_wal subdirectory to check for files placed there.")));
1124 : }
1125 : else
1126 : {
1127 10 : if (recoveryRestoreCommand == NULL ||
1128 10 : strcmp(recoveryRestoreCommand, "") == 0)
1129 0 : ereport(FATAL,
1130 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1131 : errmsg("must specify restore_command when standby mode is not enabled")));
1132 : }
1133 :
1134 : /*
1135 : * Override any inconsistent requests. Note that this is a change of
1136 : * behaviour in 9.5; prior to this we simply ignored a request to pause if
1137 : * hot_standby = off, which was surprising behaviour.
1138 : */
1139 190 : if (recoveryTargetAction == RECOVERY_TARGET_ACTION_PAUSE &&
1140 176 : !EnableHotStandby)
1141 4 : recoveryTargetAction = RECOVERY_TARGET_ACTION_SHUTDOWN;
1142 :
1143 : /*
1144 : * Final parsing of recovery_target_time string; see also
1145 : * check_recovery_target_time().
1146 : */
1147 190 : if (recoveryTarget == RECOVERY_TARGET_TIME)
1148 : {
1149 0 : recoveryTargetTime = DatumGetTimestampTz(DirectFunctionCall3(timestamptz_in,
1150 : CStringGetDatum(recovery_target_time_string),
1151 : ObjectIdGetDatum(InvalidOid),
1152 : Int32GetDatum(-1)));
1153 : }
1154 :
1155 : /*
1156 : * If user specified recovery_target_timeline, validate it or compute the
1157 : * "latest" value. We can't do this until after we've gotten the restore
1158 : * command and set InArchiveRecovery, because we need to fetch timeline
1159 : * history files from the archive.
1160 : */
1161 190 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
1162 : {
1163 0 : TimeLineID rtli = recoveryTargetTLIRequested;
1164 :
1165 : /* Timeline 1 does not have a history file, all else should */
1166 0 : if (rtli != 1 && !existsTimeLineHistory(rtli))
1167 0 : ereport(FATAL,
1168 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1169 : errmsg("recovery target timeline %u does not exist",
1170 : rtli)));
1171 0 : recoveryTargetTLI = rtli;
1172 : }
1173 190 : else if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
1174 : {
1175 : /* We start the "latest" search from pg_control's timeline */
1176 190 : recoveryTargetTLI = findNewestTimeLine(recoveryTargetTLI);
1177 : }
1178 : else
1179 : {
1180 : /*
1181 : * else we just use the recoveryTargetTLI as already read from
1182 : * ControlFile
1183 : */
1184 : Assert(recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_CONTROLFILE);
1185 : }
1186 : }
1187 :
1188 : /*
1189 : * read_backup_label: check to see if a backup_label file is present
1190 : *
1191 : * If we see a backup_label during recovery, we assume that we are recovering
1192 : * from a backup dump file, and we therefore roll forward from the checkpoint
1193 : * identified by the label file, NOT what pg_control says. This avoids the
1194 : * problem that pg_control might have been archived one or more checkpoints
1195 : * later than the start of the dump, and so if we rely on it as the start
1196 : * point, we will fail to restore a consistent database state.
1197 : *
1198 : * Returns true if a backup_label was found (and fills the checkpoint
1199 : * location and TLI into *checkPointLoc and *backupLabelTLI, respectively);
1200 : * returns false if not. If this backup_label came from a streamed backup,
1201 : * *backupEndRequired is set to true. If this backup_label was created during
1202 : * recovery, *backupFromStandby is set to true.
1203 : *
1204 : * Also sets the global variables RedoStartLSN and RedoStartTLI with the LSN
1205 : * and TLI read from the backup file.
1206 : */
1207 : static bool
1208 1520 : read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI,
1209 : bool *backupEndRequired, bool *backupFromStandby)
1210 : {
1211 : char startxlogfilename[MAXFNAMELEN];
1212 : TimeLineID tli_from_walseg,
1213 : tli_from_file;
1214 : FILE *lfp;
1215 : char ch;
1216 : char backuptype[20];
1217 : char backupfrom[20];
1218 : char backuplabel[MAXPGPATH];
1219 : char backuptime[128];
1220 : uint32 hi,
1221 : lo;
1222 :
1223 : /* suppress possible uninitialized-variable warnings */
1224 1520 : *checkPointLoc = InvalidXLogRecPtr;
1225 1520 : *backupLabelTLI = 0;
1226 1520 : *backupEndRequired = false;
1227 1520 : *backupFromStandby = false;
1228 :
1229 : /*
1230 : * See if label file is present
1231 : */
1232 1520 : lfp = AllocateFile(BACKUP_LABEL_FILE, "r");
1233 1520 : if (!lfp)
1234 : {
1235 1394 : if (errno != ENOENT)
1236 0 : ereport(FATAL,
1237 : (errcode_for_file_access(),
1238 : errmsg("could not read file \"%s\": %m",
1239 : BACKUP_LABEL_FILE)));
1240 1394 : return false; /* it's not there, all is fine */
1241 : }
1242 :
1243 : /*
1244 : * Read and parse the START WAL LOCATION and CHECKPOINT lines (this code
1245 : * is pretty crude, but we are not expecting any variability in the file
1246 : * format).
1247 : */
1248 126 : if (fscanf(lfp, "START WAL LOCATION: %X/%X (file %08X%16s)%c",
1249 126 : &hi, &lo, &tli_from_walseg, startxlogfilename, &ch) != 5 || ch != '\n')
1250 0 : ereport(FATAL,
1251 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1252 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
1253 126 : RedoStartLSN = ((uint64) hi) << 32 | lo;
1254 126 : RedoStartTLI = tli_from_walseg;
1255 126 : if (fscanf(lfp, "CHECKPOINT LOCATION: %X/%X%c",
1256 126 : &hi, &lo, &ch) != 3 || ch != '\n')
1257 0 : ereport(FATAL,
1258 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1259 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
1260 126 : *checkPointLoc = ((uint64) hi) << 32 | lo;
1261 126 : *backupLabelTLI = tli_from_walseg;
1262 :
1263 : /*
1264 : * BACKUP METHOD lets us know if this was a typical backup ("streamed",
1265 : * which could mean either pg_basebackup or the pg_backup_start/stop
1266 : * method was used) or if this label came from somewhere else (the only
1267 : * other option today being from pg_rewind). If this was a streamed
1268 : * backup then we know that we need to play through until we get to the
1269 : * end of the WAL which was generated during the backup (at which point we
1270 : * will have reached consistency and backupEndRequired will be reset to be
1271 : * false).
1272 : */
1273 126 : if (fscanf(lfp, "BACKUP METHOD: %19s\n", backuptype) == 1)
1274 : {
1275 126 : if (strcmp(backuptype, "streamed") == 0)
1276 124 : *backupEndRequired = true;
1277 : }
1278 :
1279 : /*
1280 : * BACKUP FROM lets us know if this was from a primary or a standby. If
1281 : * it was from a standby, we'll double-check that the control file state
1282 : * matches that of a standby.
1283 : */
1284 126 : if (fscanf(lfp, "BACKUP FROM: %19s\n", backupfrom) == 1)
1285 : {
1286 126 : if (strcmp(backupfrom, "standby") == 0)
1287 8 : *backupFromStandby = true;
1288 : }
1289 :
1290 : /*
1291 : * Parse START TIME and LABEL. Those are not mandatory fields for recovery
1292 : * but checking for their presence is useful for debugging and the next
1293 : * sanity checks. Cope also with the fact that the result buffers have a
1294 : * pre-allocated size, hence if the backup_label file has been generated
1295 : * with strings longer than the maximum assumed here an incorrect parsing
1296 : * happens. That's fine as only minor consistency checks are done
1297 : * afterwards.
1298 : */
1299 126 : if (fscanf(lfp, "START TIME: %127[^\n]\n", backuptime) == 1)
1300 126 : ereport(DEBUG1,
1301 : (errmsg_internal("backup time %s in file \"%s\"",
1302 : backuptime, BACKUP_LABEL_FILE)));
1303 :
1304 126 : if (fscanf(lfp, "LABEL: %1023[^\n]\n", backuplabel) == 1)
1305 124 : ereport(DEBUG1,
1306 : (errmsg_internal("backup label %s in file \"%s\"",
1307 : backuplabel, BACKUP_LABEL_FILE)));
1308 :
1309 : /*
1310 : * START TIMELINE is new as of 11. Its parsing is not mandatory, still use
1311 : * it as a sanity check if present.
1312 : */
1313 126 : if (fscanf(lfp, "START TIMELINE: %u\n", &tli_from_file) == 1)
1314 : {
1315 124 : if (tli_from_walseg != tli_from_file)
1316 0 : ereport(FATAL,
1317 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1318 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE),
1319 : errdetail("Timeline ID parsed is %u, but expected %u.",
1320 : tli_from_file, tli_from_walseg)));
1321 :
1322 124 : ereport(DEBUG1,
1323 : (errmsg_internal("backup timeline %u in file \"%s\"",
1324 : tli_from_file, BACKUP_LABEL_FILE)));
1325 : }
1326 :
1327 126 : if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%X\n", &hi, &lo) > 0)
1328 0 : ereport(FATAL,
1329 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1330 : errmsg("this is an incremental backup, not a data directory"),
1331 : errhint("Use pg_combinebackup to reconstruct a valid data directory.")));
1332 :
1333 126 : if (ferror(lfp) || FreeFile(lfp))
1334 0 : ereport(FATAL,
1335 : (errcode_for_file_access(),
1336 : errmsg("could not read file \"%s\": %m",
1337 : BACKUP_LABEL_FILE)));
1338 :
1339 126 : return true;
1340 : }
1341 :
1342 : /*
1343 : * read_tablespace_map: check to see if a tablespace_map file is present
1344 : *
1345 : * If we see a tablespace_map file during recovery, we assume that we are
1346 : * recovering from a backup dump file, and we therefore need to create symlinks
1347 : * as per the information present in tablespace_map file.
1348 : *
1349 : * Returns true if a tablespace_map file was found (and fills *tablespaces
1350 : * with a tablespaceinfo struct for each tablespace listed in the file);
1351 : * returns false if not.
1352 : */
1353 : static bool
1354 126 : read_tablespace_map(List **tablespaces)
1355 : {
1356 : tablespaceinfo *ti;
1357 : FILE *lfp;
1358 : char str[MAXPGPATH];
1359 : int ch,
1360 : i,
1361 : n;
1362 : bool was_backslash;
1363 :
1364 : /*
1365 : * See if tablespace_map file is present
1366 : */
1367 126 : lfp = AllocateFile(TABLESPACE_MAP, "r");
1368 126 : if (!lfp)
1369 : {
1370 124 : if (errno != ENOENT)
1371 0 : ereport(FATAL,
1372 : (errcode_for_file_access(),
1373 : errmsg("could not read file \"%s\": %m",
1374 : TABLESPACE_MAP)));
1375 124 : return false; /* it's not there, all is fine */
1376 : }
1377 :
1378 : /*
1379 : * Read and parse the link name and path lines from tablespace_map file
1380 : * (this code is pretty crude, but we are not expecting any variability in
1381 : * the file format). De-escape any backslashes that were inserted.
1382 : */
1383 2 : i = 0;
1384 2 : was_backslash = false;
1385 92 : while ((ch = fgetc(lfp)) != EOF)
1386 : {
1387 90 : if (!was_backslash && (ch == '\n' || ch == '\r'))
1388 : {
1389 : char *endp;
1390 :
1391 2 : if (i == 0)
1392 0 : continue; /* \r immediately followed by \n */
1393 :
1394 : /*
1395 : * The de-escaped line should contain an OID followed by exactly
1396 : * one space followed by a path. The path might start with
1397 : * spaces, so don't be too liberal about parsing.
1398 : */
1399 2 : str[i] = '\0';
1400 2 : n = 0;
1401 12 : while (str[n] && str[n] != ' ')
1402 10 : n++;
1403 2 : if (n < 1 || n >= i - 1)
1404 0 : ereport(FATAL,
1405 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1406 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1407 2 : str[n++] = '\0';
1408 :
1409 2 : ti = palloc0(sizeof(tablespaceinfo));
1410 2 : errno = 0;
1411 2 : ti->oid = strtoul(str, &endp, 10);
1412 2 : if (*endp != '\0' || errno == EINVAL || errno == ERANGE)
1413 0 : ereport(FATAL,
1414 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1415 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1416 2 : ti->path = pstrdup(str + n);
1417 2 : *tablespaces = lappend(*tablespaces, ti);
1418 :
1419 2 : i = 0;
1420 2 : continue;
1421 : }
1422 88 : else if (!was_backslash && ch == '\\')
1423 0 : was_backslash = true;
1424 : else
1425 : {
1426 88 : if (i < sizeof(str) - 1)
1427 88 : str[i++] = ch;
1428 88 : was_backslash = false;
1429 : }
1430 : }
1431 :
1432 2 : if (i != 0 || was_backslash) /* last line not terminated? */
1433 0 : ereport(FATAL,
1434 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1435 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1436 :
1437 2 : if (ferror(lfp) || FreeFile(lfp))
1438 0 : ereport(FATAL,
1439 : (errcode_for_file_access(),
1440 : errmsg("could not read file \"%s\": %m",
1441 : TABLESPACE_MAP)));
1442 :
1443 2 : return true;
1444 : }
1445 :
1446 : /*
1447 : * Finish WAL recovery.
1448 : *
1449 : * This does not close the 'xlogreader' yet, because in some cases the caller
1450 : * still wants to re-read the last checkpoint record by calling
1451 : * ReadCheckpointRecord().
1452 : *
1453 : * Returns the position of the last valid or applied record, after which new
1454 : * WAL should be appended, information about why recovery was ended, and some
1455 : * other things. See the EndOfWalRecoveryInfo struct for details.
1456 : */
1457 : EndOfWalRecoveryInfo *
1458 1420 : FinishWalRecovery(void)
1459 : {
1460 1420 : EndOfWalRecoveryInfo *result = palloc(sizeof(EndOfWalRecoveryInfo));
1461 : XLogRecPtr lastRec;
1462 : TimeLineID lastRecTLI;
1463 : XLogRecPtr endOfLog;
1464 :
1465 : /*
1466 : * Kill WAL receiver, if it's still running, before we continue to write
1467 : * the startup checkpoint and aborted-contrecord records. It will trump
1468 : * over these records and subsequent ones if it's still alive when we
1469 : * start writing WAL.
1470 : */
1471 1420 : XLogShutdownWalRcv();
1472 :
1473 : /*
1474 : * Shutdown the slot sync worker to drop any temporary slots acquired by
1475 : * it and to prevent it from keep trying to fetch the failover slots.
1476 : *
1477 : * We do not update the 'synced' column in 'pg_replication_slots' system
1478 : * view from true to false here, as any failed update could leave 'synced'
1479 : * column false for some slots. This could cause issues during slot sync
1480 : * after restarting the server as a standby. While updating the 'synced'
1481 : * column after switching to the new timeline is an option, it does not
1482 : * simplify the handling for the 'synced' column. Therefore, we retain the
1483 : * 'synced' column as true after promotion as it may provide useful
1484 : * information about the slot origin.
1485 : */
1486 1420 : ShutDownSlotSync();
1487 :
1488 : /*
1489 : * We are now done reading the xlog from stream. Turn off streaming
1490 : * recovery to force fetching the files (which would be required at end of
1491 : * recovery, e.g., timeline history file) from archive or pg_wal.
1492 : *
1493 : * Note that standby mode must be turned off after killing WAL receiver,
1494 : * i.e., calling XLogShutdownWalRcv().
1495 : */
1496 : Assert(!WalRcvStreaming());
1497 1420 : StandbyMode = false;
1498 :
1499 : /*
1500 : * Determine where to start writing WAL next.
1501 : *
1502 : * Re-fetch the last valid or last applied record, so we can identify the
1503 : * exact endpoint of what we consider the valid portion of WAL. There may
1504 : * be an incomplete continuation record after that, in which case
1505 : * 'abortedRecPtr' and 'missingContrecPtr' are set and the caller will
1506 : * write a special OVERWRITE_CONTRECORD message to mark that the rest of
1507 : * it is intentionally missing. See CreateOverwriteContrecordRecord().
1508 : *
1509 : * An important side-effect of this is to load the last page into
1510 : * xlogreader. The caller uses it to initialize the WAL for writing.
1511 : */
1512 1420 : if (!InRecovery)
1513 : {
1514 1128 : lastRec = CheckPointLoc;
1515 1128 : lastRecTLI = CheckPointTLI;
1516 : }
1517 : else
1518 : {
1519 292 : lastRec = XLogRecoveryCtl->lastReplayedReadRecPtr;
1520 292 : lastRecTLI = XLogRecoveryCtl->lastReplayedTLI;
1521 : }
1522 1420 : XLogPrefetcherBeginRead(xlogprefetcher, lastRec);
1523 1420 : (void) ReadRecord(xlogprefetcher, PANIC, false, lastRecTLI);
1524 1420 : endOfLog = xlogreader->EndRecPtr;
1525 :
1526 : /*
1527 : * Remember the TLI in the filename of the XLOG segment containing the
1528 : * end-of-log. It could be different from the timeline that endOfLog
1529 : * nominally belongs to, if there was a timeline switch in that segment,
1530 : * and we were reading the old WAL from a segment belonging to a higher
1531 : * timeline.
1532 : */
1533 1420 : result->endOfLogTLI = xlogreader->seg.ws_tli;
1534 :
1535 1420 : if (ArchiveRecoveryRequested)
1536 : {
1537 : /*
1538 : * We are no longer in archive recovery state.
1539 : *
1540 : * We are now done reading the old WAL. Turn off archive fetching if
1541 : * it was active.
1542 : */
1543 : Assert(InArchiveRecovery);
1544 90 : InArchiveRecovery = false;
1545 :
1546 : /*
1547 : * If the ending log segment is still open, close it (to avoid
1548 : * problems on Windows with trying to rename or delete an open file).
1549 : */
1550 90 : if (readFile >= 0)
1551 : {
1552 90 : close(readFile);
1553 90 : readFile = -1;
1554 : }
1555 : }
1556 :
1557 : /*
1558 : * Copy the last partial block to the caller, for initializing the WAL
1559 : * buffer for appending new WAL.
1560 : */
1561 1420 : if (endOfLog % XLOG_BLCKSZ != 0)
1562 : {
1563 : char *page;
1564 : int len;
1565 : XLogRecPtr pageBeginPtr;
1566 :
1567 1392 : pageBeginPtr = endOfLog - (endOfLog % XLOG_BLCKSZ);
1568 : Assert(readOff == XLogSegmentOffset(pageBeginPtr, wal_segment_size));
1569 :
1570 : /* Copy the valid part of the last block */
1571 1392 : len = endOfLog % XLOG_BLCKSZ;
1572 1392 : page = palloc(len);
1573 1392 : memcpy(page, xlogreader->readBuf, len);
1574 :
1575 1392 : result->lastPageBeginPtr = pageBeginPtr;
1576 1392 : result->lastPage = page;
1577 : }
1578 : else
1579 : {
1580 : /* There is no partial block to copy. */
1581 28 : result->lastPageBeginPtr = endOfLog;
1582 28 : result->lastPage = NULL;
1583 : }
1584 :
1585 : /*
1586 : * Create a comment for the history file to explain why and where timeline
1587 : * changed.
1588 : */
1589 1420 : result->recoveryStopReason = getRecoveryStopReason();
1590 :
1591 1420 : result->lastRec = lastRec;
1592 1420 : result->lastRecTLI = lastRecTLI;
1593 1420 : result->endOfLog = endOfLog;
1594 :
1595 1420 : result->abortedRecPtr = abortedRecPtr;
1596 1420 : result->missingContrecPtr = missingContrecPtr;
1597 :
1598 1420 : result->standby_signal_file_found = standby_signal_file_found;
1599 1420 : result->recovery_signal_file_found = recovery_signal_file_found;
1600 :
1601 1420 : return result;
1602 : }
1603 :
1604 : /*
1605 : * Clean up the WAL reader and leftovers from restoring WAL from archive
1606 : */
1607 : void
1608 1420 : ShutdownWalRecovery(void)
1609 : {
1610 : char recoveryPath[MAXPGPATH];
1611 :
1612 : /* Final update of pg_stat_recovery_prefetch. */
1613 1420 : XLogPrefetcherComputeStats(xlogprefetcher);
1614 :
1615 : /* Shut down xlogreader */
1616 1420 : if (readFile >= 0)
1617 : {
1618 1330 : close(readFile);
1619 1330 : readFile = -1;
1620 : }
1621 1420 : XLogReaderFree(xlogreader);
1622 1420 : XLogPrefetcherFree(xlogprefetcher);
1623 :
1624 1420 : if (ArchiveRecoveryRequested)
1625 : {
1626 : /*
1627 : * Since there might be a partial WAL segment named RECOVERYXLOG, get
1628 : * rid of it.
1629 : */
1630 90 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYXLOG");
1631 90 : unlink(recoveryPath); /* ignore any error */
1632 :
1633 : /* Get rid of any remaining recovered timeline-history file, too */
1634 90 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYHISTORY");
1635 90 : unlink(recoveryPath); /* ignore any error */
1636 : }
1637 :
1638 : /*
1639 : * We don't need the latch anymore. It's not strictly necessary to disown
1640 : * it, but let's do it for the sake of tidiness.
1641 : */
1642 1420 : if (ArchiveRecoveryRequested)
1643 90 : DisownLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
1644 1420 : }
1645 :
1646 : /*
1647 : * Perform WAL recovery.
1648 : *
1649 : * If the system was shut down cleanly, this is never called.
1650 : */
1651 : void
1652 392 : PerformWalRecovery(void)
1653 : {
1654 : XLogRecord *record;
1655 392 : bool reachedRecoveryTarget = false;
1656 : TimeLineID replayTLI;
1657 :
1658 : /*
1659 : * Initialize shared variables for tracking progress of WAL replay, as if
1660 : * we had just replayed the record before the REDO location (or the
1661 : * checkpoint record itself, if it's a shutdown checkpoint).
1662 : */
1663 392 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
1664 392 : if (RedoStartLSN < CheckPointLoc)
1665 : {
1666 200 : XLogRecoveryCtl->lastReplayedReadRecPtr = InvalidXLogRecPtr;
1667 200 : XLogRecoveryCtl->lastReplayedEndRecPtr = RedoStartLSN;
1668 200 : XLogRecoveryCtl->lastReplayedTLI = RedoStartTLI;
1669 : }
1670 : else
1671 : {
1672 192 : XLogRecoveryCtl->lastReplayedReadRecPtr = xlogreader->ReadRecPtr;
1673 192 : XLogRecoveryCtl->lastReplayedEndRecPtr = xlogreader->EndRecPtr;
1674 192 : XLogRecoveryCtl->lastReplayedTLI = CheckPointTLI;
1675 : }
1676 392 : XLogRecoveryCtl->replayEndRecPtr = XLogRecoveryCtl->lastReplayedEndRecPtr;
1677 392 : XLogRecoveryCtl->replayEndTLI = XLogRecoveryCtl->lastReplayedTLI;
1678 392 : XLogRecoveryCtl->recoveryLastXTime = 0;
1679 392 : XLogRecoveryCtl->currentChunkStartTime = 0;
1680 392 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_NOT_PAUSED;
1681 392 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
1682 :
1683 : /* Also ensure XLogReceiptTime has a sane value */
1684 392 : XLogReceiptTime = GetCurrentTimestamp();
1685 :
1686 : /*
1687 : * Let postmaster know we've started redo now, so that it can launch the
1688 : * archiver if necessary.
1689 : */
1690 392 : if (IsUnderPostmaster)
1691 374 : SendPostmasterSignal(PMSIGNAL_RECOVERY_STARTED);
1692 :
1693 : /*
1694 : * Allow read-only connections immediately if we're consistent already.
1695 : */
1696 392 : CheckRecoveryConsistency();
1697 :
1698 : /*
1699 : * Find the first record that logically follows the checkpoint --- it
1700 : * might physically precede it, though.
1701 : */
1702 392 : if (RedoStartLSN < CheckPointLoc)
1703 : {
1704 : /* back up to find the record */
1705 200 : replayTLI = RedoStartTLI;
1706 200 : XLogPrefetcherBeginRead(xlogprefetcher, RedoStartLSN);
1707 200 : record = ReadRecord(xlogprefetcher, PANIC, false, replayTLI);
1708 :
1709 : /*
1710 : * If a checkpoint record's redo pointer points back to an earlier
1711 : * LSN, the record at that LSN should be an XLOG_CHECKPOINT_REDO
1712 : * record.
1713 : */
1714 200 : if (record->xl_rmid != RM_XLOG_ID ||
1715 200 : (record->xl_info & ~XLR_INFO_MASK) != XLOG_CHECKPOINT_REDO)
1716 0 : ereport(FATAL,
1717 : (errmsg("unexpected record type found at redo point %X/%X",
1718 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr))));
1719 : }
1720 : else
1721 : {
1722 : /* just have to read next record after CheckPoint */
1723 : Assert(xlogreader->ReadRecPtr == CheckPointLoc);
1724 192 : replayTLI = CheckPointTLI;
1725 192 : record = ReadRecord(xlogprefetcher, LOG, false, replayTLI);
1726 : }
1727 :
1728 392 : if (record != NULL)
1729 : {
1730 : TimestampTz xtime;
1731 : PGRUsage ru0;
1732 :
1733 374 : pg_rusage_init(&ru0);
1734 :
1735 374 : InRedo = true;
1736 :
1737 374 : RmgrStartup();
1738 :
1739 374 : ereport(LOG,
1740 : (errmsg("redo starts at %X/%X",
1741 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr))));
1742 :
1743 : /* Prepare to report progress of the redo phase. */
1744 374 : if (!StandbyMode)
1745 204 : begin_startup_progress_phase();
1746 :
1747 : /*
1748 : * main redo apply loop
1749 : */
1750 : do
1751 : {
1752 5213706 : if (!StandbyMode)
1753 543506 : ereport_startup_progress("redo in progress, elapsed time: %ld.%02d s, current LSN: %X/%X",
1754 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr));
1755 :
1756 : #ifdef WAL_DEBUG
1757 : if (XLOG_DEBUG)
1758 : {
1759 : StringInfoData buf;
1760 :
1761 : initStringInfo(&buf);
1762 : appendStringInfo(&buf, "REDO @ %X/%X; LSN %X/%X: ",
1763 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr),
1764 : LSN_FORMAT_ARGS(xlogreader->EndRecPtr));
1765 : xlog_outrec(&buf, xlogreader);
1766 : appendStringInfoString(&buf, " - ");
1767 : xlog_outdesc(&buf, xlogreader);
1768 : elog(LOG, "%s", buf.data);
1769 : pfree(buf.data);
1770 : }
1771 : #endif
1772 :
1773 : /* Handle interrupt signals of startup process */
1774 5213706 : HandleStartupProcInterrupts();
1775 :
1776 : /*
1777 : * Pause WAL replay, if requested by a hot-standby session via
1778 : * SetRecoveryPause().
1779 : *
1780 : * Note that we intentionally don't take the info_lck spinlock
1781 : * here. We might therefore read a slightly stale value of the
1782 : * recoveryPause flag, but it can't be very stale (no worse than
1783 : * the last spinlock we did acquire). Since a pause request is a
1784 : * pretty asynchronous thing anyway, possibly responding to it one
1785 : * WAL record later than we otherwise would is a minor issue, so
1786 : * it doesn't seem worth adding another spinlock cycle to prevent
1787 : * that.
1788 : */
1789 5213706 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
1790 : RECOVERY_NOT_PAUSED)
1791 0 : recoveryPausesHere(false);
1792 :
1793 : /*
1794 : * Have we reached our recovery target?
1795 : */
1796 5213706 : if (recoveryStopsBefore(xlogreader))
1797 : {
1798 2 : reachedRecoveryTarget = true;
1799 2 : break;
1800 : }
1801 :
1802 : /*
1803 : * If we've been asked to lag the primary, wait on latch until
1804 : * enough time has passed.
1805 : */
1806 5213704 : if (recoveryApplyDelay(xlogreader))
1807 : {
1808 : /*
1809 : * We test for paused recovery again here. If user sets
1810 : * delayed apply, it may be because they expect to pause
1811 : * recovery in case of problems, so we must test again here
1812 : * otherwise pausing during the delay-wait wouldn't work.
1813 : */
1814 0 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
1815 : RECOVERY_NOT_PAUSED)
1816 0 : recoveryPausesHere(false);
1817 : }
1818 :
1819 : /*
1820 : * Apply the record
1821 : */
1822 5213704 : ApplyWalRecord(xlogreader, record, &replayTLI);
1823 :
1824 : /* Exit loop if we reached inclusive recovery target */
1825 5213700 : if (recoveryStopsAfter(xlogreader))
1826 : {
1827 10 : reachedRecoveryTarget = true;
1828 10 : break;
1829 : }
1830 :
1831 : /* Else, try to fetch the next WAL record */
1832 5213690 : record = ReadRecord(xlogprefetcher, LOG, false, replayTLI);
1833 5213596 : } while (record != NULL);
1834 :
1835 : /*
1836 : * end of main redo apply loop
1837 : */
1838 :
1839 276 : if (reachedRecoveryTarget)
1840 : {
1841 12 : if (!reachedConsistency)
1842 0 : ereport(FATAL,
1843 : (errmsg("requested recovery stop point is before consistent recovery point")));
1844 :
1845 : /*
1846 : * This is the last point where we can restart recovery with a new
1847 : * recovery target, if we shutdown and begin again. After this,
1848 : * Resource Managers may choose to do permanent corrective actions
1849 : * at end of recovery.
1850 : */
1851 12 : switch (recoveryTargetAction)
1852 : {
1853 0 : case RECOVERY_TARGET_ACTION_SHUTDOWN:
1854 :
1855 : /*
1856 : * exit with special return code to request shutdown of
1857 : * postmaster. Log messages issued from postmaster.
1858 : */
1859 0 : proc_exit(3);
1860 :
1861 0 : case RECOVERY_TARGET_ACTION_PAUSE:
1862 0 : SetRecoveryPause(true);
1863 0 : recoveryPausesHere(true);
1864 :
1865 : /* drop into promote */
1866 :
1867 12 : case RECOVERY_TARGET_ACTION_PROMOTE:
1868 12 : break;
1869 : }
1870 264 : }
1871 :
1872 276 : RmgrCleanup();
1873 :
1874 276 : ereport(LOG,
1875 : (errmsg("redo done at %X/%X system usage: %s",
1876 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr),
1877 : pg_rusage_show(&ru0))));
1878 276 : xtime = GetLatestXTime();
1879 276 : if (xtime)
1880 64 : ereport(LOG,
1881 : (errmsg("last completed transaction was at log time %s",
1882 : timestamptz_to_str(xtime))));
1883 :
1884 276 : InRedo = false;
1885 : }
1886 : else
1887 : {
1888 : /* there are no WAL records following the checkpoint */
1889 18 : ereport(LOG,
1890 : (errmsg("redo is not required")));
1891 : }
1892 :
1893 : /*
1894 : * This check is intentionally after the above log messages that indicate
1895 : * how far recovery went.
1896 : */
1897 294 : if (ArchiveRecoveryRequested &&
1898 92 : recoveryTarget != RECOVERY_TARGET_UNSET &&
1899 14 : !reachedRecoveryTarget)
1900 2 : ereport(FATAL,
1901 : (errmsg("recovery ended before configured recovery target was reached")));
1902 292 : }
1903 :
1904 : /*
1905 : * Subroutine of PerformWalRecovery, to apply one WAL record.
1906 : */
1907 : static void
1908 5213704 : ApplyWalRecord(XLogReaderState *xlogreader, XLogRecord *record, TimeLineID *replayTLI)
1909 : {
1910 : ErrorContextCallback errcallback;
1911 5213704 : bool switchedTLI = false;
1912 :
1913 : /* Setup error traceback support for ereport() */
1914 5213704 : errcallback.callback = rm_redo_error_callback;
1915 5213704 : errcallback.arg = (void *) xlogreader;
1916 5213704 : errcallback.previous = error_context_stack;
1917 5213704 : error_context_stack = &errcallback;
1918 :
1919 : /*
1920 : * TransamVariables->nextXid must be beyond record's xid.
1921 : */
1922 5213704 : AdvanceNextFullTransactionIdPastXid(record->xl_xid);
1923 :
1924 : /*
1925 : * Before replaying this record, check if this record causes the current
1926 : * timeline to change. The record is already considered to be part of the
1927 : * new timeline, so we update replayTLI before replaying it. That's
1928 : * important so that replayEndTLI, which is recorded as the minimum
1929 : * recovery point's TLI if recovery stops after this record, is set
1930 : * correctly.
1931 : */
1932 5213704 : if (record->xl_rmid == RM_XLOG_ID)
1933 : {
1934 55950 : TimeLineID newReplayTLI = *replayTLI;
1935 55950 : TimeLineID prevReplayTLI = *replayTLI;
1936 55950 : uint8 info = record->xl_info & ~XLR_INFO_MASK;
1937 :
1938 55950 : if (info == XLOG_CHECKPOINT_SHUTDOWN)
1939 : {
1940 : CheckPoint checkPoint;
1941 :
1942 58 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
1943 58 : newReplayTLI = checkPoint.ThisTimeLineID;
1944 58 : prevReplayTLI = checkPoint.PrevTimeLineID;
1945 : }
1946 55892 : else if (info == XLOG_END_OF_RECOVERY)
1947 : {
1948 : xl_end_of_recovery xlrec;
1949 :
1950 18 : memcpy(&xlrec, XLogRecGetData(xlogreader), sizeof(xl_end_of_recovery));
1951 18 : newReplayTLI = xlrec.ThisTimeLineID;
1952 18 : prevReplayTLI = xlrec.PrevTimeLineID;
1953 : }
1954 :
1955 55950 : if (newReplayTLI != *replayTLI)
1956 : {
1957 : /* Check that it's OK to switch to this TLI */
1958 20 : checkTimeLineSwitch(xlogreader->EndRecPtr,
1959 : newReplayTLI, prevReplayTLI, *replayTLI);
1960 :
1961 : /* Following WAL records should be run with new TLI */
1962 20 : *replayTLI = newReplayTLI;
1963 20 : switchedTLI = true;
1964 : }
1965 : }
1966 :
1967 : /*
1968 : * Update shared replayEndRecPtr before replaying this record, so that
1969 : * XLogFlush will update minRecoveryPoint correctly.
1970 : */
1971 5213704 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
1972 5213704 : XLogRecoveryCtl->replayEndRecPtr = xlogreader->EndRecPtr;
1973 5213704 : XLogRecoveryCtl->replayEndTLI = *replayTLI;
1974 5213704 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
1975 :
1976 : /*
1977 : * If we are attempting to enter Hot Standby mode, process XIDs we see
1978 : */
1979 5213704 : if (standbyState >= STANDBY_INITIALIZED &&
1980 4718430 : TransactionIdIsValid(record->xl_xid))
1981 4639404 : RecordKnownAssignedTransactionIds(record->xl_xid);
1982 :
1983 : /*
1984 : * Some XLOG record types that are related to recovery are processed
1985 : * directly here, rather than in xlog_redo()
1986 : */
1987 5213704 : if (record->xl_rmid == RM_XLOG_ID)
1988 55950 : xlogrecovery_redo(xlogreader, *replayTLI);
1989 :
1990 : /* Now apply the WAL record itself */
1991 5213704 : GetRmgr(record->xl_rmid).rm_redo(xlogreader);
1992 :
1993 : /*
1994 : * After redo, check whether the backup pages associated with the WAL
1995 : * record are consistent with the existing pages. This check is done only
1996 : * if consistency check is enabled for this record.
1997 : */
1998 5213700 : if ((record->xl_info & XLR_CHECK_CONSISTENCY) != 0)
1999 0 : verifyBackupPageConsistency(xlogreader);
2000 :
2001 : /* Pop the error context stack */
2002 5213700 : error_context_stack = errcallback.previous;
2003 :
2004 : /*
2005 : * Update lastReplayedEndRecPtr after this record has been successfully
2006 : * replayed.
2007 : */
2008 5213700 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
2009 5213700 : XLogRecoveryCtl->lastReplayedReadRecPtr = xlogreader->ReadRecPtr;
2010 5213700 : XLogRecoveryCtl->lastReplayedEndRecPtr = xlogreader->EndRecPtr;
2011 5213700 : XLogRecoveryCtl->lastReplayedTLI = *replayTLI;
2012 5213700 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
2013 :
2014 : /* ------
2015 : * Wakeup walsenders:
2016 : *
2017 : * On the standby, the WAL is flushed first (which will only wake up
2018 : * physical walsenders) and then applied, which will only wake up logical
2019 : * walsenders.
2020 : *
2021 : * Indeed, logical walsenders on standby can't decode and send data until
2022 : * it's been applied.
2023 : *
2024 : * Physical walsenders don't need to be woken up during replay unless
2025 : * cascading replication is allowed and time line change occurred (so that
2026 : * they can notice that they are on a new time line).
2027 : *
2028 : * That's why the wake up conditions are for:
2029 : *
2030 : * - physical walsenders in case of new time line and cascade
2031 : * replication is allowed
2032 : * - logical walsenders in case cascade replication is allowed (could not
2033 : * be created otherwise)
2034 : * ------
2035 : */
2036 5213700 : if (AllowCascadeReplication())
2037 4835388 : WalSndWakeup(switchedTLI, true);
2038 :
2039 : /*
2040 : * If rm_redo called XLogRequestWalReceiverReply, then we wake up the
2041 : * receiver so that it notices the updated lastReplayedEndRecPtr and sends
2042 : * a reply to the primary.
2043 : */
2044 5213700 : if (doRequestWalReceiverReply)
2045 : {
2046 0 : doRequestWalReceiverReply = false;
2047 0 : WalRcvForceReply();
2048 : }
2049 :
2050 : /* Allow read-only connections if we're consistent now */
2051 5213700 : CheckRecoveryConsistency();
2052 :
2053 : /* Is this a timeline switch? */
2054 5213700 : if (switchedTLI)
2055 : {
2056 : /*
2057 : * Before we continue on the new timeline, clean up any (possibly
2058 : * bogus) future WAL segments on the old timeline.
2059 : */
2060 20 : RemoveNonParentXlogFiles(xlogreader->EndRecPtr, *replayTLI);
2061 :
2062 : /* Reset the prefetcher. */
2063 20 : XLogPrefetchReconfigure();
2064 : }
2065 5213700 : }
2066 :
2067 : /*
2068 : * Some XLOG RM record types that are directly related to WAL recovery are
2069 : * handled here rather than in the xlog_redo()
2070 : */
2071 : static void
2072 55950 : xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI)
2073 : {
2074 55950 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2075 55950 : XLogRecPtr lsn = record->EndRecPtr;
2076 :
2077 : Assert(XLogRecGetRmid(record) == RM_XLOG_ID);
2078 :
2079 55950 : if (info == XLOG_OVERWRITE_CONTRECORD)
2080 : {
2081 : /* Verify the payload of a XLOG_OVERWRITE_CONTRECORD record. */
2082 : xl_overwrite_contrecord xlrec;
2083 :
2084 2 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_overwrite_contrecord));
2085 2 : if (xlrec.overwritten_lsn != record->overwrittenRecPtr)
2086 0 : elog(FATAL, "mismatching overwritten LSN %X/%X -> %X/%X",
2087 : LSN_FORMAT_ARGS(xlrec.overwritten_lsn),
2088 : LSN_FORMAT_ARGS(record->overwrittenRecPtr));
2089 :
2090 : /* We have safely skipped the aborted record */
2091 2 : abortedRecPtr = InvalidXLogRecPtr;
2092 2 : missingContrecPtr = InvalidXLogRecPtr;
2093 :
2094 2 : ereport(LOG,
2095 : (errmsg("successfully skipped missing contrecord at %X/%X, overwritten at %s",
2096 : LSN_FORMAT_ARGS(xlrec.overwritten_lsn),
2097 : timestamptz_to_str(xlrec.overwrite_time))));
2098 :
2099 : /* Verifying the record should only happen once */
2100 2 : record->overwrittenRecPtr = InvalidXLogRecPtr;
2101 : }
2102 55948 : else if (info == XLOG_BACKUP_END)
2103 : {
2104 : XLogRecPtr startpoint;
2105 :
2106 152 : memcpy(&startpoint, XLogRecGetData(record), sizeof(startpoint));
2107 :
2108 152 : if (backupStartPoint == startpoint)
2109 : {
2110 : /*
2111 : * We have reached the end of base backup, the point where
2112 : * pg_backup_stop() was done. The data on disk is now consistent
2113 : * (assuming we have also reached minRecoveryPoint). Set
2114 : * backupEndPoint to the current LSN, so that the next call to
2115 : * CheckRecoveryConsistency() will notice it and do the
2116 : * end-of-backup processing.
2117 : */
2118 122 : elog(DEBUG1, "end of backup record reached");
2119 :
2120 122 : backupEndPoint = lsn;
2121 : }
2122 : else
2123 30 : elog(DEBUG1, "saw end-of-backup record for backup starting at %X/%X, waiting for %X/%X",
2124 : LSN_FORMAT_ARGS(startpoint), LSN_FORMAT_ARGS(backupStartPoint));
2125 : }
2126 55950 : }
2127 :
2128 : /*
2129 : * Verify that, in non-test mode, ./pg_tblspc doesn't contain any real
2130 : * directories.
2131 : *
2132 : * Replay of database creation XLOG records for databases that were later
2133 : * dropped can create fake directories in pg_tblspc. By the time consistency
2134 : * is reached these directories should have been removed; here we verify
2135 : * that this did indeed happen. This is to be called at the point where
2136 : * consistent state is reached.
2137 : *
2138 : * allow_in_place_tablespaces turns the PANIC into a WARNING, which is
2139 : * useful for testing purposes, and also allows for an escape hatch in case
2140 : * things go south.
2141 : */
2142 : static void
2143 200 : CheckTablespaceDirectory(void)
2144 : {
2145 : DIR *dir;
2146 : struct dirent *de;
2147 :
2148 200 : dir = AllocateDir("pg_tblspc");
2149 610 : while ((de = ReadDir(dir, "pg_tblspc")) != NULL)
2150 : {
2151 : char path[MAXPGPATH + 10];
2152 :
2153 : /* Skip entries of non-oid names */
2154 410 : if (strspn(de->d_name, "0123456789") != strlen(de->d_name))
2155 400 : continue;
2156 :
2157 10 : snprintf(path, sizeof(path), "pg_tblspc/%s", de->d_name);
2158 :
2159 10 : if (get_dirent_type(path, de, false, ERROR) != PGFILETYPE_LNK)
2160 8 : ereport(allow_in_place_tablespaces ? WARNING : PANIC,
2161 : (errcode(ERRCODE_DATA_CORRUPTED),
2162 : errmsg("unexpected directory entry \"%s\" found in %s",
2163 : de->d_name, "pg_tblspc/"),
2164 : errdetail("All directory entries in pg_tblspc/ should be symbolic links."),
2165 : errhint("Remove those directories, or set allow_in_place_tablespaces to ON transiently to let recovery complete.")));
2166 : }
2167 200 : }
2168 :
2169 : /*
2170 : * Checks if recovery has reached a consistent state. When consistency is
2171 : * reached and we have a valid starting standby snapshot, tell postmaster
2172 : * that it can start accepting read-only connections.
2173 : */
2174 : static void
2175 5214096 : CheckRecoveryConsistency(void)
2176 : {
2177 : XLogRecPtr lastReplayedEndRecPtr;
2178 : TimeLineID lastReplayedTLI;
2179 :
2180 : /*
2181 : * During crash recovery, we don't reach a consistent state until we've
2182 : * replayed all the WAL.
2183 : */
2184 5214096 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
2185 533272 : return;
2186 :
2187 : Assert(InArchiveRecovery);
2188 :
2189 : /*
2190 : * assume that we are called in the startup process, and hence don't need
2191 : * a lock to read lastReplayedEndRecPtr
2192 : */
2193 4680824 : lastReplayedEndRecPtr = XLogRecoveryCtl->lastReplayedEndRecPtr;
2194 4680824 : lastReplayedTLI = XLogRecoveryCtl->lastReplayedTLI;
2195 :
2196 : /*
2197 : * Have we reached the point where our base backup was completed?
2198 : */
2199 4680824 : if (!XLogRecPtrIsInvalid(backupEndPoint) &&
2200 188 : backupEndPoint <= lastReplayedEndRecPtr)
2201 : {
2202 126 : XLogRecPtr saveBackupStartPoint = backupStartPoint;
2203 126 : XLogRecPtr saveBackupEndPoint = backupEndPoint;
2204 :
2205 126 : elog(DEBUG1, "end of backup reached");
2206 :
2207 : /*
2208 : * We have reached the end of base backup, as indicated by pg_control.
2209 : * Update the control file accordingly.
2210 : */
2211 126 : ReachedEndOfBackup(lastReplayedEndRecPtr, lastReplayedTLI);
2212 126 : backupStartPoint = InvalidXLogRecPtr;
2213 126 : backupEndPoint = InvalidXLogRecPtr;
2214 126 : backupEndRequired = false;
2215 :
2216 126 : ereport(LOG,
2217 : (errmsg("completed backup recovery with redo LSN %X/%X and end LSN %X/%X",
2218 : LSN_FORMAT_ARGS(saveBackupStartPoint),
2219 : LSN_FORMAT_ARGS(saveBackupEndPoint))));
2220 : }
2221 :
2222 : /*
2223 : * Have we passed our safe starting point? Note that minRecoveryPoint is
2224 : * known to be incorrectly set if recovering from a backup, until the
2225 : * XLOG_BACKUP_END arrives to advise us of the correct minRecoveryPoint.
2226 : * All we know prior to that is that we're not consistent yet.
2227 : */
2228 4680824 : if (!reachedConsistency && !backupEndRequired &&
2229 13866 : minRecoveryPoint <= lastReplayedEndRecPtr)
2230 : {
2231 : /*
2232 : * Check to see if the XLOG sequence contained any unresolved
2233 : * references to uninitialized pages.
2234 : */
2235 200 : XLogCheckInvalidPages();
2236 :
2237 : /*
2238 : * Check that pg_tblspc doesn't contain any real directories. Replay
2239 : * of Database/CREATE_* records may have created fictitious tablespace
2240 : * directories that should have been removed by the time consistency
2241 : * was reached.
2242 : */
2243 200 : CheckTablespaceDirectory();
2244 :
2245 200 : reachedConsistency = true;
2246 200 : ereport(LOG,
2247 : (errmsg("consistent recovery state reached at %X/%X",
2248 : LSN_FORMAT_ARGS(lastReplayedEndRecPtr))));
2249 : }
2250 :
2251 : /*
2252 : * Have we got a valid starting snapshot that will allow queries to be
2253 : * run? If so, we can tell postmaster that the database is consistent now,
2254 : * enabling connections.
2255 : */
2256 4680824 : if (standbyState == STANDBY_SNAPSHOT_READY &&
2257 4680424 : !LocalHotStandbyActive &&
2258 186 : reachedConsistency &&
2259 : IsUnderPostmaster)
2260 : {
2261 186 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
2262 186 : XLogRecoveryCtl->SharedHotStandbyActive = true;
2263 186 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
2264 :
2265 186 : LocalHotStandbyActive = true;
2266 :
2267 186 : SendPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY);
2268 : }
2269 : }
2270 :
2271 : /*
2272 : * Error context callback for errors occurring during rm_redo().
2273 : */
2274 : static void
2275 194 : rm_redo_error_callback(void *arg)
2276 : {
2277 194 : XLogReaderState *record = (XLogReaderState *) arg;
2278 : StringInfoData buf;
2279 :
2280 194 : initStringInfo(&buf);
2281 194 : xlog_outdesc(&buf, record);
2282 194 : xlog_block_info(&buf, record);
2283 :
2284 : /* translator: %s is a WAL record description */
2285 194 : errcontext("WAL redo at %X/%X for %s",
2286 194 : LSN_FORMAT_ARGS(record->ReadRecPtr),
2287 : buf.data);
2288 :
2289 194 : pfree(buf.data);
2290 194 : }
2291 :
2292 : /*
2293 : * Returns a string describing an XLogRecord, consisting of its identity
2294 : * optionally followed by a colon, a space, and a further description.
2295 : */
2296 : void
2297 194 : xlog_outdesc(StringInfo buf, XLogReaderState *record)
2298 : {
2299 194 : RmgrData rmgr = GetRmgr(XLogRecGetRmid(record));
2300 194 : uint8 info = XLogRecGetInfo(record);
2301 : const char *id;
2302 :
2303 194 : appendStringInfoString(buf, rmgr.rm_name);
2304 194 : appendStringInfoChar(buf, '/');
2305 :
2306 194 : id = rmgr.rm_identify(info);
2307 194 : if (id == NULL)
2308 0 : appendStringInfo(buf, "UNKNOWN (%X): ", info & ~XLR_INFO_MASK);
2309 : else
2310 194 : appendStringInfo(buf, "%s: ", id);
2311 :
2312 194 : rmgr.rm_desc(buf, record);
2313 194 : }
2314 :
2315 : #ifdef WAL_DEBUG
2316 :
2317 : static void
2318 : xlog_outrec(StringInfo buf, XLogReaderState *record)
2319 : {
2320 : appendStringInfo(buf, "prev %X/%X; xid %u",
2321 : LSN_FORMAT_ARGS(XLogRecGetPrev(record)),
2322 : XLogRecGetXid(record));
2323 :
2324 : appendStringInfo(buf, "; len %u",
2325 : XLogRecGetDataLen(record));
2326 :
2327 : xlog_block_info(buf, record);
2328 : }
2329 : #endif /* WAL_DEBUG */
2330 :
2331 : /*
2332 : * Returns a string giving information about all the blocks in an
2333 : * XLogRecord.
2334 : */
2335 : static void
2336 194 : xlog_block_info(StringInfo buf, XLogReaderState *record)
2337 : {
2338 : int block_id;
2339 :
2340 : /* decode block references */
2341 326 : for (block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
2342 : {
2343 : RelFileLocator rlocator;
2344 : ForkNumber forknum;
2345 : BlockNumber blk;
2346 :
2347 132 : if (!XLogRecGetBlockTagExtended(record, block_id,
2348 : &rlocator, &forknum, &blk, NULL))
2349 0 : continue;
2350 :
2351 132 : if (forknum != MAIN_FORKNUM)
2352 6 : appendStringInfo(buf, "; blkref #%d: rel %u/%u/%u, fork %u, blk %u",
2353 : block_id,
2354 : rlocator.spcOid, rlocator.dbOid,
2355 : rlocator.relNumber,
2356 : forknum,
2357 : blk);
2358 : else
2359 126 : appendStringInfo(buf, "; blkref #%d: rel %u/%u/%u, blk %u",
2360 : block_id,
2361 : rlocator.spcOid, rlocator.dbOid,
2362 : rlocator.relNumber,
2363 : blk);
2364 132 : if (XLogRecHasBlockImage(record, block_id))
2365 88 : appendStringInfoString(buf, " FPW");
2366 : }
2367 194 : }
2368 :
2369 :
2370 : /*
2371 : * Check that it's OK to switch to new timeline during recovery.
2372 : *
2373 : * 'lsn' is the address of the shutdown checkpoint record we're about to
2374 : * replay. (Currently, timeline can only change at a shutdown checkpoint).
2375 : */
2376 : static void
2377 20 : checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI, TimeLineID prevTLI,
2378 : TimeLineID replayTLI)
2379 : {
2380 : /* Check that the record agrees on what the current (old) timeline is */
2381 20 : if (prevTLI != replayTLI)
2382 0 : ereport(PANIC,
2383 : (errmsg("unexpected previous timeline ID %u (current timeline ID %u) in checkpoint record",
2384 : prevTLI, replayTLI)));
2385 :
2386 : /*
2387 : * The new timeline better be in the list of timelines we expect to see,
2388 : * according to the timeline history. It should also not decrease.
2389 : */
2390 20 : if (newTLI < replayTLI || !tliInHistory(newTLI, expectedTLEs))
2391 0 : ereport(PANIC,
2392 : (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
2393 : newTLI, replayTLI)));
2394 :
2395 : /*
2396 : * If we have not yet reached min recovery point, and we're about to
2397 : * switch to a timeline greater than the timeline of the min recovery
2398 : * point: trouble. After switching to the new timeline, we could not
2399 : * possibly visit the min recovery point on the correct timeline anymore.
2400 : * This can happen if there is a newer timeline in the archive that
2401 : * branched before the timeline the min recovery point is on, and you
2402 : * attempt to do PITR to the new timeline.
2403 : */
2404 20 : if (!XLogRecPtrIsInvalid(minRecoveryPoint) &&
2405 18 : lsn < minRecoveryPoint &&
2406 2 : newTLI > minRecoveryPointTLI)
2407 0 : ereport(PANIC,
2408 : (errmsg("unexpected timeline ID %u in checkpoint record, before reaching minimum recovery point %X/%X on timeline %u",
2409 : newTLI,
2410 : LSN_FORMAT_ARGS(minRecoveryPoint),
2411 : minRecoveryPointTLI)));
2412 :
2413 : /* Looks good */
2414 20 : }
2415 :
2416 :
2417 : /*
2418 : * Extract timestamp from WAL record.
2419 : *
2420 : * If the record contains a timestamp, returns true, and saves the timestamp
2421 : * in *recordXtime. If the record type has no timestamp, returns false.
2422 : * Currently, only transaction commit/abort records and restore points contain
2423 : * timestamps.
2424 : */
2425 : static bool
2426 77312 : getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime)
2427 : {
2428 77312 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2429 77312 : uint8 xact_info = info & XLOG_XACT_OPMASK;
2430 77312 : uint8 rmid = XLogRecGetRmid(record);
2431 :
2432 77312 : if (rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
2433 : {
2434 4 : *recordXtime = ((xl_restore_point *) XLogRecGetData(record))->rp_time;
2435 4 : return true;
2436 : }
2437 77308 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_COMMIT ||
2438 : xact_info == XLOG_XACT_COMMIT_PREPARED))
2439 : {
2440 71000 : *recordXtime = ((xl_xact_commit *) XLogRecGetData(record))->xact_time;
2441 71000 : return true;
2442 : }
2443 6308 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_ABORT ||
2444 : xact_info == XLOG_XACT_ABORT_PREPARED))
2445 : {
2446 6308 : *recordXtime = ((xl_xact_abort *) XLogRecGetData(record))->xact_time;
2447 6308 : return true;
2448 : }
2449 0 : return false;
2450 : }
2451 :
2452 : /*
2453 : * Checks whether the current buffer page and backup page stored in the
2454 : * WAL record are consistent or not. Before comparing the two pages, a
2455 : * masking can be applied to the pages to ignore certain areas like hint bits,
2456 : * unused space between pd_lower and pd_upper among other things. This
2457 : * function should be called once WAL replay has been completed for a
2458 : * given record.
2459 : */
2460 : static void
2461 0 : verifyBackupPageConsistency(XLogReaderState *record)
2462 : {
2463 0 : RmgrData rmgr = GetRmgr(XLogRecGetRmid(record));
2464 : RelFileLocator rlocator;
2465 : ForkNumber forknum;
2466 : BlockNumber blkno;
2467 : int block_id;
2468 :
2469 : /* Records with no backup blocks have no need for consistency checks. */
2470 0 : if (!XLogRecHasAnyBlockRefs(record))
2471 0 : return;
2472 :
2473 : Assert((XLogRecGetInfo(record) & XLR_CHECK_CONSISTENCY) != 0);
2474 :
2475 0 : for (block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
2476 : {
2477 : Buffer buf;
2478 : Page page;
2479 :
2480 0 : if (!XLogRecGetBlockTagExtended(record, block_id,
2481 : &rlocator, &forknum, &blkno, NULL))
2482 : {
2483 : /*
2484 : * WAL record doesn't contain a block reference with the given id.
2485 : * Do nothing.
2486 : */
2487 0 : continue;
2488 : }
2489 :
2490 : Assert(XLogRecHasBlockImage(record, block_id));
2491 :
2492 0 : if (XLogRecBlockImageApply(record, block_id))
2493 : {
2494 : /*
2495 : * WAL record has already applied the page, so bypass the
2496 : * consistency check as that would result in comparing the full
2497 : * page stored in the record with itself.
2498 : */
2499 0 : continue;
2500 : }
2501 :
2502 : /*
2503 : * Read the contents from the current buffer and store it in a
2504 : * temporary page.
2505 : */
2506 0 : buf = XLogReadBufferExtended(rlocator, forknum, blkno,
2507 : RBM_NORMAL_NO_LOG,
2508 : InvalidBuffer);
2509 0 : if (!BufferIsValid(buf))
2510 0 : continue;
2511 :
2512 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
2513 0 : page = BufferGetPage(buf);
2514 :
2515 : /*
2516 : * Take a copy of the local page where WAL has been applied to have a
2517 : * comparison base before masking it...
2518 : */
2519 0 : memcpy(replay_image_masked, page, BLCKSZ);
2520 :
2521 : /* No need for this page anymore now that a copy is in. */
2522 0 : UnlockReleaseBuffer(buf);
2523 :
2524 : /*
2525 : * If the block LSN is already ahead of this WAL record, we can't
2526 : * expect contents to match. This can happen if recovery is
2527 : * restarted.
2528 : */
2529 0 : if (PageGetLSN(replay_image_masked) > record->EndRecPtr)
2530 0 : continue;
2531 :
2532 : /*
2533 : * Read the contents from the backup copy, stored in WAL record and
2534 : * store it in a temporary page. There is no need to allocate a new
2535 : * page here, a local buffer is fine to hold its contents and a mask
2536 : * can be directly applied on it.
2537 : */
2538 0 : if (!RestoreBlockImage(record, block_id, primary_image_masked))
2539 0 : ereport(ERROR,
2540 : (errcode(ERRCODE_INTERNAL_ERROR),
2541 : errmsg_internal("%s", record->errormsg_buf)));
2542 :
2543 : /*
2544 : * If masking function is defined, mask both the primary and replay
2545 : * images
2546 : */
2547 0 : if (rmgr.rm_mask != NULL)
2548 : {
2549 0 : rmgr.rm_mask(replay_image_masked, blkno);
2550 0 : rmgr.rm_mask(primary_image_masked, blkno);
2551 : }
2552 :
2553 : /* Time to compare the primary and replay images. */
2554 0 : if (memcmp(replay_image_masked, primary_image_masked, BLCKSZ) != 0)
2555 : {
2556 0 : elog(FATAL,
2557 : "inconsistent page found, rel %u/%u/%u, forknum %u, blkno %u",
2558 : rlocator.spcOid, rlocator.dbOid, rlocator.relNumber,
2559 : forknum, blkno);
2560 : }
2561 : }
2562 : }
2563 :
2564 : /*
2565 : * For point-in-time recovery, this function decides whether we want to
2566 : * stop applying the XLOG before the current record.
2567 : *
2568 : * Returns true if we are stopping, false otherwise. If stopping, some
2569 : * information is saved in recoveryStopXid et al for use in annotating the
2570 : * new timeline's history file.
2571 : */
2572 : static bool
2573 5213706 : recoveryStopsBefore(XLogReaderState *record)
2574 : {
2575 5213706 : bool stopsHere = false;
2576 : uint8 xact_info;
2577 : bool isCommit;
2578 5213706 : TimestampTz recordXtime = 0;
2579 : TransactionId recordXid;
2580 :
2581 : /*
2582 : * Ignore recovery target settings when not in archive recovery (meaning
2583 : * we are in crash recovery).
2584 : */
2585 5213706 : if (!ArchiveRecoveryRequested)
2586 495246 : return false;
2587 :
2588 : /* Check if we should stop as soon as reaching consistency */
2589 4718460 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
2590 : {
2591 0 : ereport(LOG,
2592 : (errmsg("recovery stopping after reaching consistency")));
2593 :
2594 0 : recoveryStopAfter = false;
2595 0 : recoveryStopXid = InvalidTransactionId;
2596 0 : recoveryStopLSN = InvalidXLogRecPtr;
2597 0 : recoveryStopTime = 0;
2598 0 : recoveryStopName[0] = '\0';
2599 0 : return true;
2600 : }
2601 :
2602 : /* Check if target LSN has been reached */
2603 4718460 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
2604 13856 : !recoveryTargetInclusive &&
2605 666 : record->ReadRecPtr >= recoveryTargetLSN)
2606 : {
2607 2 : recoveryStopAfter = false;
2608 2 : recoveryStopXid = InvalidTransactionId;
2609 2 : recoveryStopLSN = record->ReadRecPtr;
2610 2 : recoveryStopTime = 0;
2611 2 : recoveryStopName[0] = '\0';
2612 2 : ereport(LOG,
2613 : (errmsg("recovery stopping before WAL location (LSN) \"%X/%X\"",
2614 : LSN_FORMAT_ARGS(recoveryStopLSN))));
2615 2 : return true;
2616 : }
2617 :
2618 : /* Otherwise we only consider stopping before COMMIT or ABORT records. */
2619 4718458 : if (XLogRecGetRmid(record) != RM_XACT_ID)
2620 4679392 : return false;
2621 :
2622 39066 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
2623 :
2624 39066 : if (xact_info == XLOG_XACT_COMMIT)
2625 : {
2626 35456 : isCommit = true;
2627 35456 : recordXid = XLogRecGetXid(record);
2628 : }
2629 3610 : else if (xact_info == XLOG_XACT_COMMIT_PREPARED)
2630 : {
2631 46 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
2632 : xl_xact_parsed_commit parsed;
2633 :
2634 46 : isCommit = true;
2635 46 : ParseCommitRecord(XLogRecGetInfo(record),
2636 : xlrec,
2637 : &parsed);
2638 46 : recordXid = parsed.twophase_xid;
2639 : }
2640 3564 : else if (xact_info == XLOG_XACT_ABORT)
2641 : {
2642 3132 : isCommit = false;
2643 3132 : recordXid = XLogRecGetXid(record);
2644 : }
2645 432 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
2646 : {
2647 22 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
2648 : xl_xact_parsed_abort parsed;
2649 :
2650 22 : isCommit = false;
2651 22 : ParseAbortRecord(XLogRecGetInfo(record),
2652 : xlrec,
2653 : &parsed);
2654 22 : recordXid = parsed.twophase_xid;
2655 : }
2656 : else
2657 410 : return false;
2658 :
2659 38656 : if (recoveryTarget == RECOVERY_TARGET_XID && !recoveryTargetInclusive)
2660 : {
2661 : /*
2662 : * There can be only one transaction end record with this exact
2663 : * transactionid
2664 : *
2665 : * when testing for an xid, we MUST test for equality only, since
2666 : * transactions are numbered in the order they start, not the order
2667 : * they complete. A higher numbered xid will complete before you about
2668 : * 50% of the time...
2669 : */
2670 0 : stopsHere = (recordXid == recoveryTargetXid);
2671 : }
2672 :
2673 : /*
2674 : * Note: we must fetch recordXtime regardless of recoveryTarget setting.
2675 : * We don't expect getRecordTimestamp ever to fail, since we already know
2676 : * this is a commit or abort record; but test its result anyway.
2677 : */
2678 38656 : if (getRecordTimestamp(record, &recordXtime) &&
2679 38656 : recoveryTarget == RECOVERY_TARGET_TIME)
2680 : {
2681 : /*
2682 : * There can be many transactions that share the same commit time, so
2683 : * we stop after the last one, if we are inclusive, or stop at the
2684 : * first one if we are exclusive
2685 : */
2686 0 : if (recoveryTargetInclusive)
2687 0 : stopsHere = (recordXtime > recoveryTargetTime);
2688 : else
2689 0 : stopsHere = (recordXtime >= recoveryTargetTime);
2690 : }
2691 :
2692 38656 : if (stopsHere)
2693 : {
2694 0 : recoveryStopAfter = false;
2695 0 : recoveryStopXid = recordXid;
2696 0 : recoveryStopTime = recordXtime;
2697 0 : recoveryStopLSN = InvalidXLogRecPtr;
2698 0 : recoveryStopName[0] = '\0';
2699 :
2700 0 : if (isCommit)
2701 : {
2702 0 : ereport(LOG,
2703 : (errmsg("recovery stopping before commit of transaction %u, time %s",
2704 : recoveryStopXid,
2705 : timestamptz_to_str(recoveryStopTime))));
2706 : }
2707 : else
2708 : {
2709 0 : ereport(LOG,
2710 : (errmsg("recovery stopping before abort of transaction %u, time %s",
2711 : recoveryStopXid,
2712 : timestamptz_to_str(recoveryStopTime))));
2713 : }
2714 : }
2715 :
2716 38656 : return stopsHere;
2717 : }
2718 :
2719 : /*
2720 : * Same as recoveryStopsBefore, but called after applying the record.
2721 : *
2722 : * We also track the timestamp of the latest applied COMMIT/ABORT
2723 : * record in XLogRecoveryCtl->recoveryLastXTime.
2724 : */
2725 : static bool
2726 5213700 : recoveryStopsAfter(XLogReaderState *record)
2727 : {
2728 : uint8 info;
2729 : uint8 xact_info;
2730 : uint8 rmid;
2731 5213700 : TimestampTz recordXtime = 0;
2732 :
2733 : /*
2734 : * Ignore recovery target settings when not in archive recovery (meaning
2735 : * we are in crash recovery).
2736 : */
2737 5213700 : if (!ArchiveRecoveryRequested)
2738 495246 : return false;
2739 :
2740 4718454 : info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2741 4718454 : rmid = XLogRecGetRmid(record);
2742 :
2743 : /*
2744 : * There can be many restore points that share the same name; we stop at
2745 : * the first one.
2746 : */
2747 4718454 : if (recoveryTarget == RECOVERY_TARGET_NAME &&
2748 44 : rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
2749 : {
2750 : xl_restore_point *recordRestorePointData;
2751 :
2752 6 : recordRestorePointData = (xl_restore_point *) XLogRecGetData(record);
2753 :
2754 6 : if (strcmp(recordRestorePointData->rp_name, recoveryTargetName) == 0)
2755 : {
2756 4 : recoveryStopAfter = true;
2757 4 : recoveryStopXid = InvalidTransactionId;
2758 4 : recoveryStopLSN = InvalidXLogRecPtr;
2759 4 : (void) getRecordTimestamp(record, &recoveryStopTime);
2760 4 : strlcpy(recoveryStopName, recordRestorePointData->rp_name, MAXFNAMELEN);
2761 :
2762 4 : ereport(LOG,
2763 : (errmsg("recovery stopping at restore point \"%s\", time %s",
2764 : recoveryStopName,
2765 : timestamptz_to_str(recoveryStopTime))));
2766 4 : return true;
2767 : }
2768 : }
2769 :
2770 : /* Check if the target LSN has been reached */
2771 4718450 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
2772 13190 : recoveryTargetInclusive &&
2773 13190 : record->ReadRecPtr >= recoveryTargetLSN)
2774 : {
2775 6 : recoveryStopAfter = true;
2776 6 : recoveryStopXid = InvalidTransactionId;
2777 6 : recoveryStopLSN = record->ReadRecPtr;
2778 6 : recoveryStopTime = 0;
2779 6 : recoveryStopName[0] = '\0';
2780 6 : ereport(LOG,
2781 : (errmsg("recovery stopping after WAL location (LSN) \"%X/%X\"",
2782 : LSN_FORMAT_ARGS(recoveryStopLSN))));
2783 6 : return true;
2784 : }
2785 :
2786 4718444 : if (rmid != RM_XACT_ID)
2787 4679382 : return false;
2788 :
2789 39062 : xact_info = info & XLOG_XACT_OPMASK;
2790 :
2791 39062 : if (xact_info == XLOG_XACT_COMMIT ||
2792 3564 : xact_info == XLOG_XACT_COMMIT_PREPARED ||
2793 432 : xact_info == XLOG_XACT_ABORT ||
2794 : xact_info == XLOG_XACT_ABORT_PREPARED)
2795 : {
2796 : TransactionId recordXid;
2797 :
2798 : /* Update the last applied transaction timestamp */
2799 38652 : if (getRecordTimestamp(record, &recordXtime))
2800 38652 : SetLatestXTime(recordXtime);
2801 :
2802 : /* Extract the XID of the committed/aborted transaction */
2803 38652 : if (xact_info == XLOG_XACT_COMMIT_PREPARED)
2804 : {
2805 46 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
2806 : xl_xact_parsed_commit parsed;
2807 :
2808 46 : ParseCommitRecord(XLogRecGetInfo(record),
2809 : xlrec,
2810 : &parsed);
2811 46 : recordXid = parsed.twophase_xid;
2812 : }
2813 38606 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
2814 : {
2815 22 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
2816 : xl_xact_parsed_abort parsed;
2817 :
2818 22 : ParseAbortRecord(XLogRecGetInfo(record),
2819 : xlrec,
2820 : &parsed);
2821 22 : recordXid = parsed.twophase_xid;
2822 : }
2823 : else
2824 38584 : recordXid = XLogRecGetXid(record);
2825 :
2826 : /*
2827 : * There can be only one transaction end record with this exact
2828 : * transactionid
2829 : *
2830 : * when testing for an xid, we MUST test for equality only, since
2831 : * transactions are numbered in the order they start, not the order
2832 : * they complete. A higher numbered xid will complete before you about
2833 : * 50% of the time...
2834 : */
2835 38652 : if (recoveryTarget == RECOVERY_TARGET_XID && recoveryTargetInclusive &&
2836 0 : recordXid == recoveryTargetXid)
2837 : {
2838 0 : recoveryStopAfter = true;
2839 0 : recoveryStopXid = recordXid;
2840 0 : recoveryStopTime = recordXtime;
2841 0 : recoveryStopLSN = InvalidXLogRecPtr;
2842 0 : recoveryStopName[0] = '\0';
2843 :
2844 0 : if (xact_info == XLOG_XACT_COMMIT ||
2845 : xact_info == XLOG_XACT_COMMIT_PREPARED)
2846 : {
2847 0 : ereport(LOG,
2848 : (errmsg("recovery stopping after commit of transaction %u, time %s",
2849 : recoveryStopXid,
2850 : timestamptz_to_str(recoveryStopTime))));
2851 : }
2852 0 : else if (xact_info == XLOG_XACT_ABORT ||
2853 : xact_info == XLOG_XACT_ABORT_PREPARED)
2854 : {
2855 0 : ereport(LOG,
2856 : (errmsg("recovery stopping after abort of transaction %u, time %s",
2857 : recoveryStopXid,
2858 : timestamptz_to_str(recoveryStopTime))));
2859 : }
2860 0 : return true;
2861 : }
2862 : }
2863 :
2864 : /* Check if we should stop as soon as reaching consistency */
2865 39062 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
2866 : {
2867 0 : ereport(LOG,
2868 : (errmsg("recovery stopping after reaching consistency")));
2869 :
2870 0 : recoveryStopAfter = true;
2871 0 : recoveryStopXid = InvalidTransactionId;
2872 0 : recoveryStopTime = 0;
2873 0 : recoveryStopLSN = InvalidXLogRecPtr;
2874 0 : recoveryStopName[0] = '\0';
2875 0 : return true;
2876 : }
2877 :
2878 39062 : return false;
2879 : }
2880 :
2881 : /*
2882 : * Create a comment for the history file to explain why and where
2883 : * timeline changed.
2884 : */
2885 : static char *
2886 1420 : getRecoveryStopReason(void)
2887 : {
2888 : char reason[200];
2889 :
2890 1420 : if (recoveryTarget == RECOVERY_TARGET_XID)
2891 0 : snprintf(reason, sizeof(reason),
2892 : "%s transaction %u",
2893 0 : recoveryStopAfter ? "after" : "before",
2894 : recoveryStopXid);
2895 1420 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
2896 0 : snprintf(reason, sizeof(reason),
2897 : "%s %s\n",
2898 0 : recoveryStopAfter ? "after" : "before",
2899 : timestamptz_to_str(recoveryStopTime));
2900 1420 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
2901 10 : snprintf(reason, sizeof(reason),
2902 : "%s LSN %X/%X\n",
2903 10 : recoveryStopAfter ? "after" : "before",
2904 10 : LSN_FORMAT_ARGS(recoveryStopLSN));
2905 1410 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
2906 6 : snprintf(reason, sizeof(reason),
2907 : "at restore point \"%s\"",
2908 : recoveryStopName);
2909 1404 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
2910 0 : snprintf(reason, sizeof(reason), "reached consistency");
2911 : else
2912 1404 : snprintf(reason, sizeof(reason), "no recovery target specified");
2913 :
2914 1420 : return pstrdup(reason);
2915 : }
2916 :
2917 : /*
2918 : * Wait until shared recoveryPauseState is set to RECOVERY_NOT_PAUSED.
2919 : *
2920 : * endOfRecovery is true if the recovery target is reached and
2921 : * the paused state starts at the end of recovery because of
2922 : * recovery_target_action=pause, and false otherwise.
2923 : */
2924 : static void
2925 4 : recoveryPausesHere(bool endOfRecovery)
2926 : {
2927 : /* Don't pause unless users can connect! */
2928 4 : if (!LocalHotStandbyActive)
2929 0 : return;
2930 :
2931 : /* Don't pause after standby promotion has been triggered */
2932 4 : if (LocalPromoteIsTriggered)
2933 0 : return;
2934 :
2935 4 : if (endOfRecovery)
2936 0 : ereport(LOG,
2937 : (errmsg("pausing at the end of recovery"),
2938 : errhint("Execute pg_wal_replay_resume() to promote.")));
2939 : else
2940 4 : ereport(LOG,
2941 : (errmsg("recovery has paused"),
2942 : errhint("Execute pg_wal_replay_resume() to continue.")));
2943 :
2944 : /* loop until recoveryPauseState is set to RECOVERY_NOT_PAUSED */
2945 12 : while (GetRecoveryPauseState() != RECOVERY_NOT_PAUSED)
2946 : {
2947 10 : HandleStartupProcInterrupts();
2948 10 : if (CheckForStandbyTrigger())
2949 2 : return;
2950 :
2951 : /*
2952 : * If recovery pause is requested then set it paused. While we are in
2953 : * the loop, user might resume and pause again so set this every time.
2954 : */
2955 8 : ConfirmRecoveryPaused();
2956 :
2957 : /*
2958 : * We wait on a condition variable that will wake us as soon as the
2959 : * pause ends, but we use a timeout so we can check the above exit
2960 : * condition periodically too.
2961 : */
2962 8 : ConditionVariableTimedSleep(&XLogRecoveryCtl->recoveryNotPausedCV, 1000,
2963 : WAIT_EVENT_RECOVERY_PAUSE);
2964 : }
2965 2 : ConditionVariableCancelSleep();
2966 : }
2967 :
2968 : /*
2969 : * When recovery_min_apply_delay is set, we wait long enough to make sure
2970 : * certain record types are applied at least that interval behind the primary.
2971 : *
2972 : * Returns true if we waited.
2973 : *
2974 : * Note that the delay is calculated between the WAL record log time and
2975 : * the current time on standby. We would prefer to keep track of when this
2976 : * standby received each WAL record, which would allow a more consistent
2977 : * approach and one not affected by time synchronisation issues, but that
2978 : * is significantly more effort and complexity for little actual gain in
2979 : * usability.
2980 : */
2981 : static bool
2982 5213704 : recoveryApplyDelay(XLogReaderState *record)
2983 : {
2984 : uint8 xact_info;
2985 : TimestampTz xtime;
2986 : TimestampTz delayUntil;
2987 : long msecs;
2988 :
2989 : /* nothing to do if no delay configured */
2990 5213704 : if (recovery_min_apply_delay <= 0)
2991 5213704 : return false;
2992 :
2993 : /* no delay is applied on a database not yet consistent */
2994 0 : if (!reachedConsistency)
2995 0 : return false;
2996 :
2997 : /* nothing to do if crash recovery is requested */
2998 0 : if (!ArchiveRecoveryRequested)
2999 0 : return false;
3000 :
3001 : /*
3002 : * Is it a COMMIT record?
3003 : *
3004 : * We deliberately choose not to delay aborts since they have no effect on
3005 : * MVCC. We already allow replay of records that don't have a timestamp,
3006 : * so there is already opportunity for issues caused by early conflicts on
3007 : * standbys.
3008 : */
3009 0 : if (XLogRecGetRmid(record) != RM_XACT_ID)
3010 0 : return false;
3011 :
3012 0 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
3013 :
3014 0 : if (xact_info != XLOG_XACT_COMMIT &&
3015 : xact_info != XLOG_XACT_COMMIT_PREPARED)
3016 0 : return false;
3017 :
3018 0 : if (!getRecordTimestamp(record, &xtime))
3019 0 : return false;
3020 :
3021 0 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
3022 :
3023 : /*
3024 : * Exit without arming the latch if it's already past time to apply this
3025 : * record
3026 : */
3027 0 : msecs = TimestampDifferenceMilliseconds(GetCurrentTimestamp(), delayUntil);
3028 0 : if (msecs <= 0)
3029 0 : return false;
3030 :
3031 : while (true)
3032 : {
3033 0 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3034 :
3035 : /* This might change recovery_min_apply_delay. */
3036 0 : HandleStartupProcInterrupts();
3037 :
3038 0 : if (CheckForStandbyTrigger())
3039 0 : break;
3040 :
3041 : /*
3042 : * Recalculate delayUntil as recovery_min_apply_delay could have
3043 : * changed while waiting in this loop.
3044 : */
3045 0 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
3046 :
3047 : /*
3048 : * Wait for difference between GetCurrentTimestamp() and delayUntil.
3049 : */
3050 0 : msecs = TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
3051 : delayUntil);
3052 :
3053 0 : if (msecs <= 0)
3054 0 : break;
3055 :
3056 0 : elog(DEBUG2, "recovery apply delay %ld milliseconds", msecs);
3057 :
3058 0 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3059 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
3060 : msecs,
3061 : WAIT_EVENT_RECOVERY_APPLY_DELAY);
3062 : }
3063 0 : return true;
3064 : }
3065 :
3066 : /*
3067 : * Get the current state of the recovery pause request.
3068 : */
3069 : RecoveryPauseState
3070 20 : GetRecoveryPauseState(void)
3071 : {
3072 : RecoveryPauseState state;
3073 :
3074 20 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3075 20 : state = XLogRecoveryCtl->recoveryPauseState;
3076 20 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3077 :
3078 20 : return state;
3079 : }
3080 :
3081 : /*
3082 : * Set the recovery pause state.
3083 : *
3084 : * If recovery pause is requested then sets the recovery pause state to
3085 : * 'pause requested' if it is not already 'paused'. Otherwise, sets it
3086 : * to 'not paused' to resume the recovery. The recovery pause will be
3087 : * confirmed by the ConfirmRecoveryPaused.
3088 : */
3089 : void
3090 82 : SetRecoveryPause(bool recoveryPause)
3091 : {
3092 82 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3093 :
3094 82 : if (!recoveryPause)
3095 78 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_NOT_PAUSED;
3096 4 : else if (XLogRecoveryCtl->recoveryPauseState == RECOVERY_NOT_PAUSED)
3097 4 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_PAUSE_REQUESTED;
3098 :
3099 82 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3100 :
3101 82 : if (!recoveryPause)
3102 78 : ConditionVariableBroadcast(&XLogRecoveryCtl->recoveryNotPausedCV);
3103 82 : }
3104 :
3105 : /*
3106 : * Confirm the recovery pause by setting the recovery pause state to
3107 : * RECOVERY_PAUSED.
3108 : */
3109 : static void
3110 8 : ConfirmRecoveryPaused(void)
3111 : {
3112 : /* If recovery pause is requested then set it paused */
3113 8 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3114 8 : if (XLogRecoveryCtl->recoveryPauseState == RECOVERY_PAUSE_REQUESTED)
3115 4 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_PAUSED;
3116 8 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3117 8 : }
3118 :
3119 :
3120 : /*
3121 : * Attempt to read the next XLOG record.
3122 : *
3123 : * Before first call, the reader needs to be positioned to the first record
3124 : * by calling XLogPrefetcherBeginRead().
3125 : *
3126 : * If no valid record is available, returns NULL, or fails if emode is PANIC.
3127 : * (emode must be either PANIC, LOG). In standby mode, retries until a valid
3128 : * record is available.
3129 : */
3130 : static XLogRecord *
3131 5217148 : ReadRecord(XLogPrefetcher *xlogprefetcher, int emode,
3132 : bool fetching_ckpt, TimeLineID replayTLI)
3133 : {
3134 : XLogRecord *record;
3135 5217148 : XLogReaderState *xlogreader = XLogPrefetcherGetReader(xlogprefetcher);
3136 5217148 : XLogPageReadPrivate *private = (XLogPageReadPrivate *) xlogreader->private_data;
3137 :
3138 : /* Pass through parameters to XLogPageRead */
3139 5217148 : private->fetching_ckpt = fetching_ckpt;
3140 5217148 : private->emode = emode;
3141 5217148 : private->randAccess = (xlogreader->ReadRecPtr == InvalidXLogRecPtr);
3142 5217148 : private->replayTLI = replayTLI;
3143 :
3144 : /* This is the first attempt to read this page. */
3145 5217148 : lastSourceFailed = false;
3146 :
3147 : for (;;)
3148 178 : {
3149 : char *errormsg;
3150 :
3151 5217326 : record = XLogPrefetcherReadRecord(xlogprefetcher, &errormsg);
3152 5217232 : if (record == NULL)
3153 : {
3154 : /*
3155 : * When we find that WAL ends in an incomplete record, keep track
3156 : * of that record. After recovery is done, we'll write a record
3157 : * to indicate to downstream WAL readers that that portion is to
3158 : * be ignored.
3159 : *
3160 : * However, when ArchiveRecoveryRequested = true, we're going to
3161 : * switch to a new timeline at the end of recovery. We will only
3162 : * copy WAL over to the new timeline up to the end of the last
3163 : * complete record, so if we did this, we would later create an
3164 : * overwrite contrecord in the wrong place, breaking everything.
3165 : */
3166 460 : if (!ArchiveRecoveryRequested &&
3167 202 : !XLogRecPtrIsInvalid(xlogreader->abortedRecPtr))
3168 : {
3169 20 : abortedRecPtr = xlogreader->abortedRecPtr;
3170 20 : missingContrecPtr = xlogreader->missingContrecPtr;
3171 : }
3172 :
3173 460 : if (readFile >= 0)
3174 : {
3175 424 : close(readFile);
3176 424 : readFile = -1;
3177 : }
3178 :
3179 : /*
3180 : * We only end up here without a message when XLogPageRead()
3181 : * failed - in that case we already logged something. In
3182 : * StandbyMode that only happens if we have been triggered, so we
3183 : * shouldn't loop anymore in that case.
3184 : */
3185 460 : if (errormsg)
3186 424 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3187 : (errmsg_internal("%s", errormsg) /* already translated */ ));
3188 : }
3189 :
3190 : /*
3191 : * Check page TLI is one of the expected values.
3192 : */
3193 5216772 : else if (!tliInHistory(xlogreader->latestPageTLI, expectedTLEs))
3194 : {
3195 : char fname[MAXFNAMELEN];
3196 : XLogSegNo segno;
3197 : int32 offset;
3198 :
3199 0 : XLByteToSeg(xlogreader->latestPagePtr, segno, wal_segment_size);
3200 0 : offset = XLogSegmentOffset(xlogreader->latestPagePtr,
3201 : wal_segment_size);
3202 0 : XLogFileName(fname, xlogreader->seg.ws_tli, segno,
3203 : wal_segment_size);
3204 0 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3205 : (errmsg("unexpected timeline ID %u in WAL segment %s, LSN %X/%X, offset %u",
3206 : xlogreader->latestPageTLI,
3207 : fname,
3208 : LSN_FORMAT_ARGS(xlogreader->latestPagePtr),
3209 : offset)));
3210 0 : record = NULL;
3211 : }
3212 :
3213 5217232 : if (record)
3214 : {
3215 : /* Great, got a record */
3216 5217054 : return record;
3217 : }
3218 : else
3219 : {
3220 : /* No valid record available from this source */
3221 460 : lastSourceFailed = true;
3222 :
3223 : /*
3224 : * If archive recovery was requested, but we were still doing
3225 : * crash recovery, switch to archive recovery and retry using the
3226 : * offline archive. We have now replayed all the valid WAL in
3227 : * pg_wal, so we are presumably now consistent.
3228 : *
3229 : * We require that there's at least some valid WAL present in
3230 : * pg_wal, however (!fetching_ckpt). We could recover using the
3231 : * WAL from the archive, even if pg_wal is completely empty, but
3232 : * we'd have no idea how far we'd have to replay to reach
3233 : * consistency. So err on the safe side and give up.
3234 : */
3235 460 : if (!InArchiveRecovery && ArchiveRecoveryRequested &&
3236 4 : !fetching_ckpt)
3237 : {
3238 4 : ereport(DEBUG1,
3239 : (errmsg_internal("reached end of WAL in pg_wal, entering archive recovery")));
3240 4 : InArchiveRecovery = true;
3241 4 : if (StandbyModeRequested)
3242 4 : EnableStandbyMode();
3243 :
3244 4 : SwitchIntoArchiveRecovery(xlogreader->EndRecPtr, replayTLI);
3245 4 : minRecoveryPoint = xlogreader->EndRecPtr;
3246 4 : minRecoveryPointTLI = replayTLI;
3247 :
3248 4 : CheckRecoveryConsistency();
3249 :
3250 : /*
3251 : * Before we retry, reset lastSourceFailed and currentSource
3252 : * so that we will check the archive next.
3253 : */
3254 4 : lastSourceFailed = false;
3255 4 : currentSource = XLOG_FROM_ANY;
3256 :
3257 178 : continue;
3258 : }
3259 :
3260 : /* In standby mode, loop back to retry. Otherwise, give up. */
3261 456 : if (StandbyMode && !CheckForStandbyTrigger())
3262 174 : continue;
3263 : else
3264 282 : return NULL;
3265 : }
3266 : }
3267 : }
3268 :
3269 : /*
3270 : * Read the XLOG page containing targetPagePtr into readBuf (if not read
3271 : * already). Returns number of bytes read, if the page is read successfully,
3272 : * or XLREAD_FAIL in case of errors. When errors occur, they are ereport'ed,
3273 : * but only if they have not been previously reported.
3274 : *
3275 : * See XLogReaderRoutine.page_read for more details.
3276 : *
3277 : * While prefetching, xlogreader->nonblocking may be set. In that case,
3278 : * returns XLREAD_WOULDBLOCK if we'd otherwise have to wait for more WAL.
3279 : *
3280 : * This is responsible for restoring files from archive as needed, as well
3281 : * as for waiting for the requested WAL record to arrive in standby mode.
3282 : *
3283 : * xlogreader->private_data->emode specifies the log level used for reporting
3284 : * "file not found" or "end of WAL" situations in archive recovery, or in
3285 : * standby mode when promotion is triggered. If set to WARNING or below,
3286 : * XLogPageRead() returns XLREAD_FAIL in those situations, on higher log
3287 : * levels the ereport() won't return.
3288 : *
3289 : * In standby mode, if after a successful return of XLogPageRead() the
3290 : * caller finds the record it's interested in to be broken, it should
3291 : * ereport the error with the level determined by
3292 : * emode_for_corrupt_record(), and then set lastSourceFailed
3293 : * and call XLogPageRead() again with the same arguments. This lets
3294 : * XLogPageRead() to try fetching the record from another source, or to
3295 : * sleep and retry.
3296 : */
3297 : static int
3298 174424 : XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr, int reqLen,
3299 : XLogRecPtr targetRecPtr, char *readBuf)
3300 : {
3301 174424 : XLogPageReadPrivate *private =
3302 : (XLogPageReadPrivate *) xlogreader->private_data;
3303 174424 : int emode = private->emode;
3304 : uint32 targetPageOff;
3305 : XLogSegNo targetSegNo PG_USED_FOR_ASSERTS_ONLY;
3306 : int r;
3307 :
3308 174424 : XLByteToSeg(targetPagePtr, targetSegNo, wal_segment_size);
3309 174424 : targetPageOff = XLogSegmentOffset(targetPagePtr, wal_segment_size);
3310 :
3311 : /*
3312 : * See if we need to switch to a new segment because the requested record
3313 : * is not in the currently open one.
3314 : */
3315 174424 : if (readFile >= 0 &&
3316 172230 : !XLByteInSeg(targetPagePtr, readSegNo, wal_segment_size))
3317 : {
3318 : /*
3319 : * Request a restartpoint if we've replayed too much xlog since the
3320 : * last one.
3321 : */
3322 278 : if (ArchiveRecoveryRequested && IsUnderPostmaster)
3323 : {
3324 254 : if (XLogCheckpointNeeded(readSegNo))
3325 : {
3326 68 : (void) GetRedoRecPtr();
3327 68 : if (XLogCheckpointNeeded(readSegNo))
3328 60 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
3329 : }
3330 : }
3331 :
3332 278 : close(readFile);
3333 278 : readFile = -1;
3334 278 : readSource = XLOG_FROM_ANY;
3335 : }
3336 :
3337 174424 : XLByteToSeg(targetPagePtr, readSegNo, wal_segment_size);
3338 :
3339 174432 : retry:
3340 : /* See if we need to retrieve more data */
3341 174432 : if (readFile < 0 ||
3342 171952 : (readSource == XLOG_FROM_STREAM &&
3343 147592 : flushedUpto < targetPagePtr + reqLen))
3344 : {
3345 62598 : if (readFile >= 0 &&
3346 60118 : xlogreader->nonblocking &&
3347 29662 : readSource == XLOG_FROM_STREAM &&
3348 29662 : flushedUpto < targetPagePtr + reqLen)
3349 29662 : return XLREAD_WOULDBLOCK;
3350 :
3351 32842 : switch (WaitForWALToBecomeAvailable(targetPagePtr + reqLen,
3352 32936 : private->randAccess,
3353 32936 : private->fetching_ckpt,
3354 : targetRecPtr,
3355 : private->replayTLI,
3356 : xlogreader->EndRecPtr,
3357 32936 : xlogreader->nonblocking))
3358 : {
3359 186 : case XLREAD_WOULDBLOCK:
3360 186 : return XLREAD_WOULDBLOCK;
3361 66 : case XLREAD_FAIL:
3362 66 : if (readFile >= 0)
3363 0 : close(readFile);
3364 66 : readFile = -1;
3365 66 : readLen = 0;
3366 66 : readSource = XLOG_FROM_ANY;
3367 66 : return XLREAD_FAIL;
3368 32590 : case XLREAD_SUCCESS:
3369 32590 : break;
3370 : }
3371 111834 : }
3372 :
3373 : /*
3374 : * At this point, we have the right segment open and if we're streaming we
3375 : * know the requested record is in it.
3376 : */
3377 : Assert(readFile != -1);
3378 :
3379 : /*
3380 : * If the current segment is being streamed from the primary, calculate
3381 : * how much of the current page we have received already. We know the
3382 : * requested record has been received, but this is for the benefit of
3383 : * future calls, to allow quick exit at the top of this function.
3384 : */
3385 144424 : if (readSource == XLOG_FROM_STREAM)
3386 : {
3387 118032 : if (((targetPagePtr) / XLOG_BLCKSZ) != (flushedUpto / XLOG_BLCKSZ))
3388 87942 : readLen = XLOG_BLCKSZ;
3389 : else
3390 30090 : readLen = XLogSegmentOffset(flushedUpto, wal_segment_size) -
3391 : targetPageOff;
3392 : }
3393 : else
3394 26392 : readLen = XLOG_BLCKSZ;
3395 :
3396 : /* Read the requested page */
3397 144424 : readOff = targetPageOff;
3398 :
3399 144424 : pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
3400 144424 : r = pg_pread(readFile, readBuf, XLOG_BLCKSZ, (off_t) readOff);
3401 144424 : if (r != XLOG_BLCKSZ)
3402 : {
3403 : char fname[MAXFNAMELEN];
3404 0 : int save_errno = errno;
3405 :
3406 0 : pgstat_report_wait_end();
3407 0 : XLogFileName(fname, curFileTLI, readSegNo, wal_segment_size);
3408 0 : if (r < 0)
3409 : {
3410 0 : errno = save_errno;
3411 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
3412 : (errcode_for_file_access(),
3413 : errmsg("could not read from WAL segment %s, LSN %X/%X, offset %u: %m",
3414 : fname, LSN_FORMAT_ARGS(targetPagePtr),
3415 : readOff)));
3416 : }
3417 : else
3418 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
3419 : (errcode(ERRCODE_DATA_CORRUPTED),
3420 : errmsg("could not read from WAL segment %s, LSN %X/%X, offset %u: read %d of %zu",
3421 : fname, LSN_FORMAT_ARGS(targetPagePtr),
3422 : readOff, r, (Size) XLOG_BLCKSZ)));
3423 0 : goto next_record_is_invalid;
3424 : }
3425 144424 : pgstat_report_wait_end();
3426 :
3427 : Assert(targetSegNo == readSegNo);
3428 : Assert(targetPageOff == readOff);
3429 : Assert(reqLen <= readLen);
3430 :
3431 144424 : xlogreader->seg.ws_tli = curFileTLI;
3432 :
3433 : /*
3434 : * Check the page header immediately, so that we can retry immediately if
3435 : * it's not valid. This may seem unnecessary, because ReadPageInternal()
3436 : * validates the page header anyway, and would propagate the failure up to
3437 : * ReadRecord(), which would retry. However, there's a corner case with
3438 : * continuation records, if a record is split across two pages such that
3439 : * we would need to read the two pages from different sources. For
3440 : * example, imagine a scenario where a streaming replica is started up,
3441 : * and replay reaches a record that's split across two WAL segments. The
3442 : * first page is only available locally, in pg_wal, because it's already
3443 : * been recycled on the primary. The second page, however, is not present
3444 : * in pg_wal, and we should stream it from the primary. There is a
3445 : * recycled WAL segment present in pg_wal, with garbage contents, however.
3446 : * We would read the first page from the local WAL segment, but when
3447 : * reading the second page, we would read the bogus, recycled, WAL
3448 : * segment. If we didn't catch that case here, we would never recover,
3449 : * because ReadRecord() would retry reading the whole record from the
3450 : * beginning.
3451 : *
3452 : * Of course, this only catches errors in the page header, which is what
3453 : * happens in the case of a recycled WAL segment. Other kinds of errors or
3454 : * corruption still has the same problem. But this at least fixes the
3455 : * common case, which can happen as part of normal operation.
3456 : *
3457 : * Validating the page header is cheap enough that doing it twice
3458 : * shouldn't be a big deal from a performance point of view.
3459 : *
3460 : * When not in standby mode, an invalid page header should cause recovery
3461 : * to end, not retry reading the page, so we don't need to validate the
3462 : * page header here for the retry. Instead, ReadPageInternal() is
3463 : * responsible for the validation.
3464 : */
3465 144424 : if (StandbyMode &&
3466 124364 : !XLogReaderValidatePageHeader(xlogreader, targetPagePtr, readBuf))
3467 : {
3468 : /*
3469 : * Emit this error right now then retry this page immediately. Use
3470 : * errmsg_internal() because the message was already translated.
3471 : */
3472 10 : if (xlogreader->errormsg_buf[0])
3473 10 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3474 : (errmsg_internal("%s", xlogreader->errormsg_buf)));
3475 :
3476 : /* reset any error XLogReaderValidatePageHeader() might have set */
3477 10 : XLogReaderResetError(xlogreader);
3478 10 : goto next_record_is_invalid;
3479 : }
3480 :
3481 144414 : return readLen;
3482 :
3483 10 : next_record_is_invalid:
3484 :
3485 : /*
3486 : * If we're reading ahead, give up fast. Retries and error reporting will
3487 : * be handled by a later read when recovery catches up to this point.
3488 : */
3489 10 : if (xlogreader->nonblocking)
3490 2 : return XLREAD_WOULDBLOCK;
3491 :
3492 8 : lastSourceFailed = true;
3493 :
3494 8 : if (readFile >= 0)
3495 8 : close(readFile);
3496 8 : readFile = -1;
3497 8 : readLen = 0;
3498 8 : readSource = XLOG_FROM_ANY;
3499 :
3500 : /* In standby-mode, keep trying */
3501 8 : if (StandbyMode)
3502 8 : goto retry;
3503 : else
3504 0 : return XLREAD_FAIL;
3505 : }
3506 :
3507 : /*
3508 : * Open the WAL segment containing WAL location 'RecPtr'.
3509 : *
3510 : * The segment can be fetched via restore_command, or via walreceiver having
3511 : * streamed the record, or it can already be present in pg_wal. Checking
3512 : * pg_wal is mainly for crash recovery, but it will be polled in standby mode
3513 : * too, in case someone copies a new segment directly to pg_wal. That is not
3514 : * documented or recommended, though.
3515 : *
3516 : * If 'fetching_ckpt' is true, we're fetching a checkpoint record, and should
3517 : * prepare to read WAL starting from RedoStartLSN after this.
3518 : *
3519 : * 'RecPtr' might not point to the beginning of the record we're interested
3520 : * in, it might also point to the page or segment header. In that case,
3521 : * 'tliRecPtr' is the position of the WAL record we're interested in. It is
3522 : * used to decide which timeline to stream the requested WAL from.
3523 : *
3524 : * 'replayLSN' is the current replay LSN, so that if we scan for new
3525 : * timelines, we can reject a switch to a timeline that branched off before
3526 : * this point.
3527 : *
3528 : * If the record is not immediately available, the function returns false
3529 : * if we're not in standby mode. In standby mode, waits for it to become
3530 : * available.
3531 : *
3532 : * When the requested record becomes available, the function opens the file
3533 : * containing it (if not open already), and returns XLREAD_SUCCESS. When end
3534 : * of standby mode is triggered by the user, and there is no more WAL
3535 : * available, returns XLREAD_FAIL.
3536 : *
3537 : * If nonblocking is true, then give up immediately if we can't satisfy the
3538 : * request, returning XLREAD_WOULDBLOCK instead of waiting.
3539 : */
3540 : static XLogPageReadResult
3541 32936 : WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
3542 : bool fetching_ckpt, XLogRecPtr tliRecPtr,
3543 : TimeLineID replayTLI, XLogRecPtr replayLSN,
3544 : bool nonblocking)
3545 : {
3546 : static TimestampTz last_fail_time = 0;
3547 : TimestampTz now;
3548 32936 : bool streaming_reply_sent = false;
3549 :
3550 : /*-------
3551 : * Standby mode is implemented by a state machine:
3552 : *
3553 : * 1. Read from either archive or pg_wal (XLOG_FROM_ARCHIVE), or just
3554 : * pg_wal (XLOG_FROM_PG_WAL)
3555 : * 2. Check for promotion trigger request
3556 : * 3. Read from primary server via walreceiver (XLOG_FROM_STREAM)
3557 : * 4. Rescan timelines
3558 : * 5. Sleep wal_retrieve_retry_interval milliseconds, and loop back to 1.
3559 : *
3560 : * Failure to read from the current source advances the state machine to
3561 : * the next state.
3562 : *
3563 : * 'currentSource' indicates the current state. There are no currentSource
3564 : * values for "check trigger", "rescan timelines", and "sleep" states,
3565 : * those actions are taken when reading from the previous source fails, as
3566 : * part of advancing to the next state.
3567 : *
3568 : * If standby mode is turned off while reading WAL from stream, we move
3569 : * to XLOG_FROM_ARCHIVE and reset lastSourceFailed, to force fetching
3570 : * the files (which would be required at end of recovery, e.g., timeline
3571 : * history file) from archive or pg_wal. We don't need to kill WAL receiver
3572 : * here because it's already stopped when standby mode is turned off at
3573 : * the end of recovery.
3574 : *-------
3575 : */
3576 32936 : if (!InArchiveRecovery)
3577 1542 : currentSource = XLOG_FROM_PG_WAL;
3578 31394 : else if (currentSource == XLOG_FROM_ANY ||
3579 31194 : (!StandbyMode && currentSource == XLOG_FROM_STREAM))
3580 : {
3581 200 : lastSourceFailed = false;
3582 200 : currentSource = XLOG_FROM_ARCHIVE;
3583 : }
3584 :
3585 : for (;;)
3586 30346 : {
3587 63282 : XLogSource oldSource = currentSource;
3588 63282 : bool startWalReceiver = false;
3589 :
3590 : /*
3591 : * First check if we failed to read from the current source, and
3592 : * advance the state machine if so. The failure to read might've
3593 : * happened outside this function, e.g when a CRC check fails on a
3594 : * record, or within this loop.
3595 : */
3596 63282 : if (lastSourceFailed)
3597 : {
3598 : /*
3599 : * Don't allow any retry loops to occur during nonblocking
3600 : * readahead. Let the caller process everything that has been
3601 : * decoded already first.
3602 : */
3603 748 : if (nonblocking)
3604 126 : return XLREAD_WOULDBLOCK;
3605 :
3606 622 : switch (currentSource)
3607 : {
3608 382 : case XLOG_FROM_ARCHIVE:
3609 : case XLOG_FROM_PG_WAL:
3610 :
3611 : /*
3612 : * Check to see if promotion is requested. Note that we do
3613 : * this only after failure, so when you promote, we still
3614 : * finish replaying as much as we can from archive and
3615 : * pg_wal before failover.
3616 : */
3617 382 : if (StandbyMode && CheckForStandbyTrigger())
3618 : {
3619 30 : XLogShutdownWalRcv();
3620 30 : return XLREAD_FAIL;
3621 : }
3622 :
3623 : /*
3624 : * Not in standby mode, and we've now tried the archive
3625 : * and pg_wal.
3626 : */
3627 352 : if (!StandbyMode)
3628 36 : return XLREAD_FAIL;
3629 :
3630 : /*
3631 : * Move to XLOG_FROM_STREAM state, and set to start a
3632 : * walreceiver if necessary.
3633 : */
3634 316 : currentSource = XLOG_FROM_STREAM;
3635 316 : startWalReceiver = true;
3636 316 : break;
3637 :
3638 240 : case XLOG_FROM_STREAM:
3639 :
3640 : /*
3641 : * Failure while streaming. Most likely, we got here
3642 : * because streaming replication was terminated, or
3643 : * promotion was triggered. But we also get here if we
3644 : * find an invalid record in the WAL streamed from the
3645 : * primary, in which case something is seriously wrong.
3646 : * There's little chance that the problem will just go
3647 : * away, but PANIC is not good for availability either,
3648 : * especially in hot standby mode. So, we treat that the
3649 : * same as disconnection, and retry from archive/pg_wal
3650 : * again. The WAL in the archive should be identical to
3651 : * what was streamed, so it's unlikely that it helps, but
3652 : * one can hope...
3653 : */
3654 :
3655 : /*
3656 : * We should be able to move to XLOG_FROM_STREAM only in
3657 : * standby mode.
3658 : */
3659 : Assert(StandbyMode);
3660 :
3661 : /*
3662 : * Before we leave XLOG_FROM_STREAM state, make sure that
3663 : * walreceiver is not active, so that it won't overwrite
3664 : * WAL that we restore from archive.
3665 : */
3666 240 : XLogShutdownWalRcv();
3667 :
3668 : /*
3669 : * Before we sleep, re-scan for possible new timelines if
3670 : * we were requested to recover to the latest timeline.
3671 : */
3672 240 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
3673 : {
3674 240 : if (rescanLatestTimeLine(replayTLI, replayLSN))
3675 : {
3676 12 : currentSource = XLOG_FROM_ARCHIVE;
3677 12 : break;
3678 : }
3679 : }
3680 :
3681 : /*
3682 : * XLOG_FROM_STREAM is the last state in our state
3683 : * machine, so we've exhausted all the options for
3684 : * obtaining the requested WAL. We're going to loop back
3685 : * and retry from the archive, but if it hasn't been long
3686 : * since last attempt, sleep wal_retrieve_retry_interval
3687 : * milliseconds to avoid busy-waiting.
3688 : */
3689 228 : now = GetCurrentTimestamp();
3690 228 : if (!TimestampDifferenceExceeds(last_fail_time, now,
3691 : wal_retrieve_retry_interval))
3692 : {
3693 : long wait_time;
3694 :
3695 208 : wait_time = wal_retrieve_retry_interval -
3696 104 : TimestampDifferenceMilliseconds(last_fail_time, now);
3697 :
3698 104 : elog(LOG, "waiting for WAL to become available at %X/%X",
3699 : LSN_FORMAT_ARGS(RecPtr));
3700 :
3701 : /* Do background tasks that might benefit us later. */
3702 104 : KnownAssignedTransactionIdsIdleMaintenance();
3703 :
3704 104 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3705 : WL_LATCH_SET | WL_TIMEOUT |
3706 : WL_EXIT_ON_PM_DEATH,
3707 : wait_time,
3708 : WAIT_EVENT_RECOVERY_RETRIEVE_RETRY_INTERVAL);
3709 104 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3710 104 : now = GetCurrentTimestamp();
3711 :
3712 : /* Handle interrupt signals of startup process */
3713 104 : HandleStartupProcInterrupts();
3714 : }
3715 208 : last_fail_time = now;
3716 208 : currentSource = XLOG_FROM_ARCHIVE;
3717 208 : break;
3718 :
3719 0 : default:
3720 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
3721 : }
3722 : }
3723 62534 : else if (currentSource == XLOG_FROM_PG_WAL)
3724 : {
3725 : /*
3726 : * We just successfully read a file in pg_wal. We prefer files in
3727 : * the archive over ones in pg_wal, so try the next file again
3728 : * from the archive first.
3729 : */
3730 1538 : if (InArchiveRecovery)
3731 0 : currentSource = XLOG_FROM_ARCHIVE;
3732 : }
3733 :
3734 63070 : if (currentSource != oldSource)
3735 536 : elog(DEBUG2, "switched WAL source from %s to %s after %s",
3736 : xlogSourceNames[oldSource], xlogSourceNames[currentSource],
3737 : lastSourceFailed ? "failure" : "success");
3738 :
3739 : /*
3740 : * We've now handled possible failure. Try to read from the chosen
3741 : * source.
3742 : */
3743 63070 : lastSourceFailed = false;
3744 :
3745 63070 : switch (currentSource)
3746 : {
3747 2328 : case XLOG_FROM_ARCHIVE:
3748 : case XLOG_FROM_PG_WAL:
3749 :
3750 : /*
3751 : * WAL receiver must not be running when reading WAL from
3752 : * archive or pg_wal.
3753 : */
3754 : Assert(!WalRcvStreaming());
3755 :
3756 : /* Close any old file we might have open. */
3757 2328 : if (readFile >= 0)
3758 : {
3759 134 : close(readFile);
3760 134 : readFile = -1;
3761 : }
3762 : /* Reset curFileTLI if random fetch. */
3763 2328 : if (randAccess)
3764 1806 : curFileTLI = 0;
3765 :
3766 : /*
3767 : * Try to restore the file from archive, or read an existing
3768 : * file from pg_wal.
3769 : */
3770 2328 : readFile = XLogFileReadAnyTLI(readSegNo, DEBUG2,
3771 2328 : currentSource == XLOG_FROM_ARCHIVE ? XLOG_FROM_ANY :
3772 : currentSource);
3773 2328 : if (readFile >= 0)
3774 2032 : return XLREAD_SUCCESS; /* success! */
3775 :
3776 : /*
3777 : * Nope, not found in archive or pg_wal.
3778 : */
3779 296 : lastSourceFailed = true;
3780 296 : break;
3781 :
3782 60742 : case XLOG_FROM_STREAM:
3783 : {
3784 : bool havedata;
3785 :
3786 : /*
3787 : * We should be able to move to XLOG_FROM_STREAM only in
3788 : * standby mode.
3789 : */
3790 : Assert(StandbyMode);
3791 :
3792 : /*
3793 : * First, shutdown walreceiver if its restart has been
3794 : * requested -- but no point if we're already slated for
3795 : * starting it.
3796 : */
3797 60742 : if (pendingWalRcvRestart && !startWalReceiver)
3798 : {
3799 6 : XLogShutdownWalRcv();
3800 :
3801 : /*
3802 : * Re-scan for possible new timelines if we were
3803 : * requested to recover to the latest timeline.
3804 : */
3805 6 : if (recoveryTargetTimeLineGoal ==
3806 : RECOVERY_TARGET_TIMELINE_LATEST)
3807 6 : rescanLatestTimeLine(replayTLI, replayLSN);
3808 :
3809 6 : startWalReceiver = true;
3810 : }
3811 60742 : pendingWalRcvRestart = false;
3812 :
3813 : /*
3814 : * Launch walreceiver if needed.
3815 : *
3816 : * If fetching_ckpt is true, RecPtr points to the initial
3817 : * checkpoint location. In that case, we use RedoStartLSN
3818 : * as the streaming start position instead of RecPtr, so
3819 : * that when we later jump backwards to start redo at
3820 : * RedoStartLSN, we will have the logs streamed already.
3821 : */
3822 60742 : if (startWalReceiver &&
3823 322 : PrimaryConnInfo && strcmp(PrimaryConnInfo, "") != 0)
3824 : {
3825 : XLogRecPtr ptr;
3826 : TimeLineID tli;
3827 :
3828 282 : if (fetching_ckpt)
3829 : {
3830 0 : ptr = RedoStartLSN;
3831 0 : tli = RedoStartTLI;
3832 : }
3833 : else
3834 : {
3835 282 : ptr = RecPtr;
3836 :
3837 : /*
3838 : * Use the record begin position to determine the
3839 : * TLI, rather than the position we're reading.
3840 : */
3841 282 : tli = tliOfPointInHistory(tliRecPtr, expectedTLEs);
3842 :
3843 282 : if (curFileTLI > 0 && tli < curFileTLI)
3844 0 : elog(ERROR, "according to history file, WAL location %X/%X belongs to timeline %u, but previous recovered WAL file came from timeline %u",
3845 : LSN_FORMAT_ARGS(tliRecPtr),
3846 : tli, curFileTLI);
3847 : }
3848 282 : curFileTLI = tli;
3849 282 : SetInstallXLogFileSegmentActive();
3850 282 : RequestXLogStreaming(tli, ptr, PrimaryConnInfo,
3851 : PrimarySlotName,
3852 : wal_receiver_create_temp_slot);
3853 282 : flushedUpto = 0;
3854 : }
3855 :
3856 : /*
3857 : * Check if WAL receiver is active or wait to start up.
3858 : */
3859 60742 : if (!WalRcvStreaming())
3860 : {
3861 184 : lastSourceFailed = true;
3862 184 : break;
3863 : }
3864 :
3865 : /*
3866 : * Walreceiver is active, so see if new data has arrived.
3867 : *
3868 : * We only advance XLogReceiptTime when we obtain fresh
3869 : * WAL from walreceiver and observe that we had already
3870 : * processed everything before the most recent "chunk"
3871 : * that it flushed to disk. In steady state where we are
3872 : * keeping up with the incoming data, XLogReceiptTime will
3873 : * be updated on each cycle. When we are behind,
3874 : * XLogReceiptTime will not advance, so the grace time
3875 : * allotted to conflicting queries will decrease.
3876 : */
3877 60558 : if (RecPtr < flushedUpto)
3878 314 : havedata = true;
3879 : else
3880 : {
3881 : XLogRecPtr latestChunkStart;
3882 :
3883 60244 : flushedUpto = GetWalRcvFlushRecPtr(&latestChunkStart, &receiveTLI);
3884 60244 : if (RecPtr < flushedUpto && receiveTLI == curFileTLI)
3885 : {
3886 30532 : havedata = true;
3887 30532 : if (latestChunkStart <= RecPtr)
3888 : {
3889 28780 : XLogReceiptTime = GetCurrentTimestamp();
3890 28780 : SetCurrentChunkStartTime(XLogReceiptTime);
3891 : }
3892 : }
3893 : else
3894 29712 : havedata = false;
3895 : }
3896 60558 : if (havedata)
3897 : {
3898 : /*
3899 : * Great, streamed far enough. Open the file if it's
3900 : * not open already. Also read the timeline history
3901 : * file if we haven't initialized timeline history
3902 : * yet; it should be streamed over and present in
3903 : * pg_wal by now. Use XLOG_FROM_STREAM so that source
3904 : * info is set correctly and XLogReceiptTime isn't
3905 : * changed.
3906 : *
3907 : * NB: We must set readTimeLineHistory based on
3908 : * recoveryTargetTLI, not receiveTLI. Normally they'll
3909 : * be the same, but if recovery_target_timeline is
3910 : * 'latest' and archiving is configured, then it's
3911 : * possible that we managed to retrieve one or more
3912 : * new timeline history files from the archive,
3913 : * updating recoveryTargetTLI.
3914 : */
3915 30846 : if (readFile < 0)
3916 : {
3917 288 : if (!expectedTLEs)
3918 0 : expectedTLEs = readTimeLineHistory(recoveryTargetTLI);
3919 288 : readFile = XLogFileRead(readSegNo, PANIC,
3920 : receiveTLI,
3921 : XLOG_FROM_STREAM, false);
3922 : Assert(readFile >= 0);
3923 : }
3924 : else
3925 : {
3926 : /* just make sure source info is correct... */
3927 30558 : readSource = XLOG_FROM_STREAM;
3928 30558 : XLogReceiptSource = XLOG_FROM_STREAM;
3929 30558 : return XLREAD_SUCCESS;
3930 : }
3931 288 : break;
3932 : }
3933 :
3934 : /* In nonblocking mode, return rather than sleeping. */
3935 29712 : if (nonblocking)
3936 60 : return XLREAD_WOULDBLOCK;
3937 :
3938 : /*
3939 : * Data not here yet. Check for trigger, then wait for
3940 : * walreceiver to wake us up when new WAL arrives.
3941 : */
3942 29652 : if (CheckForStandbyTrigger())
3943 : {
3944 : /*
3945 : * Note that we don't return XLREAD_FAIL immediately
3946 : * here. After being triggered, we still want to
3947 : * replay all the WAL that was already streamed. It's
3948 : * in pg_wal now, so we just treat this as a failure,
3949 : * and the state machine will move on to replay the
3950 : * streamed WAL from pg_wal, and then recheck the
3951 : * trigger and exit replay.
3952 : */
3953 56 : lastSourceFailed = true;
3954 56 : break;
3955 : }
3956 :
3957 : /*
3958 : * Since we have replayed everything we have received so
3959 : * far and are about to start waiting for more WAL, let's
3960 : * tell the upstream server our replay location now so
3961 : * that pg_stat_replication doesn't show stale
3962 : * information.
3963 : */
3964 29596 : if (!streaming_reply_sent)
3965 : {
3966 26594 : WalRcvForceReply();
3967 26594 : streaming_reply_sent = true;
3968 : }
3969 :
3970 : /* Do any background tasks that might benefit us later. */
3971 29596 : KnownAssignedTransactionIdsIdleMaintenance();
3972 :
3973 : /* Update pg_stat_recovery_prefetch before sleeping. */
3974 29596 : XLogPrefetcherComputeStats(xlogprefetcher);
3975 :
3976 : /*
3977 : * Wait for more WAL to arrive, when we will be woken
3978 : * immediately by the WAL receiver.
3979 : */
3980 29596 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3981 : WL_LATCH_SET | WL_EXIT_ON_PM_DEATH,
3982 : -1L,
3983 : WAIT_EVENT_RECOVERY_WAL_STREAM);
3984 29596 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3985 29596 : break;
3986 : }
3987 :
3988 0 : default:
3989 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
3990 : }
3991 :
3992 : /*
3993 : * Check for recovery pause here so that we can confirm more quickly
3994 : * that a requested pause has actually taken effect.
3995 : */
3996 30420 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
3997 : RECOVERY_NOT_PAUSED)
3998 4 : recoveryPausesHere(false);
3999 :
4000 : /*
4001 : * This possibly-long loop needs to handle interrupts of startup
4002 : * process.
4003 : */
4004 30420 : HandleStartupProcInterrupts();
4005 : }
4006 :
4007 : return XLREAD_FAIL; /* not reached */
4008 : }
4009 :
4010 :
4011 : /*
4012 : * Determine what log level should be used to report a corrupt WAL record
4013 : * in the current WAL page, previously read by XLogPageRead().
4014 : *
4015 : * 'emode' is the error mode that would be used to report a file-not-found
4016 : * or legitimate end-of-WAL situation. Generally, we use it as-is, but if
4017 : * we're retrying the exact same record that we've tried previously, only
4018 : * complain the first time to keep the noise down. However, we only do when
4019 : * reading from pg_wal, because we don't expect any invalid records in archive
4020 : * or in records streamed from the primary. Files in the archive should be complete,
4021 : * and we should never hit the end of WAL because we stop and wait for more WAL
4022 : * to arrive before replaying it.
4023 : *
4024 : * NOTE: This function remembers the RecPtr value it was last called with,
4025 : * to suppress repeated messages about the same record. Only call this when
4026 : * you are about to ereport(), or you might cause a later message to be
4027 : * erroneously suppressed.
4028 : */
4029 : static int
4030 434 : emode_for_corrupt_record(int emode, XLogRecPtr RecPtr)
4031 : {
4032 : static XLogRecPtr lastComplaint = 0;
4033 :
4034 434 : if (readSource == XLOG_FROM_PG_WAL && emode == LOG)
4035 : {
4036 434 : if (RecPtr == lastComplaint)
4037 80 : emode = DEBUG1;
4038 : else
4039 354 : lastComplaint = RecPtr;
4040 : }
4041 434 : return emode;
4042 : }
4043 :
4044 :
4045 : /*
4046 : * Subroutine to try to fetch and validate a prior checkpoint record.
4047 : */
4048 : static XLogRecord *
4049 1520 : ReadCheckpointRecord(XLogPrefetcher *xlogprefetcher, XLogRecPtr RecPtr,
4050 : TimeLineID replayTLI)
4051 : {
4052 : XLogRecord *record;
4053 : uint8 info;
4054 :
4055 : Assert(xlogreader != NULL);
4056 :
4057 1520 : if (!XRecOffIsValid(RecPtr))
4058 : {
4059 0 : ereport(LOG,
4060 : (errmsg("invalid checkpoint location")));
4061 0 : return NULL;
4062 : }
4063 :
4064 1520 : XLogPrefetcherBeginRead(xlogprefetcher, RecPtr);
4065 1520 : record = ReadRecord(xlogprefetcher, LOG, true, replayTLI);
4066 :
4067 1520 : if (record == NULL)
4068 : {
4069 0 : ereport(LOG,
4070 : (errmsg("invalid checkpoint record")));
4071 0 : return NULL;
4072 : }
4073 1520 : if (record->xl_rmid != RM_XLOG_ID)
4074 : {
4075 0 : ereport(LOG,
4076 : (errmsg("invalid resource manager ID in checkpoint record")));
4077 0 : return NULL;
4078 : }
4079 1520 : info = record->xl_info & ~XLR_INFO_MASK;
4080 1520 : if (info != XLOG_CHECKPOINT_SHUTDOWN &&
4081 : info != XLOG_CHECKPOINT_ONLINE)
4082 : {
4083 0 : ereport(LOG,
4084 : (errmsg("invalid xl_info in checkpoint record")));
4085 0 : return NULL;
4086 : }
4087 1520 : if (record->xl_tot_len != SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(CheckPoint))
4088 : {
4089 0 : ereport(LOG,
4090 : (errmsg("invalid length of checkpoint record")));
4091 0 : return NULL;
4092 : }
4093 1520 : return record;
4094 : }
4095 :
4096 : /*
4097 : * Scan for new timelines that might have appeared in the archive since we
4098 : * started recovery.
4099 : *
4100 : * If there are any, the function changes recovery target TLI to the latest
4101 : * one and returns 'true'.
4102 : */
4103 : static bool
4104 246 : rescanLatestTimeLine(TimeLineID replayTLI, XLogRecPtr replayLSN)
4105 : {
4106 : List *newExpectedTLEs;
4107 : bool found;
4108 : ListCell *cell;
4109 : TimeLineID newtarget;
4110 246 : TimeLineID oldtarget = recoveryTargetTLI;
4111 246 : TimeLineHistoryEntry *currentTle = NULL;
4112 :
4113 246 : newtarget = findNewestTimeLine(recoveryTargetTLI);
4114 246 : if (newtarget == recoveryTargetTLI)
4115 : {
4116 : /* No new timelines found */
4117 234 : return false;
4118 : }
4119 :
4120 : /*
4121 : * Determine the list of expected TLIs for the new TLI
4122 : */
4123 :
4124 12 : newExpectedTLEs = readTimeLineHistory(newtarget);
4125 :
4126 : /*
4127 : * If the current timeline is not part of the history of the new timeline,
4128 : * we cannot proceed to it.
4129 : */
4130 12 : found = false;
4131 24 : foreach(cell, newExpectedTLEs)
4132 : {
4133 24 : currentTle = (TimeLineHistoryEntry *) lfirst(cell);
4134 :
4135 24 : if (currentTle->tli == recoveryTargetTLI)
4136 : {
4137 12 : found = true;
4138 12 : break;
4139 : }
4140 : }
4141 12 : if (!found)
4142 : {
4143 0 : ereport(LOG,
4144 : (errmsg("new timeline %u is not a child of database system timeline %u",
4145 : newtarget,
4146 : replayTLI)));
4147 0 : return false;
4148 : }
4149 :
4150 : /*
4151 : * The current timeline was found in the history file, but check that the
4152 : * next timeline was forked off from it *after* the current recovery
4153 : * location.
4154 : */
4155 12 : if (currentTle->end < replayLSN)
4156 : {
4157 0 : ereport(LOG,
4158 : (errmsg("new timeline %u forked off current database system timeline %u before current recovery point %X/%X",
4159 : newtarget,
4160 : replayTLI,
4161 : LSN_FORMAT_ARGS(replayLSN))));
4162 0 : return false;
4163 : }
4164 :
4165 : /* The new timeline history seems valid. Switch target */
4166 12 : recoveryTargetTLI = newtarget;
4167 12 : list_free_deep(expectedTLEs);
4168 12 : expectedTLEs = newExpectedTLEs;
4169 :
4170 : /*
4171 : * As in StartupXLOG(), try to ensure we have all the history files
4172 : * between the old target and new target in pg_wal.
4173 : */
4174 12 : restoreTimeLineHistoryFiles(oldtarget + 1, newtarget);
4175 :
4176 12 : ereport(LOG,
4177 : (errmsg("new target timeline is %u",
4178 : recoveryTargetTLI)));
4179 :
4180 12 : return true;
4181 : }
4182 :
4183 :
4184 : /*
4185 : * Open a logfile segment for reading (during recovery).
4186 : *
4187 : * If source == XLOG_FROM_ARCHIVE, the segment is retrieved from archive.
4188 : * Otherwise, it's assumed to be already available in pg_wal.
4189 : */
4190 : static int
4191 3376 : XLogFileRead(XLogSegNo segno, int emode, TimeLineID tli,
4192 : XLogSource source, bool notfoundOk)
4193 : {
4194 : char xlogfname[MAXFNAMELEN];
4195 : char activitymsg[MAXFNAMELEN + 16];
4196 : char path[MAXPGPATH];
4197 : int fd;
4198 :
4199 3376 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
4200 :
4201 3376 : switch (source)
4202 : {
4203 808 : case XLOG_FROM_ARCHIVE:
4204 : /* Report recovery progress in PS display */
4205 808 : snprintf(activitymsg, sizeof(activitymsg), "waiting for %s",
4206 : xlogfname);
4207 808 : set_ps_display(activitymsg);
4208 :
4209 808 : if (!RestoreArchivedFile(path, xlogfname,
4210 : "RECOVERYXLOG",
4211 : wal_segment_size,
4212 : InRedo))
4213 742 : return -1;
4214 66 : break;
4215 :
4216 2568 : case XLOG_FROM_PG_WAL:
4217 : case XLOG_FROM_STREAM:
4218 2568 : XLogFilePath(path, tli, segno, wal_segment_size);
4219 2568 : break;
4220 :
4221 0 : default:
4222 0 : elog(ERROR, "invalid XLogFileRead source %d", source);
4223 : }
4224 :
4225 : /*
4226 : * If the segment was fetched from archival storage, replace the existing
4227 : * xlog segment (if any) with the archival version.
4228 : */
4229 2634 : if (source == XLOG_FROM_ARCHIVE)
4230 : {
4231 : Assert(!IsInstallXLogFileSegmentActive());
4232 66 : KeepFileRestoredFromArchive(path, xlogfname);
4233 :
4234 : /*
4235 : * Set path to point at the new file in pg_wal.
4236 : */
4237 66 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlogfname);
4238 : }
4239 :
4240 2634 : fd = BasicOpenFile(path, O_RDONLY | PG_BINARY);
4241 2634 : if (fd >= 0)
4242 : {
4243 : /* Success! */
4244 2320 : curFileTLI = tli;
4245 :
4246 : /* Report recovery progress in PS display */
4247 2320 : snprintf(activitymsg, sizeof(activitymsg), "recovering %s",
4248 : xlogfname);
4249 2320 : set_ps_display(activitymsg);
4250 :
4251 : /* Track source of data in assorted state variables */
4252 2320 : readSource = source;
4253 2320 : XLogReceiptSource = source;
4254 : /* In FROM_STREAM case, caller tracks receipt time, not me */
4255 2320 : if (source != XLOG_FROM_STREAM)
4256 2032 : XLogReceiptTime = GetCurrentTimestamp();
4257 :
4258 2320 : return fd;
4259 : }
4260 314 : if (errno != ENOENT || !notfoundOk) /* unexpected failure? */
4261 0 : ereport(PANIC,
4262 : (errcode_for_file_access(),
4263 : errmsg("could not open file \"%s\": %m", path)));
4264 314 : return -1;
4265 : }
4266 :
4267 : /*
4268 : * Open a logfile segment for reading (during recovery).
4269 : *
4270 : * This version searches for the segment with any TLI listed in expectedTLEs.
4271 : */
4272 : static int
4273 2328 : XLogFileReadAnyTLI(XLogSegNo segno, int emode, XLogSource source)
4274 : {
4275 : char path[MAXPGPATH];
4276 : ListCell *cell;
4277 : int fd;
4278 : List *tles;
4279 :
4280 : /*
4281 : * Loop looking for a suitable timeline ID: we might need to read any of
4282 : * the timelines listed in expectedTLEs.
4283 : *
4284 : * We expect curFileTLI on entry to be the TLI of the preceding file in
4285 : * sequence, or 0 if there was no predecessor. We do not allow curFileTLI
4286 : * to go backwards; this prevents us from picking up the wrong file when a
4287 : * parent timeline extends to higher segment numbers than the child we
4288 : * want to read.
4289 : *
4290 : * If we haven't read the timeline history file yet, read it now, so that
4291 : * we know which TLIs to scan. We don't save the list in expectedTLEs,
4292 : * however, unless we actually find a valid segment. That way if there is
4293 : * neither a timeline history file nor a WAL segment in the archive, and
4294 : * streaming replication is set up, we'll read the timeline history file
4295 : * streamed from the primary when we start streaming, instead of
4296 : * recovering with a dummy history generated here.
4297 : */
4298 2328 : if (expectedTLEs)
4299 808 : tles = expectedTLEs;
4300 : else
4301 1520 : tles = readTimeLineHistory(recoveryTargetTLI);
4302 :
4303 2654 : foreach(cell, tles)
4304 : {
4305 2362 : TimeLineHistoryEntry *hent = (TimeLineHistoryEntry *) lfirst(cell);
4306 2362 : TimeLineID tli = hent->tli;
4307 :
4308 2362 : if (tli < curFileTLI)
4309 4 : break; /* don't bother looking at too-old TLIs */
4310 :
4311 : /*
4312 : * Skip scanning the timeline ID that the logfile segment to read
4313 : * doesn't belong to
4314 : */
4315 2358 : if (hent->begin != InvalidXLogRecPtr)
4316 : {
4317 118 : XLogSegNo beginseg = 0;
4318 :
4319 118 : XLByteToSeg(hent->begin, beginseg, wal_segment_size);
4320 :
4321 : /*
4322 : * The logfile segment that doesn't belong to the timeline is
4323 : * older or newer than the segment that the timeline started or
4324 : * ended at, respectively. It's sufficient to check only the
4325 : * starting segment of the timeline here. Since the timelines are
4326 : * scanned in descending order in this loop, any segments newer
4327 : * than the ending segment should belong to newer timeline and
4328 : * have already been read before. So it's not necessary to check
4329 : * the ending segment of the timeline here.
4330 : */
4331 118 : if (segno < beginseg)
4332 12 : continue;
4333 : }
4334 :
4335 2346 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_ARCHIVE)
4336 : {
4337 808 : fd = XLogFileRead(segno, emode, tli,
4338 : XLOG_FROM_ARCHIVE, true);
4339 808 : if (fd != -1)
4340 : {
4341 66 : elog(DEBUG1, "got WAL segment from archive");
4342 66 : if (!expectedTLEs)
4343 28 : expectedTLEs = tles;
4344 2032 : return fd;
4345 : }
4346 : }
4347 :
4348 2280 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_PG_WAL)
4349 : {
4350 2280 : fd = XLogFileRead(segno, emode, tli,
4351 : XLOG_FROM_PG_WAL, true);
4352 2280 : if (fd != -1)
4353 : {
4354 1966 : if (!expectedTLEs)
4355 1492 : expectedTLEs = tles;
4356 1966 : return fd;
4357 : }
4358 : }
4359 : }
4360 :
4361 : /* Couldn't find it. For simplicity, complain about front timeline */
4362 296 : XLogFilePath(path, recoveryTargetTLI, segno, wal_segment_size);
4363 296 : errno = ENOENT;
4364 296 : ereport(emode,
4365 : (errcode_for_file_access(),
4366 : errmsg("could not open file \"%s\": %m", path)));
4367 296 : return -1;
4368 : }
4369 :
4370 : /*
4371 : * Set flag to signal the walreceiver to restart. (The startup process calls
4372 : * this on noticing a relevant configuration change.)
4373 : */
4374 : void
4375 6 : StartupRequestWalReceiverRestart(void)
4376 : {
4377 6 : if (currentSource == XLOG_FROM_STREAM && WalRcvRunning())
4378 : {
4379 6 : ereport(LOG,
4380 : (errmsg("WAL receiver process shutdown requested")));
4381 :
4382 6 : pendingWalRcvRestart = true;
4383 : }
4384 6 : }
4385 :
4386 :
4387 : /*
4388 : * Has a standby promotion already been triggered?
4389 : *
4390 : * Unlike CheckForStandbyTrigger(), this works in any process
4391 : * that's connected to shared memory.
4392 : */
4393 : bool
4394 96 : PromoteIsTriggered(void)
4395 : {
4396 : /*
4397 : * We check shared state each time only until a standby promotion is
4398 : * triggered. We can't trigger a promotion again, so there's no need to
4399 : * keep checking after the shared variable has once been seen true.
4400 : */
4401 96 : if (LocalPromoteIsTriggered)
4402 76 : return true;
4403 :
4404 20 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4405 20 : LocalPromoteIsTriggered = XLogRecoveryCtl->SharedPromoteIsTriggered;
4406 20 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4407 :
4408 20 : return LocalPromoteIsTriggered;
4409 : }
4410 :
4411 : static void
4412 76 : SetPromoteIsTriggered(void)
4413 : {
4414 76 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4415 76 : XLogRecoveryCtl->SharedPromoteIsTriggered = true;
4416 76 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4417 :
4418 : /*
4419 : * Mark the recovery pause state as 'not paused' because the paused state
4420 : * ends and promotion continues if a promotion is triggered while recovery
4421 : * is paused. Otherwise pg_get_wal_replay_pause_state() can mistakenly
4422 : * return 'paused' while a promotion is ongoing.
4423 : */
4424 76 : SetRecoveryPause(false);
4425 :
4426 76 : LocalPromoteIsTriggered = true;
4427 76 : }
4428 :
4429 : /*
4430 : * Check whether a promote request has arrived.
4431 : */
4432 : static bool
4433 30258 : CheckForStandbyTrigger(void)
4434 : {
4435 30258 : if (LocalPromoteIsTriggered)
4436 88 : return true;
4437 :
4438 30170 : if (IsPromoteSignaled() && CheckPromoteSignal())
4439 : {
4440 76 : ereport(LOG, (errmsg("received promote request")));
4441 76 : RemovePromoteSignalFiles();
4442 76 : ResetPromoteSignaled();
4443 76 : SetPromoteIsTriggered();
4444 76 : return true;
4445 : }
4446 :
4447 30094 : return false;
4448 : }
4449 :
4450 : /*
4451 : * Remove the files signaling a standby promotion request.
4452 : */
4453 : void
4454 1498 : RemovePromoteSignalFiles(void)
4455 : {
4456 1498 : unlink(PROMOTE_SIGNAL_FILE);
4457 1498 : }
4458 :
4459 : /*
4460 : * Check to see if a promote request has arrived.
4461 : */
4462 : bool
4463 1330 : CheckPromoteSignal(void)
4464 : {
4465 : struct stat stat_buf;
4466 :
4467 1330 : if (stat(PROMOTE_SIGNAL_FILE, &stat_buf) == 0)
4468 154 : return true;
4469 :
4470 1176 : return false;
4471 : }
4472 :
4473 : /*
4474 : * Wake up startup process to replay newly arrived WAL, or to notice that
4475 : * failover has been requested.
4476 : */
4477 : void
4478 38978 : WakeupRecovery(void)
4479 : {
4480 38978 : SetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
4481 38978 : }
4482 :
4483 : /*
4484 : * Schedule a walreceiver wakeup in the main recovery loop.
4485 : */
4486 : void
4487 0 : XLogRequestWalReceiverReply(void)
4488 : {
4489 0 : doRequestWalReceiverReply = true;
4490 0 : }
4491 :
4492 : /*
4493 : * Is HotStandby active yet? This is only important in special backends
4494 : * since normal backends won't ever be able to connect until this returns
4495 : * true. Postmaster knows this by way of signal, not via shared memory.
4496 : *
4497 : * Unlike testing standbyState, this works in any process that's connected to
4498 : * shared memory. (And note that standbyState alone doesn't tell the truth
4499 : * anyway.)
4500 : */
4501 : bool
4502 300 : HotStandbyActive(void)
4503 : {
4504 : /*
4505 : * We check shared state each time only until Hot Standby is active. We
4506 : * can't de-activate Hot Standby, so there's no need to keep checking
4507 : * after the shared variable has once been seen true.
4508 : */
4509 300 : if (LocalHotStandbyActive)
4510 46 : return true;
4511 : else
4512 : {
4513 : /* spinlock is essential on machines with weak memory ordering! */
4514 254 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4515 254 : LocalHotStandbyActive = XLogRecoveryCtl->SharedHotStandbyActive;
4516 254 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4517 :
4518 254 : return LocalHotStandbyActive;
4519 : }
4520 : }
4521 :
4522 : /*
4523 : * Like HotStandbyActive(), but to be used only in WAL replay code,
4524 : * where we don't need to ask any other process what the state is.
4525 : */
4526 : static bool
4527 0 : HotStandbyActiveInReplay(void)
4528 : {
4529 : Assert(AmStartupProcess() || !IsPostmasterEnvironment);
4530 0 : return LocalHotStandbyActive;
4531 : }
4532 :
4533 : /*
4534 : * Get latest redo apply position.
4535 : *
4536 : * Exported to allow WALReceiver to read the pointer directly.
4537 : */
4538 : XLogRecPtr
4539 109834 : GetXLogReplayRecPtr(TimeLineID *replayTLI)
4540 : {
4541 : XLogRecPtr recptr;
4542 : TimeLineID tli;
4543 :
4544 109834 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4545 109834 : recptr = XLogRecoveryCtl->lastReplayedEndRecPtr;
4546 109834 : tli = XLogRecoveryCtl->lastReplayedTLI;
4547 109834 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4548 :
4549 109834 : if (replayTLI)
4550 2820 : *replayTLI = tli;
4551 109834 : return recptr;
4552 : }
4553 :
4554 :
4555 : /*
4556 : * Get position of last applied, or the record being applied.
4557 : *
4558 : * This is different from GetXLogReplayRecPtr() in that if a WAL
4559 : * record is currently being applied, this includes that record.
4560 : */
4561 : XLogRecPtr
4562 12130 : GetCurrentReplayRecPtr(TimeLineID *replayEndTLI)
4563 : {
4564 : XLogRecPtr recptr;
4565 : TimeLineID tli;
4566 :
4567 12130 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4568 12130 : recptr = XLogRecoveryCtl->replayEndRecPtr;
4569 12130 : tli = XLogRecoveryCtl->replayEndTLI;
4570 12130 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4571 :
4572 12130 : if (replayEndTLI)
4573 12130 : *replayEndTLI = tli;
4574 12130 : return recptr;
4575 : }
4576 :
4577 : /*
4578 : * Save timestamp of latest processed commit/abort record.
4579 : *
4580 : * We keep this in XLogRecoveryCtl, not a simple static variable, so that it can be
4581 : * seen by processes other than the startup process. Note in particular
4582 : * that CreateRestartPoint is executed in the checkpointer.
4583 : */
4584 : static void
4585 38652 : SetLatestXTime(TimestampTz xtime)
4586 : {
4587 38652 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4588 38652 : XLogRecoveryCtl->recoveryLastXTime = xtime;
4589 38652 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4590 38652 : }
4591 :
4592 : /*
4593 : * Fetch timestamp of latest processed commit/abort record.
4594 : */
4595 : TimestampTz
4596 348 : GetLatestXTime(void)
4597 : {
4598 : TimestampTz xtime;
4599 :
4600 348 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4601 348 : xtime = XLogRecoveryCtl->recoveryLastXTime;
4602 348 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4603 :
4604 348 : return xtime;
4605 : }
4606 :
4607 : /*
4608 : * Save timestamp of the next chunk of WAL records to apply.
4609 : *
4610 : * We keep this in XLogRecoveryCtl, not a simple static variable, so that it can be
4611 : * seen by all backends.
4612 : */
4613 : static void
4614 28780 : SetCurrentChunkStartTime(TimestampTz xtime)
4615 : {
4616 28780 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4617 28780 : XLogRecoveryCtl->currentChunkStartTime = xtime;
4618 28780 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4619 28780 : }
4620 :
4621 : /*
4622 : * Fetch timestamp of latest processed commit/abort record.
4623 : * Startup process maintains an accurate local copy in XLogReceiptTime
4624 : */
4625 : TimestampTz
4626 468 : GetCurrentChunkReplayStartTime(void)
4627 : {
4628 : TimestampTz xtime;
4629 :
4630 468 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4631 468 : xtime = XLogRecoveryCtl->currentChunkStartTime;
4632 468 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4633 :
4634 468 : return xtime;
4635 : }
4636 :
4637 : /*
4638 : * Returns time of receipt of current chunk of XLOG data, as well as
4639 : * whether it was received from streaming replication or from archives.
4640 : */
4641 : void
4642 58 : GetXLogReceiptTime(TimestampTz *rtime, bool *fromStream)
4643 : {
4644 : /*
4645 : * This must be executed in the startup process, since we don't export the
4646 : * relevant state to shared memory.
4647 : */
4648 : Assert(InRecovery);
4649 :
4650 58 : *rtime = XLogReceiptTime;
4651 58 : *fromStream = (XLogReceiptSource == XLOG_FROM_STREAM);
4652 58 : }
4653 :
4654 : /*
4655 : * Note that text field supplied is a parameter name and does not require
4656 : * translation
4657 : */
4658 : void
4659 1070 : RecoveryRequiresIntParameter(const char *param_name, int currValue, int minValue)
4660 : {
4661 1070 : if (currValue < minValue)
4662 : {
4663 0 : if (HotStandbyActiveInReplay())
4664 : {
4665 0 : bool warned_for_promote = false;
4666 :
4667 0 : ereport(WARNING,
4668 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4669 : errmsg("hot standby is not possible because of insufficient parameter settings"),
4670 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4671 : param_name,
4672 : currValue,
4673 : minValue)));
4674 :
4675 0 : SetRecoveryPause(true);
4676 :
4677 0 : ereport(LOG,
4678 : (errmsg("recovery has paused"),
4679 : errdetail("If recovery is unpaused, the server will shut down."),
4680 : errhint("You can then restart the server after making the necessary configuration changes.")));
4681 :
4682 0 : while (GetRecoveryPauseState() != RECOVERY_NOT_PAUSED)
4683 : {
4684 0 : HandleStartupProcInterrupts();
4685 :
4686 0 : if (CheckForStandbyTrigger())
4687 : {
4688 0 : if (!warned_for_promote)
4689 0 : ereport(WARNING,
4690 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4691 : errmsg("promotion is not possible because of insufficient parameter settings"),
4692 :
4693 : /*
4694 : * Repeat the detail from above so it's easy to find
4695 : * in the log.
4696 : */
4697 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4698 : param_name,
4699 : currValue,
4700 : minValue),
4701 : errhint("Restart the server after making the necessary configuration changes.")));
4702 0 : warned_for_promote = true;
4703 : }
4704 :
4705 : /*
4706 : * If recovery pause is requested then set it paused. While
4707 : * we are in the loop, user might resume and pause again so
4708 : * set this every time.
4709 : */
4710 0 : ConfirmRecoveryPaused();
4711 :
4712 : /*
4713 : * We wait on a condition variable that will wake us as soon
4714 : * as the pause ends, but we use a timeout so we can check the
4715 : * above conditions periodically too.
4716 : */
4717 0 : ConditionVariableTimedSleep(&XLogRecoveryCtl->recoveryNotPausedCV, 1000,
4718 : WAIT_EVENT_RECOVERY_PAUSE);
4719 : }
4720 0 : ConditionVariableCancelSleep();
4721 : }
4722 :
4723 0 : ereport(FATAL,
4724 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4725 : errmsg("recovery aborted because of insufficient parameter settings"),
4726 : /* Repeat the detail from above so it's easy to find in the log. */
4727 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4728 : param_name,
4729 : currValue,
4730 : minValue),
4731 : errhint("You can restart the server after making the necessary configuration changes.")));
4732 : }
4733 1070 : }
4734 :
4735 :
4736 : /*
4737 : * GUC check_hook for primary_slot_name
4738 : */
4739 : bool
4740 2112 : check_primary_slot_name(char **newval, void **extra, GucSource source)
4741 : {
4742 2112 : if (*newval && strcmp(*newval, "") != 0 &&
4743 280 : !ReplicationSlotValidateName(*newval, WARNING))
4744 0 : return false;
4745 :
4746 2112 : return true;
4747 : }
4748 :
4749 : /*
4750 : * Recovery target settings: Only one of the several recovery_target* settings
4751 : * may be set. Setting a second one results in an error. The global variable
4752 : * recoveryTarget tracks which kind of recovery target was chosen. Other
4753 : * variables store the actual target value (for example a string or a xid).
4754 : * The assign functions of the parameters check whether a competing parameter
4755 : * was already set. But we want to allow setting the same parameter multiple
4756 : * times. We also want to allow unsetting a parameter and setting a different
4757 : * one, so we unset recoveryTarget when the parameter is set to an empty
4758 : * string.
4759 : *
4760 : * XXX this code is broken by design. Throwing an error from a GUC assign
4761 : * hook breaks fundamental assumptions of guc.c. So long as all the variables
4762 : * for which this can happen are PGC_POSTMASTER, the consequences are limited,
4763 : * since we'd just abort postmaster startup anyway. Nonetheless it's likely
4764 : * that we have odd behaviors such as unexpected GUC ordering dependencies.
4765 : */
4766 :
4767 : static void
4768 : pg_attribute_noreturn()
4769 2 : error_multiple_recovery_targets(void)
4770 : {
4771 2 : ereport(ERROR,
4772 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4773 : errmsg("multiple recovery targets specified"),
4774 : errdetail("At most one of recovery_target, recovery_target_lsn, recovery_target_name, recovery_target_time, recovery_target_xid may be set.")));
4775 : }
4776 :
4777 : /*
4778 : * GUC check_hook for recovery_target
4779 : */
4780 : bool
4781 1836 : check_recovery_target(char **newval, void **extra, GucSource source)
4782 : {
4783 1836 : if (strcmp(*newval, "immediate") != 0 && strcmp(*newval, "") != 0)
4784 : {
4785 0 : GUC_check_errdetail("The only allowed value is \"immediate\".");
4786 0 : return false;
4787 : }
4788 1836 : return true;
4789 : }
4790 :
4791 : /*
4792 : * GUC assign_hook for recovery_target
4793 : */
4794 : void
4795 1836 : assign_recovery_target(const char *newval, void *extra)
4796 : {
4797 1836 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4798 0 : recoveryTarget != RECOVERY_TARGET_IMMEDIATE)
4799 0 : error_multiple_recovery_targets();
4800 :
4801 1836 : if (newval && strcmp(newval, "") != 0)
4802 2 : recoveryTarget = RECOVERY_TARGET_IMMEDIATE;
4803 : else
4804 1834 : recoveryTarget = RECOVERY_TARGET_UNSET;
4805 1836 : }
4806 :
4807 : /*
4808 : * GUC check_hook for recovery_target_lsn
4809 : */
4810 : bool
4811 1842 : check_recovery_target_lsn(char **newval, void **extra, GucSource source)
4812 : {
4813 1842 : if (strcmp(*newval, "") != 0)
4814 : {
4815 : XLogRecPtr lsn;
4816 : XLogRecPtr *myextra;
4817 12 : bool have_error = false;
4818 :
4819 12 : lsn = pg_lsn_in_internal(*newval, &have_error);
4820 12 : if (have_error)
4821 0 : return false;
4822 :
4823 12 : myextra = (XLogRecPtr *) guc_malloc(ERROR, sizeof(XLogRecPtr));
4824 12 : *myextra = lsn;
4825 12 : *extra = (void *) myextra;
4826 : }
4827 1842 : return true;
4828 : }
4829 :
4830 : /*
4831 : * GUC assign_hook for recovery_target_lsn
4832 : */
4833 : void
4834 1842 : assign_recovery_target_lsn(const char *newval, void *extra)
4835 : {
4836 1842 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4837 0 : recoveryTarget != RECOVERY_TARGET_LSN)
4838 0 : error_multiple_recovery_targets();
4839 :
4840 1842 : if (newval && strcmp(newval, "") != 0)
4841 : {
4842 12 : recoveryTarget = RECOVERY_TARGET_LSN;
4843 12 : recoveryTargetLSN = *((XLogRecPtr *) extra);
4844 : }
4845 : else
4846 1830 : recoveryTarget = RECOVERY_TARGET_UNSET;
4847 1842 : }
4848 :
4849 : /*
4850 : * GUC check_hook for recovery_target_name
4851 : */
4852 : bool
4853 1848 : check_recovery_target_name(char **newval, void **extra, GucSource source)
4854 : {
4855 : /* Use the value of newval directly */
4856 1848 : if (strlen(*newval) >= MAXFNAMELEN)
4857 : {
4858 0 : GUC_check_errdetail("%s is too long (maximum %d characters).",
4859 : "recovery_target_name", MAXFNAMELEN - 1);
4860 0 : return false;
4861 : }
4862 1848 : return true;
4863 : }
4864 :
4865 : /*
4866 : * GUC assign_hook for recovery_target_name
4867 : */
4868 : void
4869 1848 : assign_recovery_target_name(const char *newval, void *extra)
4870 : {
4871 1848 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4872 0 : recoveryTarget != RECOVERY_TARGET_NAME)
4873 0 : error_multiple_recovery_targets();
4874 :
4875 1848 : if (newval && strcmp(newval, "") != 0)
4876 : {
4877 12 : recoveryTarget = RECOVERY_TARGET_NAME;
4878 12 : recoveryTargetName = newval;
4879 : }
4880 : else
4881 1836 : recoveryTarget = RECOVERY_TARGET_UNSET;
4882 1848 : }
4883 :
4884 : /*
4885 : * GUC check_hook for recovery_target_time
4886 : *
4887 : * The interpretation of the recovery_target_time string can depend on the
4888 : * time zone setting, so we need to wait until after all GUC processing is
4889 : * done before we can do the final parsing of the string. This check function
4890 : * only does a parsing pass to catch syntax errors, but we store the string
4891 : * and parse it again when we need to use it.
4892 : */
4893 : bool
4894 1840 : check_recovery_target_time(char **newval, void **extra, GucSource source)
4895 : {
4896 1840 : if (strcmp(*newval, "") != 0)
4897 : {
4898 : /* reject some special values */
4899 6 : if (strcmp(*newval, "now") == 0 ||
4900 6 : strcmp(*newval, "today") == 0 ||
4901 6 : strcmp(*newval, "tomorrow") == 0 ||
4902 6 : strcmp(*newval, "yesterday") == 0)
4903 : {
4904 0 : return false;
4905 : }
4906 :
4907 : /*
4908 : * parse timestamp value (see also timestamptz_in())
4909 : */
4910 : {
4911 6 : char *str = *newval;
4912 : fsec_t fsec;
4913 : struct pg_tm tt,
4914 6 : *tm = &tt;
4915 : int tz;
4916 : int dtype;
4917 : int nf;
4918 : int dterr;
4919 : char *field[MAXDATEFIELDS];
4920 : int ftype[MAXDATEFIELDS];
4921 : char workbuf[MAXDATELEN + MAXDATEFIELDS];
4922 : DateTimeErrorExtra dtextra;
4923 : TimestampTz timestamp;
4924 :
4925 6 : dterr = ParseDateTime(str, workbuf, sizeof(workbuf),
4926 : field, ftype, MAXDATEFIELDS, &nf);
4927 6 : if (dterr == 0)
4928 6 : dterr = DecodeDateTime(field, ftype, nf,
4929 : &dtype, tm, &fsec, &tz, &dtextra);
4930 6 : if (dterr != 0)
4931 0 : return false;
4932 6 : if (dtype != DTK_DATE)
4933 0 : return false;
4934 :
4935 6 : if (tm2timestamp(tm, fsec, &tz, ×tamp) != 0)
4936 : {
4937 0 : GUC_check_errdetail("timestamp out of range: \"%s\"", str);
4938 0 : return false;
4939 : }
4940 : }
4941 : }
4942 1840 : return true;
4943 : }
4944 :
4945 : /*
4946 : * GUC assign_hook for recovery_target_time
4947 : */
4948 : void
4949 1840 : assign_recovery_target_time(const char *newval, void *extra)
4950 : {
4951 1840 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4952 2 : recoveryTarget != RECOVERY_TARGET_TIME)
4953 2 : error_multiple_recovery_targets();
4954 :
4955 1838 : if (newval && strcmp(newval, "") != 0)
4956 4 : recoveryTarget = RECOVERY_TARGET_TIME;
4957 : else
4958 1834 : recoveryTarget = RECOVERY_TARGET_UNSET;
4959 1838 : }
4960 :
4961 : /*
4962 : * GUC check_hook for recovery_target_timeline
4963 : */
4964 : bool
4965 1836 : check_recovery_target_timeline(char **newval, void **extra, GucSource source)
4966 : {
4967 : RecoveryTargetTimeLineGoal rttg;
4968 : RecoveryTargetTimeLineGoal *myextra;
4969 :
4970 1836 : if (strcmp(*newval, "current") == 0)
4971 0 : rttg = RECOVERY_TARGET_TIMELINE_CONTROLFILE;
4972 1836 : else if (strcmp(*newval, "latest") == 0)
4973 1836 : rttg = RECOVERY_TARGET_TIMELINE_LATEST;
4974 : else
4975 : {
4976 0 : rttg = RECOVERY_TARGET_TIMELINE_NUMERIC;
4977 :
4978 0 : errno = 0;
4979 0 : strtoul(*newval, NULL, 0);
4980 0 : if (errno == EINVAL || errno == ERANGE)
4981 : {
4982 0 : GUC_check_errdetail("recovery_target_timeline is not a valid number.");
4983 0 : return false;
4984 : }
4985 : }
4986 :
4987 1836 : myextra = (RecoveryTargetTimeLineGoal *) guc_malloc(ERROR, sizeof(RecoveryTargetTimeLineGoal));
4988 1836 : *myextra = rttg;
4989 1836 : *extra = (void *) myextra;
4990 :
4991 1836 : return true;
4992 : }
4993 :
4994 : /*
4995 : * GUC assign_hook for recovery_target_timeline
4996 : */
4997 : void
4998 1836 : assign_recovery_target_timeline(const char *newval, void *extra)
4999 : {
5000 1836 : recoveryTargetTimeLineGoal = *((RecoveryTargetTimeLineGoal *) extra);
5001 1836 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
5002 0 : recoveryTargetTLIRequested = (TimeLineID) strtoul(newval, NULL, 0);
5003 : else
5004 1836 : recoveryTargetTLIRequested = 0;
5005 1836 : }
5006 :
5007 : /*
5008 : * GUC check_hook for recovery_target_xid
5009 : */
5010 : bool
5011 1836 : check_recovery_target_xid(char **newval, void **extra, GucSource source)
5012 : {
5013 1836 : if (strcmp(*newval, "") != 0)
5014 : {
5015 : TransactionId xid;
5016 : TransactionId *myextra;
5017 :
5018 2 : errno = 0;
5019 2 : xid = (TransactionId) strtou64(*newval, NULL, 0);
5020 2 : if (errno == EINVAL || errno == ERANGE)
5021 0 : return false;
5022 :
5023 2 : myextra = (TransactionId *) guc_malloc(ERROR, sizeof(TransactionId));
5024 2 : *myextra = xid;
5025 2 : *extra = (void *) myextra;
5026 : }
5027 1836 : return true;
5028 : }
5029 :
5030 : /*
5031 : * GUC assign_hook for recovery_target_xid
5032 : */
5033 : void
5034 1836 : assign_recovery_target_xid(const char *newval, void *extra)
5035 : {
5036 1836 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
5037 0 : recoveryTarget != RECOVERY_TARGET_XID)
5038 0 : error_multiple_recovery_targets();
5039 :
5040 1836 : if (newval && strcmp(newval, "") != 0)
5041 : {
5042 2 : recoveryTarget = RECOVERY_TARGET_XID;
5043 2 : recoveryTargetXid = *((TransactionId *) extra);
5044 : }
5045 : else
5046 1834 : recoveryTarget = RECOVERY_TARGET_UNSET;
5047 1836 : }
|