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) 1996-2026, 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 <time.h>
29 : #include <sys/stat.h>
30 : #include <sys/time.h>
31 : #include <unistd.h>
32 :
33 : #include "access/timeline.h"
34 : #include "access/transam.h"
35 : #include "access/xact.h"
36 : #include "access/xlog_internal.h"
37 : #include "access/xlogarchive.h"
38 : #include "access/xlogprefetcher.h"
39 : #include "access/xlogreader.h"
40 : #include "access/xlogrecovery.h"
41 : #include "access/xlogutils.h"
42 : #include "access/xlogwait.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 "nodes/miscnodes.h"
49 : #include "pgstat.h"
50 : #include "postmaster/bgwriter.h"
51 : #include "postmaster/startup.h"
52 : #include "replication/slot.h"
53 : #include "replication/slotsync.h"
54 : #include "replication/walreceiver.h"
55 : #include "storage/fd.h"
56 : #include "storage/ipc.h"
57 : #include "storage/latch.h"
58 : #include "storage/pmsignal.h"
59 : #include "storage/procarray.h"
60 : #include "storage/spin.h"
61 : #include "utils/datetime.h"
62 : #include "utils/fmgrprotos.h"
63 : #include "utils/guc_hooks.h"
64 : #include "utils/pgstat_internal.h"
65 : #include "utils/pg_lsn.h"
66 : #include "utils/ps_status.h"
67 : #include "utils/pg_rusage.h"
68 :
69 : /* Unsupported old recovery command file names (relative to $PGDATA) */
70 : #define RECOVERY_COMMAND_FILE "recovery.conf"
71 : #define RECOVERY_COMMAND_DONE "recovery.done"
72 :
73 : /*
74 : * GUC support
75 : */
76 : const struct config_enum_entry recovery_target_action_options[] = {
77 : {"pause", RECOVERY_TARGET_ACTION_PAUSE, false},
78 : {"promote", RECOVERY_TARGET_ACTION_PROMOTE, false},
79 : {"shutdown", RECOVERY_TARGET_ACTION_SHUTDOWN, false},
80 : {NULL, 0, false}
81 : };
82 :
83 : /* options formerly taken from recovery.conf for archive recovery */
84 : char *recoveryRestoreCommand = NULL;
85 : char *recoveryEndCommand = NULL;
86 : char *archiveCleanupCommand = NULL;
87 : RecoveryTargetType recoveryTarget = RECOVERY_TARGET_UNSET;
88 : bool recoveryTargetInclusive = true;
89 : int recoveryTargetAction = RECOVERY_TARGET_ACTION_PAUSE;
90 : TransactionId recoveryTargetXid;
91 : char *recovery_target_time_string;
92 : TimestampTz recoveryTargetTime;
93 : const char *recoveryTargetName;
94 : XLogRecPtr recoveryTargetLSN;
95 : int recovery_min_apply_delay = 0;
96 :
97 : /* options formerly taken from recovery.conf for XLOG streaming */
98 : char *PrimaryConnInfo = NULL;
99 : char *PrimarySlotName = NULL;
100 : bool wal_receiver_create_temp_slot = false;
101 :
102 : /*
103 : * recoveryTargetTimeLineGoal: what the user requested, if any
104 : *
105 : * recoveryTargetTLIRequested: numeric value of requested timeline, if constant
106 : *
107 : * recoveryTargetTLI: the currently understood target timeline; changes
108 : *
109 : * expectedTLEs: a list of TimeLineHistoryEntries for recoveryTargetTLI and
110 : * the timelines of its known parents, newest first (so recoveryTargetTLI is
111 : * always the first list member). Only these TLIs are expected to be seen in
112 : * the WAL segments we read, and indeed only these TLIs will be considered as
113 : * candidate WAL files to open at all.
114 : *
115 : * curFileTLI: the TLI appearing in the name of the current input WAL file.
116 : * (This is not necessarily the same as the timeline from which we are
117 : * replaying WAL, which StartupXLOG calls replayTLI, because we could be
118 : * scanning data that was copied from an ancestor timeline when the current
119 : * file was created.) During a sequential scan we do not allow this value
120 : * to decrease.
121 : */
122 : RecoveryTargetTimeLineGoal recoveryTargetTimeLineGoal = RECOVERY_TARGET_TIMELINE_LATEST;
123 : TimeLineID recoveryTargetTLIRequested = 0;
124 : TimeLineID recoveryTargetTLI = 0;
125 : static List *expectedTLEs;
126 : static TimeLineID curFileTLI;
127 :
128 : /*
129 : * When ArchiveRecoveryRequested is set, archive recovery was requested,
130 : * ie. signal files were present. When InArchiveRecovery is set, we are
131 : * currently recovering using offline XLOG archives. These variables are only
132 : * valid in the startup process.
133 : *
134 : * When ArchiveRecoveryRequested is true, but InArchiveRecovery is false, we're
135 : * currently performing crash recovery using only XLOG files in pg_wal, but
136 : * will switch to using offline XLOG archives as soon as we reach the end of
137 : * WAL in pg_wal.
138 : */
139 : bool ArchiveRecoveryRequested = false;
140 : bool InArchiveRecovery = false;
141 :
142 : /*
143 : * When StandbyModeRequested is set, standby mode was requested, i.e.
144 : * standby.signal file was present. When StandbyMode is set, we are currently
145 : * in standby mode. These variables are only valid in the startup process.
146 : * They work similarly to ArchiveRecoveryRequested and InArchiveRecovery.
147 : */
148 : static bool StandbyModeRequested = false;
149 : bool StandbyMode = false;
150 :
151 : /* was a signal file present at startup? */
152 : static bool standby_signal_file_found = false;
153 : static bool recovery_signal_file_found = false;
154 :
155 : /*
156 : * CheckPointLoc is the position of the checkpoint record that determines
157 : * where to start the replay. It comes from the backup label file or the
158 : * control file.
159 : *
160 : * RedoStartLSN is the checkpoint's REDO location, also from the backup label
161 : * file or the control file. In standby mode, XLOG streaming usually starts
162 : * from the position where an invalid record was found. But if we fail to
163 : * read even the initial checkpoint record, we use the REDO location instead
164 : * of the checkpoint location as the start position of XLOG streaming.
165 : * Otherwise we would have to jump backwards to the REDO location after
166 : * reading the checkpoint record, because the REDO record can precede the
167 : * checkpoint record.
168 : */
169 : static XLogRecPtr CheckPointLoc = InvalidXLogRecPtr;
170 : static TimeLineID CheckPointTLI = 0;
171 : static XLogRecPtr RedoStartLSN = InvalidXLogRecPtr;
172 : static TimeLineID RedoStartTLI = 0;
173 :
174 : /*
175 : * Local copy of SharedHotStandbyActive variable. False actually means "not
176 : * known, need to check the shared state".
177 : */
178 : static bool LocalHotStandbyActive = false;
179 :
180 : /*
181 : * Local copy of SharedPromoteIsTriggered variable. False actually means "not
182 : * known, need to check the shared state".
183 : */
184 : static bool LocalPromoteIsTriggered = false;
185 :
186 : /* Has the recovery code requested a walreceiver wakeup? */
187 : static bool doRequestWalReceiverReply;
188 :
189 : /* XLogReader object used to parse the WAL records */
190 : static XLogReaderState *xlogreader = NULL;
191 :
192 : /* XLogPrefetcher object used to consume WAL records with read-ahead */
193 : static XLogPrefetcher *xlogprefetcher = NULL;
194 :
195 : /* Parameters passed down from ReadRecord to the XLogPageRead callback. */
196 : typedef struct XLogPageReadPrivate
197 : {
198 : int emode;
199 : bool fetching_ckpt; /* are we fetching a checkpoint record? */
200 : bool randAccess;
201 : TimeLineID replayTLI;
202 : } XLogPageReadPrivate;
203 :
204 : /* flag to tell XLogPageRead that we have started replaying */
205 : static bool InRedo = false;
206 :
207 : /*
208 : * Codes indicating where we got a WAL file from during recovery, or where
209 : * to attempt to get one.
210 : */
211 : typedef enum
212 : {
213 : XLOG_FROM_ANY = 0, /* request to read WAL from any source */
214 : XLOG_FROM_ARCHIVE, /* restored using restore_command */
215 : XLOG_FROM_PG_WAL, /* existing file in pg_wal */
216 : XLOG_FROM_STREAM, /* streamed from primary */
217 : } XLogSource;
218 :
219 : /* human-readable names for XLogSources, for debugging output */
220 : static const char *const xlogSourceNames[] = {"any", "archive", "pg_wal", "stream"};
221 :
222 : /*
223 : * readFile is -1 or a kernel FD for the log file segment that's currently
224 : * open for reading. readSegNo identifies the segment. readOff is the offset
225 : * of the page just read, readLen indicates how much of it has been read into
226 : * readBuf, and readSource indicates where we got the currently open file from.
227 : *
228 : * Note: we could use Reserve/ReleaseExternalFD to track consumption of this
229 : * FD too (like for openLogFile in xlog.c); but it doesn't currently seem
230 : * worthwhile, since the XLOG is not read by general-purpose sessions.
231 : */
232 : static int readFile = -1;
233 : static XLogSegNo readSegNo = 0;
234 : static uint32 readOff = 0;
235 : static uint32 readLen = 0;
236 : static XLogSource readSource = XLOG_FROM_ANY;
237 :
238 : /*
239 : * Keeps track of which source we're currently reading from. This is
240 : * different from readSource in that this is always set, even when we don't
241 : * currently have a WAL file open. If lastSourceFailed is set, our last
242 : * attempt to read from currentSource failed, and we should try another source
243 : * next.
244 : *
245 : * pendingWalRcvRestart is set when a config change occurs that requires a
246 : * walreceiver restart. This is only valid in XLOG_FROM_STREAM state.
247 : */
248 : static XLogSource currentSource = XLOG_FROM_ANY;
249 : static bool lastSourceFailed = false;
250 : static bool pendingWalRcvRestart = false;
251 :
252 : /*
253 : * These variables track when we last obtained some WAL data to process,
254 : * and where we got it from. (XLogReceiptSource is initially the same as
255 : * readSource, but readSource gets reset to zero when we don't have data
256 : * to process right now. It is also different from currentSource, which
257 : * also changes when we try to read from a source and fail, while
258 : * XLogReceiptSource tracks where we last successfully read some WAL.)
259 : */
260 : static TimestampTz XLogReceiptTime = 0;
261 : static XLogSource XLogReceiptSource = XLOG_FROM_ANY;
262 :
263 : /* Local copy of WalRcv->flushedUpto */
264 : static XLogRecPtr flushedUpto = 0;
265 : static TimeLineID receiveTLI = 0;
266 :
267 : /*
268 : * Copy of minRecoveryPoint and backupEndPoint from the control file.
269 : *
270 : * In order to reach consistency, we must replay the WAL up to
271 : * minRecoveryPoint. If backupEndRequired is true, we must also reach
272 : * backupEndPoint, or if it's invalid, an end-of-backup record corresponding
273 : * to backupStartPoint.
274 : *
275 : * Note: In archive recovery, after consistency has been reached, the
276 : * functions in xlog.c will start updating minRecoveryPoint in the control
277 : * file. But this copy of minRecoveryPoint variable reflects the value at the
278 : * beginning of recovery, and is *not* updated after consistency is reached.
279 : */
280 : static XLogRecPtr minRecoveryPoint;
281 : static TimeLineID minRecoveryPointTLI;
282 :
283 : static XLogRecPtr backupStartPoint;
284 : static XLogRecPtr backupEndPoint;
285 : static bool backupEndRequired = false;
286 :
287 : /*
288 : * Have we reached a consistent database state? In crash recovery, we have
289 : * to replay all the WAL, so reachedConsistency is never set. During archive
290 : * recovery, the database is consistent once minRecoveryPoint is reached.
291 : *
292 : * Consistent state means that the system is internally consistent, all
293 : * the WAL has been replayed up to a certain point, and importantly, there
294 : * is no trace of later actions on disk.
295 : *
296 : * This flag is used only by the startup process and postmaster. When
297 : * minRecoveryPoint is reached, the startup process sets it to true and
298 : * sends a PMSIGNAL_RECOVERY_CONSISTENT signal to the postmaster,
299 : * which then sets it to true upon receiving the signal.
300 : */
301 : bool reachedConsistency = false;
302 :
303 : /* Buffers dedicated to consistency checks of size BLCKSZ */
304 : static char *replay_image_masked = NULL;
305 : static char *primary_image_masked = NULL;
306 :
307 :
308 : /*
309 : * Shared-memory state for WAL recovery.
310 : */
311 : typedef struct XLogRecoveryCtlData
312 : {
313 : /*
314 : * SharedHotStandbyActive indicates if we allow hot standby queries to be
315 : * run. Protected by info_lck.
316 : */
317 : bool SharedHotStandbyActive;
318 :
319 : /*
320 : * SharedPromoteIsTriggered indicates if a standby promotion has been
321 : * triggered. Protected by info_lck.
322 : */
323 : bool SharedPromoteIsTriggered;
324 :
325 : /*
326 : * recoveryWakeupLatch is used to wake up the startup process to continue
327 : * WAL replay, if it is waiting for WAL to arrive or promotion to be
328 : * requested.
329 : *
330 : * Note that the startup process also uses another latch, its procLatch,
331 : * to wait for recovery conflict. If we get rid of recoveryWakeupLatch for
332 : * signaling the startup process in favor of using its procLatch, which
333 : * comports better with possible generic signal handlers using that latch.
334 : * But we should not do that because the startup process doesn't assume
335 : * that it's waken up by walreceiver process or SIGHUP signal handler
336 : * while it's waiting for recovery conflict. The separate latches,
337 : * recoveryWakeupLatch and procLatch, should be used for inter-process
338 : * communication for WAL replay and recovery conflict, respectively.
339 : */
340 : Latch recoveryWakeupLatch;
341 :
342 : /*
343 : * Last record successfully replayed.
344 : */
345 : XLogRecPtr lastReplayedReadRecPtr; /* start position */
346 : XLogRecPtr lastReplayedEndRecPtr; /* end+1 position */
347 : TimeLineID lastReplayedTLI; /* timeline */
348 :
349 : /*
350 : * When we're currently replaying a record, ie. in a redo function,
351 : * replayEndRecPtr points to the end+1 of the record being replayed,
352 : * otherwise it's equal to lastReplayedEndRecPtr.
353 : */
354 : XLogRecPtr replayEndRecPtr;
355 : TimeLineID replayEndTLI;
356 : /* timestamp of last COMMIT/ABORT record replayed (or being replayed) */
357 : TimestampTz recoveryLastXTime;
358 :
359 : /*
360 : * timestamp of when we started replaying the current chunk of WAL data,
361 : * only relevant for replication or archive recovery
362 : */
363 : TimestampTz currentChunkStartTime;
364 : /* Recovery pause state */
365 : RecoveryPauseState recoveryPauseState;
366 : ConditionVariable recoveryNotPausedCV;
367 :
368 : slock_t info_lck; /* locks shared variables shown above */
369 : } XLogRecoveryCtlData;
370 :
371 : static XLogRecoveryCtlData *XLogRecoveryCtl = NULL;
372 :
373 : /*
374 : * abortedRecPtr is the start pointer of a broken record at end of WAL when
375 : * recovery completes; missingContrecPtr is the location of the first
376 : * contrecord that went missing. See CreateOverwriteContrecordRecord for
377 : * details.
378 : */
379 : static XLogRecPtr abortedRecPtr;
380 : static XLogRecPtr missingContrecPtr;
381 :
382 : /*
383 : * if recoveryStopsBefore/After returns true, it saves information of the stop
384 : * point here
385 : */
386 : static TransactionId recoveryStopXid;
387 : static TimestampTz recoveryStopTime;
388 : static XLogRecPtr recoveryStopLSN;
389 : static char recoveryStopName[MAXFNAMELEN];
390 : static bool recoveryStopAfter;
391 :
392 : /* prototypes for local functions */
393 : static void ApplyWalRecord(XLogReaderState *xlogreader, XLogRecord *record, TimeLineID *replayTLI);
394 :
395 : static void EnableStandbyMode(void);
396 : static void readRecoverySignalFile(void);
397 : static void validateRecoveryParameters(void);
398 : static bool read_backup_label(XLogRecPtr *checkPointLoc,
399 : TimeLineID *backupLabelTLI,
400 : bool *backupEndRequired, bool *backupFromStandby);
401 : static bool read_tablespace_map(List **tablespaces);
402 :
403 : static void xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI);
404 : static void CheckRecoveryConsistency(void);
405 : static void rm_redo_error_callback(void *arg);
406 : #ifdef WAL_DEBUG
407 : static void xlog_outrec(StringInfo buf, XLogReaderState *record);
408 : #endif
409 : static void xlog_block_info(StringInfo buf, XLogReaderState *record);
410 : static void checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI,
411 : TimeLineID prevTLI, TimeLineID replayTLI);
412 : static bool getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime);
413 : static void verifyBackupPageConsistency(XLogReaderState *record);
414 :
415 : static bool recoveryStopsBefore(XLogReaderState *record);
416 : static bool recoveryStopsAfter(XLogReaderState *record);
417 : static char *getRecoveryStopReason(void);
418 : static void recoveryPausesHere(bool endOfRecovery);
419 : static bool recoveryApplyDelay(XLogReaderState *record);
420 : static void ConfirmRecoveryPaused(void);
421 :
422 : static XLogRecord *ReadRecord(XLogPrefetcher *xlogprefetcher,
423 : int emode, bool fetching_ckpt,
424 : TimeLineID replayTLI);
425 :
426 : static int XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr,
427 : int reqLen, XLogRecPtr targetRecPtr, char *readBuf);
428 : static XLogPageReadResult WaitForWALToBecomeAvailable(XLogRecPtr RecPtr,
429 : bool randAccess,
430 : bool fetching_ckpt,
431 : XLogRecPtr tliRecPtr,
432 : TimeLineID replayTLI,
433 : XLogRecPtr replayLSN,
434 : bool nonblocking);
435 : static int emode_for_corrupt_record(int emode, XLogRecPtr RecPtr);
436 : static XLogRecord *ReadCheckpointRecord(XLogPrefetcher *xlogprefetcher,
437 : XLogRecPtr RecPtr, TimeLineID replayTLI);
438 : static bool rescanLatestTimeLine(TimeLineID replayTLI, XLogRecPtr replayLSN);
439 : static int XLogFileRead(XLogSegNo segno, TimeLineID tli,
440 : XLogSource source, bool notfoundOk);
441 : static int XLogFileReadAnyTLI(XLogSegNo segno, XLogSource source);
442 :
443 : static bool CheckForStandbyTrigger(void);
444 : static void SetPromoteIsTriggered(void);
445 : static bool HotStandbyActiveInReplay(void);
446 :
447 : static void SetCurrentChunkStartTime(TimestampTz xtime);
448 : static void SetLatestXTime(TimestampTz xtime);
449 :
450 : /*
451 : * Initialization of shared memory for WAL recovery
452 : */
453 : Size
454 6510 : XLogRecoveryShmemSize(void)
455 : {
456 : Size size;
457 :
458 : /* XLogRecoveryCtl */
459 6510 : size = sizeof(XLogRecoveryCtlData);
460 :
461 6510 : return size;
462 : }
463 :
464 : void
465 2272 : XLogRecoveryShmemInit(void)
466 : {
467 : bool found;
468 :
469 2272 : XLogRecoveryCtl = (XLogRecoveryCtlData *)
470 2272 : ShmemInitStruct("XLOG Recovery Ctl", XLogRecoveryShmemSize(), &found);
471 2272 : if (found)
472 0 : return;
473 2272 : memset(XLogRecoveryCtl, 0, sizeof(XLogRecoveryCtlData));
474 :
475 2272 : SpinLockInit(&XLogRecoveryCtl->info_lck);
476 2272 : InitSharedLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
477 2272 : ConditionVariableInit(&XLogRecoveryCtl->recoveryNotPausedCV);
478 : }
479 :
480 : /*
481 : * A thin wrapper to enable StandbyMode and do other preparatory work as
482 : * needed.
483 : */
484 : static void
485 224 : EnableStandbyMode(void)
486 : {
487 224 : StandbyMode = true;
488 :
489 : /*
490 : * To avoid server log bloat, we don't report recovery progress in a
491 : * standby as it will always be in recovery unless promoted. We disable
492 : * startup progress timeout in standby mode to avoid calling
493 : * startup_progress_timeout_handler() unnecessarily.
494 : */
495 224 : disable_startup_progress_timeout();
496 224 : }
497 :
498 : /*
499 : * Prepare the system for WAL recovery, if needed.
500 : *
501 : * This is called by StartupXLOG() which coordinates the server startup
502 : * sequence. This function analyzes the control file and the backup label
503 : * file, if any, and figures out whether we need to perform crash recovery or
504 : * archive recovery, and how far we need to replay the WAL to reach a
505 : * consistent state.
506 : *
507 : * This doesn't yet change the on-disk state, except for creating the symlinks
508 : * from table space map file if any, and for fetching WAL files needed to find
509 : * the checkpoint record. On entry, the caller has already read the control
510 : * file into memory, and passes it as argument. This function updates it to
511 : * reflect the recovery state, and the caller is expected to write it back to
512 : * disk does after initializing other subsystems, but before calling
513 : * PerformWalRecovery().
514 : *
515 : * This initializes some global variables like ArchiveRecoveryRequested, and
516 : * StandbyModeRequested and InRecovery.
517 : */
518 : void
519 1984 : InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr,
520 : bool *haveBackupLabel_ptr, bool *haveTblspcMap_ptr)
521 : {
522 : XLogPageReadPrivate *private;
523 : struct stat st;
524 : bool wasShutdown;
525 : XLogRecord *record;
526 : DBState dbstate_at_startup;
527 1984 : bool haveTblspcMap = false;
528 1984 : bool haveBackupLabel = false;
529 : CheckPoint checkPoint;
530 1984 : bool backupFromStandby = false;
531 :
532 1984 : dbstate_at_startup = ControlFile->state;
533 :
534 : /*
535 : * Initialize on the assumption we want to recover to the latest timeline
536 : * that's active according to pg_control.
537 : */
538 1984 : if (ControlFile->minRecoveryPointTLI >
539 1984 : ControlFile->checkPointCopy.ThisTimeLineID)
540 4 : recoveryTargetTLI = ControlFile->minRecoveryPointTLI;
541 : else
542 1980 : recoveryTargetTLI = ControlFile->checkPointCopy.ThisTimeLineID;
543 :
544 : /*
545 : * Check for signal files, and if so set up state for offline recovery
546 : */
547 1984 : readRecoverySignalFile();
548 1984 : validateRecoveryParameters();
549 :
550 : /*
551 : * Take ownership of the wakeup latch if we're going to sleep during
552 : * recovery, if required.
553 : */
554 1984 : if (ArchiveRecoveryRequested)
555 234 : OwnLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
556 :
557 : /*
558 : * Set the WAL reading processor now, as it will be needed when reading
559 : * the checkpoint record required (backup_label or not).
560 : */
561 1984 : private = palloc0_object(XLogPageReadPrivate);
562 1984 : xlogreader =
563 1984 : XLogReaderAllocate(wal_segment_size, NULL,
564 1984 : XL_ROUTINE(.page_read = &XLogPageRead,
565 : .segment_open = NULL,
566 : .segment_close = wal_segment_close),
567 : private);
568 1984 : if (!xlogreader)
569 0 : ereport(ERROR,
570 : (errcode(ERRCODE_OUT_OF_MEMORY),
571 : errmsg("out of memory"),
572 : errdetail("Failed while allocating a WAL reading processor.")));
573 1984 : xlogreader->system_identifier = ControlFile->system_identifier;
574 :
575 : /*
576 : * Set the WAL decode buffer size. This limits how far ahead we can read
577 : * in the WAL.
578 : */
579 1984 : XLogReaderSetDecodeBuffer(xlogreader, NULL, wal_decode_buffer_size);
580 :
581 : /* Create a WAL prefetcher. */
582 1984 : xlogprefetcher = XLogPrefetcherAllocate(xlogreader);
583 :
584 : /*
585 : * Allocate two page buffers dedicated to WAL consistency checks. We do
586 : * it this way, rather than just making static arrays, for two reasons:
587 : * (1) no need to waste the storage in most instantiations of the backend;
588 : * (2) a static char array isn't guaranteed to have any particular
589 : * alignment, whereas palloc() will provide MAXALIGN'd storage.
590 : */
591 1984 : replay_image_masked = (char *) palloc(BLCKSZ);
592 1984 : primary_image_masked = (char *) palloc(BLCKSZ);
593 :
594 : /*
595 : * Read the backup_label file. We want to run this part of the recovery
596 : * process after checking for signal files and after performing validation
597 : * of the recovery parameters.
598 : */
599 1984 : if (read_backup_label(&CheckPointLoc, &CheckPointTLI, &backupEndRequired,
600 : &backupFromStandby))
601 : {
602 158 : List *tablespaces = NIL;
603 :
604 : /*
605 : * Archive recovery was requested, and thanks to the backup label
606 : * file, we know how far we need to replay to reach consistency. Enter
607 : * archive recovery directly.
608 : */
609 158 : InArchiveRecovery = true;
610 158 : if (StandbyModeRequested)
611 136 : EnableStandbyMode();
612 :
613 : /*
614 : * Omitting backup_label when creating a new replica, PITR node etc.
615 : * unfortunately is a common cause of corruption. Logging that
616 : * backup_label was used makes it a bit easier to exclude that as the
617 : * cause of observed corruption.
618 : *
619 : * Do so before we try to read the checkpoint record (which can fail),
620 : * as otherwise it can be hard to understand why a checkpoint other
621 : * than ControlFile->checkPoint is used.
622 : */
623 158 : ereport(LOG,
624 : errmsg("starting backup recovery with redo LSN %X/%08X, checkpoint LSN %X/%08X, on timeline ID %u",
625 : LSN_FORMAT_ARGS(RedoStartLSN),
626 : LSN_FORMAT_ARGS(CheckPointLoc),
627 : CheckPointTLI));
628 :
629 : /*
630 : * When a backup_label file is present, we want to roll forward from
631 : * the checkpoint it identifies, rather than using pg_control.
632 : */
633 158 : record = ReadCheckpointRecord(xlogprefetcher, CheckPointLoc,
634 : CheckPointTLI);
635 158 : if (record != NULL)
636 : {
637 158 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
638 158 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
639 158 : ereport(DEBUG1,
640 : errmsg_internal("checkpoint record is at %X/%08X",
641 : LSN_FORMAT_ARGS(CheckPointLoc)));
642 158 : InRecovery = true; /* force recovery even if SHUTDOWNED */
643 :
644 : /*
645 : * Make sure that REDO location exists. This may not be the case
646 : * if there was a crash during an online backup, which left a
647 : * backup_label around that references a WAL segment that's
648 : * already been archived.
649 : */
650 158 : if (checkPoint.redo < CheckPointLoc)
651 : {
652 158 : XLogPrefetcherBeginRead(xlogprefetcher, checkPoint.redo);
653 158 : if (!ReadRecord(xlogprefetcher, LOG, false,
654 : checkPoint.ThisTimeLineID))
655 0 : ereport(FATAL,
656 : errmsg("could not find redo location %X/%08X referenced by checkpoint record at %X/%08X",
657 : LSN_FORMAT_ARGS(checkPoint.redo), LSN_FORMAT_ARGS(CheckPointLoc)),
658 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" or \"%s/standby.signal\" and add required recovery options.\n"
659 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
660 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
661 : DataDir, DataDir, DataDir, DataDir));
662 : }
663 : }
664 : else
665 : {
666 0 : ereport(FATAL,
667 : errmsg("could not locate required checkpoint record at %X/%08X",
668 : LSN_FORMAT_ARGS(CheckPointLoc)),
669 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" or \"%s/standby.signal\" and add required recovery options.\n"
670 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
671 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
672 : DataDir, DataDir, DataDir, DataDir));
673 : wasShutdown = false; /* keep compiler quiet */
674 : }
675 :
676 : /* Read the tablespace_map file if present and create symlinks. */
677 158 : if (read_tablespace_map(&tablespaces))
678 : {
679 : ListCell *lc;
680 :
681 8 : foreach(lc, tablespaces)
682 : {
683 4 : tablespaceinfo *ti = lfirst(lc);
684 : char *linkloc;
685 :
686 4 : linkloc = psprintf("%s/%u", PG_TBLSPC_DIR, ti->oid);
687 :
688 : /*
689 : * Remove the existing symlink if any and Create the symlink
690 : * under PGDATA.
691 : */
692 4 : remove_tablespace_symlink(linkloc);
693 :
694 4 : if (symlink(ti->path, linkloc) < 0)
695 0 : ereport(ERROR,
696 : (errcode_for_file_access(),
697 : errmsg("could not create symbolic link \"%s\": %m",
698 : linkloc)));
699 :
700 4 : pfree(ti->path);
701 4 : pfree(ti);
702 : }
703 :
704 : /* tell the caller to delete it later */
705 4 : haveTblspcMap = true;
706 : }
707 :
708 : /* tell the caller to delete it later */
709 158 : haveBackupLabel = true;
710 : }
711 : else
712 : {
713 : /* No backup_label file has been found if we are here. */
714 :
715 : /*
716 : * If tablespace_map file is present without backup_label file, there
717 : * is no use of such file. There is no harm in retaining it, but it
718 : * is better to get rid of the map file so that we don't have any
719 : * redundant file in data directory and it will avoid any sort of
720 : * confusion. It seems prudent though to just rename the file out of
721 : * the way rather than delete it completely, also we ignore any error
722 : * that occurs in rename operation as even if map file is present
723 : * without backup_label file, it is harmless.
724 : */
725 1826 : if (stat(TABLESPACE_MAP, &st) == 0)
726 : {
727 2 : unlink(TABLESPACE_MAP_OLD);
728 2 : if (durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, DEBUG1) == 0)
729 2 : ereport(LOG,
730 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
731 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
732 : errdetail("File \"%s\" was renamed to \"%s\".",
733 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
734 : else
735 0 : ereport(LOG,
736 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
737 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
738 : errdetail("Could not rename file \"%s\" to \"%s\": %m.",
739 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
740 : }
741 :
742 : /*
743 : * It's possible that archive recovery was requested, but we don't
744 : * know how far we need to replay the WAL before we reach consistency.
745 : * This can happen for example if a base backup is taken from a
746 : * running server using an atomic filesystem snapshot, without calling
747 : * pg_backup_start/stop. Or if you just kill a running primary server
748 : * and put it into archive recovery by creating a recovery signal
749 : * file.
750 : *
751 : * Our strategy in that case is to perform crash recovery first,
752 : * replaying all the WAL present in pg_wal, and only enter archive
753 : * recovery after that.
754 : *
755 : * But usually we already know how far we need to replay the WAL (up
756 : * to minRecoveryPoint, up to backupEndPoint, or until we see an
757 : * end-of-backup record), and we can enter archive recovery directly.
758 : */
759 1826 : if (ArchiveRecoveryRequested &&
760 88 : (XLogRecPtrIsValid(ControlFile->minRecoveryPoint) ||
761 18 : ControlFile->backupEndRequired ||
762 18 : XLogRecPtrIsValid(ControlFile->backupEndPoint) ||
763 18 : ControlFile->state == DB_SHUTDOWNED))
764 : {
765 84 : InArchiveRecovery = true;
766 84 : if (StandbyModeRequested)
767 84 : EnableStandbyMode();
768 : }
769 :
770 : /*
771 : * For the same reason as when starting up with backup_label present,
772 : * emit a log message when we continue initializing from a base
773 : * backup.
774 : */
775 1826 : if (XLogRecPtrIsValid(ControlFile->backupStartPoint))
776 0 : ereport(LOG,
777 : errmsg("restarting backup recovery with redo LSN %X/%08X",
778 : LSN_FORMAT_ARGS(ControlFile->backupStartPoint)));
779 :
780 : /* Get the last valid checkpoint record. */
781 1826 : CheckPointLoc = ControlFile->checkPoint;
782 1826 : CheckPointTLI = ControlFile->checkPointCopy.ThisTimeLineID;
783 1826 : RedoStartLSN = ControlFile->checkPointCopy.redo;
784 1826 : RedoStartTLI = ControlFile->checkPointCopy.ThisTimeLineID;
785 1826 : record = ReadCheckpointRecord(xlogprefetcher, CheckPointLoc,
786 : CheckPointTLI);
787 1826 : if (record != NULL)
788 : {
789 1826 : ereport(DEBUG1,
790 : errmsg_internal("checkpoint record is at %X/%08X",
791 : LSN_FORMAT_ARGS(CheckPointLoc)));
792 : }
793 : else
794 : {
795 : /*
796 : * We used to attempt to go back to a secondary checkpoint record
797 : * here, but only when not in standby mode. We now just fail if we
798 : * can't read the last checkpoint because this allows us to
799 : * simplify processing around checkpoints.
800 : */
801 0 : ereport(PANIC,
802 : errmsg("could not locate a valid checkpoint record at %X/%08X",
803 : LSN_FORMAT_ARGS(CheckPointLoc)));
804 : }
805 1826 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
806 1826 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
807 :
808 : /* Make sure that REDO location exists. */
809 1826 : if (checkPoint.redo < CheckPointLoc)
810 : {
811 88 : XLogPrefetcherBeginRead(xlogprefetcher, checkPoint.redo);
812 88 : if (!ReadRecord(xlogprefetcher, LOG, false, checkPoint.ThisTimeLineID))
813 2 : ereport(FATAL,
814 : errmsg("could not find redo location %X/%08X referenced by checkpoint record at %X/%08X",
815 : LSN_FORMAT_ARGS(checkPoint.redo), LSN_FORMAT_ARGS(CheckPointLoc)));
816 : }
817 : }
818 :
819 1982 : if (ArchiveRecoveryRequested)
820 : {
821 234 : if (StandbyModeRequested)
822 224 : ereport(LOG,
823 : (errmsg("entering standby mode")));
824 10 : else if (recoveryTarget == RECOVERY_TARGET_XID)
825 0 : ereport(LOG,
826 : (errmsg("starting point-in-time recovery to XID %u",
827 : recoveryTargetXid)));
828 10 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
829 0 : ereport(LOG,
830 : (errmsg("starting point-in-time recovery to %s",
831 : timestamptz_to_str(recoveryTargetTime))));
832 10 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
833 6 : ereport(LOG,
834 : (errmsg("starting point-in-time recovery to \"%s\"",
835 : recoveryTargetName)));
836 4 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
837 0 : ereport(LOG,
838 : errmsg("starting point-in-time recovery to WAL location (LSN) \"%X/%08X\"",
839 : LSN_FORMAT_ARGS(recoveryTargetLSN)));
840 4 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
841 0 : ereport(LOG,
842 : (errmsg("starting point-in-time recovery to earliest consistent point")));
843 : else
844 4 : ereport(LOG,
845 : (errmsg("starting archive recovery")));
846 : }
847 :
848 : /*
849 : * If the location of the checkpoint record is not on the expected
850 : * timeline in the history of the requested timeline, we cannot proceed:
851 : * the backup is not part of the history of the requested timeline.
852 : */
853 : Assert(expectedTLEs); /* was initialized by reading checkpoint
854 : * record */
855 1982 : if (tliOfPointInHistory(CheckPointLoc, expectedTLEs) !=
856 : CheckPointTLI)
857 : {
858 : XLogRecPtr switchpoint;
859 :
860 : /*
861 : * tliSwitchPoint will throw an error if the checkpoint's timeline is
862 : * not in expectedTLEs at all.
863 : */
864 0 : switchpoint = tliSwitchPoint(CheckPointTLI, expectedTLEs, NULL);
865 0 : ereport(FATAL,
866 : (errmsg("requested timeline %u is not a child of this server's history",
867 : recoveryTargetTLI),
868 : /* translator: %s is a backup_label file or a pg_control file */
869 : errdetail("Latest checkpoint in file \"%s\" is at %X/%08X on timeline %u, but in the history of the requested timeline, the server forked off from that timeline at %X/%08X.",
870 : haveBackupLabel ? "backup_label" : "pg_control",
871 : LSN_FORMAT_ARGS(CheckPointLoc),
872 : CheckPointTLI,
873 : LSN_FORMAT_ARGS(switchpoint))));
874 : }
875 :
876 : /*
877 : * The min recovery point should be part of the requested timeline's
878 : * history, too.
879 : */
880 1982 : if (XLogRecPtrIsValid(ControlFile->minRecoveryPoint) &&
881 84 : tliOfPointInHistory(ControlFile->minRecoveryPoint - 1, expectedTLEs) !=
882 84 : ControlFile->minRecoveryPointTLI)
883 0 : ereport(FATAL,
884 : errmsg("requested timeline %u does not contain minimum recovery point %X/%08X on timeline %u",
885 : recoveryTargetTLI,
886 : LSN_FORMAT_ARGS(ControlFile->minRecoveryPoint),
887 : ControlFile->minRecoveryPointTLI));
888 :
889 1982 : ereport(DEBUG1,
890 : errmsg_internal("redo record is at %X/%08X; shutdown %s",
891 : LSN_FORMAT_ARGS(checkPoint.redo),
892 : wasShutdown ? "true" : "false"));
893 1982 : ereport(DEBUG1,
894 : (errmsg_internal("next transaction ID: " UINT64_FORMAT "; next OID: %u",
895 : U64FromFullTransactionId(checkPoint.nextXid),
896 : checkPoint.nextOid)));
897 1982 : ereport(DEBUG1,
898 : (errmsg_internal("next MultiXactId: %u; next MultiXactOffset: %" PRIu64,
899 : checkPoint.nextMulti, checkPoint.nextMultiOffset)));
900 1982 : ereport(DEBUG1,
901 : (errmsg_internal("oldest unfrozen transaction ID: %u, in database %u",
902 : checkPoint.oldestXid, checkPoint.oldestXidDB)));
903 1982 : ereport(DEBUG1,
904 : (errmsg_internal("oldest MultiXactId: %u, in database %u",
905 : checkPoint.oldestMulti, checkPoint.oldestMultiDB)));
906 1982 : ereport(DEBUG1,
907 : (errmsg_internal("commit timestamp Xid oldest/newest: %u/%u",
908 : checkPoint.oldestCommitTsXid,
909 : checkPoint.newestCommitTsXid)));
910 1982 : if (!TransactionIdIsNormal(XidFromFullTransactionId(checkPoint.nextXid)))
911 0 : ereport(PANIC,
912 : (errmsg("invalid next transaction ID")));
913 :
914 : /* sanity check */
915 1982 : if (checkPoint.redo > CheckPointLoc)
916 0 : ereport(PANIC,
917 : (errmsg("invalid redo in checkpoint record")));
918 :
919 : /*
920 : * Check whether we need to force recovery from WAL. If it appears to
921 : * have been a clean shutdown and we did not have a recovery signal file,
922 : * then assume no recovery needed.
923 : */
924 1982 : if (checkPoint.redo < CheckPointLoc)
925 : {
926 244 : if (wasShutdown)
927 0 : ereport(PANIC,
928 : (errmsg("invalid redo record in shutdown checkpoint")));
929 244 : InRecovery = true;
930 : }
931 1738 : else if (ControlFile->state != DB_SHUTDOWNED)
932 190 : InRecovery = true;
933 1548 : else if (ArchiveRecoveryRequested)
934 : {
935 : /* force recovery due to presence of recovery signal file */
936 14 : InRecovery = true;
937 : }
938 :
939 : /*
940 : * If recovery is needed, update our in-memory copy of pg_control to show
941 : * that we are recovering and to show the selected checkpoint as the place
942 : * we are starting from. We also mark pg_control with any minimum recovery
943 : * stop point obtained from a backup history file.
944 : *
945 : * We don't write the changes to disk yet, though. Only do that after
946 : * initializing various subsystems.
947 : */
948 1982 : if (InRecovery)
949 : {
950 448 : if (InArchiveRecovery)
951 : {
952 242 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
953 : }
954 : else
955 : {
956 206 : ereport(LOG,
957 : (errmsg("database system was not properly shut down; "
958 : "automatic recovery in progress")));
959 206 : if (recoveryTargetTLI > ControlFile->checkPointCopy.ThisTimeLineID)
960 4 : ereport(LOG,
961 : (errmsg("crash recovery starts in timeline %u "
962 : "and has target timeline %u",
963 : ControlFile->checkPointCopy.ThisTimeLineID,
964 : recoveryTargetTLI)));
965 206 : ControlFile->state = DB_IN_CRASH_RECOVERY;
966 : }
967 448 : ControlFile->checkPoint = CheckPointLoc;
968 448 : ControlFile->checkPointCopy = checkPoint;
969 448 : if (InArchiveRecovery)
970 : {
971 : /* initialize minRecoveryPoint if not set yet */
972 242 : if (ControlFile->minRecoveryPoint < checkPoint.redo)
973 : {
974 162 : ControlFile->minRecoveryPoint = checkPoint.redo;
975 162 : ControlFile->minRecoveryPointTLI = checkPoint.ThisTimeLineID;
976 : }
977 : }
978 :
979 : /*
980 : * Set backupStartPoint if we're starting recovery from a base backup.
981 : *
982 : * Also set backupEndPoint and use minRecoveryPoint as the backup end
983 : * location if we're starting recovery from a base backup which was
984 : * taken from a standby. In this case, the database system status in
985 : * pg_control must indicate that the database was already in recovery.
986 : * Usually that will be DB_IN_ARCHIVE_RECOVERY but also can be
987 : * DB_SHUTDOWNED_IN_RECOVERY if recovery previously was interrupted
988 : * before reaching this point; e.g. because restore_command or
989 : * primary_conninfo were faulty.
990 : *
991 : * Any other state indicates that the backup somehow became corrupted
992 : * and we can't sensibly continue with recovery.
993 : */
994 448 : if (haveBackupLabel)
995 : {
996 158 : ControlFile->backupStartPoint = checkPoint.redo;
997 158 : ControlFile->backupEndRequired = backupEndRequired;
998 :
999 158 : if (backupFromStandby)
1000 : {
1001 10 : if (dbstate_at_startup != DB_IN_ARCHIVE_RECOVERY &&
1002 : dbstate_at_startup != DB_SHUTDOWNED_IN_RECOVERY)
1003 0 : ereport(FATAL,
1004 : (errmsg("backup_label contains data inconsistent with control file"),
1005 : errhint("This means that the backup is corrupted and you will "
1006 : "have to use another backup for recovery.")));
1007 10 : ControlFile->backupEndPoint = ControlFile->minRecoveryPoint;
1008 : }
1009 : }
1010 : }
1011 :
1012 : /* remember these, so that we know when we have reached consistency */
1013 1982 : backupStartPoint = ControlFile->backupStartPoint;
1014 1982 : backupEndRequired = ControlFile->backupEndRequired;
1015 1982 : backupEndPoint = ControlFile->backupEndPoint;
1016 1982 : if (InArchiveRecovery)
1017 : {
1018 242 : minRecoveryPoint = ControlFile->minRecoveryPoint;
1019 242 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
1020 : }
1021 : else
1022 : {
1023 1740 : minRecoveryPoint = InvalidXLogRecPtr;
1024 1740 : minRecoveryPointTLI = 0;
1025 : }
1026 :
1027 : /*
1028 : * Start recovery assuming that the final record isn't lost.
1029 : */
1030 1982 : abortedRecPtr = InvalidXLogRecPtr;
1031 1982 : missingContrecPtr = InvalidXLogRecPtr;
1032 :
1033 1982 : *wasShutdown_ptr = wasShutdown;
1034 1982 : *haveBackupLabel_ptr = haveBackupLabel;
1035 1982 : *haveTblspcMap_ptr = haveTblspcMap;
1036 1982 : }
1037 :
1038 : /*
1039 : * See if there are any recovery signal files and if so, set state for
1040 : * recovery.
1041 : *
1042 : * See if there is a recovery command file (recovery.conf), and if so
1043 : * throw an ERROR since as of PG12 we no longer recognize that.
1044 : */
1045 : static void
1046 1984 : readRecoverySignalFile(void)
1047 : {
1048 : struct stat stat_buf;
1049 :
1050 1984 : if (IsBootstrapProcessingMode())
1051 1750 : return;
1052 :
1053 : /*
1054 : * Check for old recovery API file: recovery.conf
1055 : */
1056 1882 : if (stat(RECOVERY_COMMAND_FILE, &stat_buf) == 0)
1057 0 : ereport(FATAL,
1058 : (errcode_for_file_access(),
1059 : errmsg("using recovery command file \"%s\" is not supported",
1060 : RECOVERY_COMMAND_FILE)));
1061 :
1062 : /*
1063 : * Remove unused .done file, if present. Ignore if absent.
1064 : */
1065 1882 : unlink(RECOVERY_COMMAND_DONE);
1066 :
1067 : /*
1068 : * Check for recovery signal files and if found, fsync them since they
1069 : * represent server state information. We don't sweat too much about the
1070 : * possibility of fsync failure, however.
1071 : *
1072 : * If present, standby signal file takes precedence. If neither is present
1073 : * then we won't enter archive recovery.
1074 : */
1075 1882 : if (stat(STANDBY_SIGNAL_FILE, &stat_buf) == 0)
1076 : {
1077 : int fd;
1078 :
1079 224 : fd = BasicOpenFilePerm(STANDBY_SIGNAL_FILE, O_RDWR | PG_BINARY,
1080 : S_IRUSR | S_IWUSR);
1081 224 : if (fd >= 0)
1082 : {
1083 224 : (void) pg_fsync(fd);
1084 224 : close(fd);
1085 : }
1086 224 : standby_signal_file_found = true;
1087 : }
1088 1658 : else if (stat(RECOVERY_SIGNAL_FILE, &stat_buf) == 0)
1089 : {
1090 : int fd;
1091 :
1092 10 : fd = BasicOpenFilePerm(RECOVERY_SIGNAL_FILE, O_RDWR | PG_BINARY,
1093 : S_IRUSR | S_IWUSR);
1094 10 : if (fd >= 0)
1095 : {
1096 10 : (void) pg_fsync(fd);
1097 10 : close(fd);
1098 : }
1099 10 : recovery_signal_file_found = true;
1100 : }
1101 :
1102 1882 : StandbyModeRequested = false;
1103 1882 : ArchiveRecoveryRequested = false;
1104 1882 : if (standby_signal_file_found)
1105 : {
1106 224 : StandbyModeRequested = true;
1107 224 : ArchiveRecoveryRequested = true;
1108 : }
1109 1658 : else if (recovery_signal_file_found)
1110 : {
1111 10 : StandbyModeRequested = false;
1112 10 : ArchiveRecoveryRequested = true;
1113 : }
1114 : else
1115 1648 : return;
1116 :
1117 : /*
1118 : * We don't support standby mode in standalone backends; that requires
1119 : * other processes such as the WAL receiver to be alive.
1120 : */
1121 234 : if (StandbyModeRequested && !IsUnderPostmaster)
1122 0 : ereport(FATAL,
1123 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1124 : errmsg("standby mode is not supported by single-user servers")));
1125 : }
1126 :
1127 : static void
1128 1984 : validateRecoveryParameters(void)
1129 : {
1130 1984 : if (!ArchiveRecoveryRequested)
1131 1750 : return;
1132 :
1133 : /*
1134 : * Check for compulsory parameters
1135 : */
1136 234 : if (StandbyModeRequested)
1137 : {
1138 224 : if ((PrimaryConnInfo == NULL || strcmp(PrimaryConnInfo, "") == 0) &&
1139 22 : (recoveryRestoreCommand == NULL || strcmp(recoveryRestoreCommand, "") == 0))
1140 4 : ereport(WARNING,
1141 : (errmsg("specified neither \"primary_conninfo\" nor \"restore_command\""),
1142 : errhint("The database server will regularly poll the pg_wal subdirectory to check for files placed there.")));
1143 : }
1144 : else
1145 : {
1146 10 : if (recoveryRestoreCommand == NULL ||
1147 10 : strcmp(recoveryRestoreCommand, "") == 0)
1148 0 : ereport(FATAL,
1149 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1150 : errmsg("must specify \"restore_command\" when standby mode is not enabled")));
1151 : }
1152 :
1153 : /*
1154 : * Override any inconsistent requests. Note that this is a change of
1155 : * behaviour in 9.5; prior to this we simply ignored a request to pause if
1156 : * hot_standby = off, which was surprising behaviour.
1157 : */
1158 234 : if (recoveryTargetAction == RECOVERY_TARGET_ACTION_PAUSE &&
1159 220 : !EnableHotStandby)
1160 6 : recoveryTargetAction = RECOVERY_TARGET_ACTION_SHUTDOWN;
1161 :
1162 : /*
1163 : * Final parsing of recovery_target_time string; see also
1164 : * check_recovery_target_time().
1165 : */
1166 234 : if (recoveryTarget == RECOVERY_TARGET_TIME)
1167 : {
1168 0 : recoveryTargetTime = DatumGetTimestampTz(DirectFunctionCall3(timestamptz_in,
1169 : CStringGetDatum(recovery_target_time_string),
1170 : ObjectIdGetDatum(InvalidOid),
1171 : Int32GetDatum(-1)));
1172 : }
1173 :
1174 : /*
1175 : * If user specified recovery_target_timeline, validate it or compute the
1176 : * "latest" value. We can't do this until after we've gotten the restore
1177 : * command and set InArchiveRecovery, because we need to fetch timeline
1178 : * history files from the archive.
1179 : */
1180 234 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
1181 : {
1182 0 : TimeLineID rtli = recoveryTargetTLIRequested;
1183 :
1184 : /* Timeline 1 does not have a history file, all else should */
1185 0 : if (rtli != 1 && !existsTimeLineHistory(rtli))
1186 0 : ereport(FATAL,
1187 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1188 : errmsg("recovery target timeline %u does not exist",
1189 : rtli)));
1190 0 : recoveryTargetTLI = rtli;
1191 : }
1192 234 : else if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
1193 : {
1194 : /* We start the "latest" search from pg_control's timeline */
1195 234 : recoveryTargetTLI = findNewestTimeLine(recoveryTargetTLI);
1196 : }
1197 : else
1198 : {
1199 : /*
1200 : * else we just use the recoveryTargetTLI as already read from
1201 : * ControlFile
1202 : */
1203 : Assert(recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_CONTROLFILE);
1204 : }
1205 : }
1206 :
1207 : /*
1208 : * read_backup_label: check to see if a backup_label file is present
1209 : *
1210 : * If we see a backup_label during recovery, we assume that we are recovering
1211 : * from a backup dump file, and we therefore roll forward from the checkpoint
1212 : * identified by the label file, NOT what pg_control says. This avoids the
1213 : * problem that pg_control might have been archived one or more checkpoints
1214 : * later than the start of the dump, and so if we rely on it as the start
1215 : * point, we will fail to restore a consistent database state.
1216 : *
1217 : * Returns true if a backup_label was found (and fills the checkpoint
1218 : * location and TLI into *checkPointLoc and *backupLabelTLI, respectively);
1219 : * returns false if not. If this backup_label came from a streamed backup,
1220 : * *backupEndRequired is set to true. If this backup_label was created during
1221 : * recovery, *backupFromStandby is set to true.
1222 : *
1223 : * Also sets the global variables RedoStartLSN and RedoStartTLI with the LSN
1224 : * and TLI read from the backup file.
1225 : */
1226 : static bool
1227 1984 : read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI,
1228 : bool *backupEndRequired, bool *backupFromStandby)
1229 : {
1230 : char startxlogfilename[MAXFNAMELEN];
1231 : TimeLineID tli_from_walseg,
1232 : tli_from_file;
1233 : FILE *lfp;
1234 : char ch;
1235 : char backuptype[20];
1236 : char backupfrom[20];
1237 : char backuplabel[MAXPGPATH];
1238 : char backuptime[128];
1239 : uint32 hi,
1240 : lo;
1241 :
1242 : /* suppress possible uninitialized-variable warnings */
1243 1984 : *checkPointLoc = InvalidXLogRecPtr;
1244 1984 : *backupLabelTLI = 0;
1245 1984 : *backupEndRequired = false;
1246 1984 : *backupFromStandby = false;
1247 :
1248 : /*
1249 : * See if label file is present
1250 : */
1251 1984 : lfp = AllocateFile(BACKUP_LABEL_FILE, "r");
1252 1984 : if (!lfp)
1253 : {
1254 1826 : if (errno != ENOENT)
1255 0 : ereport(FATAL,
1256 : (errcode_for_file_access(),
1257 : errmsg("could not read file \"%s\": %m",
1258 : BACKUP_LABEL_FILE)));
1259 1826 : return false; /* it's not there, all is fine */
1260 : }
1261 :
1262 : /*
1263 : * Read and parse the START WAL LOCATION and CHECKPOINT lines (this code
1264 : * is pretty crude, but we are not expecting any variability in the file
1265 : * format).
1266 : */
1267 158 : if (fscanf(lfp, "START WAL LOCATION: %X/%08X (file %08X%16s)%c",
1268 158 : &hi, &lo, &tli_from_walseg, startxlogfilename, &ch) != 5 || ch != '\n')
1269 0 : ereport(FATAL,
1270 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1271 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
1272 158 : RedoStartLSN = ((uint64) hi) << 32 | lo;
1273 158 : RedoStartTLI = tli_from_walseg;
1274 158 : if (fscanf(lfp, "CHECKPOINT LOCATION: %X/%08X%c",
1275 158 : &hi, &lo, &ch) != 3 || ch != '\n')
1276 0 : ereport(FATAL,
1277 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1278 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
1279 158 : *checkPointLoc = ((uint64) hi) << 32 | lo;
1280 158 : *backupLabelTLI = tli_from_walseg;
1281 :
1282 : /*
1283 : * BACKUP METHOD lets us know if this was a typical backup ("streamed",
1284 : * which could mean either pg_basebackup or the pg_backup_start/stop
1285 : * method was used) or if this label came from somewhere else (the only
1286 : * other option today being from pg_rewind). If this was a streamed
1287 : * backup then we know that we need to play through until we get to the
1288 : * end of the WAL which was generated during the backup (at which point we
1289 : * will have reached consistency and backupEndRequired will be reset to be
1290 : * false).
1291 : */
1292 158 : if (fscanf(lfp, "BACKUP METHOD: %19s\n", backuptype) == 1)
1293 : {
1294 158 : if (strcmp(backuptype, "streamed") == 0)
1295 156 : *backupEndRequired = true;
1296 : }
1297 :
1298 : /*
1299 : * BACKUP FROM lets us know if this was from a primary or a standby. If
1300 : * it was from a standby, we'll double-check that the control file state
1301 : * matches that of a standby.
1302 : */
1303 158 : if (fscanf(lfp, "BACKUP FROM: %19s\n", backupfrom) == 1)
1304 : {
1305 158 : if (strcmp(backupfrom, "standby") == 0)
1306 10 : *backupFromStandby = true;
1307 : }
1308 :
1309 : /*
1310 : * Parse START TIME and LABEL. Those are not mandatory fields for recovery
1311 : * but checking for their presence is useful for debugging and the next
1312 : * sanity checks. Cope also with the fact that the result buffers have a
1313 : * pre-allocated size, hence if the backup_label file has been generated
1314 : * with strings longer than the maximum assumed here an incorrect parsing
1315 : * happens. That's fine as only minor consistency checks are done
1316 : * afterwards.
1317 : */
1318 158 : if (fscanf(lfp, "START TIME: %127[^\n]\n", backuptime) == 1)
1319 158 : ereport(DEBUG1,
1320 : (errmsg_internal("backup time %s in file \"%s\"",
1321 : backuptime, BACKUP_LABEL_FILE)));
1322 :
1323 158 : if (fscanf(lfp, "LABEL: %1023[^\n]\n", backuplabel) == 1)
1324 156 : ereport(DEBUG1,
1325 : (errmsg_internal("backup label %s in file \"%s\"",
1326 : backuplabel, BACKUP_LABEL_FILE)));
1327 :
1328 : /*
1329 : * START TIMELINE is new as of 11. Its parsing is not mandatory, still use
1330 : * it as a sanity check if present.
1331 : */
1332 158 : if (fscanf(lfp, "START TIMELINE: %u\n", &tli_from_file) == 1)
1333 : {
1334 156 : if (tli_from_walseg != tli_from_file)
1335 0 : ereport(FATAL,
1336 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1337 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE),
1338 : errdetail("Timeline ID parsed is %u, but expected %u.",
1339 : tli_from_file, tli_from_walseg)));
1340 :
1341 156 : ereport(DEBUG1,
1342 : (errmsg_internal("backup timeline %u in file \"%s\"",
1343 : tli_from_file, BACKUP_LABEL_FILE)));
1344 : }
1345 :
1346 158 : if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%08X\n", &hi, &lo) > 0)
1347 0 : ereport(FATAL,
1348 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1349 : errmsg("this is an incremental backup, not a data directory"),
1350 : errhint("Use pg_combinebackup to reconstruct a valid data directory.")));
1351 :
1352 158 : if (ferror(lfp) || FreeFile(lfp))
1353 0 : ereport(FATAL,
1354 : (errcode_for_file_access(),
1355 : errmsg("could not read file \"%s\": %m",
1356 : BACKUP_LABEL_FILE)));
1357 :
1358 158 : return true;
1359 : }
1360 :
1361 : /*
1362 : * read_tablespace_map: check to see if a tablespace_map file is present
1363 : *
1364 : * If we see a tablespace_map file during recovery, we assume that we are
1365 : * recovering from a backup dump file, and we therefore need to create symlinks
1366 : * as per the information present in tablespace_map file.
1367 : *
1368 : * Returns true if a tablespace_map file was found (and fills *tablespaces
1369 : * with a tablespaceinfo struct for each tablespace listed in the file);
1370 : * returns false if not.
1371 : */
1372 : static bool
1373 158 : read_tablespace_map(List **tablespaces)
1374 : {
1375 : tablespaceinfo *ti;
1376 : FILE *lfp;
1377 : char str[MAXPGPATH];
1378 : int ch,
1379 : i,
1380 : n;
1381 : bool was_backslash;
1382 :
1383 : /*
1384 : * See if tablespace_map file is present
1385 : */
1386 158 : lfp = AllocateFile(TABLESPACE_MAP, "r");
1387 158 : if (!lfp)
1388 : {
1389 154 : if (errno != ENOENT)
1390 0 : ereport(FATAL,
1391 : (errcode_for_file_access(),
1392 : errmsg("could not read file \"%s\": %m",
1393 : TABLESPACE_MAP)));
1394 154 : return false; /* it's not there, all is fine */
1395 : }
1396 :
1397 : /*
1398 : * Read and parse the link name and path lines from tablespace_map file
1399 : * (this code is pretty crude, but we are not expecting any variability in
1400 : * the file format). De-escape any backslashes that were inserted.
1401 : */
1402 4 : i = 0;
1403 4 : was_backslash = false;
1404 154 : while ((ch = fgetc(lfp)) != EOF)
1405 : {
1406 150 : if (!was_backslash && (ch == '\n' || ch == '\r'))
1407 4 : {
1408 : char *endp;
1409 :
1410 4 : if (i == 0)
1411 0 : continue; /* \r immediately followed by \n */
1412 :
1413 : /*
1414 : * The de-escaped line should contain an OID followed by exactly
1415 : * one space followed by a path. The path might start with
1416 : * spaces, so don't be too liberal about parsing.
1417 : */
1418 4 : str[i] = '\0';
1419 4 : n = 0;
1420 24 : while (str[n] && str[n] != ' ')
1421 20 : n++;
1422 4 : if (n < 1 || n >= i - 1)
1423 0 : ereport(FATAL,
1424 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1425 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1426 4 : str[n++] = '\0';
1427 :
1428 4 : ti = palloc0_object(tablespaceinfo);
1429 4 : errno = 0;
1430 4 : ti->oid = strtoul(str, &endp, 10);
1431 4 : if (*endp != '\0' || errno == EINVAL || errno == ERANGE)
1432 0 : ereport(FATAL,
1433 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1434 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1435 4 : ti->path = pstrdup(str + n);
1436 4 : *tablespaces = lappend(*tablespaces, ti);
1437 :
1438 4 : i = 0;
1439 4 : continue;
1440 : }
1441 146 : else if (!was_backslash && ch == '\\')
1442 0 : was_backslash = true;
1443 : else
1444 : {
1445 146 : if (i < sizeof(str) - 1)
1446 146 : str[i++] = ch;
1447 146 : was_backslash = false;
1448 : }
1449 : }
1450 :
1451 4 : if (i != 0 || was_backslash) /* last line not terminated? */
1452 0 : ereport(FATAL,
1453 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1454 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1455 :
1456 4 : if (ferror(lfp) || FreeFile(lfp))
1457 0 : ereport(FATAL,
1458 : (errcode_for_file_access(),
1459 : errmsg("could not read file \"%s\": %m",
1460 : TABLESPACE_MAP)));
1461 :
1462 4 : return true;
1463 : }
1464 :
1465 : /*
1466 : * Finish WAL recovery.
1467 : *
1468 : * This does not close the 'xlogreader' yet, because in some cases the caller
1469 : * still wants to re-read the last checkpoint record by calling
1470 : * ReadCheckpointRecord().
1471 : *
1472 : * Returns the position of the last valid or applied record, after which new
1473 : * WAL should be appended, information about why recovery was ended, and some
1474 : * other things. See the EndOfWalRecoveryInfo struct for details.
1475 : */
1476 : EndOfWalRecoveryInfo *
1477 1854 : FinishWalRecovery(void)
1478 : {
1479 1854 : EndOfWalRecoveryInfo *result = palloc_object(EndOfWalRecoveryInfo);
1480 : XLogRecPtr lastRec;
1481 : TimeLineID lastRecTLI;
1482 : XLogRecPtr endOfLog;
1483 :
1484 : /*
1485 : * Kill WAL receiver, if it's still running, before we continue to write
1486 : * the startup checkpoint and aborted-contrecord records. It will trump
1487 : * over these records and subsequent ones if it's still alive when we
1488 : * start writing WAL.
1489 : */
1490 1854 : XLogShutdownWalRcv();
1491 :
1492 : /*
1493 : * Shutdown the slot sync worker to drop any temporary slots acquired by
1494 : * it and to prevent it from keep trying to fetch the failover slots.
1495 : *
1496 : * We do not update the 'synced' column in 'pg_replication_slots' system
1497 : * view from true to false here, as any failed update could leave 'synced'
1498 : * column false for some slots. This could cause issues during slot sync
1499 : * after restarting the server as a standby. While updating the 'synced'
1500 : * column after switching to the new timeline is an option, it does not
1501 : * simplify the handling for the 'synced' column. Therefore, we retain the
1502 : * 'synced' column as true after promotion as it may provide useful
1503 : * information about the slot origin.
1504 : */
1505 1854 : ShutDownSlotSync();
1506 :
1507 : /*
1508 : * We are now done reading the xlog from stream. Turn off streaming
1509 : * recovery to force fetching the files (which would be required at end of
1510 : * recovery, e.g., timeline history file) from archive or pg_wal.
1511 : *
1512 : * Note that standby mode must be turned off after killing WAL receiver,
1513 : * i.e., calling XLogShutdownWalRcv().
1514 : */
1515 : Assert(!WalRcvStreaming());
1516 1854 : StandbyMode = false;
1517 :
1518 : /*
1519 : * Determine where to start writing WAL next.
1520 : *
1521 : * Re-fetch the last valid or last applied record, so we can identify the
1522 : * exact endpoint of what we consider the valid portion of WAL. There may
1523 : * be an incomplete continuation record after that, in which case
1524 : * 'abortedRecPtr' and 'missingContrecPtr' are set and the caller will
1525 : * write a special OVERWRITE_CONTRECORD message to mark that the rest of
1526 : * it is intentionally missing. See CreateOverwriteContrecordRecord().
1527 : *
1528 : * An important side-effect of this is to load the last page into
1529 : * xlogreader. The caller uses it to initialize the WAL for writing.
1530 : */
1531 1854 : if (!InRecovery)
1532 : {
1533 1532 : lastRec = CheckPointLoc;
1534 1532 : lastRecTLI = CheckPointTLI;
1535 : }
1536 : else
1537 : {
1538 322 : lastRec = XLogRecoveryCtl->lastReplayedReadRecPtr;
1539 322 : lastRecTLI = XLogRecoveryCtl->lastReplayedTLI;
1540 : }
1541 1854 : XLogPrefetcherBeginRead(xlogprefetcher, lastRec);
1542 1854 : (void) ReadRecord(xlogprefetcher, PANIC, false, lastRecTLI);
1543 1854 : endOfLog = xlogreader->EndRecPtr;
1544 :
1545 : /*
1546 : * Remember the TLI in the filename of the XLOG segment containing the
1547 : * end-of-log. It could be different from the timeline that endOfLog
1548 : * nominally belongs to, if there was a timeline switch in that segment,
1549 : * and we were reading the old WAL from a segment belonging to a higher
1550 : * timeline.
1551 : */
1552 1854 : result->endOfLogTLI = xlogreader->seg.ws_tli;
1553 :
1554 1854 : if (ArchiveRecoveryRequested)
1555 : {
1556 : /*
1557 : * We are no longer in archive recovery state.
1558 : *
1559 : * We are now done reading the old WAL. Turn off archive fetching if
1560 : * it was active.
1561 : */
1562 : Assert(InArchiveRecovery);
1563 108 : InArchiveRecovery = false;
1564 :
1565 : /*
1566 : * If the ending log segment is still open, close it (to avoid
1567 : * problems on Windows with trying to rename or delete an open file).
1568 : */
1569 108 : if (readFile >= 0)
1570 : {
1571 108 : close(readFile);
1572 108 : readFile = -1;
1573 : }
1574 : }
1575 :
1576 : /*
1577 : * Copy the last partial block to the caller, for initializing the WAL
1578 : * buffer for appending new WAL.
1579 : */
1580 1854 : if (endOfLog % XLOG_BLCKSZ != 0)
1581 : {
1582 : char *page;
1583 : int len;
1584 : XLogRecPtr pageBeginPtr;
1585 :
1586 1812 : pageBeginPtr = endOfLog - (endOfLog % XLOG_BLCKSZ);
1587 : Assert(readOff == XLogSegmentOffset(pageBeginPtr, wal_segment_size));
1588 :
1589 : /* Copy the valid part of the last block */
1590 1812 : len = endOfLog % XLOG_BLCKSZ;
1591 1812 : page = palloc(len);
1592 1812 : memcpy(page, xlogreader->readBuf, len);
1593 :
1594 1812 : result->lastPageBeginPtr = pageBeginPtr;
1595 1812 : result->lastPage = page;
1596 : }
1597 : else
1598 : {
1599 : /* There is no partial block to copy. */
1600 42 : result->lastPageBeginPtr = endOfLog;
1601 42 : result->lastPage = NULL;
1602 : }
1603 :
1604 : /*
1605 : * Create a comment for the history file to explain why and where timeline
1606 : * changed.
1607 : */
1608 1854 : result->recoveryStopReason = getRecoveryStopReason();
1609 :
1610 1854 : result->lastRec = lastRec;
1611 1854 : result->lastRecTLI = lastRecTLI;
1612 1854 : result->endOfLog = endOfLog;
1613 :
1614 1854 : result->abortedRecPtr = abortedRecPtr;
1615 1854 : result->missingContrecPtr = missingContrecPtr;
1616 :
1617 1854 : result->standby_signal_file_found = standby_signal_file_found;
1618 1854 : result->recovery_signal_file_found = recovery_signal_file_found;
1619 :
1620 1854 : return result;
1621 : }
1622 :
1623 : /*
1624 : * Clean up the WAL reader and leftovers from restoring WAL from archive
1625 : */
1626 : void
1627 1854 : ShutdownWalRecovery(void)
1628 : {
1629 : char recoveryPath[MAXPGPATH];
1630 :
1631 : /* Final update of pg_stat_recovery_prefetch. */
1632 1854 : XLogPrefetcherComputeStats(xlogprefetcher);
1633 :
1634 : /* Shut down xlogreader */
1635 1854 : if (readFile >= 0)
1636 : {
1637 1746 : close(readFile);
1638 1746 : readFile = -1;
1639 : }
1640 1854 : pfree(xlogreader->private_data);
1641 1854 : XLogReaderFree(xlogreader);
1642 1854 : XLogPrefetcherFree(xlogprefetcher);
1643 :
1644 1854 : if (ArchiveRecoveryRequested)
1645 : {
1646 : /*
1647 : * Since there might be a partial WAL segment named RECOVERYXLOG, get
1648 : * rid of it.
1649 : */
1650 108 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYXLOG");
1651 108 : unlink(recoveryPath); /* ignore any error */
1652 :
1653 : /* Get rid of any remaining recovered timeline-history file, too */
1654 108 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYHISTORY");
1655 108 : unlink(recoveryPath); /* ignore any error */
1656 : }
1657 :
1658 : /*
1659 : * We don't need the latch anymore. It's not strictly necessary to disown
1660 : * it, but let's do it for the sake of tidiness.
1661 : */
1662 1854 : if (ArchiveRecoveryRequested)
1663 108 : DisownLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
1664 1854 : }
1665 :
1666 : /*
1667 : * Perform WAL recovery.
1668 : *
1669 : * If the system was shut down cleanly, this is never called.
1670 : */
1671 : void
1672 446 : PerformWalRecovery(void)
1673 : {
1674 : XLogRecord *record;
1675 446 : bool reachedRecoveryTarget = false;
1676 : TimeLineID replayTLI;
1677 :
1678 : /*
1679 : * Initialize shared variables for tracking progress of WAL replay, as if
1680 : * we had just replayed the record before the REDO location (or the
1681 : * checkpoint record itself, if it's a shutdown checkpoint).
1682 : */
1683 446 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
1684 446 : if (RedoStartLSN < CheckPointLoc)
1685 : {
1686 242 : XLogRecoveryCtl->lastReplayedReadRecPtr = InvalidXLogRecPtr;
1687 242 : XLogRecoveryCtl->lastReplayedEndRecPtr = RedoStartLSN;
1688 242 : XLogRecoveryCtl->lastReplayedTLI = RedoStartTLI;
1689 : }
1690 : else
1691 : {
1692 204 : XLogRecoveryCtl->lastReplayedReadRecPtr = xlogreader->ReadRecPtr;
1693 204 : XLogRecoveryCtl->lastReplayedEndRecPtr = xlogreader->EndRecPtr;
1694 204 : XLogRecoveryCtl->lastReplayedTLI = CheckPointTLI;
1695 : }
1696 446 : XLogRecoveryCtl->replayEndRecPtr = XLogRecoveryCtl->lastReplayedEndRecPtr;
1697 446 : XLogRecoveryCtl->replayEndTLI = XLogRecoveryCtl->lastReplayedTLI;
1698 446 : XLogRecoveryCtl->recoveryLastXTime = 0;
1699 446 : XLogRecoveryCtl->currentChunkStartTime = 0;
1700 446 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_NOT_PAUSED;
1701 446 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
1702 :
1703 : /* Also ensure XLogReceiptTime has a sane value */
1704 446 : XLogReceiptTime = GetCurrentTimestamp();
1705 :
1706 : /*
1707 : * Let postmaster know we've started redo now, so that it can launch the
1708 : * archiver if necessary.
1709 : */
1710 446 : if (IsUnderPostmaster)
1711 428 : SendPostmasterSignal(PMSIGNAL_RECOVERY_STARTED);
1712 :
1713 : /*
1714 : * Allow read-only connections immediately if we're consistent already.
1715 : */
1716 446 : CheckRecoveryConsistency();
1717 :
1718 : /*
1719 : * Find the first record that logically follows the checkpoint --- it
1720 : * might physically precede it, though.
1721 : */
1722 446 : if (RedoStartLSN < CheckPointLoc)
1723 : {
1724 : /* back up to find the record */
1725 242 : replayTLI = RedoStartTLI;
1726 242 : XLogPrefetcherBeginRead(xlogprefetcher, RedoStartLSN);
1727 242 : record = ReadRecord(xlogprefetcher, PANIC, false, replayTLI);
1728 :
1729 : /*
1730 : * If a checkpoint record's redo pointer points back to an earlier
1731 : * LSN, the record at that LSN should be an XLOG_CHECKPOINT_REDO
1732 : * record.
1733 : */
1734 242 : if (record->xl_rmid != RM_XLOG_ID ||
1735 242 : (record->xl_info & ~XLR_INFO_MASK) != XLOG_CHECKPOINT_REDO)
1736 0 : ereport(FATAL,
1737 : errmsg("unexpected record type found at redo point %X/%08X",
1738 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr)));
1739 : }
1740 : else
1741 : {
1742 : /* just have to read next record after CheckPoint */
1743 : Assert(xlogreader->ReadRecPtr == CheckPointLoc);
1744 204 : replayTLI = CheckPointTLI;
1745 204 : record = ReadRecord(xlogprefetcher, LOG, false, replayTLI);
1746 : }
1747 :
1748 446 : if (record != NULL)
1749 : {
1750 : TimestampTz xtime;
1751 : PGRUsage ru0;
1752 :
1753 428 : pg_rusage_init(&ru0);
1754 :
1755 428 : InRedo = true;
1756 :
1757 428 : RmgrStartup();
1758 :
1759 428 : ereport(LOG,
1760 : errmsg("redo starts at %X/%08X",
1761 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr)));
1762 :
1763 : /* Prepare to report progress of the redo phase. */
1764 428 : if (!StandbyMode)
1765 216 : begin_startup_progress_phase();
1766 :
1767 : /*
1768 : * main redo apply loop
1769 : */
1770 : do
1771 : {
1772 5537712 : if (!StandbyMode)
1773 527900 : ereport_startup_progress("redo in progress, elapsed time: %ld.%02d s, current LSN: %X/%08X",
1774 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr));
1775 :
1776 : #ifdef WAL_DEBUG
1777 : if (XLOG_DEBUG)
1778 : {
1779 : StringInfoData buf;
1780 :
1781 : initStringInfo(&buf);
1782 : appendStringInfo(&buf, "REDO @ %X/%08X; LSN %X/%08X: ",
1783 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr),
1784 : LSN_FORMAT_ARGS(xlogreader->EndRecPtr));
1785 : xlog_outrec(&buf, xlogreader);
1786 : appendStringInfoString(&buf, " - ");
1787 : xlog_outdesc(&buf, xlogreader);
1788 : elog(LOG, "%s", buf.data);
1789 : pfree(buf.data);
1790 : }
1791 : #endif
1792 :
1793 : /* Handle interrupt signals of startup process */
1794 5537712 : ProcessStartupProcInterrupts();
1795 :
1796 : /*
1797 : * Pause WAL replay, if requested by a hot-standby session via
1798 : * SetRecoveryPause().
1799 : *
1800 : * Note that we intentionally don't take the info_lck spinlock
1801 : * here. We might therefore read a slightly stale value of the
1802 : * recoveryPause flag, but it can't be very stale (no worse than
1803 : * the last spinlock we did acquire). Since a pause request is a
1804 : * pretty asynchronous thing anyway, possibly responding to it one
1805 : * WAL record later than we otherwise would is a minor issue, so
1806 : * it doesn't seem worth adding another spinlock cycle to prevent
1807 : * that.
1808 : */
1809 5537712 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
1810 : RECOVERY_NOT_PAUSED)
1811 0 : recoveryPausesHere(false);
1812 :
1813 : /*
1814 : * Have we reached our recovery target?
1815 : */
1816 5537712 : if (recoveryStopsBefore(xlogreader))
1817 : {
1818 4 : reachedRecoveryTarget = true;
1819 4 : break;
1820 : }
1821 :
1822 : /*
1823 : * If we've been asked to lag the primary, wait on latch until
1824 : * enough time has passed.
1825 : */
1826 5537708 : if (recoveryApplyDelay(xlogreader))
1827 : {
1828 : /*
1829 : * We test for paused recovery again here. If user sets
1830 : * delayed apply, it may be because they expect to pause
1831 : * recovery in case of problems, so we must test again here
1832 : * otherwise pausing during the delay-wait wouldn't work.
1833 : */
1834 56 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
1835 : RECOVERY_NOT_PAUSED)
1836 0 : recoveryPausesHere(false);
1837 : }
1838 :
1839 : /*
1840 : * Apply the record
1841 : */
1842 5537708 : ApplyWalRecord(xlogreader, record, &replayTLI);
1843 :
1844 : /* Exit loop if we reached inclusive recovery target */
1845 5537704 : if (recoveryStopsAfter(xlogreader))
1846 : {
1847 10 : reachedRecoveryTarget = true;
1848 10 : break;
1849 : }
1850 :
1851 : /*
1852 : * If we replayed an LSN that someone was waiting for then walk
1853 : * over the shared memory array and set latches to notify the
1854 : * waiters.
1855 : */
1856 11075388 : if (waitLSNState &&
1857 5537694 : (XLogRecoveryCtl->lastReplayedEndRecPtr >=
1858 5537694 : pg_atomic_read_u64(&waitLSNState->minWaitedLSN[WAIT_LSN_TYPE_STANDBY_REPLAY])))
1859 16 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_REPLAY, XLogRecoveryCtl->lastReplayedEndRecPtr);
1860 :
1861 : /* Else, try to fetch the next WAL record */
1862 5537694 : record = ReadRecord(xlogprefetcher, LOG, false, replayTLI);
1863 5537576 : } while (record != NULL);
1864 :
1865 : /*
1866 : * end of main redo apply loop
1867 : */
1868 :
1869 306 : if (reachedRecoveryTarget)
1870 : {
1871 14 : if (!reachedConsistency)
1872 0 : ereport(FATAL,
1873 : (errmsg("requested recovery stop point is before consistent recovery point")));
1874 :
1875 : /*
1876 : * This is the last point where we can restart recovery with a new
1877 : * recovery target, if we shutdown and begin again. After this,
1878 : * Resource Managers may choose to do permanent corrective actions
1879 : * at end of recovery.
1880 : */
1881 14 : switch (recoveryTargetAction)
1882 : {
1883 0 : case RECOVERY_TARGET_ACTION_SHUTDOWN:
1884 :
1885 : /*
1886 : * exit with special return code to request shutdown of
1887 : * postmaster. Log messages issued from postmaster.
1888 : */
1889 0 : proc_exit(3);
1890 :
1891 2 : case RECOVERY_TARGET_ACTION_PAUSE:
1892 2 : SetRecoveryPause(true);
1893 2 : recoveryPausesHere(true);
1894 :
1895 : /* drop into promote */
1896 :
1897 14 : case RECOVERY_TARGET_ACTION_PROMOTE:
1898 14 : break;
1899 : }
1900 : }
1901 :
1902 306 : RmgrCleanup();
1903 :
1904 306 : ereport(LOG,
1905 : errmsg("redo done at %X/%08X system usage: %s",
1906 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr),
1907 : pg_rusage_show(&ru0)));
1908 306 : xtime = GetLatestXTime();
1909 306 : if (xtime)
1910 74 : ereport(LOG,
1911 : (errmsg("last completed transaction was at log time %s",
1912 : timestamptz_to_str(xtime))));
1913 :
1914 306 : InRedo = false;
1915 : }
1916 : else
1917 : {
1918 : /* there are no WAL records following the checkpoint */
1919 18 : ereport(LOG,
1920 : (errmsg("redo is not required")));
1921 : }
1922 :
1923 : /*
1924 : * This check is intentionally after the above log messages that indicate
1925 : * how far recovery went.
1926 : */
1927 324 : if (ArchiveRecoveryRequested &&
1928 110 : recoveryTarget != RECOVERY_TARGET_UNSET &&
1929 16 : !reachedRecoveryTarget)
1930 2 : ereport(FATAL,
1931 : (errcode(ERRCODE_CONFIG_FILE_ERROR),
1932 : errmsg("recovery ended before configured recovery target was reached")));
1933 322 : }
1934 :
1935 : /*
1936 : * Subroutine of PerformWalRecovery, to apply one WAL record.
1937 : */
1938 : static void
1939 5537708 : ApplyWalRecord(XLogReaderState *xlogreader, XLogRecord *record, TimeLineID *replayTLI)
1940 : {
1941 : ErrorContextCallback errcallback;
1942 5537708 : bool switchedTLI = false;
1943 :
1944 : /* Setup error traceback support for ereport() */
1945 5537708 : errcallback.callback = rm_redo_error_callback;
1946 5537708 : errcallback.arg = xlogreader;
1947 5537708 : errcallback.previous = error_context_stack;
1948 5537708 : error_context_stack = &errcallback;
1949 :
1950 : /*
1951 : * TransamVariables->nextXid must be beyond record's xid.
1952 : */
1953 5537708 : AdvanceNextFullTransactionIdPastXid(record->xl_xid);
1954 :
1955 : /*
1956 : * Before replaying this record, check if this record causes the current
1957 : * timeline to change. The record is already considered to be part of the
1958 : * new timeline, so we update replayTLI before replaying it. That's
1959 : * important so that replayEndTLI, which is recorded as the minimum
1960 : * recovery point's TLI if recovery stops after this record, is set
1961 : * correctly.
1962 : */
1963 5537708 : if (record->xl_rmid == RM_XLOG_ID)
1964 : {
1965 87538 : TimeLineID newReplayTLI = *replayTLI;
1966 87538 : TimeLineID prevReplayTLI = *replayTLI;
1967 87538 : uint8 info = record->xl_info & ~XLR_INFO_MASK;
1968 :
1969 87538 : if (info == XLOG_CHECKPOINT_SHUTDOWN)
1970 : {
1971 : CheckPoint checkPoint;
1972 :
1973 78 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
1974 78 : newReplayTLI = checkPoint.ThisTimeLineID;
1975 78 : prevReplayTLI = checkPoint.PrevTimeLineID;
1976 : }
1977 87460 : else if (info == XLOG_END_OF_RECOVERY)
1978 : {
1979 : xl_end_of_recovery xlrec;
1980 :
1981 22 : memcpy(&xlrec, XLogRecGetData(xlogreader), sizeof(xl_end_of_recovery));
1982 22 : newReplayTLI = xlrec.ThisTimeLineID;
1983 22 : prevReplayTLI = xlrec.PrevTimeLineID;
1984 : }
1985 :
1986 87538 : if (newReplayTLI != *replayTLI)
1987 : {
1988 : /* Check that it's OK to switch to this TLI */
1989 24 : checkTimeLineSwitch(xlogreader->EndRecPtr,
1990 : newReplayTLI, prevReplayTLI, *replayTLI);
1991 :
1992 : /* Following WAL records should be run with new TLI */
1993 24 : *replayTLI = newReplayTLI;
1994 24 : switchedTLI = true;
1995 : }
1996 : }
1997 :
1998 : /*
1999 : * Update shared replayEndRecPtr before replaying this record, so that
2000 : * XLogFlush will update minRecoveryPoint correctly.
2001 : */
2002 5537708 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
2003 5537708 : XLogRecoveryCtl->replayEndRecPtr = xlogreader->EndRecPtr;
2004 5537708 : XLogRecoveryCtl->replayEndTLI = *replayTLI;
2005 5537708 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
2006 :
2007 : /*
2008 : * If we are attempting to enter Hot Standby mode, process XIDs we see
2009 : */
2010 5537708 : if (standbyState >= STANDBY_INITIALIZED &&
2011 5049598 : TransactionIdIsValid(record->xl_xid))
2012 4947146 : RecordKnownAssignedTransactionIds(record->xl_xid);
2013 :
2014 : /*
2015 : * Some XLOG record types that are related to recovery are processed
2016 : * directly here, rather than in xlog_redo()
2017 : */
2018 5537708 : if (record->xl_rmid == RM_XLOG_ID)
2019 87538 : xlogrecovery_redo(xlogreader, *replayTLI);
2020 :
2021 : /* Now apply the WAL record itself */
2022 5537708 : GetRmgr(record->xl_rmid).rm_redo(xlogreader);
2023 :
2024 : /*
2025 : * After redo, check whether the backup pages associated with the WAL
2026 : * record are consistent with the existing pages. This check is done only
2027 : * if consistency check is enabled for this record.
2028 : */
2029 5537704 : if ((record->xl_info & XLR_CHECK_CONSISTENCY) != 0)
2030 4408750 : verifyBackupPageConsistency(xlogreader);
2031 :
2032 : /* Pop the error context stack */
2033 5537704 : error_context_stack = errcallback.previous;
2034 :
2035 : /*
2036 : * Update lastReplayedEndRecPtr after this record has been successfully
2037 : * replayed.
2038 : */
2039 5537704 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
2040 5537704 : XLogRecoveryCtl->lastReplayedReadRecPtr = xlogreader->ReadRecPtr;
2041 5537704 : XLogRecoveryCtl->lastReplayedEndRecPtr = xlogreader->EndRecPtr;
2042 5537704 : XLogRecoveryCtl->lastReplayedTLI = *replayTLI;
2043 5537704 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
2044 :
2045 : /* ------
2046 : * Wakeup walsenders:
2047 : *
2048 : * On the standby, the WAL is flushed first (which will only wake up
2049 : * physical walsenders) and then applied, which will only wake up logical
2050 : * walsenders.
2051 : *
2052 : * Indeed, logical walsenders on standby can't decode and send data until
2053 : * it's been applied.
2054 : *
2055 : * Physical walsenders don't need to be woken up during replay unless
2056 : * cascading replication is allowed and time line change occurred (so that
2057 : * they can notice that they are on a new time line).
2058 : *
2059 : * That's why the wake up conditions are for:
2060 : *
2061 : * - physical walsenders in case of new time line and cascade
2062 : * replication is allowed
2063 : * - logical walsenders in case cascade replication is allowed (could not
2064 : * be created otherwise)
2065 : * ------
2066 : */
2067 5537704 : if (AllowCascadeReplication())
2068 5158888 : WalSndWakeup(switchedTLI, true);
2069 :
2070 : /*
2071 : * If rm_redo called XLogRequestWalReceiverReply, then we wake up the
2072 : * receiver so that it notices the updated lastReplayedEndRecPtr and sends
2073 : * a reply to the primary.
2074 : */
2075 5537704 : if (doRequestWalReceiverReply)
2076 : {
2077 4 : doRequestWalReceiverReply = false;
2078 4 : WalRcvForceReply();
2079 : }
2080 :
2081 : /* Allow read-only connections if we're consistent now */
2082 5537704 : CheckRecoveryConsistency();
2083 :
2084 : /* Is this a timeline switch? */
2085 5537704 : if (switchedTLI)
2086 : {
2087 : /*
2088 : * Before we continue on the new timeline, clean up any (possibly
2089 : * bogus) future WAL segments on the old timeline.
2090 : */
2091 24 : RemoveNonParentXlogFiles(xlogreader->EndRecPtr, *replayTLI);
2092 :
2093 : /* Reset the prefetcher. */
2094 24 : XLogPrefetchReconfigure();
2095 : }
2096 5537704 : }
2097 :
2098 : /*
2099 : * Some XLOG RM record types that are directly related to WAL recovery are
2100 : * handled here rather than in the xlog_redo()
2101 : */
2102 : static void
2103 87538 : xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI)
2104 : {
2105 87538 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2106 87538 : XLogRecPtr lsn = record->EndRecPtr;
2107 :
2108 : Assert(XLogRecGetRmid(record) == RM_XLOG_ID);
2109 :
2110 87538 : if (info == XLOG_OVERWRITE_CONTRECORD)
2111 : {
2112 : /* Verify the payload of a XLOG_OVERWRITE_CONTRECORD record. */
2113 : xl_overwrite_contrecord xlrec;
2114 :
2115 2 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_overwrite_contrecord));
2116 2 : if (xlrec.overwritten_lsn != record->overwrittenRecPtr)
2117 0 : elog(FATAL, "mismatching overwritten LSN %X/%08X -> %X/%08X",
2118 : LSN_FORMAT_ARGS(xlrec.overwritten_lsn),
2119 : LSN_FORMAT_ARGS(record->overwrittenRecPtr));
2120 :
2121 : /* We have safely skipped the aborted record */
2122 2 : abortedRecPtr = InvalidXLogRecPtr;
2123 2 : missingContrecPtr = InvalidXLogRecPtr;
2124 :
2125 2 : ereport(LOG,
2126 : errmsg("successfully skipped missing contrecord at %X/%08X, overwritten at %s",
2127 : LSN_FORMAT_ARGS(xlrec.overwritten_lsn),
2128 : timestamptz_to_str(xlrec.overwrite_time)));
2129 :
2130 : /* Verifying the record should only happen once */
2131 2 : record->overwrittenRecPtr = InvalidXLogRecPtr;
2132 : }
2133 87536 : else if (info == XLOG_BACKUP_END)
2134 : {
2135 : XLogRecPtr startpoint;
2136 :
2137 188 : memcpy(&startpoint, XLogRecGetData(record), sizeof(startpoint));
2138 :
2139 188 : if (backupStartPoint == startpoint)
2140 : {
2141 : /*
2142 : * We have reached the end of base backup, the point where
2143 : * pg_backup_stop() was done. The data on disk is now consistent
2144 : * (assuming we have also reached minRecoveryPoint). Set
2145 : * backupEndPoint to the current LSN, so that the next call to
2146 : * CheckRecoveryConsistency() will notice it and do the
2147 : * end-of-backup processing.
2148 : */
2149 154 : elog(DEBUG1, "end of backup record reached");
2150 :
2151 154 : backupEndPoint = lsn;
2152 : }
2153 : else
2154 34 : elog(DEBUG1, "saw end-of-backup record for backup starting at %X/%08X, waiting for %X/%08X",
2155 : LSN_FORMAT_ARGS(startpoint), LSN_FORMAT_ARGS(backupStartPoint));
2156 : }
2157 87538 : }
2158 :
2159 : /*
2160 : * Verify that, in non-test mode, ./pg_tblspc doesn't contain any real
2161 : * directories.
2162 : *
2163 : * Replay of database creation XLOG records for databases that were later
2164 : * dropped can create fake directories in pg_tblspc. By the time consistency
2165 : * is reached these directories should have been removed; here we verify
2166 : * that this did indeed happen. This is to be called at the point where
2167 : * consistent state is reached.
2168 : *
2169 : * allow_in_place_tablespaces turns the PANIC into a WARNING, which is
2170 : * useful for testing purposes, and also allows for an escape hatch in case
2171 : * things go south.
2172 : */
2173 : static void
2174 244 : CheckTablespaceDirectory(void)
2175 : {
2176 : DIR *dir;
2177 : struct dirent *de;
2178 :
2179 244 : dir = AllocateDir(PG_TBLSPC_DIR);
2180 746 : while ((de = ReadDir(dir, PG_TBLSPC_DIR)) != NULL)
2181 : {
2182 : char path[MAXPGPATH + sizeof(PG_TBLSPC_DIR)];
2183 :
2184 : /* Skip entries of non-oid names */
2185 502 : if (strspn(de->d_name, "0123456789") != strlen(de->d_name))
2186 488 : continue;
2187 :
2188 14 : snprintf(path, sizeof(path), "%s/%s", PG_TBLSPC_DIR, de->d_name);
2189 :
2190 14 : if (get_dirent_type(path, de, false, ERROR) != PGFILETYPE_LNK)
2191 8 : ereport(allow_in_place_tablespaces ? WARNING : PANIC,
2192 : (errcode(ERRCODE_DATA_CORRUPTED),
2193 : errmsg("unexpected directory entry \"%s\" found in %s",
2194 : de->d_name, PG_TBLSPC_DIR),
2195 : errdetail("All directory entries in %s/ should be symbolic links.",
2196 : PG_TBLSPC_DIR),
2197 : errhint("Remove those directories, or set \"allow_in_place_tablespaces\" to ON transiently to let recovery complete.")));
2198 : }
2199 244 : }
2200 :
2201 : /*
2202 : * Checks if recovery has reached a consistent state. When consistency is
2203 : * reached and we have a valid starting standby snapshot, tell postmaster
2204 : * that it can start accepting read-only connections.
2205 : */
2206 : static void
2207 5538154 : CheckRecoveryConsistency(void)
2208 : {
2209 : XLogRecPtr lastReplayedEndRecPtr;
2210 : TimeLineID lastReplayedTLI;
2211 :
2212 : /*
2213 : * During crash recovery, we don't reach a consistent state until we've
2214 : * replayed all the WAL.
2215 : */
2216 5538154 : if (!XLogRecPtrIsValid(minRecoveryPoint))
2217 517684 : return;
2218 :
2219 : Assert(InArchiveRecovery);
2220 :
2221 : /*
2222 : * assume that we are called in the startup process, and hence don't need
2223 : * a lock to read lastReplayedEndRecPtr
2224 : */
2225 5020470 : lastReplayedEndRecPtr = XLogRecoveryCtl->lastReplayedEndRecPtr;
2226 5020470 : lastReplayedTLI = XLogRecoveryCtl->lastReplayedTLI;
2227 :
2228 : /*
2229 : * Have we reached the point where our base backup was completed?
2230 : */
2231 5020470 : if (XLogRecPtrIsValid(backupEndPoint) &&
2232 226 : backupEndPoint <= lastReplayedEndRecPtr)
2233 : {
2234 158 : XLogRecPtr saveBackupStartPoint = backupStartPoint;
2235 158 : XLogRecPtr saveBackupEndPoint = backupEndPoint;
2236 :
2237 158 : elog(DEBUG1, "end of backup reached");
2238 :
2239 : /*
2240 : * We have reached the end of base backup, as indicated by pg_control.
2241 : * Update the control file accordingly.
2242 : */
2243 158 : ReachedEndOfBackup(lastReplayedEndRecPtr, lastReplayedTLI);
2244 158 : backupStartPoint = InvalidXLogRecPtr;
2245 158 : backupEndPoint = InvalidXLogRecPtr;
2246 158 : backupEndRequired = false;
2247 :
2248 158 : ereport(LOG,
2249 : errmsg("completed backup recovery with redo LSN %X/%08X and end LSN %X/%08X",
2250 : LSN_FORMAT_ARGS(saveBackupStartPoint),
2251 : LSN_FORMAT_ARGS(saveBackupEndPoint)));
2252 : }
2253 :
2254 : /*
2255 : * Have we passed our safe starting point? Note that minRecoveryPoint is
2256 : * known to be incorrectly set if recovering from a backup, until the
2257 : * XLOG_BACKUP_END arrives to advise us of the correct minRecoveryPoint.
2258 : * All we know prior to that is that we're not consistent yet.
2259 : */
2260 5020470 : if (!reachedConsistency && !backupEndRequired &&
2261 15358 : minRecoveryPoint <= lastReplayedEndRecPtr)
2262 : {
2263 : /*
2264 : * Check to see if the XLOG sequence contained any unresolved
2265 : * references to uninitialized pages.
2266 : */
2267 244 : XLogCheckInvalidPages();
2268 :
2269 : /*
2270 : * Check that pg_tblspc doesn't contain any real directories. Replay
2271 : * of Database/CREATE_* records may have created fictitious tablespace
2272 : * directories that should have been removed by the time consistency
2273 : * was reached.
2274 : */
2275 244 : CheckTablespaceDirectory();
2276 :
2277 244 : reachedConsistency = true;
2278 244 : SendPostmasterSignal(PMSIGNAL_RECOVERY_CONSISTENT);
2279 244 : ereport(LOG,
2280 : errmsg("consistent recovery state reached at %X/%08X",
2281 : LSN_FORMAT_ARGS(lastReplayedEndRecPtr)));
2282 : }
2283 :
2284 : /*
2285 : * Have we got a valid starting snapshot that will allow queries to be
2286 : * run? If so, we can tell postmaster that the database is consistent now,
2287 : * enabling connections.
2288 : */
2289 5020470 : if (standbyState == STANDBY_SNAPSHOT_READY &&
2290 5019992 : !LocalHotStandbyActive &&
2291 228 : reachedConsistency &&
2292 : IsUnderPostmaster)
2293 : {
2294 228 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
2295 228 : XLogRecoveryCtl->SharedHotStandbyActive = true;
2296 228 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
2297 :
2298 228 : LocalHotStandbyActive = true;
2299 :
2300 228 : SendPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY);
2301 : }
2302 : }
2303 :
2304 : /*
2305 : * Error context callback for errors occurring during rm_redo().
2306 : */
2307 : static void
2308 296 : rm_redo_error_callback(void *arg)
2309 : {
2310 296 : XLogReaderState *record = (XLogReaderState *) arg;
2311 : StringInfoData buf;
2312 :
2313 296 : initStringInfo(&buf);
2314 296 : xlog_outdesc(&buf, record);
2315 296 : xlog_block_info(&buf, record);
2316 :
2317 : /* translator: %s is a WAL record description */
2318 296 : errcontext("WAL redo at %X/%08X for %s",
2319 296 : LSN_FORMAT_ARGS(record->ReadRecPtr),
2320 : buf.data);
2321 :
2322 296 : pfree(buf.data);
2323 296 : }
2324 :
2325 : /*
2326 : * Returns a string describing an XLogRecord, consisting of its identity
2327 : * optionally followed by a colon, a space, and a further description.
2328 : */
2329 : void
2330 296 : xlog_outdesc(StringInfo buf, XLogReaderState *record)
2331 : {
2332 296 : RmgrData rmgr = GetRmgr(XLogRecGetRmid(record));
2333 296 : uint8 info = XLogRecGetInfo(record);
2334 : const char *id;
2335 :
2336 296 : appendStringInfoString(buf, rmgr.rm_name);
2337 296 : appendStringInfoChar(buf, '/');
2338 :
2339 296 : id = rmgr.rm_identify(info);
2340 296 : if (id == NULL)
2341 0 : appendStringInfo(buf, "UNKNOWN (%X): ", info & ~XLR_INFO_MASK);
2342 : else
2343 296 : appendStringInfo(buf, "%s: ", id);
2344 :
2345 296 : rmgr.rm_desc(buf, record);
2346 296 : }
2347 :
2348 : #ifdef WAL_DEBUG
2349 :
2350 : static void
2351 : xlog_outrec(StringInfo buf, XLogReaderState *record)
2352 : {
2353 : appendStringInfo(buf, "prev %X/%08X; xid %u",
2354 : LSN_FORMAT_ARGS(XLogRecGetPrev(record)),
2355 : XLogRecGetXid(record));
2356 :
2357 : appendStringInfo(buf, "; len %u",
2358 : XLogRecGetDataLen(record));
2359 :
2360 : xlog_block_info(buf, record);
2361 : }
2362 : #endif /* WAL_DEBUG */
2363 :
2364 : /*
2365 : * Returns a string giving information about all the blocks in an
2366 : * XLogRecord.
2367 : */
2368 : static void
2369 296 : xlog_block_info(StringInfo buf, XLogReaderState *record)
2370 : {
2371 : int block_id;
2372 :
2373 : /* decode block references */
2374 388 : for (block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
2375 : {
2376 : RelFileLocator rlocator;
2377 : ForkNumber forknum;
2378 : BlockNumber blk;
2379 :
2380 92 : if (!XLogRecGetBlockTagExtended(record, block_id,
2381 : &rlocator, &forknum, &blk, NULL))
2382 0 : continue;
2383 :
2384 92 : if (forknum != MAIN_FORKNUM)
2385 6 : appendStringInfo(buf, "; blkref #%d: rel %u/%u/%u, fork %u, blk %u",
2386 : block_id,
2387 : rlocator.spcOid, rlocator.dbOid,
2388 : rlocator.relNumber,
2389 : forknum,
2390 : blk);
2391 : else
2392 86 : appendStringInfo(buf, "; blkref #%d: rel %u/%u/%u, blk %u",
2393 : block_id,
2394 : rlocator.spcOid, rlocator.dbOid,
2395 : rlocator.relNumber,
2396 : blk);
2397 92 : if (XLogRecHasBlockImage(record, block_id))
2398 56 : appendStringInfoString(buf, " FPW");
2399 : }
2400 296 : }
2401 :
2402 :
2403 : /*
2404 : * Check that it's OK to switch to new timeline during recovery.
2405 : *
2406 : * 'lsn' is the address of the shutdown checkpoint record we're about to
2407 : * replay. (Currently, timeline can only change at a shutdown checkpoint).
2408 : */
2409 : static void
2410 24 : checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI, TimeLineID prevTLI,
2411 : TimeLineID replayTLI)
2412 : {
2413 : /* Check that the record agrees on what the current (old) timeline is */
2414 24 : if (prevTLI != replayTLI)
2415 0 : ereport(PANIC,
2416 : (errmsg("unexpected previous timeline ID %u (current timeline ID %u) in checkpoint record",
2417 : prevTLI, replayTLI)));
2418 :
2419 : /*
2420 : * The new timeline better be in the list of timelines we expect to see,
2421 : * according to the timeline history. It should also not decrease.
2422 : */
2423 24 : if (newTLI < replayTLI || !tliInHistory(newTLI, expectedTLEs))
2424 0 : ereport(PANIC,
2425 : (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
2426 : newTLI, replayTLI)));
2427 :
2428 : /*
2429 : * If we have not yet reached min recovery point, and we're about to
2430 : * switch to a timeline greater than the timeline of the min recovery
2431 : * point: trouble. After switching to the new timeline, we could not
2432 : * possibly visit the min recovery point on the correct timeline anymore.
2433 : * This can happen if there is a newer timeline in the archive that
2434 : * branched before the timeline the min recovery point is on, and you
2435 : * attempt to do PITR to the new timeline.
2436 : */
2437 24 : if (XLogRecPtrIsValid(minRecoveryPoint) &&
2438 20 : lsn < minRecoveryPoint &&
2439 2 : newTLI > minRecoveryPointTLI)
2440 0 : ereport(PANIC,
2441 : errmsg("unexpected timeline ID %u in checkpoint record, before reaching minimum recovery point %X/%08X on timeline %u",
2442 : newTLI,
2443 : LSN_FORMAT_ARGS(minRecoveryPoint),
2444 : minRecoveryPointTLI));
2445 :
2446 : /* Looks good */
2447 24 : }
2448 :
2449 :
2450 : /*
2451 : * Extract timestamp from WAL record.
2452 : *
2453 : * If the record contains a timestamp, returns true, and saves the timestamp
2454 : * in *recordXtime. If the record type has no timestamp, returns false.
2455 : * Currently, only transaction commit/abort records and restore points contain
2456 : * timestamps.
2457 : */
2458 : static bool
2459 88890 : getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime)
2460 : {
2461 88890 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2462 88890 : uint8 xact_info = info & XLOG_XACT_OPMASK;
2463 88890 : uint8 rmid = XLogRecGetRmid(record);
2464 :
2465 88890 : if (rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
2466 : {
2467 4 : *recordXtime = ((xl_restore_point *) XLogRecGetData(record))->rp_time;
2468 4 : return true;
2469 : }
2470 88886 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_COMMIT ||
2471 : xact_info == XLOG_XACT_COMMIT_PREPARED))
2472 : {
2473 81442 : *recordXtime = ((xl_xact_commit *) XLogRecGetData(record))->xact_time;
2474 81442 : return true;
2475 : }
2476 7444 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_ABORT ||
2477 : xact_info == XLOG_XACT_ABORT_PREPARED))
2478 : {
2479 7444 : *recordXtime = ((xl_xact_abort *) XLogRecGetData(record))->xact_time;
2480 7444 : return true;
2481 : }
2482 0 : return false;
2483 : }
2484 :
2485 : /*
2486 : * Checks whether the current buffer page and backup page stored in the
2487 : * WAL record are consistent or not. Before comparing the two pages, a
2488 : * masking can be applied to the pages to ignore certain areas like hint bits,
2489 : * unused space between pd_lower and pd_upper among other things. This
2490 : * function should be called once WAL replay has been completed for a
2491 : * given record.
2492 : */
2493 : static void
2494 4408750 : verifyBackupPageConsistency(XLogReaderState *record)
2495 : {
2496 4408750 : RmgrData rmgr = GetRmgr(XLogRecGetRmid(record));
2497 : RelFileLocator rlocator;
2498 : ForkNumber forknum;
2499 : BlockNumber blkno;
2500 : int block_id;
2501 :
2502 : /* Records with no backup blocks have no need for consistency checks. */
2503 4408750 : if (!XLogRecHasAnyBlockRefs(record))
2504 106 : return;
2505 :
2506 : Assert((XLogRecGetInfo(record) & XLR_CHECK_CONSISTENCY) != 0);
2507 :
2508 9153514 : for (block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
2509 : {
2510 : Buffer buf;
2511 : Page page;
2512 :
2513 4744870 : if (!XLogRecGetBlockTagExtended(record, block_id,
2514 : &rlocator, &forknum, &blkno, NULL))
2515 : {
2516 : /*
2517 : * WAL record doesn't contain a block reference with the given id.
2518 : * Do nothing.
2519 : */
2520 4022 : continue;
2521 : }
2522 :
2523 : Assert(XLogRecHasBlockImage(record, block_id));
2524 :
2525 4740848 : if (XLogRecBlockImageApply(record, block_id))
2526 : {
2527 : /*
2528 : * WAL record has already applied the page, so bypass the
2529 : * consistency check as that would result in comparing the full
2530 : * page stored in the record with itself.
2531 : */
2532 51084 : continue;
2533 : }
2534 :
2535 : /*
2536 : * Read the contents from the current buffer and store it in a
2537 : * temporary page.
2538 : */
2539 4689764 : buf = XLogReadBufferExtended(rlocator, forknum, blkno,
2540 : RBM_NORMAL_NO_LOG,
2541 : InvalidBuffer);
2542 4689764 : if (!BufferIsValid(buf))
2543 0 : continue;
2544 :
2545 4689764 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
2546 4689764 : page = BufferGetPage(buf);
2547 :
2548 : /*
2549 : * Take a copy of the local page where WAL has been applied to have a
2550 : * comparison base before masking it...
2551 : */
2552 4689764 : memcpy(replay_image_masked, page, BLCKSZ);
2553 :
2554 : /* No need for this page anymore now that a copy is in. */
2555 4689764 : UnlockReleaseBuffer(buf);
2556 :
2557 : /*
2558 : * If the block LSN is already ahead of this WAL record, we can't
2559 : * expect contents to match. This can happen if recovery is
2560 : * restarted.
2561 : */
2562 4689764 : if (PageGetLSN(replay_image_masked) > record->EndRecPtr)
2563 0 : continue;
2564 :
2565 : /*
2566 : * Read the contents from the backup copy, stored in WAL record and
2567 : * store it in a temporary page. There is no need to allocate a new
2568 : * page here, a local buffer is fine to hold its contents and a mask
2569 : * can be directly applied on it.
2570 : */
2571 4689764 : if (!RestoreBlockImage(record, block_id, primary_image_masked))
2572 0 : ereport(ERROR,
2573 : (errcode(ERRCODE_INTERNAL_ERROR),
2574 : errmsg_internal("%s", record->errormsg_buf)));
2575 :
2576 : /*
2577 : * If masking function is defined, mask both the primary and replay
2578 : * images
2579 : */
2580 4689764 : if (rmgr.rm_mask != NULL)
2581 : {
2582 4689764 : rmgr.rm_mask(replay_image_masked, blkno);
2583 4689764 : rmgr.rm_mask(primary_image_masked, blkno);
2584 : }
2585 :
2586 : /* Time to compare the primary and replay images. */
2587 4689764 : if (memcmp(replay_image_masked, primary_image_masked, BLCKSZ) != 0)
2588 : {
2589 0 : elog(FATAL,
2590 : "inconsistent page found, rel %u/%u/%u, forknum %u, blkno %u",
2591 : rlocator.spcOid, rlocator.dbOid, rlocator.relNumber,
2592 : forknum, blkno);
2593 : }
2594 : }
2595 : }
2596 :
2597 : /*
2598 : * For point-in-time recovery, this function decides whether we want to
2599 : * stop applying the XLOG before the current record.
2600 : *
2601 : * Returns true if we are stopping, false otherwise. If stopping, some
2602 : * information is saved in recoveryStopXid et al for use in annotating the
2603 : * new timeline's history file.
2604 : */
2605 : static bool
2606 5537712 : recoveryStopsBefore(XLogReaderState *record)
2607 : {
2608 5537712 : bool stopsHere = false;
2609 : uint8 xact_info;
2610 : bool isCommit;
2611 5537712 : TimestampTz recordXtime = 0;
2612 : TransactionId recordXid;
2613 :
2614 : /*
2615 : * Ignore recovery target settings when not in archive recovery (meaning
2616 : * we are in crash recovery).
2617 : */
2618 5537712 : if (!ArchiveRecoveryRequested)
2619 488082 : return false;
2620 :
2621 : /* Check if we should stop as soon as reaching consistency */
2622 5049630 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
2623 : {
2624 0 : ereport(LOG,
2625 : (errmsg("recovery stopping after reaching consistency")));
2626 :
2627 0 : recoveryStopAfter = false;
2628 0 : recoveryStopXid = InvalidTransactionId;
2629 0 : recoveryStopLSN = InvalidXLogRecPtr;
2630 0 : recoveryStopTime = 0;
2631 0 : recoveryStopName[0] = '\0';
2632 0 : return true;
2633 : }
2634 :
2635 : /* Check if target LSN has been reached */
2636 5049630 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
2637 17042 : !recoveryTargetInclusive &&
2638 982 : record->ReadRecPtr >= recoveryTargetLSN)
2639 : {
2640 4 : recoveryStopAfter = false;
2641 4 : recoveryStopXid = InvalidTransactionId;
2642 4 : recoveryStopLSN = record->ReadRecPtr;
2643 4 : recoveryStopTime = 0;
2644 4 : recoveryStopName[0] = '\0';
2645 4 : ereport(LOG,
2646 : errmsg("recovery stopping before WAL location (LSN) \"%X/%08X\"",
2647 : LSN_FORMAT_ARGS(recoveryStopLSN)));
2648 4 : return true;
2649 : }
2650 :
2651 : /* Otherwise we only consider stopping before COMMIT or ABORT records. */
2652 5049626 : if (XLogRecGetRmid(record) != RM_XACT_ID)
2653 5004594 : return false;
2654 :
2655 45032 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
2656 :
2657 45032 : if (xact_info == XLOG_XACT_COMMIT)
2658 : {
2659 40634 : isCommit = true;
2660 40634 : recordXid = XLogRecGetXid(record);
2661 : }
2662 4398 : else if (xact_info == XLOG_XACT_COMMIT_PREPARED)
2663 : {
2664 60 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
2665 : xl_xact_parsed_commit parsed;
2666 :
2667 60 : isCommit = true;
2668 60 : ParseCommitRecord(XLogRecGetInfo(record),
2669 : xlrec,
2670 : &parsed);
2671 60 : recordXid = parsed.twophase_xid;
2672 : }
2673 4338 : else if (xact_info == XLOG_XACT_ABORT)
2674 : {
2675 3696 : isCommit = false;
2676 3696 : recordXid = XLogRecGetXid(record);
2677 : }
2678 642 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
2679 : {
2680 26 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
2681 : xl_xact_parsed_abort parsed;
2682 :
2683 26 : isCommit = false;
2684 26 : ParseAbortRecord(XLogRecGetInfo(record),
2685 : xlrec,
2686 : &parsed);
2687 26 : recordXid = parsed.twophase_xid;
2688 : }
2689 : else
2690 616 : return false;
2691 :
2692 44416 : if (recoveryTarget == RECOVERY_TARGET_XID && !recoveryTargetInclusive)
2693 : {
2694 : /*
2695 : * There can be only one transaction end record with this exact
2696 : * transactionid
2697 : *
2698 : * when testing for an xid, we MUST test for equality only, since
2699 : * transactions are numbered in the order they start, not the order
2700 : * they complete. A higher numbered xid will complete before you about
2701 : * 50% of the time...
2702 : */
2703 0 : stopsHere = (recordXid == recoveryTargetXid);
2704 : }
2705 :
2706 : /*
2707 : * Note: we must fetch recordXtime regardless of recoveryTarget setting.
2708 : * We don't expect getRecordTimestamp ever to fail, since we already know
2709 : * this is a commit or abort record; but test its result anyway.
2710 : */
2711 44416 : if (getRecordTimestamp(record, &recordXtime) &&
2712 44416 : recoveryTarget == RECOVERY_TARGET_TIME)
2713 : {
2714 : /*
2715 : * There can be many transactions that share the same commit time, so
2716 : * we stop after the last one, if we are inclusive, or stop at the
2717 : * first one if we are exclusive
2718 : */
2719 0 : if (recoveryTargetInclusive)
2720 0 : stopsHere = (recordXtime > recoveryTargetTime);
2721 : else
2722 0 : stopsHere = (recordXtime >= recoveryTargetTime);
2723 : }
2724 :
2725 44416 : if (stopsHere)
2726 : {
2727 0 : recoveryStopAfter = false;
2728 0 : recoveryStopXid = recordXid;
2729 0 : recoveryStopTime = recordXtime;
2730 0 : recoveryStopLSN = InvalidXLogRecPtr;
2731 0 : recoveryStopName[0] = '\0';
2732 :
2733 0 : if (isCommit)
2734 : {
2735 0 : ereport(LOG,
2736 : (errmsg("recovery stopping before commit of transaction %u, time %s",
2737 : recoveryStopXid,
2738 : timestamptz_to_str(recoveryStopTime))));
2739 : }
2740 : else
2741 : {
2742 0 : ereport(LOG,
2743 : (errmsg("recovery stopping before abort of transaction %u, time %s",
2744 : recoveryStopXid,
2745 : timestamptz_to_str(recoveryStopTime))));
2746 : }
2747 : }
2748 :
2749 44416 : return stopsHere;
2750 : }
2751 :
2752 : /*
2753 : * Same as recoveryStopsBefore, but called after applying the record.
2754 : *
2755 : * We also track the timestamp of the latest applied COMMIT/ABORT
2756 : * record in XLogRecoveryCtl->recoveryLastXTime.
2757 : */
2758 : static bool
2759 5537704 : recoveryStopsAfter(XLogReaderState *record)
2760 : {
2761 : uint8 info;
2762 : uint8 xact_info;
2763 : uint8 rmid;
2764 5537704 : TimestampTz recordXtime = 0;
2765 :
2766 : /*
2767 : * Ignore recovery target settings when not in archive recovery (meaning
2768 : * we are in crash recovery).
2769 : */
2770 5537704 : if (!ArchiveRecoveryRequested)
2771 488082 : return false;
2772 :
2773 5049622 : info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2774 5049622 : rmid = XLogRecGetRmid(record);
2775 :
2776 : /*
2777 : * There can be many restore points that share the same name; we stop at
2778 : * the first one.
2779 : */
2780 5049622 : if (recoveryTarget == RECOVERY_TARGET_NAME &&
2781 40 : rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
2782 : {
2783 : xl_restore_point *recordRestorePointData;
2784 :
2785 6 : recordRestorePointData = (xl_restore_point *) XLogRecGetData(record);
2786 :
2787 6 : if (strcmp(recordRestorePointData->rp_name, recoveryTargetName) == 0)
2788 : {
2789 4 : recoveryStopAfter = true;
2790 4 : recoveryStopXid = InvalidTransactionId;
2791 4 : recoveryStopLSN = InvalidXLogRecPtr;
2792 4 : (void) getRecordTimestamp(record, &recoveryStopTime);
2793 4 : strlcpy(recoveryStopName, recordRestorePointData->rp_name, MAXFNAMELEN);
2794 :
2795 4 : ereport(LOG,
2796 : (errmsg("recovery stopping at restore point \"%s\", time %s",
2797 : recoveryStopName,
2798 : timestamptz_to_str(recoveryStopTime))));
2799 4 : return true;
2800 : }
2801 : }
2802 :
2803 : /* Check if the target LSN has been reached */
2804 5049618 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
2805 16060 : recoveryTargetInclusive &&
2806 16060 : record->ReadRecPtr >= recoveryTargetLSN)
2807 : {
2808 6 : recoveryStopAfter = true;
2809 6 : recoveryStopXid = InvalidTransactionId;
2810 6 : recoveryStopLSN = record->ReadRecPtr;
2811 6 : recoveryStopTime = 0;
2812 6 : recoveryStopName[0] = '\0';
2813 6 : ereport(LOG,
2814 : errmsg("recovery stopping after WAL location (LSN) \"%X/%08X\"",
2815 : LSN_FORMAT_ARGS(recoveryStopLSN)));
2816 6 : return true;
2817 : }
2818 :
2819 5049612 : if (rmid != RM_XACT_ID)
2820 5004584 : return false;
2821 :
2822 45028 : xact_info = info & XLOG_XACT_OPMASK;
2823 :
2824 45028 : if (xact_info == XLOG_XACT_COMMIT ||
2825 4338 : xact_info == XLOG_XACT_COMMIT_PREPARED ||
2826 642 : xact_info == XLOG_XACT_ABORT ||
2827 : xact_info == XLOG_XACT_ABORT_PREPARED)
2828 : {
2829 : TransactionId recordXid;
2830 :
2831 : /* Update the last applied transaction timestamp */
2832 44412 : if (getRecordTimestamp(record, &recordXtime))
2833 44412 : SetLatestXTime(recordXtime);
2834 :
2835 : /* Extract the XID of the committed/aborted transaction */
2836 44412 : if (xact_info == XLOG_XACT_COMMIT_PREPARED)
2837 : {
2838 60 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
2839 : xl_xact_parsed_commit parsed;
2840 :
2841 60 : ParseCommitRecord(XLogRecGetInfo(record),
2842 : xlrec,
2843 : &parsed);
2844 60 : recordXid = parsed.twophase_xid;
2845 : }
2846 44352 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
2847 : {
2848 26 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
2849 : xl_xact_parsed_abort parsed;
2850 :
2851 26 : ParseAbortRecord(XLogRecGetInfo(record),
2852 : xlrec,
2853 : &parsed);
2854 26 : recordXid = parsed.twophase_xid;
2855 : }
2856 : else
2857 44326 : recordXid = XLogRecGetXid(record);
2858 :
2859 : /*
2860 : * There can be only one transaction end record with this exact
2861 : * transactionid
2862 : *
2863 : * when testing for an xid, we MUST test for equality only, since
2864 : * transactions are numbered in the order they start, not the order
2865 : * they complete. A higher numbered xid will complete before you about
2866 : * 50% of the time...
2867 : */
2868 44412 : if (recoveryTarget == RECOVERY_TARGET_XID && recoveryTargetInclusive &&
2869 0 : recordXid == recoveryTargetXid)
2870 : {
2871 0 : recoveryStopAfter = true;
2872 0 : recoveryStopXid = recordXid;
2873 0 : recoveryStopTime = recordXtime;
2874 0 : recoveryStopLSN = InvalidXLogRecPtr;
2875 0 : recoveryStopName[0] = '\0';
2876 :
2877 0 : if (xact_info == XLOG_XACT_COMMIT ||
2878 : xact_info == XLOG_XACT_COMMIT_PREPARED)
2879 : {
2880 0 : ereport(LOG,
2881 : (errmsg("recovery stopping after commit of transaction %u, time %s",
2882 : recoveryStopXid,
2883 : timestamptz_to_str(recoveryStopTime))));
2884 : }
2885 0 : else if (xact_info == XLOG_XACT_ABORT ||
2886 : xact_info == XLOG_XACT_ABORT_PREPARED)
2887 : {
2888 0 : ereport(LOG,
2889 : (errmsg("recovery stopping after abort of transaction %u, time %s",
2890 : recoveryStopXid,
2891 : timestamptz_to_str(recoveryStopTime))));
2892 : }
2893 0 : return true;
2894 : }
2895 : }
2896 :
2897 : /* Check if we should stop as soon as reaching consistency */
2898 45028 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
2899 : {
2900 0 : ereport(LOG,
2901 : (errmsg("recovery stopping after reaching consistency")));
2902 :
2903 0 : recoveryStopAfter = true;
2904 0 : recoveryStopXid = InvalidTransactionId;
2905 0 : recoveryStopTime = 0;
2906 0 : recoveryStopLSN = InvalidXLogRecPtr;
2907 0 : recoveryStopName[0] = '\0';
2908 0 : return true;
2909 : }
2910 :
2911 45028 : return false;
2912 : }
2913 :
2914 : /*
2915 : * Create a comment for the history file to explain why and where
2916 : * timeline changed.
2917 : */
2918 : static char *
2919 1854 : getRecoveryStopReason(void)
2920 : {
2921 : char reason[200];
2922 :
2923 1854 : if (recoveryTarget == RECOVERY_TARGET_XID)
2924 0 : snprintf(reason, sizeof(reason),
2925 : "%s transaction %u",
2926 0 : recoveryStopAfter ? "after" : "before",
2927 : recoveryStopXid);
2928 1854 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
2929 0 : snprintf(reason, sizeof(reason),
2930 : "%s %s\n",
2931 0 : recoveryStopAfter ? "after" : "before",
2932 : timestamptz_to_str(recoveryStopTime));
2933 1854 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
2934 12 : snprintf(reason, sizeof(reason),
2935 : "%s LSN %X/%08X\n",
2936 12 : recoveryStopAfter ? "after" : "before",
2937 12 : LSN_FORMAT_ARGS(recoveryStopLSN));
2938 1842 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
2939 6 : snprintf(reason, sizeof(reason),
2940 : "at restore point \"%s\"",
2941 : recoveryStopName);
2942 1836 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
2943 0 : snprintf(reason, sizeof(reason), "reached consistency");
2944 : else
2945 1836 : snprintf(reason, sizeof(reason), "no recovery target specified");
2946 :
2947 1854 : return pstrdup(reason);
2948 : }
2949 :
2950 : /*
2951 : * Wait until shared recoveryPauseState is set to RECOVERY_NOT_PAUSED.
2952 : *
2953 : * endOfRecovery is true if the recovery target is reached and
2954 : * the paused state starts at the end of recovery because of
2955 : * recovery_target_action=pause, and false otherwise.
2956 : */
2957 : static void
2958 8 : recoveryPausesHere(bool endOfRecovery)
2959 : {
2960 : /* Don't pause unless users can connect! */
2961 8 : if (!LocalHotStandbyActive)
2962 0 : return;
2963 :
2964 : /* Don't pause after standby promotion has been triggered */
2965 8 : if (LocalPromoteIsTriggered)
2966 0 : return;
2967 :
2968 8 : if (endOfRecovery)
2969 2 : ereport(LOG,
2970 : (errmsg("pausing at the end of recovery"),
2971 : errhint("Execute pg_wal_replay_resume() to promote.")));
2972 : else
2973 6 : ereport(LOG,
2974 : (errmsg("recovery has paused"),
2975 : errhint("Execute pg_wal_replay_resume() to continue.")));
2976 :
2977 : /* loop until recoveryPauseState is set to RECOVERY_NOT_PAUSED */
2978 26 : while (GetRecoveryPauseState() != RECOVERY_NOT_PAUSED)
2979 : {
2980 22 : ProcessStartupProcInterrupts();
2981 22 : if (CheckForStandbyTrigger())
2982 4 : return;
2983 :
2984 : /*
2985 : * If recovery pause is requested then set it paused. While we are in
2986 : * the loop, user might resume and pause again so set this every time.
2987 : */
2988 18 : ConfirmRecoveryPaused();
2989 :
2990 : /*
2991 : * We wait on a condition variable that will wake us as soon as the
2992 : * pause ends, but we use a timeout so we can check the above exit
2993 : * condition periodically too.
2994 : */
2995 18 : ConditionVariableTimedSleep(&XLogRecoveryCtl->recoveryNotPausedCV, 1000,
2996 : WAIT_EVENT_RECOVERY_PAUSE);
2997 : }
2998 4 : ConditionVariableCancelSleep();
2999 : }
3000 :
3001 : /*
3002 : * When recovery_min_apply_delay is set, we wait long enough to make sure
3003 : * certain record types are applied at least that interval behind the primary.
3004 : *
3005 : * Returns true if we waited.
3006 : *
3007 : * Note that the delay is calculated between the WAL record log time and
3008 : * the current time on standby. We would prefer to keep track of when this
3009 : * standby received each WAL record, which would allow a more consistent
3010 : * approach and one not affected by time synchronisation issues, but that
3011 : * is significantly more effort and complexity for little actual gain in
3012 : * usability.
3013 : */
3014 : static bool
3015 5537708 : recoveryApplyDelay(XLogReaderState *record)
3016 : {
3017 : uint8 xact_info;
3018 : TimestampTz xtime;
3019 : TimestampTz delayUntil;
3020 : long msecs;
3021 :
3022 : /* nothing to do if no delay configured */
3023 5537708 : if (recovery_min_apply_delay <= 0)
3024 5537424 : return false;
3025 :
3026 : /* no delay is applied on a database not yet consistent */
3027 284 : if (!reachedConsistency)
3028 8 : return false;
3029 :
3030 : /* nothing to do if crash recovery is requested */
3031 276 : if (!ArchiveRecoveryRequested)
3032 0 : return false;
3033 :
3034 : /*
3035 : * Is it a COMMIT record?
3036 : *
3037 : * We deliberately choose not to delay aborts since they have no effect on
3038 : * MVCC. We already allow replay of records that don't have a timestamp,
3039 : * so there is already opportunity for issues caused by early conflicts on
3040 : * standbys.
3041 : */
3042 276 : if (XLogRecGetRmid(record) != RM_XACT_ID)
3043 218 : return false;
3044 :
3045 58 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
3046 :
3047 58 : if (xact_info != XLOG_XACT_COMMIT &&
3048 : xact_info != XLOG_XACT_COMMIT_PREPARED)
3049 0 : return false;
3050 :
3051 58 : if (!getRecordTimestamp(record, &xtime))
3052 0 : return false;
3053 :
3054 58 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
3055 :
3056 : /*
3057 : * Exit without arming the latch if it's already past time to apply this
3058 : * record
3059 : */
3060 58 : msecs = TimestampDifferenceMilliseconds(GetCurrentTimestamp(), delayUntil);
3061 58 : if (msecs <= 0)
3062 2 : return false;
3063 :
3064 : while (true)
3065 : {
3066 286 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3067 :
3068 : /* This might change recovery_min_apply_delay. */
3069 286 : ProcessStartupProcInterrupts();
3070 :
3071 286 : if (CheckForStandbyTrigger())
3072 0 : break;
3073 :
3074 : /*
3075 : * Recalculate delayUntil as recovery_min_apply_delay could have
3076 : * changed while waiting in this loop.
3077 : */
3078 286 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
3079 :
3080 : /*
3081 : * Wait for difference between GetCurrentTimestamp() and delayUntil.
3082 : */
3083 286 : msecs = TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
3084 : delayUntil);
3085 :
3086 286 : if (msecs <= 0)
3087 56 : break;
3088 :
3089 230 : elog(DEBUG2, "recovery apply delay %ld milliseconds", msecs);
3090 :
3091 230 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3092 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
3093 : msecs,
3094 : WAIT_EVENT_RECOVERY_APPLY_DELAY);
3095 : }
3096 56 : return true;
3097 : }
3098 :
3099 : /*
3100 : * Get the current state of the recovery pause request.
3101 : */
3102 : RecoveryPauseState
3103 58 : GetRecoveryPauseState(void)
3104 : {
3105 : RecoveryPauseState state;
3106 :
3107 58 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3108 58 : state = XLogRecoveryCtl->recoveryPauseState;
3109 58 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3110 :
3111 58 : return state;
3112 : }
3113 :
3114 : /*
3115 : * Set the recovery pause state.
3116 : *
3117 : * If recovery pause is requested then sets the recovery pause state to
3118 : * 'pause requested' if it is not already 'paused'. Otherwise, sets it
3119 : * to 'not paused' to resume the recovery. The recovery pause will be
3120 : * confirmed by the ConfirmRecoveryPaused.
3121 : */
3122 : void
3123 110 : SetRecoveryPause(bool recoveryPause)
3124 : {
3125 110 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3126 :
3127 110 : if (!recoveryPause)
3128 100 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_NOT_PAUSED;
3129 10 : else if (XLogRecoveryCtl->recoveryPauseState == RECOVERY_NOT_PAUSED)
3130 10 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_PAUSE_REQUESTED;
3131 :
3132 110 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3133 :
3134 110 : if (!recoveryPause)
3135 100 : ConditionVariableBroadcast(&XLogRecoveryCtl->recoveryNotPausedCV);
3136 110 : }
3137 :
3138 : /*
3139 : * Confirm the recovery pause by setting the recovery pause state to
3140 : * RECOVERY_PAUSED.
3141 : */
3142 : static void
3143 18 : ConfirmRecoveryPaused(void)
3144 : {
3145 : /* If recovery pause is requested then set it paused */
3146 18 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3147 18 : if (XLogRecoveryCtl->recoveryPauseState == RECOVERY_PAUSE_REQUESTED)
3148 8 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_PAUSED;
3149 18 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3150 18 : }
3151 :
3152 :
3153 : /*
3154 : * Attempt to read the next XLOG record.
3155 : *
3156 : * Before first call, the reader needs to be positioned to the first record
3157 : * by calling XLogPrefetcherBeginRead().
3158 : *
3159 : * If no valid record is available, returns NULL, or fails if emode is PANIC.
3160 : * (emode must be either PANIC, LOG). In standby mode, retries until a valid
3161 : * record is available.
3162 : */
3163 : static XLogRecord *
3164 5542224 : ReadRecord(XLogPrefetcher *xlogprefetcher, int emode,
3165 : bool fetching_ckpt, TimeLineID replayTLI)
3166 : {
3167 : XLogRecord *record;
3168 5542224 : XLogReaderState *xlogreader = XLogPrefetcherGetReader(xlogprefetcher);
3169 5542224 : XLogPageReadPrivate *private = (XLogPageReadPrivate *) xlogreader->private_data;
3170 :
3171 : Assert(AmStartupProcess() || !IsUnderPostmaster);
3172 :
3173 : /* Pass through parameters to XLogPageRead */
3174 5542224 : private->fetching_ckpt = fetching_ckpt;
3175 5542224 : private->emode = emode;
3176 5542224 : private->randAccess = !XLogRecPtrIsValid(xlogreader->ReadRecPtr);
3177 5542224 : private->replayTLI = replayTLI;
3178 :
3179 : /* This is the first attempt to read this page. */
3180 5542224 : lastSourceFailed = false;
3181 :
3182 : for (;;)
3183 290 : {
3184 : char *errormsg;
3185 :
3186 5542514 : record = XLogPrefetcherReadRecord(xlogprefetcher, &errormsg);
3187 5542396 : if (record == NULL)
3188 : {
3189 : /*
3190 : * When we find that WAL ends in an incomplete record, keep track
3191 : * of that record. After recovery is done, we'll write a record
3192 : * to indicate to downstream WAL readers that that portion is to
3193 : * be ignored.
3194 : *
3195 : * However, when ArchiveRecoveryRequested = true, we're going to
3196 : * switch to a new timeline at the end of recovery. We will only
3197 : * copy WAL over to the new timeline up to the end of the last
3198 : * complete record, so if we did this, we would later create an
3199 : * overwrite contrecord in the wrong place, breaking everything.
3200 : */
3201 602 : if (!ArchiveRecoveryRequested &&
3202 216 : XLogRecPtrIsValid(xlogreader->abortedRecPtr))
3203 : {
3204 22 : abortedRecPtr = xlogreader->abortedRecPtr;
3205 22 : missingContrecPtr = xlogreader->missingContrecPtr;
3206 : }
3207 :
3208 602 : if (readFile >= 0)
3209 : {
3210 554 : close(readFile);
3211 554 : readFile = -1;
3212 : }
3213 :
3214 : /*
3215 : * We only end up here without a message when XLogPageRead()
3216 : * failed - in that case we already logged something. In
3217 : * StandbyMode that only happens if we have been triggered, so we
3218 : * shouldn't loop anymore in that case.
3219 : */
3220 602 : if (errormsg)
3221 554 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3222 : (errmsg_internal("%s", errormsg) /* already translated */ ));
3223 : }
3224 :
3225 : /*
3226 : * Check page TLI is one of the expected values.
3227 : */
3228 5541794 : else if (!tliInHistory(xlogreader->latestPageTLI, expectedTLEs))
3229 : {
3230 : char fname[MAXFNAMELEN];
3231 : XLogSegNo segno;
3232 : int32 offset;
3233 :
3234 0 : XLByteToSeg(xlogreader->latestPagePtr, segno, wal_segment_size);
3235 0 : offset = XLogSegmentOffset(xlogreader->latestPagePtr,
3236 : wal_segment_size);
3237 0 : XLogFileName(fname, xlogreader->seg.ws_tli, segno,
3238 : wal_segment_size);
3239 0 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3240 : errmsg("unexpected timeline ID %u in WAL segment %s, LSN %X/%08X, offset %u",
3241 : xlogreader->latestPageTLI,
3242 : fname,
3243 : LSN_FORMAT_ARGS(xlogreader->latestPagePtr),
3244 : offset));
3245 0 : record = NULL;
3246 : }
3247 :
3248 5542396 : if (record)
3249 : {
3250 : /* Great, got a record */
3251 5542106 : return record;
3252 : }
3253 : else
3254 : {
3255 : /* No valid record available from this source */
3256 602 : lastSourceFailed = true;
3257 :
3258 : /*
3259 : * If archive recovery was requested, but we were still doing
3260 : * crash recovery, switch to archive recovery and retry using the
3261 : * offline archive. We have now replayed all the valid WAL in
3262 : * pg_wal, so we are presumably now consistent.
3263 : *
3264 : * We require that there's at least some valid WAL present in
3265 : * pg_wal, however (!fetching_ckpt). We could recover using the
3266 : * WAL from the archive, even if pg_wal is completely empty, but
3267 : * we'd have no idea how far we'd have to replay to reach
3268 : * consistency. So err on the safe side and give up.
3269 : */
3270 602 : if (!InArchiveRecovery && ArchiveRecoveryRequested &&
3271 4 : !fetching_ckpt)
3272 : {
3273 4 : ereport(DEBUG1,
3274 : (errmsg_internal("reached end of WAL in pg_wal, entering archive recovery")));
3275 4 : InArchiveRecovery = true;
3276 4 : if (StandbyModeRequested)
3277 4 : EnableStandbyMode();
3278 :
3279 4 : SwitchIntoArchiveRecovery(xlogreader->EndRecPtr, replayTLI);
3280 4 : minRecoveryPoint = xlogreader->EndRecPtr;
3281 4 : minRecoveryPointTLI = replayTLI;
3282 :
3283 4 : CheckRecoveryConsistency();
3284 :
3285 : /*
3286 : * Before we retry, reset lastSourceFailed and currentSource
3287 : * so that we will check the archive next.
3288 : */
3289 4 : lastSourceFailed = false;
3290 4 : currentSource = XLOG_FROM_ANY;
3291 :
3292 290 : continue;
3293 : }
3294 :
3295 : /* In standby mode, loop back to retry. Otherwise, give up. */
3296 598 : if (StandbyMode && !CheckForStandbyTrigger())
3297 286 : continue;
3298 : else
3299 312 : return NULL;
3300 : }
3301 : }
3302 : }
3303 :
3304 : /*
3305 : * Read the XLOG page containing targetPagePtr into readBuf (if not read
3306 : * already). Returns number of bytes read, if the page is read successfully,
3307 : * or XLREAD_FAIL in case of errors. When errors occur, they are ereport'ed,
3308 : * but only if they have not been previously reported.
3309 : *
3310 : * See XLogReaderRoutine.page_read for more details.
3311 : *
3312 : * While prefetching, xlogreader->nonblocking may be set. In that case,
3313 : * returns XLREAD_WOULDBLOCK if we'd otherwise have to wait for more WAL.
3314 : *
3315 : * This is responsible for restoring files from archive as needed, as well
3316 : * as for waiting for the requested WAL record to arrive in standby mode.
3317 : *
3318 : * xlogreader->private_data->emode specifies the log level used for reporting
3319 : * "file not found" or "end of WAL" situations in archive recovery, or in
3320 : * standby mode when promotion is triggered. If set to WARNING or below,
3321 : * XLogPageRead() returns XLREAD_FAIL in those situations, on higher log
3322 : * levels the ereport() won't return.
3323 : *
3324 : * In standby mode, if after a successful return of XLogPageRead() the
3325 : * caller finds the record it's interested in to be broken, it should
3326 : * ereport the error with the level determined by
3327 : * emode_for_corrupt_record(), and then set lastSourceFailed
3328 : * and call XLogPageRead() again with the same arguments. This lets
3329 : * XLogPageRead() to try fetching the record from another source, or to
3330 : * sleep and retry.
3331 : */
3332 : static int
3333 2887820 : XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr, int reqLen,
3334 : XLogRecPtr targetRecPtr, char *readBuf)
3335 : {
3336 2887820 : XLogPageReadPrivate *private =
3337 : (XLogPageReadPrivate *) xlogreader->private_data;
3338 2887820 : int emode = private->emode;
3339 : uint32 targetPageOff;
3340 : XLogSegNo targetSegNo PG_USED_FOR_ASSERTS_ONLY;
3341 : int r;
3342 : instr_time io_start;
3343 :
3344 : Assert(AmStartupProcess() || !IsUnderPostmaster);
3345 :
3346 2887820 : XLByteToSeg(targetPagePtr, targetSegNo, wal_segment_size);
3347 2887820 : targetPageOff = XLogSegmentOffset(targetPagePtr, wal_segment_size);
3348 :
3349 : /*
3350 : * See if we need to switch to a new segment because the requested record
3351 : * is not in the currently open one.
3352 : */
3353 2887820 : if (readFile >= 0 &&
3354 2884236 : !XLByteInSeg(targetPagePtr, readSegNo, wal_segment_size))
3355 : {
3356 : /*
3357 : * Request a restartpoint if we've replayed too much xlog since the
3358 : * last one.
3359 : */
3360 3046 : if (ArchiveRecoveryRequested && IsUnderPostmaster)
3361 : {
3362 3014 : if (XLogCheckpointNeeded(readSegNo))
3363 : {
3364 2784 : (void) GetRedoRecPtr();
3365 2784 : if (XLogCheckpointNeeded(readSegNo))
3366 2772 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
3367 : }
3368 : }
3369 :
3370 3046 : close(readFile);
3371 3046 : readFile = -1;
3372 3046 : readSource = XLOG_FROM_ANY;
3373 : }
3374 :
3375 2887820 : XLByteToSeg(targetPagePtr, readSegNo, wal_segment_size);
3376 :
3377 2887830 : retry:
3378 : /* See if we need to retrieve more data */
3379 2887830 : if (readFile < 0 ||
3380 2881190 : (readSource == XLOG_FROM_STREAM &&
3381 2857126 : flushedUpto < targetPagePtr + reqLen))
3382 : {
3383 29686 : if (readFile >= 0 &&
3384 23046 : xlogreader->nonblocking &&
3385 11354 : readSource == XLOG_FROM_STREAM &&
3386 11354 : flushedUpto < targetPagePtr + reqLen)
3387 11354 : return XLREAD_WOULDBLOCK;
3388 :
3389 18214 : switch (WaitForWALToBecomeAvailable(targetPagePtr + reqLen,
3390 18332 : private->randAccess,
3391 18332 : private->fetching_ckpt,
3392 : targetRecPtr,
3393 : private->replayTLI,
3394 : xlogreader->EndRecPtr,
3395 18332 : xlogreader->nonblocking))
3396 : {
3397 958 : case XLREAD_WOULDBLOCK:
3398 958 : return XLREAD_WOULDBLOCK;
3399 92 : case XLREAD_FAIL:
3400 92 : if (readFile >= 0)
3401 0 : close(readFile);
3402 92 : readFile = -1;
3403 92 : readLen = 0;
3404 92 : readSource = XLOG_FROM_ANY;
3405 92 : return XLREAD_FAIL;
3406 17164 : case XLREAD_SUCCESS:
3407 17164 : break;
3408 : }
3409 : }
3410 :
3411 : /*
3412 : * At this point, we have the right segment open and if we're streaming we
3413 : * know the requested record is in it.
3414 : */
3415 : Assert(readFile != -1);
3416 :
3417 : /*
3418 : * If the current segment is being streamed from the primary, calculate
3419 : * how much of the current page we have received already. We know the
3420 : * requested record has been received, but this is for the benefit of
3421 : * future calls, to allow quick exit at the top of this function.
3422 : */
3423 2875308 : if (readSource == XLOG_FROM_STREAM)
3424 : {
3425 2847962 : if (((targetPagePtr) / XLOG_BLCKSZ) != (flushedUpto / XLOG_BLCKSZ))
3426 2840474 : readLen = XLOG_BLCKSZ;
3427 : else
3428 7488 : readLen = XLogSegmentOffset(flushedUpto, wal_segment_size) -
3429 : targetPageOff;
3430 : }
3431 : else
3432 27346 : readLen = XLOG_BLCKSZ;
3433 :
3434 : /* Read the requested page */
3435 2875308 : readOff = targetPageOff;
3436 :
3437 : /* Measure I/O timing when reading segment */
3438 2875308 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3439 :
3440 2875308 : pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
3441 2875308 : r = pg_pread(readFile, readBuf, XLOG_BLCKSZ, (pgoff_t) readOff);
3442 2875308 : if (r != XLOG_BLCKSZ)
3443 : {
3444 : char fname[MAXFNAMELEN];
3445 0 : int save_errno = errno;
3446 :
3447 0 : pgstat_report_wait_end();
3448 :
3449 0 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_READ,
3450 : io_start, 1, r);
3451 :
3452 0 : XLogFileName(fname, curFileTLI, readSegNo, wal_segment_size);
3453 0 : if (r < 0)
3454 : {
3455 0 : errno = save_errno;
3456 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
3457 : (errcode_for_file_access(),
3458 : errmsg("could not read from WAL segment %s, LSN %X/%08X, offset %u: %m",
3459 : fname, LSN_FORMAT_ARGS(targetPagePtr),
3460 : readOff)));
3461 : }
3462 : else
3463 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
3464 : (errcode(ERRCODE_DATA_CORRUPTED),
3465 : errmsg("could not read from WAL segment %s, LSN %X/%08X, offset %u: read %d of %zu",
3466 : fname, LSN_FORMAT_ARGS(targetPagePtr),
3467 : readOff, r, (Size) XLOG_BLCKSZ)));
3468 0 : goto next_record_is_invalid;
3469 : }
3470 2875308 : pgstat_report_wait_end();
3471 :
3472 2875308 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_READ,
3473 : io_start, 1, r);
3474 :
3475 : Assert(targetSegNo == readSegNo);
3476 : Assert(targetPageOff == readOff);
3477 : Assert(reqLen <= readLen);
3478 :
3479 2875308 : xlogreader->seg.ws_tli = curFileTLI;
3480 :
3481 : /*
3482 : * Check the page header immediately, so that we can retry immediately if
3483 : * it's not valid. This may seem unnecessary, because ReadPageInternal()
3484 : * validates the page header anyway, and would propagate the failure up to
3485 : * ReadRecord(), which would retry. However, there's a corner case with
3486 : * continuation records, if a record is split across two pages such that
3487 : * we would need to read the two pages from different sources across two
3488 : * WAL segments.
3489 : *
3490 : * The first page is only available locally, in pg_wal, because it's
3491 : * already been recycled on the primary. The second page, however, is not
3492 : * present in pg_wal, and we should stream it from the primary. There is a
3493 : * recycled WAL segment present in pg_wal, with garbage contents, however.
3494 : * We would read the first page from the local WAL segment, but when
3495 : * reading the second page, we would read the bogus, recycled, WAL
3496 : * segment. If we didn't catch that case here, we would never recover,
3497 : * because ReadRecord() would retry reading the whole record from the
3498 : * beginning.
3499 : *
3500 : * Of course, this only catches errors in the page header, which is what
3501 : * happens in the case of a recycled WAL segment. Other kinds of errors or
3502 : * corruption still has the same problem. But this at least fixes the
3503 : * common case, which can happen as part of normal operation.
3504 : *
3505 : * Validating the page header is cheap enough that doing it twice
3506 : * shouldn't be a big deal from a performance point of view.
3507 : *
3508 : * When not in standby mode, an invalid page header should cause recovery
3509 : * to end, not retry reading the page, so we don't need to validate the
3510 : * page header here for the retry. Instead, ReadPageInternal() is
3511 : * responsible for the validation.
3512 : */
3513 2875308 : if (StandbyMode &&
3514 2855426 : (targetPagePtr % wal_segment_size) == 0 &&
3515 2816 : !XLogReaderValidatePageHeader(xlogreader, targetPagePtr, readBuf))
3516 : {
3517 : /*
3518 : * Emit this error right now then retry this page immediately. Use
3519 : * errmsg_internal() because the message was already translated.
3520 : */
3521 12 : if (xlogreader->errormsg_buf[0])
3522 12 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3523 : (errmsg_internal("%s", xlogreader->errormsg_buf)));
3524 :
3525 : /* reset any error XLogReaderValidatePageHeader() might have set */
3526 12 : XLogReaderResetError(xlogreader);
3527 12 : goto next_record_is_invalid;
3528 : }
3529 :
3530 2875296 : return readLen;
3531 :
3532 12 : next_record_is_invalid:
3533 :
3534 : /*
3535 : * If we're reading ahead, give up fast. Retries and error reporting will
3536 : * be handled by a later read when recovery catches up to this point.
3537 : */
3538 12 : if (xlogreader->nonblocking)
3539 2 : return XLREAD_WOULDBLOCK;
3540 :
3541 10 : lastSourceFailed = true;
3542 :
3543 10 : if (readFile >= 0)
3544 10 : close(readFile);
3545 10 : readFile = -1;
3546 10 : readLen = 0;
3547 10 : readSource = XLOG_FROM_ANY;
3548 :
3549 : /* In standby-mode, keep trying */
3550 10 : if (StandbyMode)
3551 10 : goto retry;
3552 : else
3553 0 : return XLREAD_FAIL;
3554 : }
3555 :
3556 : /*
3557 : * Open the WAL segment containing WAL location 'RecPtr'.
3558 : *
3559 : * The segment can be fetched via restore_command, or via walreceiver having
3560 : * streamed the record, or it can already be present in pg_wal. Checking
3561 : * pg_wal is mainly for crash recovery, but it will be polled in standby mode
3562 : * too, in case someone copies a new segment directly to pg_wal. That is not
3563 : * documented or recommended, though.
3564 : *
3565 : * If 'fetching_ckpt' is true, we're fetching a checkpoint record, and should
3566 : * prepare to read WAL starting from RedoStartLSN after this.
3567 : *
3568 : * 'RecPtr' might not point to the beginning of the record we're interested
3569 : * in, it might also point to the page or segment header. In that case,
3570 : * 'tliRecPtr' is the position of the WAL record we're interested in. It is
3571 : * used to decide which timeline to stream the requested WAL from.
3572 : *
3573 : * 'replayLSN' is the current replay LSN, so that if we scan for new
3574 : * timelines, we can reject a switch to a timeline that branched off before
3575 : * this point.
3576 : *
3577 : * If the record is not immediately available, the function returns false
3578 : * if we're not in standby mode. In standby mode, waits for it to become
3579 : * available.
3580 : *
3581 : * When the requested record becomes available, the function opens the file
3582 : * containing it (if not open already), and returns XLREAD_SUCCESS. When end
3583 : * of standby mode is triggered by the user, and there is no more WAL
3584 : * available, returns XLREAD_FAIL.
3585 : *
3586 : * If nonblocking is true, then give up immediately if we can't satisfy the
3587 : * request, returning XLREAD_WOULDBLOCK instead of waiting.
3588 : */
3589 : static XLogPageReadResult
3590 18332 : WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
3591 : bool fetching_ckpt, XLogRecPtr tliRecPtr,
3592 : TimeLineID replayTLI, XLogRecPtr replayLSN,
3593 : bool nonblocking)
3594 : {
3595 : static TimestampTz last_fail_time = 0;
3596 : TimestampTz now;
3597 18332 : bool streaming_reply_sent = false;
3598 :
3599 : /*-------
3600 : * Standby mode is implemented by a state machine:
3601 : *
3602 : * 1. Read from either archive or pg_wal (XLOG_FROM_ARCHIVE), or just
3603 : * pg_wal (XLOG_FROM_PG_WAL)
3604 : * 2. Check for promotion trigger request
3605 : * 3. Read from primary server via walreceiver (XLOG_FROM_STREAM)
3606 : * 4. Rescan timelines
3607 : * 5. Sleep wal_retrieve_retry_interval milliseconds, and loop back to 1.
3608 : *
3609 : * Failure to read from the current source advances the state machine to
3610 : * the next state.
3611 : *
3612 : * 'currentSource' indicates the current state. There are no currentSource
3613 : * values for "check trigger", "rescan timelines", and "sleep" states,
3614 : * those actions are taken when reading from the previous source fails, as
3615 : * part of advancing to the next state.
3616 : *
3617 : * If standby mode is turned off while reading WAL from stream, we move
3618 : * to XLOG_FROM_ARCHIVE and reset lastSourceFailed, to force fetching
3619 : * the files (which would be required at end of recovery, e.g., timeline
3620 : * history file) from archive or pg_wal. We don't need to kill WAL receiver
3621 : * here because it's already stopped when standby mode is turned off at
3622 : * the end of recovery.
3623 : *-------
3624 : */
3625 18332 : if (!InArchiveRecovery)
3626 1982 : currentSource = XLOG_FROM_PG_WAL;
3627 16350 : else if (currentSource == XLOG_FROM_ANY ||
3628 16104 : (!StandbyMode && currentSource == XLOG_FROM_STREAM))
3629 : {
3630 246 : lastSourceFailed = false;
3631 246 : currentSource = XLOG_FROM_ARCHIVE;
3632 : }
3633 :
3634 : for (;;)
3635 14984 : {
3636 33316 : XLogSource oldSource = currentSource;
3637 33316 : bool startWalReceiver = false;
3638 :
3639 : /*
3640 : * First check if we failed to read from the current source, and
3641 : * advance the state machine if so. The failure to read might've
3642 : * happened outside this function, e.g when a CRC check fails on a
3643 : * record, or within this loop.
3644 : */
3645 33316 : if (lastSourceFailed)
3646 : {
3647 : /*
3648 : * Don't allow any retry loops to occur during nonblocking
3649 : * readahead. Let the caller process everything that has been
3650 : * decoded already first.
3651 : */
3652 1126 : if (nonblocking)
3653 156 : return XLREAD_WOULDBLOCK;
3654 :
3655 970 : switch (currentSource)
3656 : {
3657 576 : case XLOG_FROM_ARCHIVE:
3658 : case XLOG_FROM_PG_WAL:
3659 :
3660 : /*
3661 : * Check to see if promotion is requested. Note that we do
3662 : * this only after failure, so when you promote, we still
3663 : * finish replaying as much as we can from archive and
3664 : * pg_wal before failover.
3665 : */
3666 576 : if (StandbyMode && CheckForStandbyTrigger())
3667 : {
3668 44 : XLogShutdownWalRcv();
3669 44 : return XLREAD_FAIL;
3670 : }
3671 :
3672 : /*
3673 : * Not in standby mode, and we've now tried the archive
3674 : * and pg_wal.
3675 : */
3676 532 : if (!StandbyMode)
3677 48 : return XLREAD_FAIL;
3678 :
3679 : /*
3680 : * Move to XLOG_FROM_STREAM state, and set to start a
3681 : * walreceiver if necessary.
3682 : */
3683 484 : currentSource = XLOG_FROM_STREAM;
3684 484 : startWalReceiver = true;
3685 484 : break;
3686 :
3687 394 : case XLOG_FROM_STREAM:
3688 :
3689 : /*
3690 : * Failure while streaming. Most likely, we got here
3691 : * because streaming replication was terminated, or
3692 : * promotion was triggered. But we also get here if we
3693 : * find an invalid record in the WAL streamed from the
3694 : * primary, in which case something is seriously wrong.
3695 : * There's little chance that the problem will just go
3696 : * away, but PANIC is not good for availability either,
3697 : * especially in hot standby mode. So, we treat that the
3698 : * same as disconnection, and retry from archive/pg_wal
3699 : * again. The WAL in the archive should be identical to
3700 : * what was streamed, so it's unlikely that it helps, but
3701 : * one can hope...
3702 : */
3703 :
3704 : /*
3705 : * We should be able to move to XLOG_FROM_STREAM only in
3706 : * standby mode.
3707 : */
3708 : Assert(StandbyMode);
3709 :
3710 : /*
3711 : * Before we leave XLOG_FROM_STREAM state, make sure that
3712 : * walreceiver is not active, so that it won't overwrite
3713 : * WAL that we restore from archive.
3714 : *
3715 : * If walreceiver is actively streaming (or attempting to
3716 : * connect), we must shut it down. However, if it's
3717 : * already in WAITING state (e.g., due to timeline
3718 : * divergence), we only need to reset the install flag to
3719 : * allow archive restoration.
3720 : */
3721 394 : if (WalRcvStreaming())
3722 68 : XLogShutdownWalRcv();
3723 : else
3724 : {
3725 : /*
3726 : * WALRCV_STOPPING state is a transient state while
3727 : * the startup process is in ShutdownWalRcv(). It
3728 : * should never appear here since we would be waiting
3729 : * for the walreceiver to reach WALRCV_STOPPED in that
3730 : * case.
3731 : */
3732 : Assert(WalRcvGetState() != WALRCV_STOPPING);
3733 326 : ResetInstallXLogFileSegmentActive();
3734 : }
3735 :
3736 : /*
3737 : * Before we sleep, re-scan for possible new timelines if
3738 : * we were requested to recover to the latest timeline.
3739 : */
3740 394 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
3741 : {
3742 394 : if (rescanLatestTimeLine(replayTLI, replayLSN))
3743 : {
3744 14 : currentSource = XLOG_FROM_ARCHIVE;
3745 14 : break;
3746 : }
3747 : }
3748 :
3749 : /*
3750 : * XLOG_FROM_STREAM is the last state in our state
3751 : * machine, so we've exhausted all the options for
3752 : * obtaining the requested WAL. We're going to loop back
3753 : * and retry from the archive, but if it hasn't been long
3754 : * since last attempt, sleep wal_retrieve_retry_interval
3755 : * milliseconds to avoid busy-waiting.
3756 : */
3757 378 : now = GetCurrentTimestamp();
3758 378 : if (!TimestampDifferenceExceeds(last_fail_time, now,
3759 : wal_retrieve_retry_interval))
3760 : {
3761 : long wait_time;
3762 :
3763 444 : wait_time = wal_retrieve_retry_interval -
3764 222 : TimestampDifferenceMilliseconds(last_fail_time, now);
3765 :
3766 222 : elog(LOG, "waiting for WAL to become available at %X/%08X",
3767 : LSN_FORMAT_ARGS(RecPtr));
3768 :
3769 : /* Do background tasks that might benefit us later. */
3770 222 : KnownAssignedTransactionIdsIdleMaintenance();
3771 :
3772 222 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3773 : WL_LATCH_SET | WL_TIMEOUT |
3774 : WL_EXIT_ON_PM_DEATH,
3775 : wait_time,
3776 : WAIT_EVENT_RECOVERY_RETRIEVE_RETRY_INTERVAL);
3777 222 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3778 222 : now = GetCurrentTimestamp();
3779 :
3780 : /* Handle interrupt signals of startup process */
3781 222 : ProcessStartupProcInterrupts();
3782 : }
3783 350 : last_fail_time = now;
3784 350 : currentSource = XLOG_FROM_ARCHIVE;
3785 350 : break;
3786 :
3787 0 : default:
3788 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
3789 : }
3790 : }
3791 32190 : else if (currentSource == XLOG_FROM_PG_WAL)
3792 : {
3793 : /*
3794 : * We just successfully read a file in pg_wal. We prefer files in
3795 : * the archive over ones in pg_wal, so try the next file again
3796 : * from the archive first.
3797 : */
3798 1974 : if (InArchiveRecovery)
3799 0 : currentSource = XLOG_FROM_ARCHIVE;
3800 : }
3801 :
3802 33038 : if (currentSource != oldSource)
3803 848 : elog(DEBUG2, "switched WAL source from %s to %s after %s",
3804 : xlogSourceNames[oldSource], xlogSourceNames[currentSource],
3805 : lastSourceFailed ? "failure" : "success");
3806 :
3807 : /*
3808 : * We've now handled possible failure. Try to read from the chosen
3809 : * source.
3810 : */
3811 33038 : lastSourceFailed = false;
3812 :
3813 33038 : switch (currentSource)
3814 : {
3815 3676 : case XLOG_FROM_ARCHIVE:
3816 : case XLOG_FROM_PG_WAL:
3817 :
3818 : /*
3819 : * WAL receiver must not be running when reading WAL from
3820 : * archive or pg_wal.
3821 : */
3822 : Assert(!WalRcvStreaming());
3823 :
3824 : /* Close any old file we might have open. */
3825 3676 : if (readFile >= 0)
3826 : {
3827 174 : close(readFile);
3828 174 : readFile = -1;
3829 : }
3830 : /* Reset curFileTLI if random fetch. */
3831 3676 : if (randAccess)
3832 2300 : curFileTLI = 0;
3833 :
3834 : /*
3835 : * Try to restore the file from archive, or read an existing
3836 : * file from pg_wal.
3837 : */
3838 3676 : readFile = XLogFileReadAnyTLI(readSegNo,
3839 3676 : currentSource == XLOG_FROM_ARCHIVE ? XLOG_FROM_ANY :
3840 : currentSource);
3841 3674 : if (readFile >= 0)
3842 3282 : return XLREAD_SUCCESS; /* success! */
3843 :
3844 : /*
3845 : * Nope, not found in archive or pg_wal.
3846 : */
3847 392 : lastSourceFailed = true;
3848 392 : break;
3849 :
3850 29362 : case XLOG_FROM_STREAM:
3851 : {
3852 : bool havedata;
3853 :
3854 : /*
3855 : * We should be able to move to XLOG_FROM_STREAM only in
3856 : * standby mode.
3857 : */
3858 : Assert(StandbyMode);
3859 :
3860 : /*
3861 : * First, shutdown walreceiver if its restart has been
3862 : * requested -- but no point if we're already slated for
3863 : * starting it.
3864 : */
3865 29362 : if (pendingWalRcvRestart && !startWalReceiver)
3866 : {
3867 16 : XLogShutdownWalRcv();
3868 :
3869 : /*
3870 : * Re-scan for possible new timelines if we were
3871 : * requested to recover to the latest timeline.
3872 : */
3873 16 : if (recoveryTargetTimeLineGoal ==
3874 : RECOVERY_TARGET_TIMELINE_LATEST)
3875 16 : rescanLatestTimeLine(replayTLI, replayLSN);
3876 :
3877 16 : startWalReceiver = true;
3878 : }
3879 29362 : pendingWalRcvRestart = false;
3880 :
3881 : /*
3882 : * Launch walreceiver if needed.
3883 : *
3884 : * If fetching_ckpt is true, RecPtr points to the initial
3885 : * checkpoint location. In that case, we use RedoStartLSN
3886 : * as the streaming start position instead of RecPtr, so
3887 : * that when we later jump backwards to start redo at
3888 : * RedoStartLSN, we will have the logs streamed already.
3889 : */
3890 29362 : if (startWalReceiver &&
3891 500 : PrimaryConnInfo && strcmp(PrimaryConnInfo, "") != 0)
3892 : {
3893 : XLogRecPtr ptr;
3894 : TimeLineID tli;
3895 :
3896 402 : if (fetching_ckpt)
3897 : {
3898 0 : ptr = RedoStartLSN;
3899 0 : tli = RedoStartTLI;
3900 : }
3901 : else
3902 : {
3903 402 : ptr = RecPtr;
3904 :
3905 : /*
3906 : * Use the record begin position to determine the
3907 : * TLI, rather than the position we're reading.
3908 : */
3909 402 : tli = tliOfPointInHistory(tliRecPtr, expectedTLEs);
3910 :
3911 402 : if (curFileTLI > 0 && tli < curFileTLI)
3912 0 : elog(ERROR, "according to history file, WAL location %X/%08X belongs to timeline %u, but previous recovered WAL file came from timeline %u",
3913 : LSN_FORMAT_ARGS(tliRecPtr),
3914 : tli, curFileTLI);
3915 : }
3916 402 : curFileTLI = tli;
3917 402 : SetInstallXLogFileSegmentActive();
3918 402 : RequestXLogStreaming(tli, ptr, PrimaryConnInfo,
3919 : PrimarySlotName,
3920 : wal_receiver_create_temp_slot);
3921 402 : flushedUpto = 0;
3922 : }
3923 :
3924 : /*
3925 : * Check if WAL receiver is active or wait to start up.
3926 : */
3927 29362 : if (!WalRcvStreaming())
3928 : {
3929 326 : lastSourceFailed = true;
3930 326 : break;
3931 : }
3932 :
3933 : /*
3934 : * Walreceiver is active, so see if new data has arrived.
3935 : *
3936 : * We only advance XLogReceiptTime when we obtain fresh
3937 : * WAL from walreceiver and observe that we had already
3938 : * processed everything before the most recent "chunk"
3939 : * that it flushed to disk. In steady state where we are
3940 : * keeping up with the incoming data, XLogReceiptTime will
3941 : * be updated on each cycle. When we are behind,
3942 : * XLogReceiptTime will not advance, so the grace time
3943 : * allotted to conflicting queries will decrease.
3944 : */
3945 29036 : if (RecPtr < flushedUpto)
3946 3608 : havedata = true;
3947 : else
3948 : {
3949 : XLogRecPtr latestChunkStart;
3950 :
3951 25428 : flushedUpto = GetWalRcvFlushRecPtr(&latestChunkStart, &receiveTLI);
3952 25428 : if (RecPtr < flushedUpto && receiveTLI == curFileTLI)
3953 : {
3954 12702 : havedata = true;
3955 12702 : if (latestChunkStart <= RecPtr)
3956 : {
3957 8682 : XLogReceiptTime = GetCurrentTimestamp();
3958 8682 : SetCurrentChunkStartTime(XLogReceiptTime);
3959 : }
3960 : }
3961 : else
3962 12726 : havedata = false;
3963 : }
3964 29036 : if (havedata)
3965 : {
3966 : /*
3967 : * Great, streamed far enough. Open the file if it's
3968 : * not open already. Also read the timeline history
3969 : * file if we haven't initialized timeline history
3970 : * yet; it should be streamed over and present in
3971 : * pg_wal by now. Use XLOG_FROM_STREAM so that source
3972 : * info is set correctly and XLogReceiptTime isn't
3973 : * changed.
3974 : *
3975 : * NB: We must set readTimeLineHistory based on
3976 : * recoveryTargetTLI, not receiveTLI. Normally they'll
3977 : * be the same, but if recovery_target_timeline is
3978 : * 'latest' and archiving is configured, then it's
3979 : * possible that we managed to retrieve one or more
3980 : * new timeline history files from the archive,
3981 : * updating recoveryTargetTLI.
3982 : */
3983 16310 : if (readFile < 0)
3984 : {
3985 2428 : if (!expectedTLEs)
3986 0 : expectedTLEs = readTimeLineHistory(recoveryTargetTLI);
3987 2428 : readFile = XLogFileRead(readSegNo, receiveTLI,
3988 : XLOG_FROM_STREAM, false);
3989 : Assert(readFile >= 0);
3990 : }
3991 : else
3992 : {
3993 : /* just make sure source info is correct... */
3994 13882 : readSource = XLOG_FROM_STREAM;
3995 13882 : XLogReceiptSource = XLOG_FROM_STREAM;
3996 13882 : return XLREAD_SUCCESS;
3997 : }
3998 2428 : break;
3999 : }
4000 :
4001 : /* In nonblocking mode, return rather than sleeping. */
4002 12726 : if (nonblocking)
4003 802 : return XLREAD_WOULDBLOCK;
4004 :
4005 : /*
4006 : * Data not here yet. Check for trigger, then wait for
4007 : * walreceiver to wake us up when new WAL arrives.
4008 : */
4009 11924 : if (CheckForStandbyTrigger())
4010 : {
4011 : /*
4012 : * Note that we don't return XLREAD_FAIL immediately
4013 : * here. After being triggered, we still want to
4014 : * replay all the WAL that was already streamed. It's
4015 : * in pg_wal now, so we just treat this as a failure,
4016 : * and the state machine will move on to replay the
4017 : * streamed WAL from pg_wal, and then recheck the
4018 : * trigger and exit replay.
4019 : */
4020 68 : lastSourceFailed = true;
4021 68 : break;
4022 : }
4023 :
4024 : /*
4025 : * Since we have replayed everything we have received so
4026 : * far and are about to start waiting for more WAL, let's
4027 : * tell the upstream server our replay location now so
4028 : * that pg_stat_replication doesn't show stale
4029 : * information.
4030 : */
4031 11856 : if (!streaming_reply_sent)
4032 : {
4033 9006 : WalRcvForceReply();
4034 9006 : streaming_reply_sent = true;
4035 : }
4036 :
4037 : /* Do any background tasks that might benefit us later. */
4038 11856 : KnownAssignedTransactionIdsIdleMaintenance();
4039 :
4040 : /* Update pg_stat_recovery_prefetch before sleeping. */
4041 11856 : XLogPrefetcherComputeStats(xlogprefetcher);
4042 :
4043 : /*
4044 : * Wait for more WAL to arrive, when we will be woken
4045 : * immediately by the WAL receiver.
4046 : */
4047 11856 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
4048 : WL_LATCH_SET | WL_EXIT_ON_PM_DEATH,
4049 : -1L,
4050 : WAIT_EVENT_RECOVERY_WAL_STREAM);
4051 11856 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
4052 11856 : break;
4053 : }
4054 :
4055 0 : default:
4056 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
4057 : }
4058 :
4059 : /*
4060 : * Check for recovery pause here so that we can confirm more quickly
4061 : * that a requested pause has actually taken effect.
4062 : */
4063 15070 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
4064 : RECOVERY_NOT_PAUSED)
4065 6 : recoveryPausesHere(false);
4066 :
4067 : /*
4068 : * This possibly-long loop needs to handle interrupts of startup
4069 : * process.
4070 : */
4071 15070 : ProcessStartupProcInterrupts();
4072 : }
4073 :
4074 : return XLREAD_FAIL; /* not reached */
4075 : }
4076 :
4077 :
4078 : /*
4079 : * Determine what log level should be used to report a corrupt WAL record
4080 : * in the current WAL page, previously read by XLogPageRead().
4081 : *
4082 : * 'emode' is the error mode that would be used to report a file-not-found
4083 : * or legitimate end-of-WAL situation. Generally, we use it as-is, but if
4084 : * we're retrying the exact same record that we've tried previously, only
4085 : * complain the first time to keep the noise down. However, we only do when
4086 : * reading from pg_wal, because we don't expect any invalid records in archive
4087 : * or in records streamed from the primary. Files in the archive should be complete,
4088 : * and we should never hit the end of WAL because we stop and wait for more WAL
4089 : * to arrive before replaying it.
4090 : *
4091 : * NOTE: This function remembers the RecPtr value it was last called with,
4092 : * to suppress repeated messages about the same record. Only call this when
4093 : * you are about to ereport(), or you might cause a later message to be
4094 : * erroneously suppressed.
4095 : */
4096 : static int
4097 566 : emode_for_corrupt_record(int emode, XLogRecPtr RecPtr)
4098 : {
4099 : static XLogRecPtr lastComplaint = 0;
4100 :
4101 566 : if (readSource == XLOG_FROM_PG_WAL && emode == LOG)
4102 : {
4103 558 : if (RecPtr == lastComplaint)
4104 152 : emode = DEBUG1;
4105 : else
4106 406 : lastComplaint = RecPtr;
4107 : }
4108 566 : return emode;
4109 : }
4110 :
4111 :
4112 : /*
4113 : * Subroutine to try to fetch and validate a prior checkpoint record.
4114 : */
4115 : static XLogRecord *
4116 1984 : ReadCheckpointRecord(XLogPrefetcher *xlogprefetcher, XLogRecPtr RecPtr,
4117 : TimeLineID replayTLI)
4118 : {
4119 : XLogRecord *record;
4120 : uint8 info;
4121 :
4122 : Assert(xlogreader != NULL);
4123 :
4124 1984 : if (!XRecOffIsValid(RecPtr))
4125 : {
4126 0 : ereport(LOG,
4127 : (errmsg("invalid checkpoint location")));
4128 0 : return NULL;
4129 : }
4130 :
4131 1984 : XLogPrefetcherBeginRead(xlogprefetcher, RecPtr);
4132 1984 : record = ReadRecord(xlogprefetcher, LOG, true, replayTLI);
4133 :
4134 1984 : if (record == NULL)
4135 : {
4136 0 : ereport(LOG,
4137 : (errmsg("invalid checkpoint record")));
4138 0 : return NULL;
4139 : }
4140 1984 : if (record->xl_rmid != RM_XLOG_ID)
4141 : {
4142 0 : ereport(LOG,
4143 : (errmsg("invalid resource manager ID in checkpoint record")));
4144 0 : return NULL;
4145 : }
4146 1984 : info = record->xl_info & ~XLR_INFO_MASK;
4147 1984 : if (info != XLOG_CHECKPOINT_SHUTDOWN &&
4148 : info != XLOG_CHECKPOINT_ONLINE)
4149 : {
4150 0 : ereport(LOG,
4151 : (errmsg("invalid xl_info in checkpoint record")));
4152 0 : return NULL;
4153 : }
4154 1984 : if (record->xl_tot_len != SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(CheckPoint))
4155 : {
4156 0 : ereport(LOG,
4157 : (errmsg("invalid length of checkpoint record")));
4158 0 : return NULL;
4159 : }
4160 1984 : return record;
4161 : }
4162 :
4163 : /*
4164 : * Scan for new timelines that might have appeared in the archive since we
4165 : * started recovery.
4166 : *
4167 : * If there are any, the function changes recovery target TLI to the latest
4168 : * one and returns 'true'.
4169 : */
4170 : static bool
4171 410 : rescanLatestTimeLine(TimeLineID replayTLI, XLogRecPtr replayLSN)
4172 : {
4173 : List *newExpectedTLEs;
4174 : bool found;
4175 : ListCell *cell;
4176 : TimeLineID newtarget;
4177 410 : TimeLineID oldtarget = recoveryTargetTLI;
4178 410 : TimeLineHistoryEntry *currentTle = NULL;
4179 :
4180 410 : newtarget = findNewestTimeLine(recoveryTargetTLI);
4181 408 : if (newtarget == recoveryTargetTLI)
4182 : {
4183 : /* No new timelines found */
4184 394 : return false;
4185 : }
4186 :
4187 : /*
4188 : * Determine the list of expected TLIs for the new TLI
4189 : */
4190 :
4191 14 : newExpectedTLEs = readTimeLineHistory(newtarget);
4192 :
4193 : /*
4194 : * If the current timeline is not part of the history of the new timeline,
4195 : * we cannot proceed to it.
4196 : */
4197 14 : found = false;
4198 28 : foreach(cell, newExpectedTLEs)
4199 : {
4200 28 : currentTle = (TimeLineHistoryEntry *) lfirst(cell);
4201 :
4202 28 : if (currentTle->tli == recoveryTargetTLI)
4203 : {
4204 14 : found = true;
4205 14 : break;
4206 : }
4207 : }
4208 14 : if (!found)
4209 : {
4210 0 : ereport(LOG,
4211 : (errmsg("new timeline %u is not a child of database system timeline %u",
4212 : newtarget,
4213 : replayTLI)));
4214 0 : return false;
4215 : }
4216 :
4217 : /*
4218 : * The current timeline was found in the history file, but check that the
4219 : * next timeline was forked off from it *after* the current recovery
4220 : * location.
4221 : */
4222 14 : if (currentTle->end < replayLSN)
4223 : {
4224 0 : ereport(LOG,
4225 : errmsg("new timeline %u forked off current database system timeline %u before current recovery point %X/%08X",
4226 : newtarget,
4227 : replayTLI,
4228 : LSN_FORMAT_ARGS(replayLSN)));
4229 0 : return false;
4230 : }
4231 :
4232 : /* The new timeline history seems valid. Switch target */
4233 14 : recoveryTargetTLI = newtarget;
4234 14 : list_free_deep(expectedTLEs);
4235 14 : expectedTLEs = newExpectedTLEs;
4236 :
4237 : /*
4238 : * As in StartupXLOG(), try to ensure we have all the history files
4239 : * between the old target and new target in pg_wal.
4240 : */
4241 14 : restoreTimeLineHistoryFiles(oldtarget + 1, newtarget);
4242 :
4243 14 : ereport(LOG,
4244 : (errmsg("new target timeline is %u",
4245 : recoveryTargetTLI)));
4246 :
4247 14 : return true;
4248 : }
4249 :
4250 :
4251 : /*
4252 : * Open a logfile segment for reading (during recovery).
4253 : *
4254 : * If source == XLOG_FROM_ARCHIVE, the segment is retrieved from archive.
4255 : * Otherwise, it's assumed to be already available in pg_wal.
4256 : */
4257 : static int
4258 7116 : XLogFileRead(XLogSegNo segno, TimeLineID tli,
4259 : XLogSource source, bool notfoundOk)
4260 : {
4261 : char xlogfname[MAXFNAMELEN];
4262 : char activitymsg[MAXFNAMELEN + 16];
4263 : char path[MAXPGPATH];
4264 : int fd;
4265 :
4266 7116 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
4267 :
4268 7116 : switch (source)
4269 : {
4270 1720 : case XLOG_FROM_ARCHIVE:
4271 : /* Report recovery progress in PS display */
4272 1720 : snprintf(activitymsg, sizeof(activitymsg), "waiting for %s",
4273 : xlogfname);
4274 1720 : set_ps_display(activitymsg);
4275 :
4276 1720 : if (!RestoreArchivedFile(path, xlogfname,
4277 : "RECOVERYXLOG",
4278 : wal_segment_size,
4279 : InRedo))
4280 994 : return -1;
4281 724 : break;
4282 :
4283 5396 : case XLOG_FROM_PG_WAL:
4284 : case XLOG_FROM_STREAM:
4285 5396 : XLogFilePath(path, tli, segno, wal_segment_size);
4286 5396 : break;
4287 :
4288 0 : default:
4289 0 : elog(ERROR, "invalid XLogFileRead source %d", source);
4290 : }
4291 :
4292 : /*
4293 : * If the segment was fetched from archival storage, replace the existing
4294 : * xlog segment (if any) with the archival version.
4295 : */
4296 6120 : if (source == XLOG_FROM_ARCHIVE)
4297 : {
4298 : Assert(!IsInstallXLogFileSegmentActive());
4299 724 : KeepFileRestoredFromArchive(path, xlogfname);
4300 :
4301 : /*
4302 : * Set path to point at the new file in pg_wal.
4303 : */
4304 724 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlogfname);
4305 : }
4306 :
4307 6120 : fd = BasicOpenFile(path, O_RDONLY | PG_BINARY);
4308 6120 : if (fd >= 0)
4309 : {
4310 : /* Success! */
4311 5710 : curFileTLI = tli;
4312 :
4313 : /* Report recovery progress in PS display */
4314 5710 : snprintf(activitymsg, sizeof(activitymsg), "recovering %s",
4315 : xlogfname);
4316 5710 : set_ps_display(activitymsg);
4317 :
4318 : /* Track source of data in assorted state variables */
4319 5710 : readSource = source;
4320 5710 : XLogReceiptSource = source;
4321 : /* In FROM_STREAM case, caller tracks receipt time, not me */
4322 5710 : if (source != XLOG_FROM_STREAM)
4323 3282 : XLogReceiptTime = GetCurrentTimestamp();
4324 :
4325 5710 : return fd;
4326 : }
4327 410 : if (errno != ENOENT || !notfoundOk) /* unexpected failure? */
4328 0 : ereport(PANIC,
4329 : (errcode_for_file_access(),
4330 : errmsg("could not open file \"%s\": %m", path)));
4331 410 : return -1;
4332 : }
4333 :
4334 : /*
4335 : * Open a logfile segment for reading (during recovery).
4336 : *
4337 : * This version searches for the segment with any TLI listed in expectedTLEs.
4338 : */
4339 : static int
4340 3676 : XLogFileReadAnyTLI(XLogSegNo segno, XLogSource source)
4341 : {
4342 : char path[MAXPGPATH];
4343 : ListCell *cell;
4344 : int fd;
4345 : List *tles;
4346 :
4347 : /*
4348 : * Loop looking for a suitable timeline ID: we might need to read any of
4349 : * the timelines listed in expectedTLEs.
4350 : *
4351 : * We expect curFileTLI on entry to be the TLI of the preceding file in
4352 : * sequence, or 0 if there was no predecessor. We do not allow curFileTLI
4353 : * to go backwards; this prevents us from picking up the wrong file when a
4354 : * parent timeline extends to higher segment numbers than the child we
4355 : * want to read.
4356 : *
4357 : * If we haven't read the timeline history file yet, read it now, so that
4358 : * we know which TLIs to scan. We don't save the list in expectedTLEs,
4359 : * however, unless we actually find a valid segment. That way if there is
4360 : * neither a timeline history file nor a WAL segment in the archive, and
4361 : * streaming replication is set up, we'll read the timeline history file
4362 : * streamed from the primary when we start streaming, instead of
4363 : * recovering with a dummy history generated here.
4364 : */
4365 3676 : if (expectedTLEs)
4366 1692 : tles = expectedTLEs;
4367 : else
4368 1984 : tles = readTimeLineHistory(recoveryTargetTLI);
4369 :
4370 4100 : foreach(cell, tles)
4371 : {
4372 3720 : TimeLineHistoryEntry *hent = (TimeLineHistoryEntry *) lfirst(cell);
4373 3720 : TimeLineID tli = hent->tli;
4374 :
4375 3720 : if (tli < curFileTLI)
4376 12 : break; /* don't bother looking at too-old TLIs */
4377 :
4378 : /*
4379 : * Skip scanning the timeline ID that the logfile segment to read
4380 : * doesn't belong to
4381 : */
4382 3708 : if (XLogRecPtrIsValid(hent->begin))
4383 : {
4384 156 : XLogSegNo beginseg = 0;
4385 :
4386 156 : XLByteToSeg(hent->begin, beginseg, wal_segment_size);
4387 :
4388 : /*
4389 : * The logfile segment that doesn't belong to the timeline is
4390 : * older or newer than the segment that the timeline started or
4391 : * ended at, respectively. It's sufficient to check only the
4392 : * starting segment of the timeline here. Since the timelines are
4393 : * scanned in descending order in this loop, any segments newer
4394 : * than the ending segment should belong to newer timeline and
4395 : * have already been read before. So it's not necessary to check
4396 : * the ending segment of the timeline here.
4397 : */
4398 156 : if (segno < beginseg)
4399 14 : continue;
4400 : }
4401 :
4402 3694 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_ARCHIVE)
4403 : {
4404 1720 : fd = XLogFileRead(segno, tli, XLOG_FROM_ARCHIVE, true);
4405 1718 : if (fd != -1)
4406 : {
4407 724 : elog(DEBUG1, "got WAL segment from archive");
4408 724 : if (!expectedTLEs)
4409 36 : expectedTLEs = tles;
4410 3282 : return fd;
4411 : }
4412 : }
4413 :
4414 2968 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_PG_WAL)
4415 : {
4416 2968 : fd = XLogFileRead(segno, tli, XLOG_FROM_PG_WAL, true);
4417 2968 : if (fd != -1)
4418 : {
4419 2558 : if (!expectedTLEs)
4420 1948 : expectedTLEs = tles;
4421 2558 : return fd;
4422 : }
4423 : }
4424 : }
4425 :
4426 : /* Couldn't find it. For simplicity, complain about front timeline */
4427 392 : XLogFilePath(path, recoveryTargetTLI, segno, wal_segment_size);
4428 392 : errno = ENOENT;
4429 392 : ereport(DEBUG2,
4430 : (errcode_for_file_access(),
4431 : errmsg("could not open file \"%s\": %m", path)));
4432 392 : return -1;
4433 : }
4434 :
4435 : /*
4436 : * Set flag to signal the walreceiver to restart. (The startup process calls
4437 : * this on noticing a relevant configuration change.)
4438 : */
4439 : void
4440 24 : StartupRequestWalReceiverRestart(void)
4441 : {
4442 24 : if (currentSource == XLOG_FROM_STREAM && WalRcvRunning())
4443 : {
4444 16 : ereport(LOG,
4445 : (errmsg("WAL receiver process shutdown requested")));
4446 :
4447 16 : pendingWalRcvRestart = true;
4448 : }
4449 24 : }
4450 :
4451 :
4452 : /*
4453 : * Has a standby promotion already been triggered?
4454 : *
4455 : * Unlike CheckForStandbyTrigger(), this works in any process
4456 : * that's connected to shared memory.
4457 : */
4458 : bool
4459 142 : PromoteIsTriggered(void)
4460 : {
4461 : /*
4462 : * We check shared state each time only until a standby promotion is
4463 : * triggered. We can't trigger a promotion again, so there's no need to
4464 : * keep checking after the shared variable has once been seen true.
4465 : */
4466 142 : if (LocalPromoteIsTriggered)
4467 100 : return true;
4468 :
4469 42 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4470 42 : LocalPromoteIsTriggered = XLogRecoveryCtl->SharedPromoteIsTriggered;
4471 42 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4472 :
4473 42 : return LocalPromoteIsTriggered;
4474 : }
4475 :
4476 : static void
4477 94 : SetPromoteIsTriggered(void)
4478 : {
4479 94 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4480 94 : XLogRecoveryCtl->SharedPromoteIsTriggered = true;
4481 94 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4482 :
4483 : /*
4484 : * Mark the recovery pause state as 'not paused' because the paused state
4485 : * ends and promotion continues if a promotion is triggered while recovery
4486 : * is paused. Otherwise pg_get_wal_replay_pause_state() can mistakenly
4487 : * return 'paused' while a promotion is ongoing.
4488 : */
4489 94 : SetRecoveryPause(false);
4490 :
4491 94 : LocalPromoteIsTriggered = true;
4492 94 : }
4493 :
4494 : /*
4495 : * Check whether a promote request has arrived.
4496 : */
4497 : static bool
4498 13138 : CheckForStandbyTrigger(void)
4499 : {
4500 13138 : if (LocalPromoteIsTriggered)
4501 114 : return true;
4502 :
4503 13024 : if (IsPromoteSignaled() && CheckPromoteSignal())
4504 : {
4505 94 : ereport(LOG, (errmsg("received promote request")));
4506 94 : RemovePromoteSignalFiles();
4507 94 : ResetPromoteSignaled();
4508 94 : SetPromoteIsTriggered();
4509 94 : return true;
4510 : }
4511 :
4512 12930 : return false;
4513 : }
4514 :
4515 : /*
4516 : * Remove the files signaling a standby promotion request.
4517 : */
4518 : void
4519 1910 : RemovePromoteSignalFiles(void)
4520 : {
4521 1910 : unlink(PROMOTE_SIGNAL_FILE);
4522 1910 : }
4523 :
4524 : /*
4525 : * Check to see if a promote request has arrived.
4526 : */
4527 : bool
4528 1742 : CheckPromoteSignal(void)
4529 : {
4530 : struct stat stat_buf;
4531 :
4532 1742 : if (stat(PROMOTE_SIGNAL_FILE, &stat_buf) == 0)
4533 190 : return true;
4534 :
4535 1552 : return false;
4536 : }
4537 :
4538 : /*
4539 : * Wake up startup process to replay newly arrived WAL, or to notice that
4540 : * failover has been requested.
4541 : */
4542 : void
4543 68972 : WakeupRecovery(void)
4544 : {
4545 68972 : SetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
4546 68972 : }
4547 :
4548 : /*
4549 : * Schedule a walreceiver wakeup in the main recovery loop.
4550 : */
4551 : void
4552 4 : XLogRequestWalReceiverReply(void)
4553 : {
4554 4 : doRequestWalReceiverReply = true;
4555 4 : }
4556 :
4557 : /*
4558 : * Is HotStandby active yet? This is only important in special backends
4559 : * since normal backends won't ever be able to connect until this returns
4560 : * true. Postmaster knows this by way of signal, not via shared memory.
4561 : *
4562 : * Unlike testing standbyState, this works in any process that's connected to
4563 : * shared memory. (And note that standbyState alone doesn't tell the truth
4564 : * anyway.)
4565 : */
4566 : bool
4567 328 : HotStandbyActive(void)
4568 : {
4569 : /*
4570 : * We check shared state each time only until Hot Standby is active. We
4571 : * can't de-activate Hot Standby, so there's no need to keep checking
4572 : * after the shared variable has once been seen true.
4573 : */
4574 328 : if (LocalHotStandbyActive)
4575 50 : return true;
4576 : else
4577 : {
4578 : /* spinlock is essential on machines with weak memory ordering! */
4579 278 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4580 278 : LocalHotStandbyActive = XLogRecoveryCtl->SharedHotStandbyActive;
4581 278 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4582 :
4583 278 : return LocalHotStandbyActive;
4584 : }
4585 : }
4586 :
4587 : /*
4588 : * Like HotStandbyActive(), but to be used only in WAL replay code,
4589 : * where we don't need to ask any other process what the state is.
4590 : */
4591 : static bool
4592 0 : HotStandbyActiveInReplay(void)
4593 : {
4594 : Assert(AmStartupProcess() || !IsPostmasterEnvironment);
4595 0 : return LocalHotStandbyActive;
4596 : }
4597 :
4598 : /*
4599 : * Get latest redo apply position.
4600 : *
4601 : * Exported to allow WALReceiver to read the pointer directly.
4602 : */
4603 : XLogRecPtr
4604 172254 : GetXLogReplayRecPtr(TimeLineID *replayTLI)
4605 : {
4606 : XLogRecPtr recptr;
4607 : TimeLineID tli;
4608 :
4609 172254 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4610 172254 : recptr = XLogRecoveryCtl->lastReplayedEndRecPtr;
4611 172254 : tli = XLogRecoveryCtl->lastReplayedTLI;
4612 172254 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4613 :
4614 172254 : if (replayTLI)
4615 7242 : *replayTLI = tli;
4616 172254 : return recptr;
4617 : }
4618 :
4619 :
4620 : /*
4621 : * Get position of last applied, or the record being applied.
4622 : *
4623 : * This is different from GetXLogReplayRecPtr() in that if a WAL
4624 : * record is currently being applied, this includes that record.
4625 : */
4626 : XLogRecPtr
4627 12486 : GetCurrentReplayRecPtr(TimeLineID *replayEndTLI)
4628 : {
4629 : XLogRecPtr recptr;
4630 : TimeLineID tli;
4631 :
4632 12486 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4633 12486 : recptr = XLogRecoveryCtl->replayEndRecPtr;
4634 12486 : tli = XLogRecoveryCtl->replayEndTLI;
4635 12486 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4636 :
4637 12486 : if (replayEndTLI)
4638 12486 : *replayEndTLI = tli;
4639 12486 : return recptr;
4640 : }
4641 :
4642 : /*
4643 : * Save timestamp of latest processed commit/abort record.
4644 : *
4645 : * We keep this in XLogRecoveryCtl, not a simple static variable, so that it can be
4646 : * seen by processes other than the startup process. Note in particular
4647 : * that CreateRestartPoint is executed in the checkpointer.
4648 : */
4649 : static void
4650 44412 : SetLatestXTime(TimestampTz xtime)
4651 : {
4652 44412 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4653 44412 : XLogRecoveryCtl->recoveryLastXTime = xtime;
4654 44412 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4655 44412 : }
4656 :
4657 : /*
4658 : * Fetch timestamp of latest processed commit/abort record.
4659 : */
4660 : TimestampTz
4661 696 : GetLatestXTime(void)
4662 : {
4663 : TimestampTz xtime;
4664 :
4665 696 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4666 696 : xtime = XLogRecoveryCtl->recoveryLastXTime;
4667 696 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4668 :
4669 696 : return xtime;
4670 : }
4671 :
4672 : /*
4673 : * Save timestamp of the next chunk of WAL records to apply.
4674 : *
4675 : * We keep this in XLogRecoveryCtl, not a simple static variable, so that it can be
4676 : * seen by all backends.
4677 : */
4678 : static void
4679 8682 : SetCurrentChunkStartTime(TimestampTz xtime)
4680 : {
4681 8682 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4682 8682 : XLogRecoveryCtl->currentChunkStartTime = xtime;
4683 8682 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4684 8682 : }
4685 :
4686 : /*
4687 : * Fetch timestamp of latest processed commit/abort record.
4688 : * Startup process maintains an accurate local copy in XLogReceiptTime
4689 : */
4690 : TimestampTz
4691 604 : GetCurrentChunkReplayStartTime(void)
4692 : {
4693 : TimestampTz xtime;
4694 :
4695 604 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4696 604 : xtime = XLogRecoveryCtl->currentChunkStartTime;
4697 604 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4698 :
4699 604 : return xtime;
4700 : }
4701 :
4702 : /*
4703 : * Returns time of receipt of current chunk of XLOG data, as well as
4704 : * whether it was received from streaming replication or from archives.
4705 : */
4706 : void
4707 54 : GetXLogReceiptTime(TimestampTz *rtime, bool *fromStream)
4708 : {
4709 : /*
4710 : * This must be executed in the startup process, since we don't export the
4711 : * relevant state to shared memory.
4712 : */
4713 : Assert(InRecovery);
4714 :
4715 54 : *rtime = XLogReceiptTime;
4716 54 : *fromStream = (XLogReceiptSource == XLOG_FROM_STREAM);
4717 54 : }
4718 :
4719 : /*
4720 : * Note that text field supplied is a parameter name and does not require
4721 : * translation
4722 : */
4723 : void
4724 1360 : RecoveryRequiresIntParameter(const char *param_name, int currValue, int minValue)
4725 : {
4726 1360 : if (currValue < minValue)
4727 : {
4728 0 : if (HotStandbyActiveInReplay())
4729 : {
4730 0 : bool warned_for_promote = false;
4731 :
4732 0 : ereport(WARNING,
4733 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4734 : errmsg("hot standby is not possible because of insufficient parameter settings"),
4735 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4736 : param_name,
4737 : currValue,
4738 : minValue)));
4739 :
4740 0 : SetRecoveryPause(true);
4741 :
4742 0 : ereport(LOG,
4743 : (errmsg("recovery has paused"),
4744 : errdetail("If recovery is unpaused, the server will shut down."),
4745 : errhint("You can then restart the server after making the necessary configuration changes.")));
4746 :
4747 0 : while (GetRecoveryPauseState() != RECOVERY_NOT_PAUSED)
4748 : {
4749 0 : ProcessStartupProcInterrupts();
4750 :
4751 0 : if (CheckForStandbyTrigger())
4752 : {
4753 0 : if (!warned_for_promote)
4754 0 : ereport(WARNING,
4755 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4756 : errmsg("promotion is not possible because of insufficient parameter settings"),
4757 :
4758 : /*
4759 : * Repeat the detail from above so it's easy to find
4760 : * in the log.
4761 : */
4762 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4763 : param_name,
4764 : currValue,
4765 : minValue),
4766 : errhint("Restart the server after making the necessary configuration changes.")));
4767 0 : warned_for_promote = true;
4768 : }
4769 :
4770 : /*
4771 : * If recovery pause is requested then set it paused. While
4772 : * we are in the loop, user might resume and pause again so
4773 : * set this every time.
4774 : */
4775 0 : ConfirmRecoveryPaused();
4776 :
4777 : /*
4778 : * We wait on a condition variable that will wake us as soon
4779 : * as the pause ends, but we use a timeout so we can check the
4780 : * above conditions periodically too.
4781 : */
4782 0 : ConditionVariableTimedSleep(&XLogRecoveryCtl->recoveryNotPausedCV, 1000,
4783 : WAIT_EVENT_RECOVERY_PAUSE);
4784 : }
4785 0 : ConditionVariableCancelSleep();
4786 : }
4787 :
4788 0 : ereport(FATAL,
4789 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4790 : errmsg("recovery aborted because of insufficient parameter settings"),
4791 : /* Repeat the detail from above so it's easy to find in the log. */
4792 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4793 : param_name,
4794 : currValue,
4795 : minValue),
4796 : errhint("You can restart the server after making the necessary configuration changes.")));
4797 : }
4798 1360 : }
4799 :
4800 :
4801 : /*
4802 : * GUC check_hook for primary_slot_name
4803 : */
4804 : bool
4805 2752 : check_primary_slot_name(char **newval, void **extra, GucSource source)
4806 : {
4807 : int err_code;
4808 2752 : char *err_msg = NULL;
4809 2752 : char *err_hint = NULL;
4810 :
4811 2752 : if (*newval && strcmp(*newval, "") != 0 &&
4812 402 : !ReplicationSlotValidateNameInternal(*newval, false, &err_code,
4813 : &err_msg, &err_hint))
4814 : {
4815 0 : GUC_check_errcode(err_code);
4816 0 : GUC_check_errdetail("%s", err_msg);
4817 0 : if (err_hint != NULL)
4818 0 : GUC_check_errhint("%s", err_hint);
4819 0 : return false;
4820 : }
4821 :
4822 2752 : return true;
4823 : }
4824 :
4825 : /*
4826 : * Recovery target settings: Only one of the several recovery_target* settings
4827 : * may be set. Setting a second one results in an error. The global variable
4828 : * recoveryTarget tracks which kind of recovery target was chosen. Other
4829 : * variables store the actual target value (for example a string or a xid).
4830 : * The assign functions of the parameters check whether a competing parameter
4831 : * was already set. But we want to allow setting the same parameter multiple
4832 : * times. We also want to allow unsetting a parameter and setting a different
4833 : * one, so we unset recoveryTarget when the parameter is set to an empty
4834 : * string.
4835 : *
4836 : * XXX this code is broken by design. Throwing an error from a GUC assign
4837 : * hook breaks fundamental assumptions of guc.c. So long as all the variables
4838 : * for which this can happen are PGC_POSTMASTER, the consequences are limited,
4839 : * since we'd just abort postmaster startup anyway. Nonetheless it's likely
4840 : * that we have odd behaviors such as unexpected GUC ordering dependencies.
4841 : */
4842 :
4843 : pg_noreturn static void
4844 2 : error_multiple_recovery_targets(void)
4845 : {
4846 2 : ereport(ERROR,
4847 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4848 : errmsg("multiple recovery targets specified"),
4849 : errdetail("At most one of \"recovery_target\", \"recovery_target_lsn\", \"recovery_target_name\", \"recovery_target_time\", \"recovery_target_xid\" may be set.")));
4850 : }
4851 :
4852 : /*
4853 : * GUC check_hook for recovery_target
4854 : */
4855 : bool
4856 2350 : check_recovery_target(char **newval, void **extra, GucSource source)
4857 : {
4858 2350 : if (strcmp(*newval, "immediate") != 0 && strcmp(*newval, "") != 0)
4859 : {
4860 0 : GUC_check_errdetail("The only allowed value is \"immediate\".");
4861 0 : return false;
4862 : }
4863 2350 : return true;
4864 : }
4865 :
4866 : /*
4867 : * GUC assign_hook for recovery_target
4868 : */
4869 : void
4870 2350 : assign_recovery_target(const char *newval, void *extra)
4871 : {
4872 2350 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4873 0 : recoveryTarget != RECOVERY_TARGET_IMMEDIATE)
4874 0 : error_multiple_recovery_targets();
4875 :
4876 2350 : if (newval && strcmp(newval, "") != 0)
4877 2 : recoveryTarget = RECOVERY_TARGET_IMMEDIATE;
4878 : else
4879 2348 : recoveryTarget = RECOVERY_TARGET_UNSET;
4880 2350 : }
4881 :
4882 : /*
4883 : * GUC check_hook for recovery_target_lsn
4884 : */
4885 : bool
4886 2362 : check_recovery_target_lsn(char **newval, void **extra, GucSource source)
4887 : {
4888 2362 : if (strcmp(*newval, "") != 0)
4889 : {
4890 : XLogRecPtr lsn;
4891 : XLogRecPtr *myextra;
4892 16 : ErrorSaveContext escontext = {T_ErrorSaveContext};
4893 :
4894 16 : lsn = pg_lsn_in_safe(*newval, (Node *) &escontext);
4895 16 : if (escontext.error_occurred)
4896 0 : return false;
4897 :
4898 16 : myextra = (XLogRecPtr *) guc_malloc(LOG, sizeof(XLogRecPtr));
4899 16 : if (!myextra)
4900 0 : return false;
4901 16 : *myextra = lsn;
4902 16 : *extra = myextra;
4903 : }
4904 2362 : return true;
4905 : }
4906 :
4907 : /*
4908 : * GUC assign_hook for recovery_target_lsn
4909 : */
4910 : void
4911 2362 : assign_recovery_target_lsn(const char *newval, void *extra)
4912 : {
4913 2362 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4914 0 : recoveryTarget != RECOVERY_TARGET_LSN)
4915 0 : error_multiple_recovery_targets();
4916 :
4917 2362 : if (newval && strcmp(newval, "") != 0)
4918 : {
4919 16 : recoveryTarget = RECOVERY_TARGET_LSN;
4920 16 : recoveryTargetLSN = *((XLogRecPtr *) extra);
4921 : }
4922 : else
4923 2346 : recoveryTarget = RECOVERY_TARGET_UNSET;
4924 2362 : }
4925 :
4926 : /*
4927 : * GUC check_hook for recovery_target_name
4928 : */
4929 : bool
4930 2362 : check_recovery_target_name(char **newval, void **extra, GucSource source)
4931 : {
4932 : /* Use the value of newval directly */
4933 2362 : if (strlen(*newval) >= MAXFNAMELEN)
4934 : {
4935 0 : GUC_check_errdetail("\"%s\" is too long (maximum %d characters).",
4936 : "recovery_target_name", MAXFNAMELEN - 1);
4937 0 : return false;
4938 : }
4939 2362 : return true;
4940 : }
4941 :
4942 : /*
4943 : * GUC assign_hook for recovery_target_name
4944 : */
4945 : void
4946 2362 : assign_recovery_target_name(const char *newval, void *extra)
4947 : {
4948 2362 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4949 0 : recoveryTarget != RECOVERY_TARGET_NAME)
4950 0 : error_multiple_recovery_targets();
4951 :
4952 2362 : if (newval && strcmp(newval, "") != 0)
4953 : {
4954 12 : recoveryTarget = RECOVERY_TARGET_NAME;
4955 12 : recoveryTargetName = newval;
4956 : }
4957 : else
4958 2350 : recoveryTarget = RECOVERY_TARGET_UNSET;
4959 2362 : }
4960 :
4961 : /*
4962 : * GUC check_hook for recovery_target_time
4963 : *
4964 : * The interpretation of the recovery_target_time string can depend on the
4965 : * time zone setting, so we need to wait until after all GUC processing is
4966 : * done before we can do the final parsing of the string. This check function
4967 : * only does a parsing pass to catch syntax errors, but we store the string
4968 : * and parse it again when we need to use it.
4969 : */
4970 : bool
4971 2354 : check_recovery_target_time(char **newval, void **extra, GucSource source)
4972 : {
4973 2354 : if (strcmp(*newval, "") != 0)
4974 : {
4975 : /* reject some special values */
4976 6 : if (strcmp(*newval, "now") == 0 ||
4977 6 : strcmp(*newval, "today") == 0 ||
4978 6 : strcmp(*newval, "tomorrow") == 0 ||
4979 6 : strcmp(*newval, "yesterday") == 0)
4980 : {
4981 0 : return false;
4982 : }
4983 :
4984 : /*
4985 : * parse timestamp value (see also timestamptz_in())
4986 : */
4987 : {
4988 6 : char *str = *newval;
4989 : fsec_t fsec;
4990 : struct pg_tm tt,
4991 6 : *tm = &tt;
4992 : int tz;
4993 : int dtype;
4994 : int nf;
4995 : int dterr;
4996 : char *field[MAXDATEFIELDS];
4997 : int ftype[MAXDATEFIELDS];
4998 : char workbuf[MAXDATELEN + MAXDATEFIELDS];
4999 : DateTimeErrorExtra dtextra;
5000 : TimestampTz timestamp;
5001 :
5002 6 : dterr = ParseDateTime(str, workbuf, sizeof(workbuf),
5003 : field, ftype, MAXDATEFIELDS, &nf);
5004 6 : if (dterr == 0)
5005 6 : dterr = DecodeDateTime(field, ftype, nf,
5006 : &dtype, tm, &fsec, &tz, &dtextra);
5007 6 : if (dterr != 0)
5008 0 : return false;
5009 6 : if (dtype != DTK_DATE)
5010 0 : return false;
5011 :
5012 6 : if (tm2timestamp(tm, fsec, &tz, ×tamp) != 0)
5013 : {
5014 0 : GUC_check_errdetail("Timestamp out of range: \"%s\".", str);
5015 0 : return false;
5016 : }
5017 : }
5018 : }
5019 2354 : return true;
5020 : }
5021 :
5022 : /*
5023 : * GUC assign_hook for recovery_target_time
5024 : */
5025 : void
5026 2354 : assign_recovery_target_time(const char *newval, void *extra)
5027 : {
5028 2354 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
5029 2 : recoveryTarget != RECOVERY_TARGET_TIME)
5030 2 : error_multiple_recovery_targets();
5031 :
5032 2352 : if (newval && strcmp(newval, "") != 0)
5033 4 : recoveryTarget = RECOVERY_TARGET_TIME;
5034 : else
5035 2348 : recoveryTarget = RECOVERY_TARGET_UNSET;
5036 2352 : }
5037 :
5038 : /*
5039 : * GUC check_hook for recovery_target_timeline
5040 : */
5041 : bool
5042 2356 : check_recovery_target_timeline(char **newval, void **extra, GucSource source)
5043 : {
5044 : RecoveryTargetTimeLineGoal rttg;
5045 : RecoveryTargetTimeLineGoal *myextra;
5046 :
5047 2356 : if (strcmp(*newval, "current") == 0)
5048 0 : rttg = RECOVERY_TARGET_TIMELINE_CONTROLFILE;
5049 2356 : else if (strcmp(*newval, "latest") == 0)
5050 2350 : rttg = RECOVERY_TARGET_TIMELINE_LATEST;
5051 : else
5052 : {
5053 : char *endp;
5054 : uint64 timeline;
5055 :
5056 6 : rttg = RECOVERY_TARGET_TIMELINE_NUMERIC;
5057 :
5058 6 : errno = 0;
5059 6 : timeline = strtou64(*newval, &endp, 0);
5060 :
5061 6 : if (*endp != '\0' || errno == EINVAL || errno == ERANGE)
5062 : {
5063 2 : GUC_check_errdetail("\"%s\" is not a valid number.",
5064 : "recovery_target_timeline");
5065 6 : return false;
5066 : }
5067 :
5068 4 : if (timeline < 1 || timeline > PG_UINT32_MAX)
5069 : {
5070 4 : GUC_check_errdetail("\"%s\" must be between %u and %u.",
5071 : "recovery_target_timeline", 1, UINT_MAX);
5072 4 : return false;
5073 : }
5074 : }
5075 :
5076 2350 : myextra = (RecoveryTargetTimeLineGoal *) guc_malloc(LOG, sizeof(RecoveryTargetTimeLineGoal));
5077 2350 : if (!myextra)
5078 0 : return false;
5079 2350 : *myextra = rttg;
5080 2350 : *extra = myextra;
5081 :
5082 2350 : return true;
5083 : }
5084 :
5085 : /*
5086 : * GUC assign_hook for recovery_target_timeline
5087 : */
5088 : void
5089 2350 : assign_recovery_target_timeline(const char *newval, void *extra)
5090 : {
5091 2350 : recoveryTargetTimeLineGoal = *((RecoveryTargetTimeLineGoal *) extra);
5092 2350 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
5093 0 : recoveryTargetTLIRequested = (TimeLineID) strtoul(newval, NULL, 0);
5094 : else
5095 2350 : recoveryTargetTLIRequested = 0;
5096 2350 : }
5097 :
5098 : /*
5099 : * GUC check_hook for recovery_target_xid
5100 : */
5101 : bool
5102 2350 : check_recovery_target_xid(char **newval, void **extra, GucSource source)
5103 : {
5104 2350 : if (strcmp(*newval, "") != 0)
5105 : {
5106 : TransactionId xid;
5107 : TransactionId *myextra;
5108 :
5109 2 : errno = 0;
5110 2 : xid = (TransactionId) strtou64(*newval, NULL, 0);
5111 2 : if (errno == EINVAL || errno == ERANGE)
5112 0 : return false;
5113 :
5114 2 : myextra = (TransactionId *) guc_malloc(LOG, sizeof(TransactionId));
5115 2 : if (!myextra)
5116 0 : return false;
5117 2 : *myextra = xid;
5118 2 : *extra = myextra;
5119 : }
5120 2350 : return true;
5121 : }
5122 :
5123 : /*
5124 : * GUC assign_hook for recovery_target_xid
5125 : */
5126 : void
5127 2350 : assign_recovery_target_xid(const char *newval, void *extra)
5128 : {
5129 2350 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
5130 0 : recoveryTarget != RECOVERY_TARGET_XID)
5131 0 : error_multiple_recovery_targets();
5132 :
5133 2350 : if (newval && strcmp(newval, "") != 0)
5134 : {
5135 2 : recoveryTarget = RECOVERY_TARGET_XID;
5136 2 : recoveryTargetXid = *((TransactionId *) extra);
5137 : }
5138 : else
5139 2348 : recoveryTarget = RECOVERY_TARGET_UNSET;
5140 2350 : }
|
