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) 1996-2025, PostgreSQL Global Development Group
18 : * Portions Copyright (c) 1994, Regents of the University of California
19 : *
20 : * src/backend/access/transam/xlogrecovery.c
21 : *
22 : *-------------------------------------------------------------------------
23 : */
24 :
25 : #include "postgres.h"
26 :
27 : #include <ctype.h>
28 : #include <math.h>
29 : #include <time.h>
30 : #include <sys/stat.h>
31 : #include <sys/time.h>
32 : #include <unistd.h>
33 :
34 : #include "access/timeline.h"
35 : #include "access/transam.h"
36 : #include "access/xact.h"
37 : #include "access/xlog_internal.h"
38 : #include "access/xlogarchive.h"
39 : #include "access/xlogprefetcher.h"
40 : #include "access/xlogreader.h"
41 : #include "access/xlogrecovery.h"
42 : #include "access/xlogutils.h"
43 : #include "backup/basebackup.h"
44 : #include "catalog/pg_control.h"
45 : #include "commands/tablespace.h"
46 : #include "common/file_utils.h"
47 : #include "miscadmin.h"
48 : #include "pgstat.h"
49 : #include "postmaster/bgwriter.h"
50 : #include "postmaster/startup.h"
51 : #include "replication/slot.h"
52 : #include "replication/slotsync.h"
53 : #include "replication/walreceiver.h"
54 : #include "storage/fd.h"
55 : #include "storage/ipc.h"
56 : #include "storage/latch.h"
57 : #include "storage/pmsignal.h"
58 : #include "storage/procarray.h"
59 : #include "storage/spin.h"
60 : #include "utils/datetime.h"
61 : #include "utils/fmgrprotos.h"
62 : #include "utils/guc_hooks.h"
63 : #include "utils/pgstat_internal.h"
64 : #include "utils/pg_lsn.h"
65 : #include "utils/ps_status.h"
66 : #include "utils/pg_rusage.h"
67 :
68 : /* Unsupported old recovery command file names (relative to $PGDATA) */
69 : #define RECOVERY_COMMAND_FILE "recovery.conf"
70 : #define RECOVERY_COMMAND_DONE "recovery.done"
71 :
72 : /*
73 : * GUC support
74 : */
75 : const struct config_enum_entry recovery_target_action_options[] = {
76 : {"pause", RECOVERY_TARGET_ACTION_PAUSE, false},
77 : {"promote", RECOVERY_TARGET_ACTION_PROMOTE, false},
78 : {"shutdown", RECOVERY_TARGET_ACTION_SHUTDOWN, false},
79 : {NULL, 0, false}
80 : };
81 :
82 : /* options formerly taken from recovery.conf for archive recovery */
83 : char *recoveryRestoreCommand = NULL;
84 : char *recoveryEndCommand = NULL;
85 : char *archiveCleanupCommand = NULL;
86 : RecoveryTargetType recoveryTarget = RECOVERY_TARGET_UNSET;
87 : bool recoveryTargetInclusive = true;
88 : int recoveryTargetAction = RECOVERY_TARGET_ACTION_PAUSE;
89 : TransactionId recoveryTargetXid;
90 : char *recovery_target_time_string;
91 : TimestampTz recoveryTargetTime;
92 : const char *recoveryTargetName;
93 : XLogRecPtr recoveryTargetLSN;
94 : int recovery_min_apply_delay = 0;
95 :
96 : /* options formerly taken from recovery.conf for XLOG streaming */
97 : char *PrimaryConnInfo = NULL;
98 : char *PrimarySlotName = NULL;
99 : bool wal_receiver_create_temp_slot = false;
100 :
101 : /*
102 : * recoveryTargetTimeLineGoal: what the user requested, if any
103 : *
104 : * recoveryTargetTLIRequested: numeric value of requested timeline, if constant
105 : *
106 : * recoveryTargetTLI: the currently understood target timeline; changes
107 : *
108 : * expectedTLEs: a list of TimeLineHistoryEntries for recoveryTargetTLI and
109 : * the timelines of its known parents, newest first (so recoveryTargetTLI is
110 : * always the first list member). Only these TLIs are expected to be seen in
111 : * the WAL segments we read, and indeed only these TLIs will be considered as
112 : * candidate WAL files to open at all.
113 : *
114 : * curFileTLI: the TLI appearing in the name of the current input WAL file.
115 : * (This is not necessarily the same as the timeline from which we are
116 : * replaying WAL, which StartupXLOG calls replayTLI, because we could be
117 : * scanning data that was copied from an ancestor timeline when the current
118 : * file was created.) During a sequential scan we do not allow this value
119 : * to decrease.
120 : */
121 : RecoveryTargetTimeLineGoal recoveryTargetTimeLineGoal = RECOVERY_TARGET_TIMELINE_LATEST;
122 : TimeLineID recoveryTargetTLIRequested = 0;
123 : TimeLineID recoveryTargetTLI = 0;
124 : static List *expectedTLEs;
125 : static TimeLineID curFileTLI;
126 :
127 : /*
128 : * When ArchiveRecoveryRequested is set, archive recovery was requested,
129 : * ie. signal files were present. When InArchiveRecovery is set, we are
130 : * currently recovering using offline XLOG archives. These variables are only
131 : * valid in the startup process.
132 : *
133 : * When ArchiveRecoveryRequested is true, but InArchiveRecovery is false, we're
134 : * currently performing crash recovery using only XLOG files in pg_wal, but
135 : * will switch to using offline XLOG archives as soon as we reach the end of
136 : * WAL in pg_wal.
137 : */
138 : bool ArchiveRecoveryRequested = false;
139 : bool InArchiveRecovery = false;
140 :
141 : /*
142 : * When StandbyModeRequested is set, standby mode was requested, i.e.
143 : * standby.signal file was present. When StandbyMode is set, we are currently
144 : * in standby mode. These variables are only valid in the startup process.
145 : * They work similarly to ArchiveRecoveryRequested and InArchiveRecovery.
146 : */
147 : static bool StandbyModeRequested = false;
148 : bool StandbyMode = false;
149 :
150 : /* was a signal file present at startup? */
151 : static bool standby_signal_file_found = false;
152 : static bool recovery_signal_file_found = false;
153 :
154 : /*
155 : * CheckPointLoc is the position of the checkpoint record that determines
156 : * where to start the replay. It comes from the backup label file or the
157 : * control file.
158 : *
159 : * RedoStartLSN is the checkpoint's REDO location, also from the backup label
160 : * file or the control file. In standby mode, XLOG streaming usually starts
161 : * from the position where an invalid record was found. But if we fail to
162 : * read even the initial checkpoint record, we use the REDO location instead
163 : * of the checkpoint location as the start position of XLOG streaming.
164 : * Otherwise we would have to jump backwards to the REDO location after
165 : * reading the checkpoint record, because the REDO record can precede the
166 : * checkpoint record.
167 : */
168 : static XLogRecPtr CheckPointLoc = InvalidXLogRecPtr;
169 : static TimeLineID CheckPointTLI = 0;
170 : static XLogRecPtr RedoStartLSN = InvalidXLogRecPtr;
171 : static TimeLineID RedoStartTLI = 0;
172 :
173 : /*
174 : * Local copy of SharedHotStandbyActive variable. False actually means "not
175 : * known, need to check the shared state".
176 : */
177 : static bool LocalHotStandbyActive = false;
178 :
179 : /*
180 : * Local copy of SharedPromoteIsTriggered variable. False actually means "not
181 : * known, need to check the shared state".
182 : */
183 : static bool LocalPromoteIsTriggered = false;
184 :
185 : /* Has the recovery code requested a walreceiver wakeup? */
186 : static bool doRequestWalReceiverReply;
187 :
188 : /* XLogReader object used to parse the WAL records */
189 : static XLogReaderState *xlogreader = NULL;
190 :
191 : /* XLogPrefetcher object used to consume WAL records with read-ahead */
192 : static XLogPrefetcher *xlogprefetcher = NULL;
193 :
194 : /* Parameters passed down from ReadRecord to the XLogPageRead callback. */
195 : typedef struct XLogPageReadPrivate
196 : {
197 : int emode;
198 : bool fetching_ckpt; /* are we fetching a checkpoint record? */
199 : bool randAccess;
200 : TimeLineID replayTLI;
201 : } XLogPageReadPrivate;
202 :
203 : /* flag to tell XLogPageRead that we have started replaying */
204 : static bool InRedo = false;
205 :
206 : /*
207 : * Codes indicating where we got a WAL file from during recovery, or where
208 : * to attempt to get one.
209 : */
210 : typedef enum
211 : {
212 : XLOG_FROM_ANY = 0, /* request to read WAL from any source */
213 : XLOG_FROM_ARCHIVE, /* restored using restore_command */
214 : XLOG_FROM_PG_WAL, /* existing file in pg_wal */
215 : XLOG_FROM_STREAM, /* streamed from primary */
216 : } XLogSource;
217 :
218 : /* human-readable names for XLogSources, for debugging output */
219 : static const char *const xlogSourceNames[] = {"any", "archive", "pg_wal", "stream"};
220 :
221 : /*
222 : * readFile is -1 or a kernel FD for the log file segment that's currently
223 : * open for reading. readSegNo identifies the segment. readOff is the offset
224 : * of the page just read, readLen indicates how much of it has been read into
225 : * readBuf, and readSource indicates where we got the currently open file from.
226 : *
227 : * Note: we could use Reserve/ReleaseExternalFD to track consumption of this
228 : * FD too (like for openLogFile in xlog.c); but it doesn't currently seem
229 : * worthwhile, since the XLOG is not read by general-purpose sessions.
230 : */
231 : static int readFile = -1;
232 : static XLogSegNo readSegNo = 0;
233 : static uint32 readOff = 0;
234 : static uint32 readLen = 0;
235 : static XLogSource readSource = XLOG_FROM_ANY;
236 :
237 : /*
238 : * Keeps track of which source we're currently reading from. This is
239 : * different from readSource in that this is always set, even when we don't
240 : * currently have a WAL file open. If lastSourceFailed is set, our last
241 : * attempt to read from currentSource failed, and we should try another source
242 : * next.
243 : *
244 : * pendingWalRcvRestart is set when a config change occurs that requires a
245 : * walreceiver restart. This is only valid in XLOG_FROM_STREAM state.
246 : */
247 : static XLogSource currentSource = XLOG_FROM_ANY;
248 : static bool lastSourceFailed = false;
249 : static bool pendingWalRcvRestart = false;
250 :
251 : /*
252 : * These variables track when we last obtained some WAL data to process,
253 : * and where we got it from. (XLogReceiptSource is initially the same as
254 : * readSource, but readSource gets reset to zero when we don't have data
255 : * to process right now. It is also different from currentSource, which
256 : * also changes when we try to read from a source and fail, while
257 : * XLogReceiptSource tracks where we last successfully read some WAL.)
258 : */
259 : static TimestampTz XLogReceiptTime = 0;
260 : static XLogSource XLogReceiptSource = XLOG_FROM_ANY;
261 :
262 : /* Local copy of WalRcv->flushedUpto */
263 : static XLogRecPtr flushedUpto = 0;
264 : static TimeLineID receiveTLI = 0;
265 :
266 : /*
267 : * Copy of minRecoveryPoint and backupEndPoint from the control file.
268 : *
269 : * In order to reach consistency, we must replay the WAL up to
270 : * minRecoveryPoint. If backupEndRequired is true, we must also reach
271 : * backupEndPoint, or if it's invalid, an end-of-backup record corresponding
272 : * to backupStartPoint.
273 : *
274 : * Note: In archive recovery, after consistency has been reached, the
275 : * functions in xlog.c will start updating minRecoveryPoint in the control
276 : * file. But this copy of minRecoveryPoint variable reflects the value at the
277 : * beginning of recovery, and is *not* updated after consistency is reached.
278 : */
279 : static XLogRecPtr minRecoveryPoint;
280 : static TimeLineID minRecoveryPointTLI;
281 :
282 : static XLogRecPtr backupStartPoint;
283 : static XLogRecPtr backupEndPoint;
284 : static bool backupEndRequired = false;
285 :
286 : /*
287 : * Have we reached a consistent database state? In crash recovery, we have
288 : * to replay all the WAL, so reachedConsistency is never set. During archive
289 : * recovery, the database is consistent once minRecoveryPoint is reached.
290 : *
291 : * Consistent state means that the system is internally consistent, all
292 : * the WAL has been replayed up to a certain point, and importantly, there
293 : * is no trace of later actions on disk.
294 : *
295 : * This flag is used only by the startup process and postmaster. When
296 : * minRecoveryPoint is reached, the startup process sets it to true and
297 : * sends a PMSIGNAL_RECOVERY_CONSISTENT signal to the postmaster,
298 : * which then sets it to true upon receiving the signal.
299 : */
300 : bool reachedConsistency = false;
301 :
302 : /* Buffers dedicated to consistency checks of size BLCKSZ */
303 : static char *replay_image_masked = NULL;
304 : static char *primary_image_masked = NULL;
305 :
306 :
307 : /*
308 : * Shared-memory state for WAL recovery.
309 : */
310 : typedef struct XLogRecoveryCtlData
311 : {
312 : /*
313 : * SharedHotStandbyActive indicates if we allow hot standby queries to be
314 : * run. Protected by info_lck.
315 : */
316 : bool SharedHotStandbyActive;
317 :
318 : /*
319 : * SharedPromoteIsTriggered indicates if a standby promotion has been
320 : * triggered. Protected by info_lck.
321 : */
322 : bool SharedPromoteIsTriggered;
323 :
324 : /*
325 : * recoveryWakeupLatch is used to wake up the startup process to continue
326 : * WAL replay, if it is waiting for WAL to arrive or promotion to be
327 : * requested.
328 : *
329 : * Note that the startup process also uses another latch, its procLatch,
330 : * to wait for recovery conflict. If we get rid of recoveryWakeupLatch for
331 : * signaling the startup process in favor of using its procLatch, which
332 : * comports better with possible generic signal handlers using that latch.
333 : * But we should not do that because the startup process doesn't assume
334 : * that it's waken up by walreceiver process or SIGHUP signal handler
335 : * while it's waiting for recovery conflict. The separate latches,
336 : * recoveryWakeupLatch and procLatch, should be used for inter-process
337 : * communication for WAL replay and recovery conflict, respectively.
338 : */
339 : Latch recoveryWakeupLatch;
340 :
341 : /*
342 : * Last record successfully replayed.
343 : */
344 : XLogRecPtr lastReplayedReadRecPtr; /* start position */
345 : XLogRecPtr lastReplayedEndRecPtr; /* end+1 position */
346 : TimeLineID lastReplayedTLI; /* timeline */
347 :
348 : /*
349 : * When we're currently replaying a record, ie. in a redo function,
350 : * replayEndRecPtr points to the end+1 of the record being replayed,
351 : * otherwise it's equal to lastReplayedEndRecPtr.
352 : */
353 : XLogRecPtr replayEndRecPtr;
354 : TimeLineID replayEndTLI;
355 : /* timestamp of last COMMIT/ABORT record replayed (or being replayed) */
356 : TimestampTz recoveryLastXTime;
357 :
358 : /*
359 : * timestamp of when we started replaying the current chunk of WAL data,
360 : * only relevant for replication or archive recovery
361 : */
362 : TimestampTz currentChunkStartTime;
363 : /* Recovery pause state */
364 : RecoveryPauseState recoveryPauseState;
365 : ConditionVariable recoveryNotPausedCV;
366 :
367 : slock_t info_lck; /* locks shared variables shown above */
368 : } XLogRecoveryCtlData;
369 :
370 : static XLogRecoveryCtlData *XLogRecoveryCtl = NULL;
371 :
372 : /*
373 : * abortedRecPtr is the start pointer of a broken record at end of WAL when
374 : * recovery completes; missingContrecPtr is the location of the first
375 : * contrecord that went missing. See CreateOverwriteContrecordRecord for
376 : * details.
377 : */
378 : static XLogRecPtr abortedRecPtr;
379 : static XLogRecPtr missingContrecPtr;
380 :
381 : /*
382 : * if recoveryStopsBefore/After returns true, it saves information of the stop
383 : * point here
384 : */
385 : static TransactionId recoveryStopXid;
386 : static TimestampTz recoveryStopTime;
387 : static XLogRecPtr recoveryStopLSN;
388 : static char recoveryStopName[MAXFNAMELEN];
389 : static bool recoveryStopAfter;
390 :
391 : /* prototypes for local functions */
392 : static void ApplyWalRecord(XLogReaderState *xlogreader, XLogRecord *record, TimeLineID *replayTLI);
393 :
394 : static void EnableStandbyMode(void);
395 : static void readRecoverySignalFile(void);
396 : static void validateRecoveryParameters(void);
397 : static bool read_backup_label(XLogRecPtr *checkPointLoc,
398 : TimeLineID *backupLabelTLI,
399 : bool *backupEndRequired, bool *backupFromStandby);
400 : static bool read_tablespace_map(List **tablespaces);
401 :
402 : static void xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI);
403 : static void CheckRecoveryConsistency(void);
404 : static void rm_redo_error_callback(void *arg);
405 : #ifdef WAL_DEBUG
406 : static void xlog_outrec(StringInfo buf, XLogReaderState *record);
407 : #endif
408 : static void xlog_block_info(StringInfo buf, XLogReaderState *record);
409 : static void checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI,
410 : TimeLineID prevTLI, TimeLineID replayTLI);
411 : static bool getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime);
412 : static void verifyBackupPageConsistency(XLogReaderState *record);
413 :
414 : static bool recoveryStopsBefore(XLogReaderState *record);
415 : static bool recoveryStopsAfter(XLogReaderState *record);
416 : static char *getRecoveryStopReason(void);
417 : static void recoveryPausesHere(bool endOfRecovery);
418 : static bool recoveryApplyDelay(XLogReaderState *record);
419 : static void ConfirmRecoveryPaused(void);
420 :
421 : static XLogRecord *ReadRecord(XLogPrefetcher *xlogprefetcher,
422 : int emode, bool fetching_ckpt,
423 : TimeLineID replayTLI);
424 :
425 : static int XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr,
426 : int reqLen, XLogRecPtr targetRecPtr, char *readBuf);
427 : static XLogPageReadResult WaitForWALToBecomeAvailable(XLogRecPtr RecPtr,
428 : bool randAccess,
429 : bool fetching_ckpt,
430 : XLogRecPtr tliRecPtr,
431 : TimeLineID replayTLI,
432 : XLogRecPtr replayLSN,
433 : bool nonblocking);
434 : static int emode_for_corrupt_record(int emode, XLogRecPtr RecPtr);
435 : static XLogRecord *ReadCheckpointRecord(XLogPrefetcher *xlogprefetcher,
436 : XLogRecPtr RecPtr, TimeLineID replayTLI);
437 : static bool rescanLatestTimeLine(TimeLineID replayTLI, XLogRecPtr replayLSN);
438 : static int XLogFileRead(XLogSegNo segno, TimeLineID tli,
439 : XLogSource source, bool notfoundOk);
440 : static int XLogFileReadAnyTLI(XLogSegNo segno, XLogSource source);
441 :
442 : static bool CheckForStandbyTrigger(void);
443 : static void SetPromoteIsTriggered(void);
444 : static bool HotStandbyActiveInReplay(void);
445 :
446 : static void SetCurrentChunkStartTime(TimestampTz xtime);
447 : static void SetLatestXTime(TimestampTz xtime);
448 :
449 : /*
450 : * Initialization of shared memory for WAL recovery
451 : */
452 : Size
453 6084 : XLogRecoveryShmemSize(void)
454 : {
455 : Size size;
456 :
457 : /* XLogRecoveryCtl */
458 6084 : size = sizeof(XLogRecoveryCtlData);
459 :
460 6084 : return size;
461 : }
462 :
463 : void
464 2128 : XLogRecoveryShmemInit(void)
465 : {
466 : bool found;
467 :
468 2128 : XLogRecoveryCtl = (XLogRecoveryCtlData *)
469 2128 : ShmemInitStruct("XLOG Recovery Ctl", XLogRecoveryShmemSize(), &found);
470 2128 : if (found)
471 0 : return;
472 2128 : memset(XLogRecoveryCtl, 0, sizeof(XLogRecoveryCtlData));
473 :
474 2128 : SpinLockInit(&XLogRecoveryCtl->info_lck);
475 2128 : InitSharedLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
476 2128 : ConditionVariableInit(&XLogRecoveryCtl->recoveryNotPausedCV);
477 : }
478 :
479 : /*
480 : * A thin wrapper to enable StandbyMode and do other preparatory work as
481 : * needed.
482 : */
483 : static void
484 206 : EnableStandbyMode(void)
485 : {
486 206 : StandbyMode = true;
487 :
488 : /*
489 : * To avoid server log bloat, we don't report recovery progress in a
490 : * standby as it will always be in recovery unless promoted. We disable
491 : * startup progress timeout in standby mode to avoid calling
492 : * startup_progress_timeout_handler() unnecessarily.
493 : */
494 206 : disable_startup_progress_timeout();
495 206 : }
496 :
497 : /*
498 : * Prepare the system for WAL recovery, if needed.
499 : *
500 : * This is called by StartupXLOG() which coordinates the server startup
501 : * sequence. This function analyzes the control file and the backup label
502 : * file, if any, and figures out whether we need to perform crash recovery or
503 : * archive recovery, and how far we need to replay the WAL to reach a
504 : * consistent state.
505 : *
506 : * This doesn't yet change the on-disk state, except for creating the symlinks
507 : * from table space map file if any, and for fetching WAL files needed to find
508 : * the checkpoint record. On entry, the caller has already read the control
509 : * file into memory, and passes it as argument. This function updates it to
510 : * reflect the recovery state, and the caller is expected to write it back to
511 : * disk does after initializing other subsystems, but before calling
512 : * PerformWalRecovery().
513 : *
514 : * This initializes some global variables like ArchiveRecoveryRequested, and
515 : * StandbyModeRequested and InRecovery.
516 : */
517 : void
518 1844 : InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr,
519 : bool *haveBackupLabel_ptr, bool *haveTblspcMap_ptr)
520 : {
521 : XLogPageReadPrivate *private;
522 : struct stat st;
523 : bool wasShutdown;
524 : XLogRecord *record;
525 : DBState dbstate_at_startup;
526 1844 : bool haveTblspcMap = false;
527 1844 : bool haveBackupLabel = false;
528 : CheckPoint checkPoint;
529 1844 : bool backupFromStandby = false;
530 :
531 1844 : dbstate_at_startup = ControlFile->state;
532 :
533 : /*
534 : * Initialize on the assumption we want to recover to the latest timeline
535 : * that's active according to pg_control.
536 : */
537 1844 : if (ControlFile->minRecoveryPointTLI >
538 1844 : ControlFile->checkPointCopy.ThisTimeLineID)
539 4 : recoveryTargetTLI = ControlFile->minRecoveryPointTLI;
540 : else
541 1840 : recoveryTargetTLI = ControlFile->checkPointCopy.ThisTimeLineID;
542 :
543 : /*
544 : * Check for signal files, and if so set up state for offline recovery
545 : */
546 1844 : readRecoverySignalFile();
547 1844 : validateRecoveryParameters();
548 :
549 : /*
550 : * Take ownership of the wakeup latch if we're going to sleep during
551 : * recovery, if required.
552 : */
553 1844 : if (ArchiveRecoveryRequested)
554 216 : OwnLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
555 :
556 : /*
557 : * Set the WAL reading processor now, as it will be needed when reading
558 : * the checkpoint record required (backup_label or not).
559 : */
560 1844 : private = palloc0(sizeof(XLogPageReadPrivate));
561 1844 : xlogreader =
562 1844 : XLogReaderAllocate(wal_segment_size, NULL,
563 1844 : XL_ROUTINE(.page_read = &XLogPageRead,
564 : .segment_open = NULL,
565 : .segment_close = wal_segment_close),
566 : private);
567 1844 : if (!xlogreader)
568 0 : ereport(ERROR,
569 : (errcode(ERRCODE_OUT_OF_MEMORY),
570 : errmsg("out of memory"),
571 : errdetail("Failed while allocating a WAL reading processor.")));
572 1844 : xlogreader->system_identifier = ControlFile->system_identifier;
573 :
574 : /*
575 : * Set the WAL decode buffer size. This limits how far ahead we can read
576 : * in the WAL.
577 : */
578 1844 : XLogReaderSetDecodeBuffer(xlogreader, NULL, wal_decode_buffer_size);
579 :
580 : /* Create a WAL prefetcher. */
581 1844 : xlogprefetcher = XLogPrefetcherAllocate(xlogreader);
582 :
583 : /*
584 : * Allocate two page buffers dedicated to WAL consistency checks. We do
585 : * it this way, rather than just making static arrays, for two reasons:
586 : * (1) no need to waste the storage in most instantiations of the backend;
587 : * (2) a static char array isn't guaranteed to have any particular
588 : * alignment, whereas palloc() will provide MAXALIGN'd storage.
589 : */
590 1844 : replay_image_masked = (char *) palloc(BLCKSZ);
591 1844 : primary_image_masked = (char *) palloc(BLCKSZ);
592 :
593 : /*
594 : * Read the backup_label file. We want to run this part of the recovery
595 : * process after checking for signal files and after performing validation
596 : * of the recovery parameters.
597 : */
598 1844 : if (read_backup_label(&CheckPointLoc, &CheckPointTLI, &backupEndRequired,
599 : &backupFromStandby))
600 : {
601 142 : List *tablespaces = NIL;
602 :
603 : /*
604 : * Archive recovery was requested, and thanks to the backup label
605 : * file, we know how far we need to replay to reach consistency. Enter
606 : * archive recovery directly.
607 : */
608 142 : InArchiveRecovery = true;
609 142 : if (StandbyModeRequested)
610 120 : EnableStandbyMode();
611 :
612 : /*
613 : * Omitting backup_label when creating a new replica, PITR node etc.
614 : * unfortunately is a common cause of corruption. Logging that
615 : * backup_label was used makes it a bit easier to exclude that as the
616 : * cause of observed corruption.
617 : *
618 : * Do so before we try to read the checkpoint record (which can fail),
619 : * as otherwise it can be hard to understand why a checkpoint other
620 : * than ControlFile->checkPoint is used.
621 : */
622 142 : ereport(LOG,
623 : errmsg("starting backup recovery with redo LSN %X/%08X, checkpoint LSN %X/%08X, on timeline ID %u",
624 : LSN_FORMAT_ARGS(RedoStartLSN),
625 : LSN_FORMAT_ARGS(CheckPointLoc),
626 : CheckPointTLI));
627 :
628 : /*
629 : * When a backup_label file is present, we want to roll forward from
630 : * the checkpoint it identifies, rather than using pg_control.
631 : */
632 142 : record = ReadCheckpointRecord(xlogprefetcher, CheckPointLoc,
633 : CheckPointTLI);
634 142 : if (record != NULL)
635 : {
636 142 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
637 142 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
638 142 : ereport(DEBUG1,
639 : errmsg_internal("checkpoint record is at %X/%08X",
640 : LSN_FORMAT_ARGS(CheckPointLoc)));
641 142 : InRecovery = true; /* force recovery even if SHUTDOWNED */
642 :
643 : /*
644 : * Make sure that REDO location exists. This may not be the case
645 : * if there was a crash during an online backup, which left a
646 : * backup_label around that references a WAL segment that's
647 : * already been archived.
648 : */
649 142 : if (checkPoint.redo < CheckPointLoc)
650 : {
651 142 : XLogPrefetcherBeginRead(xlogprefetcher, checkPoint.redo);
652 142 : if (!ReadRecord(xlogprefetcher, LOG, false,
653 : checkPoint.ThisTimeLineID))
654 0 : ereport(FATAL,
655 : errmsg("could not find redo location %X/%08X referenced by checkpoint record at %X/%08X",
656 : LSN_FORMAT_ARGS(checkPoint.redo), LSN_FORMAT_ARGS(CheckPointLoc)),
657 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" or \"%s/standby.signal\" and add required recovery options.\n"
658 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
659 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
660 : DataDir, DataDir, DataDir, DataDir));
661 : }
662 : }
663 : else
664 : {
665 0 : ereport(FATAL,
666 : errmsg("could not locate required checkpoint record at %X/%08X",
667 : LSN_FORMAT_ARGS(CheckPointLoc)),
668 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" or \"%s/standby.signal\" and add required recovery options.\n"
669 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
670 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
671 : DataDir, DataDir, DataDir, DataDir));
672 : wasShutdown = false; /* keep compiler quiet */
673 : }
674 :
675 : /* Read the tablespace_map file if present and create symlinks. */
676 142 : if (read_tablespace_map(&tablespaces))
677 : {
678 : ListCell *lc;
679 :
680 8 : foreach(lc, tablespaces)
681 : {
682 4 : tablespaceinfo *ti = lfirst(lc);
683 : char *linkloc;
684 :
685 4 : linkloc = psprintf("%s/%u", PG_TBLSPC_DIR, ti->oid);
686 :
687 : /*
688 : * Remove the existing symlink if any and Create the symlink
689 : * under PGDATA.
690 : */
691 4 : remove_tablespace_symlink(linkloc);
692 :
693 4 : if (symlink(ti->path, linkloc) < 0)
694 0 : ereport(ERROR,
695 : (errcode_for_file_access(),
696 : errmsg("could not create symbolic link \"%s\": %m",
697 : linkloc)));
698 :
699 4 : pfree(ti->path);
700 4 : pfree(ti);
701 : }
702 :
703 : /* tell the caller to delete it later */
704 4 : haveTblspcMap = true;
705 : }
706 :
707 : /* tell the caller to delete it later */
708 142 : haveBackupLabel = true;
709 : }
710 : else
711 : {
712 : /* No backup_label file has been found if we are here. */
713 :
714 : /*
715 : * If tablespace_map file is present without backup_label file, there
716 : * is no use of such file. There is no harm in retaining it, but it
717 : * is better to get rid of the map file so that we don't have any
718 : * redundant file in data directory and it will avoid any sort of
719 : * confusion. It seems prudent though to just rename the file out of
720 : * the way rather than delete it completely, also we ignore any error
721 : * that occurs in rename operation as even if map file is present
722 : * without backup_label file, it is harmless.
723 : */
724 1702 : if (stat(TABLESPACE_MAP, &st) == 0)
725 : {
726 2 : unlink(TABLESPACE_MAP_OLD);
727 2 : if (durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, DEBUG1) == 0)
728 2 : ereport(LOG,
729 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
730 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
731 : errdetail("File \"%s\" was renamed to \"%s\".",
732 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
733 : else
734 0 : ereport(LOG,
735 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
736 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
737 : errdetail("Could not rename file \"%s\" to \"%s\": %m.",
738 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
739 : }
740 :
741 : /*
742 : * It's possible that archive recovery was requested, but we don't
743 : * know how far we need to replay the WAL before we reach consistency.
744 : * This can happen for example if a base backup is taken from a
745 : * running server using an atomic filesystem snapshot, without calling
746 : * pg_backup_start/stop. Or if you just kill a running primary server
747 : * and put it into archive recovery by creating a recovery signal
748 : * file.
749 : *
750 : * Our strategy in that case is to perform crash recovery first,
751 : * replaying all the WAL present in pg_wal, and only enter archive
752 : * recovery after that.
753 : *
754 : * But usually we already know how far we need to replay the WAL (up
755 : * to minRecoveryPoint, up to backupEndPoint, or until we see an
756 : * end-of-backup record), and we can enter archive recovery directly.
757 : */
758 1702 : if (ArchiveRecoveryRequested &&
759 86 : (ControlFile->minRecoveryPoint != InvalidXLogRecPtr ||
760 18 : ControlFile->backupEndRequired ||
761 18 : ControlFile->backupEndPoint != InvalidXLogRecPtr ||
762 18 : ControlFile->state == DB_SHUTDOWNED))
763 : {
764 82 : InArchiveRecovery = true;
765 82 : if (StandbyModeRequested)
766 82 : EnableStandbyMode();
767 : }
768 :
769 : /*
770 : * For the same reason as when starting up with backup_label present,
771 : * emit a log message when we continue initializing from a base
772 : * backup.
773 : */
774 1702 : if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
775 0 : ereport(LOG,
776 : errmsg("restarting backup recovery with redo LSN %X/%08X",
777 : LSN_FORMAT_ARGS(ControlFile->backupStartPoint)));
778 :
779 : /* Get the last valid checkpoint record. */
780 1702 : CheckPointLoc = ControlFile->checkPoint;
781 1702 : CheckPointTLI = ControlFile->checkPointCopy.ThisTimeLineID;
782 1702 : RedoStartLSN = ControlFile->checkPointCopy.redo;
783 1702 : RedoStartTLI = ControlFile->checkPointCopy.ThisTimeLineID;
784 1702 : record = ReadCheckpointRecord(xlogprefetcher, CheckPointLoc,
785 : CheckPointTLI);
786 1702 : if (record != NULL)
787 : {
788 1702 : ereport(DEBUG1,
789 : errmsg_internal("checkpoint record is at %X/%08X",
790 : LSN_FORMAT_ARGS(CheckPointLoc)));
791 : }
792 : else
793 : {
794 : /*
795 : * We used to attempt to go back to a secondary checkpoint record
796 : * here, but only when not in standby mode. We now just fail if we
797 : * can't read the last checkpoint because this allows us to
798 : * simplify processing around checkpoints.
799 : */
800 0 : ereport(PANIC,
801 : errmsg("could not locate a valid checkpoint record at %X/%08X",
802 : LSN_FORMAT_ARGS(CheckPointLoc)));
803 : }
804 1702 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
805 1702 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
806 : }
807 :
808 1844 : if (ArchiveRecoveryRequested)
809 : {
810 216 : if (StandbyModeRequested)
811 206 : ereport(LOG,
812 : (errmsg("entering standby mode")));
813 10 : else if (recoveryTarget == RECOVERY_TARGET_XID)
814 0 : ereport(LOG,
815 : (errmsg("starting point-in-time recovery to XID %u",
816 : recoveryTargetXid)));
817 10 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
818 0 : ereport(LOG,
819 : (errmsg("starting point-in-time recovery to %s",
820 : timestamptz_to_str(recoveryTargetTime))));
821 10 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
822 6 : ereport(LOG,
823 : (errmsg("starting point-in-time recovery to \"%s\"",
824 : recoveryTargetName)));
825 4 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
826 0 : ereport(LOG,
827 : errmsg("starting point-in-time recovery to WAL location (LSN) \"%X/%08X\"",
828 : LSN_FORMAT_ARGS(recoveryTargetLSN)));
829 4 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
830 0 : ereport(LOG,
831 : (errmsg("starting point-in-time recovery to earliest consistent point")));
832 : else
833 4 : ereport(LOG,
834 : (errmsg("starting archive recovery")));
835 : }
836 :
837 : /*
838 : * If the location of the checkpoint record is not on the expected
839 : * timeline in the history of the requested timeline, we cannot proceed:
840 : * the backup is not part of the history of the requested timeline.
841 : */
842 : Assert(expectedTLEs); /* was initialized by reading checkpoint
843 : * record */
844 1844 : if (tliOfPointInHistory(CheckPointLoc, expectedTLEs) !=
845 : CheckPointTLI)
846 : {
847 : XLogRecPtr switchpoint;
848 :
849 : /*
850 : * tliSwitchPoint will throw an error if the checkpoint's timeline is
851 : * not in expectedTLEs at all.
852 : */
853 0 : switchpoint = tliSwitchPoint(CheckPointTLI, expectedTLEs, NULL);
854 0 : ereport(FATAL,
855 : (errmsg("requested timeline %u is not a child of this server's history",
856 : recoveryTargetTLI),
857 : /* translator: %s is a backup_label file or a pg_control file */
858 : 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.",
859 : haveBackupLabel ? "backup_label" : "pg_control",
860 : LSN_FORMAT_ARGS(CheckPointLoc),
861 : CheckPointTLI,
862 : LSN_FORMAT_ARGS(switchpoint))));
863 : }
864 :
865 : /*
866 : * The min recovery point should be part of the requested timeline's
867 : * history, too.
868 : */
869 1844 : if (!XLogRecPtrIsInvalid(ControlFile->minRecoveryPoint) &&
870 80 : tliOfPointInHistory(ControlFile->minRecoveryPoint - 1, expectedTLEs) !=
871 80 : ControlFile->minRecoveryPointTLI)
872 0 : ereport(FATAL,
873 : errmsg("requested timeline %u does not contain minimum recovery point %X/%08X on timeline %u",
874 : recoveryTargetTLI,
875 : LSN_FORMAT_ARGS(ControlFile->minRecoveryPoint),
876 : ControlFile->minRecoveryPointTLI));
877 :
878 1844 : ereport(DEBUG1,
879 : errmsg_internal("redo record is at %X/%08X; shutdown %s",
880 : LSN_FORMAT_ARGS(checkPoint.redo),
881 : wasShutdown ? "true" : "false"));
882 1844 : ereport(DEBUG1,
883 : (errmsg_internal("next transaction ID: " UINT64_FORMAT "; next OID: %u",
884 : U64FromFullTransactionId(checkPoint.nextXid),
885 : checkPoint.nextOid)));
886 1844 : ereport(DEBUG1,
887 : (errmsg_internal("next MultiXactId: %u; next MultiXactOffset: %u",
888 : checkPoint.nextMulti, checkPoint.nextMultiOffset)));
889 1844 : ereport(DEBUG1,
890 : (errmsg_internal("oldest unfrozen transaction ID: %u, in database %u",
891 : checkPoint.oldestXid, checkPoint.oldestXidDB)));
892 1844 : ereport(DEBUG1,
893 : (errmsg_internal("oldest MultiXactId: %u, in database %u",
894 : checkPoint.oldestMulti, checkPoint.oldestMultiDB)));
895 1844 : ereport(DEBUG1,
896 : (errmsg_internal("commit timestamp Xid oldest/newest: %u/%u",
897 : checkPoint.oldestCommitTsXid,
898 : checkPoint.newestCommitTsXid)));
899 1844 : if (!TransactionIdIsNormal(XidFromFullTransactionId(checkPoint.nextXid)))
900 0 : ereport(PANIC,
901 : (errmsg("invalid next transaction ID")));
902 :
903 : /* sanity check */
904 1844 : if (checkPoint.redo > CheckPointLoc)
905 0 : ereport(PANIC,
906 : (errmsg("invalid redo in checkpoint record")));
907 :
908 : /*
909 : * Check whether we need to force recovery from WAL. If it appears to
910 : * have been a clean shutdown and we did not have a recovery signal file,
911 : * then assume no recovery needed.
912 : */
913 1844 : if (checkPoint.redo < CheckPointLoc)
914 : {
915 226 : if (wasShutdown)
916 0 : ereport(PANIC,
917 : (errmsg("invalid redo record in shutdown checkpoint")));
918 226 : InRecovery = true;
919 : }
920 1618 : else if (ControlFile->state != DB_SHUTDOWNED)
921 188 : InRecovery = true;
922 1430 : else if (ArchiveRecoveryRequested)
923 : {
924 : /* force recovery due to presence of recovery signal file */
925 14 : InRecovery = true;
926 : }
927 :
928 : /*
929 : * If recovery is needed, update our in-memory copy of pg_control to show
930 : * that we are recovering and to show the selected checkpoint as the place
931 : * we are starting from. We also mark pg_control with any minimum recovery
932 : * stop point obtained from a backup history file.
933 : *
934 : * We don't write the changes to disk yet, though. Only do that after
935 : * initializing various subsystems.
936 : */
937 1844 : if (InRecovery)
938 : {
939 428 : if (InArchiveRecovery)
940 : {
941 224 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
942 : }
943 : else
944 : {
945 204 : ereport(LOG,
946 : (errmsg("database system was not properly shut down; "
947 : "automatic recovery in progress")));
948 204 : if (recoveryTargetTLI > ControlFile->checkPointCopy.ThisTimeLineID)
949 4 : ereport(LOG,
950 : (errmsg("crash recovery starts in timeline %u "
951 : "and has target timeline %u",
952 : ControlFile->checkPointCopy.ThisTimeLineID,
953 : recoveryTargetTLI)));
954 204 : ControlFile->state = DB_IN_CRASH_RECOVERY;
955 : }
956 428 : ControlFile->checkPoint = CheckPointLoc;
957 428 : ControlFile->checkPointCopy = checkPoint;
958 428 : if (InArchiveRecovery)
959 : {
960 : /* initialize minRecoveryPoint if not set yet */
961 224 : if (ControlFile->minRecoveryPoint < checkPoint.redo)
962 : {
963 148 : ControlFile->minRecoveryPoint = checkPoint.redo;
964 148 : ControlFile->minRecoveryPointTLI = checkPoint.ThisTimeLineID;
965 : }
966 : }
967 :
968 : /*
969 : * Set backupStartPoint if we're starting recovery from a base backup.
970 : *
971 : * Also set backupEndPoint and use minRecoveryPoint as the backup end
972 : * location if we're starting recovery from a base backup which was
973 : * taken from a standby. In this case, the database system status in
974 : * pg_control must indicate that the database was already in recovery.
975 : * Usually that will be DB_IN_ARCHIVE_RECOVERY but also can be
976 : * DB_SHUTDOWNED_IN_RECOVERY if recovery previously was interrupted
977 : * before reaching this point; e.g. because restore_command or
978 : * primary_conninfo were faulty.
979 : *
980 : * Any other state indicates that the backup somehow became corrupted
981 : * and we can't sensibly continue with recovery.
982 : */
983 428 : if (haveBackupLabel)
984 : {
985 142 : ControlFile->backupStartPoint = checkPoint.redo;
986 142 : ControlFile->backupEndRequired = backupEndRequired;
987 :
988 142 : if (backupFromStandby)
989 : {
990 8 : if (dbstate_at_startup != DB_IN_ARCHIVE_RECOVERY &&
991 : dbstate_at_startup != DB_SHUTDOWNED_IN_RECOVERY)
992 0 : ereport(FATAL,
993 : (errmsg("backup_label contains data inconsistent with control file"),
994 : errhint("This means that the backup is corrupted and you will "
995 : "have to use another backup for recovery.")));
996 8 : ControlFile->backupEndPoint = ControlFile->minRecoveryPoint;
997 : }
998 : }
999 : }
1000 :
1001 : /* remember these, so that we know when we have reached consistency */
1002 1844 : backupStartPoint = ControlFile->backupStartPoint;
1003 1844 : backupEndRequired = ControlFile->backupEndRequired;
1004 1844 : backupEndPoint = ControlFile->backupEndPoint;
1005 1844 : if (InArchiveRecovery)
1006 : {
1007 224 : minRecoveryPoint = ControlFile->minRecoveryPoint;
1008 224 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
1009 : }
1010 : else
1011 : {
1012 1620 : minRecoveryPoint = InvalidXLogRecPtr;
1013 1620 : minRecoveryPointTLI = 0;
1014 : }
1015 :
1016 : /*
1017 : * Start recovery assuming that the final record isn't lost.
1018 : */
1019 1844 : abortedRecPtr = InvalidXLogRecPtr;
1020 1844 : missingContrecPtr = InvalidXLogRecPtr;
1021 :
1022 1844 : *wasShutdown_ptr = wasShutdown;
1023 1844 : *haveBackupLabel_ptr = haveBackupLabel;
1024 1844 : *haveTblspcMap_ptr = haveTblspcMap;
1025 1844 : }
1026 :
1027 : /*
1028 : * See if there are any recovery signal files and if so, set state for
1029 : * recovery.
1030 : *
1031 : * See if there is a recovery command file (recovery.conf), and if so
1032 : * throw an ERROR since as of PG12 we no longer recognize that.
1033 : */
1034 : static void
1035 1844 : readRecoverySignalFile(void)
1036 : {
1037 : struct stat stat_buf;
1038 :
1039 1844 : if (IsBootstrapProcessingMode())
1040 1628 : return;
1041 :
1042 : /*
1043 : * Check for old recovery API file: recovery.conf
1044 : */
1045 1744 : if (stat(RECOVERY_COMMAND_FILE, &stat_buf) == 0)
1046 0 : ereport(FATAL,
1047 : (errcode_for_file_access(),
1048 : errmsg("using recovery command file \"%s\" is not supported",
1049 : RECOVERY_COMMAND_FILE)));
1050 :
1051 : /*
1052 : * Remove unused .done file, if present. Ignore if absent.
1053 : */
1054 1744 : unlink(RECOVERY_COMMAND_DONE);
1055 :
1056 : /*
1057 : * Check for recovery signal files and if found, fsync them since they
1058 : * represent server state information. We don't sweat too much about the
1059 : * possibility of fsync failure, however.
1060 : *
1061 : * If present, standby signal file takes precedence. If neither is present
1062 : * then we won't enter archive recovery.
1063 : */
1064 1744 : if (stat(STANDBY_SIGNAL_FILE, &stat_buf) == 0)
1065 : {
1066 : int fd;
1067 :
1068 206 : fd = BasicOpenFilePerm(STANDBY_SIGNAL_FILE, O_RDWR | PG_BINARY,
1069 : S_IRUSR | S_IWUSR);
1070 206 : if (fd >= 0)
1071 : {
1072 206 : (void) pg_fsync(fd);
1073 206 : close(fd);
1074 : }
1075 206 : standby_signal_file_found = true;
1076 : }
1077 1538 : else if (stat(RECOVERY_SIGNAL_FILE, &stat_buf) == 0)
1078 : {
1079 : int fd;
1080 :
1081 10 : fd = BasicOpenFilePerm(RECOVERY_SIGNAL_FILE, O_RDWR | PG_BINARY,
1082 : S_IRUSR | S_IWUSR);
1083 10 : if (fd >= 0)
1084 : {
1085 10 : (void) pg_fsync(fd);
1086 10 : close(fd);
1087 : }
1088 10 : recovery_signal_file_found = true;
1089 : }
1090 :
1091 1744 : StandbyModeRequested = false;
1092 1744 : ArchiveRecoveryRequested = false;
1093 1744 : if (standby_signal_file_found)
1094 : {
1095 206 : StandbyModeRequested = true;
1096 206 : ArchiveRecoveryRequested = true;
1097 : }
1098 1538 : else if (recovery_signal_file_found)
1099 : {
1100 10 : StandbyModeRequested = false;
1101 10 : ArchiveRecoveryRequested = true;
1102 : }
1103 : else
1104 1528 : return;
1105 :
1106 : /*
1107 : * We don't support standby mode in standalone backends; that requires
1108 : * other processes such as the WAL receiver to be alive.
1109 : */
1110 216 : if (StandbyModeRequested && !IsUnderPostmaster)
1111 0 : ereport(FATAL,
1112 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1113 : errmsg("standby mode is not supported by single-user servers")));
1114 : }
1115 :
1116 : static void
1117 1844 : validateRecoveryParameters(void)
1118 : {
1119 1844 : if (!ArchiveRecoveryRequested)
1120 1628 : return;
1121 :
1122 : /*
1123 : * Check for compulsory parameters
1124 : */
1125 216 : if (StandbyModeRequested)
1126 : {
1127 206 : if ((PrimaryConnInfo == NULL || strcmp(PrimaryConnInfo, "") == 0) &&
1128 22 : (recoveryRestoreCommand == NULL || strcmp(recoveryRestoreCommand, "") == 0))
1129 4 : ereport(WARNING,
1130 : (errmsg("specified neither \"primary_conninfo\" nor \"restore_command\""),
1131 : errhint("The database server will regularly poll the pg_wal subdirectory to check for files placed there.")));
1132 : }
1133 : else
1134 : {
1135 10 : if (recoveryRestoreCommand == NULL ||
1136 10 : strcmp(recoveryRestoreCommand, "") == 0)
1137 0 : ereport(FATAL,
1138 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1139 : errmsg("must specify \"restore_command\" when standby mode is not enabled")));
1140 : }
1141 :
1142 : /*
1143 : * Override any inconsistent requests. Note that this is a change of
1144 : * behaviour in 9.5; prior to this we simply ignored a request to pause if
1145 : * hot_standby = off, which was surprising behaviour.
1146 : */
1147 216 : if (recoveryTargetAction == RECOVERY_TARGET_ACTION_PAUSE &&
1148 202 : !EnableHotStandby)
1149 6 : recoveryTargetAction = RECOVERY_TARGET_ACTION_SHUTDOWN;
1150 :
1151 : /*
1152 : * Final parsing of recovery_target_time string; see also
1153 : * check_recovery_target_time().
1154 : */
1155 216 : if (recoveryTarget == RECOVERY_TARGET_TIME)
1156 : {
1157 0 : recoveryTargetTime = DatumGetTimestampTz(DirectFunctionCall3(timestamptz_in,
1158 : CStringGetDatum(recovery_target_time_string),
1159 : ObjectIdGetDatum(InvalidOid),
1160 : Int32GetDatum(-1)));
1161 : }
1162 :
1163 : /*
1164 : * If user specified recovery_target_timeline, validate it or compute the
1165 : * "latest" value. We can't do this until after we've gotten the restore
1166 : * command and set InArchiveRecovery, because we need to fetch timeline
1167 : * history files from the archive.
1168 : */
1169 216 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
1170 : {
1171 0 : TimeLineID rtli = recoveryTargetTLIRequested;
1172 :
1173 : /* Timeline 1 does not have a history file, all else should */
1174 0 : if (rtli != 1 && !existsTimeLineHistory(rtli))
1175 0 : ereport(FATAL,
1176 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1177 : errmsg("recovery target timeline %u does not exist",
1178 : rtli)));
1179 0 : recoveryTargetTLI = rtli;
1180 : }
1181 216 : else if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
1182 : {
1183 : /* We start the "latest" search from pg_control's timeline */
1184 216 : recoveryTargetTLI = findNewestTimeLine(recoveryTargetTLI);
1185 : }
1186 : else
1187 : {
1188 : /*
1189 : * else we just use the recoveryTargetTLI as already read from
1190 : * ControlFile
1191 : */
1192 : Assert(recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_CONTROLFILE);
1193 : }
1194 : }
1195 :
1196 : /*
1197 : * read_backup_label: check to see if a backup_label file is present
1198 : *
1199 : * If we see a backup_label during recovery, we assume that we are recovering
1200 : * from a backup dump file, and we therefore roll forward from the checkpoint
1201 : * identified by the label file, NOT what pg_control says. This avoids the
1202 : * problem that pg_control might have been archived one or more checkpoints
1203 : * later than the start of the dump, and so if we rely on it as the start
1204 : * point, we will fail to restore a consistent database state.
1205 : *
1206 : * Returns true if a backup_label was found (and fills the checkpoint
1207 : * location and TLI into *checkPointLoc and *backupLabelTLI, respectively);
1208 : * returns false if not. If this backup_label came from a streamed backup,
1209 : * *backupEndRequired is set to true. If this backup_label was created during
1210 : * recovery, *backupFromStandby is set to true.
1211 : *
1212 : * Also sets the global variables RedoStartLSN and RedoStartTLI with the LSN
1213 : * and TLI read from the backup file.
1214 : */
1215 : static bool
1216 1844 : read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI,
1217 : bool *backupEndRequired, bool *backupFromStandby)
1218 : {
1219 : char startxlogfilename[MAXFNAMELEN];
1220 : TimeLineID tli_from_walseg,
1221 : tli_from_file;
1222 : FILE *lfp;
1223 : char ch;
1224 : char backuptype[20];
1225 : char backupfrom[20];
1226 : char backuplabel[MAXPGPATH];
1227 : char backuptime[128];
1228 : uint32 hi,
1229 : lo;
1230 :
1231 : /* suppress possible uninitialized-variable warnings */
1232 1844 : *checkPointLoc = InvalidXLogRecPtr;
1233 1844 : *backupLabelTLI = 0;
1234 1844 : *backupEndRequired = false;
1235 1844 : *backupFromStandby = false;
1236 :
1237 : /*
1238 : * See if label file is present
1239 : */
1240 1844 : lfp = AllocateFile(BACKUP_LABEL_FILE, "r");
1241 1844 : if (!lfp)
1242 : {
1243 1702 : if (errno != ENOENT)
1244 0 : ereport(FATAL,
1245 : (errcode_for_file_access(),
1246 : errmsg("could not read file \"%s\": %m",
1247 : BACKUP_LABEL_FILE)));
1248 1702 : return false; /* it's not there, all is fine */
1249 : }
1250 :
1251 : /*
1252 : * Read and parse the START WAL LOCATION and CHECKPOINT lines (this code
1253 : * is pretty crude, but we are not expecting any variability in the file
1254 : * format).
1255 : */
1256 142 : if (fscanf(lfp, "START WAL LOCATION: %X/%08X (file %08X%16s)%c",
1257 142 : &hi, &lo, &tli_from_walseg, startxlogfilename, &ch) != 5 || ch != '\n')
1258 0 : ereport(FATAL,
1259 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1260 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
1261 142 : RedoStartLSN = ((uint64) hi) << 32 | lo;
1262 142 : RedoStartTLI = tli_from_walseg;
1263 142 : if (fscanf(lfp, "CHECKPOINT LOCATION: %X/%08X%c",
1264 142 : &hi, &lo, &ch) != 3 || ch != '\n')
1265 0 : ereport(FATAL,
1266 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1267 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
1268 142 : *checkPointLoc = ((uint64) hi) << 32 | lo;
1269 142 : *backupLabelTLI = tli_from_walseg;
1270 :
1271 : /*
1272 : * BACKUP METHOD lets us know if this was a typical backup ("streamed",
1273 : * which could mean either pg_basebackup or the pg_backup_start/stop
1274 : * method was used) or if this label came from somewhere else (the only
1275 : * other option today being from pg_rewind). If this was a streamed
1276 : * backup then we know that we need to play through until we get to the
1277 : * end of the WAL which was generated during the backup (at which point we
1278 : * will have reached consistency and backupEndRequired will be reset to be
1279 : * false).
1280 : */
1281 142 : if (fscanf(lfp, "BACKUP METHOD: %19s\n", backuptype) == 1)
1282 : {
1283 142 : if (strcmp(backuptype, "streamed") == 0)
1284 140 : *backupEndRequired = true;
1285 : }
1286 :
1287 : /*
1288 : * BACKUP FROM lets us know if this was from a primary or a standby. If
1289 : * it was from a standby, we'll double-check that the control file state
1290 : * matches that of a standby.
1291 : */
1292 142 : if (fscanf(lfp, "BACKUP FROM: %19s\n", backupfrom) == 1)
1293 : {
1294 142 : if (strcmp(backupfrom, "standby") == 0)
1295 8 : *backupFromStandby = true;
1296 : }
1297 :
1298 : /*
1299 : * Parse START TIME and LABEL. Those are not mandatory fields for recovery
1300 : * but checking for their presence is useful for debugging and the next
1301 : * sanity checks. Cope also with the fact that the result buffers have a
1302 : * pre-allocated size, hence if the backup_label file has been generated
1303 : * with strings longer than the maximum assumed here an incorrect parsing
1304 : * happens. That's fine as only minor consistency checks are done
1305 : * afterwards.
1306 : */
1307 142 : if (fscanf(lfp, "START TIME: %127[^\n]\n", backuptime) == 1)
1308 142 : ereport(DEBUG1,
1309 : (errmsg_internal("backup time %s in file \"%s\"",
1310 : backuptime, BACKUP_LABEL_FILE)));
1311 :
1312 142 : if (fscanf(lfp, "LABEL: %1023[^\n]\n", backuplabel) == 1)
1313 140 : ereport(DEBUG1,
1314 : (errmsg_internal("backup label %s in file \"%s\"",
1315 : backuplabel, BACKUP_LABEL_FILE)));
1316 :
1317 : /*
1318 : * START TIMELINE is new as of 11. Its parsing is not mandatory, still use
1319 : * it as a sanity check if present.
1320 : */
1321 142 : if (fscanf(lfp, "START TIMELINE: %u\n", &tli_from_file) == 1)
1322 : {
1323 140 : if (tli_from_walseg != tli_from_file)
1324 0 : ereport(FATAL,
1325 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1326 : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE),
1327 : errdetail("Timeline ID parsed is %u, but expected %u.",
1328 : tli_from_file, tli_from_walseg)));
1329 :
1330 140 : ereport(DEBUG1,
1331 : (errmsg_internal("backup timeline %u in file \"%s\"",
1332 : tli_from_file, BACKUP_LABEL_FILE)));
1333 : }
1334 :
1335 142 : if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%08X\n", &hi, &lo) > 0)
1336 0 : ereport(FATAL,
1337 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1338 : errmsg("this is an incremental backup, not a data directory"),
1339 : errhint("Use pg_combinebackup to reconstruct a valid data directory.")));
1340 :
1341 142 : if (ferror(lfp) || FreeFile(lfp))
1342 0 : ereport(FATAL,
1343 : (errcode_for_file_access(),
1344 : errmsg("could not read file \"%s\": %m",
1345 : BACKUP_LABEL_FILE)));
1346 :
1347 142 : return true;
1348 : }
1349 :
1350 : /*
1351 : * read_tablespace_map: check to see if a tablespace_map file is present
1352 : *
1353 : * If we see a tablespace_map file during recovery, we assume that we are
1354 : * recovering from a backup dump file, and we therefore need to create symlinks
1355 : * as per the information present in tablespace_map file.
1356 : *
1357 : * Returns true if a tablespace_map file was found (and fills *tablespaces
1358 : * with a tablespaceinfo struct for each tablespace listed in the file);
1359 : * returns false if not.
1360 : */
1361 : static bool
1362 142 : read_tablespace_map(List **tablespaces)
1363 : {
1364 : tablespaceinfo *ti;
1365 : FILE *lfp;
1366 : char str[MAXPGPATH];
1367 : int ch,
1368 : i,
1369 : n;
1370 : bool was_backslash;
1371 :
1372 : /*
1373 : * See if tablespace_map file is present
1374 : */
1375 142 : lfp = AllocateFile(TABLESPACE_MAP, "r");
1376 142 : if (!lfp)
1377 : {
1378 138 : if (errno != ENOENT)
1379 0 : ereport(FATAL,
1380 : (errcode_for_file_access(),
1381 : errmsg("could not read file \"%s\": %m",
1382 : TABLESPACE_MAP)));
1383 138 : return false; /* it's not there, all is fine */
1384 : }
1385 :
1386 : /*
1387 : * Read and parse the link name and path lines from tablespace_map file
1388 : * (this code is pretty crude, but we are not expecting any variability in
1389 : * the file format). De-escape any backslashes that were inserted.
1390 : */
1391 4 : i = 0;
1392 4 : was_backslash = false;
1393 154 : while ((ch = fgetc(lfp)) != EOF)
1394 : {
1395 150 : if (!was_backslash && (ch == '\n' || ch == '\r'))
1396 4 : {
1397 : char *endp;
1398 :
1399 4 : if (i == 0)
1400 0 : continue; /* \r immediately followed by \n */
1401 :
1402 : /*
1403 : * The de-escaped line should contain an OID followed by exactly
1404 : * one space followed by a path. The path might start with
1405 : * spaces, so don't be too liberal about parsing.
1406 : */
1407 4 : str[i] = '\0';
1408 4 : n = 0;
1409 24 : while (str[n] && str[n] != ' ')
1410 20 : n++;
1411 4 : if (n < 1 || n >= i - 1)
1412 0 : ereport(FATAL,
1413 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1414 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1415 4 : str[n++] = '\0';
1416 :
1417 4 : ti = palloc0(sizeof(tablespaceinfo));
1418 4 : errno = 0;
1419 4 : ti->oid = strtoul(str, &endp, 10);
1420 4 : if (*endp != '\0' || errno == EINVAL || errno == ERANGE)
1421 0 : ereport(FATAL,
1422 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1423 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1424 4 : ti->path = pstrdup(str + n);
1425 4 : *tablespaces = lappend(*tablespaces, ti);
1426 :
1427 4 : i = 0;
1428 4 : continue;
1429 : }
1430 146 : else if (!was_backslash && ch == '\\')
1431 0 : was_backslash = true;
1432 : else
1433 : {
1434 146 : if (i < sizeof(str) - 1)
1435 146 : str[i++] = ch;
1436 146 : was_backslash = false;
1437 : }
1438 : }
1439 :
1440 4 : if (i != 0 || was_backslash) /* last line not terminated? */
1441 0 : ereport(FATAL,
1442 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1443 : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1444 :
1445 4 : if (ferror(lfp) || FreeFile(lfp))
1446 0 : ereport(FATAL,
1447 : (errcode_for_file_access(),
1448 : errmsg("could not read file \"%s\": %m",
1449 : TABLESPACE_MAP)));
1450 :
1451 4 : return true;
1452 : }
1453 :
1454 : /*
1455 : * Finish WAL recovery.
1456 : *
1457 : * This does not close the 'xlogreader' yet, because in some cases the caller
1458 : * still wants to re-read the last checkpoint record by calling
1459 : * ReadCheckpointRecord().
1460 : *
1461 : * Returns the position of the last valid or applied record, after which new
1462 : * WAL should be appended, information about why recovery was ended, and some
1463 : * other things. See the EndOfWalRecoveryInfo struct for details.
1464 : */
1465 : EndOfWalRecoveryInfo *
1466 1726 : FinishWalRecovery(void)
1467 : {
1468 1726 : EndOfWalRecoveryInfo *result = palloc(sizeof(EndOfWalRecoveryInfo));
1469 : XLogRecPtr lastRec;
1470 : TimeLineID lastRecTLI;
1471 : XLogRecPtr endOfLog;
1472 :
1473 : /*
1474 : * Kill WAL receiver, if it's still running, before we continue to write
1475 : * the startup checkpoint and aborted-contrecord records. It will trump
1476 : * over these records and subsequent ones if it's still alive when we
1477 : * start writing WAL.
1478 : */
1479 1726 : XLogShutdownWalRcv();
1480 :
1481 : /*
1482 : * Shutdown the slot sync worker to drop any temporary slots acquired by
1483 : * it and to prevent it from keep trying to fetch the failover slots.
1484 : *
1485 : * We do not update the 'synced' column in 'pg_replication_slots' system
1486 : * view from true to false here, as any failed update could leave 'synced'
1487 : * column false for some slots. This could cause issues during slot sync
1488 : * after restarting the server as a standby. While updating the 'synced'
1489 : * column after switching to the new timeline is an option, it does not
1490 : * simplify the handling for the 'synced' column. Therefore, we retain the
1491 : * 'synced' column as true after promotion as it may provide useful
1492 : * information about the slot origin.
1493 : */
1494 1726 : ShutDownSlotSync();
1495 :
1496 : /*
1497 : * We are now done reading the xlog from stream. Turn off streaming
1498 : * recovery to force fetching the files (which would be required at end of
1499 : * recovery, e.g., timeline history file) from archive or pg_wal.
1500 : *
1501 : * Note that standby mode must be turned off after killing WAL receiver,
1502 : * i.e., calling XLogShutdownWalRcv().
1503 : */
1504 : Assert(!WalRcvStreaming());
1505 1726 : StandbyMode = false;
1506 :
1507 : /*
1508 : * Determine where to start writing WAL next.
1509 : *
1510 : * Re-fetch the last valid or last applied record, so we can identify the
1511 : * exact endpoint of what we consider the valid portion of WAL. There may
1512 : * be an incomplete continuation record after that, in which case
1513 : * 'abortedRecPtr' and 'missingContrecPtr' are set and the caller will
1514 : * write a special OVERWRITE_CONTRECORD message to mark that the rest of
1515 : * it is intentionally missing. See CreateOverwriteContrecordRecord().
1516 : *
1517 : * An important side-effect of this is to load the last page into
1518 : * xlogreader. The caller uses it to initialize the WAL for writing.
1519 : */
1520 1726 : if (!InRecovery)
1521 : {
1522 1416 : lastRec = CheckPointLoc;
1523 1416 : lastRecTLI = CheckPointTLI;
1524 : }
1525 : else
1526 : {
1527 310 : lastRec = XLogRecoveryCtl->lastReplayedReadRecPtr;
1528 310 : lastRecTLI = XLogRecoveryCtl->lastReplayedTLI;
1529 : }
1530 1726 : XLogPrefetcherBeginRead(xlogprefetcher, lastRec);
1531 1726 : (void) ReadRecord(xlogprefetcher, PANIC, false, lastRecTLI);
1532 1726 : endOfLog = xlogreader->EndRecPtr;
1533 :
1534 : /*
1535 : * Remember the TLI in the filename of the XLOG segment containing the
1536 : * end-of-log. It could be different from the timeline that endOfLog
1537 : * nominally belongs to, if there was a timeline switch in that segment,
1538 : * and we were reading the old WAL from a segment belonging to a higher
1539 : * timeline.
1540 : */
1541 1726 : result->endOfLogTLI = xlogreader->seg.ws_tli;
1542 :
1543 1726 : if (ArchiveRecoveryRequested)
1544 : {
1545 : /*
1546 : * We are no longer in archive recovery state.
1547 : *
1548 : * We are now done reading the old WAL. Turn off archive fetching if
1549 : * it was active.
1550 : */
1551 : Assert(InArchiveRecovery);
1552 98 : InArchiveRecovery = false;
1553 :
1554 : /*
1555 : * If the ending log segment is still open, close it (to avoid
1556 : * problems on Windows with trying to rename or delete an open file).
1557 : */
1558 98 : if (readFile >= 0)
1559 : {
1560 98 : close(readFile);
1561 98 : readFile = -1;
1562 : }
1563 : }
1564 :
1565 : /*
1566 : * Copy the last partial block to the caller, for initializing the WAL
1567 : * buffer for appending new WAL.
1568 : */
1569 1726 : if (endOfLog % XLOG_BLCKSZ != 0)
1570 : {
1571 : char *page;
1572 : int len;
1573 : XLogRecPtr pageBeginPtr;
1574 :
1575 1694 : pageBeginPtr = endOfLog - (endOfLog % XLOG_BLCKSZ);
1576 : Assert(readOff == XLogSegmentOffset(pageBeginPtr, wal_segment_size));
1577 :
1578 : /* Copy the valid part of the last block */
1579 1694 : len = endOfLog % XLOG_BLCKSZ;
1580 1694 : page = palloc(len);
1581 1694 : memcpy(page, xlogreader->readBuf, len);
1582 :
1583 1694 : result->lastPageBeginPtr = pageBeginPtr;
1584 1694 : result->lastPage = page;
1585 : }
1586 : else
1587 : {
1588 : /* There is no partial block to copy. */
1589 32 : result->lastPageBeginPtr = endOfLog;
1590 32 : result->lastPage = NULL;
1591 : }
1592 :
1593 : /*
1594 : * Create a comment for the history file to explain why and where timeline
1595 : * changed.
1596 : */
1597 1726 : result->recoveryStopReason = getRecoveryStopReason();
1598 :
1599 1726 : result->lastRec = lastRec;
1600 1726 : result->lastRecTLI = lastRecTLI;
1601 1726 : result->endOfLog = endOfLog;
1602 :
1603 1726 : result->abortedRecPtr = abortedRecPtr;
1604 1726 : result->missingContrecPtr = missingContrecPtr;
1605 :
1606 1726 : result->standby_signal_file_found = standby_signal_file_found;
1607 1726 : result->recovery_signal_file_found = recovery_signal_file_found;
1608 :
1609 1726 : return result;
1610 : }
1611 :
1612 : /*
1613 : * Clean up the WAL reader and leftovers from restoring WAL from archive
1614 : */
1615 : void
1616 1726 : ShutdownWalRecovery(void)
1617 : {
1618 : char recoveryPath[MAXPGPATH];
1619 :
1620 : /* Final update of pg_stat_recovery_prefetch. */
1621 1726 : XLogPrefetcherComputeStats(xlogprefetcher);
1622 :
1623 : /* Shut down xlogreader */
1624 1726 : if (readFile >= 0)
1625 : {
1626 1628 : close(readFile);
1627 1628 : readFile = -1;
1628 : }
1629 1726 : pfree(xlogreader->private_data);
1630 1726 : XLogReaderFree(xlogreader);
1631 1726 : XLogPrefetcherFree(xlogprefetcher);
1632 :
1633 1726 : if (ArchiveRecoveryRequested)
1634 : {
1635 : /*
1636 : * Since there might be a partial WAL segment named RECOVERYXLOG, get
1637 : * rid of it.
1638 : */
1639 98 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYXLOG");
1640 98 : unlink(recoveryPath); /* ignore any error */
1641 :
1642 : /* Get rid of any remaining recovered timeline-history file, too */
1643 98 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYHISTORY");
1644 98 : unlink(recoveryPath); /* ignore any error */
1645 : }
1646 :
1647 : /*
1648 : * We don't need the latch anymore. It's not strictly necessary to disown
1649 : * it, but let's do it for the sake of tidiness.
1650 : */
1651 1726 : if (ArchiveRecoveryRequested)
1652 98 : DisownLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
1653 1726 : }
1654 :
1655 : /*
1656 : * Perform WAL recovery.
1657 : *
1658 : * If the system was shut down cleanly, this is never called.
1659 : */
1660 : void
1661 426 : PerformWalRecovery(void)
1662 : {
1663 : XLogRecord *record;
1664 426 : bool reachedRecoveryTarget = false;
1665 : TimeLineID replayTLI;
1666 :
1667 : /*
1668 : * Initialize shared variables for tracking progress of WAL replay, as if
1669 : * we had just replayed the record before the REDO location (or the
1670 : * checkpoint record itself, if it's a shutdown checkpoint).
1671 : */
1672 426 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
1673 426 : if (RedoStartLSN < CheckPointLoc)
1674 : {
1675 224 : XLogRecoveryCtl->lastReplayedReadRecPtr = InvalidXLogRecPtr;
1676 224 : XLogRecoveryCtl->lastReplayedEndRecPtr = RedoStartLSN;
1677 224 : XLogRecoveryCtl->lastReplayedTLI = RedoStartTLI;
1678 : }
1679 : else
1680 : {
1681 202 : XLogRecoveryCtl->lastReplayedReadRecPtr = xlogreader->ReadRecPtr;
1682 202 : XLogRecoveryCtl->lastReplayedEndRecPtr = xlogreader->EndRecPtr;
1683 202 : XLogRecoveryCtl->lastReplayedTLI = CheckPointTLI;
1684 : }
1685 426 : XLogRecoveryCtl->replayEndRecPtr = XLogRecoveryCtl->lastReplayedEndRecPtr;
1686 426 : XLogRecoveryCtl->replayEndTLI = XLogRecoveryCtl->lastReplayedTLI;
1687 426 : XLogRecoveryCtl->recoveryLastXTime = 0;
1688 426 : XLogRecoveryCtl->currentChunkStartTime = 0;
1689 426 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_NOT_PAUSED;
1690 426 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
1691 :
1692 : /* Also ensure XLogReceiptTime has a sane value */
1693 426 : XLogReceiptTime = GetCurrentTimestamp();
1694 :
1695 : /*
1696 : * Let postmaster know we've started redo now, so that it can launch the
1697 : * archiver if necessary.
1698 : */
1699 426 : if (IsUnderPostmaster)
1700 408 : SendPostmasterSignal(PMSIGNAL_RECOVERY_STARTED);
1701 :
1702 : /*
1703 : * Allow read-only connections immediately if we're consistent already.
1704 : */
1705 426 : CheckRecoveryConsistency();
1706 :
1707 : /*
1708 : * Find the first record that logically follows the checkpoint --- it
1709 : * might physically precede it, though.
1710 : */
1711 426 : if (RedoStartLSN < CheckPointLoc)
1712 : {
1713 : /* back up to find the record */
1714 224 : replayTLI = RedoStartTLI;
1715 224 : XLogPrefetcherBeginRead(xlogprefetcher, RedoStartLSN);
1716 224 : record = ReadRecord(xlogprefetcher, PANIC, false, replayTLI);
1717 :
1718 : /*
1719 : * If a checkpoint record's redo pointer points back to an earlier
1720 : * LSN, the record at that LSN should be an XLOG_CHECKPOINT_REDO
1721 : * record.
1722 : */
1723 224 : if (record->xl_rmid != RM_XLOG_ID ||
1724 224 : (record->xl_info & ~XLR_INFO_MASK) != XLOG_CHECKPOINT_REDO)
1725 0 : ereport(FATAL,
1726 : errmsg("unexpected record type found at redo point %X/%08X",
1727 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr)));
1728 : }
1729 : else
1730 : {
1731 : /* just have to read next record after CheckPoint */
1732 : Assert(xlogreader->ReadRecPtr == CheckPointLoc);
1733 202 : replayTLI = CheckPointTLI;
1734 202 : record = ReadRecord(xlogprefetcher, LOG, false, replayTLI);
1735 : }
1736 :
1737 426 : if (record != NULL)
1738 : {
1739 : TimestampTz xtime;
1740 : PGRUsage ru0;
1741 :
1742 408 : pg_rusage_init(&ru0);
1743 :
1744 408 : InRedo = true;
1745 :
1746 408 : RmgrStartup();
1747 :
1748 408 : ereport(LOG,
1749 : errmsg("redo starts at %X/%08X",
1750 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr)));
1751 :
1752 : /* Prepare to report progress of the redo phase. */
1753 408 : if (!StandbyMode)
1754 214 : begin_startup_progress_phase();
1755 :
1756 : /*
1757 : * main redo apply loop
1758 : */
1759 : do
1760 : {
1761 5473488 : if (!StandbyMode)
1762 527202 : ereport_startup_progress("redo in progress, elapsed time: %ld.%02d s, current LSN: %X/%08X",
1763 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr));
1764 :
1765 : #ifdef WAL_DEBUG
1766 : if (XLOG_DEBUG)
1767 : {
1768 : StringInfoData buf;
1769 :
1770 : initStringInfo(&buf);
1771 : appendStringInfo(&buf, "REDO @ %X/%08X; LSN %X/%08X: ",
1772 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr),
1773 : LSN_FORMAT_ARGS(xlogreader->EndRecPtr));
1774 : xlog_outrec(&buf, xlogreader);
1775 : appendStringInfoString(&buf, " - ");
1776 : xlog_outdesc(&buf, xlogreader);
1777 : elog(LOG, "%s", buf.data);
1778 : pfree(buf.data);
1779 : }
1780 : #endif
1781 :
1782 : /* Handle interrupt signals of startup process */
1783 5473488 : ProcessStartupProcInterrupts();
1784 :
1785 : /*
1786 : * Pause WAL replay, if requested by a hot-standby session via
1787 : * SetRecoveryPause().
1788 : *
1789 : * Note that we intentionally don't take the info_lck spinlock
1790 : * here. We might therefore read a slightly stale value of the
1791 : * recoveryPause flag, but it can't be very stale (no worse than
1792 : * the last spinlock we did acquire). Since a pause request is a
1793 : * pretty asynchronous thing anyway, possibly responding to it one
1794 : * WAL record later than we otherwise would is a minor issue, so
1795 : * it doesn't seem worth adding another spinlock cycle to prevent
1796 : * that.
1797 : */
1798 5473488 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
1799 : RECOVERY_NOT_PAUSED)
1800 0 : recoveryPausesHere(false);
1801 :
1802 : /*
1803 : * Have we reached our recovery target?
1804 : */
1805 5473488 : if (recoveryStopsBefore(xlogreader))
1806 : {
1807 4 : reachedRecoveryTarget = true;
1808 4 : break;
1809 : }
1810 :
1811 : /*
1812 : * If we've been asked to lag the primary, wait on latch until
1813 : * enough time has passed.
1814 : */
1815 5473484 : if (recoveryApplyDelay(xlogreader))
1816 : {
1817 : /*
1818 : * We test for paused recovery again here. If user sets
1819 : * delayed apply, it may be because they expect to pause
1820 : * recovery in case of problems, so we must test again here
1821 : * otherwise pausing during the delay-wait wouldn't work.
1822 : */
1823 0 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
1824 : RECOVERY_NOT_PAUSED)
1825 0 : recoveryPausesHere(false);
1826 : }
1827 :
1828 : /*
1829 : * Apply the record
1830 : */
1831 5473484 : ApplyWalRecord(xlogreader, record, &replayTLI);
1832 :
1833 : /* Exit loop if we reached inclusive recovery target */
1834 5473480 : if (recoveryStopsAfter(xlogreader))
1835 : {
1836 10 : reachedRecoveryTarget = true;
1837 10 : break;
1838 : }
1839 :
1840 : /* Else, try to fetch the next WAL record */
1841 5473470 : record = ReadRecord(xlogprefetcher, LOG, false, replayTLI);
1842 5473360 : } while (record != NULL);
1843 :
1844 : /*
1845 : * end of main redo apply loop
1846 : */
1847 :
1848 294 : if (reachedRecoveryTarget)
1849 : {
1850 14 : if (!reachedConsistency)
1851 0 : ereport(FATAL,
1852 : (errmsg("requested recovery stop point is before consistent recovery point")));
1853 :
1854 : /*
1855 : * This is the last point where we can restart recovery with a new
1856 : * recovery target, if we shutdown and begin again. After this,
1857 : * Resource Managers may choose to do permanent corrective actions
1858 : * at end of recovery.
1859 : */
1860 14 : switch (recoveryTargetAction)
1861 : {
1862 0 : case RECOVERY_TARGET_ACTION_SHUTDOWN:
1863 :
1864 : /*
1865 : * exit with special return code to request shutdown of
1866 : * postmaster. Log messages issued from postmaster.
1867 : */
1868 0 : proc_exit(3);
1869 :
1870 2 : case RECOVERY_TARGET_ACTION_PAUSE:
1871 2 : SetRecoveryPause(true);
1872 2 : recoveryPausesHere(true);
1873 :
1874 : /* drop into promote */
1875 :
1876 14 : case RECOVERY_TARGET_ACTION_PROMOTE:
1877 14 : break;
1878 : }
1879 : }
1880 :
1881 294 : RmgrCleanup();
1882 :
1883 294 : ereport(LOG,
1884 : errmsg("redo done at %X/%08X system usage: %s",
1885 : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr),
1886 : pg_rusage_show(&ru0)));
1887 294 : xtime = GetLatestXTime();
1888 294 : if (xtime)
1889 70 : ereport(LOG,
1890 : (errmsg("last completed transaction was at log time %s",
1891 : timestamptz_to_str(xtime))));
1892 :
1893 294 : InRedo = false;
1894 : }
1895 : else
1896 : {
1897 : /* there are no WAL records following the checkpoint */
1898 18 : ereport(LOG,
1899 : (errmsg("redo is not required")));
1900 : }
1901 :
1902 : /*
1903 : * This check is intentionally after the above log messages that indicate
1904 : * how far recovery went.
1905 : */
1906 312 : if (ArchiveRecoveryRequested &&
1907 100 : recoveryTarget != RECOVERY_TARGET_UNSET &&
1908 16 : !reachedRecoveryTarget)
1909 2 : ereport(FATAL,
1910 : (errcode(ERRCODE_CONFIG_FILE_ERROR),
1911 : errmsg("recovery ended before configured recovery target was reached")));
1912 310 : }
1913 :
1914 : /*
1915 : * Subroutine of PerformWalRecovery, to apply one WAL record.
1916 : */
1917 : static void
1918 5473484 : ApplyWalRecord(XLogReaderState *xlogreader, XLogRecord *record, TimeLineID *replayTLI)
1919 : {
1920 : ErrorContextCallback errcallback;
1921 5473484 : bool switchedTLI = false;
1922 :
1923 : /* Setup error traceback support for ereport() */
1924 5473484 : errcallback.callback = rm_redo_error_callback;
1925 5473484 : errcallback.arg = xlogreader;
1926 5473484 : errcallback.previous = error_context_stack;
1927 5473484 : error_context_stack = &errcallback;
1928 :
1929 : /*
1930 : * TransamVariables->nextXid must be beyond record's xid.
1931 : */
1932 5473484 : AdvanceNextFullTransactionIdPastXid(record->xl_xid);
1933 :
1934 : /*
1935 : * Before replaying this record, check if this record causes the current
1936 : * timeline to change. The record is already considered to be part of the
1937 : * new timeline, so we update replayTLI before replaying it. That's
1938 : * important so that replayEndTLI, which is recorded as the minimum
1939 : * recovery point's TLI if recovery stops after this record, is set
1940 : * correctly.
1941 : */
1942 5473484 : if (record->xl_rmid == RM_XLOG_ID)
1943 : {
1944 86220 : TimeLineID newReplayTLI = *replayTLI;
1945 86220 : TimeLineID prevReplayTLI = *replayTLI;
1946 86220 : uint8 info = record->xl_info & ~XLR_INFO_MASK;
1947 :
1948 86220 : if (info == XLOG_CHECKPOINT_SHUTDOWN)
1949 : {
1950 : CheckPoint checkPoint;
1951 :
1952 68 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
1953 68 : newReplayTLI = checkPoint.ThisTimeLineID;
1954 68 : prevReplayTLI = checkPoint.PrevTimeLineID;
1955 : }
1956 86152 : else if (info == XLOG_END_OF_RECOVERY)
1957 : {
1958 : xl_end_of_recovery xlrec;
1959 :
1960 20 : memcpy(&xlrec, XLogRecGetData(xlogreader), sizeof(xl_end_of_recovery));
1961 20 : newReplayTLI = xlrec.ThisTimeLineID;
1962 20 : prevReplayTLI = xlrec.PrevTimeLineID;
1963 : }
1964 :
1965 86220 : if (newReplayTLI != *replayTLI)
1966 : {
1967 : /* Check that it's OK to switch to this TLI */
1968 22 : checkTimeLineSwitch(xlogreader->EndRecPtr,
1969 : newReplayTLI, prevReplayTLI, *replayTLI);
1970 :
1971 : /* Following WAL records should be run with new TLI */
1972 22 : *replayTLI = newReplayTLI;
1973 22 : switchedTLI = true;
1974 : }
1975 : }
1976 :
1977 : /*
1978 : * Update shared replayEndRecPtr before replaying this record, so that
1979 : * XLogFlush will update minRecoveryPoint correctly.
1980 : */
1981 5473484 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
1982 5473484 : XLogRecoveryCtl->replayEndRecPtr = xlogreader->EndRecPtr;
1983 5473484 : XLogRecoveryCtl->replayEndTLI = *replayTLI;
1984 5473484 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
1985 :
1986 : /*
1987 : * If we are attempting to enter Hot Standby mode, process XIDs we see
1988 : */
1989 5473484 : if (standbyState >= STANDBY_INITIALIZED &&
1990 4986000 : TransactionIdIsValid(record->xl_xid))
1991 4878958 : RecordKnownAssignedTransactionIds(record->xl_xid);
1992 :
1993 : /*
1994 : * Some XLOG record types that are related to recovery are processed
1995 : * directly here, rather than in xlog_redo()
1996 : */
1997 5473484 : if (record->xl_rmid == RM_XLOG_ID)
1998 86220 : xlogrecovery_redo(xlogreader, *replayTLI);
1999 :
2000 : /* Now apply the WAL record itself */
2001 5473484 : GetRmgr(record->xl_rmid).rm_redo(xlogreader);
2002 :
2003 : /*
2004 : * After redo, check whether the backup pages associated with the WAL
2005 : * record are consistent with the existing pages. This check is done only
2006 : * if consistency check is enabled for this record.
2007 : */
2008 5473480 : if ((record->xl_info & XLR_CHECK_CONSISTENCY) != 0)
2009 4312876 : verifyBackupPageConsistency(xlogreader);
2010 :
2011 : /* Pop the error context stack */
2012 5473480 : error_context_stack = errcallback.previous;
2013 :
2014 : /*
2015 : * Update lastReplayedEndRecPtr after this record has been successfully
2016 : * replayed.
2017 : */
2018 5473480 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
2019 5473480 : XLogRecoveryCtl->lastReplayedReadRecPtr = xlogreader->ReadRecPtr;
2020 5473480 : XLogRecoveryCtl->lastReplayedEndRecPtr = xlogreader->EndRecPtr;
2021 5473480 : XLogRecoveryCtl->lastReplayedTLI = *replayTLI;
2022 5473480 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
2023 :
2024 : /* ------
2025 : * Wakeup walsenders:
2026 : *
2027 : * On the standby, the WAL is flushed first (which will only wake up
2028 : * physical walsenders) and then applied, which will only wake up logical
2029 : * walsenders.
2030 : *
2031 : * Indeed, logical walsenders on standby can't decode and send data until
2032 : * it's been applied.
2033 : *
2034 : * Physical walsenders don't need to be woken up during replay unless
2035 : * cascading replication is allowed and time line change occurred (so that
2036 : * they can notice that they are on a new time line).
2037 : *
2038 : * That's why the wake up conditions are for:
2039 : *
2040 : * - physical walsenders in case of new time line and cascade
2041 : * replication is allowed
2042 : * - logical walsenders in case cascade replication is allowed (could not
2043 : * be created otherwise)
2044 : * ------
2045 : */
2046 5473480 : if (AllowCascadeReplication())
2047 5095252 : WalSndWakeup(switchedTLI, true);
2048 :
2049 : /*
2050 : * If rm_redo called XLogRequestWalReceiverReply, then we wake up the
2051 : * receiver so that it notices the updated lastReplayedEndRecPtr and sends
2052 : * a reply to the primary.
2053 : */
2054 5473480 : if (doRequestWalReceiverReply)
2055 : {
2056 4 : doRequestWalReceiverReply = false;
2057 4 : WalRcvForceReply();
2058 : }
2059 :
2060 : /* Allow read-only connections if we're consistent now */
2061 5473480 : CheckRecoveryConsistency();
2062 :
2063 : /* Is this a timeline switch? */
2064 5473480 : if (switchedTLI)
2065 : {
2066 : /*
2067 : * Before we continue on the new timeline, clean up any (possibly
2068 : * bogus) future WAL segments on the old timeline.
2069 : */
2070 22 : RemoveNonParentXlogFiles(xlogreader->EndRecPtr, *replayTLI);
2071 :
2072 : /* Reset the prefetcher. */
2073 22 : XLogPrefetchReconfigure();
2074 : }
2075 5473480 : }
2076 :
2077 : /*
2078 : * Some XLOG RM record types that are directly related to WAL recovery are
2079 : * handled here rather than in the xlog_redo()
2080 : */
2081 : static void
2082 86220 : xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI)
2083 : {
2084 86220 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2085 86220 : XLogRecPtr lsn = record->EndRecPtr;
2086 :
2087 : Assert(XLogRecGetRmid(record) == RM_XLOG_ID);
2088 :
2089 86220 : if (info == XLOG_OVERWRITE_CONTRECORD)
2090 : {
2091 : /* Verify the payload of a XLOG_OVERWRITE_CONTRECORD record. */
2092 : xl_overwrite_contrecord xlrec;
2093 :
2094 2 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_overwrite_contrecord));
2095 2 : if (xlrec.overwritten_lsn != record->overwrittenRecPtr)
2096 0 : elog(FATAL, "mismatching overwritten LSN %X/%08X -> %X/%08X",
2097 : LSN_FORMAT_ARGS(xlrec.overwritten_lsn),
2098 : LSN_FORMAT_ARGS(record->overwrittenRecPtr));
2099 :
2100 : /* We have safely skipped the aborted record */
2101 2 : abortedRecPtr = InvalidXLogRecPtr;
2102 2 : missingContrecPtr = InvalidXLogRecPtr;
2103 :
2104 2 : ereport(LOG,
2105 : errmsg("successfully skipped missing contrecord at %X/%08X, overwritten at %s",
2106 : LSN_FORMAT_ARGS(xlrec.overwritten_lsn),
2107 : timestamptz_to_str(xlrec.overwrite_time)));
2108 :
2109 : /* Verifying the record should only happen once */
2110 2 : record->overwrittenRecPtr = InvalidXLogRecPtr;
2111 : }
2112 86218 : else if (info == XLOG_BACKUP_END)
2113 : {
2114 : XLogRecPtr startpoint;
2115 :
2116 170 : memcpy(&startpoint, XLogRecGetData(record), sizeof(startpoint));
2117 :
2118 170 : if (backupStartPoint == startpoint)
2119 : {
2120 : /*
2121 : * We have reached the end of base backup, the point where
2122 : * pg_backup_stop() was done. The data on disk is now consistent
2123 : * (assuming we have also reached minRecoveryPoint). Set
2124 : * backupEndPoint to the current LSN, so that the next call to
2125 : * CheckRecoveryConsistency() will notice it and do the
2126 : * end-of-backup processing.
2127 : */
2128 138 : elog(DEBUG1, "end of backup record reached");
2129 :
2130 138 : backupEndPoint = lsn;
2131 : }
2132 : else
2133 32 : elog(DEBUG1, "saw end-of-backup record for backup starting at %X/%08X, waiting for %X/%08X",
2134 : LSN_FORMAT_ARGS(startpoint), LSN_FORMAT_ARGS(backupStartPoint));
2135 : }
2136 86220 : }
2137 :
2138 : /*
2139 : * Verify that, in non-test mode, ./pg_tblspc doesn't contain any real
2140 : * directories.
2141 : *
2142 : * Replay of database creation XLOG records for databases that were later
2143 : * dropped can create fake directories in pg_tblspc. By the time consistency
2144 : * is reached these directories should have been removed; here we verify
2145 : * that this did indeed happen. This is to be called at the point where
2146 : * consistent state is reached.
2147 : *
2148 : * allow_in_place_tablespaces turns the PANIC into a WARNING, which is
2149 : * useful for testing purposes, and also allows for an escape hatch in case
2150 : * things go south.
2151 : */
2152 : static void
2153 226 : CheckTablespaceDirectory(void)
2154 : {
2155 : DIR *dir;
2156 : struct dirent *de;
2157 :
2158 226 : dir = AllocateDir(PG_TBLSPC_DIR);
2159 692 : while ((de = ReadDir(dir, PG_TBLSPC_DIR)) != NULL)
2160 : {
2161 : char path[MAXPGPATH + sizeof(PG_TBLSPC_DIR)];
2162 :
2163 : /* Skip entries of non-oid names */
2164 466 : if (strspn(de->d_name, "0123456789") != strlen(de->d_name))
2165 452 : continue;
2166 :
2167 14 : snprintf(path, sizeof(path), "%s/%s", PG_TBLSPC_DIR, de->d_name);
2168 :
2169 14 : if (get_dirent_type(path, de, false, ERROR) != PGFILETYPE_LNK)
2170 8 : ereport(allow_in_place_tablespaces ? WARNING : PANIC,
2171 : (errcode(ERRCODE_DATA_CORRUPTED),
2172 : errmsg("unexpected directory entry \"%s\" found in %s",
2173 : de->d_name, PG_TBLSPC_DIR),
2174 : errdetail("All directory entries in %s/ should be symbolic links.",
2175 : PG_TBLSPC_DIR),
2176 : errhint("Remove those directories, or set \"allow_in_place_tablespaces\" to ON transiently to let recovery complete.")));
2177 : }
2178 226 : }
2179 :
2180 : /*
2181 : * Checks if recovery has reached a consistent state. When consistency is
2182 : * reached and we have a valid starting standby snapshot, tell postmaster
2183 : * that it can start accepting read-only connections.
2184 : */
2185 : static void
2186 5473910 : CheckRecoveryConsistency(void)
2187 : {
2188 : XLogRecPtr lastReplayedEndRecPtr;
2189 : TimeLineID lastReplayedTLI;
2190 :
2191 : /*
2192 : * During crash recovery, we don't reach a consistent state until we've
2193 : * replayed all the WAL.
2194 : */
2195 5473910 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
2196 516982 : return;
2197 :
2198 : Assert(InArchiveRecovery);
2199 :
2200 : /*
2201 : * assume that we are called in the startup process, and hence don't need
2202 : * a lock to read lastReplayedEndRecPtr
2203 : */
2204 4956928 : lastReplayedEndRecPtr = XLogRecoveryCtl->lastReplayedEndRecPtr;
2205 4956928 : lastReplayedTLI = XLogRecoveryCtl->lastReplayedTLI;
2206 :
2207 : /*
2208 : * Have we reached the point where our base backup was completed?
2209 : */
2210 4956928 : if (!XLogRecPtrIsInvalid(backupEndPoint) &&
2211 204 : backupEndPoint <= lastReplayedEndRecPtr)
2212 : {
2213 142 : XLogRecPtr saveBackupStartPoint = backupStartPoint;
2214 142 : XLogRecPtr saveBackupEndPoint = backupEndPoint;
2215 :
2216 142 : elog(DEBUG1, "end of backup reached");
2217 :
2218 : /*
2219 : * We have reached the end of base backup, as indicated by pg_control.
2220 : * Update the control file accordingly.
2221 : */
2222 142 : ReachedEndOfBackup(lastReplayedEndRecPtr, lastReplayedTLI);
2223 142 : backupStartPoint = InvalidXLogRecPtr;
2224 142 : backupEndPoint = InvalidXLogRecPtr;
2225 142 : backupEndRequired = false;
2226 :
2227 142 : ereport(LOG,
2228 : errmsg("completed backup recovery with redo LSN %X/%08X and end LSN %X/%08X",
2229 : LSN_FORMAT_ARGS(saveBackupStartPoint),
2230 : LSN_FORMAT_ARGS(saveBackupEndPoint)));
2231 : }
2232 :
2233 : /*
2234 : * Have we passed our safe starting point? Note that minRecoveryPoint is
2235 : * known to be incorrectly set if recovering from a backup, until the
2236 : * XLOG_BACKUP_END arrives to advise us of the correct minRecoveryPoint.
2237 : * All we know prior to that is that we're not consistent yet.
2238 : */
2239 4956928 : if (!reachedConsistency && !backupEndRequired &&
2240 15202 : minRecoveryPoint <= lastReplayedEndRecPtr)
2241 : {
2242 : /*
2243 : * Check to see if the XLOG sequence contained any unresolved
2244 : * references to uninitialized pages.
2245 : */
2246 226 : XLogCheckInvalidPages();
2247 :
2248 : /*
2249 : * Check that pg_tblspc doesn't contain any real directories. Replay
2250 : * of Database/CREATE_* records may have created fictitious tablespace
2251 : * directories that should have been removed by the time consistency
2252 : * was reached.
2253 : */
2254 226 : CheckTablespaceDirectory();
2255 :
2256 226 : reachedConsistency = true;
2257 226 : SendPostmasterSignal(PMSIGNAL_RECOVERY_CONSISTENT);
2258 226 : ereport(LOG,
2259 : errmsg("consistent recovery state reached at %X/%08X",
2260 : LSN_FORMAT_ARGS(lastReplayedEndRecPtr)));
2261 : }
2262 :
2263 : /*
2264 : * Have we got a valid starting snapshot that will allow queries to be
2265 : * run? If so, we can tell postmaster that the database is consistent now,
2266 : * enabling connections.
2267 : */
2268 4956928 : if (standbyState == STANDBY_SNAPSHOT_READY &&
2269 4956484 : !LocalHotStandbyActive &&
2270 210 : reachedConsistency &&
2271 : IsUnderPostmaster)
2272 : {
2273 210 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
2274 210 : XLogRecoveryCtl->SharedHotStandbyActive = true;
2275 210 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
2276 :
2277 210 : LocalHotStandbyActive = true;
2278 :
2279 210 : SendPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY);
2280 : }
2281 : }
2282 :
2283 : /*
2284 : * Error context callback for errors occurring during rm_redo().
2285 : */
2286 : static void
2287 218 : rm_redo_error_callback(void *arg)
2288 : {
2289 218 : XLogReaderState *record = (XLogReaderState *) arg;
2290 : StringInfoData buf;
2291 :
2292 218 : initStringInfo(&buf);
2293 218 : xlog_outdesc(&buf, record);
2294 218 : xlog_block_info(&buf, record);
2295 :
2296 : /* translator: %s is a WAL record description */
2297 218 : errcontext("WAL redo at %X/%08X for %s",
2298 218 : LSN_FORMAT_ARGS(record->ReadRecPtr),
2299 : buf.data);
2300 :
2301 218 : pfree(buf.data);
2302 218 : }
2303 :
2304 : /*
2305 : * Returns a string describing an XLogRecord, consisting of its identity
2306 : * optionally followed by a colon, a space, and a further description.
2307 : */
2308 : void
2309 218 : xlog_outdesc(StringInfo buf, XLogReaderState *record)
2310 : {
2311 218 : RmgrData rmgr = GetRmgr(XLogRecGetRmid(record));
2312 218 : uint8 info = XLogRecGetInfo(record);
2313 : const char *id;
2314 :
2315 218 : appendStringInfoString(buf, rmgr.rm_name);
2316 218 : appendStringInfoChar(buf, '/');
2317 :
2318 218 : id = rmgr.rm_identify(info);
2319 218 : if (id == NULL)
2320 0 : appendStringInfo(buf, "UNKNOWN (%X): ", info & ~XLR_INFO_MASK);
2321 : else
2322 218 : appendStringInfo(buf, "%s: ", id);
2323 :
2324 218 : rmgr.rm_desc(buf, record);
2325 218 : }
2326 :
2327 : #ifdef WAL_DEBUG
2328 :
2329 : static void
2330 : xlog_outrec(StringInfo buf, XLogReaderState *record)
2331 : {
2332 : appendStringInfo(buf, "prev %X/%08X; xid %u",
2333 : LSN_FORMAT_ARGS(XLogRecGetPrev(record)),
2334 : XLogRecGetXid(record));
2335 :
2336 : appendStringInfo(buf, "; len %u",
2337 : XLogRecGetDataLen(record));
2338 :
2339 : xlog_block_info(buf, record);
2340 : }
2341 : #endif /* WAL_DEBUG */
2342 :
2343 : /*
2344 : * Returns a string giving information about all the blocks in an
2345 : * XLogRecord.
2346 : */
2347 : static void
2348 218 : xlog_block_info(StringInfo buf, XLogReaderState *record)
2349 : {
2350 : int block_id;
2351 :
2352 : /* decode block references */
2353 324 : for (block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
2354 : {
2355 : RelFileLocator rlocator;
2356 : ForkNumber forknum;
2357 : BlockNumber blk;
2358 :
2359 106 : if (!XLogRecGetBlockTagExtended(record, block_id,
2360 : &rlocator, &forknum, &blk, NULL))
2361 0 : continue;
2362 :
2363 106 : if (forknum != MAIN_FORKNUM)
2364 6 : appendStringInfo(buf, "; blkref #%d: rel %u/%u/%u, fork %u, blk %u",
2365 : block_id,
2366 : rlocator.spcOid, rlocator.dbOid,
2367 : rlocator.relNumber,
2368 : forknum,
2369 : blk);
2370 : else
2371 100 : appendStringInfo(buf, "; blkref #%d: rel %u/%u/%u, blk %u",
2372 : block_id,
2373 : rlocator.spcOid, rlocator.dbOid,
2374 : rlocator.relNumber,
2375 : blk);
2376 106 : if (XLogRecHasBlockImage(record, block_id))
2377 64 : appendStringInfoString(buf, " FPW");
2378 : }
2379 218 : }
2380 :
2381 :
2382 : /*
2383 : * Check that it's OK to switch to new timeline during recovery.
2384 : *
2385 : * 'lsn' is the address of the shutdown checkpoint record we're about to
2386 : * replay. (Currently, timeline can only change at a shutdown checkpoint).
2387 : */
2388 : static void
2389 22 : checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI, TimeLineID prevTLI,
2390 : TimeLineID replayTLI)
2391 : {
2392 : /* Check that the record agrees on what the current (old) timeline is */
2393 22 : if (prevTLI != replayTLI)
2394 0 : ereport(PANIC,
2395 : (errmsg("unexpected previous timeline ID %u (current timeline ID %u) in checkpoint record",
2396 : prevTLI, replayTLI)));
2397 :
2398 : /*
2399 : * The new timeline better be in the list of timelines we expect to see,
2400 : * according to the timeline history. It should also not decrease.
2401 : */
2402 22 : if (newTLI < replayTLI || !tliInHistory(newTLI, expectedTLEs))
2403 0 : ereport(PANIC,
2404 : (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
2405 : newTLI, replayTLI)));
2406 :
2407 : /*
2408 : * If we have not yet reached min recovery point, and we're about to
2409 : * switch to a timeline greater than the timeline of the min recovery
2410 : * point: trouble. After switching to the new timeline, we could not
2411 : * possibly visit the min recovery point on the correct timeline anymore.
2412 : * This can happen if there is a newer timeline in the archive that
2413 : * branched before the timeline the min recovery point is on, and you
2414 : * attempt to do PITR to the new timeline.
2415 : */
2416 22 : if (!XLogRecPtrIsInvalid(minRecoveryPoint) &&
2417 18 : lsn < minRecoveryPoint &&
2418 2 : newTLI > minRecoveryPointTLI)
2419 0 : ereport(PANIC,
2420 : errmsg("unexpected timeline ID %u in checkpoint record, before reaching minimum recovery point %X/%08X on timeline %u",
2421 : newTLI,
2422 : LSN_FORMAT_ARGS(minRecoveryPoint),
2423 : minRecoveryPointTLI));
2424 :
2425 : /* Looks good */
2426 22 : }
2427 :
2428 :
2429 : /*
2430 : * Extract timestamp from WAL record.
2431 : *
2432 : * If the record contains a timestamp, returns true, and saves the timestamp
2433 : * in *recordXtime. If the record type has no timestamp, returns false.
2434 : * Currently, only transaction commit/abort records and restore points contain
2435 : * timestamps.
2436 : */
2437 : static bool
2438 85092 : getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime)
2439 : {
2440 85092 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2441 85092 : uint8 xact_info = info & XLOG_XACT_OPMASK;
2442 85092 : uint8 rmid = XLogRecGetRmid(record);
2443 :
2444 85092 : if (rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
2445 : {
2446 4 : *recordXtime = ((xl_restore_point *) XLogRecGetData(record))->rp_time;
2447 4 : return true;
2448 : }
2449 85088 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_COMMIT ||
2450 : xact_info == XLOG_XACT_COMMIT_PREPARED))
2451 : {
2452 77976 : *recordXtime = ((xl_xact_commit *) XLogRecGetData(record))->xact_time;
2453 77976 : return true;
2454 : }
2455 7112 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_ABORT ||
2456 : xact_info == XLOG_XACT_ABORT_PREPARED))
2457 : {
2458 7112 : *recordXtime = ((xl_xact_abort *) XLogRecGetData(record))->xact_time;
2459 7112 : return true;
2460 : }
2461 0 : return false;
2462 : }
2463 :
2464 : /*
2465 : * Checks whether the current buffer page and backup page stored in the
2466 : * WAL record are consistent or not. Before comparing the two pages, a
2467 : * masking can be applied to the pages to ignore certain areas like hint bits,
2468 : * unused space between pd_lower and pd_upper among other things. This
2469 : * function should be called once WAL replay has been completed for a
2470 : * given record.
2471 : */
2472 : static void
2473 4312876 : verifyBackupPageConsistency(XLogReaderState *record)
2474 : {
2475 4312876 : RmgrData rmgr = GetRmgr(XLogRecGetRmid(record));
2476 : RelFileLocator rlocator;
2477 : ForkNumber forknum;
2478 : BlockNumber blkno;
2479 : int block_id;
2480 :
2481 : /* Records with no backup blocks have no need for consistency checks. */
2482 4312876 : if (!XLogRecHasAnyBlockRefs(record))
2483 132 : return;
2484 :
2485 : Assert((XLogRecGetInfo(record) & XLR_CHECK_CONSISTENCY) != 0);
2486 :
2487 8961442 : for (block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
2488 : {
2489 : Buffer buf;
2490 : Page page;
2491 :
2492 4648698 : if (!XLogRecGetBlockTagExtended(record, block_id,
2493 : &rlocator, &forknum, &blkno, NULL))
2494 : {
2495 : /*
2496 : * WAL record doesn't contain a block reference with the given id.
2497 : * Do nothing.
2498 : */
2499 3068 : continue;
2500 : }
2501 :
2502 : Assert(XLogRecHasBlockImage(record, block_id));
2503 :
2504 4645630 : if (XLogRecBlockImageApply(record, block_id))
2505 : {
2506 : /*
2507 : * WAL record has already applied the page, so bypass the
2508 : * consistency check as that would result in comparing the full
2509 : * page stored in the record with itself.
2510 : */
2511 48328 : continue;
2512 : }
2513 :
2514 : /*
2515 : * Read the contents from the current buffer and store it in a
2516 : * temporary page.
2517 : */
2518 4597302 : buf = XLogReadBufferExtended(rlocator, forknum, blkno,
2519 : RBM_NORMAL_NO_LOG,
2520 : InvalidBuffer);
2521 4597302 : if (!BufferIsValid(buf))
2522 0 : continue;
2523 :
2524 4597302 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
2525 4597302 : page = BufferGetPage(buf);
2526 :
2527 : /*
2528 : * Take a copy of the local page where WAL has been applied to have a
2529 : * comparison base before masking it...
2530 : */
2531 4597302 : memcpy(replay_image_masked, page, BLCKSZ);
2532 :
2533 : /* No need for this page anymore now that a copy is in. */
2534 4597302 : UnlockReleaseBuffer(buf);
2535 :
2536 : /*
2537 : * If the block LSN is already ahead of this WAL record, we can't
2538 : * expect contents to match. This can happen if recovery is
2539 : * restarted.
2540 : */
2541 4597302 : if (PageGetLSN(replay_image_masked) > record->EndRecPtr)
2542 0 : continue;
2543 :
2544 : /*
2545 : * Read the contents from the backup copy, stored in WAL record and
2546 : * store it in a temporary page. There is no need to allocate a new
2547 : * page here, a local buffer is fine to hold its contents and a mask
2548 : * can be directly applied on it.
2549 : */
2550 4597302 : if (!RestoreBlockImage(record, block_id, primary_image_masked))
2551 0 : ereport(ERROR,
2552 : (errcode(ERRCODE_INTERNAL_ERROR),
2553 : errmsg_internal("%s", record->errormsg_buf)));
2554 :
2555 : /*
2556 : * If masking function is defined, mask both the primary and replay
2557 : * images
2558 : */
2559 4597302 : if (rmgr.rm_mask != NULL)
2560 : {
2561 4597302 : rmgr.rm_mask(replay_image_masked, blkno);
2562 4597302 : rmgr.rm_mask(primary_image_masked, blkno);
2563 : }
2564 :
2565 : /* Time to compare the primary and replay images. */
2566 4597302 : if (memcmp(replay_image_masked, primary_image_masked, BLCKSZ) != 0)
2567 : {
2568 0 : elog(FATAL,
2569 : "inconsistent page found, rel %u/%u/%u, forknum %u, blkno %u",
2570 : rlocator.spcOid, rlocator.dbOid, rlocator.relNumber,
2571 : forknum, blkno);
2572 : }
2573 : }
2574 : }
2575 :
2576 : /*
2577 : * For point-in-time recovery, this function decides whether we want to
2578 : * stop applying the XLOG before the current record.
2579 : *
2580 : * Returns true if we are stopping, false otherwise. If stopping, some
2581 : * information is saved in recoveryStopXid et al for use in annotating the
2582 : * new timeline's history file.
2583 : */
2584 : static bool
2585 5473488 : recoveryStopsBefore(XLogReaderState *record)
2586 : {
2587 5473488 : bool stopsHere = false;
2588 : uint8 xact_info;
2589 : bool isCommit;
2590 5473488 : TimestampTz recordXtime = 0;
2591 : TransactionId recordXid;
2592 :
2593 : /*
2594 : * Ignore recovery target settings when not in archive recovery (meaning
2595 : * we are in crash recovery).
2596 : */
2597 5473488 : if (!ArchiveRecoveryRequested)
2598 487456 : return false;
2599 :
2600 : /* Check if we should stop as soon as reaching consistency */
2601 4986032 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
2602 : {
2603 0 : ereport(LOG,
2604 : (errmsg("recovery stopping after reaching consistency")));
2605 :
2606 0 : recoveryStopAfter = false;
2607 0 : recoveryStopXid = InvalidTransactionId;
2608 0 : recoveryStopLSN = InvalidXLogRecPtr;
2609 0 : recoveryStopTime = 0;
2610 0 : recoveryStopName[0] = '\0';
2611 0 : return true;
2612 : }
2613 :
2614 : /* Check if target LSN has been reached */
2615 4986032 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
2616 16904 : !recoveryTargetInclusive &&
2617 840 : record->ReadRecPtr >= recoveryTargetLSN)
2618 : {
2619 4 : recoveryStopAfter = false;
2620 4 : recoveryStopXid = InvalidTransactionId;
2621 4 : recoveryStopLSN = record->ReadRecPtr;
2622 4 : recoveryStopTime = 0;
2623 4 : recoveryStopName[0] = '\0';
2624 4 : ereport(LOG,
2625 : errmsg("recovery stopping before WAL location (LSN) \"%X/%08X\"",
2626 : LSN_FORMAT_ARGS(recoveryStopLSN)));
2627 4 : return true;
2628 : }
2629 :
2630 : /* Otherwise we only consider stopping before COMMIT or ABORT records. */
2631 4986028 : if (XLogRecGetRmid(record) != RM_XACT_ID)
2632 4942912 : return false;
2633 :
2634 43116 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
2635 :
2636 43116 : if (xact_info == XLOG_XACT_COMMIT)
2637 : {
2638 38942 : isCommit = true;
2639 38942 : recordXid = XLogRecGetXid(record);
2640 : }
2641 4174 : else if (xact_info == XLOG_XACT_COMMIT_PREPARED)
2642 : {
2643 48 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
2644 : xl_xact_parsed_commit parsed;
2645 :
2646 48 : isCommit = true;
2647 48 : ParseCommitRecord(XLogRecGetInfo(record),
2648 : xlrec,
2649 : &parsed);
2650 48 : recordXid = parsed.twophase_xid;
2651 : }
2652 4126 : else if (xact_info == XLOG_XACT_ABORT)
2653 : {
2654 3534 : isCommit = false;
2655 3534 : recordXid = XLogRecGetXid(record);
2656 : }
2657 592 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
2658 : {
2659 22 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
2660 : xl_xact_parsed_abort parsed;
2661 :
2662 22 : isCommit = false;
2663 22 : ParseAbortRecord(XLogRecGetInfo(record),
2664 : xlrec,
2665 : &parsed);
2666 22 : recordXid = parsed.twophase_xid;
2667 : }
2668 : else
2669 570 : return false;
2670 :
2671 42546 : if (recoveryTarget == RECOVERY_TARGET_XID && !recoveryTargetInclusive)
2672 : {
2673 : /*
2674 : * There can be only one transaction end record with this exact
2675 : * transactionid
2676 : *
2677 : * when testing for an xid, we MUST test for equality only, since
2678 : * transactions are numbered in the order they start, not the order
2679 : * they complete. A higher numbered xid will complete before you about
2680 : * 50% of the time...
2681 : */
2682 0 : stopsHere = (recordXid == recoveryTargetXid);
2683 : }
2684 :
2685 : /*
2686 : * Note: we must fetch recordXtime regardless of recoveryTarget setting.
2687 : * We don't expect getRecordTimestamp ever to fail, since we already know
2688 : * this is a commit or abort record; but test its result anyway.
2689 : */
2690 42546 : if (getRecordTimestamp(record, &recordXtime) &&
2691 42546 : recoveryTarget == RECOVERY_TARGET_TIME)
2692 : {
2693 : /*
2694 : * There can be many transactions that share the same commit time, so
2695 : * we stop after the last one, if we are inclusive, or stop at the
2696 : * first one if we are exclusive
2697 : */
2698 0 : if (recoveryTargetInclusive)
2699 0 : stopsHere = (recordXtime > recoveryTargetTime);
2700 : else
2701 0 : stopsHere = (recordXtime >= recoveryTargetTime);
2702 : }
2703 :
2704 42546 : if (stopsHere)
2705 : {
2706 0 : recoveryStopAfter = false;
2707 0 : recoveryStopXid = recordXid;
2708 0 : recoveryStopTime = recordXtime;
2709 0 : recoveryStopLSN = InvalidXLogRecPtr;
2710 0 : recoveryStopName[0] = '\0';
2711 :
2712 0 : if (isCommit)
2713 : {
2714 0 : ereport(LOG,
2715 : (errmsg("recovery stopping before commit of transaction %u, time %s",
2716 : recoveryStopXid,
2717 : timestamptz_to_str(recoveryStopTime))));
2718 : }
2719 : else
2720 : {
2721 0 : ereport(LOG,
2722 : (errmsg("recovery stopping before abort of transaction %u, time %s",
2723 : recoveryStopXid,
2724 : timestamptz_to_str(recoveryStopTime))));
2725 : }
2726 : }
2727 :
2728 42546 : return stopsHere;
2729 : }
2730 :
2731 : /*
2732 : * Same as recoveryStopsBefore, but called after applying the record.
2733 : *
2734 : * We also track the timestamp of the latest applied COMMIT/ABORT
2735 : * record in XLogRecoveryCtl->recoveryLastXTime.
2736 : */
2737 : static bool
2738 5473480 : recoveryStopsAfter(XLogReaderState *record)
2739 : {
2740 : uint8 info;
2741 : uint8 xact_info;
2742 : uint8 rmid;
2743 5473480 : TimestampTz recordXtime = 0;
2744 :
2745 : /*
2746 : * Ignore recovery target settings when not in archive recovery (meaning
2747 : * we are in crash recovery).
2748 : */
2749 5473480 : if (!ArchiveRecoveryRequested)
2750 487456 : return false;
2751 :
2752 4986024 : info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2753 4986024 : rmid = XLogRecGetRmid(record);
2754 :
2755 : /*
2756 : * There can be many restore points that share the same name; we stop at
2757 : * the first one.
2758 : */
2759 4986024 : if (recoveryTarget == RECOVERY_TARGET_NAME &&
2760 40 : rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
2761 : {
2762 : xl_restore_point *recordRestorePointData;
2763 :
2764 6 : recordRestorePointData = (xl_restore_point *) XLogRecGetData(record);
2765 :
2766 6 : if (strcmp(recordRestorePointData->rp_name, recoveryTargetName) == 0)
2767 : {
2768 4 : recoveryStopAfter = true;
2769 4 : recoveryStopXid = InvalidTransactionId;
2770 4 : recoveryStopLSN = InvalidXLogRecPtr;
2771 4 : (void) getRecordTimestamp(record, &recoveryStopTime);
2772 4 : strlcpy(recoveryStopName, recordRestorePointData->rp_name, MAXFNAMELEN);
2773 :
2774 4 : ereport(LOG,
2775 : (errmsg("recovery stopping at restore point \"%s\", time %s",
2776 : recoveryStopName,
2777 : timestamptz_to_str(recoveryStopTime))));
2778 4 : return true;
2779 : }
2780 : }
2781 :
2782 : /* Check if the target LSN has been reached */
2783 4986020 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
2784 16064 : recoveryTargetInclusive &&
2785 16064 : record->ReadRecPtr >= recoveryTargetLSN)
2786 : {
2787 6 : recoveryStopAfter = true;
2788 6 : recoveryStopXid = InvalidTransactionId;
2789 6 : recoveryStopLSN = record->ReadRecPtr;
2790 6 : recoveryStopTime = 0;
2791 6 : recoveryStopName[0] = '\0';
2792 6 : ereport(LOG,
2793 : errmsg("recovery stopping after WAL location (LSN) \"%X/%08X\"",
2794 : LSN_FORMAT_ARGS(recoveryStopLSN)));
2795 6 : return true;
2796 : }
2797 :
2798 4986014 : if (rmid != RM_XACT_ID)
2799 4942902 : return false;
2800 :
2801 43112 : xact_info = info & XLOG_XACT_OPMASK;
2802 :
2803 43112 : if (xact_info == XLOG_XACT_COMMIT ||
2804 4126 : xact_info == XLOG_XACT_COMMIT_PREPARED ||
2805 592 : xact_info == XLOG_XACT_ABORT ||
2806 : xact_info == XLOG_XACT_ABORT_PREPARED)
2807 : {
2808 : TransactionId recordXid;
2809 :
2810 : /* Update the last applied transaction timestamp */
2811 42542 : if (getRecordTimestamp(record, &recordXtime))
2812 42542 : SetLatestXTime(recordXtime);
2813 :
2814 : /* Extract the XID of the committed/aborted transaction */
2815 42542 : if (xact_info == XLOG_XACT_COMMIT_PREPARED)
2816 : {
2817 48 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
2818 : xl_xact_parsed_commit parsed;
2819 :
2820 48 : ParseCommitRecord(XLogRecGetInfo(record),
2821 : xlrec,
2822 : &parsed);
2823 48 : recordXid = parsed.twophase_xid;
2824 : }
2825 42494 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
2826 : {
2827 22 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
2828 : xl_xact_parsed_abort parsed;
2829 :
2830 22 : ParseAbortRecord(XLogRecGetInfo(record),
2831 : xlrec,
2832 : &parsed);
2833 22 : recordXid = parsed.twophase_xid;
2834 : }
2835 : else
2836 42472 : recordXid = XLogRecGetXid(record);
2837 :
2838 : /*
2839 : * There can be only one transaction end record with this exact
2840 : * transactionid
2841 : *
2842 : * when testing for an xid, we MUST test for equality only, since
2843 : * transactions are numbered in the order they start, not the order
2844 : * they complete. A higher numbered xid will complete before you about
2845 : * 50% of the time...
2846 : */
2847 42542 : if (recoveryTarget == RECOVERY_TARGET_XID && recoveryTargetInclusive &&
2848 0 : recordXid == recoveryTargetXid)
2849 : {
2850 0 : recoveryStopAfter = true;
2851 0 : recoveryStopXid = recordXid;
2852 0 : recoveryStopTime = recordXtime;
2853 0 : recoveryStopLSN = InvalidXLogRecPtr;
2854 0 : recoveryStopName[0] = '\0';
2855 :
2856 0 : if (xact_info == XLOG_XACT_COMMIT ||
2857 : xact_info == XLOG_XACT_COMMIT_PREPARED)
2858 : {
2859 0 : ereport(LOG,
2860 : (errmsg("recovery stopping after commit of transaction %u, time %s",
2861 : recoveryStopXid,
2862 : timestamptz_to_str(recoveryStopTime))));
2863 : }
2864 0 : else if (xact_info == XLOG_XACT_ABORT ||
2865 : xact_info == XLOG_XACT_ABORT_PREPARED)
2866 : {
2867 0 : ereport(LOG,
2868 : (errmsg("recovery stopping after abort of transaction %u, time %s",
2869 : recoveryStopXid,
2870 : timestamptz_to_str(recoveryStopTime))));
2871 : }
2872 0 : return true;
2873 : }
2874 : }
2875 :
2876 : /* Check if we should stop as soon as reaching consistency */
2877 43112 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
2878 : {
2879 0 : ereport(LOG,
2880 : (errmsg("recovery stopping after reaching consistency")));
2881 :
2882 0 : recoveryStopAfter = true;
2883 0 : recoveryStopXid = InvalidTransactionId;
2884 0 : recoveryStopTime = 0;
2885 0 : recoveryStopLSN = InvalidXLogRecPtr;
2886 0 : recoveryStopName[0] = '\0';
2887 0 : return true;
2888 : }
2889 :
2890 43112 : return false;
2891 : }
2892 :
2893 : /*
2894 : * Create a comment for the history file to explain why and where
2895 : * timeline changed.
2896 : */
2897 : static char *
2898 1726 : getRecoveryStopReason(void)
2899 : {
2900 : char reason[200];
2901 :
2902 1726 : if (recoveryTarget == RECOVERY_TARGET_XID)
2903 0 : snprintf(reason, sizeof(reason),
2904 : "%s transaction %u",
2905 0 : recoveryStopAfter ? "after" : "before",
2906 : recoveryStopXid);
2907 1726 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
2908 0 : snprintf(reason, sizeof(reason),
2909 : "%s %s\n",
2910 0 : recoveryStopAfter ? "after" : "before",
2911 : timestamptz_to_str(recoveryStopTime));
2912 1726 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
2913 14 : snprintf(reason, sizeof(reason),
2914 : "%s LSN %X/%08X\n",
2915 14 : recoveryStopAfter ? "after" : "before",
2916 14 : LSN_FORMAT_ARGS(recoveryStopLSN));
2917 1712 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
2918 6 : snprintf(reason, sizeof(reason),
2919 : "at restore point \"%s\"",
2920 : recoveryStopName);
2921 1706 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
2922 0 : snprintf(reason, sizeof(reason), "reached consistency");
2923 : else
2924 1706 : snprintf(reason, sizeof(reason), "no recovery target specified");
2925 :
2926 1726 : return pstrdup(reason);
2927 : }
2928 :
2929 : /*
2930 : * Wait until shared recoveryPauseState is set to RECOVERY_NOT_PAUSED.
2931 : *
2932 : * endOfRecovery is true if the recovery target is reached and
2933 : * the paused state starts at the end of recovery because of
2934 : * recovery_target_action=pause, and false otherwise.
2935 : */
2936 : static void
2937 6 : recoveryPausesHere(bool endOfRecovery)
2938 : {
2939 : /* Don't pause unless users can connect! */
2940 6 : if (!LocalHotStandbyActive)
2941 0 : return;
2942 :
2943 : /* Don't pause after standby promotion has been triggered */
2944 6 : if (LocalPromoteIsTriggered)
2945 0 : return;
2946 :
2947 6 : if (endOfRecovery)
2948 2 : ereport(LOG,
2949 : (errmsg("pausing at the end of recovery"),
2950 : errhint("Execute pg_wal_replay_resume() to promote.")));
2951 : else
2952 4 : ereport(LOG,
2953 : (errmsg("recovery has paused"),
2954 : errhint("Execute pg_wal_replay_resume() to continue.")));
2955 :
2956 : /* loop until recoveryPauseState is set to RECOVERY_NOT_PAUSED */
2957 18 : while (GetRecoveryPauseState() != RECOVERY_NOT_PAUSED)
2958 : {
2959 16 : ProcessStartupProcInterrupts();
2960 16 : if (CheckForStandbyTrigger())
2961 4 : return;
2962 :
2963 : /*
2964 : * If recovery pause is requested then set it paused. While we are in
2965 : * the loop, user might resume and pause again so set this every time.
2966 : */
2967 12 : ConfirmRecoveryPaused();
2968 :
2969 : /*
2970 : * We wait on a condition variable that will wake us as soon as the
2971 : * pause ends, but we use a timeout so we can check the above exit
2972 : * condition periodically too.
2973 : */
2974 12 : ConditionVariableTimedSleep(&XLogRecoveryCtl->recoveryNotPausedCV, 1000,
2975 : WAIT_EVENT_RECOVERY_PAUSE);
2976 : }
2977 2 : ConditionVariableCancelSleep();
2978 : }
2979 :
2980 : /*
2981 : * When recovery_min_apply_delay is set, we wait long enough to make sure
2982 : * certain record types are applied at least that interval behind the primary.
2983 : *
2984 : * Returns true if we waited.
2985 : *
2986 : * Note that the delay is calculated between the WAL record log time and
2987 : * the current time on standby. We would prefer to keep track of when this
2988 : * standby received each WAL record, which would allow a more consistent
2989 : * approach and one not affected by time synchronisation issues, but that
2990 : * is significantly more effort and complexity for little actual gain in
2991 : * usability.
2992 : */
2993 : static bool
2994 5473484 : recoveryApplyDelay(XLogReaderState *record)
2995 : {
2996 : uint8 xact_info;
2997 : TimestampTz xtime;
2998 : TimestampTz delayUntil;
2999 : long msecs;
3000 :
3001 : /* nothing to do if no delay configured */
3002 5473484 : if (recovery_min_apply_delay <= 0)
3003 5473484 : return false;
3004 :
3005 : /* no delay is applied on a database not yet consistent */
3006 0 : if (!reachedConsistency)
3007 0 : return false;
3008 :
3009 : /* nothing to do if crash recovery is requested */
3010 0 : if (!ArchiveRecoveryRequested)
3011 0 : return false;
3012 :
3013 : /*
3014 : * Is it a COMMIT record?
3015 : *
3016 : * We deliberately choose not to delay aborts since they have no effect on
3017 : * MVCC. We already allow replay of records that don't have a timestamp,
3018 : * so there is already opportunity for issues caused by early conflicts on
3019 : * standbys.
3020 : */
3021 0 : if (XLogRecGetRmid(record) != RM_XACT_ID)
3022 0 : return false;
3023 :
3024 0 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
3025 :
3026 0 : if (xact_info != XLOG_XACT_COMMIT &&
3027 : xact_info != XLOG_XACT_COMMIT_PREPARED)
3028 0 : return false;
3029 :
3030 0 : if (!getRecordTimestamp(record, &xtime))
3031 0 : return false;
3032 :
3033 0 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
3034 :
3035 : /*
3036 : * Exit without arming the latch if it's already past time to apply this
3037 : * record
3038 : */
3039 0 : msecs = TimestampDifferenceMilliseconds(GetCurrentTimestamp(), delayUntil);
3040 0 : if (msecs <= 0)
3041 0 : return false;
3042 :
3043 : while (true)
3044 : {
3045 0 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3046 :
3047 : /* This might change recovery_min_apply_delay. */
3048 0 : ProcessStartupProcInterrupts();
3049 :
3050 0 : if (CheckForStandbyTrigger())
3051 0 : break;
3052 :
3053 : /*
3054 : * Recalculate delayUntil as recovery_min_apply_delay could have
3055 : * changed while waiting in this loop.
3056 : */
3057 0 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
3058 :
3059 : /*
3060 : * Wait for difference between GetCurrentTimestamp() and delayUntil.
3061 : */
3062 0 : msecs = TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
3063 : delayUntil);
3064 :
3065 0 : if (msecs <= 0)
3066 0 : break;
3067 :
3068 0 : elog(DEBUG2, "recovery apply delay %ld milliseconds", msecs);
3069 :
3070 0 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3071 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
3072 : msecs,
3073 : WAIT_EVENT_RECOVERY_APPLY_DELAY);
3074 : }
3075 0 : return true;
3076 : }
3077 :
3078 : /*
3079 : * Get the current state of the recovery pause request.
3080 : */
3081 : RecoveryPauseState
3082 28 : GetRecoveryPauseState(void)
3083 : {
3084 : RecoveryPauseState state;
3085 :
3086 28 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3087 28 : state = XLogRecoveryCtl->recoveryPauseState;
3088 28 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3089 :
3090 28 : return state;
3091 : }
3092 :
3093 : /*
3094 : * Set the recovery pause state.
3095 : *
3096 : * If recovery pause is requested then sets the recovery pause state to
3097 : * 'pause requested' if it is not already 'paused'. Otherwise, sets it
3098 : * to 'not paused' to resume the recovery. The recovery pause will be
3099 : * confirmed by the ConfirmRecoveryPaused.
3100 : */
3101 : void
3102 92 : SetRecoveryPause(bool recoveryPause)
3103 : {
3104 92 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3105 :
3106 92 : if (!recoveryPause)
3107 86 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_NOT_PAUSED;
3108 6 : else if (XLogRecoveryCtl->recoveryPauseState == RECOVERY_NOT_PAUSED)
3109 6 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_PAUSE_REQUESTED;
3110 :
3111 92 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3112 :
3113 92 : if (!recoveryPause)
3114 86 : ConditionVariableBroadcast(&XLogRecoveryCtl->recoveryNotPausedCV);
3115 92 : }
3116 :
3117 : /*
3118 : * Confirm the recovery pause by setting the recovery pause state to
3119 : * RECOVERY_PAUSED.
3120 : */
3121 : static void
3122 12 : ConfirmRecoveryPaused(void)
3123 : {
3124 : /* If recovery pause is requested then set it paused */
3125 12 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3126 12 : if (XLogRecoveryCtl->recoveryPauseState == RECOVERY_PAUSE_REQUESTED)
3127 6 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_PAUSED;
3128 12 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3129 12 : }
3130 :
3131 :
3132 : /*
3133 : * Attempt to read the next XLOG record.
3134 : *
3135 : * Before first call, the reader needs to be positioned to the first record
3136 : * by calling XLogPrefetcherBeginRead().
3137 : *
3138 : * If no valid record is available, returns NULL, or fails if emode is PANIC.
3139 : * (emode must be either PANIC, LOG). In standby mode, retries until a valid
3140 : * record is available.
3141 : */
3142 : static XLogRecord *
3143 5477608 : ReadRecord(XLogPrefetcher *xlogprefetcher, int emode,
3144 : bool fetching_ckpt, TimeLineID replayTLI)
3145 : {
3146 : XLogRecord *record;
3147 5477608 : XLogReaderState *xlogreader = XLogPrefetcherGetReader(xlogprefetcher);
3148 5477608 : XLogPageReadPrivate *private = (XLogPageReadPrivate *) xlogreader->private_data;
3149 :
3150 : /* Pass through parameters to XLogPageRead */
3151 5477608 : private->fetching_ckpt = fetching_ckpt;
3152 5477608 : private->emode = emode;
3153 5477608 : private->randAccess = (xlogreader->ReadRecPtr == InvalidXLogRecPtr);
3154 5477608 : private->replayTLI = replayTLI;
3155 :
3156 : /* This is the first attempt to read this page. */
3157 5477608 : lastSourceFailed = false;
3158 :
3159 : for (;;)
3160 242 : {
3161 : char *errormsg;
3162 :
3163 5477850 : record = XLogPrefetcherReadRecord(xlogprefetcher, &errormsg);
3164 5477740 : if (record == NULL)
3165 : {
3166 : /*
3167 : * When we find that WAL ends in an incomplete record, keep track
3168 : * of that record. After recovery is done, we'll write a record
3169 : * to indicate to downstream WAL readers that that portion is to
3170 : * be ignored.
3171 : *
3172 : * However, when ArchiveRecoveryRequested = true, we're going to
3173 : * switch to a new timeline at the end of recovery. We will only
3174 : * copy WAL over to the new timeline up to the end of the last
3175 : * complete record, so if we did this, we would later create an
3176 : * overwrite contrecord in the wrong place, breaking everything.
3177 : */
3178 540 : if (!ArchiveRecoveryRequested &&
3179 212 : !XLogRecPtrIsInvalid(xlogreader->abortedRecPtr))
3180 : {
3181 24 : abortedRecPtr = xlogreader->abortedRecPtr;
3182 24 : missingContrecPtr = xlogreader->missingContrecPtr;
3183 : }
3184 :
3185 540 : if (readFile >= 0)
3186 : {
3187 502 : close(readFile);
3188 502 : readFile = -1;
3189 : }
3190 :
3191 : /*
3192 : * We only end up here without a message when XLogPageRead()
3193 : * failed - in that case we already logged something. In
3194 : * StandbyMode that only happens if we have been triggered, so we
3195 : * shouldn't loop anymore in that case.
3196 : */
3197 540 : if (errormsg)
3198 502 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3199 : (errmsg_internal("%s", errormsg) /* already translated */ ));
3200 : }
3201 :
3202 : /*
3203 : * Check page TLI is one of the expected values.
3204 : */
3205 5477200 : else if (!tliInHistory(xlogreader->latestPageTLI, expectedTLEs))
3206 : {
3207 : char fname[MAXFNAMELEN];
3208 : XLogSegNo segno;
3209 : int32 offset;
3210 :
3211 0 : XLByteToSeg(xlogreader->latestPagePtr, segno, wal_segment_size);
3212 0 : offset = XLogSegmentOffset(xlogreader->latestPagePtr,
3213 : wal_segment_size);
3214 0 : XLogFileName(fname, xlogreader->seg.ws_tli, segno,
3215 : wal_segment_size);
3216 0 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3217 : errmsg("unexpected timeline ID %u in WAL segment %s, LSN %X/%08X, offset %u",
3218 : xlogreader->latestPageTLI,
3219 : fname,
3220 : LSN_FORMAT_ARGS(xlogreader->latestPagePtr),
3221 : offset));
3222 0 : record = NULL;
3223 : }
3224 :
3225 5477740 : if (record)
3226 : {
3227 : /* Great, got a record */
3228 5477498 : return record;
3229 : }
3230 : else
3231 : {
3232 : /* No valid record available from this source */
3233 540 : lastSourceFailed = true;
3234 :
3235 : /*
3236 : * If archive recovery was requested, but we were still doing
3237 : * crash recovery, switch to archive recovery and retry using the
3238 : * offline archive. We have now replayed all the valid WAL in
3239 : * pg_wal, so we are presumably now consistent.
3240 : *
3241 : * We require that there's at least some valid WAL present in
3242 : * pg_wal, however (!fetching_ckpt). We could recover using the
3243 : * WAL from the archive, even if pg_wal is completely empty, but
3244 : * we'd have no idea how far we'd have to replay to reach
3245 : * consistency. So err on the safe side and give up.
3246 : */
3247 540 : if (!InArchiveRecovery && ArchiveRecoveryRequested &&
3248 4 : !fetching_ckpt)
3249 : {
3250 4 : ereport(DEBUG1,
3251 : (errmsg_internal("reached end of WAL in pg_wal, entering archive recovery")));
3252 4 : InArchiveRecovery = true;
3253 4 : if (StandbyModeRequested)
3254 4 : EnableStandbyMode();
3255 :
3256 4 : SwitchIntoArchiveRecovery(xlogreader->EndRecPtr, replayTLI);
3257 4 : minRecoveryPoint = xlogreader->EndRecPtr;
3258 4 : minRecoveryPointTLI = replayTLI;
3259 :
3260 4 : CheckRecoveryConsistency();
3261 :
3262 : /*
3263 : * Before we retry, reset lastSourceFailed and currentSource
3264 : * so that we will check the archive next.
3265 : */
3266 4 : lastSourceFailed = false;
3267 4 : currentSource = XLOG_FROM_ANY;
3268 :
3269 242 : continue;
3270 : }
3271 :
3272 : /* In standby mode, loop back to retry. Otherwise, give up. */
3273 536 : if (StandbyMode && !CheckForStandbyTrigger())
3274 238 : continue;
3275 : else
3276 298 : return NULL;
3277 : }
3278 : }
3279 : }
3280 :
3281 : /*
3282 : * Read the XLOG page containing targetPagePtr into readBuf (if not read
3283 : * already). Returns number of bytes read, if the page is read successfully,
3284 : * or XLREAD_FAIL in case of errors. When errors occur, they are ereport'ed,
3285 : * but only if they have not been previously reported.
3286 : *
3287 : * See XLogReaderRoutine.page_read for more details.
3288 : *
3289 : * While prefetching, xlogreader->nonblocking may be set. In that case,
3290 : * returns XLREAD_WOULDBLOCK if we'd otherwise have to wait for more WAL.
3291 : *
3292 : * This is responsible for restoring files from archive as needed, as well
3293 : * as for waiting for the requested WAL record to arrive in standby mode.
3294 : *
3295 : * xlogreader->private_data->emode specifies the log level used for reporting
3296 : * "file not found" or "end of WAL" situations in archive recovery, or in
3297 : * standby mode when promotion is triggered. If set to WARNING or below,
3298 : * XLogPageRead() returns XLREAD_FAIL in those situations, on higher log
3299 : * levels the ereport() won't return.
3300 : *
3301 : * In standby mode, if after a successful return of XLogPageRead() the
3302 : * caller finds the record it's interested in to be broken, it should
3303 : * ereport the error with the level determined by
3304 : * emode_for_corrupt_record(), and then set lastSourceFailed
3305 : * and call XLogPageRead() again with the same arguments. This lets
3306 : * XLogPageRead() to try fetching the record from another source, or to
3307 : * sleep and retry.
3308 : */
3309 : static int
3310 2819416 : XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr, int reqLen,
3311 : XLogRecPtr targetRecPtr, char *readBuf)
3312 : {
3313 2819416 : XLogPageReadPrivate *private =
3314 : (XLogPageReadPrivate *) xlogreader->private_data;
3315 2819416 : int emode = private->emode;
3316 : uint32 targetPageOff;
3317 : XLogSegNo targetSegNo PG_USED_FOR_ASSERTS_ONLY;
3318 : int r;
3319 : instr_time io_start;
3320 :
3321 2819416 : XLByteToSeg(targetPagePtr, targetSegNo, wal_segment_size);
3322 2819416 : targetPageOff = XLogSegmentOffset(targetPagePtr, wal_segment_size);
3323 :
3324 : /*
3325 : * See if we need to switch to a new segment because the requested record
3326 : * is not in the currently open one.
3327 : */
3328 2819416 : if (readFile >= 0 &&
3329 2816060 : !XLByteInSeg(targetPagePtr, readSegNo, wal_segment_size))
3330 : {
3331 : /*
3332 : * Request a restartpoint if we've replayed too much xlog since the
3333 : * last one.
3334 : */
3335 2994 : if (ArchiveRecoveryRequested && IsUnderPostmaster)
3336 : {
3337 2964 : if (XLogCheckpointNeeded(readSegNo))
3338 : {
3339 2758 : (void) GetRedoRecPtr();
3340 2758 : if (XLogCheckpointNeeded(readSegNo))
3341 2742 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
3342 : }
3343 : }
3344 :
3345 2994 : close(readFile);
3346 2994 : readFile = -1;
3347 2994 : readSource = XLOG_FROM_ANY;
3348 : }
3349 :
3350 2819416 : XLByteToSeg(targetPagePtr, readSegNo, wal_segment_size);
3351 :
3352 2819422 : retry:
3353 : /* See if we need to retrieve more data */
3354 2819422 : if (readFile < 0 ||
3355 2813066 : (readSource == XLOG_FROM_STREAM &&
3356 2789206 : flushedUpto < targetPagePtr + reqLen))
3357 : {
3358 22370 : if (readFile >= 0 &&
3359 16014 : xlogreader->nonblocking &&
3360 7848 : readSource == XLOG_FROM_STREAM &&
3361 7848 : flushedUpto < targetPagePtr + reqLen)
3362 7848 : return XLREAD_WOULDBLOCK;
3363 :
3364 14412 : switch (WaitForWALToBecomeAvailable(targetPagePtr + reqLen,
3365 14522 : private->randAccess,
3366 14522 : private->fetching_ckpt,
3367 : targetRecPtr,
3368 : private->replayTLI,
3369 : xlogreader->EndRecPtr,
3370 14522 : xlogreader->nonblocking))
3371 : {
3372 940 : case XLREAD_WOULDBLOCK:
3373 940 : return XLREAD_WOULDBLOCK;
3374 72 : case XLREAD_FAIL:
3375 72 : if (readFile >= 0)
3376 0 : close(readFile);
3377 72 : readFile = -1;
3378 72 : readLen = 0;
3379 72 : readSource = XLOG_FROM_ANY;
3380 72 : return XLREAD_FAIL;
3381 13400 : case XLREAD_SUCCESS:
3382 13400 : break;
3383 : }
3384 : }
3385 :
3386 : /*
3387 : * At this point, we have the right segment open and if we're streaming we
3388 : * know the requested record is in it.
3389 : */
3390 : Assert(readFile != -1);
3391 :
3392 : /*
3393 : * If the current segment is being streamed from the primary, calculate
3394 : * how much of the current page we have received already. We know the
3395 : * requested record has been received, but this is for the benefit of
3396 : * future calls, to allow quick exit at the top of this function.
3397 : */
3398 2810452 : if (readSource == XLOG_FROM_STREAM)
3399 : {
3400 2783516 : if (((targetPagePtr) / XLOG_BLCKSZ) != (flushedUpto / XLOG_BLCKSZ))
3401 2778010 : readLen = XLOG_BLCKSZ;
3402 : else
3403 5506 : readLen = XLogSegmentOffset(flushedUpto, wal_segment_size) -
3404 : targetPageOff;
3405 : }
3406 : else
3407 26936 : readLen = XLOG_BLCKSZ;
3408 :
3409 : /* Read the requested page */
3410 2810452 : readOff = targetPageOff;
3411 :
3412 : /* Measure I/O timing when reading segment */
3413 2810452 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3414 :
3415 2810452 : pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
3416 2810452 : r = pg_pread(readFile, readBuf, XLOG_BLCKSZ, (off_t) readOff);
3417 2810452 : if (r != XLOG_BLCKSZ)
3418 : {
3419 : char fname[MAXFNAMELEN];
3420 0 : int save_errno = errno;
3421 :
3422 0 : pgstat_report_wait_end();
3423 :
3424 0 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_READ,
3425 : io_start, 1, r);
3426 :
3427 0 : XLogFileName(fname, curFileTLI, readSegNo, wal_segment_size);
3428 0 : if (r < 0)
3429 : {
3430 0 : errno = save_errno;
3431 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
3432 : (errcode_for_file_access(),
3433 : errmsg("could not read from WAL segment %s, LSN %X/%08X, offset %u: %m",
3434 : fname, LSN_FORMAT_ARGS(targetPagePtr),
3435 : readOff)));
3436 : }
3437 : else
3438 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
3439 : (errcode(ERRCODE_DATA_CORRUPTED),
3440 : errmsg("could not read from WAL segment %s, LSN %X/%08X, offset %u: read %d of %zu",
3441 : fname, LSN_FORMAT_ARGS(targetPagePtr),
3442 : readOff, r, (Size) XLOG_BLCKSZ)));
3443 0 : goto next_record_is_invalid;
3444 : }
3445 2810452 : pgstat_report_wait_end();
3446 :
3447 2810452 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_READ,
3448 : io_start, 1, r);
3449 :
3450 : Assert(targetSegNo == readSegNo);
3451 : Assert(targetPageOff == readOff);
3452 : Assert(reqLen <= readLen);
3453 :
3454 2810452 : xlogreader->seg.ws_tli = curFileTLI;
3455 :
3456 : /*
3457 : * Check the page header immediately, so that we can retry immediately if
3458 : * it's not valid. This may seem unnecessary, because ReadPageInternal()
3459 : * validates the page header anyway, and would propagate the failure up to
3460 : * ReadRecord(), which would retry. However, there's a corner case with
3461 : * continuation records, if a record is split across two pages such that
3462 : * we would need to read the two pages from different sources across two
3463 : * WAL segments.
3464 : *
3465 : * The first page is only available locally, in pg_wal, because it's
3466 : * already been recycled on the primary. The second page, however, is not
3467 : * present in pg_wal, and we should stream it from the primary. There is a
3468 : * recycled WAL segment present in pg_wal, with garbage contents, however.
3469 : * We would read the first page from the local WAL segment, but when
3470 : * reading the second page, we would read the bogus, recycled, WAL
3471 : * segment. If we didn't catch that case here, we would never recover,
3472 : * because ReadRecord() would retry reading the whole record from the
3473 : * beginning.
3474 : *
3475 : * Of course, this only catches errors in the page header, which is what
3476 : * happens in the case of a recycled WAL segment. Other kinds of errors or
3477 : * corruption still has the same problem. But this at least fixes the
3478 : * common case, which can happen as part of normal operation.
3479 : *
3480 : * Validating the page header is cheap enough that doing it twice
3481 : * shouldn't be a big deal from a performance point of view.
3482 : *
3483 : * When not in standby mode, an invalid page header should cause recovery
3484 : * to end, not retry reading the page, so we don't need to validate the
3485 : * page header here for the retry. Instead, ReadPageInternal() is
3486 : * responsible for the validation.
3487 : */
3488 2810452 : if (StandbyMode &&
3489 2790830 : (targetPagePtr % wal_segment_size) == 0 &&
3490 2698 : !XLogReaderValidatePageHeader(xlogreader, targetPagePtr, readBuf))
3491 : {
3492 : /*
3493 : * Emit this error right now then retry this page immediately. Use
3494 : * errmsg_internal() because the message was already translated.
3495 : */
3496 8 : if (xlogreader->errormsg_buf[0])
3497 8 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3498 : (errmsg_internal("%s", xlogreader->errormsg_buf)));
3499 :
3500 : /* reset any error XLogReaderValidatePageHeader() might have set */
3501 8 : XLogReaderResetError(xlogreader);
3502 8 : goto next_record_is_invalid;
3503 : }
3504 :
3505 2810444 : return readLen;
3506 :
3507 8 : next_record_is_invalid:
3508 :
3509 : /*
3510 : * If we're reading ahead, give up fast. Retries and error reporting will
3511 : * be handled by a later read when recovery catches up to this point.
3512 : */
3513 8 : if (xlogreader->nonblocking)
3514 2 : return XLREAD_WOULDBLOCK;
3515 :
3516 6 : lastSourceFailed = true;
3517 :
3518 6 : if (readFile >= 0)
3519 6 : close(readFile);
3520 6 : readFile = -1;
3521 6 : readLen = 0;
3522 6 : readSource = XLOG_FROM_ANY;
3523 :
3524 : /* In standby-mode, keep trying */
3525 6 : if (StandbyMode)
3526 6 : goto retry;
3527 : else
3528 0 : return XLREAD_FAIL;
3529 : }
3530 :
3531 : /*
3532 : * Open the WAL segment containing WAL location 'RecPtr'.
3533 : *
3534 : * The segment can be fetched via restore_command, or via walreceiver having
3535 : * streamed the record, or it can already be present in pg_wal. Checking
3536 : * pg_wal is mainly for crash recovery, but it will be polled in standby mode
3537 : * too, in case someone copies a new segment directly to pg_wal. That is not
3538 : * documented or recommended, though.
3539 : *
3540 : * If 'fetching_ckpt' is true, we're fetching a checkpoint record, and should
3541 : * prepare to read WAL starting from RedoStartLSN after this.
3542 : *
3543 : * 'RecPtr' might not point to the beginning of the record we're interested
3544 : * in, it might also point to the page or segment header. In that case,
3545 : * 'tliRecPtr' is the position of the WAL record we're interested in. It is
3546 : * used to decide which timeline to stream the requested WAL from.
3547 : *
3548 : * 'replayLSN' is the current replay LSN, so that if we scan for new
3549 : * timelines, we can reject a switch to a timeline that branched off before
3550 : * this point.
3551 : *
3552 : * If the record is not immediately available, the function returns false
3553 : * if we're not in standby mode. In standby mode, waits for it to become
3554 : * available.
3555 : *
3556 : * When the requested record becomes available, the function opens the file
3557 : * containing it (if not open already), and returns XLREAD_SUCCESS. When end
3558 : * of standby mode is triggered by the user, and there is no more WAL
3559 : * available, returns XLREAD_FAIL.
3560 : *
3561 : * If nonblocking is true, then give up immediately if we can't satisfy the
3562 : * request, returning XLREAD_WOULDBLOCK instead of waiting.
3563 : */
3564 : static XLogPageReadResult
3565 14522 : WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
3566 : bool fetching_ckpt, XLogRecPtr tliRecPtr,
3567 : TimeLineID replayTLI, XLogRecPtr replayLSN,
3568 : bool nonblocking)
3569 : {
3570 : static TimestampTz last_fail_time = 0;
3571 : TimestampTz now;
3572 14522 : bool streaming_reply_sent = false;
3573 :
3574 : /*-------
3575 : * Standby mode is implemented by a state machine:
3576 : *
3577 : * 1. Read from either archive or pg_wal (XLOG_FROM_ARCHIVE), or just
3578 : * pg_wal (XLOG_FROM_PG_WAL)
3579 : * 2. Check for promotion trigger request
3580 : * 3. Read from primary server via walreceiver (XLOG_FROM_STREAM)
3581 : * 4. Rescan timelines
3582 : * 5. Sleep wal_retrieve_retry_interval milliseconds, and loop back to 1.
3583 : *
3584 : * Failure to read from the current source advances the state machine to
3585 : * the next state.
3586 : *
3587 : * 'currentSource' indicates the current state. There are no currentSource
3588 : * values for "check trigger", "rescan timelines", and "sleep" states,
3589 : * those actions are taken when reading from the previous source fails, as
3590 : * part of advancing to the next state.
3591 : *
3592 : * If standby mode is turned off while reading WAL from stream, we move
3593 : * to XLOG_FROM_ARCHIVE and reset lastSourceFailed, to force fetching
3594 : * the files (which would be required at end of recovery, e.g., timeline
3595 : * history file) from archive or pg_wal. We don't need to kill WAL receiver
3596 : * here because it's already stopped when standby mode is turned off at
3597 : * the end of recovery.
3598 : *-------
3599 : */
3600 14522 : if (!InArchiveRecovery)
3601 1850 : currentSource = XLOG_FROM_PG_WAL;
3602 12672 : else if (currentSource == XLOG_FROM_ANY ||
3603 12444 : (!StandbyMode && currentSource == XLOG_FROM_STREAM))
3604 : {
3605 228 : lastSourceFailed = false;
3606 228 : currentSource = XLOG_FROM_ARCHIVE;
3607 : }
3608 :
3609 : for (;;)
3610 11198 : {
3611 25720 : XLogSource oldSource = currentSource;
3612 25720 : bool startWalReceiver = false;
3613 :
3614 : /*
3615 : * First check if we failed to read from the current source, and
3616 : * advance the state machine if so. The failure to read might've
3617 : * happened outside this function, e.g when a CRC check fails on a
3618 : * record, or within this loop.
3619 : */
3620 25720 : if (lastSourceFailed)
3621 : {
3622 : /*
3623 : * Don't allow any retry loops to occur during nonblocking
3624 : * readahead. Let the caller process everything that has been
3625 : * decoded already first.
3626 : */
3627 924 : if (nonblocking)
3628 140 : return XLREAD_WOULDBLOCK;
3629 :
3630 784 : switch (currentSource)
3631 : {
3632 472 : case XLOG_FROM_ARCHIVE:
3633 : case XLOG_FROM_PG_WAL:
3634 :
3635 : /*
3636 : * Check to see if promotion is requested. Note that we do
3637 : * this only after failure, so when you promote, we still
3638 : * finish replaying as much as we can from archive and
3639 : * pg_wal before failover.
3640 : */
3641 472 : if (StandbyMode && CheckForStandbyTrigger())
3642 : {
3643 32 : XLogShutdownWalRcv();
3644 32 : return XLREAD_FAIL;
3645 : }
3646 :
3647 : /*
3648 : * Not in standby mode, and we've now tried the archive
3649 : * and pg_wal.
3650 : */
3651 440 : if (!StandbyMode)
3652 40 : return XLREAD_FAIL;
3653 :
3654 : /*
3655 : * Move to XLOG_FROM_STREAM state, and set to start a
3656 : * walreceiver if necessary.
3657 : */
3658 400 : currentSource = XLOG_FROM_STREAM;
3659 400 : startWalReceiver = true;
3660 400 : break;
3661 :
3662 312 : case XLOG_FROM_STREAM:
3663 :
3664 : /*
3665 : * Failure while streaming. Most likely, we got here
3666 : * because streaming replication was terminated, or
3667 : * promotion was triggered. But we also get here if we
3668 : * find an invalid record in the WAL streamed from the
3669 : * primary, in which case something is seriously wrong.
3670 : * There's little chance that the problem will just go
3671 : * away, but PANIC is not good for availability either,
3672 : * especially in hot standby mode. So, we treat that the
3673 : * same as disconnection, and retry from archive/pg_wal
3674 : * again. The WAL in the archive should be identical to
3675 : * what was streamed, so it's unlikely that it helps, but
3676 : * one can hope...
3677 : */
3678 :
3679 : /*
3680 : * We should be able to move to XLOG_FROM_STREAM only in
3681 : * standby mode.
3682 : */
3683 : Assert(StandbyMode);
3684 :
3685 : /*
3686 : * Before we leave XLOG_FROM_STREAM state, make sure that
3687 : * walreceiver is not active, so that it won't overwrite
3688 : * WAL that we restore from archive.
3689 : */
3690 312 : XLogShutdownWalRcv();
3691 :
3692 : /*
3693 : * Before we sleep, re-scan for possible new timelines if
3694 : * we were requested to recover to the latest timeline.
3695 : */
3696 312 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
3697 : {
3698 312 : if (rescanLatestTimeLine(replayTLI, replayLSN))
3699 : {
3700 12 : currentSource = XLOG_FROM_ARCHIVE;
3701 12 : break;
3702 : }
3703 : }
3704 :
3705 : /*
3706 : * XLOG_FROM_STREAM is the last state in our state
3707 : * machine, so we've exhausted all the options for
3708 : * obtaining the requested WAL. We're going to loop back
3709 : * and retry from the archive, but if it hasn't been long
3710 : * since last attempt, sleep wal_retrieve_retry_interval
3711 : * milliseconds to avoid busy-waiting.
3712 : */
3713 300 : now = GetCurrentTimestamp();
3714 300 : if (!TimestampDifferenceExceeds(last_fail_time, now,
3715 : wal_retrieve_retry_interval))
3716 : {
3717 : long wait_time;
3718 :
3719 320 : wait_time = wal_retrieve_retry_interval -
3720 160 : TimestampDifferenceMilliseconds(last_fail_time, now);
3721 :
3722 160 : elog(LOG, "waiting for WAL to become available at %X/%08X",
3723 : LSN_FORMAT_ARGS(RecPtr));
3724 :
3725 : /* Do background tasks that might benefit us later. */
3726 160 : KnownAssignedTransactionIdsIdleMaintenance();
3727 :
3728 160 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3729 : WL_LATCH_SET | WL_TIMEOUT |
3730 : WL_EXIT_ON_PM_DEATH,
3731 : wait_time,
3732 : WAIT_EVENT_RECOVERY_RETRIEVE_RETRY_INTERVAL);
3733 160 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3734 160 : now = GetCurrentTimestamp();
3735 :
3736 : /* Handle interrupt signals of startup process */
3737 160 : ProcessStartupProcInterrupts();
3738 : }
3739 274 : last_fail_time = now;
3740 274 : currentSource = XLOG_FROM_ARCHIVE;
3741 274 : break;
3742 :
3743 0 : default:
3744 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
3745 : }
3746 : }
3747 24796 : else if (currentSource == XLOG_FROM_PG_WAL)
3748 : {
3749 : /*
3750 : * We just successfully read a file in pg_wal. We prefer files in
3751 : * the archive over ones in pg_wal, so try the next file again
3752 : * from the archive first.
3753 : */
3754 1846 : if (InArchiveRecovery)
3755 0 : currentSource = XLOG_FROM_ARCHIVE;
3756 : }
3757 :
3758 25482 : if (currentSource != oldSource)
3759 686 : elog(DEBUG2, "switched WAL source from %s to %s after %s",
3760 : xlogSourceNames[oldSource], xlogSourceNames[currentSource],
3761 : lastSourceFailed ? "failure" : "success");
3762 :
3763 : /*
3764 : * We've now handled possible failure. Try to read from the chosen
3765 : * source.
3766 : */
3767 25482 : lastSourceFailed = false;
3768 :
3769 25482 : switch (currentSource)
3770 : {
3771 3410 : case XLOG_FROM_ARCHIVE:
3772 : case XLOG_FROM_PG_WAL:
3773 :
3774 : /*
3775 : * WAL receiver must not be running when reading WAL from
3776 : * archive or pg_wal.
3777 : */
3778 : Assert(!WalRcvStreaming());
3779 :
3780 : /* Close any old file we might have open. */
3781 3410 : if (readFile >= 0)
3782 : {
3783 156 : close(readFile);
3784 156 : readFile = -1;
3785 : }
3786 : /* Reset curFileTLI if random fetch. */
3787 3410 : if (randAccess)
3788 2146 : curFileTLI = 0;
3789 :
3790 : /*
3791 : * Try to restore the file from archive, or read an existing
3792 : * file from pg_wal.
3793 : */
3794 3410 : readFile = XLogFileReadAnyTLI(readSegNo,
3795 3410 : currentSource == XLOG_FROM_ARCHIVE ? XLOG_FROM_ANY :
3796 : currentSource);
3797 3410 : if (readFile >= 0)
3798 3076 : return XLREAD_SUCCESS; /* success! */
3799 :
3800 : /*
3801 : * Nope, not found in archive or pg_wal.
3802 : */
3803 334 : lastSourceFailed = true;
3804 334 : break;
3805 :
3806 22072 : case XLOG_FROM_STREAM:
3807 : {
3808 : bool havedata;
3809 :
3810 : /*
3811 : * We should be able to move to XLOG_FROM_STREAM only in
3812 : * standby mode.
3813 : */
3814 : Assert(StandbyMode);
3815 :
3816 : /*
3817 : * First, shutdown walreceiver if its restart has been
3818 : * requested -- but no point if we're already slated for
3819 : * starting it.
3820 : */
3821 22072 : if (pendingWalRcvRestart && !startWalReceiver)
3822 : {
3823 8 : XLogShutdownWalRcv();
3824 :
3825 : /*
3826 : * Re-scan for possible new timelines if we were
3827 : * requested to recover to the latest timeline.
3828 : */
3829 8 : if (recoveryTargetTimeLineGoal ==
3830 : RECOVERY_TARGET_TIMELINE_LATEST)
3831 8 : rescanLatestTimeLine(replayTLI, replayLSN);
3832 :
3833 8 : startWalReceiver = true;
3834 : }
3835 22072 : pendingWalRcvRestart = false;
3836 :
3837 : /*
3838 : * Launch walreceiver if needed.
3839 : *
3840 : * If fetching_ckpt is true, RecPtr points to the initial
3841 : * checkpoint location. In that case, we use RedoStartLSN
3842 : * as the streaming start position instead of RecPtr, so
3843 : * that when we later jump backwards to start redo at
3844 : * RedoStartLSN, we will have the logs streamed already.
3845 : */
3846 22072 : if (startWalReceiver &&
3847 408 : PrimaryConnInfo && strcmp(PrimaryConnInfo, "") != 0)
3848 : {
3849 : XLogRecPtr ptr;
3850 : TimeLineID tli;
3851 :
3852 352 : if (fetching_ckpt)
3853 : {
3854 0 : ptr = RedoStartLSN;
3855 0 : tli = RedoStartTLI;
3856 : }
3857 : else
3858 : {
3859 352 : ptr = RecPtr;
3860 :
3861 : /*
3862 : * Use the record begin position to determine the
3863 : * TLI, rather than the position we're reading.
3864 : */
3865 352 : tli = tliOfPointInHistory(tliRecPtr, expectedTLEs);
3866 :
3867 352 : if (curFileTLI > 0 && tli < curFileTLI)
3868 0 : elog(ERROR, "according to history file, WAL location %X/%08X belongs to timeline %u, but previous recovered WAL file came from timeline %u",
3869 : LSN_FORMAT_ARGS(tliRecPtr),
3870 : tli, curFileTLI);
3871 : }
3872 352 : curFileTLI = tli;
3873 352 : SetInstallXLogFileSegmentActive();
3874 352 : RequestXLogStreaming(tli, ptr, PrimaryConnInfo,
3875 : PrimarySlotName,
3876 : wal_receiver_create_temp_slot);
3877 352 : flushedUpto = 0;
3878 : }
3879 :
3880 : /*
3881 : * Check if WAL receiver is active or wait to start up.
3882 : */
3883 22072 : if (!WalRcvStreaming())
3884 : {
3885 254 : lastSourceFailed = true;
3886 254 : break;
3887 : }
3888 :
3889 : /*
3890 : * Walreceiver is active, so see if new data has arrived.
3891 : *
3892 : * We only advance XLogReceiptTime when we obtain fresh
3893 : * WAL from walreceiver and observe that we had already
3894 : * processed everything before the most recent "chunk"
3895 : * that it flushed to disk. In steady state where we are
3896 : * keeping up with the incoming data, XLogReceiptTime will
3897 : * be updated on each cycle. When we are behind,
3898 : * XLogReceiptTime will not advance, so the grace time
3899 : * allotted to conflicting queries will decrease.
3900 : */
3901 21818 : if (RecPtr < flushedUpto)
3902 3526 : havedata = true;
3903 : else
3904 : {
3905 : XLogRecPtr latestChunkStart;
3906 :
3907 18292 : flushedUpto = GetWalRcvFlushRecPtr(&latestChunkStart, &receiveTLI);
3908 18292 : if (RecPtr < flushedUpto && receiveTLI == curFileTLI)
3909 : {
3910 9174 : havedata = true;
3911 9174 : if (latestChunkStart <= RecPtr)
3912 : {
3913 6720 : XLogReceiptTime = GetCurrentTimestamp();
3914 6720 : SetCurrentChunkStartTime(XLogReceiptTime);
3915 : }
3916 : }
3917 : else
3918 9118 : havedata = false;
3919 : }
3920 21818 : if (havedata)
3921 : {
3922 : /*
3923 : * Great, streamed far enough. Open the file if it's
3924 : * not open already. Also read the timeline history
3925 : * file if we haven't initialized timeline history
3926 : * yet; it should be streamed over and present in
3927 : * pg_wal by now. Use XLOG_FROM_STREAM so that source
3928 : * info is set correctly and XLogReceiptTime isn't
3929 : * changed.
3930 : *
3931 : * NB: We must set readTimeLineHistory based on
3932 : * recoveryTargetTLI, not receiveTLI. Normally they'll
3933 : * be the same, but if recovery_target_timeline is
3934 : * 'latest' and archiving is configured, then it's
3935 : * possible that we managed to retrieve one or more
3936 : * new timeline history files from the archive,
3937 : * updating recoveryTargetTLI.
3938 : */
3939 12700 : if (readFile < 0)
3940 : {
3941 2376 : if (!expectedTLEs)
3942 0 : expectedTLEs = readTimeLineHistory(recoveryTargetTLI);
3943 2376 : readFile = XLogFileRead(readSegNo, receiveTLI,
3944 : XLOG_FROM_STREAM, false);
3945 : Assert(readFile >= 0);
3946 : }
3947 : else
3948 : {
3949 : /* just make sure source info is correct... */
3950 10324 : readSource = XLOG_FROM_STREAM;
3951 10324 : XLogReceiptSource = XLOG_FROM_STREAM;
3952 10324 : return XLREAD_SUCCESS;
3953 : }
3954 2376 : break;
3955 : }
3956 :
3957 : /* In nonblocking mode, return rather than sleeping. */
3958 9118 : if (nonblocking)
3959 800 : return XLREAD_WOULDBLOCK;
3960 :
3961 : /*
3962 : * Data not here yet. Check for trigger, then wait for
3963 : * walreceiver to wake us up when new WAL arrives.
3964 : */
3965 8318 : if (CheckForStandbyTrigger())
3966 : {
3967 : /*
3968 : * Note that we don't return XLREAD_FAIL immediately
3969 : * here. After being triggered, we still want to
3970 : * replay all the WAL that was already streamed. It's
3971 : * in pg_wal now, so we just treat this as a failure,
3972 : * and the state machine will move on to replay the
3973 : * streamed WAL from pg_wal, and then recheck the
3974 : * trigger and exit replay.
3975 : */
3976 58 : lastSourceFailed = true;
3977 58 : break;
3978 : }
3979 :
3980 : /*
3981 : * Since we have replayed everything we have received so
3982 : * far and are about to start waiting for more WAL, let's
3983 : * tell the upstream server our replay location now so
3984 : * that pg_stat_replication doesn't show stale
3985 : * information.
3986 : */
3987 8260 : if (!streaming_reply_sent)
3988 : {
3989 6268 : WalRcvForceReply();
3990 6268 : streaming_reply_sent = true;
3991 : }
3992 :
3993 : /* Do any background tasks that might benefit us later. */
3994 8260 : KnownAssignedTransactionIdsIdleMaintenance();
3995 :
3996 : /* Update pg_stat_recovery_prefetch before sleeping. */
3997 8260 : XLogPrefetcherComputeStats(xlogprefetcher);
3998 :
3999 : /*
4000 : * Wait for more WAL to arrive, when we will be woken
4001 : * immediately by the WAL receiver.
4002 : */
4003 8260 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
4004 : WL_LATCH_SET | WL_EXIT_ON_PM_DEATH,
4005 : -1L,
4006 : WAIT_EVENT_RECOVERY_WAL_STREAM);
4007 8260 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
4008 8260 : break;
4009 : }
4010 :
4011 0 : default:
4012 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
4013 : }
4014 :
4015 : /*
4016 : * Check for recovery pause here so that we can confirm more quickly
4017 : * that a requested pause has actually taken effect.
4018 : */
4019 11282 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
4020 : RECOVERY_NOT_PAUSED)
4021 4 : recoveryPausesHere(false);
4022 :
4023 : /*
4024 : * This possibly-long loop needs to handle interrupts of startup
4025 : * process.
4026 : */
4027 11282 : ProcessStartupProcInterrupts();
4028 : }
4029 :
4030 : return XLREAD_FAIL; /* not reached */
4031 : }
4032 :
4033 :
4034 : /*
4035 : * Determine what log level should be used to report a corrupt WAL record
4036 : * in the current WAL page, previously read by XLogPageRead().
4037 : *
4038 : * 'emode' is the error mode that would be used to report a file-not-found
4039 : * or legitimate end-of-WAL situation. Generally, we use it as-is, but if
4040 : * we're retrying the exact same record that we've tried previously, only
4041 : * complain the first time to keep the noise down. However, we only do when
4042 : * reading from pg_wal, because we don't expect any invalid records in archive
4043 : * or in records streamed from the primary. Files in the archive should be complete,
4044 : * and we should never hit the end of WAL because we stop and wait for more WAL
4045 : * to arrive before replaying it.
4046 : *
4047 : * NOTE: This function remembers the RecPtr value it was last called with,
4048 : * to suppress repeated messages about the same record. Only call this when
4049 : * you are about to ereport(), or you might cause a later message to be
4050 : * erroneously suppressed.
4051 : */
4052 : static int
4053 510 : emode_for_corrupt_record(int emode, XLogRecPtr RecPtr)
4054 : {
4055 : static XLogRecPtr lastComplaint = 0;
4056 :
4057 510 : if (readSource == XLOG_FROM_PG_WAL && emode == LOG)
4058 : {
4059 502 : if (RecPtr == lastComplaint)
4060 122 : emode = DEBUG1;
4061 : else
4062 380 : lastComplaint = RecPtr;
4063 : }
4064 510 : return emode;
4065 : }
4066 :
4067 :
4068 : /*
4069 : * Subroutine to try to fetch and validate a prior checkpoint record.
4070 : */
4071 : static XLogRecord *
4072 1844 : ReadCheckpointRecord(XLogPrefetcher *xlogprefetcher, XLogRecPtr RecPtr,
4073 : TimeLineID replayTLI)
4074 : {
4075 : XLogRecord *record;
4076 : uint8 info;
4077 :
4078 : Assert(xlogreader != NULL);
4079 :
4080 1844 : if (!XRecOffIsValid(RecPtr))
4081 : {
4082 0 : ereport(LOG,
4083 : (errmsg("invalid checkpoint location")));
4084 0 : return NULL;
4085 : }
4086 :
4087 1844 : XLogPrefetcherBeginRead(xlogprefetcher, RecPtr);
4088 1844 : record = ReadRecord(xlogprefetcher, LOG, true, replayTLI);
4089 :
4090 1844 : if (record == NULL)
4091 : {
4092 0 : ereport(LOG,
4093 : (errmsg("invalid checkpoint record")));
4094 0 : return NULL;
4095 : }
4096 1844 : if (record->xl_rmid != RM_XLOG_ID)
4097 : {
4098 0 : ereport(LOG,
4099 : (errmsg("invalid resource manager ID in checkpoint record")));
4100 0 : return NULL;
4101 : }
4102 1844 : info = record->xl_info & ~XLR_INFO_MASK;
4103 1844 : if (info != XLOG_CHECKPOINT_SHUTDOWN &&
4104 : info != XLOG_CHECKPOINT_ONLINE)
4105 : {
4106 0 : ereport(LOG,
4107 : (errmsg("invalid xl_info in checkpoint record")));
4108 0 : return NULL;
4109 : }
4110 1844 : if (record->xl_tot_len != SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(CheckPoint))
4111 : {
4112 0 : ereport(LOG,
4113 : (errmsg("invalid length of checkpoint record")));
4114 0 : return NULL;
4115 : }
4116 1844 : return record;
4117 : }
4118 :
4119 : /*
4120 : * Scan for new timelines that might have appeared in the archive since we
4121 : * started recovery.
4122 : *
4123 : * If there are any, the function changes recovery target TLI to the latest
4124 : * one and returns 'true'.
4125 : */
4126 : static bool
4127 320 : rescanLatestTimeLine(TimeLineID replayTLI, XLogRecPtr replayLSN)
4128 : {
4129 : List *newExpectedTLEs;
4130 : bool found;
4131 : ListCell *cell;
4132 : TimeLineID newtarget;
4133 320 : TimeLineID oldtarget = recoveryTargetTLI;
4134 320 : TimeLineHistoryEntry *currentTle = NULL;
4135 :
4136 320 : newtarget = findNewestTimeLine(recoveryTargetTLI);
4137 320 : if (newtarget == recoveryTargetTLI)
4138 : {
4139 : /* No new timelines found */
4140 308 : return false;
4141 : }
4142 :
4143 : /*
4144 : * Determine the list of expected TLIs for the new TLI
4145 : */
4146 :
4147 12 : newExpectedTLEs = readTimeLineHistory(newtarget);
4148 :
4149 : /*
4150 : * If the current timeline is not part of the history of the new timeline,
4151 : * we cannot proceed to it.
4152 : */
4153 12 : found = false;
4154 24 : foreach(cell, newExpectedTLEs)
4155 : {
4156 24 : currentTle = (TimeLineHistoryEntry *) lfirst(cell);
4157 :
4158 24 : if (currentTle->tli == recoveryTargetTLI)
4159 : {
4160 12 : found = true;
4161 12 : break;
4162 : }
4163 : }
4164 12 : if (!found)
4165 : {
4166 0 : ereport(LOG,
4167 : (errmsg("new timeline %u is not a child of database system timeline %u",
4168 : newtarget,
4169 : replayTLI)));
4170 0 : return false;
4171 : }
4172 :
4173 : /*
4174 : * The current timeline was found in the history file, but check that the
4175 : * next timeline was forked off from it *after* the current recovery
4176 : * location.
4177 : */
4178 12 : if (currentTle->end < replayLSN)
4179 : {
4180 0 : ereport(LOG,
4181 : errmsg("new timeline %u forked off current database system timeline %u before current recovery point %X/%08X",
4182 : newtarget,
4183 : replayTLI,
4184 : LSN_FORMAT_ARGS(replayLSN)));
4185 0 : return false;
4186 : }
4187 :
4188 : /* The new timeline history seems valid. Switch target */
4189 12 : recoveryTargetTLI = newtarget;
4190 12 : list_free_deep(expectedTLEs);
4191 12 : expectedTLEs = newExpectedTLEs;
4192 :
4193 : /*
4194 : * As in StartupXLOG(), try to ensure we have all the history files
4195 : * between the old target and new target in pg_wal.
4196 : */
4197 12 : restoreTimeLineHistoryFiles(oldtarget + 1, newtarget);
4198 :
4199 12 : ereport(LOG,
4200 : (errmsg("new target timeline is %u",
4201 : recoveryTargetTLI)));
4202 :
4203 12 : return true;
4204 : }
4205 :
4206 :
4207 : /*
4208 : * Open a logfile segment for reading (during recovery).
4209 : *
4210 : * If source == XLOG_FROM_ARCHIVE, the segment is retrieved from archive.
4211 : * Otherwise, it's assumed to be already available in pg_wal.
4212 : */
4213 : static int
4214 6662 : XLogFileRead(XLogSegNo segno, TimeLineID tli,
4215 : XLogSource source, bool notfoundOk)
4216 : {
4217 : char xlogfname[MAXFNAMELEN];
4218 : char activitymsg[MAXFNAMELEN + 16];
4219 : char path[MAXPGPATH];
4220 : int fd;
4221 :
4222 6662 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
4223 :
4224 6662 : switch (source)
4225 : {
4226 1582 : case XLOG_FROM_ARCHIVE:
4227 : /* Report recovery progress in PS display */
4228 1582 : snprintf(activitymsg, sizeof(activitymsg), "waiting for %s",
4229 : xlogfname);
4230 1582 : set_ps_display(activitymsg);
4231 :
4232 1582 : if (!RestoreArchivedFile(path, xlogfname,
4233 : "RECOVERYXLOG",
4234 : wal_segment_size,
4235 : InRedo))
4236 858 : return -1;
4237 724 : break;
4238 :
4239 5080 : case XLOG_FROM_PG_WAL:
4240 : case XLOG_FROM_STREAM:
4241 5080 : XLogFilePath(path, tli, segno, wal_segment_size);
4242 5080 : break;
4243 :
4244 0 : default:
4245 0 : elog(ERROR, "invalid XLogFileRead source %d", source);
4246 : }
4247 :
4248 : /*
4249 : * If the segment was fetched from archival storage, replace the existing
4250 : * xlog segment (if any) with the archival version.
4251 : */
4252 5804 : if (source == XLOG_FROM_ARCHIVE)
4253 : {
4254 : Assert(!IsInstallXLogFileSegmentActive());
4255 724 : KeepFileRestoredFromArchive(path, xlogfname);
4256 :
4257 : /*
4258 : * Set path to point at the new file in pg_wal.
4259 : */
4260 724 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlogfname);
4261 : }
4262 :
4263 5804 : fd = BasicOpenFile(path, O_RDONLY | PG_BINARY);
4264 5804 : if (fd >= 0)
4265 : {
4266 : /* Success! */
4267 5452 : curFileTLI = tli;
4268 :
4269 : /* Report recovery progress in PS display */
4270 5452 : snprintf(activitymsg, sizeof(activitymsg), "recovering %s",
4271 : xlogfname);
4272 5452 : set_ps_display(activitymsg);
4273 :
4274 : /* Track source of data in assorted state variables */
4275 5452 : readSource = source;
4276 5452 : XLogReceiptSource = source;
4277 : /* In FROM_STREAM case, caller tracks receipt time, not me */
4278 5452 : if (source != XLOG_FROM_STREAM)
4279 3076 : XLogReceiptTime = GetCurrentTimestamp();
4280 :
4281 5452 : return fd;
4282 : }
4283 352 : if (errno != ENOENT || !notfoundOk) /* unexpected failure? */
4284 0 : ereport(PANIC,
4285 : (errcode_for_file_access(),
4286 : errmsg("could not open file \"%s\": %m", path)));
4287 352 : return -1;
4288 : }
4289 :
4290 : /*
4291 : * Open a logfile segment for reading (during recovery).
4292 : *
4293 : * This version searches for the segment with any TLI listed in expectedTLEs.
4294 : */
4295 : static int
4296 3410 : XLogFileReadAnyTLI(XLogSegNo segno, XLogSource source)
4297 : {
4298 : char path[MAXPGPATH];
4299 : ListCell *cell;
4300 : int fd;
4301 : List *tles;
4302 :
4303 : /*
4304 : * Loop looking for a suitable timeline ID: we might need to read any of
4305 : * the timelines listed in expectedTLEs.
4306 : *
4307 : * We expect curFileTLI on entry to be the TLI of the preceding file in
4308 : * sequence, or 0 if there was no predecessor. We do not allow curFileTLI
4309 : * to go backwards; this prevents us from picking up the wrong file when a
4310 : * parent timeline extends to higher segment numbers than the child we
4311 : * want to read.
4312 : *
4313 : * If we haven't read the timeline history file yet, read it now, so that
4314 : * we know which TLIs to scan. We don't save the list in expectedTLEs,
4315 : * however, unless we actually find a valid segment. That way if there is
4316 : * neither a timeline history file nor a WAL segment in the archive, and
4317 : * streaming replication is set up, we'll read the timeline history file
4318 : * streamed from the primary when we start streaming, instead of
4319 : * recovering with a dummy history generated here.
4320 : */
4321 3410 : if (expectedTLEs)
4322 1566 : tles = expectedTLEs;
4323 : else
4324 1844 : tles = readTimeLineHistory(recoveryTargetTLI);
4325 :
4326 3776 : foreach(cell, tles)
4327 : {
4328 3450 : TimeLineHistoryEntry *hent = (TimeLineHistoryEntry *) lfirst(cell);
4329 3450 : TimeLineID tli = hent->tli;
4330 :
4331 3450 : if (tli < curFileTLI)
4332 8 : break; /* don't bother looking at too-old TLIs */
4333 :
4334 : /*
4335 : * Skip scanning the timeline ID that the logfile segment to read
4336 : * doesn't belong to
4337 : */
4338 3442 : if (hent->begin != InvalidXLogRecPtr)
4339 : {
4340 144 : XLogSegNo beginseg = 0;
4341 :
4342 144 : XLByteToSeg(hent->begin, beginseg, wal_segment_size);
4343 :
4344 : /*
4345 : * The logfile segment that doesn't belong to the timeline is
4346 : * older or newer than the segment that the timeline started or
4347 : * ended at, respectively. It's sufficient to check only the
4348 : * starting segment of the timeline here. Since the timelines are
4349 : * scanned in descending order in this loop, any segments newer
4350 : * than the ending segment should belong to newer timeline and
4351 : * have already been read before. So it's not necessary to check
4352 : * the ending segment of the timeline here.
4353 : */
4354 144 : if (segno < beginseg)
4355 14 : continue;
4356 : }
4357 :
4358 3428 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_ARCHIVE)
4359 : {
4360 1582 : fd = XLogFileRead(segno, tli, XLOG_FROM_ARCHIVE, true);
4361 1582 : if (fd != -1)
4362 : {
4363 724 : elog(DEBUG1, "got WAL segment from archive");
4364 724 : if (!expectedTLEs)
4365 36 : expectedTLEs = tles;
4366 3076 : return fd;
4367 : }
4368 : }
4369 :
4370 2704 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_PG_WAL)
4371 : {
4372 2704 : fd = XLogFileRead(segno, tli, XLOG_FROM_PG_WAL, true);
4373 2704 : if (fd != -1)
4374 : {
4375 2352 : if (!expectedTLEs)
4376 1808 : expectedTLEs = tles;
4377 2352 : return fd;
4378 : }
4379 : }
4380 : }
4381 :
4382 : /* Couldn't find it. For simplicity, complain about front timeline */
4383 334 : XLogFilePath(path, recoveryTargetTLI, segno, wal_segment_size);
4384 334 : errno = ENOENT;
4385 334 : ereport(DEBUG2,
4386 : (errcode_for_file_access(),
4387 : errmsg("could not open file \"%s\": %m", path)));
4388 334 : return -1;
4389 : }
4390 :
4391 : /*
4392 : * Set flag to signal the walreceiver to restart. (The startup process calls
4393 : * this on noticing a relevant configuration change.)
4394 : */
4395 : void
4396 10 : StartupRequestWalReceiverRestart(void)
4397 : {
4398 10 : if (currentSource == XLOG_FROM_STREAM && WalRcvRunning())
4399 : {
4400 8 : ereport(LOG,
4401 : (errmsg("WAL receiver process shutdown requested")));
4402 :
4403 8 : pendingWalRcvRestart = true;
4404 : }
4405 10 : }
4406 :
4407 :
4408 : /*
4409 : * Has a standby promotion already been triggered?
4410 : *
4411 : * Unlike CheckForStandbyTrigger(), this works in any process
4412 : * that's connected to shared memory.
4413 : */
4414 : bool
4415 104 : PromoteIsTriggered(void)
4416 : {
4417 : /*
4418 : * We check shared state each time only until a standby promotion is
4419 : * triggered. We can't trigger a promotion again, so there's no need to
4420 : * keep checking after the shared variable has once been seen true.
4421 : */
4422 104 : if (LocalPromoteIsTriggered)
4423 84 : return true;
4424 :
4425 20 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4426 20 : LocalPromoteIsTriggered = XLogRecoveryCtl->SharedPromoteIsTriggered;
4427 20 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4428 :
4429 20 : return LocalPromoteIsTriggered;
4430 : }
4431 :
4432 : static void
4433 84 : SetPromoteIsTriggered(void)
4434 : {
4435 84 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4436 84 : XLogRecoveryCtl->SharedPromoteIsTriggered = true;
4437 84 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4438 :
4439 : /*
4440 : * Mark the recovery pause state as 'not paused' because the paused state
4441 : * ends and promotion continues if a promotion is triggered while recovery
4442 : * is paused. Otherwise pg_get_wal_replay_pause_state() can mistakenly
4443 : * return 'paused' while a promotion is ongoing.
4444 : */
4445 84 : SetRecoveryPause(false);
4446 :
4447 84 : LocalPromoteIsTriggered = true;
4448 84 : }
4449 :
4450 : /*
4451 : * Check whether a promote request has arrived.
4452 : */
4453 : static bool
4454 9086 : CheckForStandbyTrigger(void)
4455 : {
4456 9086 : if (LocalPromoteIsTriggered)
4457 92 : return true;
4458 :
4459 8994 : if (IsPromoteSignaled() && CheckPromoteSignal())
4460 : {
4461 84 : ereport(LOG, (errmsg("received promote request")));
4462 84 : RemovePromoteSignalFiles();
4463 84 : ResetPromoteSignaled();
4464 84 : SetPromoteIsTriggered();
4465 84 : return true;
4466 : }
4467 :
4468 8910 : return false;
4469 : }
4470 :
4471 : /*
4472 : * Remove the files signaling a standby promotion request.
4473 : */
4474 : void
4475 1784 : RemovePromoteSignalFiles(void)
4476 : {
4477 1784 : unlink(PROMOTE_SIGNAL_FILE);
4478 1784 : }
4479 :
4480 : /*
4481 : * Check to see if a promote request has arrived.
4482 : */
4483 : bool
4484 1666 : CheckPromoteSignal(void)
4485 : {
4486 : struct stat stat_buf;
4487 :
4488 1666 : if (stat(PROMOTE_SIGNAL_FILE, &stat_buf) == 0)
4489 168 : return true;
4490 :
4491 1498 : return false;
4492 : }
4493 :
4494 : /*
4495 : * Wake up startup process to replay newly arrived WAL, or to notice that
4496 : * failover has been requested.
4497 : */
4498 : void
4499 45826 : WakeupRecovery(void)
4500 : {
4501 45826 : SetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
4502 45826 : }
4503 :
4504 : /*
4505 : * Schedule a walreceiver wakeup in the main recovery loop.
4506 : */
4507 : void
4508 4 : XLogRequestWalReceiverReply(void)
4509 : {
4510 4 : doRequestWalReceiverReply = true;
4511 4 : }
4512 :
4513 : /*
4514 : * Is HotStandby active yet? This is only important in special backends
4515 : * since normal backends won't ever be able to connect until this returns
4516 : * true. Postmaster knows this by way of signal, not via shared memory.
4517 : *
4518 : * Unlike testing standbyState, this works in any process that's connected to
4519 : * shared memory. (And note that standbyState alone doesn't tell the truth
4520 : * anyway.)
4521 : */
4522 : bool
4523 314 : HotStandbyActive(void)
4524 : {
4525 : /*
4526 : * We check shared state each time only until Hot Standby is active. We
4527 : * can't de-activate Hot Standby, so there's no need to keep checking
4528 : * after the shared variable has once been seen true.
4529 : */
4530 314 : if (LocalHotStandbyActive)
4531 44 : return true;
4532 : else
4533 : {
4534 : /* spinlock is essential on machines with weak memory ordering! */
4535 270 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4536 270 : LocalHotStandbyActive = XLogRecoveryCtl->SharedHotStandbyActive;
4537 270 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4538 :
4539 270 : return LocalHotStandbyActive;
4540 : }
4541 : }
4542 :
4543 : /*
4544 : * Like HotStandbyActive(), but to be used only in WAL replay code,
4545 : * where we don't need to ask any other process what the state is.
4546 : */
4547 : static bool
4548 0 : HotStandbyActiveInReplay(void)
4549 : {
4550 : Assert(AmStartupProcess() || !IsPostmasterEnvironment);
4551 0 : return LocalHotStandbyActive;
4552 : }
4553 :
4554 : /*
4555 : * Get latest redo apply position.
4556 : *
4557 : * Exported to allow WALReceiver to read the pointer directly.
4558 : */
4559 : XLogRecPtr
4560 116390 : GetXLogReplayRecPtr(TimeLineID *replayTLI)
4561 : {
4562 : XLogRecPtr recptr;
4563 : TimeLineID tli;
4564 :
4565 116390 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4566 116390 : recptr = XLogRecoveryCtl->lastReplayedEndRecPtr;
4567 116390 : tli = XLogRecoveryCtl->lastReplayedTLI;
4568 116390 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4569 :
4570 116390 : if (replayTLI)
4571 4152 : *replayTLI = tli;
4572 116390 : return recptr;
4573 : }
4574 :
4575 :
4576 : /*
4577 : * Get position of last applied, or the record being applied.
4578 : *
4579 : * This is different from GetXLogReplayRecPtr() in that if a WAL
4580 : * record is currently being applied, this includes that record.
4581 : */
4582 : XLogRecPtr
4583 12074 : GetCurrentReplayRecPtr(TimeLineID *replayEndTLI)
4584 : {
4585 : XLogRecPtr recptr;
4586 : TimeLineID tli;
4587 :
4588 12074 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4589 12074 : recptr = XLogRecoveryCtl->replayEndRecPtr;
4590 12074 : tli = XLogRecoveryCtl->replayEndTLI;
4591 12074 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4592 :
4593 12074 : if (replayEndTLI)
4594 12074 : *replayEndTLI = tli;
4595 12074 : return recptr;
4596 : }
4597 :
4598 : /*
4599 : * Save timestamp of latest processed commit/abort record.
4600 : *
4601 : * We keep this in XLogRecoveryCtl, not a simple static variable, so that it can be
4602 : * seen by processes other than the startup process. Note in particular
4603 : * that CreateRestartPoint is executed in the checkpointer.
4604 : */
4605 : static void
4606 42542 : SetLatestXTime(TimestampTz xtime)
4607 : {
4608 42542 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4609 42542 : XLogRecoveryCtl->recoveryLastXTime = xtime;
4610 42542 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4611 42542 : }
4612 :
4613 : /*
4614 : * Fetch timestamp of latest processed commit/abort record.
4615 : */
4616 : TimestampTz
4617 672 : GetLatestXTime(void)
4618 : {
4619 : TimestampTz xtime;
4620 :
4621 672 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4622 672 : xtime = XLogRecoveryCtl->recoveryLastXTime;
4623 672 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4624 :
4625 672 : return xtime;
4626 : }
4627 :
4628 : /*
4629 : * Save timestamp of the next chunk of WAL records to apply.
4630 : *
4631 : * We keep this in XLogRecoveryCtl, not a simple static variable, so that it can be
4632 : * seen by all backends.
4633 : */
4634 : static void
4635 6720 : SetCurrentChunkStartTime(TimestampTz xtime)
4636 : {
4637 6720 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4638 6720 : XLogRecoveryCtl->currentChunkStartTime = xtime;
4639 6720 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4640 6720 : }
4641 :
4642 : /*
4643 : * Fetch timestamp of latest processed commit/abort record.
4644 : * Startup process maintains an accurate local copy in XLogReceiptTime
4645 : */
4646 : TimestampTz
4647 538 : GetCurrentChunkReplayStartTime(void)
4648 : {
4649 : TimestampTz xtime;
4650 :
4651 538 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4652 538 : xtime = XLogRecoveryCtl->currentChunkStartTime;
4653 538 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4654 :
4655 538 : return xtime;
4656 : }
4657 :
4658 : /*
4659 : * Returns time of receipt of current chunk of XLOG data, as well as
4660 : * whether it was received from streaming replication or from archives.
4661 : */
4662 : void
4663 56 : GetXLogReceiptTime(TimestampTz *rtime, bool *fromStream)
4664 : {
4665 : /*
4666 : * This must be executed in the startup process, since we don't export the
4667 : * relevant state to shared memory.
4668 : */
4669 : Assert(InRecovery);
4670 :
4671 56 : *rtime = XLogReceiptTime;
4672 56 : *fromStream = (XLogReceiptSource == XLOG_FROM_STREAM);
4673 56 : }
4674 :
4675 : /*
4676 : * Note that text field supplied is a parameter name and does not require
4677 : * translation
4678 : */
4679 : void
4680 1230 : RecoveryRequiresIntParameter(const char *param_name, int currValue, int minValue)
4681 : {
4682 1230 : if (currValue < minValue)
4683 : {
4684 0 : if (HotStandbyActiveInReplay())
4685 : {
4686 0 : bool warned_for_promote = false;
4687 :
4688 0 : ereport(WARNING,
4689 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4690 : errmsg("hot standby is not possible because of insufficient parameter settings"),
4691 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4692 : param_name,
4693 : currValue,
4694 : minValue)));
4695 :
4696 0 : SetRecoveryPause(true);
4697 :
4698 0 : ereport(LOG,
4699 : (errmsg("recovery has paused"),
4700 : errdetail("If recovery is unpaused, the server will shut down."),
4701 : errhint("You can then restart the server after making the necessary configuration changes.")));
4702 :
4703 0 : while (GetRecoveryPauseState() != RECOVERY_NOT_PAUSED)
4704 : {
4705 0 : ProcessStartupProcInterrupts();
4706 :
4707 0 : if (CheckForStandbyTrigger())
4708 : {
4709 0 : if (!warned_for_promote)
4710 0 : ereport(WARNING,
4711 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4712 : errmsg("promotion is not possible because of insufficient parameter settings"),
4713 :
4714 : /*
4715 : * Repeat the detail from above so it's easy to find
4716 : * in the log.
4717 : */
4718 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4719 : param_name,
4720 : currValue,
4721 : minValue),
4722 : errhint("Restart the server after making the necessary configuration changes.")));
4723 0 : warned_for_promote = true;
4724 : }
4725 :
4726 : /*
4727 : * If recovery pause is requested then set it paused. While
4728 : * we are in the loop, user might resume and pause again so
4729 : * set this every time.
4730 : */
4731 0 : ConfirmRecoveryPaused();
4732 :
4733 : /*
4734 : * We wait on a condition variable that will wake us as soon
4735 : * as the pause ends, but we use a timeout so we can check the
4736 : * above conditions periodically too.
4737 : */
4738 0 : ConditionVariableTimedSleep(&XLogRecoveryCtl->recoveryNotPausedCV, 1000,
4739 : WAIT_EVENT_RECOVERY_PAUSE);
4740 : }
4741 0 : ConditionVariableCancelSleep();
4742 : }
4743 :
4744 0 : ereport(FATAL,
4745 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4746 : errmsg("recovery aborted because of insufficient parameter settings"),
4747 : /* Repeat the detail from above so it's easy to find in the log. */
4748 : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4749 : param_name,
4750 : currValue,
4751 : minValue),
4752 : errhint("You can restart the server after making the necessary configuration changes.")));
4753 : }
4754 1230 : }
4755 :
4756 :
4757 : /*
4758 : * GUC check_hook for primary_slot_name
4759 : */
4760 : bool
4761 2598 : check_primary_slot_name(char **newval, void **extra, GucSource source)
4762 : {
4763 2598 : if (*newval && strcmp(*newval, "") != 0 &&
4764 390 : !ReplicationSlotValidateName(*newval, false, WARNING))
4765 0 : return false;
4766 :
4767 2598 : return true;
4768 : }
4769 :
4770 : /*
4771 : * Recovery target settings: Only one of the several recovery_target* settings
4772 : * may be set. Setting a second one results in an error. The global variable
4773 : * recoveryTarget tracks which kind of recovery target was chosen. Other
4774 : * variables store the actual target value (for example a string or a xid).
4775 : * The assign functions of the parameters check whether a competing parameter
4776 : * was already set. But we want to allow setting the same parameter multiple
4777 : * times. We also want to allow unsetting a parameter and setting a different
4778 : * one, so we unset recoveryTarget when the parameter is set to an empty
4779 : * string.
4780 : *
4781 : * XXX this code is broken by design. Throwing an error from a GUC assign
4782 : * hook breaks fundamental assumptions of guc.c. So long as all the variables
4783 : * for which this can happen are PGC_POSTMASTER, the consequences are limited,
4784 : * since we'd just abort postmaster startup anyway. Nonetheless it's likely
4785 : * that we have odd behaviors such as unexpected GUC ordering dependencies.
4786 : */
4787 :
4788 : pg_noreturn static void
4789 2 : error_multiple_recovery_targets(void)
4790 : {
4791 2 : ereport(ERROR,
4792 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4793 : errmsg("multiple recovery targets specified"),
4794 : errdetail("At most one of \"recovery_target\", \"recovery_target_lsn\", \"recovery_target_name\", \"recovery_target_time\", \"recovery_target_xid\" may be set.")));
4795 : }
4796 :
4797 : /*
4798 : * GUC check_hook for recovery_target
4799 : */
4800 : bool
4801 2210 : check_recovery_target(char **newval, void **extra, GucSource source)
4802 : {
4803 2210 : if (strcmp(*newval, "immediate") != 0 && strcmp(*newval, "") != 0)
4804 : {
4805 0 : GUC_check_errdetail("The only allowed value is \"immediate\".");
4806 0 : return false;
4807 : }
4808 2210 : return true;
4809 : }
4810 :
4811 : /*
4812 : * GUC assign_hook for recovery_target
4813 : */
4814 : void
4815 2210 : assign_recovery_target(const char *newval, void *extra)
4816 : {
4817 2210 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4818 0 : recoveryTarget != RECOVERY_TARGET_IMMEDIATE)
4819 0 : error_multiple_recovery_targets();
4820 :
4821 2210 : if (newval && strcmp(newval, "") != 0)
4822 2 : recoveryTarget = RECOVERY_TARGET_IMMEDIATE;
4823 : else
4824 2208 : recoveryTarget = RECOVERY_TARGET_UNSET;
4825 2210 : }
4826 :
4827 : /*
4828 : * GUC check_hook for recovery_target_lsn
4829 : */
4830 : bool
4831 2222 : check_recovery_target_lsn(char **newval, void **extra, GucSource source)
4832 : {
4833 2222 : if (strcmp(*newval, "") != 0)
4834 : {
4835 : XLogRecPtr lsn;
4836 : XLogRecPtr *myextra;
4837 18 : bool have_error = false;
4838 :
4839 18 : lsn = pg_lsn_in_internal(*newval, &have_error);
4840 18 : if (have_error)
4841 0 : return false;
4842 :
4843 18 : myextra = (XLogRecPtr *) guc_malloc(LOG, sizeof(XLogRecPtr));
4844 18 : if (!myextra)
4845 0 : return false;
4846 18 : *myextra = lsn;
4847 18 : *extra = myextra;
4848 : }
4849 2222 : return true;
4850 : }
4851 :
4852 : /*
4853 : * GUC assign_hook for recovery_target_lsn
4854 : */
4855 : void
4856 2222 : assign_recovery_target_lsn(const char *newval, void *extra)
4857 : {
4858 2222 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4859 0 : recoveryTarget != RECOVERY_TARGET_LSN)
4860 0 : error_multiple_recovery_targets();
4861 :
4862 2222 : if (newval && strcmp(newval, "") != 0)
4863 : {
4864 18 : recoveryTarget = RECOVERY_TARGET_LSN;
4865 18 : recoveryTargetLSN = *((XLogRecPtr *) extra);
4866 : }
4867 : else
4868 2204 : recoveryTarget = RECOVERY_TARGET_UNSET;
4869 2222 : }
4870 :
4871 : /*
4872 : * GUC check_hook for recovery_target_name
4873 : */
4874 : bool
4875 2222 : check_recovery_target_name(char **newval, void **extra, GucSource source)
4876 : {
4877 : /* Use the value of newval directly */
4878 2222 : if (strlen(*newval) >= MAXFNAMELEN)
4879 : {
4880 0 : GUC_check_errdetail("\"%s\" is too long (maximum %d characters).",
4881 : "recovery_target_name", MAXFNAMELEN - 1);
4882 0 : return false;
4883 : }
4884 2222 : return true;
4885 : }
4886 :
4887 : /*
4888 : * GUC assign_hook for recovery_target_name
4889 : */
4890 : void
4891 2222 : assign_recovery_target_name(const char *newval, void *extra)
4892 : {
4893 2222 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4894 0 : recoveryTarget != RECOVERY_TARGET_NAME)
4895 0 : error_multiple_recovery_targets();
4896 :
4897 2222 : if (newval && strcmp(newval, "") != 0)
4898 : {
4899 12 : recoveryTarget = RECOVERY_TARGET_NAME;
4900 12 : recoveryTargetName = newval;
4901 : }
4902 : else
4903 2210 : recoveryTarget = RECOVERY_TARGET_UNSET;
4904 2222 : }
4905 :
4906 : /*
4907 : * GUC check_hook for recovery_target_time
4908 : *
4909 : * The interpretation of the recovery_target_time string can depend on the
4910 : * time zone setting, so we need to wait until after all GUC processing is
4911 : * done before we can do the final parsing of the string. This check function
4912 : * only does a parsing pass to catch syntax errors, but we store the string
4913 : * and parse it again when we need to use it.
4914 : */
4915 : bool
4916 2214 : check_recovery_target_time(char **newval, void **extra, GucSource source)
4917 : {
4918 2214 : if (strcmp(*newval, "") != 0)
4919 : {
4920 : /* reject some special values */
4921 6 : if (strcmp(*newval, "now") == 0 ||
4922 6 : strcmp(*newval, "today") == 0 ||
4923 6 : strcmp(*newval, "tomorrow") == 0 ||
4924 6 : strcmp(*newval, "yesterday") == 0)
4925 : {
4926 0 : return false;
4927 : }
4928 :
4929 : /*
4930 : * parse timestamp value (see also timestamptz_in())
4931 : */
4932 : {
4933 6 : char *str = *newval;
4934 : fsec_t fsec;
4935 : struct pg_tm tt,
4936 6 : *tm = &tt;
4937 : int tz;
4938 : int dtype;
4939 : int nf;
4940 : int dterr;
4941 : char *field[MAXDATEFIELDS];
4942 : int ftype[MAXDATEFIELDS];
4943 : char workbuf[MAXDATELEN + MAXDATEFIELDS];
4944 : DateTimeErrorExtra dtextra;
4945 : TimestampTz timestamp;
4946 :
4947 6 : dterr = ParseDateTime(str, workbuf, sizeof(workbuf),
4948 : field, ftype, MAXDATEFIELDS, &nf);
4949 6 : if (dterr == 0)
4950 6 : dterr = DecodeDateTime(field, ftype, nf,
4951 : &dtype, tm, &fsec, &tz, &dtextra);
4952 6 : if (dterr != 0)
4953 0 : return false;
4954 6 : if (dtype != DTK_DATE)
4955 0 : return false;
4956 :
4957 6 : if (tm2timestamp(tm, fsec, &tz, ×tamp) != 0)
4958 : {
4959 0 : GUC_check_errdetail("Timestamp out of range: \"%s\".", str);
4960 0 : return false;
4961 : }
4962 : }
4963 : }
4964 2214 : return true;
4965 : }
4966 :
4967 : /*
4968 : * GUC assign_hook for recovery_target_time
4969 : */
4970 : void
4971 2214 : assign_recovery_target_time(const char *newval, void *extra)
4972 : {
4973 2214 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4974 2 : recoveryTarget != RECOVERY_TARGET_TIME)
4975 2 : error_multiple_recovery_targets();
4976 :
4977 2212 : if (newval && strcmp(newval, "") != 0)
4978 4 : recoveryTarget = RECOVERY_TARGET_TIME;
4979 : else
4980 2208 : recoveryTarget = RECOVERY_TARGET_UNSET;
4981 2212 : }
4982 :
4983 : /*
4984 : * GUC check_hook for recovery_target_timeline
4985 : */
4986 : bool
4987 2216 : check_recovery_target_timeline(char **newval, void **extra, GucSource source)
4988 : {
4989 : RecoveryTargetTimeLineGoal rttg;
4990 : RecoveryTargetTimeLineGoal *myextra;
4991 :
4992 2216 : if (strcmp(*newval, "current") == 0)
4993 0 : rttg = RECOVERY_TARGET_TIMELINE_CONTROLFILE;
4994 2216 : else if (strcmp(*newval, "latest") == 0)
4995 2210 : rttg = RECOVERY_TARGET_TIMELINE_LATEST;
4996 : else
4997 : {
4998 : char *endp;
4999 : uint64 timeline;
5000 :
5001 6 : rttg = RECOVERY_TARGET_TIMELINE_NUMERIC;
5002 :
5003 6 : errno = 0;
5004 6 : timeline = strtou64(*newval, &endp, 0);
5005 :
5006 6 : if (*endp != '\0' || errno == EINVAL || errno == ERANGE)
5007 : {
5008 2 : GUC_check_errdetail("\"%s\" is not a valid number.",
5009 : "recovery_target_timeline");
5010 6 : return false;
5011 : }
5012 :
5013 4 : if (timeline < 1 || timeline > PG_UINT32_MAX)
5014 : {
5015 4 : GUC_check_errdetail("\"%s\" must be between %u and %u.",
5016 : "recovery_target_timeline", 1, UINT_MAX);
5017 4 : return false;
5018 : }
5019 : }
5020 :
5021 2210 : myextra = (RecoveryTargetTimeLineGoal *) guc_malloc(LOG, sizeof(RecoveryTargetTimeLineGoal));
5022 2210 : if (!myextra)
5023 0 : return false;
5024 2210 : *myextra = rttg;
5025 2210 : *extra = myextra;
5026 :
5027 2210 : return true;
5028 : }
5029 :
5030 : /*
5031 : * GUC assign_hook for recovery_target_timeline
5032 : */
5033 : void
5034 2210 : assign_recovery_target_timeline(const char *newval, void *extra)
5035 : {
5036 2210 : recoveryTargetTimeLineGoal = *((RecoveryTargetTimeLineGoal *) extra);
5037 2210 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
5038 0 : recoveryTargetTLIRequested = (TimeLineID) strtoul(newval, NULL, 0);
5039 : else
5040 2210 : recoveryTargetTLIRequested = 0;
5041 2210 : }
5042 :
5043 : /*
5044 : * GUC check_hook for recovery_target_xid
5045 : */
5046 : bool
5047 2210 : check_recovery_target_xid(char **newval, void **extra, GucSource source)
5048 : {
5049 2210 : if (strcmp(*newval, "") != 0)
5050 : {
5051 : TransactionId xid;
5052 : TransactionId *myextra;
5053 :
5054 2 : errno = 0;
5055 2 : xid = (TransactionId) strtou64(*newval, NULL, 0);
5056 2 : if (errno == EINVAL || errno == ERANGE)
5057 0 : return false;
5058 :
5059 2 : myextra = (TransactionId *) guc_malloc(LOG, sizeof(TransactionId));
5060 2 : if (!myextra)
5061 0 : return false;
5062 2 : *myextra = xid;
5063 2 : *extra = myextra;
5064 : }
5065 2210 : return true;
5066 : }
5067 :
5068 : /*
5069 : * GUC assign_hook for recovery_target_xid
5070 : */
5071 : void
5072 2210 : assign_recovery_target_xid(const char *newval, void *extra)
5073 : {
5074 2210 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
5075 0 : recoveryTarget != RECOVERY_TARGET_XID)
5076 0 : error_multiple_recovery_targets();
5077 :
5078 2210 : if (newval && strcmp(newval, "") != 0)
5079 : {
5080 2 : recoveryTarget = RECOVERY_TARGET_XID;
5081 2 : recoveryTargetXid = *((TransactionId *) extra);
5082 : }
5083 : else
5084 2208 : recoveryTarget = RECOVERY_TARGET_UNSET;
5085 2210 : }
|
