PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
slot.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * slot.c
4  * Replication slot management.
5  *
6  *
7  * Copyright (c) 2012-2017, PostgreSQL Global Development Group
8  *
9  *
10  * IDENTIFICATION
11  * src/backend/replication/slot.c
12  *
13  * NOTES
14  *
15  * Replication slots are used to keep state about replication streams
16  * originating from this cluster. Their primary purpose is to prevent the
17  * premature removal of WAL or of old tuple versions in a manner that would
18  * interfere with replication; they are also useful for monitoring purposes.
19  * Slots need to be permanent (to allow restarts), crash-safe, and allocatable
20  * on standbys (to support cascading setups). The requirement that slots be
21  * usable on standbys precludes storing them in the system catalogs.
22  *
23  * Each replication slot gets its own directory inside the $PGDATA/pg_replslot
24  * directory. Inside that directory the state file will contain the slot's
25  * own data. Additional data can be stored alongside that file if required.
26  * While the server is running, the state data is also cached in memory for
27  * efficiency.
28  *
29  * ReplicationSlotAllocationLock must be taken in exclusive mode to allocate
30  * or free a slot. ReplicationSlotControlLock must be taken in shared mode
31  * to iterate over the slots, and in exclusive mode to change the in_use flag
32  * of a slot. The remaining data in each slot is protected by its mutex.
33  *
34  *-------------------------------------------------------------------------
35  */
36 
37 #include "postgres.h"
38 
39 #include <unistd.h>
40 #include <sys/stat.h>
41 
42 #include "access/transam.h"
43 #include "access/xlog_internal.h"
44 #include "common/string.h"
45 #include "miscadmin.h"
46 #include "pgstat.h"
47 #include "replication/slot.h"
48 #include "storage/fd.h"
49 #include "storage/proc.h"
50 #include "storage/procarray.h"
51 #include "utils/builtins.h"
52 
53 /*
54  * Replication slot on-disk data structure.
55  */
56 typedef struct ReplicationSlotOnDisk
57 {
58  /* first part of this struct needs to be version independent */
59 
60  /* data not covered by checksum */
63 
64  /* data covered by checksum */
67 
68  /*
69  * The actual data in the slot that follows can differ based on the above
70  * 'version'.
71  */
72 
75 
76 /* size of version independent data */
77 #define ReplicationSlotOnDiskConstantSize \
78  offsetof(ReplicationSlotOnDisk, slotdata)
79 /* size of the part of the slot not covered by the checksum */
80 #define SnapBuildOnDiskNotChecksummedSize \
81  offsetof(ReplicationSlotOnDisk, version)
82 /* size of the part covered by the checksum */
83 #define SnapBuildOnDiskChecksummedSize \
84  sizeof(ReplicationSlotOnDisk) - SnapBuildOnDiskNotChecksummedSize
85 /* size of the slot data that is version dependent */
86 #define ReplicationSlotOnDiskV2Size \
87  sizeof(ReplicationSlotOnDisk) - ReplicationSlotOnDiskConstantSize
88 
89 #define SLOT_MAGIC 0x1051CA1 /* format identifier */
90 #define SLOT_VERSION 2 /* version for new files */
91 
92 /* Control array for replication slot management */
94 
95 /* My backend's replication slot in the shared memory array */
97 
98 /* GUCs */
99 int max_replication_slots = 0; /* the maximum number of replication
100  * slots */
101 
102 static void ReplicationSlotDropAcquired(void);
103 static void ReplicationSlotDropPtr(ReplicationSlot *slot);
104 
105 /* internal persistency functions */
106 static void RestoreSlotFromDisk(const char *name);
107 static void CreateSlotOnDisk(ReplicationSlot *slot);
108 static void SaveSlotToPath(ReplicationSlot *slot, const char *path, int elevel);
109 
110 /*
111  * Report shared-memory space needed by ReplicationSlotShmemInit.
112  */
113 Size
115 {
116  Size size = 0;
117 
118  if (max_replication_slots == 0)
119  return size;
120 
121  size = offsetof(ReplicationSlotCtlData, replication_slots);
122  size = add_size(size,
124 
125  return size;
126 }
127 
128 /*
129  * Allocate and initialize walsender-related shared memory.
130  */
131 void
133 {
134  bool found;
135 
136  if (max_replication_slots == 0)
137  return;
138 
139  ReplicationSlotCtl = (ReplicationSlotCtlData *)
140  ShmemInitStruct("ReplicationSlot Ctl", ReplicationSlotsShmemSize(),
141  &found);
142 
144  "replication_slot_io");
145 
146  if (!found)
147  {
148  int i;
149 
150  /* First time through, so initialize */
151  MemSet(ReplicationSlotCtl, 0, ReplicationSlotsShmemSize());
152 
153  for (i = 0; i < max_replication_slots; i++)
154  {
155  ReplicationSlot *slot = &ReplicationSlotCtl->replication_slots[i];
156 
157  /* everything else is zeroed by the memset above */
158  SpinLockInit(&slot->mutex);
161  }
162  }
163 }
164 
165 /*
166  * Check whether the passed slot name is valid and report errors at elevel.
167  *
168  * Slot names may consist out of [a-z0-9_]{1,NAMEDATALEN-1} which should allow
169  * the name to be used as a directory name on every supported OS.
170  *
171  * Returns whether the directory name is valid or not if elevel < ERROR.
172  */
173 bool
175 {
176  const char *cp;
177 
178  if (strlen(name) == 0)
179  {
180  ereport(elevel,
181  (errcode(ERRCODE_INVALID_NAME),
182  errmsg("replication slot name \"%s\" is too short",
183  name)));
184  return false;
185  }
186 
187  if (strlen(name) >= NAMEDATALEN)
188  {
189  ereport(elevel,
190  (errcode(ERRCODE_NAME_TOO_LONG),
191  errmsg("replication slot name \"%s\" is too long",
192  name)));
193  return false;
194  }
195 
196  for (cp = name; *cp; cp++)
197  {
198  if (!((*cp >= 'a' && *cp <= 'z')
199  || (*cp >= '0' && *cp <= '9')
200  || (*cp == '_')))
201  {
202  ereport(elevel,
203  (errcode(ERRCODE_INVALID_NAME),
204  errmsg("replication slot name \"%s\" contains invalid character",
205  name),
206  errhint("Replication slot names may only contain lower case letters, numbers, and the underscore character.")));
207  return false;
208  }
209  }
210  return true;
211 }
212 
213 /*
214  * Create a new replication slot and mark it as used by this backend.
215  *
216  * name: Name of the slot
217  * db_specific: logical decoding is db specific; if the slot is going to
218  * be used for that pass true, otherwise false.
219  */
220 void
221 ReplicationSlotCreate(const char *name, bool db_specific,
222  ReplicationSlotPersistency persistency)
223 {
224  ReplicationSlot *slot = NULL;
225  int i;
226 
227  Assert(MyReplicationSlot == NULL);
228 
230 
231  /*
232  * If some other backend ran this code concurrently with us, we'd likely
233  * both allocate the same slot, and that would be bad. We'd also be at
234  * risk of missing a name collision. Also, we don't want to try to create
235  * a new slot while somebody's busy cleaning up an old one, because we
236  * might both be monkeying with the same directory.
237  */
238  LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
239 
240  /*
241  * Check for name collision, and identify an allocatable slot. We need to
242  * hold ReplicationSlotControlLock in shared mode for this, so that nobody
243  * else can change the in_use flags while we're looking at them.
244  */
245  LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
246  for (i = 0; i < max_replication_slots; i++)
247  {
248  ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
249 
250  if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
251  ereport(ERROR,
253  errmsg("replication slot \"%s\" already exists", name)));
254  if (!s->in_use && slot == NULL)
255  slot = s;
256  }
257  LWLockRelease(ReplicationSlotControlLock);
258 
259  /* If all slots are in use, we're out of luck. */
260  if (slot == NULL)
261  ereport(ERROR,
262  (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
263  errmsg("all replication slots are in use"),
264  errhint("Free one or increase max_replication_slots.")));
265 
266  /*
267  * Since this slot is not in use, nobody should be looking at any part of
268  * it other than the in_use field unless they're trying to allocate it.
269  * And since we hold ReplicationSlotAllocationLock, nobody except us can
270  * be doing that. So it's safe to initialize the slot.
271  */
272  Assert(!slot->in_use);
273  Assert(slot->active_pid == 0);
274 
275  /* first initialize persistent data */
276  memset(&slot->data, 0, sizeof(ReplicationSlotPersistentData));
277  StrNCpy(NameStr(slot->data.name), name, NAMEDATALEN);
278  slot->data.database = db_specific ? MyDatabaseId : InvalidOid;
279  slot->data.persistency = persistency;
280 
281  /* and then data only present in shared memory */
282  slot->just_dirtied = false;
283  slot->dirty = false;
290 
291  /*
292  * Create the slot on disk. We haven't actually marked the slot allocated
293  * yet, so no special cleanup is required if this errors out.
294  */
295  CreateSlotOnDisk(slot);
296 
297  /*
298  * We need to briefly prevent any other backend from iterating over the
299  * slots while we flip the in_use flag. We also need to set the active
300  * flag while holding the ControlLock as otherwise a concurrent
301  * SlotAcquire() could acquire the slot as well.
302  */
303  LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
304 
305  slot->in_use = true;
306 
307  /* We can now mark the slot active, and that makes it our slot. */
308  SpinLockAcquire(&slot->mutex);
309  Assert(slot->active_pid == 0);
310  slot->active_pid = MyProcPid;
311  SpinLockRelease(&slot->mutex);
312  MyReplicationSlot = slot;
313 
314  LWLockRelease(ReplicationSlotControlLock);
315 
316  /*
317  * Now that the slot has been marked as in_use and active, it's safe to
318  * let somebody else try to allocate a slot.
319  */
320  LWLockRelease(ReplicationSlotAllocationLock);
321 
322  /* Let everybody know we've modified this slot */
324 }
325 
326 /*
327  * Find a previously created slot and mark it as used by this backend.
328  */
329 void
330 ReplicationSlotAcquire(const char *name, bool nowait)
331 {
332  ReplicationSlot *slot;
333  int active_pid;
334  int i;
335 
336 retry:
337  Assert(MyReplicationSlot == NULL);
338 
339  /*
340  * Search for the named slot and mark it active if we find it. If the
341  * slot is already active, we exit the loop with active_pid set to the PID
342  * of the backend that owns it.
343  */
344  active_pid = 0;
345  slot = NULL;
346  LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
347  for (i = 0; i < max_replication_slots; i++)
348  {
349  ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
350 
351  if (s->in_use && strcmp(name, NameStr(s->data.name)) == 0)
352  {
353  /*
354  * This is the slot we want. We don't know yet if it's active, so
355  * get ready to sleep on it in case it is. (We may end up not
356  * sleeping, but we don't want to do this while holding the
357  * spinlock.)
358  */
360 
361  SpinLockAcquire(&s->mutex);
362 
363  active_pid = s->active_pid;
364  if (active_pid == 0)
365  active_pid = s->active_pid = MyProcPid;
366 
367  SpinLockRelease(&s->mutex);
368  slot = s;
369 
370  break;
371  }
372  }
373  LWLockRelease(ReplicationSlotControlLock);
374 
375  /* If we did not find the slot, error out. */
376  if (slot == NULL)
377  ereport(ERROR,
378  (errcode(ERRCODE_UNDEFINED_OBJECT),
379  errmsg("replication slot \"%s\" does not exist", name)));
380 
381  /*
382  * If we found the slot but it's already active in another backend, we
383  * either error out or retry after a short wait, as caller specified.
384  */
385  if (active_pid != MyProcPid)
386  {
387  if (nowait)
388  ereport(ERROR,
389  (errcode(ERRCODE_OBJECT_IN_USE),
390  errmsg("replication slot \"%s\" is active for PID %d",
391  name, active_pid)));
392 
393  /* Wait here until we get signaled, and then restart */
397  goto retry;
398  }
399  else
400  ConditionVariableCancelSleep(); /* no sleep needed after all */
401 
402  /* Let everybody know we've modified this slot */
404 
405  /* We made this slot active, so it's ours now. */
406  MyReplicationSlot = slot;
407 }
408 
409 /*
410  * Release the replication slot that this backend considers to own.
411  *
412  * This or another backend can re-acquire the slot later.
413  * Resources this slot requires will be preserved.
414  */
415 void
417 {
419 
420  Assert(slot != NULL && slot->active_pid != 0);
421 
422  if (slot->data.persistency == RS_EPHEMERAL)
423  {
424  /*
425  * Delete the slot. There is no !PANIC case where this is allowed to
426  * fail, all that may happen is an incomplete cleanup of the on-disk
427  * data.
428  */
430  }
431 
432  /*
433  * If slot needed to temporarily restrain both data and catalog xmin to
434  * create the catalog snapshot, remove that temporary constraint.
435  * Snapshots can only be exported while the initial snapshot is still
436  * acquired.
437  */
438  if (!TransactionIdIsValid(slot->data.xmin) &&
440  {
441  SpinLockAcquire(&slot->mutex);
443  SpinLockRelease(&slot->mutex);
445  }
446 
447  if (slot->data.persistency == RS_PERSISTENT)
448  {
449  /*
450  * Mark persistent slot inactive. We're not freeing it, just
451  * disconnecting, but wake up others that may be waiting for it.
452  */
453  SpinLockAcquire(&slot->mutex);
454  slot->active_pid = 0;
455  SpinLockRelease(&slot->mutex);
457  }
458 
459  MyReplicationSlot = NULL;
460 
461  /* might not have been set when we've been a plain slot */
462  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
464  LWLockRelease(ProcArrayLock);
465 }
466 
467 /*
468  * Cleanup all temporary slots created in current session.
469  */
470 void
472 {
473  int i;
474 
475  Assert(MyReplicationSlot == NULL);
476 
477 restart:
478  LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
479  for (i = 0; i < max_replication_slots; i++)
480  {
481  ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
482 
483  if (!s->in_use)
484  continue;
485 
486  SpinLockAcquire(&s->mutex);
487  if (s->active_pid == MyProcPid)
488  {
490  SpinLockRelease(&s->mutex);
491  LWLockRelease(ReplicationSlotControlLock); /* avoid deadlock */
492 
494 
496  goto restart;
497  }
498  else
499  SpinLockRelease(&s->mutex);
500  }
501 
502  LWLockRelease(ReplicationSlotControlLock);
503 }
504 
505 /*
506  * Permanently drop replication slot identified by the passed in name.
507  */
508 void
509 ReplicationSlotDrop(const char *name, bool nowait)
510 {
511  Assert(MyReplicationSlot == NULL);
512 
513  ReplicationSlotAcquire(name, nowait);
514 
516 }
517 
518 /*
519  * Permanently drop the currently acquired replication slot.
520  */
521 static void
523 {
525 
526  Assert(MyReplicationSlot != NULL);
527 
528  /* slot isn't acquired anymore */
529  MyReplicationSlot = NULL;
530 
532 }
533 
534 /*
535  * Permanently drop the replication slot which will be released by the point
536  * this function returns.
537  */
538 static void
540 {
541  char path[MAXPGPATH];
542  char tmppath[MAXPGPATH];
543 
544  /*
545  * If some other backend ran this code concurrently with us, we might try
546  * to delete a slot with a certain name while someone else was trying to
547  * create a slot with the same name.
548  */
549  LWLockAcquire(ReplicationSlotAllocationLock, LW_EXCLUSIVE);
550 
551  /* Generate pathnames. */
552  sprintf(path, "pg_replslot/%s", NameStr(slot->data.name));
553  sprintf(tmppath, "pg_replslot/%s.tmp", NameStr(slot->data.name));
554 
555  /*
556  * Rename the slot directory on disk, so that we'll no longer recognize
557  * this as a valid slot. Note that if this fails, we've got to mark the
558  * slot inactive before bailing out. If we're dropping an ephemeral or a
559  * temporary slot, we better never fail hard as the caller won't expect
560  * the slot to survive and this might get called during error handling.
561  */
562  if (rename(path, tmppath) == 0)
563  {
564  /*
565  * We need to fsync() the directory we just renamed and its parent to
566  * make sure that our changes are on disk in a crash-safe fashion. If
567  * fsync() fails, we can't be sure whether the changes are on disk or
568  * not. For now, we handle that by panicking;
569  * StartupReplicationSlots() will try to straighten it out after
570  * restart.
571  */
573  fsync_fname(tmppath, true);
574  fsync_fname("pg_replslot", true);
576  }
577  else
578  {
579  bool fail_softly = slot->data.persistency != RS_PERSISTENT;
580 
581  SpinLockAcquire(&slot->mutex);
582  slot->active_pid = 0;
583  SpinLockRelease(&slot->mutex);
584 
585  /* wake up anyone waiting on this slot */
587 
588  ereport(fail_softly ? WARNING : ERROR,
590  errmsg("could not rename file \"%s\" to \"%s\": %m",
591  path, tmppath)));
592  }
593 
594  /*
595  * The slot is definitely gone. Lock out concurrent scans of the array
596  * long enough to kill it. It's OK to clear the active PID here without
597  * grabbing the mutex because nobody else can be scanning the array here,
598  * and nobody can be attached to this slot and thus access it without
599  * scanning the array.
600  *
601  * Also wake up processes waiting for it.
602  */
603  LWLockAcquire(ReplicationSlotControlLock, LW_EXCLUSIVE);
604  slot->active_pid = 0;
605  slot->in_use = false;
606  LWLockRelease(ReplicationSlotControlLock);
608 
609  /*
610  * Slot is dead and doesn't prevent resource removal anymore, recompute
611  * limits.
612  */
615 
616  /*
617  * If removing the directory fails, the worst thing that will happen is
618  * that the user won't be able to create a new slot with the same name
619  * until the next server restart. We warn about it, but that's all.
620  */
621  if (!rmtree(tmppath, true))
624  errmsg("could not remove directory \"%s\"", tmppath)));
625 
626  /*
627  * We release this at the very end, so that nobody starts trying to create
628  * a slot while we're still cleaning up the detritus of the old one.
629  */
630  LWLockRelease(ReplicationSlotAllocationLock);
631 }
632 
633 /*
634  * Serialize the currently acquired slot's state from memory to disk, thereby
635  * guaranteeing the current state will survive a crash.
636  */
637 void
639 {
640  char path[MAXPGPATH];
641 
642  Assert(MyReplicationSlot != NULL);
643 
644  sprintf(path, "pg_replslot/%s", NameStr(MyReplicationSlot->data.name));
645  SaveSlotToPath(MyReplicationSlot, path, ERROR);
646 }
647 
648 /*
649  * Signal that it would be useful if the currently acquired slot would be
650  * flushed out to disk.
651  *
652  * Note that the actual flush to disk can be delayed for a long time, if
653  * required for correctness explicitly do a ReplicationSlotSave().
654  */
655 void
657 {
659 
660  Assert(MyReplicationSlot != NULL);
661 
662  SpinLockAcquire(&slot->mutex);
663  MyReplicationSlot->just_dirtied = true;
664  MyReplicationSlot->dirty = true;
665  SpinLockRelease(&slot->mutex);
666 }
667 
668 /*
669  * Convert a slot that's marked as RS_EPHEMERAL to a RS_PERSISTENT slot,
670  * guaranteeing it will be there after an eventual crash.
671  */
672 void
674 {
676 
677  Assert(slot != NULL);
679 
680  SpinLockAcquire(&slot->mutex);
682  SpinLockRelease(&slot->mutex);
683 
686 }
687 
688 /*
689  * Compute the oldest xmin across all slots and store it in the ProcArray.
690  *
691  * If already_locked is true, ProcArrayLock has already been acquired
692  * exclusively.
693  */
694 void
696 {
697  int i;
699  TransactionId agg_catalog_xmin = InvalidTransactionId;
700 
701  Assert(ReplicationSlotCtl != NULL);
702 
703  LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
704 
705  for (i = 0; i < max_replication_slots; i++)
706  {
707  ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
708  TransactionId effective_xmin;
709  TransactionId effective_catalog_xmin;
710 
711  if (!s->in_use)
712  continue;
713 
714  SpinLockAcquire(&s->mutex);
715  effective_xmin = s->effective_xmin;
716  effective_catalog_xmin = s->effective_catalog_xmin;
717  SpinLockRelease(&s->mutex);
718 
719  /* check the data xmin */
720  if (TransactionIdIsValid(effective_xmin) &&
721  (!TransactionIdIsValid(agg_xmin) ||
722  TransactionIdPrecedes(effective_xmin, agg_xmin)))
723  agg_xmin = effective_xmin;
724 
725  /* check the catalog xmin */
726  if (TransactionIdIsValid(effective_catalog_xmin) &&
727  (!TransactionIdIsValid(agg_catalog_xmin) ||
728  TransactionIdPrecedes(effective_catalog_xmin, agg_catalog_xmin)))
729  agg_catalog_xmin = effective_catalog_xmin;
730  }
731 
732  LWLockRelease(ReplicationSlotControlLock);
733 
734  ProcArraySetReplicationSlotXmin(agg_xmin, agg_catalog_xmin, already_locked);
735 }
736 
737 /*
738  * Compute the oldest restart LSN across all slots and inform xlog module.
739  */
740 void
742 {
743  int i;
744  XLogRecPtr min_required = InvalidXLogRecPtr;
745 
746  Assert(ReplicationSlotCtl != NULL);
747 
748  LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
749  for (i = 0; i < max_replication_slots; i++)
750  {
751  ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
752  XLogRecPtr restart_lsn;
753 
754  if (!s->in_use)
755  continue;
756 
757  SpinLockAcquire(&s->mutex);
758  restart_lsn = s->data.restart_lsn;
759  SpinLockRelease(&s->mutex);
760 
761  if (restart_lsn != InvalidXLogRecPtr &&
762  (min_required == InvalidXLogRecPtr ||
763  restart_lsn < min_required))
764  min_required = restart_lsn;
765  }
766  LWLockRelease(ReplicationSlotControlLock);
767 
768  XLogSetReplicationSlotMinimumLSN(min_required);
769 }
770 
771 /*
772  * Compute the oldest WAL LSN required by *logical* decoding slots..
773  *
774  * Returns InvalidXLogRecPtr if logical decoding is disabled or no logical
775  * slots exist.
776  *
777  * NB: this returns a value >= ReplicationSlotsComputeRequiredLSN(), since it
778  * ignores physical replication slots.
779  *
780  * The results aren't required frequently, so we don't maintain a precomputed
781  * value like we do for ComputeRequiredLSN() and ComputeRequiredXmin().
782  */
785 {
786  XLogRecPtr result = InvalidXLogRecPtr;
787  int i;
788 
789  if (max_replication_slots <= 0)
790  return InvalidXLogRecPtr;
791 
792  LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
793 
794  for (i = 0; i < max_replication_slots; i++)
795  {
796  ReplicationSlot *s;
797  XLogRecPtr restart_lsn;
798 
799  s = &ReplicationSlotCtl->replication_slots[i];
800 
801  /* cannot change while ReplicationSlotCtlLock is held */
802  if (!s->in_use)
803  continue;
804 
805  /* we're only interested in logical slots */
806  if (!SlotIsLogical(s))
807  continue;
808 
809  /* read once, it's ok if it increases while we're checking */
810  SpinLockAcquire(&s->mutex);
811  restart_lsn = s->data.restart_lsn;
812  SpinLockRelease(&s->mutex);
813 
814  if (result == InvalidXLogRecPtr ||
815  restart_lsn < result)
816  result = restart_lsn;
817  }
818 
819  LWLockRelease(ReplicationSlotControlLock);
820 
821  return result;
822 }
823 
824 /*
825  * ReplicationSlotsCountDBSlots -- count the number of slots that refer to the
826  * passed database oid.
827  *
828  * Returns true if there are any slots referencing the database. *nslots will
829  * be set to the absolute number of slots in the database, *nactive to ones
830  * currently active.
831  */
832 bool
833 ReplicationSlotsCountDBSlots(Oid dboid, int *nslots, int *nactive)
834 {
835  int i;
836 
837  *nslots = *nactive = 0;
838 
839  if (max_replication_slots <= 0)
840  return false;
841 
842  LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
843  for (i = 0; i < max_replication_slots; i++)
844  {
845  ReplicationSlot *s;
846 
847  s = &ReplicationSlotCtl->replication_slots[i];
848 
849  /* cannot change while ReplicationSlotCtlLock is held */
850  if (!s->in_use)
851  continue;
852 
853  /* only logical slots are database specific, skip */
854  if (!SlotIsLogical(s))
855  continue;
856 
857  /* not our database, skip */
858  if (s->data.database != dboid)
859  continue;
860 
861  /* count slots with spinlock held */
862  SpinLockAcquire(&s->mutex);
863  (*nslots)++;
864  if (s->active_pid != 0)
865  (*nactive)++;
866  SpinLockRelease(&s->mutex);
867  }
868  LWLockRelease(ReplicationSlotControlLock);
869 
870  if (*nslots > 0)
871  return true;
872  return false;
873 }
874 
875 /*
876  * ReplicationSlotsDropDBSlots -- Drop all db-specific slots relating to the
877  * passed database oid. The caller should hold an exclusive lock on the
878  * pg_database oid for the database to prevent creation of new slots on the db
879  * or replay from existing slots.
880  *
881  * Another session that concurrently acquires an existing slot on the target DB
882  * (most likely to drop it) may cause this function to ERROR. If that happens
883  * it may have dropped some but not all slots.
884  *
885  * This routine isn't as efficient as it could be - but we don't drop
886  * databases often, especially databases with lots of slots.
887  */
888 void
890 {
891  int i;
892 
893  if (max_replication_slots <= 0)
894  return;
895 
896 restart:
897  LWLockAcquire(ReplicationSlotControlLock, LW_SHARED);
898  for (i = 0; i < max_replication_slots; i++)
899  {
900  ReplicationSlot *s;
901  char *slotname;
902  int active_pid;
903 
904  s = &ReplicationSlotCtl->replication_slots[i];
905 
906  /* cannot change while ReplicationSlotCtlLock is held */
907  if (!s->in_use)
908  continue;
909 
910  /* only logical slots are database specific, skip */
911  if (!SlotIsLogical(s))
912  continue;
913 
914  /* not our database, skip */
915  if (s->data.database != dboid)
916  continue;
917 
918  /* acquire slot, so ReplicationSlotDropAcquired can be reused */
919  SpinLockAcquire(&s->mutex);
920  /* can't change while ReplicationSlotControlLock is held */
921  slotname = NameStr(s->data.name);
922  active_pid = s->active_pid;
923  if (active_pid == 0)
924  {
925  MyReplicationSlot = s;
926  s->active_pid = MyProcPid;
927  }
928  SpinLockRelease(&s->mutex);
929 
930  /*
931  * Even though we hold an exclusive lock on the database object a
932  * logical slot for that DB can still be active, e.g. if it's
933  * concurrently being dropped by a backend connected to another DB.
934  *
935  * That's fairly unlikely in practice, so we'll just bail out.
936  */
937  if (active_pid)
938  ereport(ERROR,
939  (errcode(ERRCODE_OBJECT_IN_USE),
940  errmsg("replication slot \"%s\" is active for PID %d",
941  slotname, active_pid)));
942 
943  /*
944  * To avoid duplicating ReplicationSlotDropAcquired() and to avoid
945  * holding ReplicationSlotControlLock over filesystem operations,
946  * release ReplicationSlotControlLock and use
947  * ReplicationSlotDropAcquired.
948  *
949  * As that means the set of slots could change, restart scan from the
950  * beginning each time we release the lock.
951  */
952  LWLockRelease(ReplicationSlotControlLock);
954  goto restart;
955  }
956  LWLockRelease(ReplicationSlotControlLock);
957 }
958 
959 
960 /*
961  * Check whether the server's configuration supports using replication
962  * slots.
963  */
964 void
966 {
967  if (max_replication_slots == 0)
968  ereport(ERROR,
969  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
970  (errmsg("replication slots can only be used if max_replication_slots > 0"))));
971 
973  ereport(ERROR,
974  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
975  errmsg("replication slots can only be used if wal_level >= replica")));
976 }
977 
978 /*
979  * Reserve WAL for the currently active slot.
980  *
981  * Compute and set restart_lsn in a manner that's appropriate for the type of
982  * the slot and concurrency safe.
983  */
984 void
986 {
988 
989  Assert(slot != NULL);
991 
992  /*
993  * The replication slot mechanism is used to prevent removal of required
994  * WAL. As there is no interlock between this routine and checkpoints, WAL
995  * segments could concurrently be removed when a now stale return value of
996  * ReplicationSlotsComputeRequiredLSN() is used. In the unlikely case that
997  * this happens we'll just retry.
998  */
999  while (true)
1000  {
1001  XLogSegNo segno;
1002 
1003  /*
1004  * For logical slots log a standby snapshot and start logical decoding
1005  * at exactly that position. That allows the slot to start up more
1006  * quickly.
1007  *
1008  * That's not needed (or indeed helpful) for physical slots as they'll
1009  * start replay at the last logged checkpoint anyway. Instead return
1010  * the location of the last redo LSN. While that slightly increases
1011  * the chance that we have to retry, it's where a base backup has to
1012  * start replay at.
1013  */
1014  if (!RecoveryInProgress() && SlotIsLogical(slot))
1015  {
1016  XLogRecPtr flushptr;
1017 
1018  /* start at current insert position */
1020 
1021  /* make sure we have enough information to start */
1022  flushptr = LogStandbySnapshot();
1023 
1024  /* and make sure it's fsynced to disk */
1025  XLogFlush(flushptr);
1026  }
1027  else
1028  {
1029  slot->data.restart_lsn = GetRedoRecPtr();
1030  }
1031 
1032  /* prevent WAL removal as fast as possible */
1034 
1035  /*
1036  * If all required WAL is still there, great, otherwise retry. The
1037  * slot should prevent further removal of WAL, unless there's a
1038  * concurrent ReplicationSlotsComputeRequiredLSN() after we've written
1039  * the new restart_lsn above, so normally we should never need to loop
1040  * more than twice.
1041  */
1043  if (XLogGetLastRemovedSegno() < segno)
1044  break;
1045  }
1046 }
1047 
1048 /*
1049  * Flush all replication slots to disk.
1050  *
1051  * This needn't actually be part of a checkpoint, but it's a convenient
1052  * location.
1053  */
1054 void
1056 {
1057  int i;
1058 
1059  elog(DEBUG1, "performing replication slot checkpoint");
1060 
1061  /*
1062  * Prevent any slot from being created/dropped while we're active. As we
1063  * explicitly do *not* want to block iterating over replication_slots or
1064  * acquiring a slot we cannot take the control lock - but that's OK,
1065  * because holding ReplicationSlotAllocationLock is strictly stronger, and
1066  * enough to guarantee that nobody can change the in_use bits on us.
1067  */
1068  LWLockAcquire(ReplicationSlotAllocationLock, LW_SHARED);
1069 
1070  for (i = 0; i < max_replication_slots; i++)
1071  {
1072  ReplicationSlot *s = &ReplicationSlotCtl->replication_slots[i];
1073  char path[MAXPGPATH];
1074 
1075  if (!s->in_use)
1076  continue;
1077 
1078  /* save the slot to disk, locking is handled in SaveSlotToPath() */
1079  sprintf(path, "pg_replslot/%s", NameStr(s->data.name));
1080  SaveSlotToPath(s, path, LOG);
1081  }
1082  LWLockRelease(ReplicationSlotAllocationLock);
1083 }
1084 
1085 /*
1086  * Load all replication slots from disk into memory at server startup. This
1087  * needs to be run before we start crash recovery.
1088  */
1089 void
1091 {
1092  DIR *replication_dir;
1093  struct dirent *replication_de;
1094 
1095  elog(DEBUG1, "starting up replication slots");
1096 
1097  /* restore all slots by iterating over all on-disk entries */
1098  replication_dir = AllocateDir("pg_replslot");
1099  while ((replication_de = ReadDir(replication_dir, "pg_replslot")) != NULL)
1100  {
1101  struct stat statbuf;
1102  char path[MAXPGPATH + 12];
1103 
1104  if (strcmp(replication_de->d_name, ".") == 0 ||
1105  strcmp(replication_de->d_name, "..") == 0)
1106  continue;
1107 
1108  snprintf(path, sizeof(path), "pg_replslot/%s", replication_de->d_name);
1109 
1110  /* we're only creating directories here, skip if it's not our's */
1111  if (lstat(path, &statbuf) == 0 && !S_ISDIR(statbuf.st_mode))
1112  continue;
1113 
1114  /* we crashed while a slot was being setup or deleted, clean up */
1115  if (pg_str_endswith(replication_de->d_name, ".tmp"))
1116  {
1117  if (!rmtree(path, true))
1118  {
1119  ereport(WARNING,
1121  errmsg("could not remove directory \"%s\"", path)));
1122  continue;
1123  }
1124  fsync_fname("pg_replslot", true);
1125  continue;
1126  }
1127 
1128  /* looks like a slot in a normal state, restore */
1129  RestoreSlotFromDisk(replication_de->d_name);
1130  }
1131  FreeDir(replication_dir);
1132 
1133  /* currently no slots exist, we're done. */
1134  if (max_replication_slots <= 0)
1135  return;
1136 
1137  /* Now that we have recovered all the data, compute replication xmin */
1140 }
1141 
1142 /* ----
1143  * Manipulation of on-disk state of replication slots
1144  *
1145  * NB: none of the routines below should take any notice whether a slot is the
1146  * current one or not, that's all handled a layer above.
1147  * ----
1148  */
1149 static void
1151 {
1152  char tmppath[MAXPGPATH];
1153  char path[MAXPGPATH];
1154  struct stat st;
1155 
1156  /*
1157  * No need to take out the io_in_progress_lock, nobody else can see this
1158  * slot yet, so nobody else will write. We're reusing SaveSlotToPath which
1159  * takes out the lock, if we'd take the lock here, we'd deadlock.
1160  */
1161 
1162  sprintf(path, "pg_replslot/%s", NameStr(slot->data.name));
1163  sprintf(tmppath, "pg_replslot/%s.tmp", NameStr(slot->data.name));
1164 
1165  /*
1166  * It's just barely possible that some previous effort to create or drop a
1167  * slot with this name left a temp directory lying around. If that seems
1168  * to be the case, try to remove it. If the rmtree() fails, we'll error
1169  * out at the mkdir() below, so we don't bother checking success.
1170  */
1171  if (stat(tmppath, &st) == 0 && S_ISDIR(st.st_mode))
1172  rmtree(tmppath, true);
1173 
1174  /* Create and fsync the temporary slot directory. */
1175  if (mkdir(tmppath, S_IRWXU) < 0)
1176  ereport(ERROR,
1178  errmsg("could not create directory \"%s\": %m",
1179  tmppath)));
1180  fsync_fname(tmppath, true);
1181 
1182  /* Write the actual state file. */
1183  slot->dirty = true; /* signal that we really need to write */
1184  SaveSlotToPath(slot, tmppath, ERROR);
1185 
1186  /* Rename the directory into place. */
1187  if (rename(tmppath, path) != 0)
1188  ereport(ERROR,
1190  errmsg("could not rename file \"%s\" to \"%s\": %m",
1191  tmppath, path)));
1192 
1193  /*
1194  * If we'd now fail - really unlikely - we wouldn't know whether this slot
1195  * would persist after an OS crash or not - so, force a restart. The
1196  * restart would try to fsync this again till it works.
1197  */
1199 
1200  fsync_fname(path, true);
1201  fsync_fname("pg_replslot", true);
1202 
1203  END_CRIT_SECTION();
1204 }
1205 
1206 /*
1207  * Shared functionality between saving and creating a replication slot.
1208  */
1209 static void
1210 SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
1211 {
1212  char tmppath[MAXPGPATH];
1213  char path[MAXPGPATH];
1214  int fd;
1216  bool was_dirty;
1217 
1218  /* first check whether there's something to write out */
1219  SpinLockAcquire(&slot->mutex);
1220  was_dirty = slot->dirty;
1221  slot->just_dirtied = false;
1222  SpinLockRelease(&slot->mutex);
1223 
1224  /* and don't do anything if there's nothing to write */
1225  if (!was_dirty)
1226  return;
1227 
1229 
1230  /* silence valgrind :( */
1231  memset(&cp, 0, sizeof(ReplicationSlotOnDisk));
1232 
1233  sprintf(tmppath, "%s/state.tmp", dir);
1234  sprintf(path, "%s/state", dir);
1235 
1236  fd = OpenTransientFile(tmppath, O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
1237  if (fd < 0)
1238  {
1239  ereport(elevel,
1241  errmsg("could not create file \"%s\": %m",
1242  tmppath)));
1243  return;
1244  }
1245 
1246  cp.magic = SLOT_MAGIC;
1247  INIT_CRC32C(cp.checksum);
1248  cp.version = SLOT_VERSION;
1250 
1251  SpinLockAcquire(&slot->mutex);
1252 
1253  memcpy(&cp.slotdata, &slot->data, sizeof(ReplicationSlotPersistentData));
1254 
1255  SpinLockRelease(&slot->mutex);
1256 
1257  COMP_CRC32C(cp.checksum,
1258  (char *) (&cp) + SnapBuildOnDiskNotChecksummedSize,
1260  FIN_CRC32C(cp.checksum);
1261 
1263  if ((write(fd, &cp, sizeof(cp))) != sizeof(cp))
1264  {
1265  int save_errno = errno;
1266 
1268  CloseTransientFile(fd);
1269  errno = save_errno;
1270  ereport(elevel,
1272  errmsg("could not write to file \"%s\": %m",
1273  tmppath)));
1274  return;
1275  }
1277 
1278  /* fsync the temporary file */
1280  if (pg_fsync(fd) != 0)
1281  {
1282  int save_errno = errno;
1283 
1285  CloseTransientFile(fd);
1286  errno = save_errno;
1287  ereport(elevel,
1289  errmsg("could not fsync file \"%s\": %m",
1290  tmppath)));
1291  return;
1292  }
1294 
1295  CloseTransientFile(fd);
1296 
1297  /* rename to permanent file, fsync file and directory */
1298  if (rename(tmppath, path) != 0)
1299  {
1300  ereport(elevel,
1302  errmsg("could not rename file \"%s\" to \"%s\": %m",
1303  tmppath, path)));
1304  return;
1305  }
1306 
1307  /* Check CreateSlot() for the reasoning of using a crit. section. */
1309 
1310  fsync_fname(path, false);
1311  fsync_fname(dir, true);
1312  fsync_fname("pg_replslot", true);
1313 
1314  END_CRIT_SECTION();
1315 
1316  /*
1317  * Successfully wrote, unset dirty bit, unless somebody dirtied again
1318  * already.
1319  */
1320  SpinLockAcquire(&slot->mutex);
1321  if (!slot->just_dirtied)
1322  slot->dirty = false;
1323  SpinLockRelease(&slot->mutex);
1324 
1326 }
1327 
1328 /*
1329  * Load a single slot from disk into memory.
1330  */
1331 static void
1333 {
1335  int i;
1336  char path[MAXPGPATH + 22];
1337  int fd;
1338  bool restored = false;
1339  int readBytes;
1340  pg_crc32c checksum;
1341 
1342  /* no need to lock here, no concurrent access allowed yet */
1343 
1344  /* delete temp file if it exists */
1345  sprintf(path, "pg_replslot/%s/state.tmp", name);
1346  if (unlink(path) < 0 && errno != ENOENT)
1347  ereport(PANIC,
1349  errmsg("could not remove file \"%s\": %m", path)));
1350 
1351  sprintf(path, "pg_replslot/%s/state", name);
1352 
1353  elog(DEBUG1, "restoring replication slot from \"%s\"", path);
1354 
1355  fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
1356 
1357  /*
1358  * We do not need to handle this as we are rename()ing the directory into
1359  * place only after we fsync()ed the state file.
1360  */
1361  if (fd < 0)
1362  ereport(PANIC,
1364  errmsg("could not open file \"%s\": %m", path)));
1365 
1366  /*
1367  * Sync state file before we're reading from it. We might have crashed
1368  * while it wasn't synced yet and we shouldn't continue on that basis.
1369  */
1371  if (pg_fsync(fd) != 0)
1372  {
1373  CloseTransientFile(fd);
1374  ereport(PANIC,
1376  errmsg("could not fsync file \"%s\": %m",
1377  path)));
1378  }
1380 
1381  /* Also sync the parent directory */
1383  fsync_fname(path, true);
1384  END_CRIT_SECTION();
1385 
1386  /* read part of statefile that's guaranteed to be version independent */
1388  readBytes = read(fd, &cp, ReplicationSlotOnDiskConstantSize);
1390  if (readBytes != ReplicationSlotOnDiskConstantSize)
1391  {
1392  int saved_errno = errno;
1393 
1394  CloseTransientFile(fd);
1395  errno = saved_errno;
1396  ereport(PANIC,
1398  errmsg("could not read file \"%s\", read %d of %u: %m",
1399  path, readBytes,
1401  }
1402 
1403  /* verify magic */
1404  if (cp.magic != SLOT_MAGIC)
1405  ereport(PANIC,
1407  errmsg("replication slot file \"%s\" has wrong magic number: %u instead of %u",
1408  path, cp.magic, SLOT_MAGIC)));
1409 
1410  /* verify version */
1411  if (cp.version != SLOT_VERSION)
1412  ereport(PANIC,
1414  errmsg("replication slot file \"%s\" has unsupported version %u",
1415  path, cp.version)));
1416 
1417  /* boundary check on length */
1419  ereport(PANIC,
1421  errmsg("replication slot file \"%s\" has corrupted length %u",
1422  path, cp.length)));
1423 
1424  /* Now that we know the size, read the entire file */
1426  readBytes = read(fd,
1427  (char *) &cp + ReplicationSlotOnDiskConstantSize,
1428  cp.length);
1430  if (readBytes != cp.length)
1431  {
1432  int saved_errno = errno;
1433 
1434  CloseTransientFile(fd);
1435  errno = saved_errno;
1436  ereport(PANIC,
1438  errmsg("could not read file \"%s\", read %d of %u: %m",
1439  path, readBytes, cp.length)));
1440  }
1441 
1442  CloseTransientFile(fd);
1443 
1444  /* now verify the CRC */
1445  INIT_CRC32C(checksum);
1446  COMP_CRC32C(checksum,
1447  (char *) &cp + SnapBuildOnDiskNotChecksummedSize,
1449  FIN_CRC32C(checksum);
1450 
1451  if (!EQ_CRC32C(checksum, cp.checksum))
1452  ereport(PANIC,
1453  (errmsg("checksum mismatch for replication slot file \"%s\": is %u, should be %u",
1454  path, checksum, cp.checksum)));
1455 
1456  /*
1457  * If we crashed with an ephemeral slot active, don't restore but delete
1458  * it.
1459  */
1461  {
1462  sprintf(path, "pg_replslot/%s", name);
1463 
1464  if (!rmtree(path, true))
1465  {
1466  ereport(WARNING,
1468  errmsg("could not remove directory \"%s\"", path)));
1469  }
1470  fsync_fname("pg_replslot", true);
1471  return;
1472  }
1473 
1474  /* nothing can be active yet, don't lock anything */
1475  for (i = 0; i < max_replication_slots; i++)
1476  {
1477  ReplicationSlot *slot;
1478 
1479  slot = &ReplicationSlotCtl->replication_slots[i];
1480 
1481  if (slot->in_use)
1482  continue;
1483 
1484  /* restore the entire set of persistent data */
1485  memcpy(&slot->data, &cp.slotdata,
1487 
1488  /* initialize in memory state */
1489  slot->effective_xmin = cp.slotdata.xmin;
1491 
1496 
1497  slot->in_use = true;
1498  slot->active_pid = 0;
1499 
1500  restored = true;
1501  break;
1502  }
1503 
1504  if (!restored)
1505  ereport(PANIC,
1506  (errmsg("too many replication slots active before shutdown"),
1507  errhint("Increase max_replication_slots and try again.")));
1508 }
#define INIT_CRC32C(crc)
Definition: pg_crc32c.h:41
static void RestoreSlotFromDisk(const char *name)
Definition: slot.c:1332
ReplicationSlotCtlData * ReplicationSlotCtl
Definition: slot.c:93
void CheckSlotRequirements(void)
Definition: slot.c:965
TransactionId candidate_catalog_xmin
Definition: slot.h:135
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
#define DEBUG1
Definition: elog.h:25
int MyProcPid
Definition: globals.c:39
int errhint(const char *fmt,...)
Definition: elog.c:987
Size ReplicationSlotsShmemSize(void)
Definition: slot.c:114
#define PROC_IN_LOGICAL_DECODING
Definition: proc.h:57
int wal_segment_size
Definition: xlog.c:113
uint32 TransactionId
Definition: c.h:391
bool pg_str_endswith(const char *str, const char *end)
Definition: string.c:31
#define write(a, b, c)
Definition: win32.h:14
#define SLOT_MAGIC
Definition: slot.c:89
uint32 pg_crc32c
Definition: pg_crc32c.h:38
#define mkdir(a, b)
Definition: win32.h:57
int wal_level
Definition: xlog.c:104
#define SpinLockInit(lock)
Definition: spin.h:60
void ReplicationSlotAcquire(const char *name, bool nowait)
Definition: slot.c:330
int ConditionVariableBroadcast(ConditionVariable *cv)
#define END_CRIT_SECTION()
Definition: miscadmin.h:133
void fsync_fname(const char *fname, bool isdir)
Definition: fd.c:572
void ReplicationSlotCreate(const char *name, bool db_specific, ReplicationSlotPersistency persistency)
Definition: slot.c:221
ReplicationSlotPersistency persistency
Definition: slot.h:54
#define START_CRIT_SECTION()
Definition: miscadmin.h:131
int errcode(int sqlerrcode)
Definition: elog.c:575
#define MemSet(start, val, len)
Definition: c.h:846
int snprintf(char *str, size_t count, const char *fmt,...) pg_attribute_printf(3
void ReplicationSlotSave(void)
Definition: slot.c:638
#define SnapBuildOnDiskNotChecksummedSize
Definition: slot.c:80
static void ReplicationSlotDropPtr(ReplicationSlot *slot)
Definition: slot.c:539
ReplicationSlotPersistentData data
Definition: slot.h:120
#define LOG
Definition: elog.h:26
unsigned int Oid
Definition: postgres_ext.h:31
bool RecoveryInProgress(void)
Definition: xlog.c:7954
Definition: dirent.h:9
#define PANIC
Definition: elog.h:53
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2773
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1027
static void CreateSlotOnDisk(ReplicationSlot *slot)
Definition: slot.c:1150
void ReplicationSlotsShmemInit(void)
Definition: slot.c:132
static void SaveSlotToPath(ReplicationSlot *slot, const char *path, int elevel)
Definition: slot.c:1210
void XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
Definition: xlog.c:2664
PGXACT * MyPgXact
Definition: proc.c:68
uint8 vacuumFlags
Definition: proc.h:230
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1721
void ConditionVariablePrepareToSleep(ConditionVariable *cv)
#define NAMEDATALEN
bool ReplicationSlotValidateName(const char *name, int elevel)
Definition: slot.c:174
#define SpinLockAcquire(lock)
Definition: spin.h:62
void ConditionVariableInit(ConditionVariable *cv)
XLogSegNo XLogGetLastRemovedSegno(void)
Definition: xlog.c:3807
void ReplicationSlotReserveWal(void)
Definition: slot.c:985
static void ReplicationSlotDropAcquired(void)
Definition: slot.c:522
void ReplicationSlotsComputeRequiredLSN(void)
Definition: slot.c:741
ReplicationSlotPersistentData slotdata
Definition: slot.c:73
void ConditionVariableCancelSleep(void)
XLogRecPtr LogStandbySnapshot(void)
Definition: standby.c:909
Definition: dirent.c:25
#define ERROR
Definition: elog.h:43
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2167
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:372
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:11196
#define MAXPGPATH
void ReplicationSlotPersist(void)
Definition: slot.c:673
TransactionId effective_xmin
Definition: slot.h:116
XLogRecPtr candidate_restart_valid
Definition: slot.h:137
void StartupReplicationSlots(void)
Definition: slot.c:1090
uint64 XLogSegNo
Definition: xlogdefs.h:34
int errcode_for_file_access(void)
Definition: elog.c:598
XLogRecPtr ReplicationSlotsComputeLogicalRestartLSN(void)
Definition: slot.c:784
TransactionId catalog_xmin
Definition: slot.h:70
#define InvalidTransactionId
Definition: transam.h:31
unsigned int uint32
Definition: c.h:258
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2367
void ReplicationSlotRelease(void)
Definition: slot.c:416
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1244
TransactionId xmin
Definition: slot.h:62
#define EQ_CRC32C(c1, c2)
Definition: pg_crc32c.h:42
#define SlotIsLogical(slot)
Definition: slot.h:142
#define ereport(elevel, rest)
Definition: elog.h:122
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
pg_crc32c checksum
Definition: slot.c:62
struct ReplicationSlotOnDisk ReplicationSlotOnDisk
void LWLockInitialize(LWLock *lock, int tranche_id)
Definition: lwlock.c:673
int CloseTransientFile(int fd)
Definition: fd.c:2337
#define WARNING
Definition: elog.h:40
bool rmtree(const char *path, bool rmtopdir)
Definition: rmtree.c:36
bool in_use
Definition: slot.h:96
static int elevel
Definition: vacuumlazy.c:136
#define SpinLockRelease(lock)
Definition: spin.h:64
Size mul_size(Size s1, Size s2)
Definition: shmem.c:492
bool just_dirtied
Definition: slot.h:102
Size add_size(Size s1, Size s2)
Definition: shmem.c:475
TransactionId effective_catalog_xmin
Definition: slot.h:117
Oid MyDatabaseId
Definition: globals.c:77
#define SLOT_VERSION
Definition: slot.c:90
#define InvalidOid
Definition: postgres_ext.h:36
ReplicationSlot * MyReplicationSlot
Definition: slot.c:96
int max_replication_slots
Definition: slot.c:99
void ConditionVariableSleep(ConditionVariable *cv, uint32 wait_event_info)
#define ReplicationSlotOnDiskV2Size
Definition: slot.c:86
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:664
#define StrNCpy(dst, src, len)
Definition: c.h:819
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2433
XLogRecPtr restart_lsn
Definition: slot.h:73
#define ReplicationSlotOnDiskConstantSize
Definition: slot.c:77
size_t Size
Definition: c.h:350
uint32 version
Definition: slot.c:65
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1220
#define SnapBuildOnDiskChecksummedSize
Definition: slot.c:83
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1117
void LWLockRegisterTranche(int tranche_id, char *tranche_name)
Definition: lwlock.c:598
XLogRecPtr GetRedoRecPtr(void)
Definition: xlog.c:8247
ConditionVariable active_cv
Definition: slot.h:126
const char * name
Definition: encode.c:521
bool ReplicationSlotsCountDBSlots(Oid dboid, int *nslots, int *nactive)
Definition: slot.c:833
XLogRecPtr candidate_xmin_lsn
Definition: slot.h:136
void ReplicationSlotDrop(const char *name, bool nowait)
Definition: slot.c:509
ReplicationSlotPersistency
Definition: slot.h:33
int errmsg(const char *fmt,...)
Definition: elog.c:797
pid_t active_pid
Definition: slot.h:99
int i
#define NameStr(name)
Definition: c.h:493
void ProcArraySetReplicationSlotXmin(TransactionId xmin, TransactionId catalog_xmin, bool already_locked)
Definition: procarray.c:2964
void ReplicationSlotCleanup(void)
Definition: slot.c:471
int pg_fsync(int fd)
Definition: fd.c:338
ReplicationSlot replication_slots[1]
Definition: slot.h:153
char d_name[MAX_PATH]
Definition: dirent.h:14
#define elog
Definition: elog.h:219
slock_t mutex
Definition: slot.h:93
#define TransactionIdIsValid(xid)
Definition: transam.h:41
#define COMP_CRC32C(crc, data, len)
Definition: pg_crc32c.h:73
#define ERRCODE_DUPLICATE_OBJECT
Definition: streamutil.c:33
#define lstat(path, sb)
Definition: win32.h:262
void CheckPointReplicationSlots(void)
Definition: slot.c:1055
#define FIN_CRC32C(crc)
Definition: pg_crc32c.h:78
void ReplicationSlotsDropDBSlots(Oid dboid)
Definition: slot.c:889
void ReplicationSlotsComputeRequiredXmin(bool already_locked)
Definition: slot.c:695
bool dirty
Definition: slot.h:103
#define read(a, b, c)
Definition: win32.h:13
int FreeDir(DIR *dir)
Definition: fd.c:2476
XLogRecPtr candidate_restart_lsn
Definition: slot.h:138
#define offsetof(type, field)
Definition: c.h:549
void ReplicationSlotMarkDirty(void)
Definition: slot.c:656
LWLock io_in_progress_lock
Definition: slot.h:123
#define XLByteToSeg(xlrp, logSegNo, wal_segsz_bytes)