PostgreSQL Source Code  git master
procarray.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * procarray.c
4  * POSTGRES process array code.
5  *
6  *
7  * This module maintains arrays of the PGPROC and PGXACT structures for all
8  * active backends. Although there are several uses for this, the principal
9  * one is as a means of determining the set of currently running transactions.
10  *
11  * Because of various subtle race conditions it is critical that a backend
12  * hold the correct locks while setting or clearing its MyPgXact->xid field.
13  * See notes in src/backend/access/transam/README.
14  *
15  * The process arrays now also include structures representing prepared
16  * transactions. The xid and subxids fields of these are valid, as are the
17  * myProcLocks lists. They can be distinguished from regular backend PGPROCs
18  * at need by checking for pid == 0.
19  *
20  * During hot standby, we also keep a list of XIDs representing transactions
21  * that are known to be running in the master (or more precisely, were running
22  * as of the current point in the WAL stream). This list is kept in the
23  * KnownAssignedXids array, and is updated by watching the sequence of
24  * arriving XIDs. This is necessary because if we leave those XIDs out of
25  * snapshots taken for standby queries, then they will appear to be already
26  * complete, leading to MVCC failures. Note that in hot standby, the PGPROC
27  * array represents standby processes, which by definition are not running
28  * transactions that have XIDs.
29  *
30  * It is perhaps possible for a backend on the master to terminate without
31  * writing an abort record for its transaction. While that shouldn't really
32  * happen, it would tie up KnownAssignedXids indefinitely, so we protect
33  * ourselves by pruning the array when a valid list of running XIDs arrives.
34  *
35  * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
36  * Portions Copyright (c) 1994, Regents of the University of California
37  *
38  *
39  * IDENTIFICATION
40  * src/backend/storage/ipc/procarray.c
41  *
42  *-------------------------------------------------------------------------
43  */
44 #include "postgres.h"
45 
46 #include <signal.h>
47 
48 #include "access/clog.h"
49 #include "access/subtrans.h"
50 #include "access/transam.h"
51 #include "access/twophase.h"
52 #include "access/xact.h"
53 #include "access/xlog.h"
54 #include "catalog/catalog.h"
55 #include "miscadmin.h"
56 #include "pgstat.h"
57 #include "storage/proc.h"
58 #include "storage/procarray.h"
59 #include "storage/spin.h"
60 #include "utils/builtins.h"
61 #include "utils/rel.h"
62 #include "utils/snapmgr.h"
63 
64 #define UINT32_ACCESS_ONCE(var) ((uint32)(*((volatile uint32 *)&(var))))
65 
66 /* Our shared memory area */
67 typedef struct ProcArrayStruct
68 {
69  int numProcs; /* number of valid procs entries */
70  int maxProcs; /* allocated size of procs array */
71 
72  /*
73  * Known assigned XIDs handling
74  */
75  int maxKnownAssignedXids; /* allocated size of array */
76  int numKnownAssignedXids; /* current # of valid entries */
77  int tailKnownAssignedXids; /* index of oldest valid element */
78  int headKnownAssignedXids; /* index of newest element, + 1 */
79  slock_t known_assigned_xids_lck; /* protects head/tail pointers */
80 
81  /*
82  * Highest subxid that has been removed from KnownAssignedXids array to
83  * prevent overflow; or InvalidTransactionId if none. We track this for
84  * similar reasons to tracking overflowing cached subxids in PGXACT
85  * entries. Must hold exclusive ProcArrayLock to change this, and shared
86  * lock to read it.
87  */
89 
90  /* oldest xmin of any replication slot */
92  /* oldest catalog xmin of any replication slot */
94 
95  /* indexes into allPgXact[], has PROCARRAY_MAXPROCS entries */
96  int pgprocnos[FLEXIBLE_ARRAY_MEMBER];
98 
100 
101 static PGPROC *allProcs;
103 
104 /*
105  * Bookkeeping for tracking emulated transactions in recovery
106  */
110 
111 /*
112  * If we're in STANDBY_SNAPSHOT_PENDING state, standbySnapshotPendingXmin is
113  * the highest xid that might still be running that we don't have in
114  * KnownAssignedXids.
115  */
117 
118 #ifdef XIDCACHE_DEBUG
119 
120 /* counters for XidCache measurement */
121 static long xc_by_recent_xmin = 0;
122 static long xc_by_known_xact = 0;
123 static long xc_by_my_xact = 0;
124 static long xc_by_latest_xid = 0;
125 static long xc_by_main_xid = 0;
126 static long xc_by_child_xid = 0;
127 static long xc_by_known_assigned = 0;
128 static long xc_no_overflow = 0;
129 static long xc_slow_answer = 0;
130 
131 #define xc_by_recent_xmin_inc() (xc_by_recent_xmin++)
132 #define xc_by_known_xact_inc() (xc_by_known_xact++)
133 #define xc_by_my_xact_inc() (xc_by_my_xact++)
134 #define xc_by_latest_xid_inc() (xc_by_latest_xid++)
135 #define xc_by_main_xid_inc() (xc_by_main_xid++)
136 #define xc_by_child_xid_inc() (xc_by_child_xid++)
137 #define xc_by_known_assigned_inc() (xc_by_known_assigned++)
138 #define xc_no_overflow_inc() (xc_no_overflow++)
139 #define xc_slow_answer_inc() (xc_slow_answer++)
140 
141 static void DisplayXidCache(void);
142 #else /* !XIDCACHE_DEBUG */
143 
144 #define xc_by_recent_xmin_inc() ((void) 0)
145 #define xc_by_known_xact_inc() ((void) 0)
146 #define xc_by_my_xact_inc() ((void) 0)
147 #define xc_by_latest_xid_inc() ((void) 0)
148 #define xc_by_main_xid_inc() ((void) 0)
149 #define xc_by_child_xid_inc() ((void) 0)
150 #define xc_by_known_assigned_inc() ((void) 0)
151 #define xc_no_overflow_inc() ((void) 0)
152 #define xc_slow_answer_inc() ((void) 0)
153 #endif /* XIDCACHE_DEBUG */
154 
155 /* Primitives for KnownAssignedXids array handling for standby */
156 static void KnownAssignedXidsCompress(bool force);
157 static void KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
158  bool exclusive_lock);
159 static bool KnownAssignedXidsSearch(TransactionId xid, bool remove);
160 static bool KnownAssignedXidExists(TransactionId xid);
161 static void KnownAssignedXidsRemove(TransactionId xid);
162 static void KnownAssignedXidsRemoveTree(TransactionId xid, int nsubxids,
163  TransactionId *subxids);
165 static int KnownAssignedXidsGet(TransactionId *xarray, TransactionId xmax);
167  TransactionId *xmin,
168  TransactionId xmax);
170 static void KnownAssignedXidsDisplay(int trace_level);
171 static void KnownAssignedXidsReset(void);
172 static inline void ProcArrayEndTransactionInternal(PGPROC *proc,
173  PGXACT *pgxact, TransactionId latestXid);
174 static void ProcArrayGroupClearXid(PGPROC *proc, TransactionId latestXid);
175 
176 /*
177  * Report shared-memory space needed by CreateSharedProcArray.
178  */
179 Size
181 {
182  Size size;
183 
184  /* Size of the ProcArray structure itself */
185 #define PROCARRAY_MAXPROCS (MaxBackends + max_prepared_xacts)
186 
188  size = add_size(size, mul_size(sizeof(int), PROCARRAY_MAXPROCS));
189 
190  /*
191  * During Hot Standby processing we have a data structure called
192  * KnownAssignedXids, created in shared memory. Local data structures are
193  * also created in various backends during GetSnapshotData(),
194  * TransactionIdIsInProgress() and GetRunningTransactionData(). All of the
195  * main structures created in those functions must be identically sized,
196  * since we may at times copy the whole of the data structures around. We
197  * refer to this size as TOTAL_MAX_CACHED_SUBXIDS.
198  *
199  * Ideally we'd only create this structure if we were actually doing hot
200  * standby in the current run, but we don't know that yet at the time
201  * shared memory is being set up.
202  */
203 #define TOTAL_MAX_CACHED_SUBXIDS \
204  ((PGPROC_MAX_CACHED_SUBXIDS + 1) * PROCARRAY_MAXPROCS)
205 
206  if (EnableHotStandby)
207  {
208  size = add_size(size,
209  mul_size(sizeof(TransactionId),
211  size = add_size(size,
212  mul_size(sizeof(bool), TOTAL_MAX_CACHED_SUBXIDS));
213  }
214 
215  return size;
216 }
217 
218 /*
219  * Initialize the shared PGPROC array during postmaster startup.
220  */
221 void
223 {
224  bool found;
225 
226  /* Create or attach to the ProcArray shared structure */
227  procArray = (ProcArrayStruct *)
228  ShmemInitStruct("Proc Array",
230  mul_size(sizeof(int),
232  &found);
233 
234  if (!found)
235  {
236  /*
237  * We're the first - initialize.
238  */
239  procArray->numProcs = 0;
240  procArray->maxProcs = PROCARRAY_MAXPROCS;
242  procArray->numKnownAssignedXids = 0;
243  procArray->tailKnownAssignedXids = 0;
244  procArray->headKnownAssignedXids = 0;
249  }
250 
251  allProcs = ProcGlobal->allProcs;
252  allPgXact = ProcGlobal->allPgXact;
253 
254  /* Create or attach to the KnownAssignedXids arrays too, if needed */
255  if (EnableHotStandby)
256  {
258  ShmemInitStruct("KnownAssignedXids",
259  mul_size(sizeof(TransactionId),
261  &found);
262  KnownAssignedXidsValid = (bool *)
263  ShmemInitStruct("KnownAssignedXidsValid",
264  mul_size(sizeof(bool), TOTAL_MAX_CACHED_SUBXIDS),
265  &found);
266  }
267 
269 }
270 
271 /*
272  * Add the specified PGPROC to the shared array.
273  */
274 void
276 {
277  ProcArrayStruct *arrayP = procArray;
278  int index;
279 
280  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
281 
282  if (arrayP->numProcs >= arrayP->maxProcs)
283  {
284  /*
285  * Oops, no room. (This really shouldn't happen, since there is a
286  * fixed supply of PGPROC structs too, and so we should have failed
287  * earlier.)
288  */
289  LWLockRelease(ProcArrayLock);
290  ereport(FATAL,
291  (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
292  errmsg("sorry, too many clients already")));
293  }
294 
295  /*
296  * Keep the procs array sorted by (PGPROC *) so that we can utilize
297  * locality of references much better. This is useful while traversing the
298  * ProcArray because there is an increased likelihood of finding the next
299  * PGPROC structure in the cache.
300  *
301  * Since the occurrence of adding/removing a proc is much lower than the
302  * access to the ProcArray itself, the overhead should be marginal
303  */
304  for (index = 0; index < arrayP->numProcs; index++)
305  {
306  /*
307  * If we are the first PGPROC or if we have found our right position
308  * in the array, break
309  */
310  if ((arrayP->pgprocnos[index] == -1) || (arrayP->pgprocnos[index] > proc->pgprocno))
311  break;
312  }
313 
314  memmove(&arrayP->pgprocnos[index + 1], &arrayP->pgprocnos[index],
315  (arrayP->numProcs - index) * sizeof(int));
316  arrayP->pgprocnos[index] = proc->pgprocno;
317  arrayP->numProcs++;
318 
319  LWLockRelease(ProcArrayLock);
320 }
321 
322 /*
323  * Remove the specified PGPROC from the shared array.
324  *
325  * When latestXid is a valid XID, we are removing a live 2PC gxact from the
326  * array, and thus causing it to appear as "not running" anymore. In this
327  * case we must advance latestCompletedXid. (This is essentially the same
328  * as ProcArrayEndTransaction followed by removal of the PGPROC, but we take
329  * the ProcArrayLock only once, and don't damage the content of the PGPROC;
330  * twophase.c depends on the latter.)
331  */
332 void
334 {
335  ProcArrayStruct *arrayP = procArray;
336  int index;
337 
338 #ifdef XIDCACHE_DEBUG
339  /* dump stats at backend shutdown, but not prepared-xact end */
340  if (proc->pid != 0)
341  DisplayXidCache();
342 #endif
343 
344  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
345 
346  if (TransactionIdIsValid(latestXid))
347  {
348  Assert(TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
349 
350  /* Advance global latestCompletedXid while holding the lock */
352  latestXid))
354  }
355  else
356  {
357  /* Shouldn't be trying to remove a live transaction here */
358  Assert(!TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
359  }
360 
361  for (index = 0; index < arrayP->numProcs; index++)
362  {
363  if (arrayP->pgprocnos[index] == proc->pgprocno)
364  {
365  /* Keep the PGPROC array sorted. See notes above */
366  memmove(&arrayP->pgprocnos[index], &arrayP->pgprocnos[index + 1],
367  (arrayP->numProcs - index - 1) * sizeof(int));
368  arrayP->pgprocnos[arrayP->numProcs - 1] = -1; /* for debugging */
369  arrayP->numProcs--;
370  LWLockRelease(ProcArrayLock);
371  return;
372  }
373  }
374 
375  /* Oops */
376  LWLockRelease(ProcArrayLock);
377 
378  elog(LOG, "failed to find proc %p in ProcArray", proc);
379 }
380 
381 
382 /*
383  * ProcArrayEndTransaction -- mark a transaction as no longer running
384  *
385  * This is used interchangeably for commit and abort cases. The transaction
386  * commit/abort must already be reported to WAL and pg_xact.
387  *
388  * proc is currently always MyProc, but we pass it explicitly for flexibility.
389  * latestXid is the latest Xid among the transaction's main XID and
390  * subtransactions, or InvalidTransactionId if it has no XID. (We must ask
391  * the caller to pass latestXid, instead of computing it from the PGPROC's
392  * contents, because the subxid information in the PGPROC might be
393  * incomplete.)
394  */
395 void
397 {
398  PGXACT *pgxact = &allPgXact[proc->pgprocno];
399 
400  if (TransactionIdIsValid(latestXid))
401  {
402  /*
403  * We must lock ProcArrayLock while clearing our advertised XID, so
404  * that we do not exit the set of "running" transactions while someone
405  * else is taking a snapshot. See discussion in
406  * src/backend/access/transam/README.
407  */
408  Assert(TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
409 
410  /*
411  * If we can immediately acquire ProcArrayLock, we clear our own XID
412  * and release the lock. If not, use group XID clearing to improve
413  * efficiency.
414  */
415  if (LWLockConditionalAcquire(ProcArrayLock, LW_EXCLUSIVE))
416  {
417  ProcArrayEndTransactionInternal(proc, pgxact, latestXid);
418  LWLockRelease(ProcArrayLock);
419  }
420  else
421  ProcArrayGroupClearXid(proc, latestXid);
422  }
423  else
424  {
425  /*
426  * If we have no XID, we don't need to lock, since we won't affect
427  * anyone else's calculation of a snapshot. We might change their
428  * estimate of global xmin, but that's OK.
429  */
430  Assert(!TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
431 
433  pgxact->xmin = InvalidTransactionId;
434  /* must be cleared with xid/xmin: */
436  pgxact->delayChkpt = false; /* be sure this is cleared in abort */
437  proc->recoveryConflictPending = false;
438 
439  Assert(pgxact->nxids == 0);
440  Assert(pgxact->overflowed == false);
441  }
442 }
443 
444 /*
445  * Mark a write transaction as no longer running.
446  *
447  * We don't do any locking here; caller must handle that.
448  */
449 static inline void
451  TransactionId latestXid)
452 {
453  pgxact->xid = InvalidTransactionId;
455  pgxact->xmin = InvalidTransactionId;
456  /* must be cleared with xid/xmin: */
458  pgxact->delayChkpt = false; /* be sure this is cleared in abort */
459  proc->recoveryConflictPending = false;
460 
461  /* Clear the subtransaction-XID cache too while holding the lock */
462  pgxact->nxids = 0;
463  pgxact->overflowed = false;
464 
465  /* Also advance global latestCompletedXid while holding the lock */
467  latestXid))
469 }
470 
471 /*
472  * ProcArrayGroupClearXid -- group XID clearing
473  *
474  * When we cannot immediately acquire ProcArrayLock in exclusive mode at
475  * commit time, add ourselves to a list of processes that need their XIDs
476  * cleared. The first process to add itself to the list will acquire
477  * ProcArrayLock in exclusive mode and perform ProcArrayEndTransactionInternal
478  * on behalf of all group members. This avoids a great deal of contention
479  * around ProcArrayLock when many processes are trying to commit at once,
480  * since the lock need not be repeatedly handed off from one committing
481  * process to the next.
482  */
483 static void
485 {
486  PROC_HDR *procglobal = ProcGlobal;
487  uint32 nextidx;
488  uint32 wakeidx;
489 
490  /* We should definitely have an XID to clear. */
491  Assert(TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
492 
493  /* Add ourselves to the list of processes needing a group XID clear. */
494  proc->procArrayGroupMember = true;
495  proc->procArrayGroupMemberXid = latestXid;
496  while (true)
497  {
498  nextidx = pg_atomic_read_u32(&procglobal->procArrayGroupFirst);
499  pg_atomic_write_u32(&proc->procArrayGroupNext, nextidx);
500 
502  &nextidx,
503  (uint32) proc->pgprocno))
504  break;
505  }
506 
507  /*
508  * If the list was not empty, the leader will clear our XID. It is
509  * impossible to have followers without a leader because the first process
510  * that has added itself to the list will always have nextidx as
511  * INVALID_PGPROCNO.
512  */
513  if (nextidx != INVALID_PGPROCNO)
514  {
515  int extraWaits = 0;
516 
517  /* Sleep until the leader clears our XID. */
519  for (;;)
520  {
521  /* acts as a read barrier */
522  PGSemaphoreLock(proc->sem);
523  if (!proc->procArrayGroupMember)
524  break;
525  extraWaits++;
526  }
528 
530 
531  /* Fix semaphore count for any absorbed wakeups */
532  while (extraWaits-- > 0)
533  PGSemaphoreUnlock(proc->sem);
534  return;
535  }
536 
537  /* We are the leader. Acquire the lock on behalf of everyone. */
538  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
539 
540  /*
541  * Now that we've got the lock, clear the list of processes waiting for
542  * group XID clearing, saving a pointer to the head of the list. Trying
543  * to pop elements one at a time could lead to an ABA problem.
544  */
545  nextidx = pg_atomic_exchange_u32(&procglobal->procArrayGroupFirst,
547 
548  /* Remember head of list so we can perform wakeups after dropping lock. */
549  wakeidx = nextidx;
550 
551  /* Walk the list and clear all XIDs. */
552  while (nextidx != INVALID_PGPROCNO)
553  {
554  PGPROC *proc = &allProcs[nextidx];
555  PGXACT *pgxact = &allPgXact[nextidx];
556 
558 
559  /* Move to next proc in list. */
560  nextidx = pg_atomic_read_u32(&proc->procArrayGroupNext);
561  }
562 
563  /* We're done with the lock now. */
564  LWLockRelease(ProcArrayLock);
565 
566  /*
567  * Now that we've released the lock, go back and wake everybody up. We
568  * don't do this under the lock so as to keep lock hold times to a
569  * minimum. The system calls we need to perform to wake other processes
570  * up are probably much slower than the simple memory writes we did while
571  * holding the lock.
572  */
573  while (wakeidx != INVALID_PGPROCNO)
574  {
575  PGPROC *proc = &allProcs[wakeidx];
576 
577  wakeidx = pg_atomic_read_u32(&proc->procArrayGroupNext);
579 
580  /* ensure all previous writes are visible before follower continues. */
582 
583  proc->procArrayGroupMember = false;
584 
585  if (proc != MyProc)
586  PGSemaphoreUnlock(proc->sem);
587  }
588 }
589 
590 /*
591  * ProcArrayClearTransaction -- clear the transaction fields
592  *
593  * This is used after successfully preparing a 2-phase transaction. We are
594  * not actually reporting the transaction's XID as no longer running --- it
595  * will still appear as running because the 2PC's gxact is in the ProcArray
596  * too. We just have to clear out our own PGXACT.
597  */
598 void
600 {
601  PGXACT *pgxact = &allPgXact[proc->pgprocno];
602 
603  /*
604  * We can skip locking ProcArrayLock here, because this action does not
605  * actually change anyone's view of the set of running XIDs: our entry is
606  * duplicate with the gxact that has already been inserted into the
607  * ProcArray.
608  */
609  pgxact->xid = InvalidTransactionId;
611  pgxact->xmin = InvalidTransactionId;
612  proc->recoveryConflictPending = false;
613 
614  /* redundant, but just in case */
616  pgxact->delayChkpt = false;
617 
618  /* Clear the subtransaction-XID cache too */
619  pgxact->nxids = 0;
620  pgxact->overflowed = false;
621 }
622 
623 /*
624  * ProcArrayInitRecovery -- initialize recovery xid mgmt environment
625  *
626  * Remember up to where the startup process initialized the CLOG and subtrans
627  * so we can ensure it's initialized gaplessly up to the point where necessary
628  * while in recovery.
629  */
630 void
632 {
634  Assert(TransactionIdIsNormal(initializedUptoXID));
635 
636  /*
637  * we set latestObservedXid to the xid SUBTRANS has been initialized up
638  * to, so we can extend it from that point onwards in
639  * RecordKnownAssignedTransactionIds, and when we get consistent in
640  * ProcArrayApplyRecoveryInfo().
641  */
642  latestObservedXid = initializedUptoXID;
644 }
645 
646 /*
647  * ProcArrayApplyRecoveryInfo -- apply recovery info about xids
648  *
649  * Takes us through 3 states: Initialized, Pending and Ready.
650  * Normal case is to go all the way to Ready straight away, though there
651  * are atypical cases where we need to take it in steps.
652  *
653  * Use the data about running transactions on master to create the initial
654  * state of KnownAssignedXids. We also use these records to regularly prune
655  * KnownAssignedXids because we know it is possible that some transactions
656  * with FATAL errors fail to write abort records, which could cause eventual
657  * overflow.
658  *
659  * See comments for LogStandbySnapshot().
660  */
661 void
663 {
664  TransactionId *xids;
665  int nxids;
666  int i;
667 
672 
673  /*
674  * Remove stale transactions, if any.
675  */
677 
678  /*
679  * Remove stale locks, if any.
680  */
682 
683  /*
684  * If our snapshot is already valid, nothing else to do...
685  */
687  return;
688 
689  /*
690  * If our initial RunningTransactionsData had an overflowed snapshot then
691  * we knew we were missing some subxids from our snapshot. If we continue
692  * to see overflowed snapshots then we might never be able to start up, so
693  * we make another test to see if our snapshot is now valid. We know that
694  * the missing subxids are equal to or earlier than nextXid. After we
695  * initialise we continue to apply changes during recovery, so once the
696  * oldestRunningXid is later than the nextXid from the initial snapshot we
697  * know that we no longer have missing information and can mark the
698  * snapshot as valid.
699  */
701  {
702  /*
703  * If the snapshot isn't overflowed or if its empty we can reset our
704  * pending state and use this snapshot instead.
705  */
706  if (!running->subxid_overflow || running->xcnt == 0)
707  {
708  /*
709  * If we have already collected known assigned xids, we need to
710  * throw them away before we apply the recovery snapshot.
711  */
714  }
715  else
716  {
718  running->oldestRunningXid))
719  {
722  "recovery snapshots are now enabled");
723  }
724  else
726  "recovery snapshot waiting for non-overflowed snapshot or "
727  "until oldest active xid on standby is at least %u (now %u)",
729  running->oldestRunningXid);
730  return;
731  }
732  }
733 
735 
736  /*
737  * NB: this can be reached at least twice, so make sure new code can deal
738  * with that.
739  */
740 
741  /*
742  * Nobody else is running yet, but take locks anyhow
743  */
744  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
745 
746  /*
747  * KnownAssignedXids is sorted so we cannot just add the xids, we have to
748  * sort them first.
749  *
750  * Some of the new xids are top-level xids and some are subtransactions.
751  * We don't call SubTransSetParent because it doesn't matter yet. If we
752  * aren't overflowed then all xids will fit in snapshot and so we don't
753  * need subtrans. If we later overflow, an xid assignment record will add
754  * xids to subtrans. If RunningTransactionsData is overflowed then we
755  * don't have enough information to correctly update subtrans anyway.
756  */
757 
758  /*
759  * Allocate a temporary array to avoid modifying the array passed as
760  * argument.
761  */
762  xids = palloc(sizeof(TransactionId) * (running->xcnt + running->subxcnt));
763 
764  /*
765  * Add to the temp array any xids which have not already completed.
766  */
767  nxids = 0;
768  for (i = 0; i < running->xcnt + running->subxcnt; i++)
769  {
770  TransactionId xid = running->xids[i];
771 
772  /*
773  * The running-xacts snapshot can contain xids that were still visible
774  * in the procarray when the snapshot was taken, but were already
775  * WAL-logged as completed. They're not running anymore, so ignore
776  * them.
777  */
779  continue;
780 
781  xids[nxids++] = xid;
782  }
783 
784  if (nxids > 0)
785  {
786  if (procArray->numKnownAssignedXids != 0)
787  {
788  LWLockRelease(ProcArrayLock);
789  elog(ERROR, "KnownAssignedXids is not empty");
790  }
791 
792  /*
793  * Sort the array so that we can add them safely into
794  * KnownAssignedXids.
795  */
796  qsort(xids, nxids, sizeof(TransactionId), xidComparator);
797 
798  /*
799  * Add the sorted snapshot into KnownAssignedXids. The running-xacts
800  * snapshot may include duplicated xids because of prepared
801  * transactions, so ignore them.
802  */
803  for (i = 0; i < nxids; i++)
804  {
805  if (i > 0 && TransactionIdEquals(xids[i - 1], xids[i]))
806  {
807  elog(DEBUG1,
808  "found duplicated transaction %u for KnownAssignedXids insertion",
809  xids[i]);
810  continue;
811  }
812  KnownAssignedXidsAdd(xids[i], xids[i], true);
813  }
814 
816  }
817 
818  pfree(xids);
819 
820  /*
821  * latestObservedXid is at least set to the point where SUBTRANS was
822  * started up to (cf. ProcArrayInitRecovery()) or to the biggest xid
823  * RecordKnownAssignedTransactionIds() was called for. Initialize
824  * subtrans from thereon, up to nextXid - 1.
825  *
826  * We need to duplicate parts of RecordKnownAssignedTransactionId() here,
827  * because we've just added xids to the known assigned xids machinery that
828  * haven't gone through RecordKnownAssignedTransactionId().
829  */
833  {
836  }
837  TransactionIdRetreat(latestObservedXid); /* = running->nextXid - 1 */
838 
839  /* ----------
840  * Now we've got the running xids we need to set the global values that
841  * are used to track snapshots as they evolve further.
842  *
843  * - latestCompletedXid which will be the xmax for snapshots
844  * - lastOverflowedXid which shows whether snapshots overflow
845  * - nextXid
846  *
847  * If the snapshot overflowed, then we still initialise with what we know,
848  * but the recovery snapshot isn't fully valid yet because we know there
849  * are some subxids missing. We don't know the specific subxids that are
850  * missing, so conservatively assume the last one is latestObservedXid.
851  * ----------
852  */
853  if (running->subxid_overflow)
854  {
856 
859  }
860  else
861  {
863 
865  }
866 
867  /*
868  * If a transaction wrote a commit record in the gap between taking and
869  * logging the snapshot then latestCompletedXid may already be higher than
870  * the value from the snapshot, so check before we use the incoming value.
871  */
873  running->latestCompletedXid))
875 
877 
878  LWLockRelease(ProcArrayLock);
879 
880  /* ShmemVariableCache->nextFullXid must be beyond any observed xid. */
882 
884 
887  elog(trace_recovery(DEBUG1), "recovery snapshots are now enabled");
888  else
890  "recovery snapshot waiting for non-overflowed snapshot or "
891  "until oldest active xid on standby is at least %u (now %u)",
893  running->oldestRunningXid);
894 }
895 
896 /*
897  * ProcArrayApplyXidAssignment
898  * Process an XLOG_XACT_ASSIGNMENT WAL record
899  */
900 void
902  int nsubxids, TransactionId *subxids)
903 {
904  TransactionId max_xid;
905  int i;
906 
908 
909  max_xid = TransactionIdLatest(topxid, nsubxids, subxids);
910 
911  /*
912  * Mark all the subtransactions as observed.
913  *
914  * NOTE: This will fail if the subxid contains too many previously
915  * unobserved xids to fit into known-assigned-xids. That shouldn't happen
916  * as the code stands, because xid-assignment records should never contain
917  * more than PGPROC_MAX_CACHED_SUBXIDS entries.
918  */
920 
921  /*
922  * Notice that we update pg_subtrans with the top-level xid, rather than
923  * the parent xid. This is a difference between normal processing and
924  * recovery, yet is still correct in all cases. The reason is that
925  * subtransaction commit is not marked in clog until commit processing, so
926  * all aborted subtransactions have already been clearly marked in clog.
927  * As a result we are able to refer directly to the top-level
928  * transaction's state rather than skipping through all the intermediate
929  * states in the subtransaction tree. This should be the first time we
930  * have attempted to SubTransSetParent().
931  */
932  for (i = 0; i < nsubxids; i++)
933  SubTransSetParent(subxids[i], topxid);
934 
935  /* KnownAssignedXids isn't maintained yet, so we're done for now */
937  return;
938 
939  /*
940  * Uses same locking as transaction commit
941  */
942  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
943 
944  /*
945  * Remove subxids from known-assigned-xacts.
946  */
948 
949  /*
950  * Advance lastOverflowedXid to be at least the last of these subxids.
951  */
952  if (TransactionIdPrecedes(procArray->lastOverflowedXid, max_xid))
953  procArray->lastOverflowedXid = max_xid;
954 
955  LWLockRelease(ProcArrayLock);
956 }
957 
958 /*
959  * TransactionIdIsInProgress -- is given transaction running in some backend
960  *
961  * Aside from some shortcuts such as checking RecentXmin and our own Xid,
962  * there are four possibilities for finding a running transaction:
963  *
964  * 1. The given Xid is a main transaction Id. We will find this out cheaply
965  * by looking at the PGXACT struct for each backend.
966  *
967  * 2. The given Xid is one of the cached subxact Xids in the PGPROC array.
968  * We can find this out cheaply too.
969  *
970  * 3. In Hot Standby mode, we must search the KnownAssignedXids list to see
971  * if the Xid is running on the master.
972  *
973  * 4. Search the SubTrans tree to find the Xid's topmost parent, and then see
974  * if that is running according to PGXACT or KnownAssignedXids. This is the
975  * slowest way, but sadly it has to be done always if the others failed,
976  * unless we see that the cached subxact sets are complete (none have
977  * overflowed).
978  *
979  * ProcArrayLock has to be held while we do 1, 2, 3. If we save the top Xids
980  * while doing 1 and 3, we can release the ProcArrayLock while we do 4.
981  * This buys back some concurrency (and we can't retrieve the main Xids from
982  * PGXACT again anyway; see GetNewTransactionId).
983  */
984 bool
986 {
987  static TransactionId *xids = NULL;
988  int nxids = 0;
989  ProcArrayStruct *arrayP = procArray;
990  TransactionId topxid;
991  int i,
992  j;
993 
994  /*
995  * Don't bother checking a transaction older than RecentXmin; it could not
996  * possibly still be running. (Note: in particular, this guarantees that
997  * we reject InvalidTransactionId, FrozenTransactionId, etc as not
998  * running.)
999  */
1001  {
1003  return false;
1004  }
1005 
1006  /*
1007  * We may have just checked the status of this transaction, so if it is
1008  * already known to be completed, we can fall out without any access to
1009  * shared memory.
1010  */
1012  {
1014  return false;
1015  }
1016 
1017  /*
1018  * Also, we can handle our own transaction (and subtransactions) without
1019  * any access to shared memory.
1020  */
1022  {
1024  return true;
1025  }
1026 
1027  /*
1028  * If first time through, get workspace to remember main XIDs in. We
1029  * malloc it permanently to avoid repeated palloc/pfree overhead.
1030  */
1031  if (xids == NULL)
1032  {
1033  /*
1034  * In hot standby mode, reserve enough space to hold all xids in the
1035  * known-assigned list. If we later finish recovery, we no longer need
1036  * the bigger array, but we don't bother to shrink it.
1037  */
1038  int maxxids = RecoveryInProgress() ? TOTAL_MAX_CACHED_SUBXIDS : arrayP->maxProcs;
1039 
1040  xids = (TransactionId *) malloc(maxxids * sizeof(TransactionId));
1041  if (xids == NULL)
1042  ereport(ERROR,
1043  (errcode(ERRCODE_OUT_OF_MEMORY),
1044  errmsg("out of memory")));
1045  }
1046 
1047  LWLockAcquire(ProcArrayLock, LW_SHARED);
1048 
1049  /*
1050  * Now that we have the lock, we can check latestCompletedXid; if the
1051  * target Xid is after that, it's surely still running.
1052  */
1054  {
1055  LWLockRelease(ProcArrayLock);
1057  return true;
1058  }
1059 
1060  /* No shortcuts, gotta grovel through the array */
1061  for (i = 0; i < arrayP->numProcs; i++)
1062  {
1063  int pgprocno = arrayP->pgprocnos[i];
1064  PGPROC *proc = &allProcs[pgprocno];
1065  PGXACT *pgxact = &allPgXact[pgprocno];
1066  TransactionId pxid;
1067  int pxids;
1068 
1069  /* Ignore my own proc --- dealt with it above */
1070  if (proc == MyProc)
1071  continue;
1072 
1073  /* Fetch xid just once - see GetNewTransactionId */
1074  pxid = UINT32_ACCESS_ONCE(pgxact->xid);
1075 
1076  if (!TransactionIdIsValid(pxid))
1077  continue;
1078 
1079  /*
1080  * Step 1: check the main Xid
1081  */
1082  if (TransactionIdEquals(pxid, xid))
1083  {
1084  LWLockRelease(ProcArrayLock);
1086  return true;
1087  }
1088 
1089  /*
1090  * We can ignore main Xids that are younger than the target Xid, since
1091  * the target could not possibly be their child.
1092  */
1093  if (TransactionIdPrecedes(xid, pxid))
1094  continue;
1095 
1096  /*
1097  * Step 2: check the cached child-Xids arrays
1098  */
1099  pxids = pgxact->nxids;
1100  pg_read_barrier(); /* pairs with barrier in GetNewTransactionId() */
1101  for (j = pxids - 1; j >= 0; j--)
1102  {
1103  /* Fetch xid just once - see GetNewTransactionId */
1104  TransactionId cxid = UINT32_ACCESS_ONCE(proc->subxids.xids[j]);
1105 
1106  if (TransactionIdEquals(cxid, xid))
1107  {
1108  LWLockRelease(ProcArrayLock);
1110  return true;
1111  }
1112  }
1113 
1114  /*
1115  * Save the main Xid for step 4. We only need to remember main Xids
1116  * that have uncached children. (Note: there is no race condition
1117  * here because the overflowed flag cannot be cleared, only set, while
1118  * we hold ProcArrayLock. So we can't miss an Xid that we need to
1119  * worry about.)
1120  */
1121  if (pgxact->overflowed)
1122  xids[nxids++] = pxid;
1123  }
1124 
1125  /*
1126  * Step 3: in hot standby mode, check the known-assigned-xids list. XIDs
1127  * in the list must be treated as running.
1128  */
1129  if (RecoveryInProgress())
1130  {
1131  /* none of the PGXACT entries should have XIDs in hot standby mode */
1132  Assert(nxids == 0);
1133 
1134  if (KnownAssignedXidExists(xid))
1135  {
1136  LWLockRelease(ProcArrayLock);
1138  return true;
1139  }
1140 
1141  /*
1142  * If the KnownAssignedXids overflowed, we have to check pg_subtrans
1143  * too. Fetch all xids from KnownAssignedXids that are lower than
1144  * xid, since if xid is a subtransaction its parent will always have a
1145  * lower value. Note we will collect both main and subXIDs here, but
1146  * there's no help for it.
1147  */
1148  if (TransactionIdPrecedesOrEquals(xid, procArray->lastOverflowedXid))
1149  nxids = KnownAssignedXidsGet(xids, xid);
1150  }
1151 
1152  LWLockRelease(ProcArrayLock);
1153 
1154  /*
1155  * If none of the relevant caches overflowed, we know the Xid is not
1156  * running without even looking at pg_subtrans.
1157  */
1158  if (nxids == 0)
1159  {
1161  return false;
1162  }
1163 
1164  /*
1165  * Step 4: have to check pg_subtrans.
1166  *
1167  * At this point, we know it's either a subtransaction of one of the Xids
1168  * in xids[], or it's not running. If it's an already-failed
1169  * subtransaction, we want to say "not running" even though its parent may
1170  * still be running. So first, check pg_xact to see if it's been aborted.
1171  */
1173 
1174  if (TransactionIdDidAbort(xid))
1175  return false;
1176 
1177  /*
1178  * It isn't aborted, so check whether the transaction tree it belongs to
1179  * is still running (or, more precisely, whether it was running when we
1180  * held ProcArrayLock).
1181  */
1182  topxid = SubTransGetTopmostTransaction(xid);
1183  Assert(TransactionIdIsValid(topxid));
1184  if (!TransactionIdEquals(topxid, xid))
1185  {
1186  for (i = 0; i < nxids; i++)
1187  {
1188  if (TransactionIdEquals(xids[i], topxid))
1189  return true;
1190  }
1191  }
1192 
1193  return false;
1194 }
1195 
1196 /*
1197  * TransactionIdIsActive -- is xid the top-level XID of an active backend?
1198  *
1199  * This differs from TransactionIdIsInProgress in that it ignores prepared
1200  * transactions, as well as transactions running on the master if we're in
1201  * hot standby. Also, we ignore subtransactions since that's not needed
1202  * for current uses.
1203  */
1204 bool
1206 {
1207  bool result = false;
1208  ProcArrayStruct *arrayP = procArray;
1209  int i;
1210 
1211  /*
1212  * Don't bother checking a transaction older than RecentXmin; it could not
1213  * possibly still be running.
1214  */
1216  return false;
1217 
1218  LWLockAcquire(ProcArrayLock, LW_SHARED);
1219 
1220  for (i = 0; i < arrayP->numProcs; i++)
1221  {
1222  int pgprocno = arrayP->pgprocnos[i];
1223  PGPROC *proc = &allProcs[pgprocno];
1224  PGXACT *pgxact = &allPgXact[pgprocno];
1225  TransactionId pxid;
1226 
1227  /* Fetch xid just once - see GetNewTransactionId */
1228  pxid = UINT32_ACCESS_ONCE(pgxact->xid);
1229 
1230  if (!TransactionIdIsValid(pxid))
1231  continue;
1232 
1233  if (proc->pid == 0)
1234  continue; /* ignore prepared transactions */
1235 
1236  if (TransactionIdEquals(pxid, xid))
1237  {
1238  result = true;
1239  break;
1240  }
1241  }
1242 
1243  LWLockRelease(ProcArrayLock);
1244 
1245  return result;
1246 }
1247 
1248 
1249 /*
1250  * GetOldestXmin -- returns oldest transaction that was running
1251  * when any current transaction was started.
1252  *
1253  * If rel is NULL or a shared relation, all backends are considered, otherwise
1254  * only backends running in this database are considered.
1255  *
1256  * The flags are used to ignore the backends in calculation when any of the
1257  * corresponding flags is set. Typically, if you want to ignore ones with
1258  * PROC_IN_VACUUM flag, you can use PROCARRAY_FLAGS_VACUUM.
1259  *
1260  * PROCARRAY_SLOTS_XMIN causes GetOldestXmin to ignore the xmin and
1261  * catalog_xmin of any replication slots that exist in the system when
1262  * calculating the oldest xmin.
1263  *
1264  * This is used by VACUUM to decide which deleted tuples must be preserved in
1265  * the passed in table. For shared relations backends in all databases must be
1266  * considered, but for non-shared relations that's not required, since only
1267  * backends in my own database could ever see the tuples in them. Also, we can
1268  * ignore concurrently running lazy VACUUMs because (a) they must be working
1269  * on other tables, and (b) they don't need to do snapshot-based lookups.
1270  *
1271  * This is also used to determine where to truncate pg_subtrans. For that
1272  * backends in all databases have to be considered, so rel = NULL has to be
1273  * passed in.
1274  *
1275  * Note: we include all currently running xids in the set of considered xids.
1276  * This ensures that if a just-started xact has not yet set its snapshot,
1277  * when it does set the snapshot it cannot set xmin less than what we compute.
1278  * See notes in src/backend/access/transam/README.
1279  *
1280  * Note: despite the above, it's possible for the calculated value to move
1281  * backwards on repeated calls. The calculated value is conservative, so that
1282  * anything older is definitely not considered as running by anyone anymore,
1283  * but the exact value calculated depends on a number of things. For example,
1284  * if rel = NULL and there are no transactions running in the current
1285  * database, GetOldestXmin() returns latestCompletedXid. If a transaction
1286  * begins after that, its xmin will include in-progress transactions in other
1287  * databases that started earlier, so another call will return a lower value.
1288  * Nonetheless it is safe to vacuum a table in the current database with the
1289  * first result. There are also replication-related effects: a walsender
1290  * process can set its xmin based on transactions that are no longer running
1291  * in the master but are still being replayed on the standby, thus possibly
1292  * making the GetOldestXmin reading go backwards. In this case there is a
1293  * possibility that we lose data that the standby would like to have, but
1294  * unless the standby uses a replication slot to make its xmin persistent
1295  * there is little we can do about that --- data is only protected if the
1296  * walsender runs continuously while queries are executed on the standby.
1297  * (The Hot Standby code deals with such cases by failing standby queries
1298  * that needed to access already-removed data, so there's no integrity bug.)
1299  * The return value is also adjusted with vacuum_defer_cleanup_age, so
1300  * increasing that setting on the fly is another easy way to make
1301  * GetOldestXmin() move backwards, with no consequences for data integrity.
1302  */
1304 GetOldestXmin(Relation rel, int flags)
1305 {
1306  ProcArrayStruct *arrayP = procArray;
1307  TransactionId result;
1308  int index;
1309  bool allDbs;
1310 
1313 
1314  /*
1315  * If we're not computing a relation specific limit, or if a shared
1316  * relation has been passed in, backends in all databases have to be
1317  * considered.
1318  */
1319  allDbs = rel == NULL || rel->rd_rel->relisshared;
1320 
1321  /* Cannot look for individual databases during recovery */
1322  Assert(allDbs || !RecoveryInProgress());
1323 
1324  LWLockAcquire(ProcArrayLock, LW_SHARED);
1325 
1326  /*
1327  * We initialize the MIN() calculation with latestCompletedXid + 1. This
1328  * is a lower bound for the XIDs that might appear in the ProcArray later,
1329  * and so protects us against overestimating the result due to future
1330  * additions.
1331  */
1333  Assert(TransactionIdIsNormal(result));
1334  TransactionIdAdvance(result);
1335 
1336  for (index = 0; index < arrayP->numProcs; index++)
1337  {
1338  int pgprocno = arrayP->pgprocnos[index];
1339  PGPROC *proc = &allProcs[pgprocno];
1340  PGXACT *pgxact = &allPgXact[pgprocno];
1341 
1342  if (pgxact->vacuumFlags & (flags & PROCARRAY_PROC_FLAGS_MASK))
1343  continue;
1344 
1345  if (allDbs ||
1346  proc->databaseId == MyDatabaseId ||
1347  proc->databaseId == 0) /* always include WalSender */
1348  {
1349  /* Fetch xid just once - see GetNewTransactionId */
1350  TransactionId xid = UINT32_ACCESS_ONCE(pgxact->xid);
1351 
1352  /* First consider the transaction's own Xid, if any */
1353  if (TransactionIdIsNormal(xid) &&
1354  TransactionIdPrecedes(xid, result))
1355  result = xid;
1356 
1357  /*
1358  * Also consider the transaction's Xmin, if set.
1359  *
1360  * We must check both Xid and Xmin because a transaction might
1361  * have an Xmin but not (yet) an Xid; conversely, if it has an
1362  * Xid, that could determine some not-yet-set Xmin.
1363  */
1364  xid = UINT32_ACCESS_ONCE(pgxact->xmin);
1365  if (TransactionIdIsNormal(xid) &&
1366  TransactionIdPrecedes(xid, result))
1367  result = xid;
1368  }
1369  }
1370 
1371  /*
1372  * Fetch into local variable while ProcArrayLock is held - the
1373  * LWLockRelease below is a barrier, ensuring this happens inside the
1374  * lock.
1375  */
1376  replication_slot_xmin = procArray->replication_slot_xmin;
1377  replication_slot_catalog_xmin = procArray->replication_slot_catalog_xmin;
1378 
1379  if (RecoveryInProgress())
1380  {
1381  /*
1382  * Check to see whether KnownAssignedXids contains an xid value older
1383  * than the main procarray.
1384  */
1386 
1387  LWLockRelease(ProcArrayLock);
1388 
1389  if (TransactionIdIsNormal(kaxmin) &&
1390  TransactionIdPrecedes(kaxmin, result))
1391  result = kaxmin;
1392  }
1393  else
1394  {
1395  /*
1396  * No other information needed, so release the lock immediately.
1397  */
1398  LWLockRelease(ProcArrayLock);
1399 
1400  /*
1401  * Compute the cutoff XID by subtracting vacuum_defer_cleanup_age,
1402  * being careful not to generate a "permanent" XID.
1403  *
1404  * vacuum_defer_cleanup_age provides some additional "slop" for the
1405  * benefit of hot standby queries on standby servers. This is quick
1406  * and dirty, and perhaps not all that useful unless the master has a
1407  * predictable transaction rate, but it offers some protection when
1408  * there's no walsender connection. Note that we are assuming
1409  * vacuum_defer_cleanup_age isn't large enough to cause wraparound ---
1410  * so guc.c should limit it to no more than the xidStopLimit threshold
1411  * in varsup.c. Also note that we intentionally don't apply
1412  * vacuum_defer_cleanup_age on standby servers.
1413  */
1414  result -= vacuum_defer_cleanup_age;
1415  if (!TransactionIdIsNormal(result))
1416  result = FirstNormalTransactionId;
1417  }
1418 
1419  /*
1420  * Check whether there are replication slots requiring an older xmin.
1421  */
1422  if (!(flags & PROCARRAY_SLOTS_XMIN) &&
1423  TransactionIdIsValid(replication_slot_xmin) &&
1424  NormalTransactionIdPrecedes(replication_slot_xmin, result))
1425  result = replication_slot_xmin;
1426 
1427  /*
1428  * After locks have been released and vacuum_defer_cleanup_age has been
1429  * applied, check whether we need to back up further to make logical
1430  * decoding possible. We need to do so if we're computing the global limit
1431  * (rel = NULL) or if the passed relation is a catalog relation of some
1432  * kind.
1433  */
1434  if (!(flags & PROCARRAY_SLOTS_XMIN) &&
1435  (rel == NULL ||
1437  TransactionIdIsValid(replication_slot_catalog_xmin) &&
1438  NormalTransactionIdPrecedes(replication_slot_catalog_xmin, result))
1440 
1441  return result;
1442 }
1443 
1444 /*
1445  * GetMaxSnapshotXidCount -- get max size for snapshot XID array
1446  *
1447  * We have to export this for use by snapmgr.c.
1448  */
1449 int
1451 {
1452  return procArray->maxProcs;
1453 }
1454 
1455 /*
1456  * GetMaxSnapshotSubxidCount -- get max size for snapshot sub-XID array
1457  *
1458  * We have to export this for use by snapmgr.c.
1459  */
1460 int
1462 {
1463  return TOTAL_MAX_CACHED_SUBXIDS;
1464 }
1465 
1466 /*
1467  * GetSnapshotData -- returns information about running transactions.
1468  *
1469  * The returned snapshot includes xmin (lowest still-running xact ID),
1470  * xmax (highest completed xact ID + 1), and a list of running xact IDs
1471  * in the range xmin <= xid < xmax. It is used as follows:
1472  * All xact IDs < xmin are considered finished.
1473  * All xact IDs >= xmax are considered still running.
1474  * For an xact ID xmin <= xid < xmax, consult list to see whether
1475  * it is considered running or not.
1476  * This ensures that the set of transactions seen as "running" by the
1477  * current xact will not change after it takes the snapshot.
1478  *
1479  * All running top-level XIDs are included in the snapshot, except for lazy
1480  * VACUUM processes. We also try to include running subtransaction XIDs,
1481  * but since PGPROC has only a limited cache area for subxact XIDs, full
1482  * information may not be available. If we find any overflowed subxid arrays,
1483  * we have to mark the snapshot's subxid data as overflowed, and extra work
1484  * *may* need to be done to determine what's running (see XidInMVCCSnapshot()
1485  * in heapam_visibility.c).
1486  *
1487  * We also update the following backend-global variables:
1488  * TransactionXmin: the oldest xmin of any snapshot in use in the
1489  * current transaction (this is the same as MyPgXact->xmin).
1490  * RecentXmin: the xmin computed for the most recent snapshot. XIDs
1491  * older than this are known not running any more.
1492  * RecentGlobalXmin: the global xmin (oldest TransactionXmin across all
1493  * running transactions, except those running LAZY VACUUM). This is
1494  * the same computation done by
1495  * GetOldestXmin(NULL, PROCARRAY_FLAGS_VACUUM).
1496  * RecentGlobalDataXmin: the global xmin for non-catalog tables
1497  * >= RecentGlobalXmin
1498  *
1499  * Note: this function should probably not be called with an argument that's
1500  * not statically allocated (see xip allocation below).
1501  */
1502 Snapshot
1504 {
1505  ProcArrayStruct *arrayP = procArray;
1506  TransactionId xmin;
1507  TransactionId xmax;
1508  TransactionId globalxmin;
1509  int index;
1510  int count = 0;
1511  int subcount = 0;
1512  bool suboverflowed = false;
1515 
1516  Assert(snapshot != NULL);
1517 
1518  /*
1519  * Allocating space for maxProcs xids is usually overkill; numProcs would
1520  * be sufficient. But it seems better to do the malloc while not holding
1521  * the lock, so we can't look at numProcs. Likewise, we allocate much
1522  * more subxip storage than is probably needed.
1523  *
1524  * This does open a possibility for avoiding repeated malloc/free: since
1525  * maxProcs does not change at runtime, we can simply reuse the previous
1526  * xip arrays if any. (This relies on the fact that all callers pass
1527  * static SnapshotData structs.)
1528  */
1529  if (snapshot->xip == NULL)
1530  {
1531  /*
1532  * First call for this snapshot. Snapshot is same size whether or not
1533  * we are in recovery, see later comments.
1534  */
1535  snapshot->xip = (TransactionId *)
1537  if (snapshot->xip == NULL)
1538  ereport(ERROR,
1539  (errcode(ERRCODE_OUT_OF_MEMORY),
1540  errmsg("out of memory")));
1541  Assert(snapshot->subxip == NULL);
1542  snapshot->subxip = (TransactionId *)
1544  if (snapshot->subxip == NULL)
1545  ereport(ERROR,
1546  (errcode(ERRCODE_OUT_OF_MEMORY),
1547  errmsg("out of memory")));
1548  }
1549 
1550  /*
1551  * It is sufficient to get shared lock on ProcArrayLock, even if we are
1552  * going to set MyPgXact->xmin.
1553  */
1554  LWLockAcquire(ProcArrayLock, LW_SHARED);
1555 
1556  /* xmax is always latestCompletedXid + 1 */
1559  TransactionIdAdvance(xmax);
1560 
1561  /* initialize xmin calculation with xmax */
1562  globalxmin = xmin = xmax;
1563 
1565 
1566  if (!snapshot->takenDuringRecovery)
1567  {
1568  int *pgprocnos = arrayP->pgprocnos;
1569  int numProcs;
1570 
1571  /*
1572  * Spin over procArray checking xid, xmin, and subxids. The goal is
1573  * to gather all active xids, find the lowest xmin, and try to record
1574  * subxids.
1575  */
1576  numProcs = arrayP->numProcs;
1577  for (index = 0; index < numProcs; index++)
1578  {
1579  int pgprocno = pgprocnos[index];
1580  PGXACT *pgxact = &allPgXact[pgprocno];
1581  TransactionId xid;
1582 
1583  /*
1584  * Skip over backends doing logical decoding which manages xmin
1585  * separately (check below) and ones running LAZY VACUUM.
1586  */
1587  if (pgxact->vacuumFlags &
1589  continue;
1590 
1591  /* Update globalxmin to be the smallest valid xmin */
1592  xid = UINT32_ACCESS_ONCE(pgxact->xmin);
1593  if (TransactionIdIsNormal(xid) &&
1594  NormalTransactionIdPrecedes(xid, globalxmin))
1595  globalxmin = xid;
1596 
1597  /* Fetch xid just once - see GetNewTransactionId */
1598  xid = UINT32_ACCESS_ONCE(pgxact->xid);
1599 
1600  /*
1601  * If the transaction has no XID assigned, we can skip it; it
1602  * won't have sub-XIDs either. If the XID is >= xmax, we can also
1603  * skip it; such transactions will be treated as running anyway
1604  * (and any sub-XIDs will also be >= xmax).
1605  */
1606  if (!TransactionIdIsNormal(xid)
1607  || !NormalTransactionIdPrecedes(xid, xmax))
1608  continue;
1609 
1610  /*
1611  * We don't include our own XIDs (if any) in the snapshot, but we
1612  * must include them in xmin.
1613  */
1614  if (NormalTransactionIdPrecedes(xid, xmin))
1615  xmin = xid;
1616  if (pgxact == MyPgXact)
1617  continue;
1618 
1619  /* Add XID to snapshot. */
1620  snapshot->xip[count++] = xid;
1621 
1622  /*
1623  * Save subtransaction XIDs if possible (if we've already
1624  * overflowed, there's no point). Note that the subxact XIDs must
1625  * be later than their parent, so no need to check them against
1626  * xmin. We could filter against xmax, but it seems better not to
1627  * do that much work while holding the ProcArrayLock.
1628  *
1629  * The other backend can add more subxids concurrently, but cannot
1630  * remove any. Hence it's important to fetch nxids just once.
1631  * Should be safe to use memcpy, though. (We needn't worry about
1632  * missing any xids added concurrently, because they must postdate
1633  * xmax.)
1634  *
1635  * Again, our own XIDs are not included in the snapshot.
1636  */
1637  if (!suboverflowed)
1638  {
1639  if (pgxact->overflowed)
1640  suboverflowed = true;
1641  else
1642  {
1643  int nxids = pgxact->nxids;
1644 
1645  if (nxids > 0)
1646  {
1647  PGPROC *proc = &allProcs[pgprocno];
1648 
1649  pg_read_barrier(); /* pairs with GetNewTransactionId */
1650 
1651  memcpy(snapshot->subxip + subcount,
1652  (void *) proc->subxids.xids,
1653  nxids * sizeof(TransactionId));
1654  subcount += nxids;
1655  }
1656  }
1657  }
1658  }
1659  }
1660  else
1661  {
1662  /*
1663  * We're in hot standby, so get XIDs from KnownAssignedXids.
1664  *
1665  * We store all xids directly into subxip[]. Here's why:
1666  *
1667  * In recovery we don't know which xids are top-level and which are
1668  * subxacts, a design choice that greatly simplifies xid processing.
1669  *
1670  * It seems like we would want to try to put xids into xip[] only, but
1671  * that is fairly small. We would either need to make that bigger or
1672  * to increase the rate at which we WAL-log xid assignment; neither is
1673  * an appealing choice.
1674  *
1675  * We could try to store xids into xip[] first and then into subxip[]
1676  * if there are too many xids. That only works if the snapshot doesn't
1677  * overflow because we do not search subxip[] in that case. A simpler
1678  * way is to just store all xids in the subxact array because this is
1679  * by far the bigger array. We just leave the xip array empty.
1680  *
1681  * Either way we need to change the way XidInMVCCSnapshot() works
1682  * depending upon when the snapshot was taken, or change normal
1683  * snapshot processing so it matches.
1684  *
1685  * Note: It is possible for recovery to end before we finish taking
1686  * the snapshot, and for newly assigned transaction ids to be added to
1687  * the ProcArray. xmax cannot change while we hold ProcArrayLock, so
1688  * those newly added transaction ids would be filtered away, so we
1689  * need not be concerned about them.
1690  */
1691  subcount = KnownAssignedXidsGetAndSetXmin(snapshot->subxip, &xmin,
1692  xmax);
1693 
1694  if (TransactionIdPrecedesOrEquals(xmin, procArray->lastOverflowedXid))
1695  suboverflowed = true;
1696  }
1697 
1698 
1699  /*
1700  * Fetch into local variable while ProcArrayLock is held - the
1701  * LWLockRelease below is a barrier, ensuring this happens inside the
1702  * lock.
1703  */
1704  replication_slot_xmin = procArray->replication_slot_xmin;
1705  replication_slot_catalog_xmin = procArray->replication_slot_catalog_xmin;
1706 
1708  MyPgXact->xmin = TransactionXmin = xmin;
1709 
1710  LWLockRelease(ProcArrayLock);
1711 
1712  /*
1713  * Update globalxmin to include actual process xids. This is a slightly
1714  * different way of computing it than GetOldestXmin uses, but should give
1715  * the same result.
1716  */
1717  if (TransactionIdPrecedes(xmin, globalxmin))
1718  globalxmin = xmin;
1719 
1720  /* Update global variables too */
1724 
1725  /* Check whether there's a replication slot requiring an older xmin. */
1726  if (TransactionIdIsValid(replication_slot_xmin) &&
1727  NormalTransactionIdPrecedes(replication_slot_xmin, RecentGlobalXmin))
1729 
1730  /* Non-catalog tables can be vacuumed if older than this xid */
1732 
1733  /*
1734  * Check whether there's a replication slot requiring an older catalog
1735  * xmin.
1736  */
1737  if (TransactionIdIsNormal(replication_slot_catalog_xmin) &&
1738  NormalTransactionIdPrecedes(replication_slot_catalog_xmin, RecentGlobalXmin))
1740 
1741  RecentXmin = xmin;
1742 
1743  snapshot->xmin = xmin;
1744  snapshot->xmax = xmax;
1745  snapshot->xcnt = count;
1746  snapshot->subxcnt = subcount;
1747  snapshot->suboverflowed = suboverflowed;
1748 
1749  snapshot->curcid = GetCurrentCommandId(false);
1750 
1751  /*
1752  * This is a new snapshot, so set both refcounts are zero, and mark it as
1753  * not copied in persistent memory.
1754  */
1755  snapshot->active_count = 0;
1756  snapshot->regd_count = 0;
1757  snapshot->copied = false;
1758 
1759  if (old_snapshot_threshold < 0)
1760  {
1761  /*
1762  * If not using "snapshot too old" feature, fill related fields with
1763  * dummy values that don't require any locking.
1764  */
1765  snapshot->lsn = InvalidXLogRecPtr;
1766  snapshot->whenTaken = 0;
1767  }
1768  else
1769  {
1770  /*
1771  * Capture the current time and WAL stream location in case this
1772  * snapshot becomes old enough to need to fall back on the special
1773  * "old snapshot" logic.
1774  */
1775  snapshot->lsn = GetXLogInsertRecPtr();
1776  snapshot->whenTaken = GetSnapshotCurrentTimestamp();
1777  MaintainOldSnapshotTimeMapping(snapshot->whenTaken, xmin);
1778  }
1779 
1780  return snapshot;
1781 }
1782 
1783 /*
1784  * ProcArrayInstallImportedXmin -- install imported xmin into MyPgXact->xmin
1785  *
1786  * This is called when installing a snapshot imported from another
1787  * transaction. To ensure that OldestXmin doesn't go backwards, we must
1788  * check that the source transaction is still running, and we'd better do
1789  * that atomically with installing the new xmin.
1790  *
1791  * Returns true if successful, false if source xact is no longer running.
1792  */
1793 bool
1795  VirtualTransactionId *sourcevxid)
1796 {
1797  bool result = false;
1798  ProcArrayStruct *arrayP = procArray;
1799  int index;
1800 
1802  if (!sourcevxid)
1803  return false;
1804 
1805  /* Get lock so source xact can't end while we're doing this */
1806  LWLockAcquire(ProcArrayLock, LW_SHARED);
1807 
1808  for (index = 0; index < arrayP->numProcs; index++)
1809  {
1810  int pgprocno = arrayP->pgprocnos[index];
1811  PGPROC *proc = &allProcs[pgprocno];
1812  PGXACT *pgxact = &allPgXact[pgprocno];
1813  TransactionId xid;
1814 
1815  /* Ignore procs running LAZY VACUUM */
1816  if (pgxact->vacuumFlags & PROC_IN_VACUUM)
1817  continue;
1818 
1819  /* We are only interested in the specific virtual transaction. */
1820  if (proc->backendId != sourcevxid->backendId)
1821  continue;
1822  if (proc->lxid != sourcevxid->localTransactionId)
1823  continue;
1824 
1825  /*
1826  * We check the transaction's database ID for paranoia's sake: if it's
1827  * in another DB then its xmin does not cover us. Caller should have
1828  * detected this already, so we just treat any funny cases as
1829  * "transaction not found".
1830  */
1831  if (proc->databaseId != MyDatabaseId)
1832  continue;
1833 
1834  /*
1835  * Likewise, let's just make real sure its xmin does cover us.
1836  */
1837  xid = UINT32_ACCESS_ONCE(pgxact->xmin);
1838  if (!TransactionIdIsNormal(xid) ||
1839  !TransactionIdPrecedesOrEquals(xid, xmin))
1840  continue;
1841 
1842  /*
1843  * We're good. Install the new xmin. As in GetSnapshotData, set
1844  * TransactionXmin too. (Note that because snapmgr.c called
1845  * GetSnapshotData first, we'll be overwriting a valid xmin here, so
1846  * we don't check that.)
1847  */
1848  MyPgXact->xmin = TransactionXmin = xmin;
1849 
1850  result = true;
1851  break;
1852  }
1853 
1854  LWLockRelease(ProcArrayLock);
1855 
1856  return result;
1857 }
1858 
1859 /*
1860  * ProcArrayInstallRestoredXmin -- install restored xmin into MyPgXact->xmin
1861  *
1862  * This is like ProcArrayInstallImportedXmin, but we have a pointer to the
1863  * PGPROC of the transaction from which we imported the snapshot, rather than
1864  * an XID.
1865  *
1866  * Returns true if successful, false if source xact is no longer running.
1867  */
1868 bool
1870 {
1871  bool result = false;
1872  TransactionId xid;
1873  PGXACT *pgxact;
1874 
1876  Assert(proc != NULL);
1877 
1878  /* Get lock so source xact can't end while we're doing this */
1879  LWLockAcquire(ProcArrayLock, LW_SHARED);
1880 
1881  pgxact = &allPgXact[proc->pgprocno];
1882 
1883  /*
1884  * Be certain that the referenced PGPROC has an advertised xmin which is
1885  * no later than the one we're installing, so that the system-wide xmin
1886  * can't go backwards. Also, make sure it's running in the same database,
1887  * so that the per-database xmin cannot go backwards.
1888  */
1889  xid = UINT32_ACCESS_ONCE(pgxact->xmin);
1890  if (proc->databaseId == MyDatabaseId &&
1891  TransactionIdIsNormal(xid) &&
1892  TransactionIdPrecedesOrEquals(xid, xmin))
1893  {
1894  MyPgXact->xmin = TransactionXmin = xmin;
1895  result = true;
1896  }
1897 
1898  LWLockRelease(ProcArrayLock);
1899 
1900  return result;
1901 }
1902 
1903 /*
1904  * GetRunningTransactionData -- returns information about running transactions.
1905  *
1906  * Similar to GetSnapshotData but returns more information. We include
1907  * all PGXACTs with an assigned TransactionId, even VACUUM processes and
1908  * prepared transactions.
1909  *
1910  * We acquire XidGenLock and ProcArrayLock, but the caller is responsible for
1911  * releasing them. Acquiring XidGenLock ensures that no new XIDs enter the proc
1912  * array until the caller has WAL-logged this snapshot, and releases the
1913  * lock. Acquiring ProcArrayLock ensures that no transactions commit until the
1914  * lock is released.
1915  *
1916  * The returned data structure is statically allocated; caller should not
1917  * modify it, and must not assume it is valid past the next call.
1918  *
1919  * This is never executed during recovery so there is no need to look at
1920  * KnownAssignedXids.
1921  *
1922  * Dummy PGXACTs from prepared transaction are included, meaning that this
1923  * may return entries with duplicated TransactionId values coming from
1924  * transaction finishing to prepare. Nothing is done about duplicated
1925  * entries here to not hold on ProcArrayLock more than necessary.
1926  *
1927  * We don't worry about updating other counters, we want to keep this as
1928  * simple as possible and leave GetSnapshotData() as the primary code for
1929  * that bookkeeping.
1930  *
1931  * Note that if any transaction has overflowed its cached subtransactions
1932  * then there is no real need include any subtransactions.
1933  */
1936 {
1937  /* result workspace */
1938  static RunningTransactionsData CurrentRunningXactsData;
1939 
1940  ProcArrayStruct *arrayP = procArray;
1941  RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData;
1942  TransactionId latestCompletedXid;
1943  TransactionId oldestRunningXid;
1944  TransactionId *xids;
1945  int index;
1946  int count;
1947  int subcount;
1948  bool suboverflowed;
1949 
1951 
1952  /*
1953  * Allocating space for maxProcs xids is usually overkill; numProcs would
1954  * be sufficient. But it seems better to do the malloc while not holding
1955  * the lock, so we can't look at numProcs. Likewise, we allocate much
1956  * more subxip storage than is probably needed.
1957  *
1958  * Should only be allocated in bgwriter, since only ever executed during
1959  * checkpoints.
1960  */
1961  if (CurrentRunningXacts->xids == NULL)
1962  {
1963  /*
1964  * First call
1965  */
1966  CurrentRunningXacts->xids = (TransactionId *)
1968  if (CurrentRunningXacts->xids == NULL)
1969  ereport(ERROR,
1970  (errcode(ERRCODE_OUT_OF_MEMORY),
1971  errmsg("out of memory")));
1972  }
1973 
1974  xids = CurrentRunningXacts->xids;
1975 
1976  count = subcount = 0;
1977  suboverflowed = false;
1978 
1979  /*
1980  * Ensure that no xids enter or leave the procarray while we obtain
1981  * snapshot.
1982  */
1983  LWLockAcquire(ProcArrayLock, LW_SHARED);
1984  LWLockAcquire(XidGenLock, LW_SHARED);
1985 
1986  latestCompletedXid = ShmemVariableCache->latestCompletedXid;
1987 
1989 
1990  /*
1991  * Spin over procArray collecting all xids
1992  */
1993  for (index = 0; index < arrayP->numProcs; index++)
1994  {
1995  int pgprocno = arrayP->pgprocnos[index];
1996  PGXACT *pgxact = &allPgXact[pgprocno];
1997  TransactionId xid;
1998 
1999  /* Fetch xid just once - see GetNewTransactionId */
2000  xid = UINT32_ACCESS_ONCE(pgxact->xid);
2001 
2002  /*
2003  * We don't need to store transactions that don't have a TransactionId
2004  * yet because they will not show as running on a standby server.
2005  */
2006  if (!TransactionIdIsValid(xid))
2007  continue;
2008 
2009  /*
2010  * Be careful not to exclude any xids before calculating the values of
2011  * oldestRunningXid and suboverflowed, since these are used to clean
2012  * up transaction information held on standbys.
2013  */
2014  if (TransactionIdPrecedes(xid, oldestRunningXid))
2015  oldestRunningXid = xid;
2016 
2017  if (pgxact->overflowed)
2018  suboverflowed = true;
2019 
2020  /*
2021  * If we wished to exclude xids this would be the right place for it.
2022  * Procs with the PROC_IN_VACUUM flag set don't usually assign xids,
2023  * but they do during truncation at the end when they get the lock and
2024  * truncate, so it is not much of a problem to include them if they
2025  * are seen and it is cleaner to include them.
2026  */
2027 
2028  xids[count++] = xid;
2029  }
2030 
2031  /*
2032  * Spin over procArray collecting all subxids, but only if there hasn't
2033  * been a suboverflow.
2034  */
2035  if (!suboverflowed)
2036  {
2037  for (index = 0; index < arrayP->numProcs; index++)
2038  {
2039  int pgprocno = arrayP->pgprocnos[index];
2040  PGPROC *proc = &allProcs[pgprocno];
2041  PGXACT *pgxact = &allPgXact[pgprocno];
2042  int nxids;
2043 
2044  /*
2045  * Save subtransaction XIDs. Other backends can't add or remove
2046  * entries while we're holding XidGenLock.
2047  */
2048  nxids = pgxact->nxids;
2049  if (nxids > 0)
2050  {
2051  /* barrier not really required, as XidGenLock is held, but ... */
2052  pg_read_barrier(); /* pairs with GetNewTransactionId */
2053 
2054  memcpy(&xids[count], (void *) proc->subxids.xids,
2055  nxids * sizeof(TransactionId));
2056  count += nxids;
2057  subcount += nxids;
2058 
2059  /*
2060  * Top-level XID of a transaction is always less than any of
2061  * its subxids, so we don't need to check if any of the
2062  * subxids are smaller than oldestRunningXid
2063  */
2064  }
2065  }
2066  }
2067 
2068  /*
2069  * It's important *not* to include the limits set by slots here because
2070  * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those
2071  * were to be included here the initial value could never increase because
2072  * of a circular dependency where slots only increase their limits when
2073  * running xacts increases oldestRunningXid and running xacts only
2074  * increases if slots do.
2075  */
2076 
2077  CurrentRunningXacts->xcnt = count - subcount;
2078  CurrentRunningXacts->subxcnt = subcount;
2079  CurrentRunningXacts->subxid_overflow = suboverflowed;
2081  CurrentRunningXacts->oldestRunningXid = oldestRunningXid;
2082  CurrentRunningXacts->latestCompletedXid = latestCompletedXid;
2083 
2084  Assert(TransactionIdIsValid(CurrentRunningXacts->nextXid));
2085  Assert(TransactionIdIsValid(CurrentRunningXacts->oldestRunningXid));
2086  Assert(TransactionIdIsNormal(CurrentRunningXacts->latestCompletedXid));
2087 
2088  /* We don't release the locks here, the caller is responsible for that */
2089 
2090  return CurrentRunningXacts;
2091 }
2092 
2093 /*
2094  * GetOldestActiveTransactionId()
2095  *
2096  * Similar to GetSnapshotData but returns just oldestActiveXid. We include
2097  * all PGXACTs with an assigned TransactionId, even VACUUM processes.
2098  * We look at all databases, though there is no need to include WALSender
2099  * since this has no effect on hot standby conflicts.
2100  *
2101  * This is never executed during recovery so there is no need to look at
2102  * KnownAssignedXids.
2103  *
2104  * We don't worry about updating other counters, we want to keep this as
2105  * simple as possible and leave GetSnapshotData() as the primary code for
2106  * that bookkeeping.
2107  */
2110 {
2111  ProcArrayStruct *arrayP = procArray;
2112  TransactionId oldestRunningXid;
2113  int index;
2114 
2116 
2117  /*
2118  * Read nextXid, as the upper bound of what's still active.
2119  *
2120  * Reading a TransactionId is atomic, but we must grab the lock to make
2121  * sure that all XIDs < nextXid are already present in the proc array (or
2122  * have already completed), when we spin over it.
2123  */
2124  LWLockAcquire(XidGenLock, LW_SHARED);
2126  LWLockRelease(XidGenLock);
2127 
2128  /*
2129  * Spin over procArray collecting all xids and subxids.
2130  */
2131  LWLockAcquire(ProcArrayLock, LW_SHARED);
2132  for (index = 0; index < arrayP->numProcs; index++)
2133  {
2134  int pgprocno = arrayP->pgprocnos[index];
2135  PGXACT *pgxact = &allPgXact[pgprocno];
2136  TransactionId xid;
2137 
2138  /* Fetch xid just once - see GetNewTransactionId */
2139  xid = UINT32_ACCESS_ONCE(pgxact->xid);
2140 
2141  if (!TransactionIdIsNormal(xid))
2142  continue;
2143 
2144  if (TransactionIdPrecedes(xid, oldestRunningXid))
2145  oldestRunningXid = xid;
2146 
2147  /*
2148  * Top-level XID of a transaction is always less than any of its
2149  * subxids, so we don't need to check if any of the subxids are
2150  * smaller than oldestRunningXid
2151  */
2152  }
2153  LWLockRelease(ProcArrayLock);
2154 
2155  return oldestRunningXid;
2156 }
2157 
2158 /*
2159  * GetOldestSafeDecodingTransactionId -- lowest xid not affected by vacuum
2160  *
2161  * Returns the oldest xid that we can guarantee not to have been affected by
2162  * vacuum, i.e. no rows >= that xid have been vacuumed away unless the
2163  * transaction aborted. Note that the value can (and most of the time will) be
2164  * much more conservative than what really has been affected by vacuum, but we
2165  * currently don't have better data available.
2166  *
2167  * This is useful to initialize the cutoff xid after which a new changeset
2168  * extraction replication slot can start decoding changes.
2169  *
2170  * Must be called with ProcArrayLock held either shared or exclusively,
2171  * although most callers will want to use exclusive mode since it is expected
2172  * that the caller will immediately use the xid to peg the xmin horizon.
2173  */
2176 {
2177  ProcArrayStruct *arrayP = procArray;
2178  TransactionId oldestSafeXid;
2179  int index;
2180  bool recovery_in_progress = RecoveryInProgress();
2181 
2182  Assert(LWLockHeldByMe(ProcArrayLock));
2183 
2184  /*
2185  * Acquire XidGenLock, so no transactions can acquire an xid while we're
2186  * running. If no transaction with xid were running concurrently a new xid
2187  * could influence the RecentXmin et al.
2188  *
2189  * We initialize the computation to nextXid since that's guaranteed to be
2190  * a safe, albeit pessimal, value.
2191  */
2192  LWLockAcquire(XidGenLock, LW_SHARED);
2194 
2195  /*
2196  * If there's already a slot pegging the xmin horizon, we can start with
2197  * that value, it's guaranteed to be safe since it's computed by this
2198  * routine initially and has been enforced since. We can always use the
2199  * slot's general xmin horizon, but the catalog horizon is only usable
2200  * when only catalog data is going to be looked at.
2201  */
2202  if (TransactionIdIsValid(procArray->replication_slot_xmin) &&
2204  oldestSafeXid))
2205  oldestSafeXid = procArray->replication_slot_xmin;
2206 
2207  if (catalogOnly &&
2210  oldestSafeXid))
2211  oldestSafeXid = procArray->replication_slot_catalog_xmin;
2212 
2213  /*
2214  * If we're not in recovery, we walk over the procarray and collect the
2215  * lowest xid. Since we're called with ProcArrayLock held and have
2216  * acquired XidGenLock, no entries can vanish concurrently, since
2217  * PGXACT->xid is only set with XidGenLock held and only cleared with
2218  * ProcArrayLock held.
2219  *
2220  * In recovery we can't lower the safe value besides what we've computed
2221  * above, so we'll have to wait a bit longer there. We unfortunately can
2222  * *not* use KnownAssignedXidsGetOldestXmin() since the KnownAssignedXids
2223  * machinery can miss values and return an older value than is safe.
2224  */
2225  if (!recovery_in_progress)
2226  {
2227  /*
2228  * Spin over procArray collecting all min(PGXACT->xid)
2229  */
2230  for (index = 0; index < arrayP->numProcs; index++)
2231  {
2232  int pgprocno = arrayP->pgprocnos[index];
2233  PGXACT *pgxact = &allPgXact[pgprocno];
2234  TransactionId xid;
2235 
2236  /* Fetch xid just once - see GetNewTransactionId */
2237  xid = UINT32_ACCESS_ONCE(pgxact->xid);
2238 
2239  if (!TransactionIdIsNormal(xid))
2240  continue;
2241 
2242  if (TransactionIdPrecedes(xid, oldestSafeXid))
2243  oldestSafeXid = xid;
2244  }
2245  }
2246 
2247  LWLockRelease(XidGenLock);
2248 
2249  return oldestSafeXid;
2250 }
2251 
2252 /*
2253  * GetVirtualXIDsDelayingChkpt -- Get the VXIDs of transactions that are
2254  * delaying checkpoint because they have critical actions in progress.
2255  *
2256  * Constructs an array of VXIDs of transactions that are currently in commit
2257  * critical sections, as shown by having delayChkpt set in their PGXACT.
2258  *
2259  * Returns a palloc'd array that should be freed by the caller.
2260  * *nvxids is the number of valid entries.
2261  *
2262  * Note that because backends set or clear delayChkpt without holding any lock,
2263  * the result is somewhat indeterminate, but we don't really care. Even in
2264  * a multiprocessor with delayed writes to shared memory, it should be certain
2265  * that setting of delayChkpt will propagate to shared memory when the backend
2266  * takes a lock, so we cannot fail to see a virtual xact as delayChkpt if
2267  * it's already inserted its commit record. Whether it takes a little while
2268  * for clearing of delayChkpt to propagate is unimportant for correctness.
2269  */
2272 {
2273  VirtualTransactionId *vxids;
2274  ProcArrayStruct *arrayP = procArray;
2275  int count = 0;
2276  int index;
2277 
2278  /* allocate what's certainly enough result space */
2279  vxids = (VirtualTransactionId *)
2280  palloc(sizeof(VirtualTransactionId) * arrayP->maxProcs);
2281 
2282  LWLockAcquire(ProcArrayLock, LW_SHARED);
2283 
2284  for (index = 0; index < arrayP->numProcs; index++)
2285  {
2286  int pgprocno = arrayP->pgprocnos[index];
2287  PGPROC *proc = &allProcs[pgprocno];
2288  PGXACT *pgxact = &allPgXact[pgprocno];
2289 
2290  if (pgxact->delayChkpt)
2291  {
2292  VirtualTransactionId vxid;
2293 
2294  GET_VXID_FROM_PGPROC(vxid, *proc);
2295  if (VirtualTransactionIdIsValid(vxid))
2296  vxids[count++] = vxid;
2297  }
2298  }
2299 
2300  LWLockRelease(ProcArrayLock);
2301 
2302  *nvxids = count;
2303  return vxids;
2304 }
2305 
2306 /*
2307  * HaveVirtualXIDsDelayingChkpt -- Are any of the specified VXIDs delaying?
2308  *
2309  * This is used with the results of GetVirtualXIDsDelayingChkpt to see if any
2310  * of the specified VXIDs are still in critical sections of code.
2311  *
2312  * Note: this is O(N^2) in the number of vxacts that are/were delaying, but
2313  * those numbers should be small enough for it not to be a problem.
2314  */
2315 bool
2317 {
2318  bool result = false;
2319  ProcArrayStruct *arrayP = procArray;
2320  int index;
2321 
2322  LWLockAcquire(ProcArrayLock, LW_SHARED);
2323 
2324  for (index = 0; index < arrayP->numProcs; index++)
2325  {
2326  int pgprocno = arrayP->pgprocnos[index];
2327  PGPROC *proc = &allProcs[pgprocno];
2328  PGXACT *pgxact = &allPgXact[pgprocno];
2329  VirtualTransactionId vxid;
2330 
2331  GET_VXID_FROM_PGPROC(vxid, *proc);
2332 
2333  if (pgxact->delayChkpt && VirtualTransactionIdIsValid(vxid))
2334  {
2335  int i;
2336 
2337  for (i = 0; i < nvxids; i++)
2338  {
2339  if (VirtualTransactionIdEquals(vxid, vxids[i]))
2340  {
2341  result = true;
2342  break;
2343  }
2344  }
2345  if (result)
2346  break;
2347  }
2348  }
2349 
2350  LWLockRelease(ProcArrayLock);
2351 
2352  return result;
2353 }
2354 
2355 /*
2356  * BackendPidGetProc -- get a backend's PGPROC given its PID
2357  *
2358  * Returns NULL if not found. Note that it is up to the caller to be
2359  * sure that the question remains meaningful for long enough for the
2360  * answer to be used ...
2361  */
2362 PGPROC *
2364 {
2365  PGPROC *result;
2366 
2367  if (pid == 0) /* never match dummy PGPROCs */
2368  return NULL;
2369 
2370  LWLockAcquire(ProcArrayLock, LW_SHARED);
2371 
2372  result = BackendPidGetProcWithLock(pid);
2373 
2374  LWLockRelease(ProcArrayLock);
2375 
2376  return result;
2377 }
2378 
2379 /*
2380  * BackendPidGetProcWithLock -- get a backend's PGPROC given its PID
2381  *
2382  * Same as above, except caller must be holding ProcArrayLock. The found
2383  * entry, if any, can be assumed to be valid as long as the lock remains held.
2384  */
2385 PGPROC *
2387 {
2388  PGPROC *result = NULL;
2389  ProcArrayStruct *arrayP = procArray;
2390  int index;
2391 
2392  if (pid == 0) /* never match dummy PGPROCs */
2393  return NULL;
2394 
2395  for (index = 0; index < arrayP->numProcs; index++)
2396  {
2397  PGPROC *proc = &allProcs[arrayP->pgprocnos[index]];
2398 
2399  if (proc->pid == pid)
2400  {
2401  result = proc;
2402  break;
2403  }
2404  }
2405 
2406  return result;
2407 }
2408 
2409 /*
2410  * BackendXidGetPid -- get a backend's pid given its XID
2411  *
2412  * Returns 0 if not found or it's a prepared transaction. Note that
2413  * it is up to the caller to be sure that the question remains
2414  * meaningful for long enough for the answer to be used ...
2415  *
2416  * Only main transaction Ids are considered. This function is mainly
2417  * useful for determining what backend owns a lock.
2418  *
2419  * Beware that not every xact has an XID assigned. However, as long as you
2420  * only call this using an XID found on disk, you're safe.
2421  */
2422 int
2424 {
2425  int result = 0;
2426  ProcArrayStruct *arrayP = procArray;
2427  int index;
2428 
2429  if (xid == InvalidTransactionId) /* never match invalid xid */
2430  return 0;
2431 
2432  LWLockAcquire(ProcArrayLock, LW_SHARED);
2433 
2434  for (index = 0; index < arrayP->numProcs; index++)
2435  {
2436  int pgprocno = arrayP->pgprocnos[index];
2437  PGPROC *proc = &allProcs[pgprocno];
2438  PGXACT *pgxact = &allPgXact[pgprocno];
2439 
2440  if (pgxact->xid == xid)
2441  {
2442  result = proc->pid;
2443  break;
2444  }
2445  }
2446 
2447  LWLockRelease(ProcArrayLock);
2448 
2449  return result;
2450 }
2451 
2452 /*
2453  * IsBackendPid -- is a given pid a running backend
2454  *
2455  * This is not called by the backend, but is called by external modules.
2456  */
2457 bool
2459 {
2460  return (BackendPidGetProc(pid) != NULL);
2461 }
2462 
2463 
2464 /*
2465  * GetCurrentVirtualXIDs -- returns an array of currently active VXIDs.
2466  *
2467  * The array is palloc'd. The number of valid entries is returned into *nvxids.
2468  *
2469  * The arguments allow filtering the set of VXIDs returned. Our own process
2470  * is always skipped. In addition:
2471  * If limitXmin is not InvalidTransactionId, skip processes with
2472  * xmin > limitXmin.
2473  * If excludeXmin0 is true, skip processes with xmin = 0.
2474  * If allDbs is false, skip processes attached to other databases.
2475  * If excludeVacuum isn't zero, skip processes for which
2476  * (vacuumFlags & excludeVacuum) is not zero.
2477  *
2478  * Note: the purpose of the limitXmin and excludeXmin0 parameters is to
2479  * allow skipping backends whose oldest live snapshot is no older than
2480  * some snapshot we have. Since we examine the procarray with only shared
2481  * lock, there are race conditions: a backend could set its xmin just after
2482  * we look. Indeed, on multiprocessors with weak memory ordering, the
2483  * other backend could have set its xmin *before* we look. We know however
2484  * that such a backend must have held shared ProcArrayLock overlapping our
2485  * own hold of ProcArrayLock, else we would see its xmin update. Therefore,
2486  * any snapshot the other backend is taking concurrently with our scan cannot
2487  * consider any transactions as still running that we think are committed
2488  * (since backends must hold ProcArrayLock exclusive to commit).
2489  */
2491 GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
2492  bool allDbs, int excludeVacuum,
2493  int *nvxids)
2494 {
2495  VirtualTransactionId *vxids;
2496  ProcArrayStruct *arrayP = procArray;
2497  int count = 0;
2498  int index;
2499 
2500  /* allocate what's certainly enough result space */
2501  vxids = (VirtualTransactionId *)
2502  palloc(sizeof(VirtualTransactionId) * arrayP->maxProcs);
2503 
2504  LWLockAcquire(ProcArrayLock, LW_SHARED);
2505 
2506  for (index = 0; index < arrayP->numProcs; index++)
2507  {
2508  int pgprocno = arrayP->pgprocnos[index];
2509  PGPROC *proc = &allProcs[pgprocno];
2510  PGXACT *pgxact = &allPgXact[pgprocno];
2511 
2512  if (proc == MyProc)
2513  continue;
2514 
2515  if (excludeVacuum & pgxact->vacuumFlags)
2516  continue;
2517 
2518  if (allDbs || proc->databaseId == MyDatabaseId)
2519  {
2520  /* Fetch xmin just once - might change on us */
2521  TransactionId pxmin = UINT32_ACCESS_ONCE(pgxact->xmin);
2522 
2523  if (excludeXmin0 && !TransactionIdIsValid(pxmin))
2524  continue;
2525 
2526  /*
2527  * InvalidTransactionId precedes all other XIDs, so a proc that
2528  * hasn't set xmin yet will not be rejected by this test.
2529  */
2530  if (!TransactionIdIsValid(limitXmin) ||
2531  TransactionIdPrecedesOrEquals(pxmin, limitXmin))
2532  {
2533  VirtualTransactionId vxid;
2534 
2535  GET_VXID_FROM_PGPROC(vxid, *proc);
2536  if (VirtualTransactionIdIsValid(vxid))
2537  vxids[count++] = vxid;
2538  }
2539  }
2540  }
2541 
2542  LWLockRelease(ProcArrayLock);
2543 
2544  *nvxids = count;
2545  return vxids;
2546 }
2547 
2548 /*
2549  * GetConflictingVirtualXIDs -- returns an array of currently active VXIDs.
2550  *
2551  * Usage is limited to conflict resolution during recovery on standby servers.
2552  * limitXmin is supplied as either latestRemovedXid, or InvalidTransactionId
2553  * in cases where we cannot accurately determine a value for latestRemovedXid.
2554  *
2555  * If limitXmin is InvalidTransactionId then we want to kill everybody,
2556  * so we're not worried if they have a snapshot or not, nor does it really
2557  * matter what type of lock we hold.
2558  *
2559  * All callers that are checking xmins always now supply a valid and useful
2560  * value for limitXmin. The limitXmin is always lower than the lowest
2561  * numbered KnownAssignedXid that is not already a FATAL error. This is
2562  * because we only care about cleanup records that are cleaning up tuple
2563  * versions from committed transactions. In that case they will only occur
2564  * at the point where the record is less than the lowest running xid. That
2565  * allows us to say that if any backend takes a snapshot concurrently with
2566  * us then the conflict assessment made here would never include the snapshot
2567  * that is being derived. So we take LW_SHARED on the ProcArray and allow
2568  * concurrent snapshots when limitXmin is valid. We might think about adding
2569  * Assert(limitXmin < lowest(KnownAssignedXids))
2570  * but that would not be true in the case of FATAL errors lagging in array,
2571  * but we already know those are bogus anyway, so we skip that test.
2572  *
2573  * If dbOid is valid we skip backends attached to other databases.
2574  *
2575  * Be careful to *not* pfree the result from this function. We reuse
2576  * this array sufficiently often that we use malloc for the result.
2577  */
2580 {
2581  static VirtualTransactionId *vxids;
2582  ProcArrayStruct *arrayP = procArray;
2583  int count = 0;
2584  int index;
2585 
2586  /*
2587  * If first time through, get workspace to remember main XIDs in. We
2588  * malloc it permanently to avoid repeated palloc/pfree overhead. Allow
2589  * result space, remembering room for a terminator.
2590  */
2591  if (vxids == NULL)
2592  {
2593  vxids = (VirtualTransactionId *)
2594  malloc(sizeof(VirtualTransactionId) * (arrayP->maxProcs + 1));
2595  if (vxids == NULL)
2596  ereport(ERROR,
2597  (errcode(ERRCODE_OUT_OF_MEMORY),
2598  errmsg("out of memory")));
2599  }
2600 
2601  LWLockAcquire(ProcArrayLock, LW_SHARED);
2602 
2603  for (index = 0; index < arrayP->numProcs; index++)
2604  {
2605  int pgprocno = arrayP->pgprocnos[index];
2606  PGPROC *proc = &allProcs[pgprocno];
2607  PGXACT *pgxact = &allPgXact[pgprocno];
2608 
2609  /* Exclude prepared transactions */
2610  if (proc->pid == 0)
2611  continue;
2612 
2613  if (!OidIsValid(dbOid) ||
2614  proc->databaseId == dbOid)
2615  {
2616  /* Fetch xmin just once - can't change on us, but good coding */
2617  TransactionId pxmin = UINT32_ACCESS_ONCE(pgxact->xmin);
2618 
2619  /*
2620  * We ignore an invalid pxmin because this means that backend has
2621  * no snapshot currently. We hold a Share lock to avoid contention
2622  * with users taking snapshots. That is not a problem because the
2623  * current xmin is always at least one higher than the latest
2624  * removed xid, so any new snapshot would never conflict with the
2625  * test here.
2626  */
2627  if (!TransactionIdIsValid(limitXmin) ||
2628  (TransactionIdIsValid(pxmin) && !TransactionIdFollows(pxmin, limitXmin)))
2629  {
2630  VirtualTransactionId vxid;
2631 
2632  GET_VXID_FROM_PGPROC(vxid, *proc);
2633  if (VirtualTransactionIdIsValid(vxid))
2634  vxids[count++] = vxid;
2635  }
2636  }
2637  }
2638 
2639  LWLockRelease(ProcArrayLock);
2640 
2641  /* add the terminator */
2642  vxids[count].backendId = InvalidBackendId;
2644 
2645  return vxids;
2646 }
2647 
2648 /*
2649  * CancelVirtualTransaction - used in recovery conflict processing
2650  *
2651  * Returns pid of the process signaled, or 0 if not found.
2652  */
2653 pid_t
2655 {
2656  ProcArrayStruct *arrayP = procArray;
2657  int index;
2658  pid_t pid = 0;
2659 
2660  LWLockAcquire(ProcArrayLock, LW_SHARED);
2661 
2662  for (index = 0; index < arrayP->numProcs; index++)
2663  {
2664  int pgprocno = arrayP->pgprocnos[index];
2665  PGPROC *proc = &allProcs[pgprocno];
2666  VirtualTransactionId procvxid;
2667 
2668  GET_VXID_FROM_PGPROC(procvxid, *proc);
2669 
2670  if (procvxid.backendId == vxid.backendId &&
2671  procvxid.localTransactionId == vxid.localTransactionId)
2672  {
2673  proc->recoveryConflictPending = true;
2674  pid = proc->pid;
2675  if (pid != 0)
2676  {
2677  /*
2678  * Kill the pid if it's still here. If not, that's what we
2679  * wanted so ignore any errors.
2680  */
2681  (void) SendProcSignal(pid, sigmode, vxid.backendId);
2682  }
2683  break;
2684  }
2685  }
2686 
2687  LWLockRelease(ProcArrayLock);
2688 
2689  return pid;
2690 }
2691 
2692 /*
2693  * MinimumActiveBackends --- count backends (other than myself) that are
2694  * in active transactions. Return true if the count exceeds the
2695  * minimum threshold passed. This is used as a heuristic to decide if
2696  * a pre-XLOG-flush delay is worthwhile during commit.
2697  *
2698  * Do not count backends that are blocked waiting for locks, since they are
2699  * not going to get to run until someone else commits.
2700  */
2701 bool
2703 {
2704  ProcArrayStruct *arrayP = procArray;
2705  int count = 0;
2706  int index;
2707 
2708  /* Quick short-circuit if no minimum is specified */
2709  if (min == 0)
2710  return true;
2711 
2712  /*
2713  * Note: for speed, we don't acquire ProcArrayLock. This is a little bit
2714  * bogus, but since we are only testing fields for zero or nonzero, it
2715  * should be OK. The result is only used for heuristic purposes anyway...
2716  */
2717  for (index = 0; index < arrayP->numProcs; index++)
2718  {
2719  int pgprocno = arrayP->pgprocnos[index];
2720  PGPROC *proc = &allProcs[pgprocno];
2721  PGXACT *pgxact = &allPgXact[pgprocno];
2722 
2723  /*
2724  * Since we're not holding a lock, need to be prepared to deal with
2725  * garbage, as someone could have incremented numProcs but not yet
2726  * filled the structure.
2727  *
2728  * If someone just decremented numProcs, 'proc' could also point to a
2729  * PGPROC entry that's no longer in the array. It still points to a
2730  * PGPROC struct, though, because freed PGPROC entries just go to the
2731  * free list and are recycled. Its contents are nonsense in that case,
2732  * but that's acceptable for this function.
2733  */
2734  if (pgprocno == -1)
2735  continue; /* do not count deleted entries */
2736  if (proc == MyProc)
2737  continue; /* do not count myself */
2738  if (pgxact->xid == InvalidTransactionId)
2739  continue; /* do not count if no XID assigned */
2740  if (proc->pid == 0)
2741  continue; /* do not count prepared xacts */
2742  if (proc->waitLock != NULL)
2743  continue; /* do not count if blocked on a lock */
2744  count++;
2745  if (count >= min)
2746  break;
2747  }
2748 
2749  return count >= min;
2750 }
2751 
2752 /*
2753  * CountDBBackends --- count backends that are using specified database
2754  */
2755 int
2757 {
2758  ProcArrayStruct *arrayP = procArray;
2759  int count = 0;
2760  int index;
2761 
2762  LWLockAcquire(ProcArrayLock, LW_SHARED);
2763 
2764  for (index = 0; index < arrayP->numProcs; index++)
2765  {
2766  int pgprocno = arrayP->pgprocnos[index];
2767  PGPROC *proc = &allProcs[pgprocno];
2768 
2769  if (proc->pid == 0)
2770  continue; /* do not count prepared xacts */
2771  if (!OidIsValid(databaseid) ||
2772  proc->databaseId == databaseid)
2773  count++;
2774  }
2775 
2776  LWLockRelease(ProcArrayLock);
2777 
2778  return count;
2779 }
2780 
2781 /*
2782  * CountDBConnections --- counts database backends ignoring any background
2783  * worker processes
2784  */
2785 int
2787 {
2788  ProcArrayStruct *arrayP = procArray;
2789  int count = 0;
2790  int index;
2791 
2792  LWLockAcquire(ProcArrayLock, LW_SHARED);
2793 
2794  for (index = 0; index < arrayP->numProcs; index++)
2795  {
2796  int pgprocno = arrayP->pgprocnos[index];
2797  PGPROC *proc = &allProcs[pgprocno];
2798 
2799  if (proc->pid == 0)
2800  continue; /* do not count prepared xacts */
2801  if (proc->isBackgroundWorker)
2802  continue; /* do not count background workers */
2803  if (!OidIsValid(databaseid) ||
2804  proc->databaseId == databaseid)
2805  count++;
2806  }
2807 
2808  LWLockRelease(ProcArrayLock);
2809 
2810  return count;
2811 }
2812 
2813 /*
2814  * CancelDBBackends --- cancel backends that are using specified database
2815  */
2816 void
2817 CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
2818 {
2819  ProcArrayStruct *arrayP = procArray;
2820  int index;
2821  pid_t pid = 0;
2822 
2823  /* tell all backends to die */
2824  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
2825 
2826  for (index = 0; index < arrayP->numProcs; index++)
2827  {
2828  int pgprocno = arrayP->pgprocnos[index];
2829  PGPROC *proc = &allProcs[pgprocno];
2830 
2831  if (databaseid == InvalidOid || proc->databaseId == databaseid)
2832  {
2833  VirtualTransactionId procvxid;
2834 
2835  GET_VXID_FROM_PGPROC(procvxid, *proc);
2836 
2837  proc->recoveryConflictPending = conflictPending;
2838  pid = proc->pid;
2839  if (pid != 0)
2840  {
2841  /*
2842  * Kill the pid if it's still here. If not, that's what we
2843  * wanted so ignore any errors.
2844  */
2845  (void) SendProcSignal(pid, sigmode, procvxid.backendId);
2846  }
2847  }
2848  }
2849 
2850  LWLockRelease(ProcArrayLock);
2851 }
2852 
2853 /*
2854  * CountUserBackends --- count backends that are used by specified user
2855  */
2856 int
2858 {
2859  ProcArrayStruct *arrayP = procArray;
2860  int count = 0;
2861  int index;
2862 
2863  LWLockAcquire(ProcArrayLock, LW_SHARED);
2864 
2865  for (index = 0; index < arrayP->numProcs; index++)
2866  {
2867  int pgprocno = arrayP->pgprocnos[index];
2868  PGPROC *proc = &allProcs[pgprocno];
2869 
2870  if (proc->pid == 0)
2871  continue; /* do not count prepared xacts */
2872  if (proc->isBackgroundWorker)
2873  continue; /* do not count background workers */
2874  if (proc->roleId == roleid)
2875  count++;
2876  }
2877 
2878  LWLockRelease(ProcArrayLock);
2879 
2880  return count;
2881 }
2882 
2883 /*
2884  * CountOtherDBBackends -- check for other backends running in the given DB
2885  *
2886  * If there are other backends in the DB, we will wait a maximum of 5 seconds
2887  * for them to exit. Autovacuum backends are encouraged to exit early by
2888  * sending them SIGTERM, but normal user backends are just waited for.
2889  *
2890  * The current backend is always ignored; it is caller's responsibility to
2891  * check whether the current backend uses the given DB, if it's important.
2892  *
2893  * Returns true if there are (still) other backends in the DB, false if not.
2894  * Also, *nbackends and *nprepared are set to the number of other backends
2895  * and prepared transactions in the DB, respectively.
2896  *
2897  * This function is used to interlock DROP DATABASE and related commands
2898  * against there being any active backends in the target DB --- dropping the
2899  * DB while active backends remain would be a Bad Thing. Note that we cannot
2900  * detect here the possibility of a newly-started backend that is trying to
2901  * connect to the doomed database, so additional interlocking is needed during
2902  * backend startup. The caller should normally hold an exclusive lock on the
2903  * target DB before calling this, which is one reason we mustn't wait
2904  * indefinitely.
2905  */
2906 bool
2907 CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
2908 {
2909  ProcArrayStruct *arrayP = procArray;
2910 
2911 #define MAXAUTOVACPIDS 10 /* max autovacs to SIGTERM per iteration */
2912  int autovac_pids[MAXAUTOVACPIDS];
2913  int tries;
2914 
2915  /* 50 tries with 100ms sleep between tries makes 5 sec total wait */
2916  for (tries = 0; tries < 50; tries++)
2917  {
2918  int nautovacs = 0;
2919  bool found = false;
2920  int index;
2921 
2923 
2924  *nbackends = *nprepared = 0;
2925 
2926  LWLockAcquire(ProcArrayLock, LW_SHARED);
2927 
2928  for (index = 0; index < arrayP->numProcs; index++)
2929  {
2930  int pgprocno = arrayP->pgprocnos[index];
2931  PGPROC *proc = &allProcs[pgprocno];
2932  PGXACT *pgxact = &allPgXact[pgprocno];
2933 
2934  if (proc->databaseId != databaseId)
2935  continue;
2936  if (proc == MyProc)
2937  continue;
2938 
2939  found = true;
2940 
2941  if (proc->pid == 0)
2942  (*nprepared)++;
2943  else
2944  {
2945  (*nbackends)++;
2946  if ((pgxact->vacuumFlags & PROC_IS_AUTOVACUUM) &&
2947  nautovacs < MAXAUTOVACPIDS)
2948  autovac_pids[nautovacs++] = proc->pid;
2949  }
2950  }
2951 
2952  LWLockRelease(ProcArrayLock);
2953 
2954  if (!found)
2955  return false; /* no conflicting backends, so done */
2956 
2957  /*
2958  * Send SIGTERM to any conflicting autovacuums before sleeping. We
2959  * postpone this step until after the loop because we don't want to
2960  * hold ProcArrayLock while issuing kill(). We have no idea what might
2961  * block kill() inside the kernel...
2962  */
2963  for (index = 0; index < nautovacs; index++)
2964  (void) kill(autovac_pids[index], SIGTERM); /* ignore any error */
2965 
2966  /* sleep, then try again */
2967  pg_usleep(100 * 1000L); /* 100ms */
2968  }
2969 
2970  return true; /* timed out, still conflicts */
2971 }
2972 
2973 /*
2974  * ProcArraySetReplicationSlotXmin
2975  *
2976  * Install limits to future computations of the xmin horizon to prevent vacuum
2977  * and HOT pruning from removing affected rows still needed by clients with
2978  * replication slots.
2979  */
2980 void
2982  bool already_locked)
2983 {
2984  Assert(!already_locked || LWLockHeldByMe(ProcArrayLock));
2985 
2986  if (!already_locked)
2987  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
2988 
2989  procArray->replication_slot_xmin = xmin;
2990  procArray->replication_slot_catalog_xmin = catalog_xmin;
2991 
2992  if (!already_locked)
2993  LWLockRelease(ProcArrayLock);
2994 }
2995 
2996 /*
2997  * ProcArrayGetReplicationSlotXmin
2998  *
2999  * Return the current slot xmin limits. That's useful to be able to remove
3000  * data that's older than those limits.
3001  */
3002 void
3004  TransactionId *catalog_xmin)
3005 {
3006  LWLockAcquire(ProcArrayLock, LW_SHARED);
3007 
3008  if (xmin != NULL)
3009  *xmin = procArray->replication_slot_xmin;
3010 
3011  if (catalog_xmin != NULL)
3012  *catalog_xmin = procArray->replication_slot_catalog_xmin;
3013 
3014  LWLockRelease(ProcArrayLock);
3015 }
3016 
3017 
3018 #define XidCacheRemove(i) \
3019  do { \
3020  MyProc->subxids.xids[i] = MyProc->subxids.xids[MyPgXact->nxids - 1]; \
3021  pg_write_barrier(); \
3022  MyPgXact->nxids--; \
3023  } while (0)
3024 
3025 /*
3026  * XidCacheRemoveRunningXids
3027  *
3028  * Remove a bunch of TransactionIds from the list of known-running
3029  * subtransactions for my backend. Both the specified xid and those in
3030  * the xids[] array (of length nxids) are removed from the subxids cache.
3031  * latestXid must be the latest XID among the group.
3032  */
3033 void
3035  int nxids, const TransactionId *xids,
3036  TransactionId latestXid)
3037 {
3038  int i,
3039  j;
3040 
3042 
3043  /*
3044  * We must hold ProcArrayLock exclusively in order to remove transactions
3045  * from the PGPROC array. (See src/backend/access/transam/README.) It's
3046  * possible this could be relaxed since we know this routine is only used
3047  * to abort subtransactions, but pending closer analysis we'd best be
3048  * conservative.
3049  *
3050  * Note that we do not have to be careful about memory ordering of our own
3051  * reads wrt. GetNewTransactionId() here - only this process can modify
3052  * relevant fields of MyProc/MyPgXact. But we do have to be careful about
3053  * our own writes being well ordered.
3054  */
3055  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3056 
3057  /*
3058  * Under normal circumstances xid and xids[] will be in increasing order,
3059  * as will be the entries in subxids. Scan backwards to avoid O(N^2)
3060  * behavior when removing a lot of xids.
3061  */
3062  for (i = nxids - 1; i >= 0; i--)
3063  {
3064  TransactionId anxid = xids[i];
3065 
3066  for (j = MyPgXact->nxids - 1; j >= 0; j--)
3067  {
3068  if (TransactionIdEquals(MyProc->subxids.xids[j], anxid))
3069  {
3070  XidCacheRemove(j);
3071  break;
3072  }
3073  }
3074 
3075  /*
3076  * Ordinarily we should have found it, unless the cache has
3077  * overflowed. However it's also possible for this routine to be
3078  * invoked multiple times for the same subtransaction, in case of an
3079  * error during AbortSubTransaction. So instead of Assert, emit a
3080  * debug warning.
3081  */
3082  if (j < 0 && !MyPgXact->overflowed)
3083  elog(WARNING, "did not find subXID %u in MyProc", anxid);
3084  }
3085 
3086  for (j = MyPgXact->nxids - 1; j >= 0; j--)
3087  {
3088  if (TransactionIdEquals(MyProc->subxids.xids[j], xid))
3089  {
3090  XidCacheRemove(j);
3091  break;
3092  }
3093  }
3094  /* Ordinarily we should have found it, unless the cache has overflowed */
3095  if (j < 0 && !MyPgXact->overflowed)
3096  elog(WARNING, "did not find subXID %u in MyProc", xid);
3097 
3098  /* Also advance global latestCompletedXid while holding the lock */
3100  latestXid))
3102 
3103  LWLockRelease(ProcArrayLock);
3104 }
3105 
3106 #ifdef XIDCACHE_DEBUG
3107 
3108 /*
3109  * Print stats about effectiveness of XID cache
3110  */
3111 static void
3112 DisplayXidCache(void)
3113 {
3114  fprintf(stderr,
3115  "XidCache: xmin: %ld, known: %ld, myxact: %ld, latest: %ld, mainxid: %ld, childxid: %ld, knownassigned: %ld, nooflo: %ld, slow: %ld\n",
3116  xc_by_recent_xmin,
3117  xc_by_known_xact,
3118  xc_by_my_xact,
3119  xc_by_latest_xid,
3120  xc_by_main_xid,
3121  xc_by_child_xid,
3122  xc_by_known_assigned,
3123  xc_no_overflow,
3124  xc_slow_answer);
3125 }
3126 #endif /* XIDCACHE_DEBUG */
3127 
3128 
3129 /* ----------------------------------------------
3130  * KnownAssignedTransactionIds sub-module
3131  * ----------------------------------------------
3132  */
3133 
3134 /*
3135  * In Hot Standby mode, we maintain a list of transactions that are (or were)
3136  * running in the master at the current point in WAL. These XIDs must be
3137  * treated as running by standby transactions, even though they are not in
3138  * the standby server's PGXACT array.
3139  *
3140  * We record all XIDs that we know have been assigned. That includes all the
3141  * XIDs seen in WAL records, plus all unobserved XIDs that we can deduce have
3142  * been assigned. We can deduce the existence of unobserved XIDs because we
3143  * know XIDs are assigned in sequence, with no gaps. The KnownAssignedXids
3144  * list expands as new XIDs are observed or inferred, and contracts when
3145  * transaction completion records arrive.
3146  *
3147  * During hot standby we do not fret too much about the distinction between
3148  * top-level XIDs and subtransaction XIDs. We store both together in the
3149  * KnownAssignedXids list. In backends, this is copied into snapshots in
3150  * GetSnapshotData(), taking advantage of the fact that XidInMVCCSnapshot()
3151  * doesn't care about the distinction either. Subtransaction XIDs are
3152  * effectively treated as top-level XIDs and in the typical case pg_subtrans
3153  * links are *not* maintained (which does not affect visibility).
3154  *
3155  * We have room in KnownAssignedXids and in snapshots to hold maxProcs *
3156  * (1 + PGPROC_MAX_CACHED_SUBXIDS) XIDs, so every master transaction must
3157  * report its subtransaction XIDs in a WAL XLOG_XACT_ASSIGNMENT record at
3158  * least every PGPROC_MAX_CACHED_SUBXIDS. When we receive one of these
3159  * records, we mark the subXIDs as children of the top XID in pg_subtrans,
3160  * and then remove them from KnownAssignedXids. This prevents overflow of
3161  * KnownAssignedXids and snapshots, at the cost that status checks for these
3162  * subXIDs will take a slower path through TransactionIdIsInProgress().
3163  * This means that KnownAssignedXids is not necessarily complete for subXIDs,
3164  * though it should be complete for top-level XIDs; this is the same situation
3165  * that holds with respect to the PGPROC entries in normal running.
3166  *
3167  * When we throw away subXIDs from KnownAssignedXids, we need to keep track of
3168  * that, similarly to tracking overflow of a PGPROC's subxids array. We do
3169  * that by remembering the lastOverflowedXid, ie the last thrown-away subXID.
3170  * As long as that is within the range of interesting XIDs, we have to assume
3171  * that subXIDs are missing from snapshots. (Note that subXID overflow occurs
3172  * on primary when 65th subXID arrives, whereas on standby it occurs when 64th
3173  * subXID arrives - that is not an error.)
3174  *
3175  * Should a backend on primary somehow disappear before it can write an abort
3176  * record, then we just leave those XIDs in KnownAssignedXids. They actually
3177  * aborted but we think they were running; the distinction is irrelevant
3178  * because either way any changes done by the transaction are not visible to
3179  * backends in the standby. We prune KnownAssignedXids when
3180  * XLOG_RUNNING_XACTS arrives, to forestall possible overflow of the
3181  * array due to such dead XIDs.
3182  */
3183 
3184 /*
3185  * RecordKnownAssignedTransactionIds
3186  * Record the given XID in KnownAssignedXids, as well as any preceding
3187  * unobserved XIDs.
3188  *
3189  * RecordKnownAssignedTransactionIds() should be run for *every* WAL record
3190  * associated with a transaction. Must be called for each record after we
3191  * have executed StartupCLOG() et al, since we must ExtendCLOG() etc..
3192  *
3193  * Called during recovery in analogy with and in place of GetNewTransactionId()
3194  */
3195 void
3197 {
3201 
3202  elog(trace_recovery(DEBUG4), "record known xact %u latestObservedXid %u",
3203  xid, latestObservedXid);
3204 
3205  /*
3206  * When a newly observed xid arrives, it is frequently the case that it is
3207  * *not* the next xid in sequence. When this occurs, we must treat the
3208  * intervening xids as running also.
3209  */
3211  {
3212  TransactionId next_expected_xid;
3213 
3214  /*
3215  * Extend subtrans like we do in GetNewTransactionId() during normal
3216  * operation using individual extend steps. Note that we do not need
3217  * to extend clog since its extensions are WAL logged.
3218  *
3219  * This part has to be done regardless of standbyState since we
3220  * immediately start assigning subtransactions to their toplevel
3221  * transactions.
3222  */
3223  next_expected_xid = latestObservedXid;
3224  while (TransactionIdPrecedes(next_expected_xid, xid))
3225  {
3226  TransactionIdAdvance(next_expected_xid);
3227  ExtendSUBTRANS(next_expected_xid);
3228  }
3229  Assert(next_expected_xid == xid);
3230 
3231  /*
3232  * If the KnownAssignedXids machinery isn't up yet, there's nothing
3233  * more to do since we don't track assigned xids yet.
3234  */
3236  {
3237  latestObservedXid = xid;
3238  return;
3239  }
3240 
3241  /*
3242  * Add (latestObservedXid, xid] onto the KnownAssignedXids array.
3243  */
3244  next_expected_xid = latestObservedXid;
3245  TransactionIdAdvance(next_expected_xid);
3246  KnownAssignedXidsAdd(next_expected_xid, xid, false);
3247 
3248  /*
3249  * Now we can advance latestObservedXid
3250  */
3251  latestObservedXid = xid;
3252 
3253  /* ShmemVariableCache->nextFullXid must be beyond any observed xid */
3255  next_expected_xid = latestObservedXid;
3256  TransactionIdAdvance(next_expected_xid);
3257  }
3258 }
3259 
3260 /*
3261  * ExpireTreeKnownAssignedTransactionIds
3262  * Remove the given XIDs from KnownAssignedXids.
3263  *
3264  * Called during recovery in analogy with and in place of ProcArrayEndTransaction()
3265  */
3266 void
3268  TransactionId *subxids, TransactionId max_xid)
3269 {
3271 
3272  /*
3273  * Uses same locking as transaction commit
3274  */
3275  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3276 
3277  KnownAssignedXidsRemoveTree(xid, nsubxids, subxids);
3278 
3279  /* As in ProcArrayEndTransaction, advance latestCompletedXid */
3281  max_xid))
3283 
3284  LWLockRelease(ProcArrayLock);
3285 }
3286 
3287 /*
3288  * ExpireAllKnownAssignedTransactionIds
3289  * Remove all entries in KnownAssignedXids
3290  */
3291 void
3293 {
3294  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3296  LWLockRelease(ProcArrayLock);
3297 }
3298 
3299 /*
3300  * ExpireOldKnownAssignedTransactionIds
3301  * Remove KnownAssignedXids entries preceding the given XID
3302  */
3303 void
3305 {
3306  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3308  LWLockRelease(ProcArrayLock);
3309 }
3310 
3311 
3312 /*
3313  * Private module functions to manipulate KnownAssignedXids
3314  *
3315  * There are 5 main uses of the KnownAssignedXids data structure:
3316  *
3317  * * backends taking snapshots - all valid XIDs need to be copied out
3318  * * backends seeking to determine presence of a specific XID
3319  * * startup process adding new known-assigned XIDs
3320  * * startup process removing specific XIDs as transactions end
3321  * * startup process pruning array when special WAL records arrive
3322  *
3323  * This data structure is known to be a hot spot during Hot Standby, so we
3324  * go to some lengths to make these operations as efficient and as concurrent
3325  * as possible.
3326  *
3327  * The XIDs are stored in an array in sorted order --- TransactionIdPrecedes
3328  * order, to be exact --- to allow binary search for specific XIDs. Note:
3329  * in general TransactionIdPrecedes would not provide a total order, but
3330  * we know that the entries present at any instant should not extend across
3331  * a large enough fraction of XID space to wrap around (the master would
3332  * shut down for fear of XID wrap long before that happens). So it's OK to
3333  * use TransactionIdPrecedes as a binary-search comparator.
3334  *
3335  * It's cheap to maintain the sortedness during insertions, since new known
3336  * XIDs are always reported in XID order; we just append them at the right.
3337  *
3338  * To keep individual deletions cheap, we need to allow gaps in the array.
3339  * This is implemented by marking array elements as valid or invalid using
3340  * the parallel boolean array KnownAssignedXidsValid[]. A deletion is done
3341  * by setting KnownAssignedXidsValid[i] to false, *without* clearing the
3342  * XID entry itself. This preserves the property that the XID entries are
3343  * sorted, so we can do binary searches easily. Periodically we compress
3344  * out the unused entries; that's much cheaper than having to compress the
3345  * array immediately on every deletion.
3346  *
3347  * The actually valid items in KnownAssignedXids[] and KnownAssignedXidsValid[]
3348  * are those with indexes tail <= i < head; items outside this subscript range
3349  * have unspecified contents. When head reaches the end of the array, we
3350  * force compression of unused entries rather than wrapping around, since
3351  * allowing wraparound would greatly complicate the search logic. We maintain
3352  * an explicit tail pointer so that pruning of old XIDs can be done without
3353  * immediately moving the array contents. In most cases only a small fraction
3354  * of the array contains valid entries at any instant.
3355  *
3356  * Although only the startup process can ever change the KnownAssignedXids
3357  * data structure, we still need interlocking so that standby backends will
3358  * not observe invalid intermediate states. The convention is that backends
3359  * must hold shared ProcArrayLock to examine the array. To remove XIDs from
3360  * the array, the startup process must hold ProcArrayLock exclusively, for
3361  * the usual transactional reasons (compare commit/abort of a transaction
3362  * during normal running). Compressing unused entries out of the array
3363  * likewise requires exclusive lock. To add XIDs to the array, we just insert
3364  * them into slots to the right of the head pointer and then advance the head
3365  * pointer. This wouldn't require any lock at all, except that on machines
3366  * with weak memory ordering we need to be careful that other processors
3367  * see the array element changes before they see the head pointer change.
3368  * We handle this by using a spinlock to protect reads and writes of the
3369  * head/tail pointers. (We could dispense with the spinlock if we were to
3370  * create suitable memory access barrier primitives and use those instead.)
3371  * The spinlock must be taken to read or write the head/tail pointers unless
3372  * the caller holds ProcArrayLock exclusively.
3373  *
3374  * Algorithmic analysis:
3375  *
3376  * If we have a maximum of M slots, with N XIDs currently spread across
3377  * S elements then we have N <= S <= M always.
3378  *
3379  * * Adding a new XID is O(1) and needs little locking (unless compression
3380  * must happen)
3381  * * Compressing the array is O(S) and requires exclusive lock
3382  * * Removing an XID is O(logS) and requires exclusive lock
3383  * * Taking a snapshot is O(S) and requires shared lock
3384  * * Checking for an XID is O(logS) and requires shared lock
3385  *
3386  * In comparison, using a hash table for KnownAssignedXids would mean that
3387  * taking snapshots would be O(M). If we can maintain S << M then the
3388  * sorted array technique will deliver significantly faster snapshots.
3389  * If we try to keep S too small then we will spend too much time compressing,
3390  * so there is an optimal point for any workload mix. We use a heuristic to
3391  * decide when to compress the array, though trimming also helps reduce
3392  * frequency of compressing. The heuristic requires us to track the number of
3393  * currently valid XIDs in the array.
3394  */
3395 
3396 
3397 /*
3398  * Compress KnownAssignedXids by shifting valid data down to the start of the
3399  * array, removing any gaps.
3400  *
3401  * A compression step is forced if "force" is true, otherwise we do it
3402  * only if a heuristic indicates it's a good time to do it.
3403  *
3404  * Caller must hold ProcArrayLock in exclusive mode.
3405  */
3406 static void
3408 {
3409  ProcArrayStruct *pArray = procArray;
3410  int head,
3411  tail;
3412  int compress_index;
3413  int i;
3414 
3415  /* no spinlock required since we hold ProcArrayLock exclusively */
3416  head = pArray->headKnownAssignedXids;
3417  tail = pArray->tailKnownAssignedXids;
3418 
3419  if (!force)
3420  {
3421  /*
3422  * If we can choose how much to compress, use a heuristic to avoid
3423  * compressing too often or not often enough.
3424  *
3425  * Heuristic is if we have a large enough current spread and less than
3426  * 50% of the elements are currently in use, then compress. This
3427  * should ensure we compress fairly infrequently. We could compress
3428  * less often though the virtual array would spread out more and
3429  * snapshots would become more expensive.
3430  */
3431  int nelements = head - tail;
3432 
3433  if (nelements < 4 * PROCARRAY_MAXPROCS ||
3434  nelements < 2 * pArray->numKnownAssignedXids)
3435  return;
3436  }
3437 
3438  /*
3439  * We compress the array by reading the valid values from tail to head,
3440  * re-aligning data to 0th element.
3441  */
3442  compress_index = 0;
3443  for (i = tail; i < head; i++)
3444  {
3445  if (KnownAssignedXidsValid[i])
3446  {
3447  KnownAssignedXids[compress_index] = KnownAssignedXids[i];
3448  KnownAssignedXidsValid[compress_index] = true;
3449  compress_index++;
3450  }
3451  }
3452 
3453  pArray->tailKnownAssignedXids = 0;
3454  pArray->headKnownAssignedXids = compress_index;
3455 }
3456 
3457 /*
3458  * Add xids into KnownAssignedXids at the head of the array.
3459  *
3460  * xids from from_xid to to_xid, inclusive, are added to the array.
3461  *
3462  * If exclusive_lock is true then caller already holds ProcArrayLock in
3463  * exclusive mode, so we need no extra locking here. Else caller holds no
3464  * lock, so we need to be sure we maintain sufficient interlocks against
3465  * concurrent readers. (Only the startup process ever calls this, so no need
3466  * to worry about concurrent writers.)
3467  */
3468 static void
3470  bool exclusive_lock)
3471 {
3472  ProcArrayStruct *pArray = procArray;
3473  TransactionId next_xid;
3474  int head,
3475  tail;
3476  int nxids;
3477  int i;
3478 
3479  Assert(TransactionIdPrecedesOrEquals(from_xid, to_xid));
3480 
3481  /*
3482  * Calculate how many array slots we'll need. Normally this is cheap; in
3483  * the unusual case where the XIDs cross the wrap point, we do it the hard
3484  * way.
3485  */
3486  if (to_xid >= from_xid)
3487  nxids = to_xid - from_xid + 1;
3488  else
3489  {
3490  nxids = 1;
3491  next_xid = from_xid;
3492  while (TransactionIdPrecedes(next_xid, to_xid))
3493  {
3494  nxids++;
3495  TransactionIdAdvance(next_xid);
3496  }
3497  }
3498 
3499  /*
3500  * Since only the startup process modifies the head/tail pointers, we
3501  * don't need a lock to read them here.
3502  */
3503  head = pArray->headKnownAssignedXids;
3504  tail = pArray->tailKnownAssignedXids;
3505 
3506  Assert(head >= 0 && head <= pArray->maxKnownAssignedXids);
3507  Assert(tail >= 0 && tail < pArray->maxKnownAssignedXids);
3508 
3509  /*
3510  * Verify that insertions occur in TransactionId sequence. Note that even
3511  * if the last existing element is marked invalid, it must still have a
3512  * correctly sequenced XID value.
3513  */
3514  if (head > tail &&
3515  TransactionIdFollowsOrEquals(KnownAssignedXids[head - 1], from_xid))
3516  {
3518  elog(ERROR, "out-of-order XID insertion in KnownAssignedXids");
3519  }
3520 
3521  /*
3522  * If our xids won't fit in the remaining space, compress out free space
3523  */
3524  if (head + nxids > pArray->maxKnownAssignedXids)
3525  {
3526  /* must hold lock to compress */
3527  if (!exclusive_lock)
3528  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3529 
3531 
3532  head = pArray->headKnownAssignedXids;
3533  /* note: we no longer care about the tail pointer */
3534 
3535  if (!exclusive_lock)
3536  LWLockRelease(ProcArrayLock);
3537 
3538  /*
3539  * If it still won't fit then we're out of memory
3540  */
3541  if (head + nxids > pArray->maxKnownAssignedXids)
3542  elog(ERROR, "too many KnownAssignedXids");
3543  }
3544 
3545  /* Now we can insert the xids into the space starting at head */
3546  next_xid = from_xid;
3547  for (i = 0; i < nxids; i++)
3548  {
3549  KnownAssignedXids[head] = next_xid;
3550  KnownAssignedXidsValid[head] = true;
3551  TransactionIdAdvance(next_xid);
3552  head++;
3553  }
3554 
3555  /* Adjust count of number of valid entries */
3556  pArray->numKnownAssignedXids += nxids;
3557 
3558  /*
3559  * Now update the head pointer. We use a spinlock to protect this
3560  * pointer, not because the update is likely to be non-atomic, but to
3561  * ensure that other processors see the above array updates before they
3562  * see the head pointer change.
3563  *
3564  * If we're holding ProcArrayLock exclusively, there's no need to take the
3565  * spinlock.
3566  */
3567  if (exclusive_lock)
3568  pArray->headKnownAssignedXids = head;
3569  else
3570  {
3572  pArray->headKnownAssignedXids = head;
3574  }
3575 }
3576 
3577 /*
3578  * KnownAssignedXidsSearch
3579  *
3580  * Searches KnownAssignedXids for a specific xid and optionally removes it.
3581  * Returns true if it was found, false if not.
3582  *
3583  * Caller must hold ProcArrayLock in shared or exclusive mode.
3584  * Exclusive lock must be held for remove = true.
3585  */
3586 static bool
3588 {
3589  ProcArrayStruct *pArray = procArray;
3590  int first,
3591  last;
3592  int head;
3593  int tail;
3594  int result_index = -1;
3595 
3596  if (remove)
3597  {
3598  /* we hold ProcArrayLock exclusively, so no need for spinlock */
3599  tail = pArray->tailKnownAssignedXids;
3600  head = pArray->headKnownAssignedXids;
3601  }
3602  else
3603  {
3604  /* take spinlock to ensure we see up-to-date array contents */
3606  tail = pArray->tailKnownAssignedXids;
3607  head = pArray->headKnownAssignedXids;
3609  }
3610 
3611  /*
3612  * Standard binary search. Note we can ignore the KnownAssignedXidsValid
3613  * array here, since even invalid entries will contain sorted XIDs.
3614  */
3615  first = tail;
3616  last = head - 1;
3617  while (first <= last)
3618  {
3619  int mid_index;
3620  TransactionId mid_xid;
3621 
3622  mid_index = (first + last) / 2;
3623  mid_xid = KnownAssignedXids[mid_index];
3624 
3625  if (xid == mid_xid)
3626  {
3627  result_index = mid_index;
3628  break;
3629  }
3630  else if (TransactionIdPrecedes(xid, mid_xid))
3631  last = mid_index - 1;
3632  else
3633  first = mid_index + 1;
3634  }
3635 
3636  if (result_index < 0)
3637  return false; /* not in array */
3638 
3639  if (!KnownAssignedXidsValid[result_index])
3640  return false; /* in array, but invalid */
3641 
3642  if (remove)
3643  {
3644  KnownAssignedXidsValid[result_index] = false;
3645 
3646  pArray->numKnownAssignedXids--;
3647  Assert(pArray->numKnownAssignedXids >= 0);
3648 
3649  /*
3650  * If we're removing the tail element then advance tail pointer over
3651  * any invalid elements. This will speed future searches.
3652  */
3653  if (result_index == tail)
3654  {
3655  tail++;
3656  while (tail < head && !KnownAssignedXidsValid[tail])
3657  tail++;
3658  if (tail >= head)
3659  {
3660  /* Array is empty, so we can reset both pointers */
3661  pArray->headKnownAssignedXids = 0;
3662  pArray->tailKnownAssignedXids = 0;
3663  }
3664  else
3665  {
3666  pArray->tailKnownAssignedXids = tail;
3667  }
3668  }
3669  }
3670 
3671  return true;
3672 }
3673 
3674 /*
3675  * Is the specified XID present in KnownAssignedXids[]?
3676  *
3677  * Caller must hold ProcArrayLock in shared or exclusive mode.
3678  */
3679 static bool
3681 {
3683 
3684  return KnownAssignedXidsSearch(xid, false);
3685 }
3686 
3687 /*
3688  * Remove the specified XID from KnownAssignedXids[].
3689  *
3690  * Caller must hold ProcArrayLock in exclusive mode.
3691  */
3692 static void
3694 {
3696 
3697  elog(trace_recovery(DEBUG4), "remove KnownAssignedXid %u", xid);
3698 
3699  /*
3700  * Note: we cannot consider it an error to remove an XID that's not
3701  * present. We intentionally remove subxact IDs while processing
3702  * XLOG_XACT_ASSIGNMENT, to avoid array overflow. Then those XIDs will be
3703  * removed again when the top-level xact commits or aborts.
3704  *
3705  * It might be possible to track such XIDs to distinguish this case from
3706  * actual errors, but it would be complicated and probably not worth it.
3707  * So, just ignore the search result.
3708  */
3709  (void) KnownAssignedXidsSearch(xid, true);
3710 }
3711 
3712 /*
3713  * KnownAssignedXidsRemoveTree
3714  * Remove xid (if it's not InvalidTransactionId) and all the subxids.
3715  *
3716  * Caller must hold ProcArrayLock in exclusive mode.
3717  */
3718 static void
3720  TransactionId *subxids)
3721 {
3722  int i;
3723 
3724  if (TransactionIdIsValid(xid))
3726 
3727  for (i = 0; i < nsubxids; i++)
3728  KnownAssignedXidsRemove(subxids[i]);
3729 
3730  /* Opportunistically compress the array */
3732 }
3733 
3734 /*
3735  * Prune KnownAssignedXids up to, but *not* including xid. If xid is invalid
3736  * then clear the whole table.
3737  *
3738  * Caller must hold ProcArrayLock in exclusive mode.
3739  */
3740 static void
3742 {
3743  ProcArrayStruct *pArray = procArray;
3744  int count = 0;
3745  int head,
3746  tail,
3747  i;
3748 
3749  if (!TransactionIdIsValid(removeXid))
3750  {
3751  elog(trace_recovery(DEBUG4), "removing all KnownAssignedXids");
3752  pArray->numKnownAssignedXids = 0;
3753  pArray->headKnownAssignedXids = pArray->tailKnownAssignedXids = 0;
3754  return;
3755  }
3756 
3757  elog(trace_recovery(DEBUG4), "prune KnownAssignedXids to %u", removeXid);
3758 
3759  /*
3760  * Mark entries invalid starting at the tail. Since array is sorted, we
3761  * can stop as soon as we reach an entry >= removeXid.
3762  */
3763  tail = pArray->tailKnownAssignedXids;
3764  head = pArray->headKnownAssignedXids;
3765 
3766  for (i = tail; i < head; i++)
3767  {
3768  if (KnownAssignedXidsValid[i])
3769  {
3770  TransactionId knownXid = KnownAssignedXids[i];
3771 
3772  if (TransactionIdFollowsOrEquals(knownXid, removeXid))
3773  break;
3774 
3775  if (!StandbyTransactionIdIsPrepared(knownXid))
3776  {
3777  KnownAssignedXidsValid[i] = false;
3778  count++;
3779  }
3780  }
3781  }
3782 
3783  pArray->numKnownAssignedXids -= count;
3784  Assert(pArray->numKnownAssignedXids >= 0);
3785 
3786  /*
3787  * Advance the tail pointer if we've marked the tail item invalid.
3788  */
3789  for (i = tail; i < head; i++)
3790  {
3791  if (KnownAssignedXidsValid[i])
3792  break;
3793  }
3794  if (i >= head)
3795  {
3796  /* Array is empty, so we can reset both pointers */
3797  pArray->headKnownAssignedXids = 0;
3798  pArray->tailKnownAssignedXids = 0;
3799  }
3800  else
3801  {
3802  pArray->tailKnownAssignedXids = i;
3803  }
3804 
3805  /* Opportunistically compress the array */
3807 }
3808 
3809 /*
3810  * KnownAssignedXidsGet - Get an array of xids by scanning KnownAssignedXids.
3811  * We filter out anything >= xmax.
3812  *
3813  * Returns the number of XIDs stored into xarray[]. Caller is responsible
3814  * that array is large enough.
3815  *
3816  * Caller must hold ProcArrayLock in (at least) shared mode.
3817  */
3818 static int
3820 {
3822 
3823  return KnownAssignedXidsGetAndSetXmin(xarray, &xtmp, xmax);
3824 }
3825 
3826 /*
3827  * KnownAssignedXidsGetAndSetXmin - as KnownAssignedXidsGet, plus
3828  * we reduce *xmin to the lowest xid value seen if not already lower.
3829  *
3830  * Caller must hold ProcArrayLock in (at least) shared mode.
3831  */
3832 static int
3834  TransactionId xmax)
3835 {
3836  int count = 0;
3837  int head,
3838  tail;
3839  int i;
3840 
3841  /*
3842  * Fetch head just once, since it may change while we loop. We can stop
3843  * once we reach the initially seen head, since we are certain that an xid
3844  * cannot enter and then leave the array while we hold ProcArrayLock. We
3845  * might miss newly-added xids, but they should be >= xmax so irrelevant
3846  * anyway.
3847  *
3848  * Must take spinlock to ensure we see up-to-date array contents.
3849  */
3851  tail = procArray->tailKnownAssignedXids;
3852  head = procArray->headKnownAssignedXids;
3854 
3855  for (i = tail; i < head; i++)
3856  {
3857  /* Skip any gaps in the array */
3858  if (KnownAssignedXidsValid[i])
3859  {
3860  TransactionId knownXid = KnownAssignedXids[i];
3861 
3862  /*
3863  * Update xmin if required. Only the first XID need be checked,
3864  * since the array is sorted.
3865  */
3866  if (count == 0 &&
3867  TransactionIdPrecedes(knownXid, *xmin))
3868  *xmin = knownXid;
3869 
3870  /*
3871  * Filter out anything >= xmax, again relying on sorted property
3872  * of array.
3873  */
3874  if (TransactionIdIsValid(xmax) &&
3875  TransactionIdFollowsOrEquals(knownXid, xmax))
3876  break;
3877 
3878  /* Add knownXid into output array */
3879  xarray[count++] = knownXid;
3880  }
3881  }
3882 
3883  return count;
3884 }
3885 
3886 /*
3887  * Get oldest XID in the KnownAssignedXids array, or InvalidTransactionId
3888  * if nothing there.
3889  */
3890 static TransactionId
3892 {
3893  int head,
3894  tail;
3895  int i;
3896 
3897  /*
3898  * Fetch head just once, since it may change while we loop.
3899  */
3901  tail = procArray->tailKnownAssignedXids;
3902  head = procArray->headKnownAssignedXids;
3904 
3905  for (i = tail; i < head; i++)
3906  {
3907  /* Skip any gaps in the array */
3908  if (KnownAssignedXidsValid[i])
3909  return KnownAssignedXids[i];
3910  }
3911 
3912  return InvalidTransactionId;
3913 }
3914 
3915 /*
3916  * Display KnownAssignedXids to provide debug trail
3917  *
3918  * Currently this is only called within startup process, so we need no
3919  * special locking.
3920  *
3921  * Note this is pretty expensive, and much of the expense will be incurred
3922  * even if the elog message will get discarded. It's not currently called
3923  * in any performance-critical places, however, so no need to be tenser.
3924  */
3925 static void
3927 {
3928  ProcArrayStruct *pArray = procArray;
3930  int head,
3931  tail,
3932  i;
3933  int nxids = 0;
3934 
3935  tail = pArray->tailKnownAssignedXids;
3936  head = pArray->headKnownAssignedXids;
3937 
3938  initStringInfo(&buf);
3939 
3940  for (i = tail; i < head; i++)
3941  {
3942  if (KnownAssignedXidsValid[i])
3943  {
3944  nxids++;
3945  appendStringInfo(&buf, "[%d]=%u ", i, KnownAssignedXids[i]);
3946  }
3947  }
3948 
3949  elog(trace_level, "%d KnownAssignedXids (num=%d tail=%d head=%d) %s",
3950  nxids,
3951  pArray->numKnownAssignedXids,
3952  pArray->tailKnownAssignedXids,
3953  pArray->headKnownAssignedXids,
3954  buf.data);
3955 
3956  pfree(buf.data);
3957 }
3958 
3959 /*
3960  * KnownAssignedXidsReset
3961  * Resets KnownAssignedXids to be empty
3962  */
3963 static void
3965 {
3966  ProcArrayStruct *pArray = procArray;
3967 
3968  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3969 
3970  pArray->numKnownAssignedXids = 0;
3971  pArray->tailKnownAssignedXids = 0;
3972  pArray->headKnownAssignedXids = 0;
3973 
3974  LWLockRelease(ProcArrayLock);
3975 }
#define TransactionIdAdvance(dest)
Definition: transam.h:75
int slock_t
Definition: s_lock.h:929
void ProcArrayApplyRecoveryInfo(RunningTransactions running)
Definition: procarray.c:662
#define XidCacheRemove(i)
Definition: procarray.c:3018
static TransactionId latestObservedXid
Definition: procarray.c:109
VirtualTransactionId * GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0, bool allDbs, int excludeVacuum, int *nvxids)
Definition: procarray.c:2491
TransactionId oldestRunningXid
Definition: standby.h:76
bool procArrayGroupMember
Definition: proc.h:166
pid_t CancelVirtualTransaction(VirtualTransactionId vxid, ProcSignalReason sigmode)
Definition: procarray.c:2654
#define PROCARRAY_MAXPROCS
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
TransactionId GetOldestSafeDecodingTransactionId(bool catalogOnly)
Definition: procarray.c:2175
int CountDBBackends(Oid databaseid)
Definition: procarray.c:2756
#define DEBUG1
Definition: elog.h:25
static void KnownAssignedXidsDisplay(int trace_level)
Definition: procarray.c:3926
#define GET_VXID_FROM_PGPROC(vxid, proc)
Definition: lock.h:80
#define TransactionIdEquals(id1, id2)
Definition: transam.h:43
bool TransactionIdFollows(TransactionId id1, TransactionId id2)
Definition: transam.c:334
#define PROC_IN_LOGICAL_DECODING
Definition: proc.h:57
BackendId backendId
Definition: proc.h:113
uint32 TransactionId
Definition: c.h:507
bool copied
Definition: snapshot.h:185
void AdvanceNextFullTransactionIdPastXid(TransactionId xid)
Definition: varsup.c:262
void PGSemaphoreUnlock(PGSemaphore sema)
Definition: posix_sema.c:329
Definition: proc.h:222
#define DEBUG3
Definition: elog.h:23
bool TransactionIdIsCurrentTransactionId(TransactionId xid)
Definition: xact.c:854
TransactionId SubTransGetTopmostTransaction(TransactionId xid)
Definition: subtrans.c:150
bool LWLockHeldByMe(LWLock *l)
Definition: lwlock.c:1842
TransactionId xmin
Definition: proc.h:228
static bool pg_atomic_compare_exchange_u32(volatile pg_atomic_uint32 *ptr, uint32 *expected, uint32 newval)
Definition: atomics.h:314
PGPROC * BackendPidGetProc(int pid)
Definition: procarray.c:2363
#define FullTransactionIdIsValid(x)
Definition: transam.h:51
PGXACT * allPgXact
Definition: proc.h:249
PGPROC * MyProc
Definition: proc.c:68
bool TransactionIdIsInProgress(TransactionId xid)
Definition: procarray.c:985
TransactionId xid
Definition: proc.h:224
TransactionId TransactionIdLatest(TransactionId mainxid, int nxids, const TransactionId *xids)
Definition: transam.c:365
int vacuum_defer_cleanup_age
Definition: standby.c:39
#define UINT32_ACCESS_ONCE(var)
Definition: procarray.c:64
#define SpinLockInit(lock)
Definition: spin.h:60
VirtualTransactionId * GetConflictingVirtualXIDs(TransactionId limitXmin, Oid dbOid)
Definition: procarray.c:2579
TransactionId replication_slot_catalog_xmin
Definition: procarray.c:93
XLogRecPtr lsn
Definition: snapshot.h:203
bool TransactionIdFollowsOrEquals(TransactionId id1, TransactionId id2)
Definition: transam.c:349
VirtualTransactionId * GetVirtualXIDsDelayingChkpt(int *nvxids)
Definition: procarray.c:2271
static void ProcArrayGroupClearXid(PGPROC *proc, TransactionId latestXid)
Definition: procarray.c:484
#define xc_by_my_xact_inc()
Definition: procarray.c:146
Oid roleId
Definition: proc.h:115
int errcode(int sqlerrcode)
Definition: elog.c:570
TransactionId RecentXmin
Definition: snapmgr.c:167
slock_t known_assigned_xids_lck
Definition: procarray.c:79
PROC_HDR * ProcGlobal
Definition: proc.c:81
bool suboverflowed
Definition: snapshot.h:182
TransactionId * xids
Definition: standby.h:79
#define kill(pid, sig)
Definition: win32_port.h:435
FullTransactionId nextFullXid
Definition: transam.h:164
bool TransactionIdDidCommit(TransactionId transactionId)
Definition: transam.c:125
#define LOG
Definition: elog.h:26
Form_pg_class rd_rel
Definition: rel.h:83
unsigned int Oid
Definition: postgres_ext.h:31
bool RecoveryInProgress(void)
Definition: xlog.c:7898
#define TransactionIdRetreat(dest)
Definition: transam.h:92
LocalTransactionId localTransactionId
Definition: lock.h:66
void ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
Definition: procarray.c:396
#define xc_by_child_xid_inc()
Definition: procarray.c:149
bool TransactionIdIsKnownCompleted(TransactionId transactionId)
Definition: transam.c:238
#define DEBUG4
Definition: elog.h:22
#define fprintf
Definition: port.h:196
void ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids, TransactionId *subxids, TransactionId max_xid)
Definition: procarray.c:3267
#define MAXAUTOVACPIDS
uint32 regd_count
Definition: snapshot.h:199
#define OidIsValid(objectId)
Definition: c.h:638
void ProcArrayGetReplicationSlotXmin(TransactionId *xmin, TransactionId *catalog_xmin)
Definition: procarray.c:3003
void ExpireAllKnownAssignedTransactionIds(void)
Definition: procarray.c:3292
PGXACT * MyPgXact
Definition: proc.c:69
int trace_recovery(int trace_level)
Definition: elog.c:3474
uint8 vacuumFlags
Definition: proc.h:233
#define PROC_VACUUM_STATE_MASK
Definition: proc.h:62
#define XidFromFullTransactionId(x)
Definition: transam.h:48
TransactionId TransactionXmin
Definition: snapmgr.c:166
void LWLockRegisterTranche(int tranche_id, const char *tranche_name)
Definition: lwlock.c:603
TransactionId latestCompletedXid
Definition: standby.h:77
Definition: type.h:89
#define malloc(a)
Definition: header.h:50
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1726
static uint32 pg_atomic_exchange_u32(volatile pg_atomic_uint32 *ptr, uint32 newval)
Definition: atomics.h:295
bool isBackgroundWorker
Definition: proc.h:120
#define VirtualTransactionIdEquals(vxid1, vxid2)
Definition: lock.h:74
#define xc_by_recent_xmin_inc()
Definition: procarray.c:144
#define xc_by_known_xact_inc()
Definition: procarray.c:145
bool MinimumActiveBackends(int min)
Definition: procarray.c:2702
static void KnownAssignedXidsRemovePreceding(TransactionId xid)
Definition: procarray.c:3741
PGPROC * BackendPidGetProcWithLock(int pid)
Definition: procarray.c:2386
#define SpinLockAcquire(lock)
Definition: spin.h:62
void pg_usleep(long microsec)
Definition: signal.c:53
int SendProcSignal(pid_t pid, ProcSignalReason reason, BackendId backendId)
Definition: procsignal.c:180
void pfree(void *pointer)
Definition: mcxt.c:1031
#define PROC_IN_VACUUM
Definition: proc.h:54
void appendStringInfo(StringInfo str, const char *fmt,...)
Definition: stringinfo.c:78
bool TransactionIdPrecedesOrEquals(TransactionId id1, TransactionId id2)
Definition: transam.c:319
int CountDBConnections(Oid databaseid)
Definition: procarray.c:2786
#define FirstNormalTransactionId
Definition: transam.h:34
#define ERROR
Definition: elog.h:43
void XidCacheRemoveRunningXids(TransactionId xid, int nxids, const TransactionId *xids, TransactionId latestXid)
Definition: procarray.c:3034
TimestampTz GetSnapshotCurrentTimestamp(void)
Definition: snapmgr.c:1720
void ProcArrayClearTransaction(PGPROC *proc)
Definition: procarray.c:599
#define PROCARRAY_SLOTS_XMIN
Definition: procarray.h:37
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:372
#define PROCARRAY_PROC_FLAGS_MASK
Definition: procarray.h:44
void ExtendSUBTRANS(TransactionId newestXact)
Definition: subtrans.c:326
void ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
Definition: procarray.c:333
#define FATAL
Definition: elog.h:52
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:11149
void CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
Definition: procarray.c:2817
#define xc_by_main_xid_inc()
Definition: procarray.c:148
bool LWLockConditionalAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1294
static char * buf
Definition: pg_test_fsync.c:68
bool recoveryConflictPending
Definition: proc.h:127
#define memmove(d, s, c)
Definition: c.h:1238
#define xc_by_known_assigned_inc()
Definition: procarray.c:150
TransactionId RecentGlobalXmin
Definition: snapmgr.c:168
uint8 nxids
Definition: proc.h:238
VariableCache ShmemVariableCache
Definition: varsup.c:34
int maxKnownAssignedXids
Definition: procarray.c:75
#define InvalidTransactionId
Definition: transam.h:31
bool StandbyTransactionIdIsPrepared(TransactionId xid)
Definition: twophase.c:1427
static PGPROC * allProcs
Definition: procarray.c:101
Oid databaseId
Definition: proc.h:114
unsigned int uint32
Definition: c.h:358
TransactionId xmax
Definition: snapshot.h:158
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1342
TransactionId xmin
Definition: snapshot.h:157
static void KnownAssignedXidsReset(void)
Definition: procarray.c:3964
LOCK * waitLock
Definition: proc.h:139
int numKnownAssignedXids
Definition: procarray.c:76
TransactionId RecentGlobalDataXmin
Definition: snapmgr.c:169
static bool * KnownAssignedXidsValid
Definition: procarray.c:108
struct XidCache subxids
Definition: proc.h:162
TransactionId lastOverflowedXid
Definition: procarray.c:88
#define ereport(elevel, rest)
Definition: elog.h:141
bool TransactionIdDidAbort(TransactionId transactionId)
Definition: transam.c:181
#define xc_by_latest_xid_inc()
Definition: procarray.c:147
bool delayChkpt
Definition: proc.h:235
#define INVALID_PGPROCNO
Definition: proc.h:77
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
TransactionId * xip
Definition: snapshot.h:168
static void KnownAssignedXidsRemoveTree(TransactionId xid, int nsubxids, TransactionId *subxids)
Definition: procarray.c:3719
pg_atomic_uint32 procArrayGroupNext
Definition: proc.h:168
Definition: proc.h:244
void initStringInfo(StringInfo str)
Definition: stringinfo.c:46
static ProcArrayStruct * procArray
Definition: procarray.c:99
#define WARNING
Definition: elog.h:40
#define VirtualTransactionIdIsValid(vxid)
Definition: lock.h:71
#define SpinLockRelease(lock)
Definition: spin.h:64
TransactionId replication_slot_xmin
Definition: procarray.c:91
Size mul_size(Size s1, Size s2)
Definition: shmem.c:492
int BackendXidGetPid(TransactionId xid)
Definition: procarray.c:2423
#define InvalidBackendId
Definition: backendid.h:23
static PGXACT * allPgXact
Definition: procarray.c:102
#define RelationIsAccessibleInLogicalDecoding(relation)
Definition: rel.h:572
static int KnownAssignedXidsGet(TransactionId *xarray, TransactionId xmax)
Definition: procarray.c:3819
Size add_size(Size s1, Size s2)
Definition: shmem.c:475
Oid MyDatabaseId
Definition: globals.c:85
static TransactionId KnownAssignedXidsGetOldestXmin(void)
Definition: procarray.c:3891
bool overflowed
Definition: proc.h:234
static void ProcArrayEndTransactionInternal(PGPROC *proc, PGXACT *pgxact, TransactionId latestXid)
Definition: procarray.c:450
#define InvalidOid
Definition: postgres_ext.h:36
CommandId curcid
Definition: snapshot.h:187
int GetMaxSnapshotXidCount(void)
Definition: procarray.c:1450
TransactionId GetOldestXmin(Relation rel, int flags)
Definition: procarray.c:1304
int pgprocnos[FLEXIBLE_ARRAY_MEMBER]
Definition: procarray.c:96
TransactionId xids[PGPROC_MAX_CACHED_SUBXIDS]
Definition: proc.h:40
#define TOTAL_MAX_CACHED_SUBXIDS
#define Assert(condition)
Definition: c.h:732
static TransactionId * KnownAssignedXids
Definition: procarray.c:107
BackendId backendId
Definition: lock.h:65
#define pg_read_barrier()
Definition: atomics.h:161
void CreateSharedProcArray(void)
Definition: procarray.c:222
bool takenDuringRecovery
Definition: snapshot.h:184
size_t Size
Definition: c.h:466
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1318
Snapshot GetSnapshotData(Snapshot snapshot)
Definition: procarray.c:1503
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1122
static int KnownAssignedXidsGetAndSetXmin(TransactionId *xarray, TransactionId *xmin, TransactionId xmax)
Definition: procarray.c:3833
static void KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid, bool exclusive_lock)
Definition: procarray.c:3469
bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
Definition: procarray.c:1869
#define NormalTransactionIdPrecedes(id1, id2)
Definition: transam.h:98
#define xc_no_overflow_inc()
Definition: procarray.c:151
bool EnableHotStandby
Definition: xlog.c:93
void PGSemaphoreLock(PGSemaphore sema)
Definition: posix_sema.c:309
static void KnownAssignedXidsCompress(bool force)
Definition: procarray.c:3407
int CountUserBackends(Oid roleid)
Definition: procarray.c:2857
static bool KnownAssignedXidExists(TransactionId xid)
Definition: procarray.c:3680
int pgprocno
Definition: proc.h:110
TransactionId nextXid
Definition: standby.h:75
bool TransactionIdIsActive(TransactionId xid)
Definition: procarray.c:1205
#define xc_slow_answer_inc()
Definition: procarray.c:152
pg_atomic_uint32 procArrayGroupFirst
Definition: proc.h:261
void ProcArrayInitRecovery(TransactionId initializedUptoXID)
Definition: procarray.c:631
uint32 xcnt
Definition: snapshot.h:169
void * palloc(Size size)
Definition: mcxt.c:924
int errmsg(const char *fmt,...)
Definition: elog.c:784
struct ProcArrayStruct ProcArrayStruct
static bool KnownAssignedXidsSearch(TransactionId xid, bool remove)
Definition: procarray.c:3587
static void KnownAssignedXidsRemove(TransactionId xid)
Definition: procarray.c:3693
#define elog(elevel,...)
Definition: elog.h:226
int old_snapshot_threshold
Definition: snapmgr.c:75
#define InvalidLocalTransactionId
Definition: lock.h:69
int i
void ExpireOldKnownAssignedTransactionIds(TransactionId xid)
Definition: procarray.c:3304
TransactionId GetOldestActiveTransactionId(void)
Definition: procarray.c:2109
bool IsBackendPid(int pid)
Definition: procarray.c:2458
#define pg_write_barrier()
Definition: atomics.h:162
ProcSignalReason
Definition: procsignal.h:30
void ProcArraySetReplicationSlotXmin(TransactionId xmin, TransactionId catalog_xmin, bool already_locked)
Definition: procarray.c:2981
int GetMaxSnapshotSubxidCount(void)
Definition: procarray.c:1461
RunningTransactions GetRunningTransactionData(void)
Definition: procarray.c:1935
void ProcArrayApplyXidAssignment(TransactionId topxid, int nsubxids, TransactionId *subxids)
Definition: procarray.c:901
TimestampTz whenTaken
Definition: snapshot.h:202
PGPROC * allProcs
Definition: proc.h:247
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:99
bool CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
Definition: procarray.c:2907
CommandId GetCurrentCommandId(bool used)
Definition: xact.c:746
#define qsort(a, b, c, d)
Definition: port.h:491
#define TransactionIdIsValid(xid)
Definition: transam.h:41
static void pg_atomic_write_u32(volatile pg_atomic_uint32 *ptr, uint32 val)
Definition: atomics.h:261
void MaintainOldSnapshotTimeMapping(TimestampTz whenTaken, TransactionId xmin)
Definition: snapmgr.c:1882
PGSemaphore sem
Definition: proc.h:101
bool HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids)
Definition: procarray.c:2316
void RecordKnownAssignedTransactionIds(TransactionId xid)
Definition: procarray.c:3196
#define TransactionIdIsNormal(xid)
Definition: transam.h:42
void SubTransSetParent(TransactionId xid, TransactionId parent)
Definition: subtrans.c:74
bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid)
Definition: procarray.c:1794
int tailKnownAssignedXids
Definition: procarray.c:77
static TransactionId standbySnapshotPendingXmin
Definition: procarray.c:116
Definition: proc.h:95
int pid
Definition: proc.h:109
HotStandbyState standbyState
Definition: xlog.c:203
void ProcArrayAdd(PGPROC *proc)
Definition: procarray.c:275
#define PROC_IS_AUTOVACUUM
Definition: proc.h:53
#define offsetof(type, field)
Definition: c.h:655
TransactionId procArrayGroupMemberXid
Definition: proc.h:174
Size ProcArrayShmemSize(void)
Definition: procarray.c:180
void StandbyReleaseOldLocks(TransactionId oldxid)
Definition: standby.c:751
TransactionId * subxip
Definition: snapshot.h:180
uint32 active_count
Definition: snapshot.h:198
int headKnownAssignedXids
Definition: procarray.c:78
int xidComparator(const void *arg1, const void *arg2)
Definition: xid.c:138
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:244
int32 subxcnt
Definition: snapshot.h:181
LocalTransactionId lxid
Definition: proc.h:106
TransactionId latestCompletedXid
Definition: transam.h:182