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) 1996-2018, 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 
268  /* Register and initialize fields of ProcLWLockTranche */
270 }
271 
272 /*
273  * Add the specified PGPROC to the shared array.
274  */
275 void
277 {
278  ProcArrayStruct *arrayP = procArray;
279  int index;
280 
281  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
282 
283  if (arrayP->numProcs >= arrayP->maxProcs)
284  {
285  /*
286  * Oops, no room. (This really shouldn't happen, since there is a
287  * fixed supply of PGPROC structs too, and so we should have failed
288  * earlier.)
289  */
290  LWLockRelease(ProcArrayLock);
291  ereport(FATAL,
292  (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
293  errmsg("sorry, too many clients already")));
294  }
295 
296  /*
297  * Keep the procs array sorted by (PGPROC *) so that we can utilize
298  * locality of references much better. This is useful while traversing the
299  * ProcArray because there is an increased likelihood of finding the next
300  * PGPROC structure in the cache.
301  *
302  * Since the occurrence of adding/removing a proc is much lower than the
303  * access to the ProcArray itself, the overhead should be marginal
304  */
305  for (index = 0; index < arrayP->numProcs; index++)
306  {
307  /*
308  * If we are the first PGPROC or if we have found our right position
309  * in the array, break
310  */
311  if ((arrayP->pgprocnos[index] == -1) || (arrayP->pgprocnos[index] > proc->pgprocno))
312  break;
313  }
314 
315  memmove(&arrayP->pgprocnos[index + 1], &arrayP->pgprocnos[index],
316  (arrayP->numProcs - index) * sizeof(int));
317  arrayP->pgprocnos[index] = proc->pgprocno;
318  arrayP->numProcs++;
319 
320  LWLockRelease(ProcArrayLock);
321 }
322 
323 /*
324  * Remove the specified PGPROC from the shared array.
325  *
326  * When latestXid is a valid XID, we are removing a live 2PC gxact from the
327  * array, and thus causing it to appear as "not running" anymore. In this
328  * case we must advance latestCompletedXid. (This is essentially the same
329  * as ProcArrayEndTransaction followed by removal of the PGPROC, but we take
330  * the ProcArrayLock only once, and don't damage the content of the PGPROC;
331  * twophase.c depends on the latter.)
332  */
333 void
335 {
336  ProcArrayStruct *arrayP = procArray;
337  int index;
338 
339 #ifdef XIDCACHE_DEBUG
340  /* dump stats at backend shutdown, but not prepared-xact end */
341  if (proc->pid != 0)
342  DisplayXidCache();
343 #endif
344 
345  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
346 
347  if (TransactionIdIsValid(latestXid))
348  {
349  Assert(TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
350 
351  /* Advance global latestCompletedXid while holding the lock */
353  latestXid))
355  }
356  else
357  {
358  /* Shouldn't be trying to remove a live transaction here */
359  Assert(!TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
360  }
361 
362  for (index = 0; index < arrayP->numProcs; index++)
363  {
364  if (arrayP->pgprocnos[index] == proc->pgprocno)
365  {
366  /* Keep the PGPROC array sorted. See notes above */
367  memmove(&arrayP->pgprocnos[index], &arrayP->pgprocnos[index + 1],
368  (arrayP->numProcs - index - 1) * sizeof(int));
369  arrayP->pgprocnos[arrayP->numProcs - 1] = -1; /* for debugging */
370  arrayP->numProcs--;
371  LWLockRelease(ProcArrayLock);
372  return;
373  }
374  }
375 
376  /* Oops */
377  LWLockRelease(ProcArrayLock);
378 
379  elog(LOG, "failed to find proc %p in ProcArray", proc);
380 }
381 
382 
383 /*
384  * ProcArrayEndTransaction -- mark a transaction as no longer running
385  *
386  * This is used interchangeably for commit and abort cases. The transaction
387  * commit/abort must already be reported to WAL and pg_xact.
388  *
389  * proc is currently always MyProc, but we pass it explicitly for flexibility.
390  * latestXid is the latest Xid among the transaction's main XID and
391  * subtransactions, or InvalidTransactionId if it has no XID. (We must ask
392  * the caller to pass latestXid, instead of computing it from the PGPROC's
393  * contents, because the subxid information in the PGPROC might be
394  * incomplete.)
395  */
396 void
398 {
399  PGXACT *pgxact = &allPgXact[proc->pgprocno];
400 
401  if (TransactionIdIsValid(latestXid))
402  {
403  /*
404  * We must lock ProcArrayLock while clearing our advertised XID, so
405  * that we do not exit the set of "running" transactions while someone
406  * else is taking a snapshot. See discussion in
407  * src/backend/access/transam/README.
408  */
409  Assert(TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
410 
411  /*
412  * If we can immediately acquire ProcArrayLock, we clear our own XID
413  * and release the lock. If not, use group XID clearing to improve
414  * efficiency.
415  */
416  if (LWLockConditionalAcquire(ProcArrayLock, LW_EXCLUSIVE))
417  {
418  ProcArrayEndTransactionInternal(proc, pgxact, latestXid);
419  LWLockRelease(ProcArrayLock);
420  }
421  else
422  ProcArrayGroupClearXid(proc, latestXid);
423  }
424  else
425  {
426  /*
427  * If we have no XID, we don't need to lock, since we won't affect
428  * anyone else's calculation of a snapshot. We might change their
429  * estimate of global xmin, but that's OK.
430  */
431  Assert(!TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
432 
434  pgxact->xmin = InvalidTransactionId;
435  /* must be cleared with xid/xmin: */
437  pgxact->delayChkpt = false; /* be sure this is cleared in abort */
438  proc->recoveryConflictPending = false;
439 
440  Assert(pgxact->nxids == 0);
441  Assert(pgxact->overflowed == false);
442  }
443 }
444 
445 /*
446  * Mark a write transaction as no longer running.
447  *
448  * We don't do any locking here; caller must handle that.
449  */
450 static inline void
452  TransactionId latestXid)
453 {
454  pgxact->xid = InvalidTransactionId;
456  pgxact->xmin = InvalidTransactionId;
457  /* must be cleared with xid/xmin: */
459  pgxact->delayChkpt = false; /* be sure this is cleared in abort */
460  proc->recoveryConflictPending = false;
461 
462  /* Clear the subtransaction-XID cache too while holding the lock */
463  pgxact->nxids = 0;
464  pgxact->overflowed = false;
465 
466  /* Also advance global latestCompletedXid while holding the lock */
468  latestXid))
470 }
471 
472 /*
473  * ProcArrayGroupClearXid -- group XID clearing
474  *
475  * When we cannot immediately acquire ProcArrayLock in exclusive mode at
476  * commit time, add ourselves to a list of processes that need their XIDs
477  * cleared. The first process to add itself to the list will acquire
478  * ProcArrayLock in exclusive mode and perform ProcArrayEndTransactionInternal
479  * on behalf of all group members. This avoids a great deal of contention
480  * around ProcArrayLock when many processes are trying to commit at once,
481  * since the lock need not be repeatedly handed off from one committing
482  * process to the next.
483  */
484 static void
486 {
487  PROC_HDR *procglobal = ProcGlobal;
488  uint32 nextidx;
489  uint32 wakeidx;
490 
491  /* We should definitely have an XID to clear. */
492  Assert(TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
493 
494  /* Add ourselves to the list of processes needing a group XID clear. */
495  proc->procArrayGroupMember = true;
496  proc->procArrayGroupMemberXid = latestXid;
497  while (true)
498  {
499  nextidx = pg_atomic_read_u32(&procglobal->procArrayGroupFirst);
500  pg_atomic_write_u32(&proc->procArrayGroupNext, nextidx);
501 
503  &nextidx,
504  (uint32) proc->pgprocno))
505  break;
506  }
507 
508  /*
509  * If the list was not empty, the leader will clear our XID. It is
510  * impossible to have followers without a leader because the first process
511  * that has added itself to the list will always have nextidx as
512  * INVALID_PGPROCNO.
513  */
514  if (nextidx != INVALID_PGPROCNO)
515  {
516  int extraWaits = 0;
517 
518  /* Sleep until the leader clears our XID. */
520  for (;;)
521  {
522  /* acts as a read barrier */
523  PGSemaphoreLock(proc->sem);
524  if (!proc->procArrayGroupMember)
525  break;
526  extraWaits++;
527  }
529 
531 
532  /* Fix semaphore count for any absorbed wakeups */
533  while (extraWaits-- > 0)
534  PGSemaphoreUnlock(proc->sem);
535  return;
536  }
537 
538  /* We are the leader. Acquire the lock on behalf of everyone. */
539  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
540 
541  /*
542  * Now that we've got the lock, clear the list of processes waiting for
543  * group XID clearing, saving a pointer to the head of the list. Trying
544  * to pop elements one at a time could lead to an ABA problem.
545  */
546  nextidx = pg_atomic_exchange_u32(&procglobal->procArrayGroupFirst,
548 
549  /* Remember head of list so we can perform wakeups after dropping lock. */
550  wakeidx = nextidx;
551 
552  /* Walk the list and clear all XIDs. */
553  while (nextidx != INVALID_PGPROCNO)
554  {
555  PGPROC *proc = &allProcs[nextidx];
556  PGXACT *pgxact = &allPgXact[nextidx];
557 
559 
560  /* Move to next proc in list. */
561  nextidx = pg_atomic_read_u32(&proc->procArrayGroupNext);
562  }
563 
564  /* We're done with the lock now. */
565  LWLockRelease(ProcArrayLock);
566 
567  /*
568  * Now that we've released the lock, go back and wake everybody up. We
569  * don't do this under the lock so as to keep lock hold times to a
570  * minimum. The system calls we need to perform to wake other processes
571  * up are probably much slower than the simple memory writes we did while
572  * holding the lock.
573  */
574  while (wakeidx != INVALID_PGPROCNO)
575  {
576  PGPROC *proc = &allProcs[wakeidx];
577 
578  wakeidx = pg_atomic_read_u32(&proc->procArrayGroupNext);
580 
581  /* ensure all previous writes are visible before follower continues. */
583 
584  proc->procArrayGroupMember = false;
585 
586  if (proc != MyProc)
587  PGSemaphoreUnlock(proc->sem);
588  }
589 }
590 
591 /*
592  * ProcArrayClearTransaction -- clear the transaction fields
593  *
594  * This is used after successfully preparing a 2-phase transaction. We are
595  * not actually reporting the transaction's XID as no longer running --- it
596  * will still appear as running because the 2PC's gxact is in the ProcArray
597  * too. We just have to clear out our own PGXACT.
598  */
599 void
601 {
602  PGXACT *pgxact = &allPgXact[proc->pgprocno];
603 
604  /*
605  * We can skip locking ProcArrayLock here, because this action does not
606  * actually change anyone's view of the set of running XIDs: our entry is
607  * duplicate with the gxact that has already been inserted into the
608  * ProcArray.
609  */
610  pgxact->xid = InvalidTransactionId;
612  pgxact->xmin = InvalidTransactionId;
613  proc->recoveryConflictPending = false;
614 
615  /* redundant, but just in case */
617  pgxact->delayChkpt = false;
618 
619  /* Clear the subtransaction-XID cache too */
620  pgxact->nxids = 0;
621  pgxact->overflowed = false;
622 }
623 
624 /*
625  * ProcArrayInitRecovery -- initialize recovery xid mgmt environment
626  *
627  * Remember up to where the startup process initialized the CLOG and subtrans
628  * so we can ensure it's initialized gaplessly up to the point where necessary
629  * while in recovery.
630  */
631 void
633 {
635  Assert(TransactionIdIsNormal(initializedUptoXID));
636 
637  /*
638  * we set latestObservedXid to the xid SUBTRANS has been initialized up
639  * to, so we can extend it from that point onwards in
640  * RecordKnownAssignedTransactionIds, and when we get consistent in
641  * ProcArrayApplyRecoveryInfo().
642  */
643  latestObservedXid = initializedUptoXID;
645 }
646 
647 /*
648  * ProcArrayApplyRecoveryInfo -- apply recovery info about xids
649  *
650  * Takes us through 3 states: Initialized, Pending and Ready.
651  * Normal case is to go all the way to Ready straight away, though there
652  * are atypical cases where we need to take it in steps.
653  *
654  * Use the data about running transactions on master to create the initial
655  * state of KnownAssignedXids. We also use these records to regularly prune
656  * KnownAssignedXids because we know it is possible that some transactions
657  * with FATAL errors fail to write abort records, which could cause eventual
658  * overflow.
659  *
660  * See comments for LogStandbySnapshot().
661  */
662 void
664 {
665  TransactionId *xids;
666  int nxids;
667  TransactionId nextXid;
668  int i;
669 
674 
675  /*
676  * Remove stale transactions, if any.
677  */
679 
680  /*
681  * Remove stale locks, if any.
682  */
684 
685  /*
686  * If our snapshot is already valid, nothing else to do...
687  */
689  return;
690 
691  /*
692  * If our initial RunningTransactionsData had an overflowed snapshot then
693  * we knew we were missing some subxids from our snapshot. If we continue
694  * to see overflowed snapshots then we might never be able to start up, so
695  * we make another test to see if our snapshot is now valid. We know that
696  * the missing subxids are equal to or earlier than nextXid. After we
697  * initialise we continue to apply changes during recovery, so once the
698  * oldestRunningXid is later than the nextXid from the initial snapshot we
699  * know that we no longer have missing information and can mark the
700  * snapshot as valid.
701  */
703  {
704  /*
705  * If the snapshot isn't overflowed or if its empty we can reset our
706  * pending state and use this snapshot instead.
707  */
708  if (!running->subxid_overflow || running->xcnt == 0)
709  {
710  /*
711  * If we have already collected known assigned xids, we need to
712  * throw them away before we apply the recovery snapshot.
713  */
716  }
717  else
718  {
720  running->oldestRunningXid))
721  {
724  "recovery snapshots are now enabled");
725  }
726  else
728  "recovery snapshot waiting for non-overflowed snapshot or "
729  "until oldest active xid on standby is at least %u (now %u)",
731  running->oldestRunningXid);
732  return;
733  }
734  }
735 
737 
738  /*
739  * OK, we need to initialise from the RunningTransactionsData record.
740  *
741  * NB: this can be reached at least twice, so make sure new code can deal
742  * with that.
743  */
744 
745  /*
746  * Nobody else is running yet, but take locks anyhow
747  */
748  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
749 
750  /*
751  * KnownAssignedXids is sorted so we cannot just add the xids, we have to
752  * sort them first.
753  *
754  * Some of the new xids are top-level xids and some are subtransactions.
755  * We don't call SubtransSetParent because it doesn't matter yet. If we
756  * aren't overflowed then all xids will fit in snapshot and so we don't
757  * need subtrans. If we later overflow, an xid assignment record will add
758  * xids to subtrans. If RunningXacts is overflowed then we don't have
759  * enough information to correctly update subtrans anyway.
760  */
761 
762  /*
763  * Allocate a temporary array to avoid modifying the array passed as
764  * argument.
765  */
766  xids = palloc(sizeof(TransactionId) * (running->xcnt + running->subxcnt));
767 
768  /*
769  * Add to the temp array any xids which have not already completed.
770  */
771  nxids = 0;
772  for (i = 0; i < running->xcnt + running->subxcnt; i++)
773  {
774  TransactionId xid = running->xids[i];
775 
776  /*
777  * The running-xacts snapshot can contain xids that were still visible
778  * in the procarray when the snapshot was taken, but were already
779  * WAL-logged as completed. They're not running anymore, so ignore
780  * them.
781  */
783  continue;
784 
785  xids[nxids++] = xid;
786  }
787 
788  if (nxids > 0)
789  {
790  if (procArray->numKnownAssignedXids != 0)
791  {
792  LWLockRelease(ProcArrayLock);
793  elog(ERROR, "KnownAssignedXids is not empty");
794  }
795 
796  /*
797  * Sort the array so that we can add them safely into
798  * KnownAssignedXids.
799  */
800  qsort(xids, nxids, sizeof(TransactionId), xidComparator);
801 
802  /*
803  * Add the sorted snapshot into KnownAssignedXids. The running-xacts
804  * snapshot may include duplicated xids because of prepared
805  * transactions, so ignore them.
806  */
807  for (i = 0; i < nxids; i++)
808  {
809  if (i > 0 && TransactionIdEquals(xids[i - 1], xids[i]))
810  {
811  elog(DEBUG1,
812  "found duplicated transaction %u for KnownAssignedXids insertion",
813  xids[i]);
814  continue;
815  }
816  KnownAssignedXidsAdd(xids[i], xids[i], true);
817  }
818 
820  }
821 
822  pfree(xids);
823 
824  /*
825  * latestObservedXid is at least set to the point where SUBTRANS was
826  * started up to (cf. ProcArrayInitRecovery()) or to the biggest xid
827  * RecordKnownAssignedTransactionIds() was called for. Initialize
828  * subtrans from thereon, up to nextXid - 1.
829  *
830  * We need to duplicate parts of RecordKnownAssignedTransactionId() here,
831  * because we've just added xids to the known assigned xids machinery that
832  * haven't gone through RecordKnownAssignedTransactionId().
833  */
837  {
840  }
841  TransactionIdRetreat(latestObservedXid); /* = running->nextXid - 1 */
842 
843  /* ----------
844  * Now we've got the running xids we need to set the global values that
845  * are used to track snapshots as they evolve further.
846  *
847  * - latestCompletedXid which will be the xmax for snapshots
848  * - lastOverflowedXid which shows whether snapshots overflow
849  * - nextXid
850  *
851  * If the snapshot overflowed, then we still initialise with what we know,
852  * but the recovery snapshot isn't fully valid yet because we know there
853  * are some subxids missing. We don't know the specific subxids that are
854  * missing, so conservatively assume the last one is latestObservedXid.
855  * ----------
856  */
857  if (running->subxid_overflow)
858  {
860 
863  }
864  else
865  {
867 
869  }
870 
871  /*
872  * If a transaction wrote a commit record in the gap between taking and
873  * logging the snapshot then latestCompletedXid may already be higher than
874  * the value from the snapshot, so check before we use the incoming value.
875  */
877  running->latestCompletedXid))
879 
881 
882  LWLockRelease(ProcArrayLock);
883 
884  /*
885  * ShmemVariableCache->nextXid must be beyond any observed xid.
886  *
887  * We don't expect anyone else to modify nextXid, hence we don't need to
888  * hold a lock while examining it. We still acquire the lock to modify
889  * it, though.
890  */
891  nextXid = latestObservedXid;
892  TransactionIdAdvance(nextXid);
894  {
895  LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
896  ShmemVariableCache->nextXid = nextXid;
897  LWLockRelease(XidGenLock);
898  }
899 
901 
904  elog(trace_recovery(DEBUG1), "recovery snapshots are now enabled");
905  else
907  "recovery snapshot waiting for non-overflowed snapshot or "
908  "until oldest active xid on standby is at least %u (now %u)",
910  running->oldestRunningXid);
911 }
912 
913 /*
914  * ProcArrayApplyXidAssignment
915  * Process an XLOG_XACT_ASSIGNMENT WAL record
916  */
917 void
919  int nsubxids, TransactionId *subxids)
920 {
921  TransactionId max_xid;
922  int i;
923 
925 
926  max_xid = TransactionIdLatest(topxid, nsubxids, subxids);
927 
928  /*
929  * Mark all the subtransactions as observed.
930  *
931  * NOTE: This will fail if the subxid contains too many previously
932  * unobserved xids to fit into known-assigned-xids. That shouldn't happen
933  * as the code stands, because xid-assignment records should never contain
934  * more than PGPROC_MAX_CACHED_SUBXIDS entries.
935  */
937 
938  /*
939  * Notice that we update pg_subtrans with the top-level xid, rather than
940  * the parent xid. This is a difference between normal processing and
941  * recovery, yet is still correct in all cases. The reason is that
942  * subtransaction commit is not marked in clog until commit processing, so
943  * all aborted subtransactions have already been clearly marked in clog.
944  * As a result we are able to refer directly to the top-level
945  * transaction's state rather than skipping through all the intermediate
946  * states in the subtransaction tree. This should be the first time we
947  * have attempted to SubTransSetParent().
948  */
949  for (i = 0; i < nsubxids; i++)
950  SubTransSetParent(subxids[i], topxid);
951 
952  /* KnownAssignedXids isn't maintained yet, so we're done for now */
954  return;
955 
956  /*
957  * Uses same locking as transaction commit
958  */
959  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
960 
961  /*
962  * Remove subxids from known-assigned-xacts.
963  */
965 
966  /*
967  * Advance lastOverflowedXid to be at least the last of these subxids.
968  */
969  if (TransactionIdPrecedes(procArray->lastOverflowedXid, max_xid))
970  procArray->lastOverflowedXid = max_xid;
971 
972  LWLockRelease(ProcArrayLock);
973 }
974 
975 /*
976  * TransactionIdIsInProgress -- is given transaction running in some backend
977  *
978  * Aside from some shortcuts such as checking RecentXmin and our own Xid,
979  * there are four possibilities for finding a running transaction:
980  *
981  * 1. The given Xid is a main transaction Id. We will find this out cheaply
982  * by looking at the PGXACT struct for each backend.
983  *
984  * 2. The given Xid is one of the cached subxact Xids in the PGPROC array.
985  * We can find this out cheaply too.
986  *
987  * 3. In Hot Standby mode, we must search the KnownAssignedXids list to see
988  * if the Xid is running on the master.
989  *
990  * 4. Search the SubTrans tree to find the Xid's topmost parent, and then see
991  * if that is running according to PGXACT or KnownAssignedXids. This is the
992  * slowest way, but sadly it has to be done always if the others failed,
993  * unless we see that the cached subxact sets are complete (none have
994  * overflowed).
995  *
996  * ProcArrayLock has to be held while we do 1, 2, 3. If we save the top Xids
997  * while doing 1 and 3, we can release the ProcArrayLock while we do 4.
998  * This buys back some concurrency (and we can't retrieve the main Xids from
999  * PGXACT again anyway; see GetNewTransactionId).
1000  */
1001 bool
1003 {
1004  static TransactionId *xids = NULL;
1005  int nxids = 0;
1006  ProcArrayStruct *arrayP = procArray;
1007  TransactionId topxid;
1008  int i,
1009  j;
1010 
1011  /*
1012  * Don't bother checking a transaction older than RecentXmin; it could not
1013  * possibly still be running. (Note: in particular, this guarantees that
1014  * we reject InvalidTransactionId, FrozenTransactionId, etc as not
1015  * running.)
1016  */
1018  {
1020  return false;
1021  }
1022 
1023  /*
1024  * We may have just checked the status of this transaction, so if it is
1025  * already known to be completed, we can fall out without any access to
1026  * shared memory.
1027  */
1029  {
1031  return false;
1032  }
1033 
1034  /*
1035  * Also, we can handle our own transaction (and subtransactions) without
1036  * any access to shared memory.
1037  */
1039  {
1041  return true;
1042  }
1043 
1044  /*
1045  * If first time through, get workspace to remember main XIDs in. We
1046  * malloc it permanently to avoid repeated palloc/pfree overhead.
1047  */
1048  if (xids == NULL)
1049  {
1050  /*
1051  * In hot standby mode, reserve enough space to hold all xids in the
1052  * known-assigned list. If we later finish recovery, we no longer need
1053  * the bigger array, but we don't bother to shrink it.
1054  */
1055  int maxxids = RecoveryInProgress() ? TOTAL_MAX_CACHED_SUBXIDS : arrayP->maxProcs;
1056 
1057  xids = (TransactionId *) malloc(maxxids * sizeof(TransactionId));
1058  if (xids == NULL)
1059  ereport(ERROR,
1060  (errcode(ERRCODE_OUT_OF_MEMORY),
1061  errmsg("out of memory")));
1062  }
1063 
1064  LWLockAcquire(ProcArrayLock, LW_SHARED);
1065 
1066  /*
1067  * Now that we have the lock, we can check latestCompletedXid; if the
1068  * target Xid is after that, it's surely still running.
1069  */
1071  {
1072  LWLockRelease(ProcArrayLock);
1074  return true;
1075  }
1076 
1077  /* No shortcuts, gotta grovel through the array */
1078  for (i = 0; i < arrayP->numProcs; i++)
1079  {
1080  int pgprocno = arrayP->pgprocnos[i];
1081  PGPROC *proc = &allProcs[pgprocno];
1082  PGXACT *pgxact = &allPgXact[pgprocno];
1083  TransactionId pxid;
1084  int pxids;
1085 
1086  /* Ignore my own proc --- dealt with it above */
1087  if (proc == MyProc)
1088  continue;
1089 
1090  /* Fetch xid just once - see GetNewTransactionId */
1091  pxid = UINT32_ACCESS_ONCE(pgxact->xid);
1092 
1093  if (!TransactionIdIsValid(pxid))
1094  continue;
1095 
1096  /*
1097  * Step 1: check the main Xid
1098  */
1099  if (TransactionIdEquals(pxid, xid))
1100  {
1101  LWLockRelease(ProcArrayLock);
1103  return true;
1104  }
1105 
1106  /*
1107  * We can ignore main Xids that are younger than the target Xid, since
1108  * the target could not possibly be their child.
1109  */
1110  if (TransactionIdPrecedes(xid, pxid))
1111  continue;
1112 
1113  /*
1114  * Step 2: check the cached child-Xids arrays
1115  */
1116  pxids = pgxact->nxids;
1117  pg_read_barrier(); /* pairs with barrier in GetNewTransactionId() */
1118  for (j = pxids - 1; j >= 0; j--)
1119  {
1120  /* Fetch xid just once - see GetNewTransactionId */
1121  TransactionId cxid = UINT32_ACCESS_ONCE(proc->subxids.xids[j]);
1122 
1123  if (TransactionIdEquals(cxid, xid))
1124  {
1125  LWLockRelease(ProcArrayLock);
1127  return true;
1128  }
1129  }
1130 
1131  /*
1132  * Save the main Xid for step 4. We only need to remember main Xids
1133  * that have uncached children. (Note: there is no race condition
1134  * here because the overflowed flag cannot be cleared, only set, while
1135  * we hold ProcArrayLock. So we can't miss an Xid that we need to
1136  * worry about.)
1137  */
1138  if (pgxact->overflowed)
1139  xids[nxids++] = pxid;
1140  }
1141 
1142  /*
1143  * Step 3: in hot standby mode, check the known-assigned-xids list. XIDs
1144  * in the list must be treated as running.
1145  */
1146  if (RecoveryInProgress())
1147  {
1148  /* none of the PGXACT entries should have XIDs in hot standby mode */
1149  Assert(nxids == 0);
1150 
1151  if (KnownAssignedXidExists(xid))
1152  {
1153  LWLockRelease(ProcArrayLock);
1155  return true;
1156  }
1157 
1158  /*
1159  * If the KnownAssignedXids overflowed, we have to check pg_subtrans
1160  * too. Fetch all xids from KnownAssignedXids that are lower than
1161  * xid, since if xid is a subtransaction its parent will always have a
1162  * lower value. Note we will collect both main and subXIDs here, but
1163  * there's no help for it.
1164  */
1165  if (TransactionIdPrecedesOrEquals(xid, procArray->lastOverflowedXid))
1166  nxids = KnownAssignedXidsGet(xids, xid);
1167  }
1168 
1169  LWLockRelease(ProcArrayLock);
1170 
1171  /*
1172  * If none of the relevant caches overflowed, we know the Xid is not
1173  * running without even looking at pg_subtrans.
1174  */
1175  if (nxids == 0)
1176  {
1178  return false;
1179  }
1180 
1181  /*
1182  * Step 4: have to check pg_subtrans.
1183  *
1184  * At this point, we know it's either a subtransaction of one of the Xids
1185  * in xids[], or it's not running. If it's an already-failed
1186  * subtransaction, we want to say "not running" even though its parent may
1187  * still be running. So first, check pg_xact to see if it's been aborted.
1188  */
1190 
1191  if (TransactionIdDidAbort(xid))
1192  return false;
1193 
1194  /*
1195  * It isn't aborted, so check whether the transaction tree it belongs to
1196  * is still running (or, more precisely, whether it was running when we
1197  * held ProcArrayLock).
1198  */
1199  topxid = SubTransGetTopmostTransaction(xid);
1200  Assert(TransactionIdIsValid(topxid));
1201  if (!TransactionIdEquals(topxid, xid))
1202  {
1203  for (i = 0; i < nxids; i++)
1204  {
1205  if (TransactionIdEquals(xids[i], topxid))
1206  return true;
1207  }
1208  }
1209 
1210  return false;
1211 }
1212 
1213 /*
1214  * TransactionIdIsActive -- is xid the top-level XID of an active backend?
1215  *
1216  * This differs from TransactionIdIsInProgress in that it ignores prepared
1217  * transactions, as well as transactions running on the master if we're in
1218  * hot standby. Also, we ignore subtransactions since that's not needed
1219  * for current uses.
1220  */
1221 bool
1223 {
1224  bool result = false;
1225  ProcArrayStruct *arrayP = procArray;
1226  int i;
1227 
1228  /*
1229  * Don't bother checking a transaction older than RecentXmin; it could not
1230  * possibly still be running.
1231  */
1233  return false;
1234 
1235  LWLockAcquire(ProcArrayLock, LW_SHARED);
1236 
1237  for (i = 0; i < arrayP->numProcs; i++)
1238  {
1239  int pgprocno = arrayP->pgprocnos[i];
1240  PGPROC *proc = &allProcs[pgprocno];
1241  PGXACT *pgxact = &allPgXact[pgprocno];
1242  TransactionId pxid;
1243 
1244  /* Fetch xid just once - see GetNewTransactionId */
1245  pxid = UINT32_ACCESS_ONCE(pgxact->xid);
1246 
1247  if (!TransactionIdIsValid(pxid))
1248  continue;
1249 
1250  if (proc->pid == 0)
1251  continue; /* ignore prepared transactions */
1252 
1253  if (TransactionIdEquals(pxid, xid))
1254  {
1255  result = true;
1256  break;
1257  }
1258  }
1259 
1260  LWLockRelease(ProcArrayLock);
1261 
1262  return result;
1263 }
1264 
1265 
1266 /*
1267  * GetOldestXmin -- returns oldest transaction that was running
1268  * when any current transaction was started.
1269  *
1270  * If rel is NULL or a shared relation, all backends are considered, otherwise
1271  * only backends running in this database are considered.
1272  *
1273  * The flags are used to ignore the backends in calculation when any of the
1274  * corresponding flags is set. Typically, if you want to ignore ones with
1275  * PROC_IN_VACUUM flag, you can use PROCARRAY_FLAGS_VACUUM.
1276  *
1277  * PROCARRAY_SLOTS_XMIN causes GetOldestXmin to ignore the xmin and
1278  * catalog_xmin of any replication slots that exist in the system when
1279  * calculating the oldest xmin.
1280  *
1281  * This is used by VACUUM to decide which deleted tuples must be preserved in
1282  * the passed in table. For shared relations backends in all databases must be
1283  * considered, but for non-shared relations that's not required, since only
1284  * backends in my own database could ever see the tuples in them. Also, we can
1285  * ignore concurrently running lazy VACUUMs because (a) they must be working
1286  * on other tables, and (b) they don't need to do snapshot-based lookups.
1287  *
1288  * This is also used to determine where to truncate pg_subtrans. For that
1289  * backends in all databases have to be considered, so rel = NULL has to be
1290  * passed in.
1291  *
1292  * Note: we include all currently running xids in the set of considered xids.
1293  * This ensures that if a just-started xact has not yet set its snapshot,
1294  * when it does set the snapshot it cannot set xmin less than what we compute.
1295  * See notes in src/backend/access/transam/README.
1296  *
1297  * Note: despite the above, it's possible for the calculated value to move
1298  * backwards on repeated calls. The calculated value is conservative, so that
1299  * anything older is definitely not considered as running by anyone anymore,
1300  * but the exact value calculated depends on a number of things. For example,
1301  * if rel = NULL and there are no transactions running in the current
1302  * database, GetOldestXmin() returns latestCompletedXid. If a transaction
1303  * begins after that, its xmin will include in-progress transactions in other
1304  * databases that started earlier, so another call will return a lower value.
1305  * Nonetheless it is safe to vacuum a table in the current database with the
1306  * first result. There are also replication-related effects: a walsender
1307  * process can set its xmin based on transactions that are no longer running
1308  * in the master but are still being replayed on the standby, thus possibly
1309  * making the GetOldestXmin reading go backwards. In this case there is a
1310  * possibility that we lose data that the standby would like to have, but
1311  * unless the standby uses a replication slot to make its xmin persistent
1312  * there is little we can do about that --- data is only protected if the
1313  * walsender runs continuously while queries are executed on the standby.
1314  * (The Hot Standby code deals with such cases by failing standby queries
1315  * that needed to access already-removed data, so there's no integrity bug.)
1316  * The return value is also adjusted with vacuum_defer_cleanup_age, so
1317  * increasing that setting on the fly is another easy way to make
1318  * GetOldestXmin() move backwards, with no consequences for data integrity.
1319  */
1321 GetOldestXmin(Relation rel, int flags)
1322 {
1323  ProcArrayStruct *arrayP = procArray;
1324  TransactionId result;
1325  int index;
1326  bool allDbs;
1327 
1330 
1331  /*
1332  * If we're not computing a relation specific limit, or if a shared
1333  * relation has been passed in, backends in all databases have to be
1334  * considered.
1335  */
1336  allDbs = rel == NULL || rel->rd_rel->relisshared;
1337 
1338  /* Cannot look for individual databases during recovery */
1339  Assert(allDbs || !RecoveryInProgress());
1340 
1341  LWLockAcquire(ProcArrayLock, LW_SHARED);
1342 
1343  /*
1344  * We initialize the MIN() calculation with latestCompletedXid + 1. This
1345  * is a lower bound for the XIDs that might appear in the ProcArray later,
1346  * and so protects us against overestimating the result due to future
1347  * additions.
1348  */
1350  Assert(TransactionIdIsNormal(result));
1351  TransactionIdAdvance(result);
1352 
1353  for (index = 0; index < arrayP->numProcs; index++)
1354  {
1355  int pgprocno = arrayP->pgprocnos[index];
1356  PGPROC *proc = &allProcs[pgprocno];
1357  PGXACT *pgxact = &allPgXact[pgprocno];
1358 
1359  if (pgxact->vacuumFlags & (flags & PROCARRAY_PROC_FLAGS_MASK))
1360  continue;
1361 
1362  if (allDbs ||
1363  proc->databaseId == MyDatabaseId ||
1364  proc->databaseId == 0) /* always include WalSender */
1365  {
1366  /* Fetch xid just once - see GetNewTransactionId */
1367  TransactionId xid = UINT32_ACCESS_ONCE(pgxact->xid);
1368 
1369  /* First consider the transaction's own Xid, if any */
1370  if (TransactionIdIsNormal(xid) &&
1371  TransactionIdPrecedes(xid, result))
1372  result = xid;
1373 
1374  /*
1375  * Also consider the transaction's Xmin, if set.
1376  *
1377  * We must check both Xid and Xmin because a transaction might
1378  * have an Xmin but not (yet) an Xid; conversely, if it has an
1379  * Xid, that could determine some not-yet-set Xmin.
1380  */
1381  xid = UINT32_ACCESS_ONCE(pgxact->xmin);
1382  if (TransactionIdIsNormal(xid) &&
1383  TransactionIdPrecedes(xid, result))
1384  result = xid;
1385  }
1386  }
1387 
1388  /*
1389  * Fetch into local variable while ProcArrayLock is held - the
1390  * LWLockRelease below is a barrier, ensuring this happens inside the
1391  * lock.
1392  */
1393  replication_slot_xmin = procArray->replication_slot_xmin;
1394  replication_slot_catalog_xmin = procArray->replication_slot_catalog_xmin;
1395 
1396  if (RecoveryInProgress())
1397  {
1398  /*
1399  * Check to see whether KnownAssignedXids contains an xid value older
1400  * than the main procarray.
1401  */
1403 
1404  LWLockRelease(ProcArrayLock);
1405 
1406  if (TransactionIdIsNormal(kaxmin) &&
1407  TransactionIdPrecedes(kaxmin, result))
1408  result = kaxmin;
1409  }
1410  else
1411  {
1412  /*
1413  * No other information needed, so release the lock immediately.
1414  */
1415  LWLockRelease(ProcArrayLock);
1416 
1417  /*
1418  * Compute the cutoff XID by subtracting vacuum_defer_cleanup_age,
1419  * being careful not to generate a "permanent" XID.
1420  *
1421  * vacuum_defer_cleanup_age provides some additional "slop" for the
1422  * benefit of hot standby queries on standby servers. This is quick
1423  * and dirty, and perhaps not all that useful unless the master has a
1424  * predictable transaction rate, but it offers some protection when
1425  * there's no walsender connection. Note that we are assuming
1426  * vacuum_defer_cleanup_age isn't large enough to cause wraparound ---
1427  * so guc.c should limit it to no more than the xidStopLimit threshold
1428  * in varsup.c. Also note that we intentionally don't apply
1429  * vacuum_defer_cleanup_age on standby servers.
1430  */
1431  result -= vacuum_defer_cleanup_age;
1432  if (!TransactionIdIsNormal(result))
1433  result = FirstNormalTransactionId;
1434  }
1435 
1436  /*
1437  * Check whether there are replication slots requiring an older xmin.
1438  */
1439  if (!(flags & PROCARRAY_SLOTS_XMIN) &&
1440  TransactionIdIsValid(replication_slot_xmin) &&
1441  NormalTransactionIdPrecedes(replication_slot_xmin, result))
1442  result = replication_slot_xmin;
1443 
1444  /*
1445  * After locks have been released and defer_cleanup_age has been applied,
1446  * check whether we need to back up further to make logical decoding
1447  * possible. We need to do so if we're computing the global limit (rel =
1448  * NULL) or if the passed relation is a catalog relation of some kind.
1449  */
1450  if (!(flags & PROCARRAY_SLOTS_XMIN) &&
1451  (rel == NULL ||
1453  TransactionIdIsValid(replication_slot_catalog_xmin) &&
1454  NormalTransactionIdPrecedes(replication_slot_catalog_xmin, result))
1456 
1457  return result;
1458 }
1459 
1460 /*
1461  * GetMaxSnapshotXidCount -- get max size for snapshot XID array
1462  *
1463  * We have to export this for use by snapmgr.c.
1464  */
1465 int
1467 {
1468  return procArray->maxProcs;
1469 }
1470 
1471 /*
1472  * GetMaxSnapshotSubxidCount -- get max size for snapshot sub-XID array
1473  *
1474  * We have to export this for use by snapmgr.c.
1475  */
1476 int
1478 {
1479  return TOTAL_MAX_CACHED_SUBXIDS;
1480 }
1481 
1482 /*
1483  * GetSnapshotData -- returns information about running transactions.
1484  *
1485  * The returned snapshot includes xmin (lowest still-running xact ID),
1486  * xmax (highest completed xact ID + 1), and a list of running xact IDs
1487  * in the range xmin <= xid < xmax. It is used as follows:
1488  * All xact IDs < xmin are considered finished.
1489  * All xact IDs >= xmax are considered still running.
1490  * For an xact ID xmin <= xid < xmax, consult list to see whether
1491  * it is considered running or not.
1492  * This ensures that the set of transactions seen as "running" by the
1493  * current xact will not change after it takes the snapshot.
1494  *
1495  * All running top-level XIDs are included in the snapshot, except for lazy
1496  * VACUUM processes. We also try to include running subtransaction XIDs,
1497  * but since PGPROC has only a limited cache area for subxact XIDs, full
1498  * information may not be available. If we find any overflowed subxid arrays,
1499  * we have to mark the snapshot's subxid data as overflowed, and extra work
1500  * *may* need to be done to determine what's running (see XidInMVCCSnapshot()
1501  * in tqual.c).
1502  *
1503  * We also update the following backend-global variables:
1504  * TransactionXmin: the oldest xmin of any snapshot in use in the
1505  * current transaction (this is the same as MyPgXact->xmin).
1506  * RecentXmin: the xmin computed for the most recent snapshot. XIDs
1507  * older than this are known not running any more.
1508  * RecentGlobalXmin: the global xmin (oldest TransactionXmin across all
1509  * running transactions, except those running LAZY VACUUM). This is
1510  * the same computation done by
1511  * GetOldestXmin(NULL, PROCARRAY_FLAGS_VACUUM).
1512  * RecentGlobalDataXmin: the global xmin for non-catalog tables
1513  * >= RecentGlobalXmin
1514  *
1515  * Note: this function should probably not be called with an argument that's
1516  * not statically allocated (see xip allocation below).
1517  */
1518 Snapshot
1520 {
1521  ProcArrayStruct *arrayP = procArray;
1522  TransactionId xmin;
1523  TransactionId xmax;
1524  TransactionId globalxmin;
1525  int index;
1526  int count = 0;
1527  int subcount = 0;
1528  bool suboverflowed = false;
1531 
1532  Assert(snapshot != NULL);
1533 
1534  /*
1535  * Allocating space for maxProcs xids is usually overkill; numProcs would
1536  * be sufficient. But it seems better to do the malloc while not holding
1537  * the lock, so we can't look at numProcs. Likewise, we allocate much
1538  * more subxip storage than is probably needed.
1539  *
1540  * This does open a possibility for avoiding repeated malloc/free: since
1541  * maxProcs does not change at runtime, we can simply reuse the previous
1542  * xip arrays if any. (This relies on the fact that all callers pass
1543  * static SnapshotData structs.)
1544  */
1545  if (snapshot->xip == NULL)
1546  {
1547  /*
1548  * First call for this snapshot. Snapshot is same size whether or not
1549  * we are in recovery, see later comments.
1550  */
1551  snapshot->xip = (TransactionId *)
1553  if (snapshot->xip == NULL)
1554  ereport(ERROR,
1555  (errcode(ERRCODE_OUT_OF_MEMORY),
1556  errmsg("out of memory")));
1557  Assert(snapshot->subxip == NULL);
1558  snapshot->subxip = (TransactionId *)
1560  if (snapshot->subxip == NULL)
1561  ereport(ERROR,
1562  (errcode(ERRCODE_OUT_OF_MEMORY),
1563  errmsg("out of memory")));
1564  }
1565 
1566  /*
1567  * It is sufficient to get shared lock on ProcArrayLock, even if we are
1568  * going to set MyPgXact->xmin.
1569  */
1570  LWLockAcquire(ProcArrayLock, LW_SHARED);
1571 
1572  /* xmax is always latestCompletedXid + 1 */
1575  TransactionIdAdvance(xmax);
1576 
1577  /* initialize xmin calculation with xmax */
1578  globalxmin = xmin = xmax;
1579 
1581 
1582  if (!snapshot->takenDuringRecovery)
1583  {
1584  int *pgprocnos = arrayP->pgprocnos;
1585  int numProcs;
1586 
1587  /*
1588  * Spin over procArray checking xid, xmin, and subxids. The goal is
1589  * to gather all active xids, find the lowest xmin, and try to record
1590  * subxids.
1591  */
1592  numProcs = arrayP->numProcs;
1593  for (index = 0; index < numProcs; index++)
1594  {
1595  int pgprocno = pgprocnos[index];
1596  PGXACT *pgxact = &allPgXact[pgprocno];
1597  TransactionId xid;
1598 
1599  /*
1600  * Skip over backends doing logical decoding which manages xmin
1601  * separately (check below) and ones running LAZY VACUUM.
1602  */
1603  if (pgxact->vacuumFlags &
1605  continue;
1606 
1607  /* Update globalxmin to be the smallest valid xmin */
1608  xid = UINT32_ACCESS_ONCE(pgxact->xmin);
1609  if (TransactionIdIsNormal(xid) &&
1610  NormalTransactionIdPrecedes(xid, globalxmin))
1611  globalxmin = xid;
1612 
1613  /* Fetch xid just once - see GetNewTransactionId */
1614  xid = UINT32_ACCESS_ONCE(pgxact->xid);
1615 
1616  /*
1617  * If the transaction has no XID assigned, we can skip it; it
1618  * won't have sub-XIDs either. If the XID is >= xmax, we can also
1619  * skip it; such transactions will be treated as running anyway
1620  * (and any sub-XIDs will also be >= xmax).
1621  */
1622  if (!TransactionIdIsNormal(xid)
1623  || !NormalTransactionIdPrecedes(xid, xmax))
1624  continue;
1625 
1626  /*
1627  * We don't include our own XIDs (if any) in the snapshot, but we
1628  * must include them in xmin.
1629  */
1630  if (NormalTransactionIdPrecedes(xid, xmin))
1631  xmin = xid;
1632  if (pgxact == MyPgXact)
1633  continue;
1634 
1635  /* Add XID to snapshot. */
1636  snapshot->xip[count++] = xid;
1637 
1638  /*
1639  * Save subtransaction XIDs if possible (if we've already
1640  * overflowed, there's no point). Note that the subxact XIDs must
1641  * be later than their parent, so no need to check them against
1642  * xmin. We could filter against xmax, but it seems better not to
1643  * do that much work while holding the ProcArrayLock.
1644  *
1645  * The other backend can add more subxids concurrently, but cannot
1646  * remove any. Hence it's important to fetch nxids just once.
1647  * Should be safe to use memcpy, though. (We needn't worry about
1648  * missing any xids added concurrently, because they must postdate
1649  * xmax.)
1650  *
1651  * Again, our own XIDs are not included in the snapshot.
1652  */
1653  if (!suboverflowed)
1654  {
1655  if (pgxact->overflowed)
1656  suboverflowed = true;
1657  else
1658  {
1659  int nxids = pgxact->nxids;
1660 
1661  if (nxids > 0)
1662  {
1663  PGPROC *proc = &allProcs[pgprocno];
1664 
1665  pg_read_barrier(); /* pairs with GetNewTransactionId */
1666 
1667  memcpy(snapshot->subxip + subcount,
1668  (void *) proc->subxids.xids,
1669  nxids * sizeof(TransactionId));
1670  subcount += nxids;
1671  }
1672  }
1673  }
1674  }
1675  }
1676  else
1677  {
1678  /*
1679  * We're in hot standby, so get XIDs from KnownAssignedXids.
1680  *
1681  * We store all xids directly into subxip[]. Here's why:
1682  *
1683  * In recovery we don't know which xids are top-level and which are
1684  * subxacts, a design choice that greatly simplifies xid processing.
1685  *
1686  * It seems like we would want to try to put xids into xip[] only, but
1687  * that is fairly small. We would either need to make that bigger or
1688  * to increase the rate at which we WAL-log xid assignment; neither is
1689  * an appealing choice.
1690  *
1691  * We could try to store xids into xip[] first and then into subxip[]
1692  * if there are too many xids. That only works if the snapshot doesn't
1693  * overflow because we do not search subxip[] in that case. A simpler
1694  * way is to just store all xids in the subxact array because this is
1695  * by far the bigger array. We just leave the xip array empty.
1696  *
1697  * Either way we need to change the way XidInMVCCSnapshot() works
1698  * depending upon when the snapshot was taken, or change normal
1699  * snapshot processing so it matches.
1700  *
1701  * Note: It is possible for recovery to end before we finish taking
1702  * the snapshot, and for newly assigned transaction ids to be added to
1703  * the ProcArray. xmax cannot change while we hold ProcArrayLock, so
1704  * those newly added transaction ids would be filtered away, so we
1705  * need not be concerned about them.
1706  */
1707  subcount = KnownAssignedXidsGetAndSetXmin(snapshot->subxip, &xmin,
1708  xmax);
1709 
1710  if (TransactionIdPrecedesOrEquals(xmin, procArray->lastOverflowedXid))
1711  suboverflowed = true;
1712  }
1713 
1714 
1715  /*
1716  * Fetch into local variable while ProcArrayLock is held - the
1717  * LWLockRelease below is a barrier, ensuring this happens inside the
1718  * lock.
1719  */
1720  replication_slot_xmin = procArray->replication_slot_xmin;
1721  replication_slot_catalog_xmin = procArray->replication_slot_catalog_xmin;
1722 
1724  MyPgXact->xmin = TransactionXmin = xmin;
1725 
1726  LWLockRelease(ProcArrayLock);
1727 
1728  /*
1729  * Update globalxmin to include actual process xids. This is a slightly
1730  * different way of computing it than GetOldestXmin uses, but should give
1731  * the same result.
1732  */
1733  if (TransactionIdPrecedes(xmin, globalxmin))
1734  globalxmin = xmin;
1735 
1736  /* Update global variables too */
1740 
1741  /* Check whether there's a replication slot requiring an older xmin. */
1742  if (TransactionIdIsValid(replication_slot_xmin) &&
1743  NormalTransactionIdPrecedes(replication_slot_xmin, RecentGlobalXmin))
1745 
1746  /* Non-catalog tables can be vacuumed if older than this xid */
1748 
1749  /*
1750  * Check whether there's a replication slot requiring an older catalog
1751  * xmin.
1752  */
1753  if (TransactionIdIsNormal(replication_slot_catalog_xmin) &&
1754  NormalTransactionIdPrecedes(replication_slot_catalog_xmin, RecentGlobalXmin))
1756 
1757  RecentXmin = xmin;
1758 
1759  snapshot->xmin = xmin;
1760  snapshot->xmax = xmax;
1761  snapshot->xcnt = count;
1762  snapshot->subxcnt = subcount;
1763  snapshot->suboverflowed = suboverflowed;
1764 
1765  snapshot->curcid = GetCurrentCommandId(false);
1766 
1767  /*
1768  * This is a new snapshot, so set both refcounts are zero, and mark it as
1769  * not copied in persistent memory.
1770  */
1771  snapshot->active_count = 0;
1772  snapshot->regd_count = 0;
1773  snapshot->copied = false;
1774 
1775  if (old_snapshot_threshold < 0)
1776  {
1777  /*
1778  * If not using "snapshot too old" feature, fill related fields with
1779  * dummy values that don't require any locking.
1780  */
1781  snapshot->lsn = InvalidXLogRecPtr;
1782  snapshot->whenTaken = 0;
1783  }
1784  else
1785  {
1786  /*
1787  * Capture the current time and WAL stream location in case this
1788  * snapshot becomes old enough to need to fall back on the special
1789  * "old snapshot" logic.
1790  */
1791  snapshot->lsn = GetXLogInsertRecPtr();
1792  snapshot->whenTaken = GetSnapshotCurrentTimestamp();
1793  MaintainOldSnapshotTimeMapping(snapshot->whenTaken, xmin);
1794  }
1795 
1796  return snapshot;
1797 }
1798 
1799 /*
1800  * ProcArrayInstallImportedXmin -- install imported xmin into MyPgXact->xmin
1801  *
1802  * This is called when installing a snapshot imported from another
1803  * transaction. To ensure that OldestXmin doesn't go backwards, we must
1804  * check that the source transaction is still running, and we'd better do
1805  * that atomically with installing the new xmin.
1806  *
1807  * Returns true if successful, false if source xact is no longer running.
1808  */
1809 bool
1811  VirtualTransactionId *sourcevxid)
1812 {
1813  bool result = false;
1814  ProcArrayStruct *arrayP = procArray;
1815  int index;
1816 
1818  if (!sourcevxid)
1819  return false;
1820 
1821  /* Get lock so source xact can't end while we're doing this */
1822  LWLockAcquire(ProcArrayLock, LW_SHARED);
1823 
1824  for (index = 0; index < arrayP->numProcs; index++)
1825  {
1826  int pgprocno = arrayP->pgprocnos[index];
1827  PGPROC *proc = &allProcs[pgprocno];
1828  PGXACT *pgxact = &allPgXact[pgprocno];
1829  TransactionId xid;
1830 
1831  /* Ignore procs running LAZY VACUUM */
1832  if (pgxact->vacuumFlags & PROC_IN_VACUUM)
1833  continue;
1834 
1835  /* We are only interested in the specific virtual transaction. */
1836  if (proc->backendId != sourcevxid->backendId)
1837  continue;
1838  if (proc->lxid != sourcevxid->localTransactionId)
1839  continue;
1840 
1841  /*
1842  * We check the transaction's database ID for paranoia's sake: if it's
1843  * in another DB then its xmin does not cover us. Caller should have
1844  * detected this already, so we just treat any funny cases as
1845  * "transaction not found".
1846  */
1847  if (proc->databaseId != MyDatabaseId)
1848  continue;
1849 
1850  /*
1851  * Likewise, let's just make real sure its xmin does cover us.
1852  */
1853  xid = UINT32_ACCESS_ONCE(pgxact->xmin);
1854  if (!TransactionIdIsNormal(xid) ||
1855  !TransactionIdPrecedesOrEquals(xid, xmin))
1856  continue;
1857 
1858  /*
1859  * We're good. Install the new xmin. As in GetSnapshotData, set
1860  * TransactionXmin too. (Note that because snapmgr.c called
1861  * GetSnapshotData first, we'll be overwriting a valid xmin here, so
1862  * we don't check that.)
1863  */
1864  MyPgXact->xmin = TransactionXmin = xmin;
1865 
1866  result = true;
1867  break;
1868  }
1869 
1870  LWLockRelease(ProcArrayLock);
1871 
1872  return result;
1873 }
1874 
1875 /*
1876  * ProcArrayInstallRestoredXmin -- install restored xmin into MyPgXact->xmin
1877  *
1878  * This is like ProcArrayInstallImportedXmin, but we have a pointer to the
1879  * PGPROC of the transaction from which we imported the snapshot, rather than
1880  * an XID.
1881  *
1882  * Returns true if successful, false if source xact is no longer running.
1883  */
1884 bool
1886 {
1887  bool result = false;
1888  TransactionId xid;
1889  PGXACT *pgxact;
1890 
1892  Assert(proc != NULL);
1893 
1894  /* Get lock so source xact can't end while we're doing this */
1895  LWLockAcquire(ProcArrayLock, LW_SHARED);
1896 
1897  pgxact = &allPgXact[proc->pgprocno];
1898 
1899  /*
1900  * Be certain that the referenced PGPROC has an advertised xmin which is
1901  * no later than the one we're installing, so that the system-wide xmin
1902  * can't go backwards. Also, make sure it's running in the same database,
1903  * so that the per-database xmin cannot go backwards.
1904  */
1905  xid = UINT32_ACCESS_ONCE(pgxact->xmin);
1906  if (proc->databaseId == MyDatabaseId &&
1907  TransactionIdIsNormal(xid) &&
1908  TransactionIdPrecedesOrEquals(xid, xmin))
1909  {
1910  MyPgXact->xmin = TransactionXmin = xmin;
1911  result = true;
1912  }
1913 
1914  LWLockRelease(ProcArrayLock);
1915 
1916  return result;
1917 }
1918 
1919 /*
1920  * GetRunningTransactionData -- returns information about running transactions.
1921  *
1922  * Similar to GetSnapshotData but returns more information. We include
1923  * all PGXACTs with an assigned TransactionId, even VACUUM processes and
1924  * prepared transactions.
1925  *
1926  * We acquire XidGenLock and ProcArrayLock, but the caller is responsible for
1927  * releasing them. Acquiring XidGenLock ensures that no new XIDs enter the proc
1928  * array until the caller has WAL-logged this snapshot, and releases the
1929  * lock. Acquiring ProcArrayLock ensures that no transactions commit until the
1930  * lock is released.
1931  *
1932  * The returned data structure is statically allocated; caller should not
1933  * modify it, and must not assume it is valid past the next call.
1934  *
1935  * This is never executed during recovery so there is no need to look at
1936  * KnownAssignedXids.
1937  *
1938  * Dummy PGXACTs from prepared transaction are included, meaning that this
1939  * may return entries with duplicated TransactionId values coming from
1940  * transaction finishing to prepare. Nothing is done about duplicated
1941  * entries here to not hold on ProcArrayLock more than necessary.
1942  *
1943  * We don't worry about updating other counters, we want to keep this as
1944  * simple as possible and leave GetSnapshotData() as the primary code for
1945  * that bookkeeping.
1946  *
1947  * Note that if any transaction has overflowed its cached subtransactions
1948  * then there is no real need include any subtransactions.
1949  */
1952 {
1953  /* result workspace */
1954  static RunningTransactionsData CurrentRunningXactsData;
1955 
1956  ProcArrayStruct *arrayP = procArray;
1957  RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData;
1958  TransactionId latestCompletedXid;
1959  TransactionId oldestRunningXid;
1960  TransactionId *xids;
1961  int index;
1962  int count;
1963  int subcount;
1964  bool suboverflowed;
1965 
1967 
1968  /*
1969  * Allocating space for maxProcs xids is usually overkill; numProcs would
1970  * be sufficient. But it seems better to do the malloc while not holding
1971  * the lock, so we can't look at numProcs. Likewise, we allocate much
1972  * more subxip storage than is probably needed.
1973  *
1974  * Should only be allocated in bgwriter, since only ever executed during
1975  * checkpoints.
1976  */
1977  if (CurrentRunningXacts->xids == NULL)
1978  {
1979  /*
1980  * First call
1981  */
1982  CurrentRunningXacts->xids = (TransactionId *)
1984  if (CurrentRunningXacts->xids == NULL)
1985  ereport(ERROR,
1986  (errcode(ERRCODE_OUT_OF_MEMORY),
1987  errmsg("out of memory")));
1988  }
1989 
1990  xids = CurrentRunningXacts->xids;
1991 
1992  count = subcount = 0;
1993  suboverflowed = false;
1994 
1995  /*
1996  * Ensure that no xids enter or leave the procarray while we obtain
1997  * snapshot.
1998  */
1999  LWLockAcquire(ProcArrayLock, LW_SHARED);
2000  LWLockAcquire(XidGenLock, LW_SHARED);
2001 
2002  latestCompletedXid = ShmemVariableCache->latestCompletedXid;
2003 
2004  oldestRunningXid = ShmemVariableCache->nextXid;
2005 
2006  /*
2007  * Spin over procArray collecting all xids
2008  */
2009  for (index = 0; index < arrayP->numProcs; index++)
2010  {
2011  int pgprocno = arrayP->pgprocnos[index];
2012  PGXACT *pgxact = &allPgXact[pgprocno];
2013  TransactionId xid;
2014 
2015  /* Fetch xid just once - see GetNewTransactionId */
2016  xid = UINT32_ACCESS_ONCE(pgxact->xid);
2017 
2018  /*
2019  * We don't need to store transactions that don't have a TransactionId
2020  * yet because they will not show as running on a standby server.
2021  */
2022  if (!TransactionIdIsValid(xid))
2023  continue;
2024 
2025  /*
2026  * Be careful not to exclude any xids before calculating the values of
2027  * oldestRunningXid and suboverflowed, since these are used to clean
2028  * up transaction information held on standbys.
2029  */
2030  if (TransactionIdPrecedes(xid, oldestRunningXid))
2031  oldestRunningXid = xid;
2032 
2033  if (pgxact->overflowed)
2034  suboverflowed = true;
2035 
2036  /*
2037  * If we wished to exclude xids this would be the right place for it.
2038  * Procs with the PROC_IN_VACUUM flag set don't usually assign xids,
2039  * but they do during truncation at the end when they get the lock and
2040  * truncate, so it is not much of a problem to include them if they
2041  * are seen and it is cleaner to include them.
2042  */
2043 
2044  xids[count++] = xid;
2045  }
2046 
2047  /*
2048  * Spin over procArray collecting all subxids, but only if there hasn't
2049  * been a suboverflow.
2050  */
2051  if (!suboverflowed)
2052  {
2053  for (index = 0; index < arrayP->numProcs; index++)
2054  {
2055  int pgprocno = arrayP->pgprocnos[index];
2056  PGPROC *proc = &allProcs[pgprocno];
2057  PGXACT *pgxact = &allPgXact[pgprocno];
2058  int nxids;
2059 
2060  /*
2061  * Save subtransaction XIDs. Other backends can't add or remove
2062  * entries while we're holding XidGenLock.
2063  */
2064  nxids = pgxact->nxids;
2065  if (nxids > 0)
2066  {
2067  /* barrier not really required, as XidGenLock is held, but ... */
2068  pg_read_barrier(); /* pairs with GetNewTransactionId */
2069 
2070  memcpy(&xids[count], (void *) proc->subxids.xids,
2071  nxids * sizeof(TransactionId));
2072  count += nxids;
2073  subcount += nxids;
2074 
2075  /*
2076  * Top-level XID of a transaction is always less than any of
2077  * its subxids, so we don't need to check if any of the
2078  * subxids are smaller than oldestRunningXid
2079  */
2080  }
2081  }
2082  }
2083 
2084  /*
2085  * It's important *not* to include the limits set by slots here because
2086  * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those
2087  * were to be included here the initial value could never increase because
2088  * of a circular dependency where slots only increase their limits when
2089  * running xacts increases oldestRunningXid and running xacts only
2090  * increases if slots do.
2091  */
2092 
2093  CurrentRunningXacts->xcnt = count - subcount;
2094  CurrentRunningXacts->subxcnt = subcount;
2095  CurrentRunningXacts->subxid_overflow = suboverflowed;
2096  CurrentRunningXacts->nextXid = ShmemVariableCache->nextXid;
2097  CurrentRunningXacts->oldestRunningXid = oldestRunningXid;
2098  CurrentRunningXacts->latestCompletedXid = latestCompletedXid;
2099 
2100  Assert(TransactionIdIsValid(CurrentRunningXacts->nextXid));
2101  Assert(TransactionIdIsValid(CurrentRunningXacts->oldestRunningXid));
2102  Assert(TransactionIdIsNormal(CurrentRunningXacts->latestCompletedXid));
2103 
2104  /* We don't release the locks here, the caller is responsible for that */
2105 
2106  return CurrentRunningXacts;
2107 }
2108 
2109 /*
2110  * GetOldestActiveTransactionId()
2111  *
2112  * Similar to GetSnapshotData but returns just oldestActiveXid. We include
2113  * all PGXACTs with an assigned TransactionId, even VACUUM processes.
2114  * We look at all databases, though there is no need to include WALSender
2115  * since this has no effect on hot standby conflicts.
2116  *
2117  * This is never executed during recovery so there is no need to look at
2118  * KnownAssignedXids.
2119  *
2120  * We don't worry about updating other counters, we want to keep this as
2121  * simple as possible and leave GetSnapshotData() as the primary code for
2122  * that bookkeeping.
2123  */
2126 {
2127  ProcArrayStruct *arrayP = procArray;
2128  TransactionId oldestRunningXid;
2129  int index;
2130 
2132 
2133  /*
2134  * Read nextXid, as the upper bound of what's still active.
2135  *
2136  * Reading a TransactionId is atomic, but we must grab the lock to make
2137  * sure that all XIDs < nextXid are already present in the proc array (or
2138  * have already completed), when we spin over it.
2139  */
2140  LWLockAcquire(XidGenLock, LW_SHARED);
2141  oldestRunningXid = ShmemVariableCache->nextXid;
2142  LWLockRelease(XidGenLock);
2143 
2144  /*
2145  * Spin over procArray collecting all xids and subxids.
2146  */
2147  LWLockAcquire(ProcArrayLock, LW_SHARED);
2148  for (index = 0; index < arrayP->numProcs; index++)
2149  {
2150  int pgprocno = arrayP->pgprocnos[index];
2151  PGXACT *pgxact = &allPgXact[pgprocno];
2152  TransactionId xid;
2153 
2154  /* Fetch xid just once - see GetNewTransactionId */
2155  xid = UINT32_ACCESS_ONCE(pgxact->xid);
2156 
2157  if (!TransactionIdIsNormal(xid))
2158  continue;
2159 
2160  if (TransactionIdPrecedes(xid, oldestRunningXid))
2161  oldestRunningXid = xid;
2162 
2163  /*
2164  * Top-level XID of a transaction is always less than any of its
2165  * subxids, so we don't need to check if any of the subxids are
2166  * smaller than oldestRunningXid
2167  */
2168  }
2169  LWLockRelease(ProcArrayLock);
2170 
2171  return oldestRunningXid;
2172 }
2173 
2174 /*
2175  * GetOldestSafeDecodingTransactionId -- lowest xid not affected by vacuum
2176  *
2177  * Returns the oldest xid that we can guarantee not to have been affected by
2178  * vacuum, i.e. no rows >= that xid have been vacuumed away unless the
2179  * transaction aborted. Note that the value can (and most of the time will) be
2180  * much more conservative than what really has been affected by vacuum, but we
2181  * currently don't have better data available.
2182  *
2183  * This is useful to initialize the cutoff xid after which a new changeset
2184  * extraction replication slot can start decoding changes.
2185  *
2186  * Must be called with ProcArrayLock held either shared or exclusively,
2187  * although most callers will want to use exclusive mode since it is expected
2188  * that the caller will immediately use the xid to peg the xmin horizon.
2189  */
2192 {
2193  ProcArrayStruct *arrayP = procArray;
2194  TransactionId oldestSafeXid;
2195  int index;
2196  bool recovery_in_progress = RecoveryInProgress();
2197 
2198  Assert(LWLockHeldByMe(ProcArrayLock));
2199 
2200  /*
2201  * Acquire XidGenLock, so no transactions can acquire an xid while we're
2202  * running. If no transaction with xid were running concurrently a new xid
2203  * could influence the RecentXmin et al.
2204  *
2205  * We initialize the computation to nextXid since that's guaranteed to be
2206  * a safe, albeit pessimal, value.
2207  */
2208  LWLockAcquire(XidGenLock, LW_SHARED);
2209  oldestSafeXid = ShmemVariableCache->nextXid;
2210 
2211  /*
2212  * If there's already a slot pegging the xmin horizon, we can start with
2213  * that value, it's guaranteed to be safe since it's computed by this
2214  * routine initially and has been enforced since. We can always use the
2215  * slot's general xmin horizon, but the catalog horizon is only usable
2216  * when only catalog data is going to be looked at.
2217  */
2218  if (TransactionIdIsValid(procArray->replication_slot_xmin) &&
2220  oldestSafeXid))
2221  oldestSafeXid = procArray->replication_slot_xmin;
2222 
2223  if (catalogOnly &&
2226  oldestSafeXid))
2227  oldestSafeXid = procArray->replication_slot_catalog_xmin;
2228 
2229  /*
2230  * If we're not in recovery, we walk over the procarray and collect the
2231  * lowest xid. Since we're called with ProcArrayLock held and have
2232  * acquired XidGenLock, no entries can vanish concurrently, since
2233  * PGXACT->xid is only set with XidGenLock held and only cleared with
2234  * ProcArrayLock held.
2235  *
2236  * In recovery we can't lower the safe value besides what we've computed
2237  * above, so we'll have to wait a bit longer there. We unfortunately can
2238  * *not* use KnownAssignedXidsGetOldestXmin() since the KnownAssignedXids
2239  * machinery can miss values and return an older value than is safe.
2240  */
2241  if (!recovery_in_progress)
2242  {
2243  /*
2244  * Spin over procArray collecting all min(PGXACT->xid)
2245  */
2246  for (index = 0; index < arrayP->numProcs; index++)
2247  {
2248  int pgprocno = arrayP->pgprocnos[index];
2249  PGXACT *pgxact = &allPgXact[pgprocno];
2250  TransactionId xid;
2251 
2252  /* Fetch xid just once - see GetNewTransactionId */
2253  xid = UINT32_ACCESS_ONCE(pgxact->xid);
2254 
2255  if (!TransactionIdIsNormal(xid))
2256  continue;
2257 
2258  if (TransactionIdPrecedes(xid, oldestSafeXid))
2259  oldestSafeXid = xid;
2260  }
2261  }
2262 
2263  LWLockRelease(XidGenLock);
2264 
2265  return oldestSafeXid;
2266 }
2267 
2268 /*
2269  * GetVirtualXIDsDelayingChkpt -- Get the VXIDs of transactions that are
2270  * delaying checkpoint because they have critical actions in progress.
2271  *
2272  * Constructs an array of VXIDs of transactions that are currently in commit
2273  * critical sections, as shown by having delayChkpt set in their PGXACT.
2274  *
2275  * Returns a palloc'd array that should be freed by the caller.
2276  * *nvxids is the number of valid entries.
2277  *
2278  * Note that because backends set or clear delayChkpt without holding any lock,
2279  * the result is somewhat indeterminate, but we don't really care. Even in
2280  * a multiprocessor with delayed writes to shared memory, it should be certain
2281  * that setting of delayChkpt will propagate to shared memory when the backend
2282  * takes a lock, so we cannot fail to see a virtual xact as delayChkpt if
2283  * it's already inserted its commit record. Whether it takes a little while
2284  * for clearing of delayChkpt to propagate is unimportant for correctness.
2285  */
2288 {
2289  VirtualTransactionId *vxids;
2290  ProcArrayStruct *arrayP = procArray;
2291  int count = 0;
2292  int index;
2293 
2294  /* allocate what's certainly enough result space */
2295  vxids = (VirtualTransactionId *)
2296  palloc(sizeof(VirtualTransactionId) * arrayP->maxProcs);
2297 
2298  LWLockAcquire(ProcArrayLock, LW_SHARED);
2299 
2300  for (index = 0; index < arrayP->numProcs; index++)
2301  {
2302  int pgprocno = arrayP->pgprocnos[index];
2303  PGPROC *proc = &allProcs[pgprocno];
2304  PGXACT *pgxact = &allPgXact[pgprocno];
2305 
2306  if (pgxact->delayChkpt)
2307  {
2308  VirtualTransactionId vxid;
2309 
2310  GET_VXID_FROM_PGPROC(vxid, *proc);
2311  if (VirtualTransactionIdIsValid(vxid))
2312  vxids[count++] = vxid;
2313  }
2314  }
2315 
2316  LWLockRelease(ProcArrayLock);
2317 
2318  *nvxids = count;
2319  return vxids;
2320 }
2321 
2322 /*
2323  * HaveVirtualXIDsDelayingChkpt -- Are any of the specified VXIDs delaying?
2324  *
2325  * This is used with the results of GetVirtualXIDsDelayingChkpt to see if any
2326  * of the specified VXIDs are still in critical sections of code.
2327  *
2328  * Note: this is O(N^2) in the number of vxacts that are/were delaying, but
2329  * those numbers should be small enough for it not to be a problem.
2330  */
2331 bool
2333 {
2334  bool result = false;
2335  ProcArrayStruct *arrayP = procArray;
2336  int index;
2337 
2338  LWLockAcquire(ProcArrayLock, LW_SHARED);
2339 
2340  for (index = 0; index < arrayP->numProcs; index++)
2341  {
2342  int pgprocno = arrayP->pgprocnos[index];
2343  PGPROC *proc = &allProcs[pgprocno];
2344  PGXACT *pgxact = &allPgXact[pgprocno];
2345  VirtualTransactionId vxid;
2346 
2347  GET_VXID_FROM_PGPROC(vxid, *proc);
2348 
2349  if (pgxact->delayChkpt && VirtualTransactionIdIsValid(vxid))
2350  {
2351  int i;
2352 
2353  for (i = 0; i < nvxids; i++)
2354  {
2355  if (VirtualTransactionIdEquals(vxid, vxids[i]))
2356  {
2357  result = true;
2358  break;
2359  }
2360  }
2361  if (result)
2362  break;
2363  }
2364  }
2365 
2366  LWLockRelease(ProcArrayLock);
2367 
2368  return result;
2369 }
2370 
2371 /*
2372  * BackendPidGetProc -- get a backend's PGPROC given its PID
2373  *
2374  * Returns NULL if not found. Note that it is up to the caller to be
2375  * sure that the question remains meaningful for long enough for the
2376  * answer to be used ...
2377  */
2378 PGPROC *
2380 {
2381  PGPROC *result;
2382 
2383  if (pid == 0) /* never match dummy PGPROCs */
2384  return NULL;
2385 
2386  LWLockAcquire(ProcArrayLock, LW_SHARED);
2387 
2388  result = BackendPidGetProcWithLock(pid);
2389 
2390  LWLockRelease(ProcArrayLock);
2391 
2392  return result;
2393 }
2394 
2395 /*
2396  * BackendPidGetProcWithLock -- get a backend's PGPROC given its PID
2397  *
2398  * Same as above, except caller must be holding ProcArrayLock. The found
2399  * entry, if any, can be assumed to be valid as long as the lock remains held.
2400  */
2401 PGPROC *
2403 {
2404  PGPROC *result = NULL;
2405  ProcArrayStruct *arrayP = procArray;
2406  int index;
2407 
2408  if (pid == 0) /* never match dummy PGPROCs */
2409  return NULL;
2410 
2411  for (index = 0; index < arrayP->numProcs; index++)
2412  {
2413  PGPROC *proc = &allProcs[arrayP->pgprocnos[index]];
2414 
2415  if (proc->pid == pid)
2416  {
2417  result = proc;
2418  break;
2419  }
2420  }
2421 
2422  return result;
2423 }
2424 
2425 /*
2426  * BackendXidGetPid -- get a backend's pid given its XID
2427  *
2428  * Returns 0 if not found or it's a prepared transaction. Note that
2429  * it is up to the caller to be sure that the question remains
2430  * meaningful for long enough for the answer to be used ...
2431  *
2432  * Only main transaction Ids are considered. This function is mainly
2433  * useful for determining what backend owns a lock.
2434  *
2435  * Beware that not every xact has an XID assigned. However, as long as you
2436  * only call this using an XID found on disk, you're safe.
2437  */
2438 int
2440 {
2441  int result = 0;
2442  ProcArrayStruct *arrayP = procArray;
2443  int index;
2444 
2445  if (xid == InvalidTransactionId) /* never match invalid xid */
2446  return 0;
2447 
2448  LWLockAcquire(ProcArrayLock, LW_SHARED);
2449 
2450  for (index = 0; index < arrayP->numProcs; index++)
2451  {
2452  int pgprocno = arrayP->pgprocnos[index];
2453  PGPROC *proc = &allProcs[pgprocno];
2454  PGXACT *pgxact = &allPgXact[pgprocno];
2455 
2456  if (pgxact->xid == xid)
2457  {
2458  result = proc->pid;
2459  break;
2460  }
2461  }
2462 
2463  LWLockRelease(ProcArrayLock);
2464 
2465  return result;
2466 }
2467 
2468 /*
2469  * IsBackendPid -- is a given pid a running backend
2470  *
2471  * This is not called by the backend, but is called by external modules.
2472  */
2473 bool
2475 {
2476  return (BackendPidGetProc(pid) != NULL);
2477 }
2478 
2479 
2480 /*
2481  * GetCurrentVirtualXIDs -- returns an array of currently active VXIDs.
2482  *
2483  * The array is palloc'd. The number of valid entries is returned into *nvxids.
2484  *
2485  * The arguments allow filtering the set of VXIDs returned. Our own process
2486  * is always skipped. In addition:
2487  * If limitXmin is not InvalidTransactionId, skip processes with
2488  * xmin > limitXmin.
2489  * If excludeXmin0 is true, skip processes with xmin = 0.
2490  * If allDbs is false, skip processes attached to other databases.
2491  * If excludeVacuum isn't zero, skip processes for which
2492  * (vacuumFlags & excludeVacuum) is not zero.
2493  *
2494  * Note: the purpose of the limitXmin and excludeXmin0 parameters is to
2495  * allow skipping backends whose oldest live snapshot is no older than
2496  * some snapshot we have. Since we examine the procarray with only shared
2497  * lock, there are race conditions: a backend could set its xmin just after
2498  * we look. Indeed, on multiprocessors with weak memory ordering, the
2499  * other backend could have set its xmin *before* we look. We know however
2500  * that such a backend must have held shared ProcArrayLock overlapping our
2501  * own hold of ProcArrayLock, else we would see its xmin update. Therefore,
2502  * any snapshot the other backend is taking concurrently with our scan cannot
2503  * consider any transactions as still running that we think are committed
2504  * (since backends must hold ProcArrayLock exclusive to commit).
2505  */
2507 GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
2508  bool allDbs, int excludeVacuum,
2509  int *nvxids)
2510 {
2511  VirtualTransactionId *vxids;
2512  ProcArrayStruct *arrayP = procArray;
2513  int count = 0;
2514  int index;
2515 
2516  /* allocate what's certainly enough result space */
2517  vxids = (VirtualTransactionId *)
2518  palloc(sizeof(VirtualTransactionId) * arrayP->maxProcs);
2519 
2520  LWLockAcquire(ProcArrayLock, LW_SHARED);
2521 
2522  for (index = 0; index < arrayP->numProcs; index++)
2523  {
2524  int pgprocno = arrayP->pgprocnos[index];
2525  PGPROC *proc = &allProcs[pgprocno];
2526  PGXACT *pgxact = &allPgXact[pgprocno];
2527 
2528  if (proc == MyProc)
2529  continue;
2530 
2531  if (excludeVacuum & pgxact->vacuumFlags)
2532  continue;
2533 
2534  if (allDbs || proc->databaseId == MyDatabaseId)
2535  {
2536  /* Fetch xmin just once - might change on us */
2537  TransactionId pxmin = UINT32_ACCESS_ONCE(pgxact->xmin);
2538 
2539  if (excludeXmin0 && !TransactionIdIsValid(pxmin))
2540  continue;
2541 
2542  /*
2543  * InvalidTransactionId precedes all other XIDs, so a proc that
2544  * hasn't set xmin yet will not be rejected by this test.
2545  */
2546  if (!TransactionIdIsValid(limitXmin) ||
2547  TransactionIdPrecedesOrEquals(pxmin, limitXmin))
2548  {
2549  VirtualTransactionId vxid;
2550 
2551  GET_VXID_FROM_PGPROC(vxid, *proc);
2552  if (VirtualTransactionIdIsValid(vxid))
2553  vxids[count++] = vxid;
2554  }
2555  }
2556  }
2557 
2558  LWLockRelease(ProcArrayLock);
2559 
2560  *nvxids = count;
2561  return vxids;
2562 }
2563 
2564 /*
2565  * GetConflictingVirtualXIDs -- returns an array of currently active VXIDs.
2566  *
2567  * Usage is limited to conflict resolution during recovery on standby servers.
2568  * limitXmin is supplied as either latestRemovedXid, or InvalidTransactionId
2569  * in cases where we cannot accurately determine a value for latestRemovedXid.
2570  *
2571  * If limitXmin is InvalidTransactionId then we want to kill everybody,
2572  * so we're not worried if they have a snapshot or not, nor does it really
2573  * matter what type of lock we hold.
2574  *
2575  * All callers that are checking xmins always now supply a valid and useful
2576  * value for limitXmin. The limitXmin is always lower than the lowest
2577  * numbered KnownAssignedXid that is not already a FATAL error. This is
2578  * because we only care about cleanup records that are cleaning up tuple
2579  * versions from committed transactions. In that case they will only occur
2580  * at the point where the record is less than the lowest running xid. That
2581  * allows us to say that if any backend takes a snapshot concurrently with
2582  * us then the conflict assessment made here would never include the snapshot
2583  * that is being derived. So we take LW_SHARED on the ProcArray and allow
2584  * concurrent snapshots when limitXmin is valid. We might think about adding
2585  * Assert(limitXmin < lowest(KnownAssignedXids))
2586  * but that would not be true in the case of FATAL errors lagging in array,
2587  * but we already know those are bogus anyway, so we skip that test.
2588  *
2589  * If dbOid is valid we skip backends attached to other databases.
2590  *
2591  * Be careful to *not* pfree the result from this function. We reuse
2592  * this array sufficiently often that we use malloc for the result.
2593  */
2596 {
2597  static VirtualTransactionId *vxids;
2598  ProcArrayStruct *arrayP = procArray;
2599  int count = 0;
2600  int index;
2601 
2602  /*
2603  * If first time through, get workspace to remember main XIDs in. We
2604  * malloc it permanently to avoid repeated palloc/pfree overhead. Allow
2605  * result space, remembering room for a terminator.
2606  */
2607  if (vxids == NULL)
2608  {
2609  vxids = (VirtualTransactionId *)
2610  malloc(sizeof(VirtualTransactionId) * (arrayP->maxProcs + 1));
2611  if (vxids == NULL)
2612  ereport(ERROR,
2613  (errcode(ERRCODE_OUT_OF_MEMORY),
2614  errmsg("out of memory")));
2615  }
2616 
2617  LWLockAcquire(ProcArrayLock, LW_SHARED);
2618 
2619  for (index = 0; index < arrayP->numProcs; index++)
2620  {
2621  int pgprocno = arrayP->pgprocnos[index];
2622  PGPROC *proc = &allProcs[pgprocno];
2623  PGXACT *pgxact = &allPgXact[pgprocno];
2624 
2625  /* Exclude prepared transactions */
2626  if (proc->pid == 0)
2627  continue;
2628 
2629  if (!OidIsValid(dbOid) ||
2630  proc->databaseId == dbOid)
2631  {
2632  /* Fetch xmin just once - can't change on us, but good coding */
2633  TransactionId pxmin = UINT32_ACCESS_ONCE(pgxact->xmin);
2634 
2635  /*
2636  * We ignore an invalid pxmin because this means that backend has
2637  * no snapshot currently. We hold a Share lock to avoid contention
2638  * with users taking snapshots. That is not a problem because the
2639  * current xmin is always at least one higher than the latest
2640  * removed xid, so any new snapshot would never conflict with the
2641  * test here.
2642  */
2643  if (!TransactionIdIsValid(limitXmin) ||
2644  (TransactionIdIsValid(pxmin) && !TransactionIdFollows(pxmin, limitXmin)))
2645  {
2646  VirtualTransactionId vxid;
2647 
2648  GET_VXID_FROM_PGPROC(vxid, *proc);
2649  if (VirtualTransactionIdIsValid(vxid))
2650  vxids[count++] = vxid;
2651  }
2652  }
2653  }
2654 
2655  LWLockRelease(ProcArrayLock);
2656 
2657  /* add the terminator */
2658  vxids[count].backendId = InvalidBackendId;
2660 
2661  return vxids;
2662 }
2663 
2664 /*
2665  * CancelVirtualTransaction - used in recovery conflict processing
2666  *
2667  * Returns pid of the process signaled, or 0 if not found.
2668  */
2669 pid_t
2671 {
2672  ProcArrayStruct *arrayP = procArray;
2673  int index;
2674  pid_t pid = 0;
2675 
2676  LWLockAcquire(ProcArrayLock, LW_SHARED);
2677 
2678  for (index = 0; index < arrayP->numProcs; index++)
2679  {
2680  int pgprocno = arrayP->pgprocnos[index];
2681  PGPROC *proc = &allProcs[pgprocno];
2682  VirtualTransactionId procvxid;
2683 
2684  GET_VXID_FROM_PGPROC(procvxid, *proc);
2685 
2686  if (procvxid.backendId == vxid.backendId &&
2687  procvxid.localTransactionId == vxid.localTransactionId)
2688  {
2689  proc->recoveryConflictPending = true;
2690  pid = proc->pid;
2691  if (pid != 0)
2692  {
2693  /*
2694  * Kill the pid if it's still here. If not, that's what we
2695  * wanted so ignore any errors.
2696  */
2697  (void) SendProcSignal(pid, sigmode, vxid.backendId);
2698  }
2699  break;
2700  }
2701  }
2702 
2703  LWLockRelease(ProcArrayLock);
2704 
2705  return pid;
2706 }
2707 
2708 /*
2709  * MinimumActiveBackends --- count backends (other than myself) that are
2710  * in active transactions. Return true if the count exceeds the
2711  * minimum threshold passed. This is used as a heuristic to decide if
2712  * a pre-XLOG-flush delay is worthwhile during commit.
2713  *
2714  * Do not count backends that are blocked waiting for locks, since they are
2715  * not going to get to run until someone else commits.
2716  */
2717 bool
2719 {
2720  ProcArrayStruct *arrayP = procArray;
2721  int count = 0;
2722  int index;
2723 
2724  /* Quick short-circuit if no minimum is specified */
2725  if (min == 0)
2726  return true;
2727 
2728  /*
2729  * Note: for speed, we don't acquire ProcArrayLock. This is a little bit
2730  * bogus, but since we are only testing fields for zero or nonzero, it
2731  * should be OK. The result is only used for heuristic purposes anyway...
2732  */
2733  for (index = 0; index < arrayP->numProcs; index++)
2734  {
2735  int pgprocno = arrayP->pgprocnos[index];
2736  PGPROC *proc = &allProcs[pgprocno];
2737  PGXACT *pgxact = &allPgXact[pgprocno];
2738 
2739  /*
2740  * Since we're not holding a lock, need to be prepared to deal with
2741  * garbage, as someone could have incremented numProcs but not yet
2742  * filled the structure.
2743  *
2744  * If someone just decremented numProcs, 'proc' could also point to a
2745  * PGPROC entry that's no longer in the array. It still points to a
2746  * PGPROC struct, though, because freed PGPROC entries just go to the
2747  * free list and are recycled. Its contents are nonsense in that case,
2748  * but that's acceptable for this function.
2749  */
2750  if (pgprocno == -1)
2751  continue; /* do not count deleted entries */
2752  if (proc == MyProc)
2753  continue; /* do not count myself */
2754  if (pgxact->xid == InvalidTransactionId)
2755  continue; /* do not count if no XID assigned */
2756  if (proc->pid == 0)
2757  continue; /* do not count prepared xacts */
2758  if (proc->waitLock != NULL)
2759  continue; /* do not count if blocked on a lock */
2760  count++;
2761  if (count >= min)
2762  break;
2763  }
2764 
2765  return count >= min;
2766 }
2767 
2768 /*
2769  * CountDBBackends --- count backends that are using specified database
2770  */
2771 int
2773 {
2774  ProcArrayStruct *arrayP = procArray;
2775  int count = 0;
2776  int index;
2777 
2778  LWLockAcquire(ProcArrayLock, LW_SHARED);
2779 
2780  for (index = 0; index < arrayP->numProcs; index++)
2781  {
2782  int pgprocno = arrayP->pgprocnos[index];
2783  PGPROC *proc = &allProcs[pgprocno];
2784 
2785  if (proc->pid == 0)
2786  continue; /* do not count prepared xacts */
2787  if (!OidIsValid(databaseid) ||
2788  proc->databaseId == databaseid)
2789  count++;
2790  }
2791 
2792  LWLockRelease(ProcArrayLock);
2793 
2794  return count;
2795 }
2796 
2797 /*
2798  * CountDBConnections --- counts database backends ignoring any background
2799  * worker processes
2800  */
2801 int
2803 {
2804  ProcArrayStruct *arrayP = procArray;
2805  int count = 0;
2806  int index;
2807 
2808  LWLockAcquire(ProcArrayLock, LW_SHARED);
2809 
2810  for (index = 0; index < arrayP->numProcs; index++)
2811  {
2812  int pgprocno = arrayP->pgprocnos[index];
2813  PGPROC *proc = &allProcs[pgprocno];
2814 
2815  if (proc->pid == 0)
2816  continue; /* do not count prepared xacts */
2817  if (proc->isBackgroundWorker)
2818  continue; /* do not count background workers */
2819  if (!OidIsValid(databaseid) ||
2820  proc->databaseId == databaseid)
2821  count++;
2822  }
2823 
2824  LWLockRelease(ProcArrayLock);
2825 
2826  return count;
2827 }
2828 
2829 /*
2830  * CancelDBBackends --- cancel backends that are using specified database
2831  */
2832 void
2833 CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
2834 {
2835  ProcArrayStruct *arrayP = procArray;
2836  int index;
2837  pid_t pid = 0;
2838 
2839  /* tell all backends to die */
2840  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
2841 
2842  for (index = 0; index < arrayP->numProcs; index++)
2843  {
2844  int pgprocno = arrayP->pgprocnos[index];
2845  PGPROC *proc = &allProcs[pgprocno];
2846 
2847  if (databaseid == InvalidOid || proc->databaseId == databaseid)
2848  {
2849  VirtualTransactionId procvxid;
2850 
2851  GET_VXID_FROM_PGPROC(procvxid, *proc);
2852 
2853  proc->recoveryConflictPending = conflictPending;
2854  pid = proc->pid;
2855  if (pid != 0)
2856  {
2857  /*
2858  * Kill the pid if it's still here. If not, that's what we
2859  * wanted so ignore any errors.
2860  */
2861  (void) SendProcSignal(pid, sigmode, procvxid.backendId);
2862  }
2863  }
2864  }
2865 
2866  LWLockRelease(ProcArrayLock);
2867 }
2868 
2869 /*
2870  * CountUserBackends --- count backends that are used by specified user
2871  */
2872 int
2874 {
2875  ProcArrayStruct *arrayP = procArray;
2876  int count = 0;
2877  int index;
2878 
2879  LWLockAcquire(ProcArrayLock, LW_SHARED);
2880 
2881  for (index = 0; index < arrayP->numProcs; index++)
2882  {
2883  int pgprocno = arrayP->pgprocnos[index];
2884  PGPROC *proc = &allProcs[pgprocno];
2885 
2886  if (proc->pid == 0)
2887  continue; /* do not count prepared xacts */
2888  if (proc->isBackgroundWorker)
2889  continue; /* do not count background workers */
2890  if (proc->roleId == roleid)
2891  count++;
2892  }
2893 
2894  LWLockRelease(ProcArrayLock);
2895 
2896  return count;
2897 }
2898 
2899 /*
2900  * CountOtherDBBackends -- check for other backends running in the given DB
2901  *
2902  * If there are other backends in the DB, we will wait a maximum of 5 seconds
2903  * for them to exit. Autovacuum backends are encouraged to exit early by
2904  * sending them SIGTERM, but normal user backends are just waited for.
2905  *
2906  * The current backend is always ignored; it is caller's responsibility to
2907  * check whether the current backend uses the given DB, if it's important.
2908  *
2909  * Returns true if there are (still) other backends in the DB, false if not.
2910  * Also, *nbackends and *nprepared are set to the number of other backends
2911  * and prepared transactions in the DB, respectively.
2912  *
2913  * This function is used to interlock DROP DATABASE and related commands
2914  * against there being any active backends in the target DB --- dropping the
2915  * DB while active backends remain would be a Bad Thing. Note that we cannot
2916  * detect here the possibility of a newly-started backend that is trying to
2917  * connect to the doomed database, so additional interlocking is needed during
2918  * backend startup. The caller should normally hold an exclusive lock on the
2919  * target DB before calling this, which is one reason we mustn't wait
2920  * indefinitely.
2921  */
2922 bool
2923 CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
2924 {
2925  ProcArrayStruct *arrayP = procArray;
2926 
2927 #define MAXAUTOVACPIDS 10 /* max autovacs to SIGTERM per iteration */
2928  int autovac_pids[MAXAUTOVACPIDS];
2929  int tries;
2930 
2931  /* 50 tries with 100ms sleep between tries makes 5 sec total wait */
2932  for (tries = 0; tries < 50; tries++)
2933  {
2934  int nautovacs = 0;
2935  bool found = false;
2936  int index;
2937 
2939 
2940  *nbackends = *nprepared = 0;
2941 
2942  LWLockAcquire(ProcArrayLock, LW_SHARED);
2943 
2944  for (index = 0; index < arrayP->numProcs; index++)
2945  {
2946  int pgprocno = arrayP->pgprocnos[index];
2947  PGPROC *proc = &allProcs[pgprocno];
2948  PGXACT *pgxact = &allPgXact[pgprocno];
2949 
2950  if (proc->databaseId != databaseId)
2951  continue;
2952  if (proc == MyProc)
2953  continue;
2954 
2955  found = true;
2956 
2957  if (proc->pid == 0)
2958  (*nprepared)++;
2959  else
2960  {
2961  (*nbackends)++;
2962  if ((pgxact->vacuumFlags & PROC_IS_AUTOVACUUM) &&
2963  nautovacs < MAXAUTOVACPIDS)
2964  autovac_pids[nautovacs++] = proc->pid;
2965  }
2966  }
2967 
2968  LWLockRelease(ProcArrayLock);
2969 
2970  if (!found)
2971  return false; /* no conflicting backends, so done */
2972 
2973  /*
2974  * Send SIGTERM to any conflicting autovacuums before sleeping. We
2975  * postpone this step until after the loop because we don't want to
2976  * hold ProcArrayLock while issuing kill(). We have no idea what might
2977  * block kill() inside the kernel...
2978  */
2979  for (index = 0; index < nautovacs; index++)
2980  (void) kill(autovac_pids[index], SIGTERM); /* ignore any error */
2981 
2982  /* sleep, then try again */
2983  pg_usleep(100 * 1000L); /* 100ms */
2984  }
2985 
2986  return true; /* timed out, still conflicts */
2987 }
2988 
2989 /*
2990  * ProcArraySetReplicationSlotXmin
2991  *
2992  * Install limits to future computations of the xmin horizon to prevent vacuum
2993  * and HOT pruning from removing affected rows still needed by clients with
2994  * replication slots.
2995  */
2996 void
2998  bool already_locked)
2999 {
3000  Assert(!already_locked || LWLockHeldByMe(ProcArrayLock));
3001 
3002  if (!already_locked)
3003  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3004 
3005  procArray->replication_slot_xmin = xmin;
3006  procArray->replication_slot_catalog_xmin = catalog_xmin;
3007 
3008  if (!already_locked)
3009  LWLockRelease(ProcArrayLock);
3010 }
3011 
3012 /*
3013  * ProcArrayGetReplicationSlotXmin
3014  *
3015  * Return the current slot xmin limits. That's useful to be able to remove
3016  * data that's older than those limits.
3017  */
3018 void
3020  TransactionId *catalog_xmin)
3021 {
3022  LWLockAcquire(ProcArrayLock, LW_SHARED);
3023 
3024  if (xmin != NULL)
3025  *xmin = procArray->replication_slot_xmin;
3026 
3027  if (catalog_xmin != NULL)
3028  *catalog_xmin = procArray->replication_slot_catalog_xmin;
3029 
3030  LWLockRelease(ProcArrayLock);
3031 }
3032 
3033 
3034 #define XidCacheRemove(i) \
3035  do { \
3036  MyProc->subxids.xids[i] = MyProc->subxids.xids[MyPgXact->nxids - 1]; \
3037  pg_write_barrier(); \
3038  MyPgXact->nxids--; \
3039  } while (0)
3040 
3041 /*
3042  * XidCacheRemoveRunningXids
3043  *
3044  * Remove a bunch of TransactionIds from the list of known-running
3045  * subtransactions for my backend. Both the specified xid and those in
3046  * the xids[] array (of length nxids) are removed from the subxids cache.
3047  * latestXid must be the latest XID among the group.
3048  */
3049 void
3051  int nxids, const TransactionId *xids,
3052  TransactionId latestXid)
3053 {
3054  int i,
3055  j;
3056 
3058 
3059  /*
3060  * We must hold ProcArrayLock exclusively in order to remove transactions
3061  * from the PGPROC array. (See src/backend/access/transam/README.) It's
3062  * possible this could be relaxed since we know this routine is only used
3063  * to abort subtransactions, but pending closer analysis we'd best be
3064  * conservative.
3065  *
3066  * Note that we do not have to be careful about memory ordering of our own
3067  * reads wrt. GetNewTransactionId() here - only this process can modify
3068  * relevant fields of MyProc/MyPgXact. But we do have to be careful about
3069  * our own writes being well ordered.
3070  */
3071  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3072 
3073  /*
3074  * Under normal circumstances xid and xids[] will be in increasing order,
3075  * as will be the entries in subxids. Scan backwards to avoid O(N^2)
3076  * behavior when removing a lot of xids.
3077  */
3078  for (i = nxids - 1; i >= 0; i--)
3079  {
3080  TransactionId anxid = xids[i];
3081 
3082  for (j = MyPgXact->nxids - 1; j >= 0; j--)
3083  {
3084  if (TransactionIdEquals(MyProc->subxids.xids[j], anxid))
3085  {
3086  XidCacheRemove(j);
3087  break;
3088  }
3089  }
3090 
3091  /*
3092  * Ordinarily we should have found it, unless the cache has
3093  * overflowed. However it's also possible for this routine to be
3094  * invoked multiple times for the same subtransaction, in case of an
3095  * error during AbortSubTransaction. So instead of Assert, emit a
3096  * debug warning.
3097  */
3098  if (j < 0 && !MyPgXact->overflowed)
3099  elog(WARNING, "did not find subXID %u in MyProc", anxid);
3100  }
3101 
3102  for (j = MyPgXact->nxids - 1; j >= 0; j--)
3103  {
3104  if (TransactionIdEquals(MyProc->subxids.xids[j], xid))
3105  {
3106  XidCacheRemove(j);
3107  break;
3108  }
3109  }
3110  /* Ordinarily we should have found it, unless the cache has overflowed */
3111  if (j < 0 && !MyPgXact->overflowed)
3112  elog(WARNING, "did not find subXID %u in MyProc", xid);
3113 
3114  /* Also advance global latestCompletedXid while holding the lock */
3116  latestXid))
3118 
3119  LWLockRelease(ProcArrayLock);
3120 }
3121 
3122 #ifdef XIDCACHE_DEBUG
3123 
3124 /*
3125  * Print stats about effectiveness of XID cache
3126  */
3127 static void
3128 DisplayXidCache(void)
3129 {
3130  fprintf(stderr,
3131  "XidCache: xmin: %ld, known: %ld, myxact: %ld, latest: %ld, mainxid: %ld, childxid: %ld, knownassigned: %ld, nooflo: %ld, slow: %ld\n",
3132  xc_by_recent_xmin,
3133  xc_by_known_xact,
3134  xc_by_my_xact,
3135  xc_by_latest_xid,
3136  xc_by_main_xid,
3137  xc_by_child_xid,
3138  xc_by_known_assigned,
3139  xc_no_overflow,
3140  xc_slow_answer);
3141 }
3142 #endif /* XIDCACHE_DEBUG */
3143 
3144 
3145 /* ----------------------------------------------
3146  * KnownAssignedTransactions sub-module
3147  * ----------------------------------------------
3148  */
3149 
3150 /*
3151  * In Hot Standby mode, we maintain a list of transactions that are (or were)
3152  * running in the master at the current point in WAL. These XIDs must be
3153  * treated as running by standby transactions, even though they are not in
3154  * the standby server's PGXACT array.
3155  *
3156  * We record all XIDs that we know have been assigned. That includes all the
3157  * XIDs seen in WAL records, plus all unobserved XIDs that we can deduce have
3158  * been assigned. We can deduce the existence of unobserved XIDs because we
3159  * know XIDs are assigned in sequence, with no gaps. The KnownAssignedXids
3160  * list expands as new XIDs are observed or inferred, and contracts when
3161  * transaction completion records arrive.
3162  *
3163  * During hot standby we do not fret too much about the distinction between
3164  * top-level XIDs and subtransaction XIDs. We store both together in the
3165  * KnownAssignedXids list. In backends, this is copied into snapshots in
3166  * GetSnapshotData(), taking advantage of the fact that XidInMVCCSnapshot()
3167  * doesn't care about the distinction either. Subtransaction XIDs are
3168  * effectively treated as top-level XIDs and in the typical case pg_subtrans
3169  * links are *not* maintained (which does not affect visibility).
3170  *
3171  * We have room in KnownAssignedXids and in snapshots to hold maxProcs *
3172  * (1 + PGPROC_MAX_CACHED_SUBXIDS) XIDs, so every master transaction must
3173  * report its subtransaction XIDs in a WAL XLOG_XACT_ASSIGNMENT record at
3174  * least every PGPROC_MAX_CACHED_SUBXIDS. When we receive one of these
3175  * records, we mark the subXIDs as children of the top XID in pg_subtrans,
3176  * and then remove them from KnownAssignedXids. This prevents overflow of
3177  * KnownAssignedXids and snapshots, at the cost that status checks for these
3178  * subXIDs will take a slower path through TransactionIdIsInProgress().
3179  * This means that KnownAssignedXids is not necessarily complete for subXIDs,
3180  * though it should be complete for top-level XIDs; this is the same situation
3181  * that holds with respect to the PGPROC entries in normal running.
3182  *
3183  * When we throw away subXIDs from KnownAssignedXids, we need to keep track of
3184  * that, similarly to tracking overflow of a PGPROC's subxids array. We do
3185  * that by remembering the lastOverflowedXID, ie the last thrown-away subXID.
3186  * As long as that is within the range of interesting XIDs, we have to assume
3187  * that subXIDs are missing from snapshots. (Note that subXID overflow occurs
3188  * on primary when 65th subXID arrives, whereas on standby it occurs when 64th
3189  * subXID arrives - that is not an error.)
3190  *
3191  * Should a backend on primary somehow disappear before it can write an abort
3192  * record, then we just leave those XIDs in KnownAssignedXids. They actually
3193  * aborted but we think they were running; the distinction is irrelevant
3194  * because either way any changes done by the transaction are not visible to
3195  * backends in the standby. We prune KnownAssignedXids when
3196  * XLOG_RUNNING_XACTS arrives, to forestall possible overflow of the
3197  * array due to such dead XIDs.
3198  */
3199 
3200 /*
3201  * RecordKnownAssignedTransactionIds
3202  * Record the given XID in KnownAssignedXids, as well as any preceding
3203  * unobserved XIDs.
3204  *
3205  * RecordKnownAssignedTransactionIds() should be run for *every* WAL record
3206  * associated with a transaction. Must be called for each record after we
3207  * have executed StartupCLOG() et al, since we must ExtendCLOG() etc..
3208  *
3209  * Called during recovery in analogy with and in place of GetNewTransactionId()
3210  */
3211 void
3213 {
3217 
3218  elog(trace_recovery(DEBUG4), "record known xact %u latestObservedXid %u",
3219  xid, latestObservedXid);
3220 
3221  /*
3222  * When a newly observed xid arrives, it is frequently the case that it is
3223  * *not* the next xid in sequence. When this occurs, we must treat the
3224  * intervening xids as running also.
3225  */
3227  {
3228  TransactionId next_expected_xid;
3229 
3230  /*
3231  * Extend subtrans like we do in GetNewTransactionId() during normal
3232  * operation using individual extend steps. Note that we do not need
3233  * to extend clog since its extensions are WAL logged.
3234  *
3235  * This part has to be done regardless of standbyState since we
3236  * immediately start assigning subtransactions to their toplevel
3237  * transactions.
3238  */
3239  next_expected_xid = latestObservedXid;
3240  while (TransactionIdPrecedes(next_expected_xid, xid))
3241  {
3242  TransactionIdAdvance(next_expected_xid);
3243  ExtendSUBTRANS(next_expected_xid);
3244  }
3245  Assert(next_expected_xid == xid);
3246 
3247  /*
3248  * If the KnownAssignedXids machinery isn't up yet, there's nothing
3249  * more to do since we don't track assigned xids yet.
3250  */
3252  {
3253  latestObservedXid = xid;
3254  return;
3255  }
3256 
3257  /*
3258  * Add (latestObservedXid, xid] onto the KnownAssignedXids array.
3259  */
3260  next_expected_xid = latestObservedXid;
3261  TransactionIdAdvance(next_expected_xid);
3262  KnownAssignedXidsAdd(next_expected_xid, xid, false);
3263 
3264  /*
3265  * Now we can advance latestObservedXid
3266  */
3267  latestObservedXid = xid;
3268 
3269  /* ShmemVariableCache->nextXid must be beyond any observed xid */
3270  next_expected_xid = latestObservedXid;
3271  TransactionIdAdvance(next_expected_xid);
3272  LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
3273  ShmemVariableCache->nextXid = next_expected_xid;
3274  LWLockRelease(XidGenLock);
3275  }
3276 }
3277 
3278 /*
3279  * ExpireTreeKnownAssignedTransactionIds
3280  * Remove the given XIDs from KnownAssignedXids.
3281  *
3282  * Called during recovery in analogy with and in place of ProcArrayEndTransaction()
3283  */
3284 void
3286  TransactionId *subxids, TransactionId max_xid)
3287 {
3289 
3290  /*
3291  * Uses same locking as transaction commit
3292  */
3293  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3294 
3295  KnownAssignedXidsRemoveTree(xid, nsubxids, subxids);
3296 
3297  /* As in ProcArrayEndTransaction, advance latestCompletedXid */
3299  max_xid))
3301 
3302  LWLockRelease(ProcArrayLock);
3303 }
3304 
3305 /*
3306  * ExpireAllKnownAssignedTransactionIds
3307  * Remove all entries in KnownAssignedXids
3308  */
3309 void
3311 {
3312  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3314  LWLockRelease(ProcArrayLock);
3315 }
3316 
3317 /*
3318  * ExpireOldKnownAssignedTransactionIds
3319  * Remove KnownAssignedXids entries preceding the given XID
3320  */
3321 void
3323 {
3324  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3326  LWLockRelease(ProcArrayLock);
3327 }
3328 
3329 
3330 /*
3331  * Private module functions to manipulate KnownAssignedXids
3332  *
3333  * There are 5 main uses of the KnownAssignedXids data structure:
3334  *
3335  * * backends taking snapshots - all valid XIDs need to be copied out
3336  * * backends seeking to determine presence of a specific XID
3337  * * startup process adding new known-assigned XIDs
3338  * * startup process removing specific XIDs as transactions end
3339  * * startup process pruning array when special WAL records arrive
3340  *
3341  * This data structure is known to be a hot spot during Hot Standby, so we
3342  * go to some lengths to make these operations as efficient and as concurrent
3343  * as possible.
3344  *
3345  * The XIDs are stored in an array in sorted order --- TransactionIdPrecedes
3346  * order, to be exact --- to allow binary search for specific XIDs. Note:
3347  * in general TransactionIdPrecedes would not provide a total order, but
3348  * we know that the entries present at any instant should not extend across
3349  * a large enough fraction of XID space to wrap around (the master would
3350  * shut down for fear of XID wrap long before that happens). So it's OK to
3351  * use TransactionIdPrecedes as a binary-search comparator.
3352  *
3353  * It's cheap to maintain the sortedness during insertions, since new known
3354  * XIDs are always reported in XID order; we just append them at the right.
3355  *
3356  * To keep individual deletions cheap, we need to allow gaps in the array.
3357  * This is implemented by marking array elements as valid or invalid using
3358  * the parallel boolean array KnownAssignedXidsValid[]. A deletion is done
3359  * by setting KnownAssignedXidsValid[i] to false, *without* clearing the
3360  * XID entry itself. This preserves the property that the XID entries are
3361  * sorted, so we can do binary searches easily. Periodically we compress
3362  * out the unused entries; that's much cheaper than having to compress the
3363  * array immediately on every deletion.
3364  *
3365  * The actually valid items in KnownAssignedXids[] and KnownAssignedXidsValid[]
3366  * are those with indexes tail <= i < head; items outside this subscript range
3367  * have unspecified contents. When head reaches the end of the array, we
3368  * force compression of unused entries rather than wrapping around, since
3369  * allowing wraparound would greatly complicate the search logic. We maintain
3370  * an explicit tail pointer so that pruning of old XIDs can be done without
3371  * immediately moving the array contents. In most cases only a small fraction
3372  * of the array contains valid entries at any instant.
3373  *
3374  * Although only the startup process can ever change the KnownAssignedXids
3375  * data structure, we still need interlocking so that standby backends will
3376  * not observe invalid intermediate states. The convention is that backends
3377  * must hold shared ProcArrayLock to examine the array. To remove XIDs from
3378  * the array, the startup process must hold ProcArrayLock exclusively, for
3379  * the usual transactional reasons (compare commit/abort of a transaction
3380  * during normal running). Compressing unused entries out of the array
3381  * likewise requires exclusive lock. To add XIDs to the array, we just insert
3382  * them into slots to the right of the head pointer and then advance the head
3383  * pointer. This wouldn't require any lock at all, except that on machines
3384  * with weak memory ordering we need to be careful that other processors
3385  * see the array element changes before they see the head pointer change.
3386  * We handle this by using a spinlock to protect reads and writes of the
3387  * head/tail pointers. (We could dispense with the spinlock if we were to
3388  * create suitable memory access barrier primitives and use those instead.)
3389  * The spinlock must be taken to read or write the head/tail pointers unless
3390  * the caller holds ProcArrayLock exclusively.
3391  *
3392  * Algorithmic analysis:
3393  *
3394  * If we have a maximum of M slots, with N XIDs currently spread across
3395  * S elements then we have N <= S <= M always.
3396  *
3397  * * Adding a new XID is O(1) and needs little locking (unless compression
3398  * must happen)
3399  * * Compressing the array is O(S) and requires exclusive lock
3400  * * Removing an XID is O(logS) and requires exclusive lock
3401  * * Taking a snapshot is O(S) and requires shared lock
3402  * * Checking for an XID is O(logS) and requires shared lock
3403  *
3404  * In comparison, using a hash table for KnownAssignedXids would mean that
3405  * taking snapshots would be O(M). If we can maintain S << M then the
3406  * sorted array technique will deliver significantly faster snapshots.
3407  * If we try to keep S too small then we will spend too much time compressing,
3408  * so there is an optimal point for any workload mix. We use a heuristic to
3409  * decide when to compress the array, though trimming also helps reduce
3410  * frequency of compressing. The heuristic requires us to track the number of
3411  * currently valid XIDs in the array.
3412  */
3413 
3414 
3415 /*
3416  * Compress KnownAssignedXids by shifting valid data down to the start of the
3417  * array, removing any gaps.
3418  *
3419  * A compression step is forced if "force" is true, otherwise we do it
3420  * only if a heuristic indicates it's a good time to do it.
3421  *
3422  * Caller must hold ProcArrayLock in exclusive mode.
3423  */
3424 static void
3426 {
3427  ProcArrayStruct *pArray = procArray;
3428  int head,
3429  tail;
3430  int compress_index;
3431  int i;
3432 
3433  /* no spinlock required since we hold ProcArrayLock exclusively */
3434  head = pArray->headKnownAssignedXids;
3435  tail = pArray->tailKnownAssignedXids;
3436 
3437  if (!force)
3438  {
3439  /*
3440  * If we can choose how much to compress, use a heuristic to avoid
3441  * compressing too often or not often enough.
3442  *
3443  * Heuristic is if we have a large enough current spread and less than
3444  * 50% of the elements are currently in use, then compress. This
3445  * should ensure we compress fairly infrequently. We could compress
3446  * less often though the virtual array would spread out more and
3447  * snapshots would become more expensive.
3448  */
3449  int nelements = head - tail;
3450 
3451  if (nelements < 4 * PROCARRAY_MAXPROCS ||
3452  nelements < 2 * pArray->numKnownAssignedXids)
3453  return;
3454  }
3455 
3456  /*
3457  * We compress the array by reading the valid values from tail to head,
3458  * re-aligning data to 0th element.
3459  */
3460  compress_index = 0;
3461  for (i = tail; i < head; i++)
3462  {
3463  if (KnownAssignedXidsValid[i])
3464  {
3465  KnownAssignedXids[compress_index] = KnownAssignedXids[i];
3466  KnownAssignedXidsValid[compress_index] = true;
3467  compress_index++;
3468  }
3469  }
3470 
3471  pArray->tailKnownAssignedXids = 0;
3472  pArray->headKnownAssignedXids = compress_index;
3473 }
3474 
3475 /*
3476  * Add xids into KnownAssignedXids at the head of the array.
3477  *
3478  * xids from from_xid to to_xid, inclusive, are added to the array.
3479  *
3480  * If exclusive_lock is true then caller already holds ProcArrayLock in
3481  * exclusive mode, so we need no extra locking here. Else caller holds no
3482  * lock, so we need to be sure we maintain sufficient interlocks against
3483  * concurrent readers. (Only the startup process ever calls this, so no need
3484  * to worry about concurrent writers.)
3485  */
3486 static void
3488  bool exclusive_lock)
3489 {
3490  ProcArrayStruct *pArray = procArray;
3491  TransactionId next_xid;
3492  int head,
3493  tail;
3494  int nxids;
3495  int i;
3496 
3497  Assert(TransactionIdPrecedesOrEquals(from_xid, to_xid));
3498 
3499  /*
3500  * Calculate how many array slots we'll need. Normally this is cheap; in
3501  * the unusual case where the XIDs cross the wrap point, we do it the hard
3502  * way.
3503  */
3504  if (to_xid >= from_xid)
3505  nxids = to_xid - from_xid + 1;
3506  else
3507  {
3508  nxids = 1;
3509  next_xid = from_xid;
3510  while (TransactionIdPrecedes(next_xid, to_xid))
3511  {
3512  nxids++;
3513  TransactionIdAdvance(next_xid);
3514  }
3515  }
3516 
3517  /*
3518  * Since only the startup process modifies the head/tail pointers, we
3519  * don't need a lock to read them here.
3520  */
3521  head = pArray->headKnownAssignedXids;
3522  tail = pArray->tailKnownAssignedXids;
3523 
3524  Assert(head >= 0 && head <= pArray->maxKnownAssignedXids);
3525  Assert(tail >= 0 && tail < pArray->maxKnownAssignedXids);
3526 
3527  /*
3528  * Verify that insertions occur in TransactionId sequence. Note that even
3529  * if the last existing element is marked invalid, it must still have a
3530  * correctly sequenced XID value.
3531  */
3532  if (head > tail &&
3533  TransactionIdFollowsOrEquals(KnownAssignedXids[head - 1], from_xid))
3534  {
3536  elog(ERROR, "out-of-order XID insertion in KnownAssignedXids");
3537  }
3538 
3539  /*
3540  * If our xids won't fit in the remaining space, compress out free space
3541  */
3542  if (head + nxids > pArray->maxKnownAssignedXids)
3543  {
3544  /* must hold lock to compress */
3545  if (!exclusive_lock)
3546  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3547 
3549 
3550  head = pArray->headKnownAssignedXids;
3551  /* note: we no longer care about the tail pointer */
3552 
3553  if (!exclusive_lock)
3554  LWLockRelease(ProcArrayLock);
3555 
3556  /*
3557  * If it still won't fit then we're out of memory
3558  */
3559  if (head + nxids > pArray->maxKnownAssignedXids)
3560  elog(ERROR, "too many KnownAssignedXids");
3561  }
3562 
3563  /* Now we can insert the xids into the space starting at head */
3564  next_xid = from_xid;
3565  for (i = 0; i < nxids; i++)
3566  {
3567  KnownAssignedXids[head] = next_xid;
3568  KnownAssignedXidsValid[head] = true;
3569  TransactionIdAdvance(next_xid);
3570  head++;
3571  }
3572 
3573  /* Adjust count of number of valid entries */
3574  pArray->numKnownAssignedXids += nxids;
3575 
3576  /*
3577  * Now update the head pointer. We use a spinlock to protect this
3578  * pointer, not because the update is likely to be non-atomic, but to
3579  * ensure that other processors see the above array updates before they
3580  * see the head pointer change.
3581  *
3582  * If we're holding ProcArrayLock exclusively, there's no need to take the
3583  * spinlock.
3584  */
3585  if (exclusive_lock)
3586  pArray->headKnownAssignedXids = head;
3587  else
3588  {
3590  pArray->headKnownAssignedXids = head;
3592  }
3593 }
3594 
3595 /*
3596  * KnownAssignedXidsSearch
3597  *
3598  * Searches KnownAssignedXids for a specific xid and optionally removes it.
3599  * Returns true if it was found, false if not.
3600  *
3601  * Caller must hold ProcArrayLock in shared or exclusive mode.
3602  * Exclusive lock must be held for remove = true.
3603  */
3604 static bool
3606 {
3607  ProcArrayStruct *pArray = procArray;
3608  int first,
3609  last;
3610  int head;
3611  int tail;
3612  int result_index = -1;
3613 
3614  if (remove)
3615  {
3616  /* we hold ProcArrayLock exclusively, so no need for spinlock */
3617  tail = pArray->tailKnownAssignedXids;
3618  head = pArray->headKnownAssignedXids;
3619  }
3620  else
3621  {
3622  /* take spinlock to ensure we see up-to-date array contents */
3624  tail = pArray->tailKnownAssignedXids;
3625  head = pArray->headKnownAssignedXids;
3627  }
3628 
3629  /*
3630  * Standard binary search. Note we can ignore the KnownAssignedXidsValid
3631  * array here, since even invalid entries will contain sorted XIDs.
3632  */
3633  first = tail;
3634  last = head - 1;
3635  while (first <= last)
3636  {
3637  int mid_index;
3638  TransactionId mid_xid;
3639 
3640  mid_index = (first + last) / 2;
3641  mid_xid = KnownAssignedXids[mid_index];
3642 
3643  if (xid == mid_xid)
3644  {
3645  result_index = mid_index;
3646  break;
3647  }
3648  else if (TransactionIdPrecedes(xid, mid_xid))
3649  last = mid_index - 1;
3650  else
3651  first = mid_index + 1;
3652  }
3653 
3654  if (result_index < 0)
3655  return false; /* not in array */
3656 
3657  if (!KnownAssignedXidsValid[result_index])
3658  return false; /* in array, but invalid */
3659 
3660  if (remove)
3661  {
3662  KnownAssignedXidsValid[result_index] = false;
3663 
3664  pArray->numKnownAssignedXids--;
3665  Assert(pArray->numKnownAssignedXids >= 0);
3666 
3667  /*
3668  * If we're removing the tail element then advance tail pointer over
3669  * any invalid elements. This will speed future searches.
3670  */
3671  if (result_index == tail)
3672  {
3673  tail++;
3674  while (tail < head && !KnownAssignedXidsValid[tail])
3675  tail++;
3676  if (tail >= head)
3677  {
3678  /* Array is empty, so we can reset both pointers */
3679  pArray->headKnownAssignedXids = 0;
3680  pArray->tailKnownAssignedXids = 0;
3681  }
3682  else
3683  {
3684  pArray->tailKnownAssignedXids = tail;
3685  }
3686  }
3687  }
3688 
3689  return true;
3690 }
3691 
3692 /*
3693  * Is the specified XID present in KnownAssignedXids[]?
3694  *
3695  * Caller must hold ProcArrayLock in shared or exclusive mode.
3696  */
3697 static bool
3699 {
3701 
3702  return KnownAssignedXidsSearch(xid, false);
3703 }
3704 
3705 /*
3706  * Remove the specified XID from KnownAssignedXids[].
3707  *
3708  * Caller must hold ProcArrayLock in exclusive mode.
3709  */
3710 static void
3712 {
3714 
3715  elog(trace_recovery(DEBUG4), "remove KnownAssignedXid %u", xid);
3716 
3717  /*
3718  * Note: we cannot consider it an error to remove an XID that's not
3719  * present. We intentionally remove subxact IDs while processing
3720  * XLOG_XACT_ASSIGNMENT, to avoid array overflow. Then those XIDs will be
3721  * removed again when the top-level xact commits or aborts.
3722  *
3723  * It might be possible to track such XIDs to distinguish this case from
3724  * actual errors, but it would be complicated and probably not worth it.
3725  * So, just ignore the search result.
3726  */
3727  (void) KnownAssignedXidsSearch(xid, true);
3728 }
3729 
3730 /*
3731  * KnownAssignedXidsRemoveTree
3732  * Remove xid (if it's not InvalidTransactionId) and all the subxids.
3733  *
3734  * Caller must hold ProcArrayLock in exclusive mode.
3735  */
3736 static void
3738  TransactionId *subxids)
3739 {
3740  int i;
3741 
3742  if (TransactionIdIsValid(xid))
3744 
3745  for (i = 0; i < nsubxids; i++)
3746  KnownAssignedXidsRemove(subxids[i]);
3747 
3748  /* Opportunistically compress the array */
3750 }
3751 
3752 /*
3753  * Prune KnownAssignedXids up to, but *not* including xid. If xid is invalid
3754  * then clear the whole table.
3755  *
3756  * Caller must hold ProcArrayLock in exclusive mode.
3757  */
3758 static void
3760 {
3761  ProcArrayStruct *pArray = procArray;
3762  int count = 0;
3763  int head,
3764  tail,
3765  i;
3766 
3767  if (!TransactionIdIsValid(removeXid))
3768  {
3769  elog(trace_recovery(DEBUG4), "removing all KnownAssignedXids");
3770  pArray->numKnownAssignedXids = 0;
3771  pArray->headKnownAssignedXids = pArray->tailKnownAssignedXids = 0;
3772  return;
3773  }
3774 
3775  elog(trace_recovery(DEBUG4), "prune KnownAssignedXids to %u", removeXid);
3776 
3777  /*
3778  * Mark entries invalid starting at the tail. Since array is sorted, we
3779  * can stop as soon as we reach an entry >= removeXid.
3780  */
3781  tail = pArray->tailKnownAssignedXids;
3782  head = pArray->headKnownAssignedXids;
3783 
3784  for (i = tail; i < head; i++)
3785  {
3786  if (KnownAssignedXidsValid[i])
3787  {
3788  TransactionId knownXid = KnownAssignedXids[i];
3789 
3790  if (TransactionIdFollowsOrEquals(knownXid, removeXid))
3791  break;
3792 
3793  if (!StandbyTransactionIdIsPrepared(knownXid))
3794  {
3795  KnownAssignedXidsValid[i] = false;
3796  count++;
3797  }
3798  }
3799  }
3800 
3801  pArray->numKnownAssignedXids -= count;
3802  Assert(pArray->numKnownAssignedXids >= 0);
3803 
3804  /*
3805  * Advance the tail pointer if we've marked the tail item invalid.
3806  */
3807  for (i = tail; i < head; i++)
3808  {
3809  if (KnownAssignedXidsValid[i])
3810  break;
3811  }
3812  if (i >= head)
3813  {
3814  /* Array is empty, so we can reset both pointers */
3815  pArray->headKnownAssignedXids = 0;
3816  pArray->tailKnownAssignedXids = 0;
3817  }
3818  else
3819  {
3820  pArray->tailKnownAssignedXids = i;
3821  }
3822 
3823  /* Opportunistically compress the array */
3825 }
3826 
3827 /*
3828  * KnownAssignedXidsGet - Get an array of xids by scanning KnownAssignedXids.
3829  * We filter out anything >= xmax.
3830  *
3831  * Returns the number of XIDs stored into xarray[]. Caller is responsible
3832  * that array is large enough.
3833  *
3834  * Caller must hold ProcArrayLock in (at least) shared mode.
3835  */
3836 static int
3838 {
3840 
3841  return KnownAssignedXidsGetAndSetXmin(xarray, &xtmp, xmax);
3842 }
3843 
3844 /*
3845  * KnownAssignedXidsGetAndSetXmin - as KnownAssignedXidsGet, plus
3846  * we reduce *xmin to the lowest xid value seen if not already lower.
3847  *
3848  * Caller must hold ProcArrayLock in (at least) shared mode.
3849  */
3850 static int
3852  TransactionId xmax)
3853 {
3854  int count = 0;
3855  int head,
3856  tail;
3857  int i;
3858 
3859  /*
3860  * Fetch head just once, since it may change while we loop. We can stop
3861  * once we reach the initially seen head, since we are certain that an xid
3862  * cannot enter and then leave the array while we hold ProcArrayLock. We
3863  * might miss newly-added xids, but they should be >= xmax so irrelevant
3864  * anyway.
3865  *
3866  * Must take spinlock to ensure we see up-to-date array contents.
3867  */
3869  tail = procArray->tailKnownAssignedXids;
3870  head = procArray->headKnownAssignedXids;
3872 
3873  for (i = tail; i < head; i++)
3874  {
3875  /* Skip any gaps in the array */
3876  if (KnownAssignedXidsValid[i])
3877  {
3878  TransactionId knownXid = KnownAssignedXids[i];
3879 
3880  /*
3881  * Update xmin if required. Only the first XID need be checked,
3882  * since the array is sorted.
3883  */
3884  if (count == 0 &&
3885  TransactionIdPrecedes(knownXid, *xmin))
3886  *xmin = knownXid;
3887 
3888  /*
3889  * Filter out anything >= xmax, again relying on sorted property
3890  * of array.
3891  */
3892  if (TransactionIdIsValid(xmax) &&
3893  TransactionIdFollowsOrEquals(knownXid, xmax))
3894  break;
3895 
3896  /* Add knownXid into output array */
3897  xarray[count++] = knownXid;
3898  }
3899  }
3900 
3901  return count;
3902 }
3903 
3904 /*
3905  * Get oldest XID in the KnownAssignedXids array, or InvalidTransactionId
3906  * if nothing there.
3907  */
3908 static TransactionId
3910 {
3911  int head,
3912  tail;
3913  int i;
3914 
3915  /*
3916  * Fetch head just once, since it may change while we loop.
3917  */
3919  tail = procArray->tailKnownAssignedXids;
3920  head = procArray->headKnownAssignedXids;
3922 
3923  for (i = tail; i < head; i++)
3924  {
3925  /* Skip any gaps in the array */
3926  if (KnownAssignedXidsValid[i])
3927  return KnownAssignedXids[i];
3928  }
3929 
3930  return InvalidTransactionId;
3931 }
3932 
3933 /*
3934  * Display KnownAssignedXids to provide debug trail
3935  *
3936  * Currently this is only called within startup process, so we need no
3937  * special locking.
3938  *
3939  * Note this is pretty expensive, and much of the expense will be incurred
3940  * even if the elog message will get discarded. It's not currently called
3941  * in any performance-critical places, however, so no need to be tenser.
3942  */
3943 static void
3945 {
3946  ProcArrayStruct *pArray = procArray;
3948  int head,
3949  tail,
3950  i;
3951  int nxids = 0;
3952 
3953  tail = pArray->tailKnownAssignedXids;
3954  head = pArray->headKnownAssignedXids;
3955 
3956  initStringInfo(&buf);
3957 
3958  for (i = tail; i < head; i++)
3959  {
3960  if (KnownAssignedXidsValid[i])
3961  {
3962  nxids++;
3963  appendStringInfo(&buf, "[%d]=%u ", i, KnownAssignedXids[i]);
3964  }
3965  }
3966 
3967  elog(trace_level, "%d KnownAssignedXids (num=%d tail=%d head=%d) %s",
3968  nxids,
3969  pArray->numKnownAssignedXids,
3970  pArray->tailKnownAssignedXids,
3971  pArray->headKnownAssignedXids,
3972  buf.data);
3973 
3974  pfree(buf.data);
3975 }
3976 
3977 /*
3978  * KnownAssignedXidsReset
3979  * Resets KnownAssignedXids to be empty
3980  */
3981 static void
3983 {
3984  ProcArrayStruct *pArray = procArray;
3985 
3986  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3987 
3988  pArray->numKnownAssignedXids = 0;
3989  pArray->tailKnownAssignedXids = 0;
3990  pArray->headKnownAssignedXids = 0;
3991 
3992  LWLockRelease(ProcArrayLock);
3993 }
#define TransactionIdAdvance(dest)
Definition: transam.h:48
int slock_t
Definition: s_lock.h:912
void ProcArrayApplyRecoveryInfo(RunningTransactions running)
Definition: procarray.c:663
#define XidCacheRemove(i)
Definition: procarray.c:3034
static TransactionId latestObservedXid
Definition: procarray.c:109
VirtualTransactionId * GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0, bool allDbs, int excludeVacuum, int *nvxids)
Definition: procarray.c:2507
TransactionId oldestRunningXid
Definition: standby.h:76
bool procArrayGroupMember
Definition: proc.h:166
pid_t CancelVirtualTransaction(VirtualTransactionId vxid, ProcSignalReason sigmode)
Definition: procarray.c:2670
#define PROCARRAY_MAXPROCS
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
TransactionId GetOldestSafeDecodingTransactionId(bool catalogOnly)
Definition: procarray.c:2191
int CountDBBackends(Oid databaseid)
Definition: procarray.c:2772
#define DEBUG1
Definition: elog.h:25
static void KnownAssignedXidsDisplay(int trace_level)
Definition: procarray.c:3944
#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:96
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:769
TransactionId SubTransGetTopmostTransaction(TransactionId xid)
Definition: subtrans.c:150
bool LWLockHeldByMe(LWLock *l)
Definition: lwlock.c:1841
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:2379
PGXACT * allPgXact
Definition: proc.h:249
PGPROC * MyProc
Definition: proc.c:67
bool TransactionIdIsInProgress(TransactionId xid)
Definition: procarray.c:1002
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:2595
TransactionId replication_slot_catalog_xmin
Definition: procarray.c:93
XLogRecPtr lsn
Definition: snapshot.h:114
bool TransactionIdFollowsOrEquals(TransactionId id1, TransactionId id2)
Definition: transam.c:349
VirtualTransactionId * GetVirtualXIDsDelayingChkpt(int *nvxids)
Definition: procarray.c:2287
static void ProcArrayGroupClearXid(PGPROC *proc, TransactionId latestXid)
Definition: procarray.c:485
#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:165
slock_t known_assigned_xids_lck
Definition: procarray.c:79
PROC_HDR * ProcGlobal
Definition: proc.c:80
bool suboverflowed
Definition: snapshot.h:93
TransactionId * xids
Definition: standby.h:79
#define kill(pid, sig)
Definition: win32_port.h:437
bool TransactionIdDidCommit(TransactionId transactionId)
Definition: transam.c:125
#define LOG
Definition: elog.h:26
Form_pg_class rd_rel
Definition: rel.h:84
unsigned int Oid
Definition: postgres_ext.h:31
bool RecoveryInProgress(void)
Definition: xlog.c:8065
#define TransactionIdRetreat(dest)
Definition: transam.h:56
LocalTransactionId localTransactionId
Definition: lock.h:66
void ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
Definition: procarray.c:397
#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:3285
#define MAXAUTOVACPIDS
uint32 regd_count
Definition: snapshot.h:110
#define OidIsValid(objectId)
Definition: c.h:638
void ProcArrayGetReplicationSlotXmin(TransactionId *xmin, TransactionId *catalog_xmin)
Definition: procarray.c:3019
void ExpireAllKnownAssignedTransactionIds(void)
Definition: procarray.c:3310
PGXACT * MyPgXact
Definition: proc.c:68
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
TransactionId TransactionXmin
Definition: snapmgr.c:164
void LWLockRegisterTranche(int tranche_id, const char *tranche_name)
Definition: lwlock.c:602
TransactionId latestCompletedXid
Definition: standby.h:77
Definition: type.h:89
#define malloc(a)
Definition: header.h:50
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1725
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:2718
static void KnownAssignedXidsRemovePreceding(TransactionId xid)
Definition: procarray.c:3759
PGPROC * BackendPidGetProcWithLock(int pid)
Definition: procarray.c:2402
#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:2802
#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:3050
TimestampTz GetSnapshotCurrentTimestamp(void)
Definition: snapmgr.c:1686
void ProcArrayClearTransaction(PGPROC *proc)
Definition: procarray.c:600
#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:324
void ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
Definition: procarray.c:334
#define FATAL
Definition: elog.h:52
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:11350
TransactionId nextXid
Definition: transam.h:117
void CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
Definition: procarray.c:2833
#define xc_by_main_xid_inc()
Definition: procarray.c:148
bool LWLockConditionalAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1293
static char * buf
Definition: pg_test_fsync.c:67
bool recoveryConflictPending
Definition: proc.h:127
#define memmove(d, s, c)
Definition: c.h:1218
#define xc_by_known_assigned_inc()
Definition: procarray.c:150
TransactionId RecentGlobalXmin
Definition: snapmgr.c:166
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:1411
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:69
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1262
TransactionId xmin
Definition: snapshot.h:68
static void KnownAssignedXidsReset(void)
Definition: procarray.c:3982
LOCK * waitLock
Definition: proc.h:139
int numKnownAssignedXids
Definition: procarray.c:76
TransactionId RecentGlobalDataXmin
Definition: snapmgr.c:167
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:79
static void KnownAssignedXidsRemoveTree(TransactionId xid, int nsubxids, TransactionId *subxids)
Definition: procarray.c:3737
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:2439
#define InvalidBackendId
Definition: backendid.h:23
static PGXACT * allPgXact
Definition: procarray.c:102
#define RelationIsAccessibleInLogicalDecoding(relation)
Definition: rel.h:565
static int KnownAssignedXidsGet(TransactionId *xarray, TransactionId xmax)
Definition: procarray.c:3837
Size add_size(Size s1, Size s2)
Definition: shmem.c:475
Oid MyDatabaseId
Definition: globals.c:85
static TransactionId KnownAssignedXidsGetOldestXmin(void)
Definition: procarray.c:3909
bool overflowed
Definition: proc.h:234
static void ProcArrayEndTransactionInternal(PGPROC *proc, PGXACT *pgxact, TransactionId latestXid)
Definition: procarray.c:451
#define InvalidOid
Definition: postgres_ext.h:36
CommandId curcid
Definition: snapshot.h:98
int GetMaxSnapshotXidCount(void)
Definition: procarray.c:1466
TransactionId GetOldestXmin(Relation rel, int flags)
Definition: procarray.c:1321
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:95
size_t Size
Definition: c.h:466
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1238
Snapshot GetSnapshotData(Snapshot snapshot)
Definition: procarray.c:1519
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1121
static int KnownAssignedXidsGetAndSetXmin(TransactionId *xarray, TransactionId *xmin, TransactionId xmax)
Definition: procarray.c:3851
static void KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid, bool exclusive_lock)
Definition: procarray.c:3487
bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
Definition: procarray.c:1885
#define NormalTransactionIdPrecedes(id1, id2)
Definition: transam.h:62
#define xc_no_overflow_inc()
Definition: procarray.c:151
bool EnableHotStandby
Definition: xlog.c:90
void PGSemaphoreLock(PGSemaphore sema)
Definition: posix_sema.c:309
static void KnownAssignedXidsCompress(bool force)
Definition: procarray.c:3425
int CountUserBackends(Oid roleid)
Definition: procarray.c:2873
static bool KnownAssignedXidExists(TransactionId xid)
Definition: procarray.c:3698
int pgprocno
Definition: proc.h:110
TransactionId nextXid
Definition: standby.h:75
bool TransactionIdIsActive(TransactionId xid)
Definition: procarray.c:1222
#define xc_slow_answer_inc()
Definition: procarray.c:152
pg_atomic_uint32 procArrayGroupFirst
Definition: proc.h:259
void ProcArrayInitRecovery(TransactionId initializedUptoXID)
Definition: procarray.c:632
uint32 xcnt
Definition: snapshot.h:80
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:3605
static void KnownAssignedXidsRemove(TransactionId xid)
Definition: procarray.c:3711
#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:3322
TransactionId GetOldestActiveTransactionId(void)
Definition: procarray.c:2125
bool IsBackendPid(int pid)
Definition: procarray.c:2474
#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:2997
int GetMaxSnapshotSubxidCount(void)
Definition: procarray.c:1477
RunningTransactions GetRunningTransactionData(void)
Definition: procarray.c:1951
void ProcArrayApplyXidAssignment(TransactionId topxid, int nsubxids, TransactionId *subxids)
Definition: procarray.c:918
TimestampTz whenTaken
Definition: snapshot.h:113
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:2923
CommandId GetCurrentCommandId(bool used)
Definition: xact.c:661
#define qsort(a, b, c, d)
Definition: port.h:479
#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:1848
PGSemaphore sem
Definition: proc.h:101
bool HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids)
Definition: procarray.c:2332
void RecordKnownAssignedTransactionIds(TransactionId xid)
Definition: procarray.c:3212
#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:1810
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:191
void ProcArrayAdd(PGPROC *proc)
Definition: procarray.c:276
#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:91
uint32 active_count
Definition: snapshot.h:109
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:92
LocalTransactionId lxid
Definition: proc.h:106
TransactionId latestCompletedXid
Definition: transam.h:135