1 /*-------------------------------------------------------------------------
4 * postgres OID & XID variables support routines
6 * Copyright (c) 2000-2018, PostgreSQL Global Development Group
9 * src/backend/access/transam/varsup.c
11 *-------------------------------------------------------------------------
16 #include "access/clog.h"
17 #include "access/commit_ts.h"
18 #include "access/subtrans.h"
19 #include "access/transam.h"
20 #include "access/xact.h"
21 #include "access/xlog.h"
22 #include "commands/dbcommands.h"
23 #include "miscadmin.h"
24 #include "postmaster/autovacuum.h"
25 #include "storage/pmsignal.h"
26 #include "storage/proc.h"
27 #include "utils/syscache.h"
30 /* Number of OIDs to prefetch (preallocate) per XLOG write */
31 #define VAR_OID_PREFETCH 8192
33 /* pointer to "variable cache" in shared memory (set up by shmem.c) */
34 VariableCache ShmemVariableCache = NULL;
38 * Allocate the next XID for a new transaction or subtransaction.
40 * The new XID is also stored into MyPgXact before returning.
42 * Note: when this is called, we are actually already inside a valid
43 * transaction, since XIDs are now not allocated until the transaction
44 * does something. So it is safe to do a database lookup if we want to
45 * issue a warning about XID wrap.
48 GetNewTransactionId(bool isSubXact)
53 * Workers synchronize transaction state at the beginning of each parallel
54 * operation, so we can't account for new XIDs after that point.
56 if (IsInParallelMode())
57 elog(ERROR, "cannot assign TransactionIds during a parallel operation");
60 * During bootstrap initialization, we return the special bootstrap
63 if (IsBootstrapProcessingMode())
66 MyPgXact->xid = BootstrapTransactionId;
67 return BootstrapTransactionId;
70 /* safety check, we should never get this far in a HS standby */
71 if (RecoveryInProgress())
72 elog(ERROR, "cannot assign TransactionIds during recovery");
74 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
76 xid = ShmemVariableCache->nextXid;
79 * Check to see if it's safe to assign another XID. This protects against
80 * catastrophic data loss due to XID wraparound. The basic rules are:
82 * If we're past xidVacLimit, start trying to force autovacuum cycles.
83 * If we're past xidWarnLimit, start issuing warnings.
84 * If we're past xidStopLimit, refuse to execute transactions, unless
85 * we are running in single-user mode (which gives an escape hatch
86 * to the DBA who somehow got past the earlier defenses).
88 * Note that this coding also appears in GetNewMultiXactId.
91 if (TransactionIdFollowsOrEquals(xid, ShmemVariableCache->xidVacLimit))
94 * For safety's sake, we release XidGenLock while sending signals,
95 * warnings, etc. This is not so much because we care about
96 * preserving concurrency in this situation, as to avoid any
97 * possibility of deadlock while doing get_database_name(). First,
98 * copy all the shared values we'll need in this path.
100 TransactionId xidWarnLimit = ShmemVariableCache->xidWarnLimit;
101 TransactionId xidStopLimit = ShmemVariableCache->xidStopLimit;
102 TransactionId xidWrapLimit = ShmemVariableCache->xidWrapLimit;
103 Oid oldest_datoid = ShmemVariableCache->oldestXidDB;
105 LWLockRelease(XidGenLock);
108 * To avoid swamping the postmaster with signals, we issue the autovac
109 * request only once per 64K transaction starts. This still gives
110 * plenty of chances before we get into real trouble.
112 if (IsUnderPostmaster && (xid % 65536) == 0)
113 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
115 if (IsUnderPostmaster &&
116 TransactionIdFollowsOrEquals(xid, xidStopLimit))
118 char *oldest_datname = get_database_name(oldest_datoid);
120 /* complain even if that DB has disappeared */
123 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
124 errmsg("database is not accepting commands to avoid wraparound data loss in database \"%s\"",
126 errhint("Stop the postmaster and vacuum that database in single-user mode.\n"
127 "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
130 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
131 errmsg("database is not accepting commands to avoid wraparound data loss in database with OID %u",
133 errhint("Stop the postmaster and vacuum that database in single-user mode.\n"
134 "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
136 else if (TransactionIdFollowsOrEquals(xid, xidWarnLimit))
138 char *oldest_datname = get_database_name(oldest_datoid);
140 /* complain even if that DB has disappeared */
143 (errmsg("database \"%s\" must be vacuumed within %u transactions",
146 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
147 "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
150 (errmsg("database with OID %u must be vacuumed within %u transactions",
153 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
154 "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
157 /* Re-acquire lock and start over */
158 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
159 xid = ShmemVariableCache->nextXid;
163 * If we are allocating the first XID of a new page of the commit log,
164 * zero out that commit-log page before returning. We must do this while
165 * holding XidGenLock, else another xact could acquire and commit a later
166 * XID before we zero the page. Fortunately, a page of the commit log
167 * holds 32K or more transactions, so we don't have to do this very often.
169 * Extend pg_subtrans and pg_commit_ts too.
176 * Now advance the nextXid counter. This must not happen until after we
177 * have successfully completed ExtendCLOG() --- if that routine fails, we
178 * want the next incoming transaction to try it again. We cannot assign
179 * more XIDs until there is CLOG space for them.
181 TransactionIdAdvance(ShmemVariableCache->nextXid);
184 * We must store the new XID into the shared ProcArray before releasing
185 * XidGenLock. This ensures that every active XID older than
186 * latestCompletedXid is present in the ProcArray, which is essential for
187 * correct OldestXmin tracking; see src/backend/access/transam/README.
189 * Note that readers of PGXACT xid fields should be careful to fetch the
190 * value only once, rather than assume they can read a value multiple
191 * times and get the same answer each time. Note we are assuming that
192 * TransactionId and int fetch/store are atomic.
194 * The same comments apply to the subxact xid count and overflow fields.
196 * Use of a write barrier prevents dangerous code rearrangement in this
197 * function; other backends could otherwise e.g. be examining my subxids
198 * info concurrently, and we don't want them to see an invalid
199 * intermediate state, such as an incremented nxids before the array entry
202 * Other processes that read nxids should do so before reading xids
203 * elements with a pg_read_barrier() in between, so that they can be sure
204 * not to read an uninitialized array element; see
205 * src/backend/storage/lmgr/README.barrier.
207 * If there's no room to fit a subtransaction XID into PGPROC, set the
208 * cache-overflowed flag instead. This forces readers to look in
209 * pg_subtrans to map subtransaction XIDs up to top-level XIDs. There is a
210 * race-condition window, in that the new XID will not appear as running
211 * until its parent link has been placed into pg_subtrans. However, that
212 * will happen before anyone could possibly have a reason to inquire about
213 * the status of the XID, so it seems OK. (Snapshots taken during this
214 * window *will* include the parent XID, so they will deliver the correct
215 * answer later on when someone does have a reason to inquire.)
218 MyPgXact->xid = xid; /* LWLockRelease acts as barrier */
221 int nxids = MyPgXact->nxids;
223 if (nxids < PGPROC_MAX_CACHED_SUBXIDS)
225 MyProc->subxids.xids[nxids] = xid;
227 MyPgXact->nxids = nxids + 1;
230 MyPgXact->overflowed = true;
233 LWLockRelease(XidGenLock);
239 * Read nextXid but don't allocate it.
242 ReadNewTransactionId(void)
246 LWLockAcquire(XidGenLock, LW_SHARED);
247 xid = ShmemVariableCache->nextXid;
248 LWLockRelease(XidGenLock);
254 * Advance the cluster-wide value for the oldest valid clog entry.
256 * We must acquire CLogTruncationLock to advance the oldestClogXid. It's not
257 * necessary to hold the lock during the actual clog truncation, only when we
258 * advance the limit, as code looking up arbitrary xids is required to hold
259 * CLogTruncationLock from when it tests oldestClogXid through to when it
260 * completes the clog lookup.
263 AdvanceOldestClogXid(TransactionId oldest_datfrozenxid)
265 LWLockAcquire(CLogTruncationLock, LW_EXCLUSIVE);
266 if (TransactionIdPrecedes(ShmemVariableCache->oldestClogXid,
267 oldest_datfrozenxid))
269 ShmemVariableCache->oldestClogXid = oldest_datfrozenxid;
271 LWLockRelease(CLogTruncationLock);
275 * Determine the last safe XID to allocate using the currently oldest
276 * datfrozenxid (ie, the oldest XID that might exist in any database
277 * of our cluster), and the OID of the (or a) database with that value.
280 SetTransactionIdLimit(TransactionId oldest_datfrozenxid, Oid oldest_datoid)
282 TransactionId xidVacLimit;
283 TransactionId xidWarnLimit;
284 TransactionId xidStopLimit;
285 TransactionId xidWrapLimit;
286 TransactionId curXid;
288 Assert(TransactionIdIsNormal(oldest_datfrozenxid));
291 * The place where we actually get into deep trouble is halfway around
292 * from the oldest potentially-existing XID. (This calculation is
293 * probably off by one or two counts, because the special XIDs reduce the
294 * size of the loop a little bit. But we throw in plenty of slop below,
295 * so it doesn't matter.)
297 xidWrapLimit = oldest_datfrozenxid + (MaxTransactionId >> 1);
298 if (xidWrapLimit < FirstNormalTransactionId)
299 xidWrapLimit += FirstNormalTransactionId;
302 * We'll refuse to continue assigning XIDs in interactive mode once we get
303 * within 1M transactions of data loss. This leaves lots of room for the
304 * DBA to fool around fixing things in a standalone backend, while not
305 * being significant compared to total XID space. (Note that since
306 * vacuuming requires one transaction per table cleaned, we had better be
307 * sure there's lots of XIDs left...)
309 xidStopLimit = xidWrapLimit - 1000000;
310 if (xidStopLimit < FirstNormalTransactionId)
311 xidStopLimit -= FirstNormalTransactionId;
314 * We'll start complaining loudly when we get within 10M transactions of
315 * the stop point. This is kind of arbitrary, but if you let your gas
316 * gauge get down to 1% of full, would you be looking for the next gas
317 * station? We need to be fairly liberal about this number because there
318 * are lots of scenarios where most transactions are done by automatic
319 * clients that won't pay attention to warnings. (No, we're not gonna make
320 * this configurable. If you know enough to configure it, you know enough
321 * to not get in this kind of trouble in the first place.)
323 xidWarnLimit = xidStopLimit - 10000000;
324 if (xidWarnLimit < FirstNormalTransactionId)
325 xidWarnLimit -= FirstNormalTransactionId;
328 * We'll start trying to force autovacuums when oldest_datfrozenxid gets
329 * to be more than autovacuum_freeze_max_age transactions old.
331 * Note: guc.c ensures that autovacuum_freeze_max_age is in a sane range,
332 * so that xidVacLimit will be well before xidWarnLimit.
334 * Note: autovacuum_freeze_max_age is a PGC_POSTMASTER parameter so that
335 * we don't have to worry about dealing with on-the-fly changes in its
336 * value. It doesn't look practical to update shared state from a GUC
337 * assign hook (too many processes would try to execute the hook,
338 * resulting in race conditions as well as crashes of those not connected
339 * to shared memory). Perhaps this can be improved someday. See also
340 * SetMultiXactIdLimit.
342 xidVacLimit = oldest_datfrozenxid + autovacuum_freeze_max_age;
343 if (xidVacLimit < FirstNormalTransactionId)
344 xidVacLimit += FirstNormalTransactionId;
346 /* Grab lock for just long enough to set the new limit values */
347 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
348 ShmemVariableCache->oldestXid = oldest_datfrozenxid;
349 ShmemVariableCache->xidVacLimit = xidVacLimit;
350 ShmemVariableCache->xidWarnLimit = xidWarnLimit;
351 ShmemVariableCache->xidStopLimit = xidStopLimit;
352 ShmemVariableCache->xidWrapLimit = xidWrapLimit;
353 ShmemVariableCache->oldestXidDB = oldest_datoid;
354 curXid = ShmemVariableCache->nextXid;
355 LWLockRelease(XidGenLock);
359 (errmsg("transaction ID wrap limit is %u, limited by database with OID %u",
360 xidWrapLimit, oldest_datoid)));
363 * If past the autovacuum force point, immediately signal an autovac
364 * request. The reason for this is that autovac only processes one
365 * database per invocation. Once it's finished cleaning up the oldest
366 * database, it'll call here, and we'll signal the postmaster to start
367 * another iteration immediately if there are still any old databases.
369 if (TransactionIdFollowsOrEquals(curXid, xidVacLimit) &&
370 IsUnderPostmaster && !InRecovery)
371 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
373 /* Give an immediate warning if past the wrap warn point */
374 if (TransactionIdFollowsOrEquals(curXid, xidWarnLimit) && !InRecovery)
376 char *oldest_datname;
379 * We can be called when not inside a transaction, for example during
380 * StartupXLOG(). In such a case we cannot do database access, so we
381 * must just report the oldest DB's OID.
383 * Note: it's also possible that get_database_name fails and returns
384 * NULL, for example because the database just got dropped. We'll
385 * still warn, even though the warning might now be unnecessary.
387 if (IsTransactionState())
388 oldest_datname = get_database_name(oldest_datoid);
390 oldest_datname = NULL;
394 (errmsg("database \"%s\" must be vacuumed within %u transactions",
396 xidWrapLimit - curXid),
397 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
398 "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
401 (errmsg("database with OID %u must be vacuumed within %u transactions",
403 xidWrapLimit - curXid),
404 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
405 "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
411 * ForceTransactionIdLimitUpdate -- does the XID wrap-limit data need updating?
413 * We primarily check whether oldestXidDB is valid. The cases we have in
414 * mind are that that database was dropped, or the field was reset to zero
415 * by pg_resetwal. In either case we should force recalculation of the
416 * wrap limit. Also do it if oldestXid is old enough to be forcing
417 * autovacuums or other actions; this ensures we update our state as soon
418 * as possible once extra overhead is being incurred.
421 ForceTransactionIdLimitUpdate(void)
423 TransactionId nextXid;
424 TransactionId xidVacLimit;
425 TransactionId oldestXid;
428 /* Locking is probably not really necessary, but let's be careful */
429 LWLockAcquire(XidGenLock, LW_SHARED);
430 nextXid = ShmemVariableCache->nextXid;
431 xidVacLimit = ShmemVariableCache->xidVacLimit;
432 oldestXid = ShmemVariableCache->oldestXid;
433 oldestXidDB = ShmemVariableCache->oldestXidDB;
434 LWLockRelease(XidGenLock);
436 if (!TransactionIdIsNormal(oldestXid))
437 return true; /* shouldn't happen, but just in case */
438 if (!TransactionIdIsValid(xidVacLimit))
439 return true; /* this shouldn't happen anymore either */
440 if (TransactionIdFollowsOrEquals(nextXid, xidVacLimit))
441 return true; /* past VacLimit, don't delay updating */
442 if (!SearchSysCacheExists1(DATABASEOID, ObjectIdGetDatum(oldestXidDB)))
443 return true; /* could happen, per comments above */
449 * GetNewObjectId -- allocate a new OID
451 * OIDs are generated by a cluster-wide counter. Since they are only 32 bits
452 * wide, counter wraparound will occur eventually, and therefore it is unwise
453 * to assume they are unique unless precautions are taken to make them so.
454 * Hence, this routine should generally not be used directly. The only direct
455 * callers should be GetNewOidWithIndex() and GetNewRelFileNode() in
463 /* safety check, we should never get this far in a HS standby */
464 if (RecoveryInProgress())
465 elog(ERROR, "cannot assign OIDs during recovery");
467 LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
470 * Check for wraparound of the OID counter. We *must* not return 0
471 * (InvalidOid), and in normal operation we mustn't return anything below
472 * FirstNormalObjectId since that range is reserved for initdb (see
473 * IsCatalogClass()). Note we are relying on unsigned comparison.
475 * During initdb, we start the OID generator at FirstBootstrapObjectId, so
476 * we only wrap if before that point when in bootstrap or standalone mode.
477 * The first time through this routine after normal postmaster start, the
478 * counter will be forced up to FirstNormalObjectId. This mechanism
479 * leaves the OIDs between FirstBootstrapObjectId and FirstNormalObjectId
480 * available for automatic assignment during initdb, while ensuring they
481 * will never conflict with user-assigned OIDs.
483 if (ShmemVariableCache->nextOid < ((Oid) FirstNormalObjectId))
485 if (IsPostmasterEnvironment)
487 /* wraparound, or first post-initdb assignment, in normal mode */
488 ShmemVariableCache->nextOid = FirstNormalObjectId;
489 ShmemVariableCache->oidCount = 0;
493 /* we may be bootstrapping, so don't enforce the full range */
494 if (ShmemVariableCache->nextOid < ((Oid) FirstBootstrapObjectId))
496 /* wraparound in standalone mode (unlikely but possible) */
497 ShmemVariableCache->nextOid = FirstNormalObjectId;
498 ShmemVariableCache->oidCount = 0;
503 /* If we run out of logged for use oids then we must log more */
504 if (ShmemVariableCache->oidCount == 0)
506 XLogPutNextOid(ShmemVariableCache->nextOid + VAR_OID_PREFETCH);
507 ShmemVariableCache->oidCount = VAR_OID_PREFETCH;
510 result = ShmemVariableCache->nextOid;
512 (ShmemVariableCache->nextOid)++;
513 (ShmemVariableCache->oidCount)--;
515 LWLockRelease(OidGenLock);