1 /*-------------------------------------------------------------------------
4 * postgres OID & XID variables support routines
6 * Copyright (c) 2000-2017, 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 slave */
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.")));
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.")));
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.")));
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.")));
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 * XXX by storing xid into MyPgXact without acquiring ProcArrayLock, we
190 * are relying on fetch/store of an xid to be atomic, else other backends
191 * might see a partially-set xid here. But holding both locks at once
192 * would be a nasty concurrency hit. So for now, assume atomicity.
194 * Note that readers of PGXACT xid fields should be careful to fetch the
195 * value only once, rather than assume they can read a value multiple
196 * times and get the same answer each time.
198 * The same comments apply to the subxact xid count and overflow fields.
200 * A solution to the atomic-store problem would be to give each PGXACT its
201 * own spinlock used only for fetching/storing that PGXACT's xid and
204 * If there's no room to fit a subtransaction XID into PGPROC, set the
205 * cache-overflowed flag instead. This forces readers to look in
206 * pg_subtrans to map subtransaction XIDs up to top-level XIDs. There is a
207 * race-condition window, in that the new XID will not appear as running
208 * until its parent link has been placed into pg_subtrans. However, that
209 * will happen before anyone could possibly have a reason to inquire about
210 * the status of the XID, so it seems OK. (Snapshots taken during this
211 * window *will* include the parent XID, so they will deliver the correct
212 * answer later on when someone does have a reason to inquire.)
216 * Use volatile pointer to prevent code rearrangement; other backends
217 * could be examining my subxids info concurrently, and we don't want
218 * them to see an invalid intermediate state, such as incrementing
219 * nxids before filling the array entry. Note we are assuming that
220 * TransactionId and int fetch/store are atomic.
222 volatile PGPROC *myproc = MyProc;
223 volatile PGXACT *mypgxact = MyPgXact;
229 int nxids = mypgxact->nxids;
231 if (nxids < PGPROC_MAX_CACHED_SUBXIDS)
233 myproc->subxids.xids[nxids] = xid;
234 mypgxact->nxids = nxids + 1;
237 mypgxact->overflowed = true;
241 LWLockRelease(XidGenLock);
247 * Read nextXid but don't allocate it.
250 ReadNewTransactionId(void)
254 LWLockAcquire(XidGenLock, LW_SHARED);
255 xid = ShmemVariableCache->nextXid;
256 LWLockRelease(XidGenLock);
262 * Advance the cluster-wide value for the oldest valid clog entry.
264 * We must acquire CLogTruncationLock to advance the oldestClogXid. It's not
265 * necessary to hold the lock during the actual clog truncation, only when we
266 * advance the limit, as code looking up arbitrary xids is required to hold
267 * CLogTruncationLock from when it tests oldestClogXid through to when it
268 * completes the clog lookup.
271 AdvanceOldestClogXid(TransactionId oldest_datfrozenxid)
273 LWLockAcquire(CLogTruncationLock, LW_EXCLUSIVE);
274 if (TransactionIdPrecedes(ShmemVariableCache->oldestClogXid,
275 oldest_datfrozenxid))
277 ShmemVariableCache->oldestClogXid = oldest_datfrozenxid;
279 LWLockRelease(CLogTruncationLock);
283 * Determine the last safe XID to allocate using the currently oldest
284 * datfrozenxid (ie, the oldest XID that might exist in any database
285 * of our cluster), and the OID of the (or a) database with that value.
288 SetTransactionIdLimit(TransactionId oldest_datfrozenxid, Oid oldest_datoid)
290 TransactionId xidVacLimit;
291 TransactionId xidWarnLimit;
292 TransactionId xidStopLimit;
293 TransactionId xidWrapLimit;
294 TransactionId curXid;
296 Assert(TransactionIdIsNormal(oldest_datfrozenxid));
299 * The place where we actually get into deep trouble is halfway around
300 * from the oldest potentially-existing XID. (This calculation is
301 * probably off by one or two counts, because the special XIDs reduce the
302 * size of the loop a little bit. But we throw in plenty of slop below,
303 * so it doesn't matter.)
305 xidWrapLimit = oldest_datfrozenxid + (MaxTransactionId >> 1);
306 if (xidWrapLimit < FirstNormalTransactionId)
307 xidWrapLimit += FirstNormalTransactionId;
310 * We'll refuse to continue assigning XIDs in interactive mode once we get
311 * within 1M transactions of data loss. This leaves lots of room for the
312 * DBA to fool around fixing things in a standalone backend, while not
313 * being significant compared to total XID space. (Note that since
314 * vacuuming requires one transaction per table cleaned, we had better be
315 * sure there's lots of XIDs left...)
317 xidStopLimit = xidWrapLimit - 1000000;
318 if (xidStopLimit < FirstNormalTransactionId)
319 xidStopLimit -= FirstNormalTransactionId;
322 * We'll start complaining loudly when we get within 10M transactions of
323 * the stop point. This is kind of arbitrary, but if you let your gas
324 * gauge get down to 1% of full, would you be looking for the next gas
325 * station? We need to be fairly liberal about this number because there
326 * are lots of scenarios where most transactions are done by automatic
327 * clients that won't pay attention to warnings. (No, we're not gonna make
328 * this configurable. If you know enough to configure it, you know enough
329 * to not get in this kind of trouble in the first place.)
331 xidWarnLimit = xidStopLimit - 10000000;
332 if (xidWarnLimit < FirstNormalTransactionId)
333 xidWarnLimit -= FirstNormalTransactionId;
336 * We'll start trying to force autovacuums when oldest_datfrozenxid gets
337 * to be more than autovacuum_freeze_max_age transactions old.
339 * Note: guc.c ensures that autovacuum_freeze_max_age is in a sane range,
340 * so that xidVacLimit will be well before xidWarnLimit.
342 * Note: autovacuum_freeze_max_age is a PGC_POSTMASTER parameter so that
343 * we don't have to worry about dealing with on-the-fly changes in its
344 * value. It doesn't look practical to update shared state from a GUC
345 * assign hook (too many processes would try to execute the hook,
346 * resulting in race conditions as well as crashes of those not connected
347 * to shared memory). Perhaps this can be improved someday. See also
348 * SetMultiXactIdLimit.
350 xidVacLimit = oldest_datfrozenxid + autovacuum_freeze_max_age;
351 if (xidVacLimit < FirstNormalTransactionId)
352 xidVacLimit += FirstNormalTransactionId;
354 /* Grab lock for just long enough to set the new limit values */
355 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
356 ShmemVariableCache->oldestXid = oldest_datfrozenxid;
357 ShmemVariableCache->xidVacLimit = xidVacLimit;
358 ShmemVariableCache->xidWarnLimit = xidWarnLimit;
359 ShmemVariableCache->xidStopLimit = xidStopLimit;
360 ShmemVariableCache->xidWrapLimit = xidWrapLimit;
361 ShmemVariableCache->oldestXidDB = oldest_datoid;
362 curXid = ShmemVariableCache->nextXid;
363 LWLockRelease(XidGenLock);
367 (errmsg("transaction ID wrap limit is %u, limited by database with OID %u",
368 xidWrapLimit, oldest_datoid)));
371 * If past the autovacuum force point, immediately signal an autovac
372 * request. The reason for this is that autovac only processes one
373 * database per invocation. Once it's finished cleaning up the oldest
374 * database, it'll call here, and we'll signal the postmaster to start
375 * another iteration immediately if there are still any old databases.
377 if (TransactionIdFollowsOrEquals(curXid, xidVacLimit) &&
378 IsUnderPostmaster && !InRecovery)
379 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
381 /* Give an immediate warning if past the wrap warn point */
382 if (TransactionIdFollowsOrEquals(curXid, xidWarnLimit) && !InRecovery)
384 char *oldest_datname;
387 * We can be called when not inside a transaction, for example during
388 * StartupXLOG(). In such a case we cannot do database access, so we
389 * must just report the oldest DB's OID.
391 * Note: it's also possible that get_database_name fails and returns
392 * NULL, for example because the database just got dropped. We'll
393 * still warn, even though the warning might now be unnecessary.
395 if (IsTransactionState())
396 oldest_datname = get_database_name(oldest_datoid);
398 oldest_datname = NULL;
402 (errmsg("database \"%s\" must be vacuumed within %u transactions",
404 xidWrapLimit - curXid),
405 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
406 "You might also need to commit or roll back old prepared transactions.")));
409 (errmsg("database with OID %u must be vacuumed within %u transactions",
411 xidWrapLimit - curXid),
412 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
413 "You might also need to commit or roll back old prepared transactions.")));
419 * ForceTransactionIdLimitUpdate -- does the XID wrap-limit data need updating?
421 * We primarily check whether oldestXidDB is valid. The cases we have in
422 * mind are that that database was dropped, or the field was reset to zero
423 * by pg_resetwal. In either case we should force recalculation of the
424 * wrap limit. Also do it if oldestXid is old enough to be forcing
425 * autovacuums or other actions; this ensures we update our state as soon
426 * as possible once extra overhead is being incurred.
429 ForceTransactionIdLimitUpdate(void)
431 TransactionId nextXid;
432 TransactionId xidVacLimit;
433 TransactionId oldestXid;
436 /* Locking is probably not really necessary, but let's be careful */
437 LWLockAcquire(XidGenLock, LW_SHARED);
438 nextXid = ShmemVariableCache->nextXid;
439 xidVacLimit = ShmemVariableCache->xidVacLimit;
440 oldestXid = ShmemVariableCache->oldestXid;
441 oldestXidDB = ShmemVariableCache->oldestXidDB;
442 LWLockRelease(XidGenLock);
444 if (!TransactionIdIsNormal(oldestXid))
445 return true; /* shouldn't happen, but just in case */
446 if (!TransactionIdIsValid(xidVacLimit))
447 return true; /* this shouldn't happen anymore either */
448 if (TransactionIdFollowsOrEquals(nextXid, xidVacLimit))
449 return true; /* past VacLimit, don't delay updating */
450 if (!SearchSysCacheExists1(DATABASEOID, ObjectIdGetDatum(oldestXidDB)))
451 return true; /* could happen, per comments above */
457 * GetNewObjectId -- allocate a new OID
459 * OIDs are generated by a cluster-wide counter. Since they are only 32 bits
460 * wide, counter wraparound will occur eventually, and therefore it is unwise
461 * to assume they are unique unless precautions are taken to make them so.
462 * Hence, this routine should generally not be used directly. The only
463 * direct callers should be GetNewOid() and GetNewRelFileNode() in
471 /* safety check, we should never get this far in a HS slave */
472 if (RecoveryInProgress())
473 elog(ERROR, "cannot assign OIDs during recovery");
475 LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
478 * Check for wraparound of the OID counter. We *must* not return 0
479 * (InvalidOid); and as long as we have to check that, it seems a good
480 * idea to skip over everything below FirstNormalObjectId too. (This
481 * basically just avoids lots of collisions with bootstrap-assigned OIDs
482 * right after a wrap occurs, so as to avoid a possibly large number of
483 * iterations in GetNewOid.) Note we are relying on unsigned comparison.
485 * During initdb, we start the OID generator at FirstBootstrapObjectId, so
486 * we only wrap if before that point when in bootstrap or standalone mode.
487 * The first time through this routine after normal postmaster start, the
488 * counter will be forced up to FirstNormalObjectId. This mechanism
489 * leaves the OIDs between FirstBootstrapObjectId and FirstNormalObjectId
490 * available for automatic assignment during initdb, while ensuring they
491 * will never conflict with user-assigned OIDs.
493 if (ShmemVariableCache->nextOid < ((Oid) FirstNormalObjectId))
495 if (IsPostmasterEnvironment)
497 /* wraparound, or first post-initdb assignment, in normal mode */
498 ShmemVariableCache->nextOid = FirstNormalObjectId;
499 ShmemVariableCache->oidCount = 0;
503 /* we may be bootstrapping, so don't enforce the full range */
504 if (ShmemVariableCache->nextOid < ((Oid) FirstBootstrapObjectId))
506 /* wraparound in standalone mode (unlikely but possible) */
507 ShmemVariableCache->nextOid = FirstNormalObjectId;
508 ShmemVariableCache->oidCount = 0;
513 /* If we run out of logged for use oids then we must log more */
514 if (ShmemVariableCache->oidCount == 0)
516 XLogPutNextOid(ShmemVariableCache->nextOid + VAR_OID_PREFETCH);
517 ShmemVariableCache->oidCount = VAR_OID_PREFETCH;
520 result = ShmemVariableCache->nextOid;
522 (ShmemVariableCache->nextOid)++;
523 (ShmemVariableCache->oidCount)--;
525 LWLockRelease(OidGenLock);