1 /*-------------------------------------------------------------------------
4 * postgres OID & XID variables support routines
6 * Copyright (c) 2000-2014, PostgreSQL Global Development Group
9 * src/backend/access/transam/varsup.c
11 *-------------------------------------------------------------------------
16 #include "access/clog.h"
17 #include "access/subtrans.h"
18 #include "access/transam.h"
19 #include "access/xact.h"
20 #include "access/xlog.h"
21 #include "commands/dbcommands.h"
22 #include "miscadmin.h"
23 #include "postmaster/autovacuum.h"
24 #include "storage/pmsignal.h"
25 #include "storage/proc.h"
26 #include "utils/syscache.h"
29 /* Number of OIDs to prefetch (preallocate) per XLOG write */
30 #define VAR_OID_PREFETCH 8192
32 /* pointer to "variable cache" in shared memory (set up by shmem.c) */
33 VariableCache ShmemVariableCache = NULL;
37 * Allocate the next XID for a new transaction or subtransaction.
39 * The new XID is also stored into MyPgXact before returning.
41 * Note: when this is called, we are actually already inside a valid
42 * transaction, since XIDs are now not allocated until the transaction
43 * does something. So it is safe to do a database lookup if we want to
44 * issue a warning about XID wrap.
47 GetNewTransactionId(bool isSubXact)
52 * During bootstrap initialization, we return the special bootstrap
55 if (IsBootstrapProcessingMode())
58 MyPgXact->xid = BootstrapTransactionId;
59 return BootstrapTransactionId;
62 /* safety check, we should never get this far in a HS slave */
63 if (RecoveryInProgress())
64 elog(ERROR, "cannot assign TransactionIds during recovery");
66 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
68 xid = ShmemVariableCache->nextXid;
71 * Check to see if it's safe to assign another XID. This protects against
72 * catastrophic data loss due to XID wraparound. The basic rules are:
74 * If we're past xidVacLimit, start trying to force autovacuum cycles.
75 * If we're past xidWarnLimit, start issuing warnings.
76 * If we're past xidStopLimit, refuse to execute transactions, unless
77 * we are running in single-user mode (which gives an escape hatch
78 * to the DBA who somehow got past the earlier defenses).
80 * Note that this coding also appears in GetNewMultiXactId.
83 if (TransactionIdFollowsOrEquals(xid, ShmemVariableCache->xidVacLimit))
86 * For safety's sake, we release XidGenLock while sending signals,
87 * warnings, etc. This is not so much because we care about
88 * preserving concurrency in this situation, as to avoid any
89 * possibility of deadlock while doing get_database_name(). First,
90 * copy all the shared values we'll need in this path.
92 TransactionId xidWarnLimit = ShmemVariableCache->xidWarnLimit;
93 TransactionId xidStopLimit = ShmemVariableCache->xidStopLimit;
94 TransactionId xidWrapLimit = ShmemVariableCache->xidWrapLimit;
95 Oid oldest_datoid = ShmemVariableCache->oldestXidDB;
97 LWLockRelease(XidGenLock);
100 * To avoid swamping the postmaster with signals, we issue the autovac
101 * request only once per 64K transaction starts. This still gives
102 * plenty of chances before we get into real trouble.
104 if (IsUnderPostmaster && (xid % 65536) == 0)
105 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
107 if (IsUnderPostmaster &&
108 TransactionIdFollowsOrEquals(xid, xidStopLimit))
110 char *oldest_datname = get_database_name(oldest_datoid);
112 /* complain even if that DB has disappeared */
115 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
116 errmsg("database is not accepting commands to avoid wraparound data loss in database \"%s\"",
118 errhint("Stop the postmaster and vacuum that database in single-user mode.\n"
119 "You might also need to commit or roll back old prepared transactions.")));
122 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
123 errmsg("database is not accepting commands to avoid wraparound data loss in database with OID %u",
125 errhint("Stop the postmaster and vacuum that database in single-user mode.\n"
126 "You might also need to commit or roll back old prepared transactions.")));
128 else if (TransactionIdFollowsOrEquals(xid, xidWarnLimit))
130 char *oldest_datname = get_database_name(oldest_datoid);
132 /* complain even if that DB has disappeared */
135 (errmsg("database \"%s\" must be vacuumed within %u transactions",
138 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
139 "You might also need to commit or roll back old prepared transactions.")));
142 (errmsg("database with OID %u must be vacuumed within %u transactions",
145 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
146 "You might also need to commit or roll back old prepared transactions.")));
149 /* Re-acquire lock and start over */
150 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
151 xid = ShmemVariableCache->nextXid;
155 * If we are allocating the first XID of a new page of the commit log,
156 * zero out that commit-log page before returning. We must do this while
157 * holding XidGenLock, else another xact could acquire and commit a later
158 * XID before we zero the page. Fortunately, a page of the commit log
159 * holds 32K or more transactions, so we don't have to do this very often.
161 * Extend pg_subtrans too.
167 * Now advance the nextXid counter. This must not happen until after we
168 * have successfully completed ExtendCLOG() --- if that routine fails, we
169 * want the next incoming transaction to try it again. We cannot assign
170 * more XIDs until there is CLOG space for them.
172 TransactionIdAdvance(ShmemVariableCache->nextXid);
175 * We must store the new XID into the shared ProcArray before releasing
176 * XidGenLock. This ensures that every active XID older than
177 * latestCompletedXid is present in the ProcArray, which is essential for
178 * correct OldestXmin tracking; see src/backend/access/transam/README.
180 * XXX by storing xid into MyPgXact without acquiring ProcArrayLock, we
181 * are relying on fetch/store of an xid to be atomic, else other backends
182 * might see a partially-set xid here. But holding both locks at once
183 * would be a nasty concurrency hit. So for now, assume atomicity.
185 * Note that readers of PGXACT xid fields should be careful to fetch the
186 * value only once, rather than assume they can read a value multiple
187 * times and get the same answer each time.
189 * The same comments apply to the subxact xid count and overflow fields.
191 * A solution to the atomic-store problem would be to give each PGXACT its
192 * own spinlock used only for fetching/storing that PGXACT's xid and
195 * If there's no room to fit a subtransaction XID into PGPROC, set the
196 * cache-overflowed flag instead. This forces readers to look in
197 * pg_subtrans to map subtransaction XIDs up to top-level XIDs. There is a
198 * race-condition window, in that the new XID will not appear as running
199 * until its parent link has been placed into pg_subtrans. However, that
200 * will happen before anyone could possibly have a reason to inquire about
201 * the status of the XID, so it seems OK. (Snapshots taken during this
202 * window *will* include the parent XID, so they will deliver the correct
203 * answer later on when someone does have a reason to inquire.)
207 * Use volatile pointer to prevent code rearrangement; other backends
208 * could be examining my subxids info concurrently, and we don't want
209 * them to see an invalid intermediate state, such as incrementing
210 * nxids before filling the array entry. Note we are assuming that
211 * TransactionId and int fetch/store are atomic.
213 volatile PGPROC *myproc = MyProc;
214 volatile PGXACT *mypgxact = MyPgXact;
220 int nxids = mypgxact->nxids;
222 if (nxids < PGPROC_MAX_CACHED_SUBXIDS)
224 myproc->subxids.xids[nxids] = xid;
225 mypgxact->nxids = nxids + 1;
228 mypgxact->overflowed = true;
232 LWLockRelease(XidGenLock);
238 * Read nextXid but don't allocate it.
241 ReadNewTransactionId(void)
245 LWLockAcquire(XidGenLock, LW_SHARED);
246 xid = ShmemVariableCache->nextXid;
247 LWLockRelease(XidGenLock);
253 * Determine the last safe XID to allocate given the currently oldest
254 * datfrozenxid (ie, the oldest XID that might exist in any database
255 * of our cluster), and the OID of the (or a) database with that value.
258 SetTransactionIdLimit(TransactionId oldest_datfrozenxid, Oid oldest_datoid)
260 TransactionId xidVacLimit;
261 TransactionId xidWarnLimit;
262 TransactionId xidStopLimit;
263 TransactionId xidWrapLimit;
264 TransactionId curXid;
266 Assert(TransactionIdIsNormal(oldest_datfrozenxid));
269 * The place where we actually get into deep trouble is halfway around
270 * from the oldest potentially-existing XID. (This calculation is
271 * probably off by one or two counts, because the special XIDs reduce the
272 * size of the loop a little bit. But we throw in plenty of slop below,
273 * so it doesn't matter.)
275 xidWrapLimit = oldest_datfrozenxid + (MaxTransactionId >> 1);
276 if (xidWrapLimit < FirstNormalTransactionId)
277 xidWrapLimit += FirstNormalTransactionId;
280 * We'll refuse to continue assigning XIDs in interactive mode once we get
281 * within 1M transactions of data loss. This leaves lots of room for the
282 * DBA to fool around fixing things in a standalone backend, while not
283 * being significant compared to total XID space. (Note that since
284 * vacuuming requires one transaction per table cleaned, we had better be
285 * sure there's lots of XIDs left...)
287 xidStopLimit = xidWrapLimit - 1000000;
288 if (xidStopLimit < FirstNormalTransactionId)
289 xidStopLimit -= FirstNormalTransactionId;
292 * We'll start complaining loudly when we get within 10M transactions of
293 * the stop point. This is kind of arbitrary, but if you let your gas
294 * gauge get down to 1% of full, would you be looking for the next gas
295 * station? We need to be fairly liberal about this number because there
296 * are lots of scenarios where most transactions are done by automatic
297 * clients that won't pay attention to warnings. (No, we're not gonna make
298 * this configurable. If you know enough to configure it, you know enough
299 * to not get in this kind of trouble in the first place.)
301 xidWarnLimit = xidStopLimit - 10000000;
302 if (xidWarnLimit < FirstNormalTransactionId)
303 xidWarnLimit -= FirstNormalTransactionId;
306 * We'll start trying to force autovacuums when oldest_datfrozenxid gets
307 * to be more than autovacuum_freeze_max_age transactions old.
309 * Note: guc.c ensures that autovacuum_freeze_max_age is in a sane range,
310 * so that xidVacLimit will be well before xidWarnLimit.
312 * Note: autovacuum_freeze_max_age is a PGC_POSTMASTER parameter so that
313 * we don't have to worry about dealing with on-the-fly changes in its
314 * value. It doesn't look practical to update shared state from a GUC
315 * assign hook (too many processes would try to execute the hook,
316 * resulting in race conditions as well as crashes of those not connected
317 * to shared memory). Perhaps this can be improved someday. See also
318 * SetMultiXactIdLimit.
320 xidVacLimit = oldest_datfrozenxid + autovacuum_freeze_max_age;
321 if (xidVacLimit < FirstNormalTransactionId)
322 xidVacLimit += FirstNormalTransactionId;
324 /* Grab lock for just long enough to set the new limit values */
325 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
326 ShmemVariableCache->oldestXid = oldest_datfrozenxid;
327 ShmemVariableCache->xidVacLimit = xidVacLimit;
328 ShmemVariableCache->xidWarnLimit = xidWarnLimit;
329 ShmemVariableCache->xidStopLimit = xidStopLimit;
330 ShmemVariableCache->xidWrapLimit = xidWrapLimit;
331 ShmemVariableCache->oldestXidDB = oldest_datoid;
332 curXid = ShmemVariableCache->nextXid;
333 LWLockRelease(XidGenLock);
337 (errmsg("transaction ID wrap limit is %u, limited by database with OID %u",
338 xidWrapLimit, oldest_datoid)));
341 * If past the autovacuum force point, immediately signal an autovac
342 * request. The reason for this is that autovac only processes one
343 * database per invocation. Once it's finished cleaning up the oldest
344 * database, it'll call here, and we'll signal the postmaster to start
345 * another iteration immediately if there are still any old databases.
347 if (TransactionIdFollowsOrEquals(curXid, xidVacLimit) &&
348 IsUnderPostmaster && !InRecovery)
349 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
351 /* Give an immediate warning if past the wrap warn point */
352 if (TransactionIdFollowsOrEquals(curXid, xidWarnLimit) && !InRecovery)
354 char *oldest_datname;
357 * We can be called when not inside a transaction, for example during
358 * StartupXLOG(). In such a case we cannot do database access, so we
359 * must just report the oldest DB's OID.
361 * Note: it's also possible that get_database_name fails and returns
362 * NULL, for example because the database just got dropped. We'll
363 * still warn, even though the warning might now be unnecessary.
365 if (IsTransactionState())
366 oldest_datname = get_database_name(oldest_datoid);
368 oldest_datname = NULL;
372 (errmsg("database \"%s\" must be vacuumed within %u transactions",
374 xidWrapLimit - curXid),
375 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
376 "You might also need to commit or roll back old prepared transactions.")));
379 (errmsg("database with OID %u must be vacuumed within %u transactions",
381 xidWrapLimit - curXid),
382 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
383 "You might also need to commit or roll back old prepared transactions.")));
389 * ForceTransactionIdLimitUpdate -- does the XID wrap-limit data need updating?
391 * We primarily check whether oldestXidDB is valid. The cases we have in
392 * mind are that that database was dropped, or the field was reset to zero
393 * by pg_resetxlog. In either case we should force recalculation of the
394 * wrap limit. Also do it if oldestXid is old enough to be forcing
395 * autovacuums or other actions; this ensures we update our state as soon
396 * as possible once extra overhead is being incurred.
399 ForceTransactionIdLimitUpdate(void)
401 TransactionId nextXid;
402 TransactionId xidVacLimit;
403 TransactionId oldestXid;
406 /* Locking is probably not really necessary, but let's be careful */
407 LWLockAcquire(XidGenLock, LW_SHARED);
408 nextXid = ShmemVariableCache->nextXid;
409 xidVacLimit = ShmemVariableCache->xidVacLimit;
410 oldestXid = ShmemVariableCache->oldestXid;
411 oldestXidDB = ShmemVariableCache->oldestXidDB;
412 LWLockRelease(XidGenLock);
414 if (!TransactionIdIsNormal(oldestXid))
415 return true; /* shouldn't happen, but just in case */
416 if (!TransactionIdIsValid(xidVacLimit))
417 return true; /* this shouldn't happen anymore either */
418 if (TransactionIdFollowsOrEquals(nextXid, xidVacLimit))
419 return true; /* past VacLimit, don't delay updating */
420 if (!SearchSysCacheExists1(DATABASEOID, ObjectIdGetDatum(oldestXidDB)))
421 return true; /* could happen, per comments above */
427 * GetNewObjectId -- allocate a new OID
429 * OIDs are generated by a cluster-wide counter. Since they are only 32 bits
430 * wide, counter wraparound will occur eventually, and therefore it is unwise
431 * to assume they are unique unless precautions are taken to make them so.
432 * Hence, this routine should generally not be used directly. The only
433 * direct callers should be GetNewOid() and GetNewRelFileNode() in
441 /* safety check, we should never get this far in a HS slave */
442 if (RecoveryInProgress())
443 elog(ERROR, "cannot assign OIDs during recovery");
445 LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
448 * Check for wraparound of the OID counter. We *must* not return 0
449 * (InvalidOid); and as long as we have to check that, it seems a good
450 * idea to skip over everything below FirstNormalObjectId too. (This
451 * basically just avoids lots of collisions with bootstrap-assigned OIDs
452 * right after a wrap occurs, so as to avoid a possibly large number of
453 * iterations in GetNewOid.) Note we are relying on unsigned comparison.
455 * During initdb, we start the OID generator at FirstBootstrapObjectId, so
456 * we only wrap if before that point when in bootstrap or standalone mode.
457 * The first time through this routine after normal postmaster start, the
458 * counter will be forced up to FirstNormalObjectId. This mechanism
459 * leaves the OIDs between FirstBootstrapObjectId and FirstNormalObjectId
460 * available for automatic assignment during initdb, while ensuring they
461 * will never conflict with user-assigned OIDs.
463 if (ShmemVariableCache->nextOid < ((Oid) FirstNormalObjectId))
465 if (IsPostmasterEnvironment)
467 /* wraparound, or first post-initdb assignment, in normal mode */
468 ShmemVariableCache->nextOid = FirstNormalObjectId;
469 ShmemVariableCache->oidCount = 0;
473 /* we may be bootstrapping, so don't enforce the full range */
474 if (ShmemVariableCache->nextOid < ((Oid) FirstBootstrapObjectId))
476 /* wraparound in standalone mode (unlikely but possible) */
477 ShmemVariableCache->nextOid = FirstNormalObjectId;
478 ShmemVariableCache->oidCount = 0;
483 /* If we run out of logged for use oids then we must log more */
484 if (ShmemVariableCache->oidCount == 0)
486 XLogPutNextOid(ShmemVariableCache->nextOid + VAR_OID_PREFETCH);
487 ShmemVariableCache->oidCount = VAR_OID_PREFETCH;
490 result = ShmemVariableCache->nextOid;
492 (ShmemVariableCache->nextOid)++;
493 (ShmemVariableCache->oidCount)--;
495 LWLockRelease(OidGenLock);