1 /*-------------------------------------------------------------------------
4 * postgres OID & XID variables support routines
6 * Copyright (c) 2000-2010, 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 "commands/dbcommands.h"
20 #include "miscadmin.h"
21 #include "postmaster/autovacuum.h"
22 #include "storage/pmsignal.h"
23 #include "storage/proc.h"
24 #include "utils/builtins.h"
25 #include "utils/syscache.h"
28 /* Number of OIDs to prefetch (preallocate) per XLOG write */
29 #define VAR_OID_PREFETCH 8192
31 /* pointer to "variable cache" in shared memory (set up by shmem.c) */
32 VariableCache ShmemVariableCache = NULL;
36 * Allocate the next XID for a new transaction or subtransaction.
38 * The new XID is also stored into MyProc before returning.
40 * Note: when this is called, we are actually already inside a valid
41 * transaction, since XIDs are now not allocated until the transaction
42 * does something. So it is safe to do a database lookup if we want to
43 * issue a warning about XID wrap.
46 GetNewTransactionId(bool isSubXact)
51 * During bootstrap initialization, we return the special bootstrap
54 if (IsBootstrapProcessingMode())
57 MyProc->xid = BootstrapTransactionId;
58 return BootstrapTransactionId;
61 /* safety check, we should never get this far in a HS slave */
62 if (RecoveryInProgress())
63 elog(ERROR, "cannot assign TransactionIds during recovery");
65 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
67 xid = ShmemVariableCache->nextXid;
70 * Check to see if it's safe to assign another XID. This protects against
71 * catastrophic data loss due to XID wraparound. The basic rules are:
73 * If we're past xidVacLimit, start trying to force autovacuum cycles.
74 * If we're past xidWarnLimit, start issuing warnings.
75 * If we're past xidStopLimit, refuse to execute transactions, unless
76 * we are running in a standalone backend (which gives an escape hatch
77 * to the DBA who somehow got past the earlier defenses).
80 if (TransactionIdFollowsOrEquals(xid, ShmemVariableCache->xidVacLimit))
83 * For safety's sake, we release XidGenLock while sending signals,
84 * warnings, etc. This is not so much because we care about
85 * preserving concurrency in this situation, as to avoid any
86 * possibility of deadlock while doing get_database_name(). First,
87 * copy all the shared values we'll need in this path.
89 TransactionId xidWarnLimit = ShmemVariableCache->xidWarnLimit;
90 TransactionId xidStopLimit = ShmemVariableCache->xidStopLimit;
91 TransactionId xidWrapLimit = ShmemVariableCache->xidWrapLimit;
92 Oid oldest_datoid = ShmemVariableCache->oldestXidDB;
94 LWLockRelease(XidGenLock);
97 * To avoid swamping the postmaster with signals, we issue the autovac
98 * request only once per 64K transaction starts. This still gives
99 * plenty of chances before we get into real trouble.
101 if (IsUnderPostmaster && (xid % 65536) == 0)
102 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
104 if (IsUnderPostmaster &&
105 TransactionIdFollowsOrEquals(xid, xidStopLimit))
107 char *oldest_datname = get_database_name(oldest_datoid);
109 /* complain even if that DB has disappeared */
112 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
113 errmsg("database is not accepting commands to avoid wraparound data loss in database \"%s\"",
115 errhint("Stop the postmaster and use a standalone backend to vacuum that database.\n"
116 "You might also need to commit or roll back old prepared transactions.")));
119 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
120 errmsg("database is not accepting commands to avoid wraparound data loss in database with OID %u",
122 errhint("Stop the postmaster and use a standalone backend to vacuum that database.\n"
123 "You might also need to commit or roll back old prepared transactions.")));
125 else if (TransactionIdFollowsOrEquals(xid, xidWarnLimit))
127 char *oldest_datname = get_database_name(oldest_datoid);
129 /* complain even if that DB has disappeared */
132 (errmsg("database \"%s\" must be vacuumed within %u transactions",
135 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
136 "You might also need to commit or roll back old prepared transactions.")));
139 (errmsg("database with OID %u must be vacuumed within %u transactions",
142 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
143 "You might also need to commit or roll back old prepared transactions.")));
146 /* Re-acquire lock and start over */
147 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
148 xid = ShmemVariableCache->nextXid;
152 * If we are allocating the first XID of a new page of the commit log,
153 * zero out that commit-log page before returning. We must do this while
154 * holding XidGenLock, else another xact could acquire and commit a later
155 * XID before we zero the page. Fortunately, a page of the commit log
156 * holds 32K or more transactions, so we don't have to do this very often.
158 * Extend pg_subtrans too.
164 * Now advance the nextXid counter. This must not happen until after we
165 * have successfully completed ExtendCLOG() --- if that routine fails, we
166 * want the next incoming transaction to try it again. We cannot assign
167 * more XIDs until there is CLOG space for them.
169 TransactionIdAdvance(ShmemVariableCache->nextXid);
172 * We must store the new XID into the shared ProcArray before releasing
173 * XidGenLock. This ensures that every active XID older than
174 * latestCompletedXid is present in the ProcArray, which is essential for
175 * correct OldestXmin tracking; see src/backend/access/transam/README.
177 * XXX by storing xid into MyProc without acquiring ProcArrayLock, we are
178 * relying on fetch/store of an xid to be atomic, else other backends
179 * might see a partially-set xid here. But holding both locks at once
180 * would be a nasty concurrency hit. So for now, assume atomicity.
182 * Note that readers of PGPROC xid fields should be careful to fetch the
183 * value only once, rather than assume they can read a value multiple
184 * times and get the same answer each time.
186 * The same comments apply to the subxact xid count and overflow fields.
188 * A solution to the atomic-store problem would be to give each PGPROC its
189 * own spinlock used only for fetching/storing that PGPROC's xid and
192 * If there's no room to fit a subtransaction XID into PGPROC, set the
193 * cache-overflowed flag instead. This forces readers to look in
194 * pg_subtrans to map subtransaction XIDs up to top-level XIDs. There is a
195 * race-condition window, in that the new XID will not appear as running
196 * until its parent link has been placed into pg_subtrans. However, that
197 * will happen before anyone could possibly have a reason to inquire about
198 * the status of the XID, so it seems OK. (Snapshots taken during this
199 * window *will* include the parent XID, so they will deliver the correct
200 * answer later on when someone does have a reason to inquire.)
204 * Use volatile pointer to prevent code rearrangement; other backends
205 * could be examining my subxids info concurrently, and we don't want
206 * them to see an invalid intermediate state, such as incrementing
207 * nxids before filling the array entry. Note we are assuming that
208 * TransactionId and int fetch/store are atomic.
210 volatile PGPROC *myproc = MyProc;
216 int nxids = myproc->subxids.nxids;
218 if (nxids < PGPROC_MAX_CACHED_SUBXIDS)
220 myproc->subxids.xids[nxids] = xid;
221 myproc->subxids.nxids = nxids + 1;
224 myproc->subxids.overflowed = true;
228 LWLockRelease(XidGenLock);
234 * Read nextXid but don't allocate it.
237 ReadNewTransactionId(void)
241 LWLockAcquire(XidGenLock, LW_SHARED);
242 xid = ShmemVariableCache->nextXid;
243 LWLockRelease(XidGenLock);
249 * Determine the last safe XID to allocate given the currently oldest
250 * datfrozenxid (ie, the oldest XID that might exist in any database
251 * of our cluster), and the OID of the (or a) database with that value.
254 SetTransactionIdLimit(TransactionId oldest_datfrozenxid, Oid oldest_datoid)
256 TransactionId xidVacLimit;
257 TransactionId xidWarnLimit;
258 TransactionId xidStopLimit;
259 TransactionId xidWrapLimit;
260 TransactionId curXid;
262 Assert(TransactionIdIsNormal(oldest_datfrozenxid));
265 * The place where we actually get into deep trouble is halfway around
266 * from the oldest potentially-existing XID. (This calculation is
267 * probably off by one or two counts, because the special XIDs reduce the
268 * size of the loop a little bit. But we throw in plenty of slop below,
269 * so it doesn't matter.)
271 xidWrapLimit = oldest_datfrozenxid + (MaxTransactionId >> 1);
272 if (xidWrapLimit < FirstNormalTransactionId)
273 xidWrapLimit += FirstNormalTransactionId;
276 * We'll refuse to continue assigning XIDs in interactive mode once we get
277 * within 1M transactions of data loss. This leaves lots of room for the
278 * DBA to fool around fixing things in a standalone backend, while not
279 * being significant compared to total XID space. (Note that since
280 * vacuuming requires one transaction per table cleaned, we had better be
281 * sure there's lots of XIDs left...)
283 xidStopLimit = xidWrapLimit - 1000000;
284 if (xidStopLimit < FirstNormalTransactionId)
285 xidStopLimit -= FirstNormalTransactionId;
288 * We'll start complaining loudly when we get within 10M transactions of
289 * the stop point. This is kind of arbitrary, but if you let your gas
290 * gauge get down to 1% of full, would you be looking for the next gas
291 * station? We need to be fairly liberal about this number because there
292 * are lots of scenarios where most transactions are done by automatic
293 * clients that won't pay attention to warnings. (No, we're not gonna make
294 * this configurable. If you know enough to configure it, you know enough
295 * to not get in this kind of trouble in the first place.)
297 xidWarnLimit = xidStopLimit - 10000000;
298 if (xidWarnLimit < FirstNormalTransactionId)
299 xidWarnLimit -= FirstNormalTransactionId;
302 * We'll start trying to force autovacuums when oldest_datfrozenxid gets
303 * to be more than autovacuum_freeze_max_age transactions old.
305 * Note: guc.c ensures that autovacuum_freeze_max_age is in a sane range,
306 * so that xidVacLimit will be well before xidWarnLimit.
308 * Note: autovacuum_freeze_max_age is a PGC_POSTMASTER parameter so that
309 * we don't have to worry about dealing with on-the-fly changes in its
310 * value. It doesn't look practical to update shared state from a GUC
311 * assign hook (too many processes would try to execute the hook,
312 * resulting in race conditions as well as crashes of those not connected
313 * to shared memory). Perhaps this can be improved someday.
315 xidVacLimit = oldest_datfrozenxid + autovacuum_freeze_max_age;
316 if (xidVacLimit < FirstNormalTransactionId)
317 xidVacLimit += FirstNormalTransactionId;
319 /* Grab lock for just long enough to set the new limit values */
320 LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
321 ShmemVariableCache->oldestXid = oldest_datfrozenxid;
322 ShmemVariableCache->xidVacLimit = xidVacLimit;
323 ShmemVariableCache->xidWarnLimit = xidWarnLimit;
324 ShmemVariableCache->xidStopLimit = xidStopLimit;
325 ShmemVariableCache->xidWrapLimit = xidWrapLimit;
326 ShmemVariableCache->oldestXidDB = oldest_datoid;
327 curXid = ShmemVariableCache->nextXid;
328 LWLockRelease(XidGenLock);
332 (errmsg("transaction ID wrap limit is %u, limited by database with OID %u",
333 xidWrapLimit, oldest_datoid)));
336 * If past the autovacuum force point, immediately signal an autovac
337 * request. The reason for this is that autovac only processes one
338 * database per invocation. Once it's finished cleaning up the oldest
339 * database, it'll call here, and we'll signal the postmaster to start
340 * another iteration immediately if there are still any old databases.
342 if (TransactionIdFollowsOrEquals(curXid, xidVacLimit) &&
343 IsUnderPostmaster && !InRecovery)
344 SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
346 /* Give an immediate warning if past the wrap warn point */
347 if (TransactionIdFollowsOrEquals(curXid, xidWarnLimit) && !InRecovery)
349 char *oldest_datname = get_database_name(oldest_datoid);
352 * Note: it's possible that get_database_name fails and returns NULL,
353 * for example because the database just got dropped. We'll still
354 * warn, even though the warning might now be unnecessary.
358 (errmsg("database \"%s\" must be vacuumed within %u transactions",
360 xidWrapLimit - curXid),
361 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
362 "You might also need to commit or roll back old prepared transactions.")));
365 (errmsg("database with OID %u must be vacuumed within %u transactions",
367 xidWrapLimit - curXid),
368 errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
369 "You might also need to commit or roll back old prepared transactions.")));
375 * ForceTransactionIdLimitUpdate -- does the XID wrap-limit data need updating?
377 * We primarily check whether oldestXidDB is valid. The cases we have in
378 * mind are that that database was dropped, or the field was reset to zero
379 * by pg_resetxlog. In either case we should force recalculation of the
380 * wrap limit. Also do it if oldestXid is old enough to be forcing
381 * autovacuums or other actions; this ensures we update our state as soon
382 * as possible once extra overhead is being incurred.
385 ForceTransactionIdLimitUpdate(void)
387 TransactionId nextXid;
388 TransactionId xidVacLimit;
389 TransactionId oldestXid;
392 /* Locking is probably not really necessary, but let's be careful */
393 LWLockAcquire(XidGenLock, LW_SHARED);
394 nextXid = ShmemVariableCache->nextXid;
395 xidVacLimit = ShmemVariableCache->xidVacLimit;
396 oldestXid = ShmemVariableCache->oldestXid;
397 oldestXidDB = ShmemVariableCache->oldestXidDB;
398 LWLockRelease(XidGenLock);
400 if (!TransactionIdIsNormal(oldestXid))
401 return true; /* shouldn't happen, but just in case */
402 if (!TransactionIdIsValid(xidVacLimit))
403 return true; /* this shouldn't happen anymore either */
404 if (TransactionIdFollowsOrEquals(nextXid, xidVacLimit))
405 return true; /* past VacLimit, don't delay updating */
406 if (!SearchSysCacheExists1(DATABASEOID, ObjectIdGetDatum(oldestXidDB)))
407 return true; /* could happen, per comments above */
413 * GetNewObjectId -- allocate a new OID
415 * OIDs are generated by a cluster-wide counter. Since they are only 32 bits
416 * wide, counter wraparound will occur eventually, and therefore it is unwise
417 * to assume they are unique unless precautions are taken to make them so.
418 * Hence, this routine should generally not be used directly. The only
419 * direct callers should be GetNewOid() and GetNewRelFileNode() in
427 /* safety check, we should never get this far in a HS slave */
428 if (RecoveryInProgress())
429 elog(ERROR, "cannot assign OIDs during recovery");
431 LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
434 * Check for wraparound of the OID counter. We *must* not return 0
435 * (InvalidOid); and as long as we have to check that, it seems a good
436 * idea to skip over everything below FirstNormalObjectId too. (This
437 * basically just avoids lots of collisions with bootstrap-assigned OIDs
438 * right after a wrap occurs, so as to avoid a possibly large number of
439 * iterations in GetNewOid.) Note we are relying on unsigned comparison.
441 * During initdb, we start the OID generator at FirstBootstrapObjectId, so
442 * we only enforce wrapping to that point when in bootstrap or standalone
443 * mode. The first time through this routine after normal postmaster
444 * start, the counter will be forced up to FirstNormalObjectId. This
445 * mechanism leaves the OIDs between FirstBootstrapObjectId and
446 * FirstNormalObjectId available for automatic assignment during initdb,
447 * while ensuring they will never conflict with user-assigned OIDs.
449 if (ShmemVariableCache->nextOid < ((Oid) FirstNormalObjectId))
451 if (IsPostmasterEnvironment)
453 /* wraparound in normal environment */
454 ShmemVariableCache->nextOid = FirstNormalObjectId;
455 ShmemVariableCache->oidCount = 0;
459 /* we may be bootstrapping, so don't enforce the full range */
460 if (ShmemVariableCache->nextOid < ((Oid) FirstBootstrapObjectId))
462 /* wraparound in standalone environment? */
463 ShmemVariableCache->nextOid = FirstBootstrapObjectId;
464 ShmemVariableCache->oidCount = 0;
469 /* If we run out of logged for use oids then we must log more */
470 if (ShmemVariableCache->oidCount == 0)
472 XLogPutNextOid(ShmemVariableCache->nextOid + VAR_OID_PREFETCH);
473 ShmemVariableCache->oidCount = VAR_OID_PREFETCH;
476 result = ShmemVariableCache->nextOid;
478 (ShmemVariableCache->nextOid)++;
479 (ShmemVariableCache->oidCount)--;
481 LWLockRelease(OidGenLock);