| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* varsup.c |
| 4 |
* postgres OID & XID variables support routines |
| 5 |
* |
| 6 |
* Copyright (c) 2000-2025, PostgreSQL Global Development Group |
| 7 |
* |
| 8 |
* IDENTIFICATION |
| 9 |
* src/backend/access/transam/varsup.c |
| 10 |
* |
| 11 |
*------------------------------------------------------------------------- |
| 12 |
*/ |
| 13 |
|
| 14 |
#include "postgres.h" |
| 15 |
|
| 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/xlogutils.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" |
| 28 |
|
| 29 |
|
| 30 |
/* Number of OIDs to prefetch (preallocate) per XLOG write */ |
| 31 |
#define VAR_OID_PREFETCH 8192 |
| 32 |
|
| 33 |
/* pointer to variables struct in shared memory */ |
| 34 |
TransamVariablesData *TransamVariables = NULL; |
| 35 |
|
| 36 |
|
| 37 |
/* |
| 38 |
* Initialization of shared memory for TransamVariables. |
| 39 |
*/ |
| 40 |
Size |
| 41 |
VarsupShmemSize(void) |
| 42 |
{ |
| 43 |
return sizeof(TransamVariablesData); |
| 44 |
} |
| 45 |
|
| 46 |
void |
| 47 |
VarsupShmemInit(void) |
| 48 |
{ |
| 49 |
bool found; |
| 50 |
|
| 51 |
/* Initialize our shared state struct */ |
| 52 |
TransamVariables = ShmemInitStruct("TransamVariables", |
| 53 |
sizeof(TransamVariablesData), |
| 54 |
&found); |
| 55 |
if (!IsUnderPostmaster) |
| 56 |
{ |
| 57 |
Assert(!found); |
| 58 |
memset(TransamVariables, 0, sizeof(TransamVariablesData)); |
| 59 |
} |
| 60 |
else |
| 61 |
Assert(found); |
| 62 |
} |
| 63 |
|
| 64 |
/* |
| 65 |
* Allocate the next FullTransactionId for a new transaction or |
| 66 |
* subtransaction. |
| 67 |
* |
| 68 |
* The new XID is also stored into MyProc->xid/ProcGlobal->xids[] before |
| 69 |
* returning. |
| 70 |
* |
| 71 |
* Note: when this is called, we are actually already inside a valid |
| 72 |
* transaction, since XIDs are now not allocated until the transaction |
| 73 |
* does something. So it is safe to do a database lookup if we want to |
| 74 |
* issue a warning about XID wrap. |
| 75 |
*/ |
| 76 |
FullTransactionId |
| 77 |
GetNewTransactionId(bool isSubXact) |
| 78 |
{ |
| 79 |
FullTransactionId full_xid; |
| 80 |
TransactionId xid; |
| 81 |
|
| 82 |
/* |
| 83 |
* Workers synchronize transaction state at the beginning of each parallel |
| 84 |
* operation, so we can't account for new XIDs after that point. |
| 85 |
*/ |
| 86 |
if (IsInParallelMode()) |
| 87 |
elog(ERROR, "cannot assign TransactionIds during a parallel operation"); |
| 88 |
|
| 89 |
/* |
| 90 |
* During bootstrap initialization, we return the special bootstrap |
| 91 |
* transaction id. |
| 92 |
*/ |
| 93 |
if (IsBootstrapProcessingMode()) |
| 94 |
{ |
| 95 |
Assert(!isSubXact); |
| 96 |
MyProc->xid = BootstrapTransactionId; |
| 97 |
ProcGlobal->xids[MyProc->pgxactoff] = BootstrapTransactionId; |
| 98 |
return FullTransactionIdFromEpochAndXid(0, BootstrapTransactionId); |
| 99 |
} |
| 100 |
|
| 101 |
/* safety check, we should never get this far in a HS standby */ |
| 102 |
if (RecoveryInProgress()) |
| 103 |
elog(ERROR, "cannot assign TransactionIds during recovery"); |
| 104 |
|
| 105 |
LWLockAcquire(XidGenLock, LW_EXCLUSIVE); |
| 106 |
|
| 107 |
full_xid = TransamVariables->nextXid; |
| 108 |
xid = XidFromFullTransactionId(full_xid); |
| 109 |
|
| 110 |
/*---------- |
| 111 |
* Check to see if it's safe to assign another XID. This protects against |
| 112 |
* catastrophic data loss due to XID wraparound. The basic rules are: |
| 113 |
* |
| 114 |
* If we're past xidVacLimit, start trying to force autovacuum cycles. |
| 115 |
* If we're past xidWarnLimit, start issuing warnings. |
| 116 |
* If we're past xidStopLimit, refuse to execute transactions, unless |
| 117 |
* we are running in single-user mode (which gives an escape hatch |
| 118 |
* to the DBA who somehow got past the earlier defenses). |
| 119 |
* |
| 120 |
* Note that this coding also appears in GetNewMultiXactId. |
| 121 |
*---------- |
| 122 |
*/ |
| 123 |
if (TransactionIdFollowsOrEquals(xid, TransamVariables->xidVacLimit)) |
| 124 |
{ |
| 125 |
/* |
| 126 |
* For safety's sake, we release XidGenLock while sending signals, |
| 127 |
* warnings, etc. This is not so much because we care about |
| 128 |
* preserving concurrency in this situation, as to avoid any |
| 129 |
* possibility of deadlock while doing get_database_name(). First, |
| 130 |
* copy all the shared values we'll need in this path. |
| 131 |
*/ |
| 132 |
TransactionId xidWarnLimit = TransamVariables->xidWarnLimit; |
| 133 |
TransactionId xidStopLimit = TransamVariables->xidStopLimit; |
| 134 |
TransactionId xidWrapLimit = TransamVariables->xidWrapLimit; |
| 135 |
Oid oldest_datoid = TransamVariables->oldestXidDB; |
| 136 |
|
| 137 |
LWLockRelease(XidGenLock); |
| 138 |
|
| 139 |
/* |
| 140 |
* To avoid swamping the postmaster with signals, we issue the autovac |
| 141 |
* request only once per 64K transaction starts. This still gives |
| 142 |
* plenty of chances before we get into real trouble. |
| 143 |
*/ |
| 144 |
if (IsUnderPostmaster && (xid % 65536) == 0) |
| 145 |
SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER); |
| 146 |
|
| 147 |
if (IsUnderPostmaster && |
| 148 |
TransactionIdFollowsOrEquals(xid, xidStopLimit)) |
| 149 |
{ |
| 150 |
char *oldest_datname = get_database_name(oldest_datoid); |
| 151 |
|
| 152 |
/* complain even if that DB has disappeared */ |
| 153 |
if (oldest_datname) |
| 154 |
ereport(ERROR, |
| 155 |
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
| 156 |
errmsg("database is not accepting commands that assign new transaction IDs to avoid wraparound data loss in database \"%s\"", |
| 157 |
oldest_datname), |
| 158 |
errhint("Execute a database-wide VACUUM in that database.\n" |
| 159 |
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots."))); |
| 160 |
else |
| 161 |
ereport(ERROR, |
| 162 |
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED), |
| 163 |
errmsg("database is not accepting commands that assign new transaction IDs to avoid wraparound data loss in database with OID %u", |
| 164 |
oldest_datoid), |
| 165 |
errhint("Execute a database-wide VACUUM in that database.\n" |
| 166 |
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots."))); |
| 167 |
} |
| 168 |
else if (TransactionIdFollowsOrEquals(xid, xidWarnLimit)) |
| 169 |
{ |
| 170 |
char *oldest_datname = get_database_name(oldest_datoid); |
| 171 |
|
| 172 |
/* complain even if that DB has disappeared */ |
| 173 |
if (oldest_datname) |
| 174 |
ereport(WARNING, |
| 175 |
(errmsg("database \"%s\" must be vacuumed within %u transactions", |
| 176 |
oldest_datname, |
| 177 |
xidWrapLimit - xid), |
| 178 |
errhint("To avoid transaction ID assignment failures, execute a database-wide VACUUM in that database.\n" |
| 179 |
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots."))); |
| 180 |
else |
| 181 |
ereport(WARNING, |
| 182 |
(errmsg("database with OID %u must be vacuumed within %u transactions", |
| 183 |
oldest_datoid, |
| 184 |
xidWrapLimit - xid), |
| 185 |
errhint("To avoid XID assignment failures, execute a database-wide VACUUM in that database.\n" |
| 186 |
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots."))); |
| 187 |
} |
| 188 |
|
| 189 |
/* Re-acquire lock and start over */ |
| 190 |
LWLockAcquire(XidGenLock, LW_EXCLUSIVE); |
| 191 |
full_xid = TransamVariables->nextXid; |
| 192 |
xid = XidFromFullTransactionId(full_xid); |
| 193 |
} |
| 194 |
|
| 195 |
/* |
| 196 |
* If we are allocating the first XID of a new page of the commit log, |
| 197 |
* zero out that commit-log page before returning. We must do this while |
| 198 |
* holding XidGenLock, else another xact could acquire and commit a later |
| 199 |
* XID before we zero the page. Fortunately, a page of the commit log |
| 200 |
* holds 32K or more transactions, so we don't have to do this very often. |
| 201 |
* |
| 202 |
* Extend pg_subtrans and pg_commit_ts too. |
| 203 |
*/ |
| 204 |
ExtendCLOG(xid); |
| 205 |
ExtendCommitTs(xid); |
| 206 |
ExtendSUBTRANS(xid); |
| 207 |
|
| 208 |
/* |
| 209 |
* Now advance the nextXid counter. This must not happen until after we |
| 210 |
* have successfully completed ExtendCLOG() --- if that routine fails, we |
| 211 |
* want the next incoming transaction to try it again. We cannot assign |
| 212 |
* more XIDs until there is CLOG space for them. |
| 213 |
*/ |
| 214 |
FullTransactionIdAdvance(&TransamVariables->nextXid); |
| 215 |
|
| 216 |
/* |
| 217 |
* We must store the new XID into the shared ProcArray before releasing |
| 218 |
* XidGenLock. This ensures that every active XID older than |
| 219 |
* latestCompletedXid is present in the ProcArray, which is essential for |
| 220 |
* correct OldestXmin tracking; see src/backend/access/transam/README. |
| 221 |
* |
| 222 |
* Note that readers of ProcGlobal->xids/PGPROC->xid should be careful to |
| 223 |
* fetch the value for each proc only once, rather than assume they can |
| 224 |
* read a value multiple times and get the same answer each time. Note we |
| 225 |
* are assuming that TransactionId and int fetch/store are atomic. |
| 226 |
* |
| 227 |
* The same comments apply to the subxact xid count and overflow fields. |
| 228 |
* |
| 229 |
* Use of a write barrier prevents dangerous code rearrangement in this |
| 230 |
* function; other backends could otherwise e.g. be examining my subxids |
| 231 |
* info concurrently, and we don't want them to see an invalid |
| 232 |
* intermediate state, such as an incremented nxids before the array entry |
| 233 |
* is filled. |
| 234 |
* |
| 235 |
* Other processes that read nxids should do so before reading xids |
| 236 |
* elements with a pg_read_barrier() in between, so that they can be sure |
| 237 |
* not to read an uninitialized array element; see |
| 238 |
* src/backend/storage/lmgr/README.barrier. |
| 239 |
* |
| 240 |
* If there's no room to fit a subtransaction XID into PGPROC, set the |
| 241 |
* cache-overflowed flag instead. This forces readers to look in |
| 242 |
* pg_subtrans to map subtransaction XIDs up to top-level XIDs. There is a |
| 243 |
* race-condition window, in that the new XID will not appear as running |
| 244 |
* until its parent link has been placed into pg_subtrans. However, that |
| 245 |
* will happen before anyone could possibly have a reason to inquire about |
| 246 |
* the status of the XID, so it seems OK. (Snapshots taken during this |
| 247 |
* window *will* include the parent XID, so they will deliver the correct |
| 248 |
* answer later on when someone does have a reason to inquire.) |
| 249 |
*/ |
| 250 |
if (!isSubXact) |
| 251 |
{ |
| 252 |
Assert(ProcGlobal->subxidStates[MyProc->pgxactoff].count == 0); |
| 253 |
Assert(!ProcGlobal->subxidStates[MyProc->pgxactoff].overflowed); |
| 254 |
Assert(MyProc->subxidStatus.count == 0); |
| 255 |
Assert(!MyProc->subxidStatus.overflowed); |
| 256 |
|
| 257 |
/* LWLockRelease acts as barrier */ |
| 258 |
MyProc->xid = xid; |
| 259 |
ProcGlobal->xids[MyProc->pgxactoff] = xid; |
| 260 |
} |
| 261 |
else |
| 262 |
{ |
| 263 |
XidCacheStatus *substat = &ProcGlobal->subxidStates[MyProc->pgxactoff]; |
| 264 |
int nxids = MyProc->subxidStatus.count; |
| 265 |
|
| 266 |
Assert(substat->count == MyProc->subxidStatus.count); |
| 267 |
Assert(substat->overflowed == MyProc->subxidStatus.overflowed); |
| 268 |
|
| 269 |
if (nxids < PGPROC_MAX_CACHED_SUBXIDS) |
| 270 |
{ |
| 271 |
MyProc->subxids.xids[nxids] = xid; |
| 272 |
pg_write_barrier(); |
| 273 |
MyProc->subxidStatus.count = substat->count = nxids + 1; |
| 274 |
} |
| 275 |
else |
| 276 |
MyProc->subxidStatus.overflowed = substat->overflowed = true; |
| 277 |
} |
| 278 |
|
| 279 |
LWLockRelease(XidGenLock); |
| 280 |
|
| 281 |
return full_xid; |
| 282 |
} |
| 283 |
|
| 284 |
/* |
| 285 |
* Read nextXid but don't allocate it. |
| 286 |
*/ |
| 287 |
FullTransactionId |
| 288 |
ReadNextFullTransactionId(void) |
| 289 |
{ |
| 290 |
FullTransactionId fullXid; |
| 291 |
|
| 292 |
LWLockAcquire(XidGenLock, LW_SHARED); |
| 293 |
fullXid = TransamVariables->nextXid; |
| 294 |
LWLockRelease(XidGenLock); |
| 295 |
|
| 296 |
return fullXid; |
| 297 |
} |
| 298 |
|
| 299 |
/* |
| 300 |
* Advance nextXid to the value after a given xid. The epoch is inferred. |
| 301 |
* This must only be called during recovery or from two-phase start-up code. |
| 302 |
*/ |
| 303 |
void |
| 304 |
AdvanceNextFullTransactionIdPastXid(TransactionId xid) |
| 305 |
{ |
| 306 |
FullTransactionId newNextFullXid; |
| 307 |
TransactionId next_xid; |
| 308 |
uint32 epoch; |
| 309 |
|
| 310 |
/* |
| 311 |
* It is safe to read nextXid without a lock, because this is only called |
| 312 |
* from the startup process or single-process mode, meaning that no other |
| 313 |
* process can modify it. |
| 314 |
*/ |
| 315 |
Assert(AmStartupProcess() || !IsUnderPostmaster); |
| 316 |
|
| 317 |
/* Fast return if this isn't an xid high enough to move the needle. */ |
| 318 |
next_xid = XidFromFullTransactionId(TransamVariables->nextXid); |
| 319 |
if (!TransactionIdFollowsOrEquals(xid, next_xid)) |
| 320 |
return; |
| 321 |
|
| 322 |
/* |
| 323 |
* Compute the FullTransactionId that comes after the given xid. To do |
| 324 |
* this, we preserve the existing epoch, but detect when we've wrapped |
| 325 |
* into a new epoch. This is necessary because WAL records and 2PC state |
| 326 |
* currently contain 32 bit xids. The wrap logic is safe in those cases |
| 327 |
* because the span of active xids cannot exceed one epoch at any given |
| 328 |
* point in the WAL stream. |
| 329 |
*/ |
| 330 |
TransactionIdAdvance(xid); |
| 331 |
epoch = EpochFromFullTransactionId(TransamVariables->nextXid); |
| 332 |
if (unlikely(xid < next_xid)) |
| 333 |
++epoch; |
| 334 |
newNextFullXid = FullTransactionIdFromEpochAndXid(epoch, xid); |
| 335 |
|
| 336 |
/* |
| 337 |
* We still need to take a lock to modify the value when there are |
| 338 |
* concurrent readers. |
| 339 |
*/ |
| 340 |
LWLockAcquire(XidGenLock, LW_EXCLUSIVE); |
| 341 |
TransamVariables->nextXid = newNextFullXid; |
| 342 |
LWLockRelease(XidGenLock); |
| 343 |
} |
| 344 |
|
| 345 |
/* |
| 346 |
* Advance the cluster-wide value for the oldest valid clog entry. |
| 347 |
* |
| 348 |
* We must acquire XactTruncationLock to advance the oldestClogXid. It's not |
| 349 |
* necessary to hold the lock during the actual clog truncation, only when we |
| 350 |
* advance the limit, as code looking up arbitrary xids is required to hold |
| 351 |
* XactTruncationLock from when it tests oldestClogXid through to when it |
| 352 |
* completes the clog lookup. |
| 353 |
*/ |
| 354 |
void |
| 355 |
AdvanceOldestClogXid(TransactionId oldest_datfrozenxid) |
| 356 |
{ |
| 357 |
LWLockAcquire(XactTruncationLock, LW_EXCLUSIVE); |
| 358 |
if (TransactionIdPrecedes(TransamVariables->oldestClogXid, |
| 359 |
oldest_datfrozenxid)) |
| 360 |
{ |
| 361 |
TransamVariables->oldestClogXid = oldest_datfrozenxid; |
| 362 |
} |
| 363 |
LWLockRelease(XactTruncationLock); |
| 364 |
} |
| 365 |
|
| 366 |
/* |
| 367 |
* Determine the last safe XID to allocate using the currently oldest |
| 368 |
* datfrozenxid (ie, the oldest XID that might exist in any database |
| 369 |
* of our cluster), and the OID of the (or a) database with that value. |
| 370 |
*/ |
| 371 |
void |
| 372 |
SetTransactionIdLimit(TransactionId oldest_datfrozenxid, Oid oldest_datoid) |
| 373 |
{ |
| 374 |
TransactionId xidVacLimit; |
| 375 |
TransactionId xidWarnLimit; |
| 376 |
TransactionId xidStopLimit; |
| 377 |
TransactionId xidWrapLimit; |
| 378 |
TransactionId curXid; |
| 379 |
|
| 380 |
Assert(TransactionIdIsNormal(oldest_datfrozenxid)); |
| 381 |
|
| 382 |
/* |
| 383 |
* The place where we actually get into deep trouble is halfway around |
| 384 |
* from the oldest potentially-existing XID. (This calculation is |
| 385 |
* probably off by one or two counts, because the special XIDs reduce the |
| 386 |
* size of the loop a little bit. But we throw in plenty of slop below, |
| 387 |
* so it doesn't matter.) |
| 388 |
*/ |
| 389 |
xidWrapLimit = oldest_datfrozenxid + (MaxTransactionId >> 1); |
| 390 |
if (xidWrapLimit < FirstNormalTransactionId) |
| 391 |
xidWrapLimit += FirstNormalTransactionId; |
| 392 |
|
| 393 |
/* |
| 394 |
* We'll refuse to continue assigning XIDs in interactive mode once we get |
| 395 |
* within 3M transactions of data loss. This leaves lots of room for the |
| 396 |
* DBA to fool around fixing things in a standalone backend, while not |
| 397 |
* being significant compared to total XID space. (VACUUM requires an XID |
| 398 |
* if it truncates at wal_level!=minimal. "VACUUM (ANALYZE)", which a DBA |
| 399 |
* might do by reflex, assigns an XID. Hence, we had better be sure |
| 400 |
* there's lots of XIDs left...) Also, at default BLCKSZ, this leaves two |
| 401 |
* completely-idle segments. In the event of edge-case bugs involving |
| 402 |
* page or segment arithmetic, idle segments render the bugs unreachable |
| 403 |
* outside of single-user mode. |
| 404 |
*/ |
| 405 |
xidStopLimit = xidWrapLimit - 3000000; |
| 406 |
if (xidStopLimit < FirstNormalTransactionId) |
| 407 |
xidStopLimit -= FirstNormalTransactionId; |
| 408 |
|
| 409 |
/* |
| 410 |
* We'll start complaining loudly when we get within 40M transactions of |
| 411 |
* data loss. This is kind of arbitrary, but if you let your gas gauge |
| 412 |
* get down to 2% of full, would you be looking for the next gas station? |
| 413 |
* We need to be fairly liberal about this number because there are lots |
| 414 |
* of scenarios where most transactions are done by automatic clients that |
| 415 |
* won't pay attention to warnings. (No, we're not gonna make this |
| 416 |
* configurable. If you know enough to configure it, you know enough to |
| 417 |
* not get in this kind of trouble in the first place.) |
| 418 |
*/ |
| 419 |
xidWarnLimit = xidWrapLimit - 40000000; |
| 420 |
if (xidWarnLimit < FirstNormalTransactionId) |
| 421 |
xidWarnLimit -= FirstNormalTransactionId; |
| 422 |
|
| 423 |
/* |
| 424 |
* We'll start trying to force autovacuums when oldest_datfrozenxid gets |
| 425 |
* to be more than autovacuum_freeze_max_age transactions old. |
| 426 |
* |
| 427 |
* Note: guc.c ensures that autovacuum_freeze_max_age is in a sane range, |
| 428 |
* so that xidVacLimit will be well before xidWarnLimit. |
| 429 |
* |
| 430 |
* Note: autovacuum_freeze_max_age is a PGC_POSTMASTER parameter so that |
| 431 |
* we don't have to worry about dealing with on-the-fly changes in its |
| 432 |
* value. It doesn't look practical to update shared state from a GUC |
| 433 |
* assign hook (too many processes would try to execute the hook, |
| 434 |
* resulting in race conditions as well as crashes of those not connected |
| 435 |
* to shared memory). Perhaps this can be improved someday. See also |
| 436 |
* SetMultiXactIdLimit. |
| 437 |
*/ |
| 438 |
xidVacLimit = oldest_datfrozenxid + autovacuum_freeze_max_age; |
| 439 |
if (xidVacLimit < FirstNormalTransactionId) |
| 440 |
xidVacLimit += FirstNormalTransactionId; |
| 441 |
|
| 442 |
/* Grab lock for just long enough to set the new limit values */ |
| 443 |
LWLockAcquire(XidGenLock, LW_EXCLUSIVE); |
| 444 |
TransamVariables->oldestXid = oldest_datfrozenxid; |
| 445 |
TransamVariables->xidVacLimit = xidVacLimit; |
| 446 |
TransamVariables->xidWarnLimit = xidWarnLimit; |
| 447 |
TransamVariables->xidStopLimit = xidStopLimit; |
| 448 |
TransamVariables->xidWrapLimit = xidWrapLimit; |
| 449 |
TransamVariables->oldestXidDB = oldest_datoid; |
| 450 |
curXid = XidFromFullTransactionId(TransamVariables->nextXid); |
| 451 |
LWLockRelease(XidGenLock); |
| 452 |
|
| 453 |
/* Log the info */ |
| 454 |
ereport(DEBUG1, |
| 455 |
(errmsg_internal("transaction ID wrap limit is %u, limited by database with OID %u", |
| 456 |
xidWrapLimit, oldest_datoid))); |
| 457 |
|
| 458 |
/* |
| 459 |
* If past the autovacuum force point, immediately signal an autovac |
| 460 |
* request. The reason for this is that autovac only processes one |
| 461 |
* database per invocation. Once it's finished cleaning up the oldest |
| 462 |
* database, it'll call here, and we'll signal the postmaster to start |
| 463 |
* another iteration immediately if there are still any old databases. |
| 464 |
*/ |
| 465 |
if (TransactionIdFollowsOrEquals(curXid, xidVacLimit) && |
| 466 |
IsUnderPostmaster && !InRecovery) |
| 467 |
SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER); |
| 468 |
|
| 469 |
/* Give an immediate warning if past the wrap warn point */ |
| 470 |
if (TransactionIdFollowsOrEquals(curXid, xidWarnLimit) && !InRecovery) |
| 471 |
{ |
| 472 |
char *oldest_datname; |
| 473 |
|
| 474 |
/* |
| 475 |
* We can be called when not inside a transaction, for example during |
| 476 |
* StartupXLOG(). In such a case we cannot do database access, so we |
| 477 |
* must just report the oldest DB's OID. |
| 478 |
* |
| 479 |
* Note: it's also possible that get_database_name fails and returns |
| 480 |
* NULL, for example because the database just got dropped. We'll |
| 481 |
* still warn, even though the warning might now be unnecessary. |
| 482 |
*/ |
| 483 |
if (IsTransactionState()) |
| 484 |
oldest_datname = get_database_name(oldest_datoid); |
| 485 |
else |
| 486 |
oldest_datname = NULL; |
| 487 |
|
| 488 |
if (oldest_datname) |
| 489 |
ereport(WARNING, |
| 490 |
(errmsg("database \"%s\" must be vacuumed within %u transactions", |
| 491 |
oldest_datname, |
| 492 |
xidWrapLimit - curXid), |
| 493 |
errhint("To avoid XID assignment failures, execute a database-wide VACUUM in that database.\n" |
| 494 |
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots."))); |
| 495 |
else |
| 496 |
ereport(WARNING, |
| 497 |
(errmsg("database with OID %u must be vacuumed within %u transactions", |
| 498 |
oldest_datoid, |
| 499 |
xidWrapLimit - curXid), |
| 500 |
errhint("To avoid XID assignment failures, execute a database-wide VACUUM in that database.\n" |
| 501 |
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots."))); |
| 502 |
} |
| 503 |
} |
| 504 |
|
| 505 |
|
| 506 |
/* |
| 507 |
* ForceTransactionIdLimitUpdate -- does the XID wrap-limit data need updating? |
| 508 |
* |
| 509 |
* We primarily check whether oldestXidDB is valid. The cases we have in |
| 510 |
* mind are that that database was dropped, or the field was reset to zero |
| 511 |
* by pg_resetwal. In either case we should force recalculation of the |
| 512 |
* wrap limit. Also do it if oldestXid is old enough to be forcing |
| 513 |
* autovacuums or other actions; this ensures we update our state as soon |
| 514 |
* as possible once extra overhead is being incurred. |
| 515 |
*/ |
| 516 |
bool |
| 517 |
ForceTransactionIdLimitUpdate(void) |
| 518 |
{ |
| 519 |
TransactionId nextXid; |
| 520 |
TransactionId xidVacLimit; |
| 521 |
TransactionId oldestXid; |
| 522 |
Oid oldestXidDB; |
| 523 |
|
| 524 |
/* Locking is probably not really necessary, but let's be careful */ |
| 525 |
LWLockAcquire(XidGenLock, LW_SHARED); |
| 526 |
nextXid = XidFromFullTransactionId(TransamVariables->nextXid); |
| 527 |
xidVacLimit = TransamVariables->xidVacLimit; |
| 528 |
oldestXid = TransamVariables->oldestXid; |
| 529 |
oldestXidDB = TransamVariables->oldestXidDB; |
| 530 |
LWLockRelease(XidGenLock); |
| 531 |
|
| 532 |
if (!TransactionIdIsNormal(oldestXid)) |
| 533 |
return true; /* shouldn't happen, but just in case */ |
| 534 |
if (!TransactionIdIsValid(xidVacLimit)) |
| 535 |
return true; /* this shouldn't happen anymore either */ |
| 536 |
if (TransactionIdFollowsOrEquals(nextXid, xidVacLimit)) |
| 537 |
return true; /* past xidVacLimit, don't delay updating */ |
| 538 |
if (!SearchSysCacheExists1(DATABASEOID, ObjectIdGetDatum(oldestXidDB))) |
| 539 |
return true; /* could happen, per comments above */ |
| 540 |
return false; |
| 541 |
} |
| 542 |
|
| 543 |
|
| 544 |
/* |
| 545 |
* GetNewObjectId -- allocate a new OID |
| 546 |
* |
| 547 |
* OIDs are generated by a cluster-wide counter. Since they are only 32 bits |
| 548 |
* wide, counter wraparound will occur eventually, and therefore it is unwise |
| 549 |
* to assume they are unique unless precautions are taken to make them so. |
| 550 |
* Hence, this routine should generally not be used directly. The only direct |
| 551 |
* callers should be GetNewOidWithIndex() and GetNewRelFileNumber() in |
| 552 |
* catalog/catalog.c. |
| 553 |
*/ |
| 554 |
Oid |
| 555 |
GetNewObjectId(void) |
| 556 |
{ |
| 557 |
Oid result; |
| 558 |
|
| 559 |
/* safety check, we should never get this far in a HS standby */ |
| 560 |
if (RecoveryInProgress()) |
| 561 |
elog(ERROR, "cannot assign OIDs during recovery"); |
| 562 |
|
| 563 |
LWLockAcquire(OidGenLock, LW_EXCLUSIVE); |
| 564 |
|
| 565 |
/* |
| 566 |
* Check for wraparound of the OID counter. We *must* not return 0 |
| 567 |
* (InvalidOid), and in normal operation we mustn't return anything below |
| 568 |
* FirstNormalObjectId since that range is reserved for initdb (see |
| 569 |
* IsCatalogRelationOid()). Note we are relying on unsigned comparison. |
| 570 |
* |
| 571 |
* During initdb, we start the OID generator at FirstGenbkiObjectId, so we |
| 572 |
* only wrap if before that point when in bootstrap or standalone mode. |
| 573 |
* The first time through this routine after normal postmaster start, the |
| 574 |
* counter will be forced up to FirstNormalObjectId. This mechanism |
| 575 |
* leaves the OIDs between FirstGenbkiObjectId and FirstNormalObjectId |
| 576 |
* available for automatic assignment during initdb, while ensuring they |
| 577 |
* will never conflict with user-assigned OIDs. |
| 578 |
*/ |
| 579 |
if (TransamVariables->nextOid < ((Oid) FirstNormalObjectId)) |
| 580 |
{ |
| 581 |
if (IsPostmasterEnvironment) |
| 582 |
{ |
| 583 |
/* wraparound, or first post-initdb assignment, in normal mode */ |
| 584 |
TransamVariables->nextOid = FirstNormalObjectId; |
| 585 |
TransamVariables->oidCount = 0; |
| 586 |
} |
| 587 |
else |
| 588 |
{ |
| 589 |
/* we may be bootstrapping, so don't enforce the full range */ |
| 590 |
if (TransamVariables->nextOid < ((Oid) FirstGenbkiObjectId)) |
| 591 |
{ |
| 592 |
/* wraparound in standalone mode (unlikely but possible) */ |
| 593 |
TransamVariables->nextOid = FirstNormalObjectId; |
| 594 |
TransamVariables->oidCount = 0; |
| 595 |
} |
| 596 |
} |
| 597 |
} |
| 598 |
|
| 599 |
/* If we run out of logged for use oids then we must log more */ |
| 600 |
if (TransamVariables->oidCount == 0) |
| 601 |
{ |
| 602 |
XLogPutNextOid(TransamVariables->nextOid + VAR_OID_PREFETCH); |
| 603 |
TransamVariables->oidCount = VAR_OID_PREFETCH; |
| 604 |
} |
| 605 |
|
| 606 |
result = TransamVariables->nextOid; |
| 607 |
|
| 608 |
(TransamVariables->nextOid)++; |
| 609 |
(TransamVariables->oidCount)--; |
| 610 |
|
| 611 |
LWLockRelease(OidGenLock); |
| 612 |
|
| 613 |
return result; |
| 614 |
} |
| 615 |
|
| 616 |
/* |
| 617 |
* SetNextObjectId |
| 618 |
* |
| 619 |
* This may only be called during initdb; it advances the OID counter |
| 620 |
* to the specified value. |
| 621 |
*/ |
| 622 |
static void |
| 623 |
SetNextObjectId(Oid nextOid) |
| 624 |
{ |
| 625 |
/* Safety check, this is only allowable during initdb */ |
| 626 |
if (IsPostmasterEnvironment) |
| 627 |
elog(ERROR, "cannot advance OID counter anymore"); |
| 628 |
|
| 629 |
/* Taking the lock is, therefore, just pro forma; but do it anyway */ |
| 630 |
LWLockAcquire(OidGenLock, LW_EXCLUSIVE); |
| 631 |
|
| 632 |
if (TransamVariables->nextOid > nextOid) |
| 633 |
elog(ERROR, "too late to advance OID counter to %u, it is now %u", |
| 634 |
nextOid, TransamVariables->nextOid); |
| 635 |
|
| 636 |
TransamVariables->nextOid = nextOid; |
| 637 |
TransamVariables->oidCount = 0; |
| 638 |
|
| 639 |
LWLockRelease(OidGenLock); |
| 640 |
} |
| 641 |
|
| 642 |
/* |
| 643 |
* StopGeneratingPinnedObjectIds |
| 644 |
* |
| 645 |
* This is called once during initdb to force the OID counter up to |
| 646 |
* FirstUnpinnedObjectId. This supports letting initdb's post-bootstrap |
| 647 |
* processing create some pinned objects early on. Once it's done doing |
| 648 |
* so, it calls this (via pg_stop_making_pinned_objects()) so that the |
| 649 |
* remaining objects it makes will be considered un-pinned. |
| 650 |
*/ |
| 651 |
void |
| 652 |
StopGeneratingPinnedObjectIds(void) |
| 653 |
{ |
| 654 |
SetNextObjectId(FirstUnpinnedObjectId); |
| 655 |
} |
| 656 |
|
| 657 |
|
| 658 |
#ifdef USE_ASSERT_CHECKING |
| 659 |
|
| 660 |
/* |
| 661 |
* Assert that xid is between [oldestXid, nextXid], which is the range we |
| 662 |
* expect XIDs coming from tables etc to be in. |
| 663 |
* |
| 664 |
* As TransamVariables->oldestXid could change just after this call without |
| 665 |
* further precautions, and as a wrapped-around xid could again fall within |
| 666 |
* the valid range, this assertion can only detect if something is definitely |
| 667 |
* wrong, but not establish correctness. |
| 668 |
* |
| 669 |
* This intentionally does not expose a return value, to avoid code being |
| 670 |
* introduced that depends on the return value. |
| 671 |
*/ |
| 672 |
void |
| 673 |
AssertTransactionIdInAllowableRange(TransactionId xid) |
| 674 |
{ |
| 675 |
TransactionId oldest_xid; |
| 676 |
TransactionId next_xid; |
| 677 |
|
| 678 |
Assert(TransactionIdIsValid(xid)); |
| 679 |
|
| 680 |
/* we may see bootstrap / frozen */ |
| 681 |
if (!TransactionIdIsNormal(xid)) |
| 682 |
return; |
| 683 |
|
| 684 |
/* |
| 685 |
* We can't acquire XidGenLock, as this may be called with XidGenLock |
| 686 |
* already held (or with other locks that don't allow XidGenLock to be |
| 687 |
* nested). That's ok for our purposes though, since we already rely on |
| 688 |
* 32bit reads to be atomic. While nextXid is 64 bit, we only look at the |
| 689 |
* lower 32bit, so a skewed read doesn't hurt. |
| 690 |
* |
| 691 |
* There's no increased danger of falling outside [oldest, next] by |
| 692 |
* accessing them without a lock. xid needs to have been created with |
| 693 |
* GetNewTransactionId() in the originating session, and the locks there |
| 694 |
* pair with the memory barrier below. We do however accept xid to be <= |
| 695 |
* to next_xid, instead of just <, as xid could be from the procarray, |
| 696 |
* before we see the updated nextXid value. |
| 697 |
*/ |
| 698 |
pg_memory_barrier(); |
| 699 |
oldest_xid = TransamVariables->oldestXid; |
| 700 |
next_xid = XidFromFullTransactionId(TransamVariables->nextXid); |
| 701 |
|
| 702 |
Assert(TransactionIdFollowsOrEquals(xid, oldest_xid) || |
| 703 |
TransactionIdPrecedesOrEquals(xid, next_xid)); |
| 704 |
} |
| 705 |
#endif |