| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* snapbuild.c |
| 4 |
* |
| 5 |
* Infrastructure for building historic catalog snapshots based on contents |
| 6 |
* of the WAL, for the purpose of decoding heapam.c style values in the |
| 7 |
* WAL. |
| 8 |
* |
| 9 |
* NOTES: |
| 10 |
* |
| 11 |
* We build snapshots which can *only* be used to read catalog contents and we |
| 12 |
* do so by reading and interpreting the WAL stream. The aim is to build a |
| 13 |
* snapshot that behaves the same as a freshly taken MVCC snapshot would have |
| 14 |
* at the time the XLogRecord was generated. |
| 15 |
* |
| 16 |
* To build the snapshots we reuse the infrastructure built for Hot |
| 17 |
* Standby. The in-memory snapshots we build look different than HS' because |
| 18 |
* we have different needs. To successfully decode data from the WAL we only |
| 19 |
* need to access catalog tables and (sys|rel|cat)cache, not the actual user |
| 20 |
* tables since the data we decode is wholly contained in the WAL |
| 21 |
* records. Also, our snapshots need to be different in comparison to normal |
| 22 |
* MVCC ones because in contrast to those we cannot fully rely on the clog and |
| 23 |
* pg_subtrans for information about committed transactions because they might |
| 24 |
* commit in the future from the POV of the WAL entry we're currently |
| 25 |
* decoding. This definition has the advantage that we only need to prevent |
| 26 |
* removal of catalog rows, while normal table's rows can still be |
| 27 |
* removed. This is achieved by using the replication slot mechanism. |
| 28 |
* |
| 29 |
* As the percentage of transactions modifying the catalog normally is fairly |
| 30 |
* small in comparisons to ones only manipulating user data, we keep track of |
| 31 |
* the committed catalog modifying ones inside [xmin, xmax) instead of keeping |
| 32 |
* track of all running transactions like it's done in a normal snapshot. Note |
| 33 |
* that we're generally only looking at transactions that have acquired an |
| 34 |
* xid. That is we keep a list of transactions between snapshot->(xmin, xmax) |
| 35 |
* that we consider committed, everything else is considered aborted/in |
| 36 |
* progress. That also allows us not to care about subtransactions before they |
| 37 |
* have committed which means this module, in contrast to HS, doesn't have to |
| 38 |
* care about suboverflowed subtransactions and similar. |
| 39 |
* |
| 40 |
* One complexity of doing this is that to e.g. handle mixed DDL/DML |
| 41 |
* transactions we need Snapshots that see intermediate versions of the |
| 42 |
* catalog in a transaction. During normal operation this is achieved by using |
| 43 |
* CommandIds/cmin/cmax. The problem with that however is that for space |
| 44 |
* efficiency reasons, the cmin and cmax are not included in WAL records. We |
| 45 |
* cannot read the cmin/cmax from the tuple itself, either, because it is |
| 46 |
* reset on crash recovery. Even if we could, we could not decode combocids |
| 47 |
* which are only tracked in the original backend's memory. To work around |
| 48 |
* that, heapam writes an extra WAL record (XLOG_HEAP2_NEW_CID) every time a |
| 49 |
* catalog row is modified, which includes the cmin and cmax of the |
| 50 |
* tuple. During decoding, we insert the ctid->(cmin,cmax) mappings into the |
| 51 |
* reorder buffer, and use them at visibility checks instead of the cmin/cmax |
| 52 |
* on the tuple itself. Check the reorderbuffer.c's comment above |
| 53 |
* ResolveCminCmaxDuringDecoding() for details. |
| 54 |
* |
| 55 |
* To facilitate all this we need our own visibility routine, as the normal |
| 56 |
* ones are optimized for different usecases. |
| 57 |
* |
| 58 |
* To replace the normal catalog snapshots with decoding ones use the |
| 59 |
* SetupHistoricSnapshot() and TeardownHistoricSnapshot() functions. |
| 60 |
* |
| 61 |
* |
| 62 |
* |
| 63 |
* The snapbuild machinery is starting up in several stages, as illustrated |
| 64 |
* by the following graph describing the SnapBuild->state transitions: |
| 65 |
* |
| 66 |
* +-------------------------+ |
| 67 |
* +----| START |-------------+ |
| 68 |
* | +-------------------------+ | |
| 69 |
* | | | |
| 70 |
* | | | |
| 71 |
* | running_xacts #1 | |
| 72 |
* | | | |
| 73 |
* | | | |
| 74 |
* | v | |
| 75 |
* | +-------------------------+ v |
| 76 |
* | | BUILDING_SNAPSHOT |------------>| |
| 77 |
* | +-------------------------+ | |
| 78 |
* | | | |
| 79 |
* | | | |
| 80 |
* | running_xacts #2, xacts from #1 finished | |
| 81 |
* | | | |
| 82 |
* | | | |
| 83 |
* | v | |
| 84 |
* | +-------------------------+ v |
| 85 |
* | | FULL_SNAPSHOT |------------>| |
| 86 |
* | +-------------------------+ | |
| 87 |
* | | | |
| 88 |
* running_xacts | saved snapshot |
| 89 |
* with zero xacts | at running_xacts's lsn |
| 90 |
* | | | |
| 91 |
* | running_xacts with xacts from #2 finished | |
| 92 |
* | | | |
| 93 |
* | v | |
| 94 |
* | +-------------------------+ | |
| 95 |
* +--->|SNAPBUILD_CONSISTENT |<------------+ |
| 96 |
* +-------------------------+ |
| 97 |
* |
| 98 |
* Initially the machinery is in the START stage. When an xl_running_xacts |
| 99 |
* record is read that is sufficiently new (above the safe xmin horizon), |
| 100 |
* there's a state transition. If there were no running xacts when the |
| 101 |
* xl_running_xacts record was generated, we'll directly go into CONSISTENT |
| 102 |
* state, otherwise we'll switch to the BUILDING_SNAPSHOT state. Having a full |
| 103 |
* snapshot means that all transactions that start henceforth can be decoded |
| 104 |
* in their entirety, but transactions that started previously can't. In |
| 105 |
* FULL_SNAPSHOT we'll switch into CONSISTENT once all those previously |
| 106 |
* running transactions have committed or aborted. |
| 107 |
* |
| 108 |
* Only transactions that commit after CONSISTENT state has been reached will |
| 109 |
* be replayed, even though they might have started while still in |
| 110 |
* FULL_SNAPSHOT. That ensures that we'll reach a point where no previous |
| 111 |
* changes has been exported, but all the following ones will be. That point |
| 112 |
* is a convenient point to initialize replication from, which is why we |
| 113 |
* export a snapshot at that point, which *can* be used to read normal data. |
| 114 |
* |
| 115 |
* Copyright (c) 2012-2025, PostgreSQL Global Development Group |
| 116 |
* |
| 117 |
* IDENTIFICATION |
| 118 |
* src/backend/replication/logical/snapbuild.c |
| 119 |
* |
| 120 |
*------------------------------------------------------------------------- |
| 121 |
*/ |
| 122 |
|
| 123 |
#include "postgres.h" |
| 124 |
|
| 125 |
#include <sys/stat.h> |
| 126 |
#include <unistd.h> |
| 127 |
|
| 128 |
#include "access/heapam_xlog.h" |
| 129 |
#include "access/transam.h" |
| 130 |
#include "access/xact.h" |
| 131 |
#include "common/file_utils.h" |
| 132 |
#include "miscadmin.h" |
| 133 |
#include "pgstat.h" |
| 134 |
#include "replication/logical.h" |
| 135 |
#include "replication/reorderbuffer.h" |
| 136 |
#include "replication/snapbuild.h" |
| 137 |
#include "replication/snapbuild_internal.h" |
| 138 |
#include "storage/fd.h" |
| 139 |
#include "storage/lmgr.h" |
| 140 |
#include "storage/proc.h" |
| 141 |
#include "storage/procarray.h" |
| 142 |
#include "storage/standby.h" |
| 143 |
#include "utils/builtins.h" |
| 144 |
#include "utils/memutils.h" |
| 145 |
#include "utils/snapmgr.h" |
| 146 |
#include "utils/snapshot.h" |
| 147 |
/* |
| 148 |
* Starting a transaction -- which we need to do while exporting a snapshot -- |
| 149 |
* removes knowledge about the previously used resowner, so we save it here. |
| 150 |
*/ |
| 151 |
static ResourceOwner SavedResourceOwnerDuringExport = NULL; |
| 152 |
static bool ExportInProgress = false; |
| 153 |
|
| 154 |
/* ->committed and ->catchange manipulation */ |
| 155 |
static void SnapBuildPurgeOlderTxn(SnapBuild *builder); |
| 156 |
|
| 157 |
/* snapshot building/manipulation/distribution functions */ |
| 158 |
static Snapshot SnapBuildBuildSnapshot(SnapBuild *builder); |
| 159 |
|
| 160 |
static void SnapBuildFreeSnapshot(Snapshot snap); |
| 161 |
|
| 162 |
static void SnapBuildSnapIncRefcount(Snapshot snap); |
| 163 |
|
| 164 |
static void SnapBuildDistributeSnapshotAndInval(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid); |
| 165 |
|
| 166 |
static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, |
| 167 |
uint32 xinfo); |
| 168 |
|
| 169 |
/* xlog reading helper functions for SnapBuildProcessRunningXacts */ |
| 170 |
static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); |
| 171 |
static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); |
| 172 |
|
| 173 |
/* serialization functions */ |
| 174 |
static void SnapBuildSerialize(SnapBuild *builder, XLogRecPtr lsn); |
| 175 |
static bool SnapBuildRestore(SnapBuild *builder, XLogRecPtr lsn); |
| 176 |
static void SnapBuildRestoreContents(int fd, void *dest, Size size, const char *path); |
| 177 |
|
| 178 |
/* |
| 179 |
* Allocate a new snapshot builder. |
| 180 |
* |
| 181 |
* xmin_horizon is the xid >= which we can be sure no catalog rows have been |
| 182 |
* removed, start_lsn is the LSN >= we want to replay commits. |
| 183 |
*/ |
| 184 |
SnapBuild * |
| 185 |
AllocateSnapshotBuilder(ReorderBuffer *reorder, |
| 186 |
TransactionId xmin_horizon, |
| 187 |
XLogRecPtr start_lsn, |
| 188 |
bool need_full_snapshot, |
| 189 |
bool in_slot_creation, |
| 190 |
XLogRecPtr two_phase_at) |
| 191 |
{ |
| 192 |
MemoryContext context; |
| 193 |
MemoryContext oldcontext; |
| 194 |
SnapBuild *builder; |
| 195 |
|
| 196 |
/* allocate memory in own context, to have better accountability */ |
| 197 |
context = AllocSetContextCreate(CurrentMemoryContext, |
| 198 |
"snapshot builder context", |
| 199 |
ALLOCSET_DEFAULT_SIZES); |
| 200 |
oldcontext = MemoryContextSwitchTo(context); |
| 201 |
|
| 202 |
builder = palloc0(sizeof(SnapBuild)); |
| 203 |
|
| 204 |
builder->state = SNAPBUILD_START; |
| 205 |
builder->context = context; |
| 206 |
builder->reorder = reorder; |
| 207 |
/* Other struct members initialized by zeroing via palloc0 above */ |
| 208 |
|
| 209 |
builder->committed.xcnt = 0; |
| 210 |
builder->committed.xcnt_space = 128; /* arbitrary number */ |
| 211 |
builder->committed.xip = |
| 212 |
palloc0(builder->committed.xcnt_space * sizeof(TransactionId)); |
| 213 |
builder->committed.includes_all_transactions = true; |
| 214 |
|
| 215 |
builder->catchange.xcnt = 0; |
| 216 |
builder->catchange.xip = NULL; |
| 217 |
|
| 218 |
builder->initial_xmin_horizon = xmin_horizon; |
| 219 |
builder->start_decoding_at = start_lsn; |
| 220 |
builder->in_slot_creation = in_slot_creation; |
| 221 |
builder->building_full_snapshot = need_full_snapshot; |
| 222 |
builder->two_phase_at = two_phase_at; |
| 223 |
|
| 224 |
MemoryContextSwitchTo(oldcontext); |
| 225 |
|
| 226 |
return builder; |
| 227 |
} |
| 228 |
|
| 229 |
/* |
| 230 |
* Free a snapshot builder. |
| 231 |
*/ |
| 232 |
void |
| 233 |
FreeSnapshotBuilder(SnapBuild *builder) |
| 234 |
{ |
| 235 |
MemoryContext context = builder->context; |
| 236 |
|
| 237 |
/* free snapshot explicitly, that contains some error checking */ |
| 238 |
if (builder->snapshot != NULL) |
| 239 |
{ |
| 240 |
SnapBuildSnapDecRefcount(builder->snapshot); |
| 241 |
builder->snapshot = NULL; |
| 242 |
} |
| 243 |
|
| 244 |
/* other resources are deallocated via memory context reset */ |
| 245 |
MemoryContextDelete(context); |
| 246 |
} |
| 247 |
|
| 248 |
/* |
| 249 |
* Free an unreferenced snapshot that has previously been built by us. |
| 250 |
*/ |
| 251 |
static void |
| 252 |
SnapBuildFreeSnapshot(Snapshot snap) |
| 253 |
{ |
| 254 |
/* make sure we don't get passed an external snapshot */ |
| 255 |
Assert(snap->snapshot_type == SNAPSHOT_HISTORIC_MVCC); |
| 256 |
|
| 257 |
/* make sure nobody modified our snapshot */ |
| 258 |
Assert(snap->curcid == FirstCommandId); |
| 259 |
Assert(!snap->suboverflowed); |
| 260 |
Assert(!snap->takenDuringRecovery); |
| 261 |
Assert(snap->regd_count == 0); |
| 262 |
|
| 263 |
/* slightly more likely, so it's checked even without c-asserts */ |
| 264 |
if (snap->copied) |
| 265 |
elog(ERROR, "cannot free a copied snapshot"); |
| 266 |
|
| 267 |
if (snap->active_count) |
| 268 |
elog(ERROR, "cannot free an active snapshot"); |
| 269 |
|
| 270 |
pfree(snap); |
| 271 |
} |
| 272 |
|
| 273 |
/* |
| 274 |
* In which state of snapshot building are we? |
| 275 |
*/ |
| 276 |
SnapBuildState |
| 277 |
SnapBuildCurrentState(SnapBuild *builder) |
| 278 |
{ |
| 279 |
return builder->state; |
| 280 |
} |
| 281 |
|
| 282 |
/* |
| 283 |
* Return the LSN at which the two-phase decoding was first enabled. |
| 284 |
*/ |
| 285 |
XLogRecPtr |
| 286 |
SnapBuildGetTwoPhaseAt(SnapBuild *builder) |
| 287 |
{ |
| 288 |
return builder->two_phase_at; |
| 289 |
} |
| 290 |
|
| 291 |
/* |
| 292 |
* Set the LSN at which two-phase decoding is enabled. |
| 293 |
*/ |
| 294 |
void |
| 295 |
SnapBuildSetTwoPhaseAt(SnapBuild *builder, XLogRecPtr ptr) |
| 296 |
{ |
| 297 |
builder->two_phase_at = ptr; |
| 298 |
} |
| 299 |
|
| 300 |
/* |
| 301 |
* Should the contents of transaction ending at 'ptr' be decoded? |
| 302 |
*/ |
| 303 |
bool |
| 304 |
SnapBuildXactNeedsSkip(SnapBuild *builder, XLogRecPtr ptr) |
| 305 |
{ |
| 306 |
return ptr < builder->start_decoding_at; |
| 307 |
} |
| 308 |
|
| 309 |
/* |
| 310 |
* Increase refcount of a snapshot. |
| 311 |
* |
| 312 |
* This is used when handing out a snapshot to some external resource or when |
| 313 |
* adding a Snapshot as builder->snapshot. |
| 314 |
*/ |
| 315 |
static void |
| 316 |
SnapBuildSnapIncRefcount(Snapshot snap) |
| 317 |
{ |
| 318 |
snap->active_count++; |
| 319 |
} |
| 320 |
|
| 321 |
/* |
| 322 |
* Decrease refcount of a snapshot and free if the refcount reaches zero. |
| 323 |
* |
| 324 |
* Externally visible, so that external resources that have been handed an |
| 325 |
* IncRef'ed Snapshot can adjust its refcount easily. |
| 326 |
*/ |
| 327 |
void |
| 328 |
SnapBuildSnapDecRefcount(Snapshot snap) |
| 329 |
{ |
| 330 |
/* make sure we don't get passed an external snapshot */ |
| 331 |
Assert(snap->snapshot_type == SNAPSHOT_HISTORIC_MVCC); |
| 332 |
|
| 333 |
/* make sure nobody modified our snapshot */ |
| 334 |
Assert(snap->curcid == FirstCommandId); |
| 335 |
Assert(!snap->suboverflowed); |
| 336 |
Assert(!snap->takenDuringRecovery); |
| 337 |
|
| 338 |
Assert(snap->regd_count == 0); |
| 339 |
|
| 340 |
Assert(snap->active_count > 0); |
| 341 |
|
| 342 |
/* slightly more likely, so it's checked even without casserts */ |
| 343 |
if (snap->copied) |
| 344 |
elog(ERROR, "cannot free a copied snapshot"); |
| 345 |
|
| 346 |
snap->active_count--; |
| 347 |
if (snap->active_count == 0) |
| 348 |
SnapBuildFreeSnapshot(snap); |
| 349 |
} |
| 350 |
|
| 351 |
/* |
| 352 |
* Build a new snapshot, based on currently committed catalog-modifying |
| 353 |
* transactions. |
| 354 |
* |
| 355 |
* In-progress transactions with catalog access are *not* allowed to modify |
| 356 |
* these snapshots; they have to copy them and fill in appropriate ->curcid |
| 357 |
* and ->subxip/subxcnt values. |
| 358 |
*/ |
| 359 |
static Snapshot |
| 360 |
SnapBuildBuildSnapshot(SnapBuild *builder) |
| 361 |
{ |
| 362 |
Snapshot snapshot; |
| 363 |
Size ssize; |
| 364 |
|
| 365 |
Assert(builder->state >= SNAPBUILD_FULL_SNAPSHOT); |
| 366 |
|
| 367 |
ssize = sizeof(SnapshotData) |
| 368 |
+ sizeof(TransactionId) * builder->committed.xcnt |
| 369 |
+ sizeof(TransactionId) * 1 /* toplevel xid */ ; |
| 370 |
|
| 371 |
snapshot = MemoryContextAllocZero(builder->context, ssize); |
| 372 |
|
| 373 |
snapshot->snapshot_type = SNAPSHOT_HISTORIC_MVCC; |
| 374 |
|
| 375 |
/* |
| 376 |
* We misuse the original meaning of SnapshotData's xip and subxip fields |
| 377 |
* to make the more fitting for our needs. |
| 378 |
* |
| 379 |
* In the 'xip' array we store transactions that have to be treated as |
| 380 |
* committed. Since we will only ever look at tuples from transactions |
| 381 |
* that have modified the catalog it's more efficient to store those few |
| 382 |
* that exist between xmin and xmax (frequently there are none). |
| 383 |
* |
| 384 |
* Snapshots that are used in transactions that have modified the catalog |
| 385 |
* also use the 'subxip' array to store their toplevel xid and all the |
| 386 |
* subtransaction xids so we can recognize when we need to treat rows as |
| 387 |
* visible that are not in xip but still need to be visible. Subxip only |
| 388 |
* gets filled when the transaction is copied into the context of a |
| 389 |
* catalog modifying transaction since we otherwise share a snapshot |
| 390 |
* between transactions. As long as a txn hasn't modified the catalog it |
| 391 |
* doesn't need to treat any uncommitted rows as visible, so there is no |
| 392 |
* need for those xids. |
| 393 |
* |
| 394 |
* Both arrays are qsort'ed so that we can use bsearch() on them. |
| 395 |
*/ |
| 396 |
Assert(TransactionIdIsNormal(builder->xmin)); |
| 397 |
Assert(TransactionIdIsNormal(builder->xmax)); |
| 398 |
|
| 399 |
snapshot->xmin = builder->xmin; |
| 400 |
snapshot->xmax = builder->xmax; |
| 401 |
|
| 402 |
/* store all transactions to be treated as committed by this snapshot */ |
| 403 |
snapshot->xip = |
| 404 |
(TransactionId *) ((char *) snapshot + sizeof(SnapshotData)); |
| 405 |
snapshot->xcnt = builder->committed.xcnt; |
| 406 |
memcpy(snapshot->xip, |
| 407 |
builder->committed.xip, |
| 408 |
builder->committed.xcnt * sizeof(TransactionId)); |
| 409 |
|
| 410 |
/* sort so we can bsearch() */ |
| 411 |
qsort(snapshot->xip, snapshot->xcnt, sizeof(TransactionId), xidComparator); |
| 412 |
|
| 413 |
/* |
| 414 |
* Initially, subxip is empty, i.e. it's a snapshot to be used by |
| 415 |
* transactions that don't modify the catalog. Will be filled by |
| 416 |
* ReorderBufferCopySnap() if necessary. |
| 417 |
*/ |
| 418 |
snapshot->subxcnt = 0; |
| 419 |
snapshot->subxip = NULL; |
| 420 |
|
| 421 |
snapshot->suboverflowed = false; |
| 422 |
snapshot->takenDuringRecovery = false; |
| 423 |
snapshot->copied = false; |
| 424 |
snapshot->curcid = FirstCommandId; |
| 425 |
snapshot->active_count = 0; |
| 426 |
snapshot->regd_count = 0; |
| 427 |
snapshot->snapXactCompletionCount = 0; |
| 428 |
|
| 429 |
return snapshot; |
| 430 |
} |
| 431 |
|
| 432 |
/* |
| 433 |
* Build the initial slot snapshot and convert it to a normal snapshot that |
| 434 |
* is understood by HeapTupleSatisfiesMVCC. |
| 435 |
* |
| 436 |
* The snapshot will be usable directly in current transaction or exported |
| 437 |
* for loading in different transaction. |
| 438 |
*/ |
| 439 |
Snapshot |
| 440 |
SnapBuildInitialSnapshot(SnapBuild *builder) |
| 441 |
{ |
| 442 |
Snapshot snap; |
| 443 |
TransactionId xid; |
| 444 |
TransactionId safeXid; |
| 445 |
TransactionId *newxip; |
| 446 |
int newxcnt = 0; |
| 447 |
|
| 448 |
Assert(XactIsoLevel == XACT_REPEATABLE_READ); |
| 449 |
Assert(builder->building_full_snapshot); |
| 450 |
|
| 451 |
/* don't allow older snapshots */ |
| 452 |
InvalidateCatalogSnapshot(); /* about to overwrite MyProc->xmin */ |
| 453 |
if (HaveRegisteredOrActiveSnapshot()) |
| 454 |
elog(ERROR, "cannot build an initial slot snapshot when snapshots exist"); |
| 455 |
Assert(!HistoricSnapshotActive()); |
| 456 |
|
| 457 |
if (builder->state != SNAPBUILD_CONSISTENT) |
| 458 |
elog(ERROR, "cannot build an initial slot snapshot before reaching a consistent state"); |
| 459 |
|
| 460 |
if (!builder->committed.includes_all_transactions) |
| 461 |
elog(ERROR, "cannot build an initial slot snapshot, not all transactions are monitored anymore"); |
| 462 |
|
| 463 |
/* so we don't overwrite the existing value */ |
| 464 |
if (TransactionIdIsValid(MyProc->xmin)) |
| 465 |
elog(ERROR, "cannot build an initial slot snapshot when MyProc->xmin already is valid"); |
| 466 |
|
| 467 |
snap = SnapBuildBuildSnapshot(builder); |
| 468 |
|
| 469 |
/* |
| 470 |
* We know that snap->xmin is alive, enforced by the logical xmin |
| 471 |
* mechanism. Due to that we can do this without locks, we're only |
| 472 |
* changing our own value. |
| 473 |
* |
| 474 |
* Building an initial snapshot is expensive and an unenforced xmin |
| 475 |
* horizon would have bad consequences, therefore always double-check that |
| 476 |
* the horizon is enforced. |
| 477 |
*/ |
| 478 |
LWLockAcquire(ProcArrayLock, LW_SHARED); |
| 479 |
safeXid = GetOldestSafeDecodingTransactionId(false); |
| 480 |
LWLockRelease(ProcArrayLock); |
| 481 |
|
| 482 |
if (TransactionIdFollows(safeXid, snap->xmin)) |
| 483 |
elog(ERROR, "cannot build an initial slot snapshot as oldest safe xid %u follows snapshot's xmin %u", |
| 484 |
safeXid, snap->xmin); |
| 485 |
|
| 486 |
MyProc->xmin = snap->xmin; |
| 487 |
|
| 488 |
/* allocate in transaction context */ |
| 489 |
newxip = (TransactionId *) |
| 490 |
palloc(sizeof(TransactionId) * GetMaxSnapshotXidCount()); |
| 491 |
|
| 492 |
/* |
| 493 |
* snapbuild.c builds transactions in an "inverted" manner, which means it |
| 494 |
* stores committed transactions in ->xip, not ones in progress. Build a |
| 495 |
* classical snapshot by marking all non-committed transactions as |
| 496 |
* in-progress. This can be expensive. |
| 497 |
*/ |
| 498 |
for (xid = snap->xmin; NormalTransactionIdPrecedes(xid, snap->xmax);) |
| 499 |
{ |
| 500 |
void *test; |
| 501 |
|
| 502 |
/* |
| 503 |
* Check whether transaction committed using the decoding snapshot |
| 504 |
* meaning of ->xip. |
| 505 |
*/ |
| 506 |
test = bsearch(&xid, snap->xip, snap->xcnt, |
| 507 |
sizeof(TransactionId), xidComparator); |
| 508 |
|
| 509 |
if (test == NULL) |
| 510 |
{ |
| 511 |
if (newxcnt >= GetMaxSnapshotXidCount()) |
| 512 |
ereport(ERROR, |
| 513 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 514 |
errmsg("initial slot snapshot too large"))); |
| 515 |
|
| 516 |
newxip[newxcnt++] = xid; |
| 517 |
} |
| 518 |
|
| 519 |
TransactionIdAdvance(xid); |
| 520 |
} |
| 521 |
|
| 522 |
/* adjust remaining snapshot fields as needed */ |
| 523 |
snap->snapshot_type = SNAPSHOT_MVCC; |
| 524 |
snap->xcnt = newxcnt; |
| 525 |
snap->xip = newxip; |
| 526 |
|
| 527 |
return snap; |
| 528 |
} |
| 529 |
|
| 530 |
/* |
| 531 |
* Export a snapshot so it can be set in another session with SET TRANSACTION |
| 532 |
* SNAPSHOT. |
| 533 |
* |
| 534 |
* For that we need to start a transaction in the current backend as the |
| 535 |
* importing side checks whether the source transaction is still open to make |
| 536 |
* sure the xmin horizon hasn't advanced since then. |
| 537 |
*/ |
| 538 |
const char * |
| 539 |
SnapBuildExportSnapshot(SnapBuild *builder) |
| 540 |
{ |
| 541 |
Snapshot snap; |
| 542 |
char *snapname; |
| 543 |
|
| 544 |
if (IsTransactionOrTransactionBlock()) |
| 545 |
elog(ERROR, "cannot export a snapshot from within a transaction"); |
| 546 |
|
| 547 |
if (SavedResourceOwnerDuringExport) |
| 548 |
elog(ERROR, "can only export one snapshot at a time"); |
| 549 |
|
| 550 |
SavedResourceOwnerDuringExport = CurrentResourceOwner; |
| 551 |
ExportInProgress = true; |
| 552 |
|
| 553 |
StartTransactionCommand(); |
| 554 |
|
| 555 |
/* There doesn't seem to a nice API to set these */ |
| 556 |
XactIsoLevel = XACT_REPEATABLE_READ; |
| 557 |
XactReadOnly = true; |
| 558 |
|
| 559 |
snap = SnapBuildInitialSnapshot(builder); |
| 560 |
|
| 561 |
/* |
| 562 |
* now that we've built a plain snapshot, make it active and use the |
| 563 |
* normal mechanisms for exporting it |
| 564 |
*/ |
| 565 |
snapname = ExportSnapshot(snap); |
| 566 |
|
| 567 |
ereport(LOG, |
| 568 |
(errmsg_plural("exported logical decoding snapshot: \"%s\" with %u transaction ID", |
| 569 |
"exported logical decoding snapshot: \"%s\" with %u transaction IDs", |
| 570 |
snap->xcnt, |
| 571 |
snapname, snap->xcnt))); |
| 572 |
return snapname; |
| 573 |
} |
| 574 |
|
| 575 |
/* |
| 576 |
* Ensure there is a snapshot and if not build one for current transaction. |
| 577 |
*/ |
| 578 |
Snapshot |
| 579 |
SnapBuildGetOrBuildSnapshot(SnapBuild *builder) |
| 580 |
{ |
| 581 |
Assert(builder->state == SNAPBUILD_CONSISTENT); |
| 582 |
|
| 583 |
/* only build a new snapshot if we don't have a prebuilt one */ |
| 584 |
if (builder->snapshot == NULL) |
| 585 |
{ |
| 586 |
builder->snapshot = SnapBuildBuildSnapshot(builder); |
| 587 |
/* increase refcount for the snapshot builder */ |
| 588 |
SnapBuildSnapIncRefcount(builder->snapshot); |
| 589 |
} |
| 590 |
|
| 591 |
return builder->snapshot; |
| 592 |
} |
| 593 |
|
| 594 |
/* |
| 595 |
* Reset a previously SnapBuildExportSnapshot()'ed snapshot if there is |
| 596 |
* any. Aborts the previously started transaction and resets the resource |
| 597 |
* owner back to its original value. |
| 598 |
*/ |
| 599 |
void |
| 600 |
SnapBuildClearExportedSnapshot(void) |
| 601 |
{ |
| 602 |
ResourceOwner tmpResOwner; |
| 603 |
|
| 604 |
/* nothing exported, that is the usual case */ |
| 605 |
if (!ExportInProgress) |
| 606 |
return; |
| 607 |
|
| 608 |
if (!IsTransactionState()) |
| 609 |
elog(ERROR, "clearing exported snapshot in wrong transaction state"); |
| 610 |
|
| 611 |
/* |
| 612 |
* AbortCurrentTransaction() takes care of resetting the snapshot state, |
| 613 |
* so remember SavedResourceOwnerDuringExport. |
| 614 |
*/ |
| 615 |
tmpResOwner = SavedResourceOwnerDuringExport; |
| 616 |
|
| 617 |
/* make sure nothing could have ever happened */ |
| 618 |
AbortCurrentTransaction(); |
| 619 |
|
| 620 |
CurrentResourceOwner = tmpResOwner; |
| 621 |
} |
| 622 |
|
| 623 |
/* |
| 624 |
* Clear snapshot export state during transaction abort. |
| 625 |
*/ |
| 626 |
void |
| 627 |
SnapBuildResetExportedSnapshotState(void) |
| 628 |
{ |
| 629 |
SavedResourceOwnerDuringExport = NULL; |
| 630 |
ExportInProgress = false; |
| 631 |
} |
| 632 |
|
| 633 |
/* |
| 634 |
* Handle the effects of a single heap change, appropriate to the current state |
| 635 |
* of the snapshot builder and returns whether changes made at (xid, lsn) can |
| 636 |
* be decoded. |
| 637 |
*/ |
| 638 |
bool |
| 639 |
SnapBuildProcessChange(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn) |
| 640 |
{ |
| 641 |
/* |
| 642 |
* We can't handle data in transactions if we haven't built a snapshot |
| 643 |
* yet, so don't store them. |
| 644 |
*/ |
| 645 |
if (builder->state < SNAPBUILD_FULL_SNAPSHOT) |
| 646 |
return false; |
| 647 |
|
| 648 |
/* |
| 649 |
* No point in keeping track of changes in transactions that we don't have |
| 650 |
* enough information about to decode. This means that they started before |
| 651 |
* we got into the SNAPBUILD_FULL_SNAPSHOT state. |
| 652 |
*/ |
| 653 |
if (builder->state < SNAPBUILD_CONSISTENT && |
| 654 |
TransactionIdPrecedes(xid, builder->next_phase_at)) |
| 655 |
return false; |
| 656 |
|
| 657 |
/* |
| 658 |
* If the reorderbuffer doesn't yet have a snapshot, add one now, it will |
| 659 |
* be needed to decode the change we're currently processing. |
| 660 |
*/ |
| 661 |
if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, xid)) |
| 662 |
{ |
| 663 |
/* only build a new snapshot if we don't have a prebuilt one */ |
| 664 |
if (builder->snapshot == NULL) |
| 665 |
{ |
| 666 |
builder->snapshot = SnapBuildBuildSnapshot(builder); |
| 667 |
/* increase refcount for the snapshot builder */ |
| 668 |
SnapBuildSnapIncRefcount(builder->snapshot); |
| 669 |
} |
| 670 |
|
| 671 |
/* |
| 672 |
* Increase refcount for the transaction we're handing the snapshot |
| 673 |
* out to. |
| 674 |
*/ |
| 675 |
SnapBuildSnapIncRefcount(builder->snapshot); |
| 676 |
ReorderBufferSetBaseSnapshot(builder->reorder, xid, lsn, |
| 677 |
builder->snapshot); |
| 678 |
} |
| 679 |
|
| 680 |
return true; |
| 681 |
} |
| 682 |
|
| 683 |
/* |
| 684 |
* Do CommandId/combo CID handling after reading an xl_heap_new_cid record. |
| 685 |
* This implies that a transaction has done some form of write to system |
| 686 |
* catalogs. |
| 687 |
*/ |
| 688 |
void |
| 689 |
SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, |
| 690 |
XLogRecPtr lsn, xl_heap_new_cid *xlrec) |
| 691 |
{ |
| 692 |
CommandId cid; |
| 693 |
|
| 694 |
/* |
| 695 |
* we only log new_cid's if a catalog tuple was modified, so mark the |
| 696 |
* transaction as containing catalog modifications |
| 697 |
*/ |
| 698 |
ReorderBufferXidSetCatalogChanges(builder->reorder, xid, lsn); |
| 699 |
|
| 700 |
ReorderBufferAddNewTupleCids(builder->reorder, xlrec->top_xid, lsn, |
| 701 |
xlrec->target_locator, xlrec->target_tid, |
| 702 |
xlrec->cmin, xlrec->cmax, |
| 703 |
xlrec->combocid); |
| 704 |
|
| 705 |
/* figure out new command id */ |
| 706 |
if (xlrec->cmin != InvalidCommandId && |
| 707 |
xlrec->cmax != InvalidCommandId) |
| 708 |
cid = Max(xlrec->cmin, xlrec->cmax); |
| 709 |
else if (xlrec->cmax != InvalidCommandId) |
| 710 |
cid = xlrec->cmax; |
| 711 |
else if (xlrec->cmin != InvalidCommandId) |
| 712 |
cid = xlrec->cmin; |
| 713 |
else |
| 714 |
{ |
| 715 |
cid = InvalidCommandId; /* silence compiler */ |
| 716 |
elog(ERROR, "xl_heap_new_cid record without a valid CommandId"); |
| 717 |
} |
| 718 |
|
| 719 |
ReorderBufferAddNewCommandId(builder->reorder, xid, lsn, cid + 1); |
| 720 |
} |
| 721 |
|
| 722 |
/* |
| 723 |
* Add a new Snapshot and invalidation messages to all transactions we're |
| 724 |
* decoding that currently are in-progress so they can see new catalog contents |
| 725 |
* made by the transaction that just committed. This is necessary because those |
| 726 |
* in-progress transactions will use the new catalog's contents from here on |
| 727 |
* (at the very least everything they do needs to be compatible with newer |
| 728 |
* catalog contents). |
| 729 |
*/ |
| 730 |
static void |
| 731 |
SnapBuildDistributeSnapshotAndInval(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid) |
| 732 |
{ |
| 733 |
dlist_iter txn_i; |
| 734 |
ReorderBufferTXN *txn; |
| 735 |
|
| 736 |
/* |
| 737 |
* Iterate through all toplevel transactions. This can include |
| 738 |
* subtransactions which we just don't yet know to be that, but that's |
| 739 |
* fine, they will just get an unnecessary snapshot and invalidations |
| 740 |
* queued. |
| 741 |
*/ |
| 742 |
dlist_foreach(txn_i, &builder->reorder->toplevel_by_lsn) |
| 743 |
{ |
| 744 |
txn = dlist_container(ReorderBufferTXN, node, txn_i.cur); |
| 745 |
|
| 746 |
Assert(TransactionIdIsValid(txn->xid)); |
| 747 |
|
| 748 |
/* |
| 749 |
* If we don't have a base snapshot yet, there are no changes in this |
| 750 |
* transaction which in turn implies we don't yet need a snapshot at |
| 751 |
* all. We'll add a snapshot when the first change gets queued. |
| 752 |
* |
| 753 |
* Similarly, we don't need to add invalidations to a transaction |
| 754 |
* whose base snapshot is not yet set. Once a base snapshot is built, |
| 755 |
* it will include the xids of committed transactions that have |
| 756 |
* modified the catalog, thus reflecting the new catalog contents. The |
| 757 |
* existing catalog cache will have already been invalidated after |
| 758 |
* processing the invalidations in the transaction that modified |
| 759 |
* catalogs, ensuring that a fresh cache is constructed during |
| 760 |
* decoding. |
| 761 |
* |
| 762 |
* NB: This works correctly even for subtransactions because |
| 763 |
* ReorderBufferAssignChild() takes care to transfer the base snapshot |
| 764 |
* to the top-level transaction, and while iterating the changequeue |
| 765 |
* we'll get the change from the subtxn. |
| 766 |
*/ |
| 767 |
if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, txn->xid)) |
| 768 |
continue; |
| 769 |
|
| 770 |
/* |
| 771 |
* We don't need to add snapshot or invalidations to prepared |
| 772 |
* transactions as they should not see the new catalog contents. |
| 773 |
*/ |
| 774 |
if (rbtxn_is_prepared(txn)) |
| 775 |
continue; |
| 776 |
|
| 777 |
elog(DEBUG2, "adding a new snapshot and invalidations to %u at %X/%X", |
| 778 |
txn->xid, LSN_FORMAT_ARGS(lsn)); |
| 779 |
|
| 780 |
/* |
| 781 |
* increase the snapshot's refcount for the transaction we are handing |
| 782 |
* it out to |
| 783 |
*/ |
| 784 |
SnapBuildSnapIncRefcount(builder->snapshot); |
| 785 |
ReorderBufferAddSnapshot(builder->reorder, txn->xid, lsn, |
| 786 |
builder->snapshot); |
| 787 |
|
| 788 |
/* |
| 789 |
* Add invalidation messages to the reorder buffer of in-progress |
| 790 |
* transactions except the current committed transaction, for which we |
| 791 |
* will execute invalidations at the end. |
| 792 |
* |
| 793 |
* It is required, otherwise, we will end up using the stale catcache |
| 794 |
* contents built by the current transaction even after its decoding, |
| 795 |
* which should have been invalidated due to concurrent catalog |
| 796 |
* changing transaction. |
| 797 |
* |
| 798 |
* Distribute only the invalidation messages generated by the current |
| 799 |
* committed transaction. Invalidation messages received from other |
| 800 |
* transactions would have already been propagated to the relevant |
| 801 |
* in-progress transactions. This transaction would have processed |
| 802 |
* those invalidations, ensuring that subsequent transactions observe |
| 803 |
* a consistent cache state. |
| 804 |
*/ |
| 805 |
if (txn->xid != xid) |
| 806 |
{ |
| 807 |
uint32 ninvalidations; |
| 808 |
SharedInvalidationMessage *msgs = NULL; |
| 809 |
|
| 810 |
ninvalidations = ReorderBufferGetInvalidations(builder->reorder, |
| 811 |
xid, &msgs); |
| 812 |
|
| 813 |
if (ninvalidations > 0) |
| 814 |
{ |
| 815 |
Assert(msgs != NULL); |
| 816 |
|
| 817 |
ReorderBufferAddDistributedInvalidations(builder->reorder, |
| 818 |
txn->xid, lsn, |
| 819 |
ninvalidations, msgs); |
| 820 |
} |
| 821 |
} |
| 822 |
} |
| 823 |
} |
| 824 |
|
| 825 |
/* |
| 826 |
* Keep track of a new catalog changing transaction that has committed. |
| 827 |
*/ |
| 828 |
static void |
| 829 |
SnapBuildAddCommittedTxn(SnapBuild *builder, TransactionId xid) |
| 830 |
{ |
| 831 |
Assert(TransactionIdIsValid(xid)); |
| 832 |
|
| 833 |
if (builder->committed.xcnt == builder->committed.xcnt_space) |
| 834 |
{ |
| 835 |
builder->committed.xcnt_space = builder->committed.xcnt_space * 2 + 1; |
| 836 |
|
| 837 |
elog(DEBUG1, "increasing space for committed transactions to %u", |
| 838 |
(uint32) builder->committed.xcnt_space); |
| 839 |
|
| 840 |
builder->committed.xip = repalloc(builder->committed.xip, |
| 841 |
builder->committed.xcnt_space * sizeof(TransactionId)); |
| 842 |
} |
| 843 |
|
| 844 |
/* |
| 845 |
* TODO: It might make sense to keep the array sorted here instead of |
| 846 |
* doing it every time we build a new snapshot. On the other hand this |
| 847 |
* gets called repeatedly when a transaction with subtransactions commits. |
| 848 |
*/ |
| 849 |
builder->committed.xip[builder->committed.xcnt++] = xid; |
| 850 |
} |
| 851 |
|
| 852 |
/* |
| 853 |
* Remove knowledge about transactions we treat as committed or containing catalog |
| 854 |
* changes that are smaller than ->xmin. Those won't ever get checked via |
| 855 |
* the ->committed or ->catchange array, respectively. The committed xids will |
| 856 |
* get checked via the clog machinery. |
| 857 |
* |
| 858 |
* We can ideally remove the transaction from catchange array once it is |
| 859 |
* finished (committed/aborted) but that could be costly as we need to maintain |
| 860 |
* the xids order in the array. |
| 861 |
*/ |
| 862 |
static void |
| 863 |
SnapBuildPurgeOlderTxn(SnapBuild *builder) |
| 864 |
{ |
| 865 |
int off; |
| 866 |
TransactionId *workspace; |
| 867 |
int surviving_xids = 0; |
| 868 |
|
| 869 |
/* not ready yet */ |
| 870 |
if (!TransactionIdIsNormal(builder->xmin)) |
| 871 |
return; |
| 872 |
|
| 873 |
/* TODO: Neater algorithm than just copying and iterating? */ |
| 874 |
workspace = |
| 875 |
MemoryContextAlloc(builder->context, |
| 876 |
builder->committed.xcnt * sizeof(TransactionId)); |
| 877 |
|
| 878 |
/* copy xids that still are interesting to workspace */ |
| 879 |
for (off = 0; off < builder->committed.xcnt; off++) |
| 880 |
{ |
| 881 |
if (NormalTransactionIdPrecedes(builder->committed.xip[off], |
| 882 |
builder->xmin)) |
| 883 |
; /* remove */ |
| 884 |
else |
| 885 |
workspace[surviving_xids++] = builder->committed.xip[off]; |
| 886 |
} |
| 887 |
|
| 888 |
/* copy workspace back to persistent state */ |
| 889 |
memcpy(builder->committed.xip, workspace, |
| 890 |
surviving_xids * sizeof(TransactionId)); |
| 891 |
|
| 892 |
elog(DEBUG3, "purged committed transactions from %u to %u, xmin: %u, xmax: %u", |
| 893 |
(uint32) builder->committed.xcnt, (uint32) surviving_xids, |
| 894 |
builder->xmin, builder->xmax); |
| 895 |
builder->committed.xcnt = surviving_xids; |
| 896 |
|
| 897 |
pfree(workspace); |
| 898 |
|
| 899 |
/* |
| 900 |
* Purge xids in ->catchange as well. The purged array must also be sorted |
| 901 |
* in xidComparator order. |
| 902 |
*/ |
| 903 |
if (builder->catchange.xcnt > 0) |
| 904 |
{ |
| 905 |
/* |
| 906 |
* Since catchange.xip is sorted, we find the lower bound of xids that |
| 907 |
* are still interesting. |
| 908 |
*/ |
| 909 |
for (off = 0; off < builder->catchange.xcnt; off++) |
| 910 |
{ |
| 911 |
if (TransactionIdFollowsOrEquals(builder->catchange.xip[off], |
| 912 |
builder->xmin)) |
| 913 |
break; |
| 914 |
} |
| 915 |
|
| 916 |
surviving_xids = builder->catchange.xcnt - off; |
| 917 |
|
| 918 |
if (surviving_xids > 0) |
| 919 |
{ |
| 920 |
memmove(builder->catchange.xip, &(builder->catchange.xip[off]), |
| 921 |
surviving_xids * sizeof(TransactionId)); |
| 922 |
} |
| 923 |
else |
| 924 |
{ |
| 925 |
pfree(builder->catchange.xip); |
| 926 |
builder->catchange.xip = NULL; |
| 927 |
} |
| 928 |
|
| 929 |
elog(DEBUG3, "purged catalog modifying transactions from %u to %u, xmin: %u, xmax: %u", |
| 930 |
(uint32) builder->catchange.xcnt, (uint32) surviving_xids, |
| 931 |
builder->xmin, builder->xmax); |
| 932 |
builder->catchange.xcnt = surviving_xids; |
| 933 |
} |
| 934 |
} |
| 935 |
|
| 936 |
/* |
| 937 |
* Handle everything that needs to be done when a transaction commits |
| 938 |
*/ |
| 939 |
void |
| 940 |
SnapBuildCommitTxn(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid, |
| 941 |
int nsubxacts, TransactionId *subxacts, uint32 xinfo) |
| 942 |
{ |
| 943 |
int nxact; |
| 944 |
|
| 945 |
bool needs_snapshot = false; |
| 946 |
bool needs_timetravel = false; |
| 947 |
bool sub_needs_timetravel = false; |
| 948 |
|
| 949 |
TransactionId xmax = xid; |
| 950 |
|
| 951 |
/* |
| 952 |
* Transactions preceding BUILDING_SNAPSHOT will neither be decoded, nor |
| 953 |
* will they be part of a snapshot. So we don't need to record anything. |
| 954 |
*/ |
| 955 |
if (builder->state == SNAPBUILD_START || |
| 956 |
(builder->state == SNAPBUILD_BUILDING_SNAPSHOT && |
| 957 |
TransactionIdPrecedes(xid, builder->next_phase_at))) |
| 958 |
{ |
| 959 |
/* ensure that only commits after this are getting replayed */ |
| 960 |
if (builder->start_decoding_at <= lsn) |
| 961 |
builder->start_decoding_at = lsn + 1; |
| 962 |
return; |
| 963 |
} |
| 964 |
|
| 965 |
if (builder->state < SNAPBUILD_CONSISTENT) |
| 966 |
{ |
| 967 |
/* ensure that only commits after this are getting replayed */ |
| 968 |
if (builder->start_decoding_at <= lsn) |
| 969 |
builder->start_decoding_at = lsn + 1; |
| 970 |
|
| 971 |
/* |
| 972 |
* If building an exportable snapshot, force xid to be tracked, even |
| 973 |
* if the transaction didn't modify the catalog. |
| 974 |
*/ |
| 975 |
if (builder->building_full_snapshot) |
| 976 |
{ |
| 977 |
needs_timetravel = true; |
| 978 |
} |
| 979 |
} |
| 980 |
|
| 981 |
for (nxact = 0; nxact < nsubxacts; nxact++) |
| 982 |
{ |
| 983 |
TransactionId subxid = subxacts[nxact]; |
| 984 |
|
| 985 |
/* |
| 986 |
* Add subtransaction to base snapshot if catalog modifying, we don't |
| 987 |
* distinguish to toplevel transactions there. |
| 988 |
*/ |
| 989 |
if (SnapBuildXidHasCatalogChanges(builder, subxid, xinfo)) |
| 990 |
{ |
| 991 |
sub_needs_timetravel = true; |
| 992 |
needs_snapshot = true; |
| 993 |
|
| 994 |
elog(DEBUG1, "found subtransaction %u:%u with catalog changes", |
| 995 |
xid, subxid); |
| 996 |
|
| 997 |
SnapBuildAddCommittedTxn(builder, subxid); |
| 998 |
|
| 999 |
if (NormalTransactionIdFollows(subxid, xmax)) |
| 1000 |
xmax = subxid; |
| 1001 |
} |
| 1002 |
|
| 1003 |
/* |
| 1004 |
* If we're forcing timetravel we also need visibility information |
| 1005 |
* about subtransaction, so keep track of subtransaction's state, even |
| 1006 |
* if not catalog modifying. Don't need to distribute a snapshot in |
| 1007 |
* that case. |
| 1008 |
*/ |
| 1009 |
else if (needs_timetravel) |
| 1010 |
{ |
| 1011 |
SnapBuildAddCommittedTxn(builder, subxid); |
| 1012 |
if (NormalTransactionIdFollows(subxid, xmax)) |
| 1013 |
xmax = subxid; |
| 1014 |
} |
| 1015 |
} |
| 1016 |
|
| 1017 |
/* if top-level modified catalog, it'll need a snapshot */ |
| 1018 |
if (SnapBuildXidHasCatalogChanges(builder, xid, xinfo)) |
| 1019 |
{ |
| 1020 |
elog(DEBUG2, "found top level transaction %u, with catalog changes", |
| 1021 |
xid); |
| 1022 |
needs_snapshot = true; |
| 1023 |
needs_timetravel = true; |
| 1024 |
SnapBuildAddCommittedTxn(builder, xid); |
| 1025 |
} |
| 1026 |
else if (sub_needs_timetravel) |
| 1027 |
{ |
| 1028 |
/* track toplevel txn as well, subxact alone isn't meaningful */ |
| 1029 |
elog(DEBUG2, "forced transaction %u to do timetravel due to one of its subtransactions", |
| 1030 |
xid); |
| 1031 |
needs_timetravel = true; |
| 1032 |
SnapBuildAddCommittedTxn(builder, xid); |
| 1033 |
} |
| 1034 |
else if (needs_timetravel) |
| 1035 |
{ |
| 1036 |
elog(DEBUG2, "forced transaction %u to do timetravel", xid); |
| 1037 |
|
| 1038 |
SnapBuildAddCommittedTxn(builder, xid); |
| 1039 |
} |
| 1040 |
|
| 1041 |
if (!needs_timetravel) |
| 1042 |
{ |
| 1043 |
/* record that we cannot export a general snapshot anymore */ |
| 1044 |
builder->committed.includes_all_transactions = false; |
| 1045 |
} |
| 1046 |
|
| 1047 |
Assert(!needs_snapshot || needs_timetravel); |
| 1048 |
|
| 1049 |
/* |
| 1050 |
* Adjust xmax of the snapshot builder, we only do that for committed, |
| 1051 |
* catalog modifying, transactions, everything else isn't interesting for |
| 1052 |
* us since we'll never look at the respective rows. |
| 1053 |
*/ |
| 1054 |
if (needs_timetravel && |
| 1055 |
(!TransactionIdIsValid(builder->xmax) || |
| 1056 |
TransactionIdFollowsOrEquals(xmax, builder->xmax))) |
| 1057 |
{ |
| 1058 |
builder->xmax = xmax; |
| 1059 |
TransactionIdAdvance(builder->xmax); |
| 1060 |
} |
| 1061 |
|
| 1062 |
/* if there's any reason to build a historic snapshot, do so now */ |
| 1063 |
if (needs_snapshot) |
| 1064 |
{ |
| 1065 |
/* |
| 1066 |
* If we haven't built a complete snapshot yet there's no need to hand |
| 1067 |
* it out, it wouldn't (and couldn't) be used anyway. |
| 1068 |
*/ |
| 1069 |
if (builder->state < SNAPBUILD_FULL_SNAPSHOT) |
| 1070 |
return; |
| 1071 |
|
| 1072 |
/* |
| 1073 |
* Decrease the snapshot builder's refcount of the old snapshot, note |
| 1074 |
* that it still will be used if it has been handed out to the |
| 1075 |
* reorderbuffer earlier. |
| 1076 |
*/ |
| 1077 |
if (builder->snapshot) |
| 1078 |
SnapBuildSnapDecRefcount(builder->snapshot); |
| 1079 |
|
| 1080 |
builder->snapshot = SnapBuildBuildSnapshot(builder); |
| 1081 |
|
| 1082 |
/* we might need to execute invalidations, add snapshot */ |
| 1083 |
if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, xid)) |
| 1084 |
{ |
| 1085 |
SnapBuildSnapIncRefcount(builder->snapshot); |
| 1086 |
ReorderBufferSetBaseSnapshot(builder->reorder, xid, lsn, |
| 1087 |
builder->snapshot); |
| 1088 |
} |
| 1089 |
|
| 1090 |
/* refcount of the snapshot builder for the new snapshot */ |
| 1091 |
SnapBuildSnapIncRefcount(builder->snapshot); |
| 1092 |
|
| 1093 |
/* |
| 1094 |
* Add a new catalog snapshot and invalidations messages to all |
| 1095 |
* currently running transactions. |
| 1096 |
*/ |
| 1097 |
SnapBuildDistributeSnapshotAndInval(builder, lsn, xid); |
| 1098 |
} |
| 1099 |
} |
| 1100 |
|
| 1101 |
/* |
| 1102 |
* Check the reorder buffer and the snapshot to see if the given transaction has |
| 1103 |
* modified catalogs. |
| 1104 |
*/ |
| 1105 |
static inline bool |
| 1106 |
SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, |
| 1107 |
uint32 xinfo) |
| 1108 |
{ |
| 1109 |
if (ReorderBufferXidHasCatalogChanges(builder->reorder, xid)) |
| 1110 |
return true; |
| 1111 |
|
| 1112 |
/* |
| 1113 |
* The transactions that have changed catalogs must have invalidation |
| 1114 |
* info. |
| 1115 |
*/ |
| 1116 |
if (!(xinfo & XACT_XINFO_HAS_INVALS)) |
| 1117 |
return false; |
| 1118 |
|
| 1119 |
/* Check the catchange XID array */ |
| 1120 |
return ((builder->catchange.xcnt > 0) && |
| 1121 |
(bsearch(&xid, builder->catchange.xip, builder->catchange.xcnt, |
| 1122 |
sizeof(TransactionId), xidComparator) != NULL)); |
| 1123 |
} |
| 1124 |
|
| 1125 |
/* ----------------------------------- |
| 1126 |
* Snapshot building functions dealing with xlog records |
| 1127 |
* ----------------------------------- |
| 1128 |
*/ |
| 1129 |
|
| 1130 |
/* |
| 1131 |
* Process a running xacts record, and use its information to first build a |
| 1132 |
* historic snapshot and later to release resources that aren't needed |
| 1133 |
* anymore. |
| 1134 |
*/ |
| 1135 |
void |
| 1136 |
SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) |
| 1137 |
{ |
| 1138 |
ReorderBufferTXN *txn; |
| 1139 |
TransactionId xmin; |
| 1140 |
|
| 1141 |
/* |
| 1142 |
* If we're not consistent yet, inspect the record to see whether it |
| 1143 |
* allows to get closer to being consistent. If we are consistent, dump |
| 1144 |
* our snapshot so others or we, after a restart, can use it. |
| 1145 |
*/ |
| 1146 |
if (builder->state < SNAPBUILD_CONSISTENT) |
| 1147 |
{ |
| 1148 |
/* returns false if there's no point in performing cleanup just yet */ |
| 1149 |
if (!SnapBuildFindSnapshot(builder, lsn, running)) |
| 1150 |
return; |
| 1151 |
} |
| 1152 |
else |
| 1153 |
SnapBuildSerialize(builder, lsn); |
| 1154 |
|
| 1155 |
/* |
| 1156 |
* Update range of interesting xids based on the running xacts |
| 1157 |
* information. We don't increase ->xmax using it, because once we are in |
| 1158 |
* a consistent state we can do that ourselves and much more efficiently |
| 1159 |
* so, because we only need to do it for catalog transactions since we |
| 1160 |
* only ever look at those. |
| 1161 |
* |
| 1162 |
* NB: We only increase xmax when a catalog modifying transaction commits |
| 1163 |
* (see SnapBuildCommitTxn). Because of this, xmax can be lower than |
| 1164 |
* xmin, which looks odd but is correct and actually more efficient, since |
| 1165 |
* we hit fast paths in heapam_visibility.c. |
| 1166 |
*/ |
| 1167 |
builder->xmin = running->oldestRunningXid; |
| 1168 |
|
| 1169 |
/* Remove transactions we don't need to keep track off anymore */ |
| 1170 |
SnapBuildPurgeOlderTxn(builder); |
| 1171 |
|
| 1172 |
/* |
| 1173 |
* Advance the xmin limit for the current replication slot, to allow |
| 1174 |
* vacuum to clean up the tuples this slot has been protecting. |
| 1175 |
* |
| 1176 |
* The reorderbuffer might have an xmin among the currently running |
| 1177 |
* snapshots; use it if so. If not, we need only consider the snapshots |
| 1178 |
* we'll produce later, which can't be less than the oldest running xid in |
| 1179 |
* the record we're reading now. |
| 1180 |
*/ |
| 1181 |
xmin = ReorderBufferGetOldestXmin(builder->reorder); |
| 1182 |
if (xmin == InvalidTransactionId) |
| 1183 |
xmin = running->oldestRunningXid; |
| 1184 |
elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", |
| 1185 |
builder->xmin, builder->xmax, running->oldestRunningXid, xmin); |
| 1186 |
LogicalIncreaseXminForSlot(lsn, xmin); |
| 1187 |
|
| 1188 |
/* |
| 1189 |
* Also tell the slot where we can restart decoding from. We don't want to |
| 1190 |
* do that after every commit because changing that implies an fsync of |
| 1191 |
* the logical slot's state file, so we only do it every time we see a |
| 1192 |
* running xacts record. |
| 1193 |
* |
| 1194 |
* Do so by looking for the oldest in progress transaction (determined by |
| 1195 |
* the first LSN of any of its relevant records). Every transaction |
| 1196 |
* remembers the last location we stored the snapshot to disk before its |
| 1197 |
* beginning. That point is where we can restart from. |
| 1198 |
*/ |
| 1199 |
|
| 1200 |
/* |
| 1201 |
* Can't know about a serialized snapshot's location if we're not |
| 1202 |
* consistent. |
| 1203 |
*/ |
| 1204 |
if (builder->state < SNAPBUILD_CONSISTENT) |
| 1205 |
return; |
| 1206 |
|
| 1207 |
txn = ReorderBufferGetOldestTXN(builder->reorder); |
| 1208 |
|
| 1209 |
/* |
| 1210 |
* oldest ongoing txn might have started when we didn't yet serialize |
| 1211 |
* anything because we hadn't reached a consistent state yet. |
| 1212 |
*/ |
| 1213 |
if (txn != NULL && txn->restart_decoding_lsn != InvalidXLogRecPtr) |
| 1214 |
LogicalIncreaseRestartDecodingForSlot(lsn, txn->restart_decoding_lsn); |
| 1215 |
|
| 1216 |
/* |
| 1217 |
* No in-progress transaction, can reuse the last serialized snapshot if |
| 1218 |
* we have one. |
| 1219 |
*/ |
| 1220 |
else if (txn == NULL && |
| 1221 |
builder->reorder->current_restart_decoding_lsn != InvalidXLogRecPtr && |
| 1222 |
builder->last_serialized_snapshot != InvalidXLogRecPtr) |
| 1223 |
LogicalIncreaseRestartDecodingForSlot(lsn, |
| 1224 |
builder->last_serialized_snapshot); |
| 1225 |
} |
| 1226 |
|
| 1227 |
|
| 1228 |
/* |
| 1229 |
* Build the start of a snapshot that's capable of decoding the catalog. |
| 1230 |
* |
| 1231 |
* Helper function for SnapBuildProcessRunningXacts() while we're not yet |
| 1232 |
* consistent. |
| 1233 |
* |
| 1234 |
* Returns true if there is a point in performing internal maintenance/cleanup |
| 1235 |
* using the xl_running_xacts record. |
| 1236 |
*/ |
| 1237 |
static bool |
| 1238 |
SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) |
| 1239 |
{ |
| 1240 |
/* --- |
| 1241 |
* Build catalog decoding snapshot incrementally using information about |
| 1242 |
* the currently running transactions. There are several ways to do that: |
| 1243 |
* |
| 1244 |
* a) There were no running transactions when the xl_running_xacts record |
| 1245 |
* was inserted, jump to CONSISTENT immediately. We might find such a |
| 1246 |
* state while waiting on c)'s sub-states. |
| 1247 |
* |
| 1248 |
* b) This (in a previous run) or another decoding slot serialized a |
| 1249 |
* snapshot to disk that we can use. Can't use this method while finding |
| 1250 |
* the start point for decoding changes as the restart LSN would be an |
| 1251 |
* arbitrary LSN but we need to find the start point to extract changes |
| 1252 |
* where we won't see the data for partial transactions. Also, we cannot |
| 1253 |
* use this method when a slot needs a full snapshot for export or direct |
| 1254 |
* use, as that snapshot will only contain catalog modifying transactions. |
| 1255 |
* |
| 1256 |
* c) First incrementally build a snapshot for catalog tuples |
| 1257 |
* (BUILDING_SNAPSHOT), that requires all, already in-progress, |
| 1258 |
* transactions to finish. Every transaction starting after that |
| 1259 |
* (FULL_SNAPSHOT state), has enough information to be decoded. But |
| 1260 |
* for older running transactions no viable snapshot exists yet, so |
| 1261 |
* CONSISTENT will only be reached once all of those have finished. |
| 1262 |
* --- |
| 1263 |
*/ |
| 1264 |
|
| 1265 |
/* |
| 1266 |
* xl_running_xacts record is older than what we can use, we might not |
| 1267 |
* have all necessary catalog rows anymore. |
| 1268 |
*/ |
| 1269 |
if (TransactionIdIsNormal(builder->initial_xmin_horizon) && |
| 1270 |
NormalTransactionIdPrecedes(running->oldestRunningXid, |
| 1271 |
builder->initial_xmin_horizon)) |
| 1272 |
{ |
| 1273 |
ereport(DEBUG1, |
| 1274 |
(errmsg_internal("skipping snapshot at %X/%X while building logical decoding snapshot, xmin horizon too low", |
| 1275 |
LSN_FORMAT_ARGS(lsn)), |
| 1276 |
errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", |
| 1277 |
builder->initial_xmin_horizon, running->oldestRunningXid))); |
| 1278 |
|
| 1279 |
|
| 1280 |
SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); |
| 1281 |
|
| 1282 |
return true; |
| 1283 |
} |
| 1284 |
|
| 1285 |
/* |
| 1286 |
* a) No transaction were running, we can jump to consistent. |
| 1287 |
* |
| 1288 |
* This is not affected by races around xl_running_xacts, because we can |
| 1289 |
* miss transaction commits, but currently not transactions starting. |
| 1290 |
* |
| 1291 |
* NB: We might have already started to incrementally assemble a snapshot, |
| 1292 |
* so we need to be careful to deal with that. |
| 1293 |
*/ |
| 1294 |
if (running->oldestRunningXid == running->nextXid) |
| 1295 |
{ |
| 1296 |
if (builder->start_decoding_at == InvalidXLogRecPtr || |
| 1297 |
builder->start_decoding_at <= lsn) |
| 1298 |
/* can decode everything after this */ |
| 1299 |
builder->start_decoding_at = lsn + 1; |
| 1300 |
|
| 1301 |
/* As no transactions were running xmin/xmax can be trivially set. */ |
| 1302 |
builder->xmin = running->nextXid; /* < are finished */ |
| 1303 |
builder->xmax = running->nextXid; /* >= are running */ |
| 1304 |
|
| 1305 |
/* so we can safely use the faster comparisons */ |
| 1306 |
Assert(TransactionIdIsNormal(builder->xmin)); |
| 1307 |
Assert(TransactionIdIsNormal(builder->xmax)); |
| 1308 |
|
| 1309 |
builder->state = SNAPBUILD_CONSISTENT; |
| 1310 |
builder->next_phase_at = InvalidTransactionId; |
| 1311 |
|
| 1312 |
ereport(LOG, |
| 1313 |
(errmsg("logical decoding found consistent point at %X/%X", |
| 1314 |
LSN_FORMAT_ARGS(lsn)), |
| 1315 |
errdetail("There are no running transactions."))); |
| 1316 |
|
| 1317 |
return false; |
| 1318 |
} |
| 1319 |
|
| 1320 |
/* |
| 1321 |
* b) valid on disk state and while neither building full snapshot nor |
| 1322 |
* creating a slot. |
| 1323 |
*/ |
| 1324 |
else if (!builder->building_full_snapshot && |
| 1325 |
!builder->in_slot_creation && |
| 1326 |
SnapBuildRestore(builder, lsn)) |
| 1327 |
{ |
| 1328 |
/* there won't be any state to cleanup */ |
| 1329 |
return false; |
| 1330 |
} |
| 1331 |
|
| 1332 |
/* |
| 1333 |
* c) transition from START to BUILDING_SNAPSHOT. |
| 1334 |
* |
| 1335 |
* In START state, and a xl_running_xacts record with running xacts is |
| 1336 |
* encountered. In that case, switch to BUILDING_SNAPSHOT state, and |
| 1337 |
* record xl_running_xacts->nextXid. Once all running xacts have finished |
| 1338 |
* (i.e. they're all >= nextXid), we have a complete catalog snapshot. It |
| 1339 |
* might look that we could use xl_running_xacts's ->xids information to |
| 1340 |
* get there quicker, but that is problematic because transactions marked |
| 1341 |
* as running, might already have inserted their commit record - it's |
| 1342 |
* infeasible to change that with locking. |
| 1343 |
*/ |
| 1344 |
else if (builder->state == SNAPBUILD_START) |
| 1345 |
{ |
| 1346 |
builder->state = SNAPBUILD_BUILDING_SNAPSHOT; |
| 1347 |
builder->next_phase_at = running->nextXid; |
| 1348 |
|
| 1349 |
/* |
| 1350 |
* Start with an xmin/xmax that's correct for future, when all the |
| 1351 |
* currently running transactions have finished. We'll update both |
| 1352 |
* while waiting for the pending transactions to finish. |
| 1353 |
*/ |
| 1354 |
builder->xmin = running->nextXid; /* < are finished */ |
| 1355 |
builder->xmax = running->nextXid; /* >= are running */ |
| 1356 |
|
| 1357 |
/* so we can safely use the faster comparisons */ |
| 1358 |
Assert(TransactionIdIsNormal(builder->xmin)); |
| 1359 |
Assert(TransactionIdIsNormal(builder->xmax)); |
| 1360 |
|
| 1361 |
ereport(LOG, |
| 1362 |
(errmsg("logical decoding found initial starting point at %X/%X", |
| 1363 |
LSN_FORMAT_ARGS(lsn)), |
| 1364 |
errdetail("Waiting for transactions (approximately %d) older than %u to end.", |
| 1365 |
running->xcnt, running->nextXid))); |
| 1366 |
|
| 1367 |
SnapBuildWaitSnapshot(running, running->nextXid); |
| 1368 |
} |
| 1369 |
|
| 1370 |
/* |
| 1371 |
* c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. |
| 1372 |
* |
| 1373 |
* In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid |
| 1374 |
* is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This |
| 1375 |
* means all transactions starting afterwards have enough information to |
| 1376 |
* be decoded. Switch to FULL_SNAPSHOT. |
| 1377 |
*/ |
| 1378 |
else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && |
| 1379 |
TransactionIdPrecedesOrEquals(builder->next_phase_at, |
| 1380 |
running->oldestRunningXid)) |
| 1381 |
{ |
| 1382 |
builder->state = SNAPBUILD_FULL_SNAPSHOT; |
| 1383 |
builder->next_phase_at = running->nextXid; |
| 1384 |
|
| 1385 |
ereport(LOG, |
| 1386 |
(errmsg("logical decoding found initial consistent point at %X/%X", |
| 1387 |
LSN_FORMAT_ARGS(lsn)), |
| 1388 |
errdetail("Waiting for transactions (approximately %d) older than %u to end.", |
| 1389 |
running->xcnt, running->nextXid))); |
| 1390 |
|
| 1391 |
SnapBuildWaitSnapshot(running, running->nextXid); |
| 1392 |
} |
| 1393 |
|
| 1394 |
/* |
| 1395 |
* c) transition from FULL_SNAPSHOT to CONSISTENT. |
| 1396 |
* |
| 1397 |
* In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is |
| 1398 |
* >= than nextXid from when we switched to FULL_SNAPSHOT. This means all |
| 1399 |
* transactions that are currently in progress have a catalog snapshot, |
| 1400 |
* and all their changes have been collected. Switch to CONSISTENT. |
| 1401 |
*/ |
| 1402 |
else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && |
| 1403 |
TransactionIdPrecedesOrEquals(builder->next_phase_at, |
| 1404 |
running->oldestRunningXid)) |
| 1405 |
{ |
| 1406 |
builder->state = SNAPBUILD_CONSISTENT; |
| 1407 |
builder->next_phase_at = InvalidTransactionId; |
| 1408 |
|
| 1409 |
ereport(LOG, |
| 1410 |
(errmsg("logical decoding found consistent point at %X/%X", |
| 1411 |
LSN_FORMAT_ARGS(lsn)), |
| 1412 |
errdetail("There are no old transactions anymore."))); |
| 1413 |
} |
| 1414 |
|
| 1415 |
/* |
| 1416 |
* We already started to track running xacts and need to wait for all |
| 1417 |
* in-progress ones to finish. We fall through to the normal processing of |
| 1418 |
* records so incremental cleanup can be performed. |
| 1419 |
*/ |
| 1420 |
return true; |
| 1421 |
} |
| 1422 |
|
| 1423 |
/* --- |
| 1424 |
* Iterate through xids in record, wait for all older than the cutoff to |
| 1425 |
* finish. Then, if possible, log a new xl_running_xacts record. |
| 1426 |
* |
| 1427 |
* This isn't required for the correctness of decoding, but to: |
| 1428 |
* a) allow isolationtester to notice that we're currently waiting for |
| 1429 |
* something. |
| 1430 |
* b) log a new xl_running_xacts record where it'd be helpful, without having |
| 1431 |
* to wait for bgwriter or checkpointer. |
| 1432 |
* --- |
| 1433 |
*/ |
| 1434 |
static void |
| 1435 |
SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) |
| 1436 |
{ |
| 1437 |
int off; |
| 1438 |
|
| 1439 |
for (off = 0; off < running->xcnt; off++) |
| 1440 |
{ |
| 1441 |
TransactionId xid = running->xids[off]; |
| 1442 |
|
| 1443 |
/* |
| 1444 |
* Upper layers should prevent that we ever need to wait on ourselves. |
| 1445 |
* Check anyway, since failing to do so would either result in an |
| 1446 |
* endless wait or an Assert() failure. |
| 1447 |
*/ |
| 1448 |
if (TransactionIdIsCurrentTransactionId(xid)) |
| 1449 |
elog(ERROR, "waiting for ourselves"); |
| 1450 |
|
| 1451 |
if (TransactionIdFollows(xid, cutoff)) |
| 1452 |
continue; |
| 1453 |
|
| 1454 |
XactLockTableWait(xid, NULL, NULL, XLTW_None); |
| 1455 |
} |
| 1456 |
|
| 1457 |
/* |
| 1458 |
* All transactions we needed to finish finished - try to ensure there is |
| 1459 |
* another xl_running_xacts record in a timely manner, without having to |
| 1460 |
* wait for bgwriter or checkpointer to log one. During recovery we can't |
| 1461 |
* enforce that, so we'll have to wait. |
| 1462 |
*/ |
| 1463 |
if (!RecoveryInProgress()) |
| 1464 |
{ |
| 1465 |
LogStandbySnapshot(); |
| 1466 |
} |
| 1467 |
} |
| 1468 |
|
| 1469 |
#define SnapBuildOnDiskConstantSize \ |
| 1470 |
offsetof(SnapBuildOnDisk, builder) |
| 1471 |
#define SnapBuildOnDiskNotChecksummedSize \ |
| 1472 |
offsetof(SnapBuildOnDisk, version) |
| 1473 |
|
| 1474 |
#define SNAPBUILD_MAGIC 0x51A1E001 |
| 1475 |
#define SNAPBUILD_VERSION 6 |
| 1476 |
|
| 1477 |
/* |
| 1478 |
* Store/Load a snapshot from disk, depending on the snapshot builder's state. |
| 1479 |
* |
| 1480 |
* Supposed to be used by external (i.e. not snapbuild.c) code that just read |
| 1481 |
* a record that's a potential location for a serialized snapshot. |
| 1482 |
*/ |
| 1483 |
void |
| 1484 |
SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn) |
| 1485 |
{ |
| 1486 |
if (builder->state < SNAPBUILD_CONSISTENT) |
| 1487 |
SnapBuildRestore(builder, lsn); |
| 1488 |
else |
| 1489 |
SnapBuildSerialize(builder, lsn); |
| 1490 |
} |
| 1491 |
|
| 1492 |
/* |
| 1493 |
* Serialize the snapshot 'builder' at the location 'lsn' if it hasn't already |
| 1494 |
* been done by another decoding process. |
| 1495 |
*/ |
| 1496 |
static void |
| 1497 |
SnapBuildSerialize(SnapBuild *builder, XLogRecPtr lsn) |
| 1498 |
{ |
| 1499 |
Size needed_length; |
| 1500 |
SnapBuildOnDisk *ondisk = NULL; |
| 1501 |
TransactionId *catchange_xip = NULL; |
| 1502 |
MemoryContext old_ctx; |
| 1503 |
size_t catchange_xcnt; |
| 1504 |
char *ondisk_c; |
| 1505 |
int fd; |
| 1506 |
char tmppath[MAXPGPATH]; |
| 1507 |
char path[MAXPGPATH]; |
| 1508 |
int ret; |
| 1509 |
struct stat stat_buf; |
| 1510 |
Size sz; |
| 1511 |
|
| 1512 |
Assert(lsn != InvalidXLogRecPtr); |
| 1513 |
Assert(builder->last_serialized_snapshot == InvalidXLogRecPtr || |
| 1514 |
builder->last_serialized_snapshot <= lsn); |
| 1515 |
|
| 1516 |
/* |
| 1517 |
* no point in serializing if we cannot continue to work immediately after |
| 1518 |
* restoring the snapshot |
| 1519 |
*/ |
| 1520 |
if (builder->state < SNAPBUILD_CONSISTENT) |
| 1521 |
return; |
| 1522 |
|
| 1523 |
/* consistent snapshots have no next phase */ |
| 1524 |
Assert(builder->next_phase_at == InvalidTransactionId); |
| 1525 |
|
| 1526 |
/* |
| 1527 |
* We identify snapshots by the LSN they are valid for. We don't need to |
| 1528 |
* include timelines in the name as each LSN maps to exactly one timeline |
| 1529 |
* unless the user used pg_resetwal or similar. If a user did so, there's |
| 1530 |
* no hope continuing to decode anyway. |
| 1531 |
*/ |
| 1532 |
sprintf(path, "%s/%X-%X.snap", |
| 1533 |
PG_LOGICAL_SNAPSHOTS_DIR, |
| 1534 |
LSN_FORMAT_ARGS(lsn)); |
| 1535 |
|
| 1536 |
/* |
| 1537 |
* first check whether some other backend already has written the snapshot |
| 1538 |
* for this LSN. It's perfectly fine if there's none, so we accept ENOENT |
| 1539 |
* as a valid state. Everything else is an unexpected error. |
| 1540 |
*/ |
| 1541 |
ret = stat(path, &stat_buf); |
| 1542 |
|
| 1543 |
if (ret != 0 && errno != ENOENT) |
| 1544 |
ereport(ERROR, |
| 1545 |
(errcode_for_file_access(), |
| 1546 |
errmsg("could not stat file \"%s\": %m", path))); |
| 1547 |
|
| 1548 |
else if (ret == 0) |
| 1549 |
{ |
| 1550 |
/* |
| 1551 |
* somebody else has already serialized to this point, don't overwrite |
| 1552 |
* but remember location, so we don't need to read old data again. |
| 1553 |
* |
| 1554 |
* To be sure it has been synced to disk after the rename() from the |
| 1555 |
* tempfile filename to the real filename, we just repeat the fsync. |
| 1556 |
* That ought to be cheap because in most scenarios it should already |
| 1557 |
* be safely on disk. |
| 1558 |
*/ |
| 1559 |
fsync_fname(path, false); |
| 1560 |
fsync_fname(PG_LOGICAL_SNAPSHOTS_DIR, true); |
| 1561 |
|
| 1562 |
builder->last_serialized_snapshot = lsn; |
| 1563 |
goto out; |
| 1564 |
} |
| 1565 |
|
| 1566 |
/* |
| 1567 |
* there is an obvious race condition here between the time we stat(2) the |
| 1568 |
* file and us writing the file. But we rename the file into place |
| 1569 |
* atomically and all files created need to contain the same data anyway, |
| 1570 |
* so this is perfectly fine, although a bit of a resource waste. Locking |
| 1571 |
* seems like pointless complication. |
| 1572 |
*/ |
| 1573 |
elog(DEBUG1, "serializing snapshot to %s", path); |
| 1574 |
|
| 1575 |
/* to make sure only we will write to this tempfile, include pid */ |
| 1576 |
sprintf(tmppath, "%s/%X-%X.snap.%d.tmp", |
| 1577 |
PG_LOGICAL_SNAPSHOTS_DIR, |
| 1578 |
LSN_FORMAT_ARGS(lsn), MyProcPid); |
| 1579 |
|
| 1580 |
/* |
| 1581 |
* Unlink temporary file if it already exists, needs to have been before a |
| 1582 |
* crash/error since we won't enter this function twice from within a |
| 1583 |
* single decoding slot/backend and the temporary file contains the pid of |
| 1584 |
* the current process. |
| 1585 |
*/ |
| 1586 |
if (unlink(tmppath) != 0 && errno != ENOENT) |
| 1587 |
ereport(ERROR, |
| 1588 |
(errcode_for_file_access(), |
| 1589 |
errmsg("could not remove file \"%s\": %m", tmppath))); |
| 1590 |
|
| 1591 |
old_ctx = MemoryContextSwitchTo(builder->context); |
| 1592 |
|
| 1593 |
/* Get the catalog modifying transactions that are yet not committed */ |
| 1594 |
catchange_xip = ReorderBufferGetCatalogChangesXacts(builder->reorder); |
| 1595 |
catchange_xcnt = dclist_count(&builder->reorder->catchange_txns); |
| 1596 |
|
| 1597 |
needed_length = sizeof(SnapBuildOnDisk) + |
| 1598 |
sizeof(TransactionId) * (builder->committed.xcnt + catchange_xcnt); |
| 1599 |
|
| 1600 |
ondisk_c = palloc0(needed_length); |
| 1601 |
ondisk = (SnapBuildOnDisk *) ondisk_c; |
| 1602 |
ondisk->magic = SNAPBUILD_MAGIC; |
| 1603 |
ondisk->version = SNAPBUILD_VERSION; |
| 1604 |
ondisk->length = needed_length; |
| 1605 |
INIT_CRC32C(ondisk->checksum); |
| 1606 |
COMP_CRC32C(ondisk->checksum, |
| 1607 |
((char *) ondisk) + SnapBuildOnDiskNotChecksummedSize, |
| 1608 |
SnapBuildOnDiskConstantSize - SnapBuildOnDiskNotChecksummedSize); |
| 1609 |
ondisk_c += sizeof(SnapBuildOnDisk); |
| 1610 |
|
| 1611 |
memcpy(&ondisk->builder, builder, sizeof(SnapBuild)); |
| 1612 |
/* NULL-ify memory-only data */ |
| 1613 |
ondisk->builder.context = NULL; |
| 1614 |
ondisk->builder.snapshot = NULL; |
| 1615 |
ondisk->builder.reorder = NULL; |
| 1616 |
ondisk->builder.committed.xip = NULL; |
| 1617 |
ondisk->builder.catchange.xip = NULL; |
| 1618 |
/* update catchange only on disk data */ |
| 1619 |
ondisk->builder.catchange.xcnt = catchange_xcnt; |
| 1620 |
|
| 1621 |
COMP_CRC32C(ondisk->checksum, |
| 1622 |
&ondisk->builder, |
| 1623 |
sizeof(SnapBuild)); |
| 1624 |
|
| 1625 |
/* copy committed xacts */ |
| 1626 |
if (builder->committed.xcnt > 0) |
| 1627 |
{ |
| 1628 |
sz = sizeof(TransactionId) * builder->committed.xcnt; |
| 1629 |
memcpy(ondisk_c, builder->committed.xip, sz); |
| 1630 |
COMP_CRC32C(ondisk->checksum, ondisk_c, sz); |
| 1631 |
ondisk_c += sz; |
| 1632 |
} |
| 1633 |
|
| 1634 |
/* copy catalog modifying xacts */ |
| 1635 |
if (catchange_xcnt > 0) |
| 1636 |
{ |
| 1637 |
sz = sizeof(TransactionId) * catchange_xcnt; |
| 1638 |
memcpy(ondisk_c, catchange_xip, sz); |
| 1639 |
COMP_CRC32C(ondisk->checksum, ondisk_c, sz); |
| 1640 |
ondisk_c += sz; |
| 1641 |
} |
| 1642 |
|
| 1643 |
FIN_CRC32C(ondisk->checksum); |
| 1644 |
|
| 1645 |
/* we have valid data now, open tempfile and write it there */ |
| 1646 |
fd = OpenTransientFile(tmppath, |
| 1647 |
O_CREAT | O_EXCL | O_WRONLY | PG_BINARY); |
| 1648 |
if (fd < 0) |
| 1649 |
ereport(ERROR, |
| 1650 |
(errcode_for_file_access(), |
| 1651 |
errmsg("could not open file \"%s\": %m", tmppath))); |
| 1652 |
|
| 1653 |
errno = 0; |
| 1654 |
pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_WRITE); |
| 1655 |
if ((write(fd, ondisk, needed_length)) != needed_length) |
| 1656 |
{ |
| 1657 |
int save_errno = errno; |
| 1658 |
|
| 1659 |
CloseTransientFile(fd); |
| 1660 |
|
| 1661 |
/* if write didn't set errno, assume problem is no disk space */ |
| 1662 |
errno = save_errno ? save_errno : ENOSPC; |
| 1663 |
ereport(ERROR, |
| 1664 |
(errcode_for_file_access(), |
| 1665 |
errmsg("could not write to file \"%s\": %m", tmppath))); |
| 1666 |
} |
| 1667 |
pgstat_report_wait_end(); |
| 1668 |
|
| 1669 |
/* |
| 1670 |
* fsync the file before renaming so that even if we crash after this we |
| 1671 |
* have either a fully valid file or nothing. |
| 1672 |
* |
| 1673 |
* It's safe to just ERROR on fsync() here because we'll retry the whole |
| 1674 |
* operation including the writes. |
| 1675 |
* |
| 1676 |
* TODO: Do the fsync() via checkpoints/restartpoints, doing it here has |
| 1677 |
* some noticeable overhead since it's performed synchronously during |
| 1678 |
* decoding? |
| 1679 |
*/ |
| 1680 |
pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_SYNC); |
| 1681 |
if (pg_fsync(fd) != 0) |
| 1682 |
{ |
| 1683 |
int save_errno = errno; |
| 1684 |
|
| 1685 |
CloseTransientFile(fd); |
| 1686 |
errno = save_errno; |
| 1687 |
ereport(ERROR, |
| 1688 |
(errcode_for_file_access(), |
| 1689 |
errmsg("could not fsync file \"%s\": %m", tmppath))); |
| 1690 |
} |
| 1691 |
pgstat_report_wait_end(); |
| 1692 |
|
| 1693 |
if (CloseTransientFile(fd) != 0) |
| 1694 |
ereport(ERROR, |
| 1695 |
(errcode_for_file_access(), |
| 1696 |
errmsg("could not close file \"%s\": %m", tmppath))); |
| 1697 |
|
| 1698 |
fsync_fname(PG_LOGICAL_SNAPSHOTS_DIR, true); |
| 1699 |
|
| 1700 |
/* |
| 1701 |
* We may overwrite the work from some other backend, but that's ok, our |
| 1702 |
* snapshot is valid as well, we'll just have done some superfluous work. |
| 1703 |
*/ |
| 1704 |
if (rename(tmppath, path) != 0) |
| 1705 |
{ |
| 1706 |
ereport(ERROR, |
| 1707 |
(errcode_for_file_access(), |
| 1708 |
errmsg("could not rename file \"%s\" to \"%s\": %m", |
| 1709 |
tmppath, path))); |
| 1710 |
} |
| 1711 |
|
| 1712 |
/* make sure we persist */ |
| 1713 |
fsync_fname(path, false); |
| 1714 |
fsync_fname(PG_LOGICAL_SNAPSHOTS_DIR, true); |
| 1715 |
|
| 1716 |
/* |
| 1717 |
* Now there's no way we can lose the dumped state anymore, remember this |
| 1718 |
* as a serialization point. |
| 1719 |
*/ |
| 1720 |
builder->last_serialized_snapshot = lsn; |
| 1721 |
|
| 1722 |
MemoryContextSwitchTo(old_ctx); |
| 1723 |
|
| 1724 |
out: |
| 1725 |
ReorderBufferSetRestartPoint(builder->reorder, |
| 1726 |
builder->last_serialized_snapshot); |
| 1727 |
/* be tidy */ |
| 1728 |
if (ondisk) |
| 1729 |
pfree(ondisk); |
| 1730 |
if (catchange_xip) |
| 1731 |
pfree(catchange_xip); |
| 1732 |
} |
| 1733 |
|
| 1734 |
/* |
| 1735 |
* Restore the logical snapshot file contents to 'ondisk'. |
| 1736 |
* |
| 1737 |
* 'context' is the memory context where the catalog modifying/committed xid |
| 1738 |
* will live. |
| 1739 |
* If 'missing_ok' is true, will not throw an error if the file is not found. |
| 1740 |
*/ |
| 1741 |
bool |
| 1742 |
SnapBuildRestoreSnapshot(SnapBuildOnDisk *ondisk, XLogRecPtr lsn, |
| 1743 |
MemoryContext context, bool missing_ok) |
| 1744 |
{ |
| 1745 |
int fd; |
| 1746 |
pg_crc32c checksum; |
| 1747 |
Size sz; |
| 1748 |
char path[MAXPGPATH]; |
| 1749 |
|
| 1750 |
sprintf(path, "%s/%X-%X.snap", |
| 1751 |
PG_LOGICAL_SNAPSHOTS_DIR, |
| 1752 |
LSN_FORMAT_ARGS(lsn)); |
| 1753 |
|
| 1754 |
fd = OpenTransientFile(path, O_RDONLY | PG_BINARY); |
| 1755 |
|
| 1756 |
if (fd < 0) |
| 1757 |
{ |
| 1758 |
if (missing_ok && errno == ENOENT) |
| 1759 |
return false; |
| 1760 |
|
| 1761 |
ereport(ERROR, |
| 1762 |
(errcode_for_file_access(), |
| 1763 |
errmsg("could not open file \"%s\": %m", path))); |
| 1764 |
} |
| 1765 |
|
| 1766 |
/* ---- |
| 1767 |
* Make sure the snapshot had been stored safely to disk, that's normally |
| 1768 |
* cheap. |
| 1769 |
* Note that we do not need PANIC here, nobody will be able to use the |
| 1770 |
* slot without fsyncing, and saving it won't succeed without an fsync() |
| 1771 |
* either... |
| 1772 |
* ---- |
| 1773 |
*/ |
| 1774 |
fsync_fname(path, false); |
| 1775 |
fsync_fname(PG_LOGICAL_SNAPSHOTS_DIR, true); |
| 1776 |
|
| 1777 |
/* read statically sized portion of snapshot */ |
| 1778 |
SnapBuildRestoreContents(fd, ondisk, SnapBuildOnDiskConstantSize, path); |
| 1779 |
|
| 1780 |
if (ondisk->magic != SNAPBUILD_MAGIC) |
| 1781 |
ereport(ERROR, |
| 1782 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1783 |
errmsg("snapbuild state file \"%s\" has wrong magic number: %u instead of %u", |
| 1784 |
path, ondisk->magic, SNAPBUILD_MAGIC))); |
| 1785 |
|
| 1786 |
if (ondisk->version != SNAPBUILD_VERSION) |
| 1787 |
ereport(ERROR, |
| 1788 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1789 |
errmsg("snapbuild state file \"%s\" has unsupported version: %u instead of %u", |
| 1790 |
path, ondisk->version, SNAPBUILD_VERSION))); |
| 1791 |
|
| 1792 |
INIT_CRC32C(checksum); |
| 1793 |
COMP_CRC32C(checksum, |
| 1794 |
((char *) ondisk) + SnapBuildOnDiskNotChecksummedSize, |
| 1795 |
SnapBuildOnDiskConstantSize - SnapBuildOnDiskNotChecksummedSize); |
| 1796 |
|
| 1797 |
/* read SnapBuild */ |
| 1798 |
SnapBuildRestoreContents(fd, &ondisk->builder, sizeof(SnapBuild), path); |
| 1799 |
COMP_CRC32C(checksum, &ondisk->builder, sizeof(SnapBuild)); |
| 1800 |
|
| 1801 |
/* restore committed xacts information */ |
| 1802 |
if (ondisk->builder.committed.xcnt > 0) |
| 1803 |
{ |
| 1804 |
sz = sizeof(TransactionId) * ondisk->builder.committed.xcnt; |
| 1805 |
ondisk->builder.committed.xip = MemoryContextAllocZero(context, sz); |
| 1806 |
SnapBuildRestoreContents(fd, ondisk->builder.committed.xip, sz, path); |
| 1807 |
COMP_CRC32C(checksum, ondisk->builder.committed.xip, sz); |
| 1808 |
} |
| 1809 |
|
| 1810 |
/* restore catalog modifying xacts information */ |
| 1811 |
if (ondisk->builder.catchange.xcnt > 0) |
| 1812 |
{ |
| 1813 |
sz = sizeof(TransactionId) * ondisk->builder.catchange.xcnt; |
| 1814 |
ondisk->builder.catchange.xip = MemoryContextAllocZero(context, sz); |
| 1815 |
SnapBuildRestoreContents(fd, ondisk->builder.catchange.xip, sz, path); |
| 1816 |
COMP_CRC32C(checksum, ondisk->builder.catchange.xip, sz); |
| 1817 |
} |
| 1818 |
|
| 1819 |
if (CloseTransientFile(fd) != 0) |
| 1820 |
ereport(ERROR, |
| 1821 |
(errcode_for_file_access(), |
| 1822 |
errmsg("could not close file \"%s\": %m", path))); |
| 1823 |
|
| 1824 |
FIN_CRC32C(checksum); |
| 1825 |
|
| 1826 |
/* verify checksum of what we've read */ |
| 1827 |
if (!EQ_CRC32C(checksum, ondisk->checksum)) |
| 1828 |
ereport(ERROR, |
| 1829 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1830 |
errmsg("checksum mismatch for snapbuild state file \"%s\": is %u, should be %u", |
| 1831 |
path, checksum, ondisk->checksum))); |
| 1832 |
|
| 1833 |
return true; |
| 1834 |
} |
| 1835 |
|
| 1836 |
/* |
| 1837 |
* Restore a snapshot into 'builder' if previously one has been stored at the |
| 1838 |
* location indicated by 'lsn'. Returns true if successful, false otherwise. |
| 1839 |
*/ |
| 1840 |
static bool |
| 1841 |
SnapBuildRestore(SnapBuild *builder, XLogRecPtr lsn) |
| 1842 |
{ |
| 1843 |
SnapBuildOnDisk ondisk; |
| 1844 |
|
| 1845 |
/* no point in loading a snapshot if we're already there */ |
| 1846 |
if (builder->state == SNAPBUILD_CONSISTENT) |
| 1847 |
return false; |
| 1848 |
|
| 1849 |
/* validate and restore the snapshot to 'ondisk' */ |
| 1850 |
if (!SnapBuildRestoreSnapshot(&ondisk, lsn, builder->context, true)) |
| 1851 |
return false; |
| 1852 |
|
| 1853 |
/* |
| 1854 |
* ok, we now have a sensible snapshot here, figure out if it has more |
| 1855 |
* information than we have. |
| 1856 |
*/ |
| 1857 |
|
| 1858 |
/* |
| 1859 |
* We are only interested in consistent snapshots for now, comparing |
| 1860 |
* whether one incomplete snapshot is more "advanced" seems to be |
| 1861 |
* unnecessarily complex. |
| 1862 |
*/ |
| 1863 |
if (ondisk.builder.state < SNAPBUILD_CONSISTENT) |
| 1864 |
goto snapshot_not_interesting; |
| 1865 |
|
| 1866 |
/* |
| 1867 |
* Don't use a snapshot that requires an xmin that we cannot guarantee to |
| 1868 |
* be available. |
| 1869 |
*/ |
| 1870 |
if (TransactionIdPrecedes(ondisk.builder.xmin, builder->initial_xmin_horizon)) |
| 1871 |
goto snapshot_not_interesting; |
| 1872 |
|
| 1873 |
/* |
| 1874 |
* Consistent snapshots have no next phase. Reset next_phase_at as it is |
| 1875 |
* possible that an old value may remain. |
| 1876 |
*/ |
| 1877 |
Assert(ondisk.builder.next_phase_at == InvalidTransactionId); |
| 1878 |
builder->next_phase_at = InvalidTransactionId; |
| 1879 |
|
| 1880 |
/* ok, we think the snapshot is sensible, copy over everything important */ |
| 1881 |
builder->xmin = ondisk.builder.xmin; |
| 1882 |
builder->xmax = ondisk.builder.xmax; |
| 1883 |
builder->state = ondisk.builder.state; |
| 1884 |
|
| 1885 |
builder->committed.xcnt = ondisk.builder.committed.xcnt; |
| 1886 |
/* We only allocated/stored xcnt, not xcnt_space xids ! */ |
| 1887 |
/* don't overwrite preallocated xip, if we don't have anything here */ |
| 1888 |
if (builder->committed.xcnt > 0) |
| 1889 |
{ |
| 1890 |
pfree(builder->committed.xip); |
| 1891 |
builder->committed.xcnt_space = ondisk.builder.committed.xcnt; |
| 1892 |
builder->committed.xip = ondisk.builder.committed.xip; |
| 1893 |
} |
| 1894 |
ondisk.builder.committed.xip = NULL; |
| 1895 |
|
| 1896 |
/* set catalog modifying transactions */ |
| 1897 |
if (builder->catchange.xip) |
| 1898 |
pfree(builder->catchange.xip); |
| 1899 |
builder->catchange.xcnt = ondisk.builder.catchange.xcnt; |
| 1900 |
builder->catchange.xip = ondisk.builder.catchange.xip; |
| 1901 |
ondisk.builder.catchange.xip = NULL; |
| 1902 |
|
| 1903 |
/* our snapshot is not interesting anymore, build a new one */ |
| 1904 |
if (builder->snapshot != NULL) |
| 1905 |
{ |
| 1906 |
SnapBuildSnapDecRefcount(builder->snapshot); |
| 1907 |
} |
| 1908 |
builder->snapshot = SnapBuildBuildSnapshot(builder); |
| 1909 |
SnapBuildSnapIncRefcount(builder->snapshot); |
| 1910 |
|
| 1911 |
ReorderBufferSetRestartPoint(builder->reorder, lsn); |
| 1912 |
|
| 1913 |
Assert(builder->state == SNAPBUILD_CONSISTENT); |
| 1914 |
|
| 1915 |
ereport(LOG, |
| 1916 |
(errmsg("logical decoding found consistent point at %X/%X", |
| 1917 |
LSN_FORMAT_ARGS(lsn)), |
| 1918 |
errdetail("Logical decoding will begin using saved snapshot."))); |
| 1919 |
return true; |
| 1920 |
|
| 1921 |
snapshot_not_interesting: |
| 1922 |
if (ondisk.builder.committed.xip != NULL) |
| 1923 |
pfree(ondisk.builder.committed.xip); |
| 1924 |
if (ondisk.builder.catchange.xip != NULL) |
| 1925 |
pfree(ondisk.builder.catchange.xip); |
| 1926 |
return false; |
| 1927 |
} |
| 1928 |
|
| 1929 |
/* |
| 1930 |
* Read the contents of the serialized snapshot to 'dest'. |
| 1931 |
*/ |
| 1932 |
static void |
| 1933 |
SnapBuildRestoreContents(int fd, void *dest, Size size, const char *path) |
| 1934 |
{ |
| 1935 |
int readBytes; |
| 1936 |
|
| 1937 |
pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_READ); |
| 1938 |
readBytes = read(fd, dest, size); |
| 1939 |
pgstat_report_wait_end(); |
| 1940 |
if (readBytes != size) |
| 1941 |
{ |
| 1942 |
int save_errno = errno; |
| 1943 |
|
| 1944 |
CloseTransientFile(fd); |
| 1945 |
|
| 1946 |
if (readBytes < 0) |
| 1947 |
{ |
| 1948 |
errno = save_errno; |
| 1949 |
ereport(ERROR, |
| 1950 |
(errcode_for_file_access(), |
| 1951 |
errmsg("could not read file \"%s\": %m", path))); |
| 1952 |
} |
| 1953 |
else |
| 1954 |
ereport(ERROR, |
| 1955 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1956 |
errmsg("could not read file \"%s\": read %d of %zu", |
| 1957 |
path, readBytes, size))); |
| 1958 |
} |
| 1959 |
} |
| 1960 |
|
| 1961 |
/* |
| 1962 |
* Remove all serialized snapshots that are not required anymore because no |
| 1963 |
* slot can need them. This doesn't actually have to run during a checkpoint, |
| 1964 |
* but it's a convenient point to schedule this. |
| 1965 |
* |
| 1966 |
* NB: We run this during checkpoints even if logical decoding is disabled so |
| 1967 |
* we cleanup old slots at some point after it got disabled. |
| 1968 |
*/ |
| 1969 |
void |
| 1970 |
CheckPointSnapBuild(void) |
| 1971 |
{ |
| 1972 |
XLogRecPtr cutoff; |
| 1973 |
XLogRecPtr redo; |
| 1974 |
DIR *snap_dir; |
| 1975 |
struct dirent *snap_de; |
| 1976 |
char path[MAXPGPATH + sizeof(PG_LOGICAL_SNAPSHOTS_DIR)]; |
| 1977 |
|
| 1978 |
/* |
| 1979 |
* We start off with a minimum of the last redo pointer. No new |
| 1980 |
* replication slot will start before that, so that's a safe upper bound |
| 1981 |
* for removal. |
| 1982 |
*/ |
| 1983 |
redo = GetRedoRecPtr(); |
| 1984 |
|
| 1985 |
/* now check for the restart ptrs from existing slots */ |
| 1986 |
cutoff = ReplicationSlotsComputeLogicalRestartLSN(); |
| 1987 |
|
| 1988 |
/* don't start earlier than the restart lsn */ |
| 1989 |
if (redo < cutoff) |
| 1990 |
cutoff = redo; |
| 1991 |
|
| 1992 |
snap_dir = AllocateDir(PG_LOGICAL_SNAPSHOTS_DIR); |
| 1993 |
while ((snap_de = ReadDir(snap_dir, PG_LOGICAL_SNAPSHOTS_DIR)) != NULL) |
| 1994 |
{ |
| 1995 |
uint32 hi; |
| 1996 |
uint32 lo; |
| 1997 |
XLogRecPtr lsn; |
| 1998 |
PGFileType de_type; |
| 1999 |
|
| 2000 |
if (strcmp(snap_de->d_name, ".") == 0 || |
| 2001 |
strcmp(snap_de->d_name, "..") == 0) |
| 2002 |
continue; |
| 2003 |
|
| 2004 |
snprintf(path, sizeof(path), "%s/%s", PG_LOGICAL_SNAPSHOTS_DIR, snap_de->d_name); |
| 2005 |
de_type = get_dirent_type(path, snap_de, false, DEBUG1); |
| 2006 |
|
| 2007 |
if (de_type != PGFILETYPE_ERROR && de_type != PGFILETYPE_REG) |
| 2008 |
{ |
| 2009 |
elog(DEBUG1, "only regular files expected: %s", path); |
| 2010 |
continue; |
| 2011 |
} |
| 2012 |
|
| 2013 |
/* |
| 2014 |
* temporary filenames from SnapBuildSerialize() include the LSN and |
| 2015 |
* everything but are postfixed by .$pid.tmp. We can just remove them |
| 2016 |
* the same as other files because there can be none that are |
| 2017 |
* currently being written that are older than cutoff. |
| 2018 |
* |
| 2019 |
* We just log a message if a file doesn't fit the pattern, it's |
| 2020 |
* probably some editors lock/state file or similar... |
| 2021 |
*/ |
| 2022 |
if (sscanf(snap_de->d_name, "%X-%X.snap", &hi, &lo) != 2) |
| 2023 |
{ |
| 2024 |
ereport(LOG, |
| 2025 |
(errmsg("could not parse file name \"%s\"", path))); |
| 2026 |
continue; |
| 2027 |
} |
| 2028 |
|
| 2029 |
lsn = ((uint64) hi) << 32 | lo; |
| 2030 |
|
| 2031 |
/* check whether we still need it */ |
| 2032 |
if (lsn < cutoff || cutoff == InvalidXLogRecPtr) |
| 2033 |
{ |
| 2034 |
elog(DEBUG1, "removing snapbuild snapshot %s", path); |
| 2035 |
|
| 2036 |
/* |
| 2037 |
* It's not particularly harmful, though strange, if we can't |
| 2038 |
* remove the file here. Don't prevent the checkpoint from |
| 2039 |
* completing, that'd be a cure worse than the disease. |
| 2040 |
*/ |
| 2041 |
if (unlink(path) < 0) |
| 2042 |
{ |
| 2043 |
ereport(LOG, |
| 2044 |
(errcode_for_file_access(), |
| 2045 |
errmsg("could not remove file \"%s\": %m", |
| 2046 |
path))); |
| 2047 |
continue; |
| 2048 |
} |
| 2049 |
} |
| 2050 |
} |
| 2051 |
FreeDir(snap_dir); |
| 2052 |
} |
| 2053 |
|
| 2054 |
/* |
| 2055 |
* Check if a logical snapshot at the specified point has been serialized. |
| 2056 |
*/ |
| 2057 |
bool |
| 2058 |
SnapBuildSnapshotExists(XLogRecPtr lsn) |
| 2059 |
{ |
| 2060 |
char path[MAXPGPATH]; |
| 2061 |
int ret; |
| 2062 |
struct stat stat_buf; |
| 2063 |
|
| 2064 |
sprintf(path, "%s/%X-%X.snap", |
| 2065 |
PG_LOGICAL_SNAPSHOTS_DIR, |
| 2066 |
LSN_FORMAT_ARGS(lsn)); |
| 2067 |
|
| 2068 |
ret = stat(path, &stat_buf); |
| 2069 |
|
| 2070 |
if (ret != 0 && errno != ENOENT) |
| 2071 |
ereport(ERROR, |
| 2072 |
(errcode_for_file_access(), |
| 2073 |
errmsg("could not stat file \"%s\": %m", path))); |
| 2074 |
|
| 2075 |
return ret == 0; |
| 2076 |
} |