Line data Source code
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) 2012-2024, 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 "storage/fd.h"
138 : #include "storage/lmgr.h"
139 : #include "storage/proc.h"
140 : #include "storage/procarray.h"
141 : #include "storage/standby.h"
142 : #include "utils/builtins.h"
143 : #include "utils/memutils.h"
144 : #include "utils/snapmgr.h"
145 : #include "utils/snapshot.h"
146 :
147 : /*
148 : * This struct contains the current state of the snapshot building
149 : * machinery. Besides a forward declaration in the header, it is not exposed
150 : * to the public, so we can easily change its contents.
151 : */
152 : struct SnapBuild
153 : {
154 : /* how far are we along building our first full snapshot */
155 : SnapBuildState state;
156 :
157 : /* private memory context used to allocate memory for this module. */
158 : MemoryContext context;
159 :
160 : /* all transactions < than this have committed/aborted */
161 : TransactionId xmin;
162 :
163 : /* all transactions >= than this are uncommitted */
164 : TransactionId xmax;
165 :
166 : /*
167 : * Don't replay commits from an LSN < this LSN. This can be set externally
168 : * but it will also be advanced (never retreat) from within snapbuild.c.
169 : */
170 : XLogRecPtr start_decoding_at;
171 :
172 : /*
173 : * LSN at which two-phase decoding was enabled or LSN at which we found a
174 : * consistent point at the time of slot creation.
175 : *
176 : * The prepared transactions, that were skipped because previously
177 : * two-phase was not enabled or are not covered by initial snapshot, need
178 : * to be sent later along with commit prepared and they must be before
179 : * this point.
180 : */
181 : XLogRecPtr two_phase_at;
182 :
183 : /*
184 : * Don't start decoding WAL until the "xl_running_xacts" information
185 : * indicates there are no running xids with an xid smaller than this.
186 : */
187 : TransactionId initial_xmin_horizon;
188 :
189 : /* Indicates if we are building full snapshot or just catalog one. */
190 : bool building_full_snapshot;
191 :
192 : /*
193 : * Snapshot that's valid to see the catalog state seen at this moment.
194 : */
195 : Snapshot snapshot;
196 :
197 : /*
198 : * LSN of the last location we are sure a snapshot has been serialized to.
199 : */
200 : XLogRecPtr last_serialized_snapshot;
201 :
202 : /*
203 : * The reorderbuffer we need to update with usable snapshots et al.
204 : */
205 : ReorderBuffer *reorder;
206 :
207 : /*
208 : * TransactionId at which the next phase of initial snapshot building will
209 : * happen. InvalidTransactionId if not known (i.e. SNAPBUILD_START), or
210 : * when no next phase necessary (SNAPBUILD_CONSISTENT).
211 : */
212 : TransactionId next_phase_at;
213 :
214 : /*
215 : * Array of transactions which could have catalog changes that committed
216 : * between xmin and xmax.
217 : */
218 : struct
219 : {
220 : /* number of committed transactions */
221 : size_t xcnt;
222 :
223 : /* available space for committed transactions */
224 : size_t xcnt_space;
225 :
226 : /*
227 : * Until we reach a CONSISTENT state, we record commits of all
228 : * transactions, not just the catalog changing ones. Record when that
229 : * changes so we know we cannot export a snapshot safely anymore.
230 : */
231 : bool includes_all_transactions;
232 :
233 : /*
234 : * Array of committed transactions that have modified the catalog.
235 : *
236 : * As this array is frequently modified we do *not* keep it in
237 : * xidComparator order. Instead we sort the array when building &
238 : * distributing a snapshot.
239 : *
240 : * TODO: It's unclear whether that reasoning has much merit. Every
241 : * time we add something here after becoming consistent will also
242 : * require distributing a snapshot. Storing them sorted would
243 : * potentially also make it easier to purge (but more complicated wrt
244 : * wraparound?). Should be improved if sorting while building the
245 : * snapshot shows up in profiles.
246 : */
247 : TransactionId *xip;
248 : } committed;
249 :
250 : /*
251 : * Array of transactions and subtransactions that had modified catalogs
252 : * and were running when the snapshot was serialized.
253 : *
254 : * We normally rely on some WAL record types such as HEAP2_NEW_CID to know
255 : * if the transaction has changed the catalog. But it could happen that
256 : * the logical decoding decodes only the commit record of the transaction
257 : * after restoring the previously serialized snapshot in which case we
258 : * will miss adding the xid to the snapshot and end up looking at the
259 : * catalogs with the wrong snapshot.
260 : *
261 : * Now to avoid the above problem, we serialize the transactions that had
262 : * modified the catalogs and are still running at the time of snapshot
263 : * serialization. We fill this array while restoring the snapshot and then
264 : * refer it while decoding commit to ensure if the xact has modified the
265 : * catalog. We discard this array when all the xids in the list become old
266 : * enough to matter. See SnapBuildPurgeOlderTxn for details.
267 : */
268 : struct
269 : {
270 : /* number of transactions */
271 : size_t xcnt;
272 :
273 : /* This array must be sorted in xidComparator order */
274 : TransactionId *xip;
275 : } catchange;
276 : };
277 :
278 : /*
279 : * Starting a transaction -- which we need to do while exporting a snapshot --
280 : * removes knowledge about the previously used resowner, so we save it here.
281 : */
282 : static ResourceOwner SavedResourceOwnerDuringExport = NULL;
283 : static bool ExportInProgress = false;
284 :
285 : /* ->committed and ->catchange manipulation */
286 : static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
287 :
288 : /* snapshot building/manipulation/distribution functions */
289 : static Snapshot SnapBuildBuildSnapshot(SnapBuild *builder);
290 :
291 : static void SnapBuildFreeSnapshot(Snapshot snap);
292 :
293 : static void SnapBuildSnapIncRefcount(Snapshot snap);
294 :
295 : static void SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn);
296 :
297 : static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
298 : uint32 xinfo);
299 :
300 : /* xlog reading helper functions for SnapBuildProcessRunningXacts */
301 : static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
302 : static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
303 :
304 : /* serialization functions */
305 : static void SnapBuildSerialize(SnapBuild *builder, XLogRecPtr lsn);
306 : static bool SnapBuildRestore(SnapBuild *builder, XLogRecPtr lsn);
307 : static void SnapBuildRestoreContents(int fd, char *dest, Size size, const char *path);
308 :
309 : /*
310 : * Allocate a new snapshot builder.
311 : *
312 : * xmin_horizon is the xid >= which we can be sure no catalog rows have been
313 : * removed, start_lsn is the LSN >= we want to replay commits.
314 : */
315 : SnapBuild *
316 1856 : AllocateSnapshotBuilder(ReorderBuffer *reorder,
317 : TransactionId xmin_horizon,
318 : XLogRecPtr start_lsn,
319 : bool need_full_snapshot,
320 : XLogRecPtr two_phase_at)
321 : {
322 : MemoryContext context;
323 : MemoryContext oldcontext;
324 : SnapBuild *builder;
325 :
326 : /* allocate memory in own context, to have better accountability */
327 1856 : context = AllocSetContextCreate(CurrentMemoryContext,
328 : "snapshot builder context",
329 : ALLOCSET_DEFAULT_SIZES);
330 1856 : oldcontext = MemoryContextSwitchTo(context);
331 :
332 1856 : builder = palloc0(sizeof(SnapBuild));
333 :
334 1856 : builder->state = SNAPBUILD_START;
335 1856 : builder->context = context;
336 1856 : builder->reorder = reorder;
337 : /* Other struct members initialized by zeroing via palloc0 above */
338 :
339 1856 : builder->committed.xcnt = 0;
340 1856 : builder->committed.xcnt_space = 128; /* arbitrary number */
341 1856 : builder->committed.xip =
342 1856 : palloc0(builder->committed.xcnt_space * sizeof(TransactionId));
343 1856 : builder->committed.includes_all_transactions = true;
344 :
345 1856 : builder->catchange.xcnt = 0;
346 1856 : builder->catchange.xip = NULL;
347 :
348 1856 : builder->initial_xmin_horizon = xmin_horizon;
349 1856 : builder->start_decoding_at = start_lsn;
350 1856 : builder->building_full_snapshot = need_full_snapshot;
351 1856 : builder->two_phase_at = two_phase_at;
352 :
353 1856 : MemoryContextSwitchTo(oldcontext);
354 :
355 1856 : return builder;
356 : }
357 :
358 : /*
359 : * Free a snapshot builder.
360 : */
361 : void
362 1532 : FreeSnapshotBuilder(SnapBuild *builder)
363 : {
364 1532 : MemoryContext context = builder->context;
365 :
366 : /* free snapshot explicitly, that contains some error checking */
367 1532 : if (builder->snapshot != NULL)
368 : {
369 380 : SnapBuildSnapDecRefcount(builder->snapshot);
370 380 : builder->snapshot = NULL;
371 : }
372 :
373 : /* other resources are deallocated via memory context reset */
374 1532 : MemoryContextDelete(context);
375 1532 : }
376 :
377 : /*
378 : * Free an unreferenced snapshot that has previously been built by us.
379 : */
380 : static void
381 2238 : SnapBuildFreeSnapshot(Snapshot snap)
382 : {
383 : /* make sure we don't get passed an external snapshot */
384 : Assert(snap->snapshot_type == SNAPSHOT_HISTORIC_MVCC);
385 :
386 : /* make sure nobody modified our snapshot */
387 : Assert(snap->curcid == FirstCommandId);
388 : Assert(!snap->suboverflowed);
389 : Assert(!snap->takenDuringRecovery);
390 : Assert(snap->regd_count == 0);
391 :
392 : /* slightly more likely, so it's checked even without c-asserts */
393 2238 : if (snap->copied)
394 0 : elog(ERROR, "cannot free a copied snapshot");
395 :
396 2238 : if (snap->active_count)
397 0 : elog(ERROR, "cannot free an active snapshot");
398 :
399 2238 : pfree(snap);
400 2238 : }
401 :
402 : /*
403 : * In which state of snapshot building are we?
404 : */
405 : SnapBuildState
406 4643166 : SnapBuildCurrentState(SnapBuild *builder)
407 : {
408 4643166 : return builder->state;
409 : }
410 :
411 : /*
412 : * Return the LSN at which the two-phase decoding was first enabled.
413 : */
414 : XLogRecPtr
415 58 : SnapBuildGetTwoPhaseAt(SnapBuild *builder)
416 : {
417 58 : return builder->two_phase_at;
418 : }
419 :
420 : /*
421 : * Set the LSN at which two-phase decoding is enabled.
422 : */
423 : void
424 12 : SnapBuildSetTwoPhaseAt(SnapBuild *builder, XLogRecPtr ptr)
425 : {
426 12 : builder->two_phase_at = ptr;
427 12 : }
428 :
429 : /*
430 : * Should the contents of transaction ending at 'ptr' be decoded?
431 : */
432 : bool
433 982996 : SnapBuildXactNeedsSkip(SnapBuild *builder, XLogRecPtr ptr)
434 : {
435 982996 : return ptr < builder->start_decoding_at;
436 : }
437 :
438 : /*
439 : * Increase refcount of a snapshot.
440 : *
441 : * This is used when handing out a snapshot to some external resource or when
442 : * adding a Snapshot as builder->snapshot.
443 : */
444 : static void
445 9642 : SnapBuildSnapIncRefcount(Snapshot snap)
446 : {
447 9642 : snap->active_count++;
448 9642 : }
449 :
450 : /*
451 : * Decrease refcount of a snapshot and free if the refcount reaches zero.
452 : *
453 : * Externally visible, so that external resources that have been handed an
454 : * IncRef'ed Snapshot can adjust its refcount easily.
455 : */
456 : void
457 9296 : SnapBuildSnapDecRefcount(Snapshot snap)
458 : {
459 : /* make sure we don't get passed an external snapshot */
460 : Assert(snap->snapshot_type == SNAPSHOT_HISTORIC_MVCC);
461 :
462 : /* make sure nobody modified our snapshot */
463 : Assert(snap->curcid == FirstCommandId);
464 : Assert(!snap->suboverflowed);
465 : Assert(!snap->takenDuringRecovery);
466 :
467 : Assert(snap->regd_count == 0);
468 :
469 : Assert(snap->active_count > 0);
470 :
471 : /* slightly more likely, so it's checked even without casserts */
472 9296 : if (snap->copied)
473 0 : elog(ERROR, "cannot free a copied snapshot");
474 :
475 9296 : snap->active_count--;
476 9296 : if (snap->active_count == 0)
477 2238 : SnapBuildFreeSnapshot(snap);
478 9296 : }
479 :
480 : /*
481 : * Build a new snapshot, based on currently committed catalog-modifying
482 : * transactions.
483 : *
484 : * In-progress transactions with catalog access are *not* allowed to modify
485 : * these snapshots; they have to copy them and fill in appropriate ->curcid
486 : * and ->subxip/subxcnt values.
487 : */
488 : static Snapshot
489 2908 : SnapBuildBuildSnapshot(SnapBuild *builder)
490 : {
491 : Snapshot snapshot;
492 : Size ssize;
493 :
494 : Assert(builder->state >= SNAPBUILD_FULL_SNAPSHOT);
495 :
496 2908 : ssize = sizeof(SnapshotData)
497 2908 : + sizeof(TransactionId) * builder->committed.xcnt
498 2908 : + sizeof(TransactionId) * 1 /* toplevel xid */ ;
499 :
500 2908 : snapshot = MemoryContextAllocZero(builder->context, ssize);
501 :
502 2908 : snapshot->snapshot_type = SNAPSHOT_HISTORIC_MVCC;
503 :
504 : /*
505 : * We misuse the original meaning of SnapshotData's xip and subxip fields
506 : * to make the more fitting for our needs.
507 : *
508 : * In the 'xip' array we store transactions that have to be treated as
509 : * committed. Since we will only ever look at tuples from transactions
510 : * that have modified the catalog it's more efficient to store those few
511 : * that exist between xmin and xmax (frequently there are none).
512 : *
513 : * Snapshots that are used in transactions that have modified the catalog
514 : * also use the 'subxip' array to store their toplevel xid and all the
515 : * subtransaction xids so we can recognize when we need to treat rows as
516 : * visible that are not in xip but still need to be visible. Subxip only
517 : * gets filled when the transaction is copied into the context of a
518 : * catalog modifying transaction since we otherwise share a snapshot
519 : * between transactions. As long as a txn hasn't modified the catalog it
520 : * doesn't need to treat any uncommitted rows as visible, so there is no
521 : * need for those xids.
522 : *
523 : * Both arrays are qsort'ed so that we can use bsearch() on them.
524 : */
525 : Assert(TransactionIdIsNormal(builder->xmin));
526 : Assert(TransactionIdIsNormal(builder->xmax));
527 :
528 2908 : snapshot->xmin = builder->xmin;
529 2908 : snapshot->xmax = builder->xmax;
530 :
531 : /* store all transactions to be treated as committed by this snapshot */
532 2908 : snapshot->xip =
533 2908 : (TransactionId *) ((char *) snapshot + sizeof(SnapshotData));
534 2908 : snapshot->xcnt = builder->committed.xcnt;
535 2908 : memcpy(snapshot->xip,
536 2908 : builder->committed.xip,
537 2908 : builder->committed.xcnt * sizeof(TransactionId));
538 :
539 : /* sort so we can bsearch() */
540 2908 : qsort(snapshot->xip, snapshot->xcnt, sizeof(TransactionId), xidComparator);
541 :
542 : /*
543 : * Initially, subxip is empty, i.e. it's a snapshot to be used by
544 : * transactions that don't modify the catalog. Will be filled by
545 : * ReorderBufferCopySnap() if necessary.
546 : */
547 2908 : snapshot->subxcnt = 0;
548 2908 : snapshot->subxip = NULL;
549 :
550 2908 : snapshot->suboverflowed = false;
551 2908 : snapshot->takenDuringRecovery = false;
552 2908 : snapshot->copied = false;
553 2908 : snapshot->curcid = FirstCommandId;
554 2908 : snapshot->active_count = 0;
555 2908 : snapshot->regd_count = 0;
556 2908 : snapshot->snapXactCompletionCount = 0;
557 :
558 2908 : return snapshot;
559 : }
560 :
561 : /*
562 : * Build the initial slot snapshot and convert it to a normal snapshot that
563 : * is understood by HeapTupleSatisfiesMVCC.
564 : *
565 : * The snapshot will be usable directly in current transaction or exported
566 : * for loading in different transaction.
567 : */
568 : Snapshot
569 342 : SnapBuildInitialSnapshot(SnapBuild *builder)
570 : {
571 : Snapshot snap;
572 : TransactionId xid;
573 : TransactionId safeXid;
574 : TransactionId *newxip;
575 342 : int newxcnt = 0;
576 :
577 : Assert(XactIsoLevel == XACT_REPEATABLE_READ);
578 : Assert(builder->building_full_snapshot);
579 :
580 : /* don't allow older snapshots */
581 342 : InvalidateCatalogSnapshot(); /* about to overwrite MyProc->xmin */
582 342 : if (HaveRegisteredOrActiveSnapshot())
583 0 : elog(ERROR, "cannot build an initial slot snapshot when snapshots exist");
584 : Assert(!HistoricSnapshotActive());
585 :
586 342 : if (builder->state != SNAPBUILD_CONSISTENT)
587 0 : elog(ERROR, "cannot build an initial slot snapshot before reaching a consistent state");
588 :
589 342 : if (!builder->committed.includes_all_transactions)
590 0 : elog(ERROR, "cannot build an initial slot snapshot, not all transactions are monitored anymore");
591 :
592 : /* so we don't overwrite the existing value */
593 342 : if (TransactionIdIsValid(MyProc->xmin))
594 0 : elog(ERROR, "cannot build an initial slot snapshot when MyProc->xmin already is valid");
595 :
596 342 : snap = SnapBuildBuildSnapshot(builder);
597 :
598 : /*
599 : * We know that snap->xmin is alive, enforced by the logical xmin
600 : * mechanism. Due to that we can do this without locks, we're only
601 : * changing our own value.
602 : *
603 : * Building an initial snapshot is expensive and an unenforced xmin
604 : * horizon would have bad consequences, therefore always double-check that
605 : * the horizon is enforced.
606 : */
607 342 : LWLockAcquire(ProcArrayLock, LW_SHARED);
608 342 : safeXid = GetOldestSafeDecodingTransactionId(false);
609 342 : LWLockRelease(ProcArrayLock);
610 :
611 342 : if (TransactionIdFollows(safeXid, snap->xmin))
612 0 : elog(ERROR, "cannot build an initial slot snapshot as oldest safe xid %u follows snapshot's xmin %u",
613 : safeXid, snap->xmin);
614 :
615 342 : MyProc->xmin = snap->xmin;
616 :
617 : /* allocate in transaction context */
618 : newxip = (TransactionId *)
619 342 : palloc(sizeof(TransactionId) * GetMaxSnapshotXidCount());
620 :
621 : /*
622 : * snapbuild.c builds transactions in an "inverted" manner, which means it
623 : * stores committed transactions in ->xip, not ones in progress. Build a
624 : * classical snapshot by marking all non-committed transactions as
625 : * in-progress. This can be expensive.
626 : */
627 342 : for (xid = snap->xmin; NormalTransactionIdPrecedes(xid, snap->xmax);)
628 : {
629 : void *test;
630 :
631 : /*
632 : * Check whether transaction committed using the decoding snapshot
633 : * meaning of ->xip.
634 : */
635 0 : test = bsearch(&xid, snap->xip, snap->xcnt,
636 : sizeof(TransactionId), xidComparator);
637 :
638 0 : if (test == NULL)
639 : {
640 0 : if (newxcnt >= GetMaxSnapshotXidCount())
641 0 : ereport(ERROR,
642 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
643 : errmsg("initial slot snapshot too large")));
644 :
645 0 : newxip[newxcnt++] = xid;
646 : }
647 :
648 0 : TransactionIdAdvance(xid);
649 : }
650 :
651 : /* adjust remaining snapshot fields as needed */
652 342 : snap->snapshot_type = SNAPSHOT_MVCC;
653 342 : snap->xcnt = newxcnt;
654 342 : snap->xip = newxip;
655 :
656 342 : return snap;
657 : }
658 :
659 : /*
660 : * Export a snapshot so it can be set in another session with SET TRANSACTION
661 : * SNAPSHOT.
662 : *
663 : * For that we need to start a transaction in the current backend as the
664 : * importing side checks whether the source transaction is still open to make
665 : * sure the xmin horizon hasn't advanced since then.
666 : */
667 : const char *
668 0 : SnapBuildExportSnapshot(SnapBuild *builder)
669 : {
670 : Snapshot snap;
671 : char *snapname;
672 :
673 0 : if (IsTransactionOrTransactionBlock())
674 0 : elog(ERROR, "cannot export a snapshot from within a transaction");
675 :
676 0 : if (SavedResourceOwnerDuringExport)
677 0 : elog(ERROR, "can only export one snapshot at a time");
678 :
679 0 : SavedResourceOwnerDuringExport = CurrentResourceOwner;
680 0 : ExportInProgress = true;
681 :
682 0 : StartTransactionCommand();
683 :
684 : /* There doesn't seem to a nice API to set these */
685 0 : XactIsoLevel = XACT_REPEATABLE_READ;
686 0 : XactReadOnly = true;
687 :
688 0 : snap = SnapBuildInitialSnapshot(builder);
689 :
690 : /*
691 : * now that we've built a plain snapshot, make it active and use the
692 : * normal mechanisms for exporting it
693 : */
694 0 : snapname = ExportSnapshot(snap);
695 :
696 0 : ereport(LOG,
697 : (errmsg_plural("exported logical decoding snapshot: \"%s\" with %u transaction ID",
698 : "exported logical decoding snapshot: \"%s\" with %u transaction IDs",
699 : snap->xcnt,
700 : snapname, snap->xcnt)));
701 0 : return snapname;
702 : }
703 :
704 : /*
705 : * Ensure there is a snapshot and if not build one for current transaction.
706 : */
707 : Snapshot
708 16 : SnapBuildGetOrBuildSnapshot(SnapBuild *builder)
709 : {
710 : Assert(builder->state == SNAPBUILD_CONSISTENT);
711 :
712 : /* only build a new snapshot if we don't have a prebuilt one */
713 16 : if (builder->snapshot == NULL)
714 : {
715 2 : builder->snapshot = SnapBuildBuildSnapshot(builder);
716 : /* increase refcount for the snapshot builder */
717 2 : SnapBuildSnapIncRefcount(builder->snapshot);
718 : }
719 :
720 16 : return builder->snapshot;
721 : }
722 :
723 : /*
724 : * Reset a previously SnapBuildExportSnapshot()'ed snapshot if there is
725 : * any. Aborts the previously started transaction and resets the resource
726 : * owner back to its original value.
727 : */
728 : void
729 8930 : SnapBuildClearExportedSnapshot(void)
730 : {
731 : ResourceOwner tmpResOwner;
732 :
733 : /* nothing exported, that is the usual case */
734 8930 : if (!ExportInProgress)
735 8930 : return;
736 :
737 0 : if (!IsTransactionState())
738 0 : elog(ERROR, "clearing exported snapshot in wrong transaction state");
739 :
740 : /*
741 : * AbortCurrentTransaction() takes care of resetting the snapshot state,
742 : * so remember SavedResourceOwnerDuringExport.
743 : */
744 0 : tmpResOwner = SavedResourceOwnerDuringExport;
745 :
746 : /* make sure nothing could have ever happened */
747 0 : AbortCurrentTransaction();
748 :
749 0 : CurrentResourceOwner = tmpResOwner;
750 : }
751 :
752 : /*
753 : * Clear snapshot export state during transaction abort.
754 : */
755 : void
756 45124 : SnapBuildResetExportedSnapshotState(void)
757 : {
758 45124 : SavedResourceOwnerDuringExport = NULL;
759 45124 : ExportInProgress = false;
760 45124 : }
761 :
762 : /*
763 : * Handle the effects of a single heap change, appropriate to the current state
764 : * of the snapshot builder and returns whether changes made at (xid, lsn) can
765 : * be decoded.
766 : */
767 : bool
768 3216568 : SnapBuildProcessChange(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn)
769 : {
770 : /*
771 : * We can't handle data in transactions if we haven't built a snapshot
772 : * yet, so don't store them.
773 : */
774 3216568 : if (builder->state < SNAPBUILD_FULL_SNAPSHOT)
775 0 : return false;
776 :
777 : /*
778 : * No point in keeping track of changes in transactions that we don't have
779 : * enough information about to decode. This means that they started before
780 : * we got into the SNAPBUILD_FULL_SNAPSHOT state.
781 : */
782 3216574 : if (builder->state < SNAPBUILD_CONSISTENT &&
783 6 : TransactionIdPrecedes(xid, builder->next_phase_at))
784 0 : return false;
785 :
786 : /*
787 : * If the reorderbuffer doesn't yet have a snapshot, add one now, it will
788 : * be needed to decode the change we're currently processing.
789 : */
790 3216568 : if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, xid))
791 : {
792 : /* only build a new snapshot if we don't have a prebuilt one */
793 5096 : if (builder->snapshot == NULL)
794 : {
795 618 : builder->snapshot = SnapBuildBuildSnapshot(builder);
796 : /* increase refcount for the snapshot builder */
797 618 : SnapBuildSnapIncRefcount(builder->snapshot);
798 : }
799 :
800 : /*
801 : * Increase refcount for the transaction we're handing the snapshot
802 : * out to.
803 : */
804 5096 : SnapBuildSnapIncRefcount(builder->snapshot);
805 5096 : ReorderBufferSetBaseSnapshot(builder->reorder, xid, lsn,
806 : builder->snapshot);
807 : }
808 :
809 3216568 : return true;
810 : }
811 :
812 : /*
813 : * Do CommandId/combo CID handling after reading an xl_heap_new_cid record.
814 : * This implies that a transaction has done some form of write to system
815 : * catalogs.
816 : */
817 : void
818 42644 : SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
819 : XLogRecPtr lsn, xl_heap_new_cid *xlrec)
820 : {
821 : CommandId cid;
822 :
823 : /*
824 : * we only log new_cid's if a catalog tuple was modified, so mark the
825 : * transaction as containing catalog modifications
826 : */
827 42644 : ReorderBufferXidSetCatalogChanges(builder->reorder, xid, lsn);
828 :
829 42644 : ReorderBufferAddNewTupleCids(builder->reorder, xlrec->top_xid, lsn,
830 : xlrec->target_locator, xlrec->target_tid,
831 : xlrec->cmin, xlrec->cmax,
832 : xlrec->combocid);
833 :
834 : /* figure out new command id */
835 42644 : if (xlrec->cmin != InvalidCommandId &&
836 36542 : xlrec->cmax != InvalidCommandId)
837 6118 : cid = Max(xlrec->cmin, xlrec->cmax);
838 36526 : else if (xlrec->cmax != InvalidCommandId)
839 6102 : cid = xlrec->cmax;
840 30424 : else if (xlrec->cmin != InvalidCommandId)
841 30424 : cid = xlrec->cmin;
842 : else
843 : {
844 0 : cid = InvalidCommandId; /* silence compiler */
845 0 : elog(ERROR, "xl_heap_new_cid record without a valid CommandId");
846 : }
847 :
848 42644 : ReorderBufferAddNewCommandId(builder->reorder, xid, lsn, cid + 1);
849 42644 : }
850 :
851 : /*
852 : * Add a new Snapshot to all transactions we're decoding that currently are
853 : * in-progress so they can see new catalog contents made by the transaction
854 : * that just committed. This is necessary because those in-progress
855 : * transactions will use the new catalog's contents from here on (at the very
856 : * least everything they do needs to be compatible with newer catalog
857 : * contents).
858 : */
859 : static void
860 1932 : SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn)
861 : {
862 : dlist_iter txn_i;
863 : ReorderBufferTXN *txn;
864 :
865 : /*
866 : * Iterate through all toplevel transactions. This can include
867 : * subtransactions which we just don't yet know to be that, but that's
868 : * fine, they will just get an unnecessary snapshot queued.
869 : */
870 3920 : dlist_foreach(txn_i, &builder->reorder->toplevel_by_lsn)
871 : {
872 1988 : txn = dlist_container(ReorderBufferTXN, node, txn_i.cur);
873 :
874 : Assert(TransactionIdIsValid(txn->xid));
875 :
876 : /*
877 : * If we don't have a base snapshot yet, there are no changes in this
878 : * transaction which in turn implies we don't yet need a snapshot at
879 : * all. We'll add a snapshot when the first change gets queued.
880 : *
881 : * NB: This works correctly even for subtransactions because
882 : * ReorderBufferAssignChild() takes care to transfer the base snapshot
883 : * to the top-level transaction, and while iterating the changequeue
884 : * we'll get the change from the subtxn.
885 : */
886 1988 : if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, txn->xid))
887 4 : continue;
888 :
889 : /*
890 : * We don't need to add snapshot to prepared transactions as they
891 : * should not see the new catalog contents.
892 : */
893 1984 : if (rbtxn_prepared(txn) || rbtxn_skip_prepared(txn))
894 44 : continue;
895 :
896 1940 : elog(DEBUG2, "adding a new snapshot to %u at %X/%X",
897 : txn->xid, LSN_FORMAT_ARGS(lsn));
898 :
899 : /*
900 : * increase the snapshot's refcount for the transaction we are handing
901 : * it out to
902 : */
903 1940 : SnapBuildSnapIncRefcount(builder->snapshot);
904 1940 : ReorderBufferAddSnapshot(builder->reorder, txn->xid, lsn,
905 : builder->snapshot);
906 : }
907 1932 : }
908 :
909 : /*
910 : * Keep track of a new catalog changing transaction that has committed.
911 : */
912 : static void
913 1950 : SnapBuildAddCommittedTxn(SnapBuild *builder, TransactionId xid)
914 : {
915 : Assert(TransactionIdIsValid(xid));
916 :
917 1950 : if (builder->committed.xcnt == builder->committed.xcnt_space)
918 : {
919 0 : builder->committed.xcnt_space = builder->committed.xcnt_space * 2 + 1;
920 :
921 0 : elog(DEBUG1, "increasing space for committed transactions to %u",
922 : (uint32) builder->committed.xcnt_space);
923 :
924 0 : builder->committed.xip = repalloc(builder->committed.xip,
925 0 : builder->committed.xcnt_space * sizeof(TransactionId));
926 : }
927 :
928 : /*
929 : * TODO: It might make sense to keep the array sorted here instead of
930 : * doing it every time we build a new snapshot. On the other hand this
931 : * gets called repeatedly when a transaction with subtransactions commits.
932 : */
933 1950 : builder->committed.xip[builder->committed.xcnt++] = xid;
934 1950 : }
935 :
936 : /*
937 : * Remove knowledge about transactions we treat as committed or containing catalog
938 : * changes that are smaller than ->xmin. Those won't ever get checked via
939 : * the ->committed or ->catchange array, respectively. The committed xids will
940 : * get checked via the clog machinery.
941 : *
942 : * We can ideally remove the transaction from catchange array once it is
943 : * finished (committed/aborted) but that could be costly as we need to maintain
944 : * the xids order in the array.
945 : */
946 : static void
947 618 : SnapBuildPurgeOlderTxn(SnapBuild *builder)
948 : {
949 : int off;
950 : TransactionId *workspace;
951 618 : int surviving_xids = 0;
952 :
953 : /* not ready yet */
954 618 : if (!TransactionIdIsNormal(builder->xmin))
955 0 : return;
956 :
957 : /* TODO: Neater algorithm than just copying and iterating? */
958 : workspace =
959 618 : MemoryContextAlloc(builder->context,
960 618 : builder->committed.xcnt * sizeof(TransactionId));
961 :
962 : /* copy xids that still are interesting to workspace */
963 1002 : for (off = 0; off < builder->committed.xcnt; off++)
964 : {
965 384 : if (NormalTransactionIdPrecedes(builder->committed.xip[off],
966 : builder->xmin))
967 : ; /* remove */
968 : else
969 0 : workspace[surviving_xids++] = builder->committed.xip[off];
970 : }
971 :
972 : /* copy workspace back to persistent state */
973 618 : memcpy(builder->committed.xip, workspace,
974 : surviving_xids * sizeof(TransactionId));
975 :
976 618 : elog(DEBUG3, "purged committed transactions from %u to %u, xmin: %u, xmax: %u",
977 : (uint32) builder->committed.xcnt, (uint32) surviving_xids,
978 : builder->xmin, builder->xmax);
979 618 : builder->committed.xcnt = surviving_xids;
980 :
981 618 : pfree(workspace);
982 :
983 : /*
984 : * Purge xids in ->catchange as well. The purged array must also be sorted
985 : * in xidComparator order.
986 : */
987 618 : if (builder->catchange.xcnt > 0)
988 : {
989 : /*
990 : * Since catchange.xip is sorted, we find the lower bound of xids that
991 : * are still interesting.
992 : */
993 14 : for (off = 0; off < builder->catchange.xcnt; off++)
994 : {
995 10 : if (TransactionIdFollowsOrEquals(builder->catchange.xip[off],
996 : builder->xmin))
997 2 : break;
998 : }
999 :
1000 6 : surviving_xids = builder->catchange.xcnt - off;
1001 :
1002 6 : if (surviving_xids > 0)
1003 : {
1004 2 : memmove(builder->catchange.xip, &(builder->catchange.xip[off]),
1005 : surviving_xids * sizeof(TransactionId));
1006 : }
1007 : else
1008 : {
1009 4 : pfree(builder->catchange.xip);
1010 4 : builder->catchange.xip = NULL;
1011 : }
1012 :
1013 6 : elog(DEBUG3, "purged catalog modifying transactions from %u to %u, xmin: %u, xmax: %u",
1014 : (uint32) builder->catchange.xcnt, (uint32) surviving_xids,
1015 : builder->xmin, builder->xmax);
1016 6 : builder->catchange.xcnt = surviving_xids;
1017 : }
1018 : }
1019 :
1020 : /*
1021 : * Handle everything that needs to be done when a transaction commits
1022 : */
1023 : void
1024 5032 : SnapBuildCommitTxn(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid,
1025 : int nsubxacts, TransactionId *subxacts, uint32 xinfo)
1026 : {
1027 : int nxact;
1028 :
1029 5032 : bool needs_snapshot = false;
1030 5032 : bool needs_timetravel = false;
1031 5032 : bool sub_needs_timetravel = false;
1032 :
1033 5032 : TransactionId xmax = xid;
1034 :
1035 : /*
1036 : * Transactions preceding BUILDING_SNAPSHOT will neither be decoded, nor
1037 : * will they be part of a snapshot. So we don't need to record anything.
1038 : */
1039 5032 : if (builder->state == SNAPBUILD_START ||
1040 5032 : (builder->state == SNAPBUILD_BUILDING_SNAPSHOT &&
1041 0 : TransactionIdPrecedes(xid, builder->next_phase_at)))
1042 : {
1043 : /* ensure that only commits after this are getting replayed */
1044 0 : if (builder->start_decoding_at <= lsn)
1045 0 : builder->start_decoding_at = lsn + 1;
1046 0 : return;
1047 : }
1048 :
1049 5032 : if (builder->state < SNAPBUILD_CONSISTENT)
1050 : {
1051 : /* ensure that only commits after this are getting replayed */
1052 10 : if (builder->start_decoding_at <= lsn)
1053 4 : builder->start_decoding_at = lsn + 1;
1054 :
1055 : /*
1056 : * If building an exportable snapshot, force xid to be tracked, even
1057 : * if the transaction didn't modify the catalog.
1058 : */
1059 10 : if (builder->building_full_snapshot)
1060 : {
1061 0 : needs_timetravel = true;
1062 : }
1063 : }
1064 :
1065 7496 : for (nxact = 0; nxact < nsubxacts; nxact++)
1066 : {
1067 2464 : TransactionId subxid = subxacts[nxact];
1068 :
1069 : /*
1070 : * Add subtransaction to base snapshot if catalog modifying, we don't
1071 : * distinguish to toplevel transactions there.
1072 : */
1073 2464 : if (SnapBuildXidHasCatalogChanges(builder, subxid, xinfo))
1074 : {
1075 18 : sub_needs_timetravel = true;
1076 18 : needs_snapshot = true;
1077 :
1078 18 : elog(DEBUG1, "found subtransaction %u:%u with catalog changes",
1079 : xid, subxid);
1080 :
1081 18 : SnapBuildAddCommittedTxn(builder, subxid);
1082 :
1083 18 : if (NormalTransactionIdFollows(subxid, xmax))
1084 18 : xmax = subxid;
1085 : }
1086 :
1087 : /*
1088 : * If we're forcing timetravel we also need visibility information
1089 : * about subtransaction, so keep track of subtransaction's state, even
1090 : * if not catalog modifying. Don't need to distribute a snapshot in
1091 : * that case.
1092 : */
1093 2446 : else if (needs_timetravel)
1094 : {
1095 0 : SnapBuildAddCommittedTxn(builder, subxid);
1096 0 : if (NormalTransactionIdFollows(subxid, xmax))
1097 0 : xmax = subxid;
1098 : }
1099 : }
1100 :
1101 : /* if top-level modified catalog, it'll need a snapshot */
1102 5032 : if (SnapBuildXidHasCatalogChanges(builder, xid, xinfo))
1103 : {
1104 1930 : elog(DEBUG2, "found top level transaction %u, with catalog changes",
1105 : xid);
1106 1930 : needs_snapshot = true;
1107 1930 : needs_timetravel = true;
1108 1930 : SnapBuildAddCommittedTxn(builder, xid);
1109 : }
1110 3102 : else if (sub_needs_timetravel)
1111 : {
1112 : /* track toplevel txn as well, subxact alone isn't meaningful */
1113 2 : elog(DEBUG2, "forced transaction %u to do timetravel due to one of its subtransactions",
1114 : xid);
1115 2 : needs_timetravel = true;
1116 2 : SnapBuildAddCommittedTxn(builder, xid);
1117 : }
1118 3100 : else if (needs_timetravel)
1119 : {
1120 0 : elog(DEBUG2, "forced transaction %u to do timetravel", xid);
1121 :
1122 0 : SnapBuildAddCommittedTxn(builder, xid);
1123 : }
1124 :
1125 5032 : if (!needs_timetravel)
1126 : {
1127 : /* record that we cannot export a general snapshot anymore */
1128 3100 : builder->committed.includes_all_transactions = false;
1129 : }
1130 :
1131 : Assert(!needs_snapshot || needs_timetravel);
1132 :
1133 : /*
1134 : * Adjust xmax of the snapshot builder, we only do that for committed,
1135 : * catalog modifying, transactions, everything else isn't interesting for
1136 : * us since we'll never look at the respective rows.
1137 : */
1138 5032 : if (needs_timetravel &&
1139 3864 : (!TransactionIdIsValid(builder->xmax) ||
1140 1932 : TransactionIdFollowsOrEquals(xmax, builder->xmax)))
1141 : {
1142 1932 : builder->xmax = xmax;
1143 1932 : TransactionIdAdvance(builder->xmax);
1144 : }
1145 :
1146 : /* if there's any reason to build a historic snapshot, do so now */
1147 5032 : if (needs_snapshot)
1148 : {
1149 : /*
1150 : * If we haven't built a complete snapshot yet there's no need to hand
1151 : * it out, it wouldn't (and couldn't) be used anyway.
1152 : */
1153 1932 : if (builder->state < SNAPBUILD_FULL_SNAPSHOT)
1154 0 : return;
1155 :
1156 : /*
1157 : * Decrease the snapshot builder's refcount of the old snapshot, note
1158 : * that it still will be used if it has been handed out to the
1159 : * reorderbuffer earlier.
1160 : */
1161 1932 : if (builder->snapshot)
1162 1920 : SnapBuildSnapDecRefcount(builder->snapshot);
1163 :
1164 1932 : builder->snapshot = SnapBuildBuildSnapshot(builder);
1165 :
1166 : /* we might need to execute invalidations, add snapshot */
1167 1932 : if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, xid))
1168 : {
1169 40 : SnapBuildSnapIncRefcount(builder->snapshot);
1170 40 : ReorderBufferSetBaseSnapshot(builder->reorder, xid, lsn,
1171 : builder->snapshot);
1172 : }
1173 :
1174 : /* refcount of the snapshot builder for the new snapshot */
1175 1932 : SnapBuildSnapIncRefcount(builder->snapshot);
1176 :
1177 : /* add a new catalog snapshot to all currently running transactions */
1178 1932 : SnapBuildDistributeNewCatalogSnapshot(builder, lsn);
1179 : }
1180 : }
1181 :
1182 : /*
1183 : * Check the reorder buffer and the snapshot to see if the given transaction has
1184 : * modified catalogs.
1185 : */
1186 : static inline bool
1187 7496 : SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
1188 : uint32 xinfo)
1189 : {
1190 7496 : if (ReorderBufferXidHasCatalogChanges(builder->reorder, xid))
1191 1940 : return true;
1192 :
1193 : /*
1194 : * The transactions that have changed catalogs must have invalidation
1195 : * info.
1196 : */
1197 5556 : if (!(xinfo & XACT_XINFO_HAS_INVALS))
1198 5540 : return false;
1199 :
1200 : /* Check the catchange XID array */
1201 24 : return ((builder->catchange.xcnt > 0) &&
1202 8 : (bsearch(&xid, builder->catchange.xip, builder->catchange.xcnt,
1203 : sizeof(TransactionId), xidComparator) != NULL));
1204 : }
1205 :
1206 : /* -----------------------------------
1207 : * Snapshot building functions dealing with xlog records
1208 : * -----------------------------------
1209 : */
1210 :
1211 : /*
1212 : * Process a running xacts record, and use its information to first build a
1213 : * historic snapshot and later to release resources that aren't needed
1214 : * anymore.
1215 : */
1216 : void
1217 2428 : SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
1218 : {
1219 : ReorderBufferTXN *txn;
1220 : TransactionId xmin;
1221 :
1222 : /*
1223 : * If we're not consistent yet, inspect the record to see whether it
1224 : * allows to get closer to being consistent. If we are consistent, dump
1225 : * our snapshot so others or we, after a restart, can use it.
1226 : */
1227 2428 : if (builder->state < SNAPBUILD_CONSISTENT)
1228 : {
1229 : /* returns false if there's no point in performing cleanup just yet */
1230 1854 : if (!SnapBuildFindSnapshot(builder, lsn, running))
1231 1806 : return;
1232 : }
1233 : else
1234 574 : SnapBuildSerialize(builder, lsn);
1235 :
1236 : /*
1237 : * Update range of interesting xids based on the running xacts
1238 : * information. We don't increase ->xmax using it, because once we are in
1239 : * a consistent state we can do that ourselves and much more efficiently
1240 : * so, because we only need to do it for catalog transactions since we
1241 : * only ever look at those.
1242 : *
1243 : * NB: We only increase xmax when a catalog modifying transaction commits
1244 : * (see SnapBuildCommitTxn). Because of this, xmax can be lower than
1245 : * xmin, which looks odd but is correct and actually more efficient, since
1246 : * we hit fast paths in heapam_visibility.c.
1247 : */
1248 618 : builder->xmin = running->oldestRunningXid;
1249 :
1250 : /* Remove transactions we don't need to keep track off anymore */
1251 618 : SnapBuildPurgeOlderTxn(builder);
1252 :
1253 : /*
1254 : * Advance the xmin limit for the current replication slot, to allow
1255 : * vacuum to clean up the tuples this slot has been protecting.
1256 : *
1257 : * The reorderbuffer might have an xmin among the currently running
1258 : * snapshots; use it if so. If not, we need only consider the snapshots
1259 : * we'll produce later, which can't be less than the oldest running xid in
1260 : * the record we're reading now.
1261 : */
1262 618 : xmin = ReorderBufferGetOldestXmin(builder->reorder);
1263 618 : if (xmin == InvalidTransactionId)
1264 550 : xmin = running->oldestRunningXid;
1265 618 : elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u",
1266 : builder->xmin, builder->xmax, running->oldestRunningXid, xmin);
1267 618 : LogicalIncreaseXminForSlot(lsn, xmin);
1268 :
1269 : /*
1270 : * Also tell the slot where we can restart decoding from. We don't want to
1271 : * do that after every commit because changing that implies an fsync of
1272 : * the logical slot's state file, so we only do it every time we see a
1273 : * running xacts record.
1274 : *
1275 : * Do so by looking for the oldest in progress transaction (determined by
1276 : * the first LSN of any of its relevant records). Every transaction
1277 : * remembers the last location we stored the snapshot to disk before its
1278 : * beginning. That point is where we can restart from.
1279 : */
1280 :
1281 : /*
1282 : * Can't know about a serialized snapshot's location if we're not
1283 : * consistent.
1284 : */
1285 618 : if (builder->state < SNAPBUILD_CONSISTENT)
1286 34 : return;
1287 :
1288 584 : txn = ReorderBufferGetOldestTXN(builder->reorder);
1289 :
1290 : /*
1291 : * oldest ongoing txn might have started when we didn't yet serialize
1292 : * anything because we hadn't reached a consistent state yet.
1293 : */
1294 584 : if (txn != NULL && txn->restart_decoding_lsn != InvalidXLogRecPtr)
1295 46 : LogicalIncreaseRestartDecodingForSlot(lsn, txn->restart_decoding_lsn);
1296 :
1297 : /*
1298 : * No in-progress transaction, can reuse the last serialized snapshot if
1299 : * we have one.
1300 : */
1301 538 : else if (txn == NULL &&
1302 496 : builder->reorder->current_restart_decoding_lsn != InvalidXLogRecPtr &&
1303 492 : builder->last_serialized_snapshot != InvalidXLogRecPtr)
1304 492 : LogicalIncreaseRestartDecodingForSlot(lsn,
1305 : builder->last_serialized_snapshot);
1306 : }
1307 :
1308 :
1309 : /*
1310 : * Build the start of a snapshot that's capable of decoding the catalog.
1311 : *
1312 : * Helper function for SnapBuildProcessRunningXacts() while we're not yet
1313 : * consistent.
1314 : *
1315 : * Returns true if there is a point in performing internal maintenance/cleanup
1316 : * using the xl_running_xacts record.
1317 : */
1318 : static bool
1319 1854 : SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
1320 : {
1321 : /* ---
1322 : * Build catalog decoding snapshot incrementally using information about
1323 : * the currently running transactions. There are several ways to do that:
1324 : *
1325 : * a) There were no running transactions when the xl_running_xacts record
1326 : * was inserted, jump to CONSISTENT immediately. We might find such a
1327 : * state while waiting on c)'s sub-states.
1328 : *
1329 : * b) This (in a previous run) or another decoding slot serialized a
1330 : * snapshot to disk that we can use. Can't use this method for the
1331 : * initial snapshot when slot is being created and needs full snapshot
1332 : * for export or direct use, as that snapshot will only contain catalog
1333 : * modifying transactions.
1334 : *
1335 : * c) First incrementally build a snapshot for catalog tuples
1336 : * (BUILDING_SNAPSHOT), that requires all, already in-progress,
1337 : * transactions to finish. Every transaction starting after that
1338 : * (FULL_SNAPSHOT state), has enough information to be decoded. But
1339 : * for older running transactions no viable snapshot exists yet, so
1340 : * CONSISTENT will only be reached once all of those have finished.
1341 : * ---
1342 : */
1343 :
1344 : /*
1345 : * xl_running_xacts record is older than what we can use, we might not
1346 : * have all necessary catalog rows anymore.
1347 : */
1348 1854 : if (TransactionIdIsNormal(builder->initial_xmin_horizon) &&
1349 798 : NormalTransactionIdPrecedes(running->oldestRunningXid,
1350 : builder->initial_xmin_horizon))
1351 : {
1352 0 : ereport(DEBUG1,
1353 : (errmsg_internal("skipping snapshot at %X/%X while building logical decoding snapshot, xmin horizon too low",
1354 : LSN_FORMAT_ARGS(lsn)),
1355 : errdetail_internal("initial xmin horizon of %u vs the snapshot's %u",
1356 : builder->initial_xmin_horizon, running->oldestRunningXid)));
1357 :
1358 :
1359 0 : SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon);
1360 :
1361 0 : return true;
1362 : }
1363 :
1364 : /*
1365 : * a) No transaction were running, we can jump to consistent.
1366 : *
1367 : * This is not affected by races around xl_running_xacts, because we can
1368 : * miss transaction commits, but currently not transactions starting.
1369 : *
1370 : * NB: We might have already started to incrementally assemble a snapshot,
1371 : * so we need to be careful to deal with that.
1372 : */
1373 1854 : if (running->oldestRunningXid == running->nextXid)
1374 : {
1375 1792 : if (builder->start_decoding_at == InvalidXLogRecPtr ||
1376 1010 : builder->start_decoding_at <= lsn)
1377 : /* can decode everything after this */
1378 784 : builder->start_decoding_at = lsn + 1;
1379 :
1380 : /* As no transactions were running xmin/xmax can be trivially set. */
1381 1792 : builder->xmin = running->nextXid; /* < are finished */
1382 1792 : builder->xmax = running->nextXid; /* >= are running */
1383 :
1384 : /* so we can safely use the faster comparisons */
1385 : Assert(TransactionIdIsNormal(builder->xmin));
1386 : Assert(TransactionIdIsNormal(builder->xmax));
1387 :
1388 1792 : builder->state = SNAPBUILD_CONSISTENT;
1389 1792 : builder->next_phase_at = InvalidTransactionId;
1390 :
1391 1792 : ereport(LOG,
1392 : (errmsg("logical decoding found consistent point at %X/%X",
1393 : LSN_FORMAT_ARGS(lsn)),
1394 : errdetail("There are no running transactions.")));
1395 :
1396 1792 : return false;
1397 : }
1398 : /* b) valid on disk state and not building full snapshot */
1399 122 : else if (!builder->building_full_snapshot &&
1400 60 : SnapBuildRestore(builder, lsn))
1401 : {
1402 : /* there won't be any state to cleanup */
1403 14 : return false;
1404 : }
1405 :
1406 : /*
1407 : * c) transition from START to BUILDING_SNAPSHOT.
1408 : *
1409 : * In START state, and a xl_running_xacts record with running xacts is
1410 : * encountered. In that case, switch to BUILDING_SNAPSHOT state, and
1411 : * record xl_running_xacts->nextXid. Once all running xacts have finished
1412 : * (i.e. they're all >= nextXid), we have a complete catalog snapshot. It
1413 : * might look that we could use xl_running_xacts's ->xids information to
1414 : * get there quicker, but that is problematic because transactions marked
1415 : * as running, might already have inserted their commit record - it's
1416 : * infeasible to change that with locking.
1417 : */
1418 48 : else if (builder->state == SNAPBUILD_START)
1419 : {
1420 28 : builder->state = SNAPBUILD_BUILDING_SNAPSHOT;
1421 28 : builder->next_phase_at = running->nextXid;
1422 :
1423 : /*
1424 : * Start with an xmin/xmax that's correct for future, when all the
1425 : * currently running transactions have finished. We'll update both
1426 : * while waiting for the pending transactions to finish.
1427 : */
1428 28 : builder->xmin = running->nextXid; /* < are finished */
1429 28 : builder->xmax = running->nextXid; /* >= are running */
1430 :
1431 : /* so we can safely use the faster comparisons */
1432 : Assert(TransactionIdIsNormal(builder->xmin));
1433 : Assert(TransactionIdIsNormal(builder->xmax));
1434 :
1435 28 : ereport(LOG,
1436 : (errmsg("logical decoding found initial starting point at %X/%X",
1437 : LSN_FORMAT_ARGS(lsn)),
1438 : errdetail("Waiting for transactions (approximately %d) older than %u to end.",
1439 : running->xcnt, running->nextXid)));
1440 :
1441 28 : SnapBuildWaitSnapshot(running, running->nextXid);
1442 : }
1443 :
1444 : /*
1445 : * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT.
1446 : *
1447 : * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid
1448 : * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This
1449 : * means all transactions starting afterwards have enough information to
1450 : * be decoded. Switch to FULL_SNAPSHOT.
1451 : */
1452 30 : else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT &&
1453 10 : TransactionIdPrecedesOrEquals(builder->next_phase_at,
1454 : running->oldestRunningXid))
1455 : {
1456 10 : builder->state = SNAPBUILD_FULL_SNAPSHOT;
1457 10 : builder->next_phase_at = running->nextXid;
1458 :
1459 10 : ereport(LOG,
1460 : (errmsg("logical decoding found initial consistent point at %X/%X",
1461 : LSN_FORMAT_ARGS(lsn)),
1462 : errdetail("Waiting for transactions (approximately %d) older than %u to end.",
1463 : running->xcnt, running->nextXid)));
1464 :
1465 10 : SnapBuildWaitSnapshot(running, running->nextXid);
1466 : }
1467 :
1468 : /*
1469 : * c) transition from FULL_SNAPSHOT to CONSISTENT.
1470 : *
1471 : * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is
1472 : * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all
1473 : * transactions that are currently in progress have a catalog snapshot,
1474 : * and all their changes have been collected. Switch to CONSISTENT.
1475 : */
1476 20 : else if (builder->state == SNAPBUILD_FULL_SNAPSHOT &&
1477 10 : TransactionIdPrecedesOrEquals(builder->next_phase_at,
1478 : running->oldestRunningXid))
1479 : {
1480 10 : builder->state = SNAPBUILD_CONSISTENT;
1481 10 : builder->next_phase_at = InvalidTransactionId;
1482 :
1483 10 : ereport(LOG,
1484 : (errmsg("logical decoding found consistent point at %X/%X",
1485 : LSN_FORMAT_ARGS(lsn)),
1486 : errdetail("There are no old transactions anymore.")));
1487 : }
1488 :
1489 : /*
1490 : * We already started to track running xacts and need to wait for all
1491 : * in-progress ones to finish. We fall through to the normal processing of
1492 : * records so incremental cleanup can be performed.
1493 : */
1494 44 : return true;
1495 : }
1496 :
1497 : /* ---
1498 : * Iterate through xids in record, wait for all older than the cutoff to
1499 : * finish. Then, if possible, log a new xl_running_xacts record.
1500 : *
1501 : * This isn't required for the correctness of decoding, but to:
1502 : * a) allow isolationtester to notice that we're currently waiting for
1503 : * something.
1504 : * b) log a new xl_running_xacts record where it'd be helpful, without having
1505 : * to wait for bgwriter or checkpointer.
1506 : * ---
1507 : */
1508 : static void
1509 38 : SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
1510 : {
1511 : int off;
1512 :
1513 72 : for (off = 0; off < running->xcnt; off++)
1514 : {
1515 38 : TransactionId xid = running->xids[off];
1516 :
1517 : /*
1518 : * Upper layers should prevent that we ever need to wait on ourselves.
1519 : * Check anyway, since failing to do so would either result in an
1520 : * endless wait or an Assert() failure.
1521 : */
1522 38 : if (TransactionIdIsCurrentTransactionId(xid))
1523 0 : elog(ERROR, "waiting for ourselves");
1524 :
1525 38 : if (TransactionIdFollows(xid, cutoff))
1526 0 : continue;
1527 :
1528 38 : XactLockTableWait(xid, NULL, NULL, XLTW_None);
1529 : }
1530 :
1531 : /*
1532 : * All transactions we needed to finish finished - try to ensure there is
1533 : * another xl_running_xacts record in a timely manner, without having to
1534 : * wait for bgwriter or checkpointer to log one. During recovery we can't
1535 : * enforce that, so we'll have to wait.
1536 : */
1537 34 : if (!RecoveryInProgress())
1538 : {
1539 34 : LogStandbySnapshot();
1540 : }
1541 34 : }
1542 :
1543 : /* -----------------------------------
1544 : * Snapshot serialization support
1545 : * -----------------------------------
1546 : */
1547 :
1548 : /*
1549 : * We store current state of struct SnapBuild on disk in the following manner:
1550 : *
1551 : * struct SnapBuildOnDisk;
1552 : * TransactionId * committed.xcnt; (*not xcnt_space*)
1553 : * TransactionId * catchange.xcnt;
1554 : *
1555 : */
1556 : typedef struct SnapBuildOnDisk
1557 : {
1558 : /* first part of this struct needs to be version independent */
1559 :
1560 : /* data not covered by checksum */
1561 : uint32 magic;
1562 : pg_crc32c checksum;
1563 :
1564 : /* data covered by checksum */
1565 :
1566 : /* version, in case we want to support pg_upgrade */
1567 : uint32 version;
1568 : /* how large is the on disk data, excluding the constant sized part */
1569 : uint32 length;
1570 :
1571 : /* version dependent part */
1572 : SnapBuild builder;
1573 :
1574 : /* variable amount of TransactionIds follows */
1575 : } SnapBuildOnDisk;
1576 :
1577 : #define SnapBuildOnDiskConstantSize \
1578 : offsetof(SnapBuildOnDisk, builder)
1579 : #define SnapBuildOnDiskNotChecksummedSize \
1580 : offsetof(SnapBuildOnDisk, version)
1581 :
1582 : #define SNAPBUILD_MAGIC 0x51A1E001
1583 : #define SNAPBUILD_VERSION 5
1584 :
1585 : /*
1586 : * Store/Load a snapshot from disk, depending on the snapshot builder's state.
1587 : *
1588 : * Supposed to be used by external (i.e. not snapbuild.c) code that just read
1589 : * a record that's a potential location for a serialized snapshot.
1590 : */
1591 : void
1592 74 : SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn)
1593 : {
1594 74 : if (builder->state < SNAPBUILD_CONSISTENT)
1595 0 : SnapBuildRestore(builder, lsn);
1596 : else
1597 74 : SnapBuildSerialize(builder, lsn);
1598 74 : }
1599 :
1600 : /*
1601 : * Serialize the snapshot 'builder' at the location 'lsn' if it hasn't already
1602 : * been done by another decoding process.
1603 : */
1604 : static void
1605 648 : SnapBuildSerialize(SnapBuild *builder, XLogRecPtr lsn)
1606 : {
1607 : Size needed_length;
1608 648 : SnapBuildOnDisk *ondisk = NULL;
1609 648 : TransactionId *catchange_xip = NULL;
1610 : MemoryContext old_ctx;
1611 : size_t catchange_xcnt;
1612 : char *ondisk_c;
1613 : int fd;
1614 : char tmppath[MAXPGPATH];
1615 : char path[MAXPGPATH];
1616 : int ret;
1617 : struct stat stat_buf;
1618 : Size sz;
1619 :
1620 : Assert(lsn != InvalidXLogRecPtr);
1621 : Assert(builder->last_serialized_snapshot == InvalidXLogRecPtr ||
1622 : builder->last_serialized_snapshot <= lsn);
1623 :
1624 : /*
1625 : * no point in serializing if we cannot continue to work immediately after
1626 : * restoring the snapshot
1627 : */
1628 648 : if (builder->state < SNAPBUILD_CONSISTENT)
1629 0 : return;
1630 :
1631 : /* consistent snapshots have no next phase */
1632 : Assert(builder->next_phase_at == InvalidTransactionId);
1633 :
1634 : /*
1635 : * We identify snapshots by the LSN they are valid for. We don't need to
1636 : * include timelines in the name as each LSN maps to exactly one timeline
1637 : * unless the user used pg_resetwal or similar. If a user did so, there's
1638 : * no hope continuing to decode anyway.
1639 : */
1640 648 : sprintf(path, "pg_logical/snapshots/%X-%X.snap",
1641 648 : LSN_FORMAT_ARGS(lsn));
1642 :
1643 : /*
1644 : * first check whether some other backend already has written the snapshot
1645 : * for this LSN. It's perfectly fine if there's none, so we accept ENOENT
1646 : * as a valid state. Everything else is an unexpected error.
1647 : */
1648 648 : ret = stat(path, &stat_buf);
1649 :
1650 648 : if (ret != 0 && errno != ENOENT)
1651 0 : ereport(ERROR,
1652 : (errcode_for_file_access(),
1653 : errmsg("could not stat file \"%s\": %m", path)));
1654 :
1655 648 : else if (ret == 0)
1656 : {
1657 : /*
1658 : * somebody else has already serialized to this point, don't overwrite
1659 : * but remember location, so we don't need to read old data again.
1660 : *
1661 : * To be sure it has been synced to disk after the rename() from the
1662 : * tempfile filename to the real filename, we just repeat the fsync.
1663 : * That ought to be cheap because in most scenarios it should already
1664 : * be safely on disk.
1665 : */
1666 146 : fsync_fname(path, false);
1667 146 : fsync_fname("pg_logical/snapshots", true);
1668 :
1669 146 : builder->last_serialized_snapshot = lsn;
1670 146 : goto out;
1671 : }
1672 :
1673 : /*
1674 : * there is an obvious race condition here between the time we stat(2) the
1675 : * file and us writing the file. But we rename the file into place
1676 : * atomically and all files created need to contain the same data anyway,
1677 : * so this is perfectly fine, although a bit of a resource waste. Locking
1678 : * seems like pointless complication.
1679 : */
1680 502 : elog(DEBUG1, "serializing snapshot to %s", path);
1681 :
1682 : /* to make sure only we will write to this tempfile, include pid */
1683 502 : sprintf(tmppath, "pg_logical/snapshots/%X-%X.snap.%d.tmp",
1684 502 : LSN_FORMAT_ARGS(lsn), MyProcPid);
1685 :
1686 : /*
1687 : * Unlink temporary file if it already exists, needs to have been before a
1688 : * crash/error since we won't enter this function twice from within a
1689 : * single decoding slot/backend and the temporary file contains the pid of
1690 : * the current process.
1691 : */
1692 502 : if (unlink(tmppath) != 0 && errno != ENOENT)
1693 0 : ereport(ERROR,
1694 : (errcode_for_file_access(),
1695 : errmsg("could not remove file \"%s\": %m", tmppath)));
1696 :
1697 502 : old_ctx = MemoryContextSwitchTo(builder->context);
1698 :
1699 : /* Get the catalog modifying transactions that are yet not committed */
1700 502 : catchange_xip = ReorderBufferGetCatalogChangesXacts(builder->reorder);
1701 502 : catchange_xcnt = dclist_count(&builder->reorder->catchange_txns);
1702 :
1703 502 : needed_length = sizeof(SnapBuildOnDisk) +
1704 502 : sizeof(TransactionId) * (builder->committed.xcnt + catchange_xcnt);
1705 :
1706 502 : ondisk_c = palloc0(needed_length);
1707 502 : ondisk = (SnapBuildOnDisk *) ondisk_c;
1708 502 : ondisk->magic = SNAPBUILD_MAGIC;
1709 502 : ondisk->version = SNAPBUILD_VERSION;
1710 502 : ondisk->length = needed_length;
1711 502 : INIT_CRC32C(ondisk->checksum);
1712 502 : COMP_CRC32C(ondisk->checksum,
1713 : ((char *) ondisk) + SnapBuildOnDiskNotChecksummedSize,
1714 : SnapBuildOnDiskConstantSize - SnapBuildOnDiskNotChecksummedSize);
1715 502 : ondisk_c += sizeof(SnapBuildOnDisk);
1716 :
1717 502 : memcpy(&ondisk->builder, builder, sizeof(SnapBuild));
1718 : /* NULL-ify memory-only data */
1719 502 : ondisk->builder.context = NULL;
1720 502 : ondisk->builder.snapshot = NULL;
1721 502 : ondisk->builder.reorder = NULL;
1722 502 : ondisk->builder.committed.xip = NULL;
1723 502 : ondisk->builder.catchange.xip = NULL;
1724 : /* update catchange only on disk data */
1725 502 : ondisk->builder.catchange.xcnt = catchange_xcnt;
1726 :
1727 502 : COMP_CRC32C(ondisk->checksum,
1728 : &ondisk->builder,
1729 : sizeof(SnapBuild));
1730 :
1731 : /* copy committed xacts */
1732 502 : if (builder->committed.xcnt > 0)
1733 : {
1734 86 : sz = sizeof(TransactionId) * builder->committed.xcnt;
1735 86 : memcpy(ondisk_c, builder->committed.xip, sz);
1736 86 : COMP_CRC32C(ondisk->checksum, ondisk_c, sz);
1737 86 : ondisk_c += sz;
1738 : }
1739 :
1740 : /* copy catalog modifying xacts */
1741 502 : if (catchange_xcnt > 0)
1742 : {
1743 14 : sz = sizeof(TransactionId) * catchange_xcnt;
1744 14 : memcpy(ondisk_c, catchange_xip, sz);
1745 14 : COMP_CRC32C(ondisk->checksum, ondisk_c, sz);
1746 14 : ondisk_c += sz;
1747 : }
1748 :
1749 502 : FIN_CRC32C(ondisk->checksum);
1750 :
1751 : /* we have valid data now, open tempfile and write it there */
1752 502 : fd = OpenTransientFile(tmppath,
1753 : O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
1754 502 : if (fd < 0)
1755 0 : ereport(ERROR,
1756 : (errcode_for_file_access(),
1757 : errmsg("could not open file \"%s\": %m", tmppath)));
1758 :
1759 502 : errno = 0;
1760 502 : pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_WRITE);
1761 502 : if ((write(fd, ondisk, needed_length)) != needed_length)
1762 : {
1763 0 : int save_errno = errno;
1764 :
1765 0 : CloseTransientFile(fd);
1766 :
1767 : /* if write didn't set errno, assume problem is no disk space */
1768 0 : errno = save_errno ? save_errno : ENOSPC;
1769 0 : ereport(ERROR,
1770 : (errcode_for_file_access(),
1771 : errmsg("could not write to file \"%s\": %m", tmppath)));
1772 : }
1773 502 : pgstat_report_wait_end();
1774 :
1775 : /*
1776 : * fsync the file before renaming so that even if we crash after this we
1777 : * have either a fully valid file or nothing.
1778 : *
1779 : * It's safe to just ERROR on fsync() here because we'll retry the whole
1780 : * operation including the writes.
1781 : *
1782 : * TODO: Do the fsync() via checkpoints/restartpoints, doing it here has
1783 : * some noticeable overhead since it's performed synchronously during
1784 : * decoding?
1785 : */
1786 502 : pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_SYNC);
1787 502 : if (pg_fsync(fd) != 0)
1788 : {
1789 0 : int save_errno = errno;
1790 :
1791 0 : CloseTransientFile(fd);
1792 0 : errno = save_errno;
1793 0 : ereport(ERROR,
1794 : (errcode_for_file_access(),
1795 : errmsg("could not fsync file \"%s\": %m", tmppath)));
1796 : }
1797 502 : pgstat_report_wait_end();
1798 :
1799 502 : if (CloseTransientFile(fd) != 0)
1800 0 : ereport(ERROR,
1801 : (errcode_for_file_access(),
1802 : errmsg("could not close file \"%s\": %m", tmppath)));
1803 :
1804 502 : fsync_fname("pg_logical/snapshots", true);
1805 :
1806 : /*
1807 : * We may overwrite the work from some other backend, but that's ok, our
1808 : * snapshot is valid as well, we'll just have done some superfluous work.
1809 : */
1810 502 : if (rename(tmppath, path) != 0)
1811 : {
1812 0 : ereport(ERROR,
1813 : (errcode_for_file_access(),
1814 : errmsg("could not rename file \"%s\" to \"%s\": %m",
1815 : tmppath, path)));
1816 : }
1817 :
1818 : /* make sure we persist */
1819 502 : fsync_fname(path, false);
1820 502 : fsync_fname("pg_logical/snapshots", true);
1821 :
1822 : /*
1823 : * Now there's no way we can lose the dumped state anymore, remember this
1824 : * as a serialization point.
1825 : */
1826 502 : builder->last_serialized_snapshot = lsn;
1827 :
1828 502 : MemoryContextSwitchTo(old_ctx);
1829 :
1830 648 : out:
1831 648 : ReorderBufferSetRestartPoint(builder->reorder,
1832 : builder->last_serialized_snapshot);
1833 : /* be tidy */
1834 648 : if (ondisk)
1835 502 : pfree(ondisk);
1836 648 : if (catchange_xip)
1837 14 : pfree(catchange_xip);
1838 : }
1839 :
1840 : /*
1841 : * Restore a snapshot into 'builder' if previously one has been stored at the
1842 : * location indicated by 'lsn'. Returns true if successful, false otherwise.
1843 : */
1844 : static bool
1845 60 : SnapBuildRestore(SnapBuild *builder, XLogRecPtr lsn)
1846 : {
1847 : SnapBuildOnDisk ondisk;
1848 : int fd;
1849 : char path[MAXPGPATH];
1850 : Size sz;
1851 : pg_crc32c checksum;
1852 :
1853 : /* no point in loading a snapshot if we're already there */
1854 60 : if (builder->state == SNAPBUILD_CONSISTENT)
1855 0 : return false;
1856 :
1857 60 : sprintf(path, "pg_logical/snapshots/%X-%X.snap",
1858 60 : LSN_FORMAT_ARGS(lsn));
1859 :
1860 60 : fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
1861 :
1862 60 : if (fd < 0 && errno == ENOENT)
1863 46 : return false;
1864 14 : else if (fd < 0)
1865 0 : ereport(ERROR,
1866 : (errcode_for_file_access(),
1867 : errmsg("could not open file \"%s\": %m", path)));
1868 :
1869 : /* ----
1870 : * Make sure the snapshot had been stored safely to disk, that's normally
1871 : * cheap.
1872 : * Note that we do not need PANIC here, nobody will be able to use the
1873 : * slot without fsyncing, and saving it won't succeed without an fsync()
1874 : * either...
1875 : * ----
1876 : */
1877 14 : fsync_fname(path, false);
1878 14 : fsync_fname("pg_logical/snapshots", true);
1879 :
1880 :
1881 : /* read statically sized portion of snapshot */
1882 14 : SnapBuildRestoreContents(fd, (char *) &ondisk, SnapBuildOnDiskConstantSize, path);
1883 :
1884 14 : if (ondisk.magic != SNAPBUILD_MAGIC)
1885 0 : ereport(ERROR,
1886 : (errcode(ERRCODE_DATA_CORRUPTED),
1887 : errmsg("snapbuild state file \"%s\" has wrong magic number: %u instead of %u",
1888 : path, ondisk.magic, SNAPBUILD_MAGIC)));
1889 :
1890 14 : if (ondisk.version != SNAPBUILD_VERSION)
1891 0 : ereport(ERROR,
1892 : (errcode(ERRCODE_DATA_CORRUPTED),
1893 : errmsg("snapbuild state file \"%s\" has unsupported version: %u instead of %u",
1894 : path, ondisk.version, SNAPBUILD_VERSION)));
1895 :
1896 14 : INIT_CRC32C(checksum);
1897 14 : COMP_CRC32C(checksum,
1898 : ((char *) &ondisk) + SnapBuildOnDiskNotChecksummedSize,
1899 : SnapBuildOnDiskConstantSize - SnapBuildOnDiskNotChecksummedSize);
1900 :
1901 : /* read SnapBuild */
1902 14 : SnapBuildRestoreContents(fd, (char *) &ondisk.builder, sizeof(SnapBuild), path);
1903 14 : COMP_CRC32C(checksum, &ondisk.builder, sizeof(SnapBuild));
1904 :
1905 : /* restore committed xacts information */
1906 14 : if (ondisk.builder.committed.xcnt > 0)
1907 : {
1908 4 : sz = sizeof(TransactionId) * ondisk.builder.committed.xcnt;
1909 4 : ondisk.builder.committed.xip = MemoryContextAllocZero(builder->context, sz);
1910 4 : SnapBuildRestoreContents(fd, (char *) ondisk.builder.committed.xip, sz, path);
1911 4 : COMP_CRC32C(checksum, ondisk.builder.committed.xip, sz);
1912 : }
1913 :
1914 : /* restore catalog modifying xacts information */
1915 14 : if (ondisk.builder.catchange.xcnt > 0)
1916 : {
1917 6 : sz = sizeof(TransactionId) * ondisk.builder.catchange.xcnt;
1918 6 : ondisk.builder.catchange.xip = MemoryContextAllocZero(builder->context, sz);
1919 6 : SnapBuildRestoreContents(fd, (char *) ondisk.builder.catchange.xip, sz, path);
1920 6 : COMP_CRC32C(checksum, ondisk.builder.catchange.xip, sz);
1921 : }
1922 :
1923 14 : if (CloseTransientFile(fd) != 0)
1924 0 : ereport(ERROR,
1925 : (errcode_for_file_access(),
1926 : errmsg("could not close file \"%s\": %m", path)));
1927 :
1928 14 : FIN_CRC32C(checksum);
1929 :
1930 : /* verify checksum of what we've read */
1931 14 : if (!EQ_CRC32C(checksum, ondisk.checksum))
1932 0 : ereport(ERROR,
1933 : (errcode(ERRCODE_DATA_CORRUPTED),
1934 : errmsg("checksum mismatch for snapbuild state file \"%s\": is %u, should be %u",
1935 : path, checksum, ondisk.checksum)));
1936 :
1937 : /*
1938 : * ok, we now have a sensible snapshot here, figure out if it has more
1939 : * information than we have.
1940 : */
1941 :
1942 : /*
1943 : * We are only interested in consistent snapshots for now, comparing
1944 : * whether one incomplete snapshot is more "advanced" seems to be
1945 : * unnecessarily complex.
1946 : */
1947 14 : if (ondisk.builder.state < SNAPBUILD_CONSISTENT)
1948 0 : goto snapshot_not_interesting;
1949 :
1950 : /*
1951 : * Don't use a snapshot that requires an xmin that we cannot guarantee to
1952 : * be available.
1953 : */
1954 14 : if (TransactionIdPrecedes(ondisk.builder.xmin, builder->initial_xmin_horizon))
1955 0 : goto snapshot_not_interesting;
1956 :
1957 : /*
1958 : * Consistent snapshots have no next phase. Reset next_phase_at as it is
1959 : * possible that an old value may remain.
1960 : */
1961 : Assert(ondisk.builder.next_phase_at == InvalidTransactionId);
1962 14 : builder->next_phase_at = InvalidTransactionId;
1963 :
1964 : /* ok, we think the snapshot is sensible, copy over everything important */
1965 14 : builder->xmin = ondisk.builder.xmin;
1966 14 : builder->xmax = ondisk.builder.xmax;
1967 14 : builder->state = ondisk.builder.state;
1968 :
1969 14 : builder->committed.xcnt = ondisk.builder.committed.xcnt;
1970 : /* We only allocated/stored xcnt, not xcnt_space xids ! */
1971 : /* don't overwrite preallocated xip, if we don't have anything here */
1972 14 : if (builder->committed.xcnt > 0)
1973 : {
1974 4 : pfree(builder->committed.xip);
1975 4 : builder->committed.xcnt_space = ondisk.builder.committed.xcnt;
1976 4 : builder->committed.xip = ondisk.builder.committed.xip;
1977 : }
1978 14 : ondisk.builder.committed.xip = NULL;
1979 :
1980 : /* set catalog modifying transactions */
1981 14 : if (builder->catchange.xip)
1982 0 : pfree(builder->catchange.xip);
1983 14 : builder->catchange.xcnt = ondisk.builder.catchange.xcnt;
1984 14 : builder->catchange.xip = ondisk.builder.catchange.xip;
1985 14 : ondisk.builder.catchange.xip = NULL;
1986 :
1987 : /* our snapshot is not interesting anymore, build a new one */
1988 14 : if (builder->snapshot != NULL)
1989 : {
1990 0 : SnapBuildSnapDecRefcount(builder->snapshot);
1991 : }
1992 14 : builder->snapshot = SnapBuildBuildSnapshot(builder);
1993 14 : SnapBuildSnapIncRefcount(builder->snapshot);
1994 :
1995 14 : ReorderBufferSetRestartPoint(builder->reorder, lsn);
1996 :
1997 : Assert(builder->state == SNAPBUILD_CONSISTENT);
1998 :
1999 14 : ereport(LOG,
2000 : (errmsg("logical decoding found consistent point at %X/%X",
2001 : LSN_FORMAT_ARGS(lsn)),
2002 : errdetail("Logical decoding will begin using saved snapshot.")));
2003 14 : return true;
2004 :
2005 0 : snapshot_not_interesting:
2006 0 : if (ondisk.builder.committed.xip != NULL)
2007 0 : pfree(ondisk.builder.committed.xip);
2008 0 : if (ondisk.builder.catchange.xip != NULL)
2009 0 : pfree(ondisk.builder.catchange.xip);
2010 0 : return false;
2011 : }
2012 :
2013 : /*
2014 : * Read the contents of the serialized snapshot to 'dest'.
2015 : */
2016 : static void
2017 38 : SnapBuildRestoreContents(int fd, char *dest, Size size, const char *path)
2018 : {
2019 : int readBytes;
2020 :
2021 38 : pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_READ);
2022 38 : readBytes = read(fd, dest, size);
2023 38 : pgstat_report_wait_end();
2024 38 : if (readBytes != size)
2025 : {
2026 0 : int save_errno = errno;
2027 :
2028 0 : CloseTransientFile(fd);
2029 :
2030 0 : if (readBytes < 0)
2031 : {
2032 0 : errno = save_errno;
2033 0 : ereport(ERROR,
2034 : (errcode_for_file_access(),
2035 : errmsg("could not read file \"%s\": %m", path)));
2036 : }
2037 : else
2038 0 : ereport(ERROR,
2039 : (errcode(ERRCODE_DATA_CORRUPTED),
2040 : errmsg("could not read file \"%s\": read %d of %zu",
2041 : path, readBytes, size)));
2042 : }
2043 38 : }
2044 :
2045 : /*
2046 : * Remove all serialized snapshots that are not required anymore because no
2047 : * slot can need them. This doesn't actually have to run during a checkpoint,
2048 : * but it's a convenient point to schedule this.
2049 : *
2050 : * NB: We run this during checkpoints even if logical decoding is disabled so
2051 : * we cleanup old slots at some point after it got disabled.
2052 : */
2053 : void
2054 1704 : CheckPointSnapBuild(void)
2055 : {
2056 : XLogRecPtr cutoff;
2057 : XLogRecPtr redo;
2058 : DIR *snap_dir;
2059 : struct dirent *snap_de;
2060 : char path[MAXPGPATH + 21];
2061 :
2062 : /*
2063 : * We start off with a minimum of the last redo pointer. No new
2064 : * replication slot will start before that, so that's a safe upper bound
2065 : * for removal.
2066 : */
2067 1704 : redo = GetRedoRecPtr();
2068 :
2069 : /* now check for the restart ptrs from existing slots */
2070 1704 : cutoff = ReplicationSlotsComputeLogicalRestartLSN();
2071 :
2072 : /* don't start earlier than the restart lsn */
2073 1704 : if (redo < cutoff)
2074 0 : cutoff = redo;
2075 :
2076 1704 : snap_dir = AllocateDir("pg_logical/snapshots");
2077 5548 : while ((snap_de = ReadDir(snap_dir, "pg_logical/snapshots")) != NULL)
2078 : {
2079 : uint32 hi;
2080 : uint32 lo;
2081 : XLogRecPtr lsn;
2082 : PGFileType de_type;
2083 :
2084 3844 : if (strcmp(snap_de->d_name, ".") == 0 ||
2085 2140 : strcmp(snap_de->d_name, "..") == 0)
2086 3408 : continue;
2087 :
2088 436 : snprintf(path, sizeof(path), "pg_logical/snapshots/%s", snap_de->d_name);
2089 436 : de_type = get_dirent_type(path, snap_de, false, DEBUG1);
2090 :
2091 436 : if (de_type != PGFILETYPE_ERROR && de_type != PGFILETYPE_REG)
2092 : {
2093 0 : elog(DEBUG1, "only regular files expected: %s", path);
2094 0 : continue;
2095 : }
2096 :
2097 : /*
2098 : * temporary filenames from SnapBuildSerialize() include the LSN and
2099 : * everything but are postfixed by .$pid.tmp. We can just remove them
2100 : * the same as other files because there can be none that are
2101 : * currently being written that are older than cutoff.
2102 : *
2103 : * We just log a message if a file doesn't fit the pattern, it's
2104 : * probably some editors lock/state file or similar...
2105 : */
2106 436 : if (sscanf(snap_de->d_name, "%X-%X.snap", &hi, &lo) != 2)
2107 : {
2108 0 : ereport(LOG,
2109 : (errmsg("could not parse file name \"%s\"", path)));
2110 0 : continue;
2111 : }
2112 :
2113 436 : lsn = ((uint64) hi) << 32 | lo;
2114 :
2115 : /* check whether we still need it */
2116 436 : if (lsn < cutoff || cutoff == InvalidXLogRecPtr)
2117 : {
2118 314 : elog(DEBUG1, "removing snapbuild snapshot %s", path);
2119 :
2120 : /*
2121 : * It's not particularly harmful, though strange, if we can't
2122 : * remove the file here. Don't prevent the checkpoint from
2123 : * completing, that'd be a cure worse than the disease.
2124 : */
2125 314 : if (unlink(path) < 0)
2126 : {
2127 0 : ereport(LOG,
2128 : (errcode_for_file_access(),
2129 : errmsg("could not remove file \"%s\": %m",
2130 : path)));
2131 0 : continue;
2132 : }
2133 : }
2134 : }
2135 1704 : FreeDir(snap_dir);
2136 1704 : }
2137 :
2138 : /*
2139 : * Check if a logical snapshot at the specified point has been serialized.
2140 : */
2141 : bool
2142 20 : SnapBuildSnapshotExists(XLogRecPtr lsn)
2143 : {
2144 : char path[MAXPGPATH];
2145 : int ret;
2146 : struct stat stat_buf;
2147 :
2148 20 : sprintf(path, "pg_logical/snapshots/%X-%X.snap",
2149 20 : LSN_FORMAT_ARGS(lsn));
2150 :
2151 20 : ret = stat(path, &stat_buf);
2152 :
2153 20 : if (ret != 0 && errno != ENOENT)
2154 0 : ereport(ERROR,
2155 : (errcode_for_file_access(),
2156 : errmsg("could not stat file \"%s\": %m", path)));
2157 :
2158 20 : return ret == 0;
2159 : }
|
