Age Owner Branch data TLA 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) 2012-2025, 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 *
4205 rhaas@postgresql.org 185 :CBC 1051 : 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 : 1051 : context = AllocSetContextCreate(CurrentMemoryContext,
198 : : "snapshot builder context",
199 : : ALLOCSET_DEFAULT_SIZES);
200 : 1051 : oldcontext = MemoryContextSwitchTo(context);
201 : :
202 : 1051 : builder = palloc0(sizeof(SnapBuild));
203 : :
204 : 1051 : builder->state = SNAPBUILD_START;
205 : 1051 : builder->context = context;
206 : 1051 : builder->reorder = reorder;
207 : : /* Other struct members initialized by zeroing via palloc0 above */
208 : :
209 : 1051 : builder->committed.xcnt = 0;
2999 tgl@sss.pgh.pa.us 210 : 1051 : builder->committed.xcnt_space = 128; /* arbitrary number */
4205 rhaas@postgresql.org 211 : 1051 : builder->committed.xip =
212 : 1051 : palloc0(builder->committed.xcnt_space * sizeof(TransactionId));
213 : 1051 : builder->committed.includes_all_transactions = true;
214 : :
1122 akapila@postgresql.o 215 : 1051 : builder->catchange.xcnt = 0;
216 : 1051 : builder->catchange.xip = NULL;
217 : :
4205 rhaas@postgresql.org 218 : 1051 : builder->initial_xmin_horizon = xmin_horizon;
4111 andres@anarazel.de 219 : 1051 : builder->start_decoding_at = start_lsn;
422 msawada@postgresql.o 220 : 1051 : builder->in_slot_creation = in_slot_creation;
3054 andres@anarazel.de 221 : 1051 : builder->building_full_snapshot = need_full_snapshot;
1515 akapila@postgresql.o 222 : 1051 : builder->two_phase_at = two_phase_at;
223 : :
4205 rhaas@postgresql.org 224 : 1051 : MemoryContextSwitchTo(oldcontext);
225 : :
226 : 1051 : return builder;
227 : : }
228 : :
229 : : /*
230 : : * Free a snapshot builder.
231 : : */
232 : : void
233 : 839 : FreeSnapshotBuilder(SnapBuild *builder)
234 : : {
235 : 839 : MemoryContext context = builder->context;
236 : :
237 : : /* free snapshot explicitly, that contains some error checking */
238 [ + + ]: 839 : if (builder->snapshot != NULL)
239 : : {
240 : 209 : SnapBuildSnapDecRefcount(builder->snapshot);
241 : 209 : builder->snapshot = NULL;
242 : : }
243 : :
244 : : /* other resources are deallocated via memory context reset */
245 : 839 : MemoryContextDelete(context);
246 : 839 : }
247 : :
248 : : /*
249 : : * Free an unreferenced snapshot that has previously been built by us.
250 : : */
251 : : static void
252 : 1359 : SnapBuildFreeSnapshot(Snapshot snap)
253 : : {
254 : : /* make sure we don't get passed an external snapshot */
2420 andres@anarazel.de 255 [ - + ]: 1359 : Assert(snap->snapshot_type == SNAPSHOT_HISTORIC_MVCC);
256 : :
257 : : /* make sure nobody modified our snapshot */
4205 rhaas@postgresql.org 258 [ - + ]: 1359 : Assert(snap->curcid == FirstCommandId);
259 [ - + ]: 1359 : Assert(!snap->suboverflowed);
260 [ - + ]: 1359 : Assert(!snap->takenDuringRecovery);
3796 heikki.linnakangas@i 261 [ - + ]: 1359 : Assert(snap->regd_count == 0);
262 : :
263 : : /* slightly more likely, so it's checked even without c-asserts */
4205 rhaas@postgresql.org 264 [ - + ]: 1359 : if (snap->copied)
4205 rhaas@postgresql.org 265 [ # # ]:UBC 0 : elog(ERROR, "cannot free a copied snapshot");
266 : :
4205 rhaas@postgresql.org 267 [ - + ]:CBC 1359 : if (snap->active_count)
4205 rhaas@postgresql.org 268 [ # # ]:UBC 0 : elog(ERROR, "cannot free an active snapshot");
269 : :
4205 rhaas@postgresql.org 270 :CBC 1359 : pfree(snap);
271 : 1359 : }
272 : :
273 : : /*
274 : : * In which state of snapshot building are we?
275 : : */
276 : : SnapBuildState
277 : 1776712 : SnapBuildCurrentState(SnapBuild *builder)
278 : : {
279 : 1776712 : return builder->state;
280 : : }
281 : :
282 : : /*
283 : : * Return the LSN at which the two-phase decoding was first enabled.
284 : : */
285 : : XLogRecPtr
1515 akapila@postgresql.o 286 : 31 : SnapBuildGetTwoPhaseAt(SnapBuild *builder)
287 : : {
288 : 31 : return builder->two_phase_at;
289 : : }
290 : :
291 : : /*
292 : : * Set the LSN at which two-phase decoding is enabled.
293 : : */
294 : : void
295 : 8 : SnapBuildSetTwoPhaseAt(SnapBuild *builder, XLogRecPtr ptr)
296 : : {
297 : 8 : builder->two_phase_at = ptr;
1650 298 : 8 : }
299 : :
300 : : /*
301 : : * Should the contents of transaction ending at 'ptr' be decoded?
302 : : */
303 : : bool
4205 rhaas@postgresql.org 304 : 331952 : SnapBuildXactNeedsSkip(SnapBuild *builder, XLogRecPtr ptr)
305 : : {
4111 andres@anarazel.de 306 : 331952 : 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
4205 rhaas@postgresql.org 316 : 5928 : SnapBuildSnapIncRefcount(Snapshot snap)
317 : : {
318 : 5928 : snap->active_count++;
319 : 5928 : }
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 : 5686 : SnapBuildSnapDecRefcount(Snapshot snap)
329 : : {
330 : : /* make sure we don't get passed an external snapshot */
2420 andres@anarazel.de 331 [ - + ]: 5686 : Assert(snap->snapshot_type == SNAPSHOT_HISTORIC_MVCC);
332 : :
333 : : /* make sure nobody modified our snapshot */
4205 rhaas@postgresql.org 334 [ - + ]: 5686 : Assert(snap->curcid == FirstCommandId);
335 [ - + ]: 5686 : Assert(!snap->suboverflowed);
336 [ - + ]: 5686 : Assert(!snap->takenDuringRecovery);
337 : :
3796 heikki.linnakangas@i 338 [ - + ]: 5686 : Assert(snap->regd_count == 0);
339 : :
340 [ - + ]: 5686 : Assert(snap->active_count > 0);
341 : :
342 : : /* slightly more likely, so it's checked even without casserts */
4205 rhaas@postgresql.org 343 [ - + ]: 5686 : if (snap->copied)
4205 rhaas@postgresql.org 344 [ # # ]:UBC 0 : elog(ERROR, "cannot free a copied snapshot");
345 : :
4205 rhaas@postgresql.org 346 :CBC 5686 : snap->active_count--;
3796 heikki.linnakangas@i 347 [ + + ]: 5686 : if (snap->active_count == 0)
4205 rhaas@postgresql.org 348 : 1359 : SnapBuildFreeSnapshot(snap);
349 : 5686 : }
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
3006 andres@anarazel.de 360 : 1777 : SnapBuildBuildSnapshot(SnapBuild *builder)
361 : : {
362 : : Snapshot snapshot;
363 : : Size ssize;
364 : :
4205 rhaas@postgresql.org 365 [ - + ]: 1777 : Assert(builder->state >= SNAPBUILD_FULL_SNAPSHOT);
366 : :
367 : 1777 : ssize = sizeof(SnapshotData)
368 : 1777 : + sizeof(TransactionId) * builder->committed.xcnt
369 : 1777 : + sizeof(TransactionId) * 1 /* toplevel xid */ ;
370 : :
371 : 1777 : snapshot = MemoryContextAllocZero(builder->context, ssize);
372 : :
2420 andres@anarazel.de 373 : 1777 : 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 : : */
4205 rhaas@postgresql.org 396 [ - + ]: 1777 : Assert(TransactionIdIsNormal(builder->xmin));
397 [ - + ]: 1777 : Assert(TransactionIdIsNormal(builder->xmax));
398 : :
399 : 1777 : snapshot->xmin = builder->xmin;
400 : 1777 : snapshot->xmax = builder->xmax;
401 : :
402 : : /* store all transactions to be treated as committed by this snapshot */
403 : 1777 : snapshot->xip =
404 : 1777 : (TransactionId *) ((char *) snapshot + sizeof(SnapshotData));
405 : 1777 : snapshot->xcnt = builder->committed.xcnt;
406 : 1777 : memcpy(snapshot->xip,
407 : 1777 : builder->committed.xip,
408 : 1777 : builder->committed.xcnt * sizeof(TransactionId));
409 : :
410 : : /* sort so we can bsearch() */
411 : 1777 : 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 : 1777 : snapshot->subxcnt = 0;
419 : 1777 : snapshot->subxip = NULL;
420 : :
421 : 1777 : snapshot->suboverflowed = false;
422 : 1777 : snapshot->takenDuringRecovery = false;
423 : 1777 : snapshot->copied = false;
424 : 1777 : snapshot->curcid = FirstCommandId;
425 : 1777 : snapshot->active_count = 0;
3796 heikki.linnakangas@i 426 : 1777 : snapshot->regd_count = 0;
1846 andres@anarazel.de 427 : 1777 : snapshot->snapXactCompletionCount = 0;
428 : :
4205 rhaas@postgresql.org 429 : 1777 : 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
3086 tgl@sss.pgh.pa.us 440 : 195 : SnapBuildInitialSnapshot(SnapBuild *builder)
441 : : {
442 : : Snapshot snap;
443 : : TransactionId xid;
444 : : TransactionId safeXid;
445 : : TransactionId *newxip;
4205 rhaas@postgresql.org 446 : 195 : int newxcnt = 0;
447 : :
3086 tgl@sss.pgh.pa.us 448 [ - + ]: 195 : Assert(XactIsoLevel == XACT_REPEATABLE_READ);
1020 akapila@postgresql.o 449 [ - + ]: 195 : Assert(builder->building_full_snapshot);
450 : :
451 : : /* don't allow older snapshots */
841 tgl@sss.pgh.pa.us 452 : 195 : InvalidateCatalogSnapshot(); /* about to overwrite MyProc->xmin */
1020 akapila@postgresql.o 453 [ - + ]: 195 : if (HaveRegisteredOrActiveSnapshot())
1020 akapila@postgresql.o 454 [ # # ]:UBC 0 : elog(ERROR, "cannot build an initial slot snapshot when snapshots exist");
1020 akapila@postgresql.o 455 [ - + ]:CBC 195 : Assert(!HistoricSnapshotActive());
456 : :
4205 rhaas@postgresql.org 457 [ - + ]: 195 : if (builder->state != SNAPBUILD_CONSISTENT)
3089 peter_e@gmx.net 458 [ # # ]:UBC 0 : elog(ERROR, "cannot build an initial slot snapshot before reaching a consistent state");
459 : :
4205 rhaas@postgresql.org 460 [ - + ]:CBC 195 : if (!builder->committed.includes_all_transactions)
3089 peter_e@gmx.net 461 [ # # ]:UBC 0 : elog(ERROR, "cannot build an initial slot snapshot, not all transactions are monitored anymore");
462 : :
463 : : /* so we don't overwrite the existing value */
1850 andres@anarazel.de 464 [ - + ]:CBC 195 : if (TransactionIdIsValid(MyProc->xmin))
1850 andres@anarazel.de 465 [ # # ]:UBC 0 : elog(ERROR, "cannot build an initial slot snapshot when MyProc->xmin already is valid");
466 : :
3006 andres@anarazel.de 467 :CBC 195 : 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 : : */
1020 akapila@postgresql.o 478 : 195 : LWLockAcquire(ProcArrayLock, LW_SHARED);
479 : 195 : safeXid = GetOldestSafeDecodingTransactionId(false);
480 : 195 : LWLockRelease(ProcArrayLock);
481 : :
482 [ - + ]: 195 : if (TransactionIdFollows(safeXid, snap->xmin))
1020 akapila@postgresql.o 483 [ # # ]:UBC 0 : elog(ERROR, "cannot build an initial slot snapshot as oldest safe xid %u follows snapshot's xmin %u",
484 : : safeXid, snap->xmin);
485 : :
1850 andres@anarazel.de 486 :CBC 195 : MyProc->xmin = snap->xmin;
487 : :
488 : : /* allocate in transaction context */
489 : : newxip = (TransactionId *)
4205 rhaas@postgresql.org 490 : 195 : 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 [ + - - + : 195 : 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 : : */
4205 rhaas@postgresql.org 506 :UBC 0 : test = bsearch(&xid, snap->xip, snap->xcnt,
507 : : sizeof(TransactionId), xidComparator);
508 : :
509 [ # # ]: 0 : if (test == NULL)
510 : : {
511 [ # # ]: 0 : if (newxcnt >= GetMaxSnapshotXidCount())
3089 peter_e@gmx.net 512 [ # # ]: 0 : ereport(ERROR,
513 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
514 : : errmsg("initial slot snapshot too large")));
515 : :
4205 rhaas@postgresql.org 516 : 0 : newxip[newxcnt++] = xid;
517 : : }
518 : :
519 [ # # ]: 0 : TransactionIdAdvance(xid);
520 : : }
521 : :
522 : : /* adjust remaining snapshot fields as needed */
2390 michael@paquier.xyz 523 :CBC 195 : snap->snapshot_type = SNAPSHOT_MVCC;
4205 rhaas@postgresql.org 524 : 195 : snap->xcnt = newxcnt;
525 : 195 : snap->xip = newxip;
526 : :
3089 peter_e@gmx.net 527 : 195 : 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 : 1 : SnapBuildExportSnapshot(SnapBuild *builder)
540 : : {
541 : : Snapshot snap;
542 : : char *snapname;
543 : :
544 [ - + ]: 1 : if (IsTransactionOrTransactionBlock())
3089 peter_e@gmx.net 545 [ # # ]:UBC 0 : elog(ERROR, "cannot export a snapshot from within a transaction");
546 : :
3089 peter_e@gmx.net 547 [ - + ]:CBC 1 : if (SavedResourceOwnerDuringExport)
3089 peter_e@gmx.net 548 [ # # ]:UBC 0 : elog(ERROR, "can only export one snapshot at a time");
549 : :
3089 peter_e@gmx.net 550 :CBC 1 : SavedResourceOwnerDuringExport = CurrentResourceOwner;
551 : 1 : ExportInProgress = true;
552 : :
553 : 1 : StartTransactionCommand();
554 : :
555 : : /* There doesn't seem to a nice API to set these */
556 : 1 : XactIsoLevel = XACT_REPEATABLE_READ;
557 : 1 : XactReadOnly = true;
558 : :
3086 tgl@sss.pgh.pa.us 559 : 1 : 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 : : */
4205 rhaas@postgresql.org 565 : 1 : snapname = ExportSnapshot(snap);
566 : :
567 [ + - ]: 1 : 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 : 1 : return snapname;
573 : : }
574 : :
575 : : /*
576 : : * Ensure there is a snapshot and if not build one for current transaction.
577 : : */
578 : : Snapshot
1076 akapila@postgresql.o 579 : 8 : SnapBuildGetOrBuildSnapshot(SnapBuild *builder)
580 : : {
3440 simon@2ndQuadrant.co 581 [ - + ]: 8 : Assert(builder->state == SNAPBUILD_CONSISTENT);
582 : :
583 : : /* only build a new snapshot if we don't have a prebuilt one */
584 [ + + ]: 8 : if (builder->snapshot == NULL)
585 : : {
3006 andres@anarazel.de 586 : 1 : builder->snapshot = SnapBuildBuildSnapshot(builder);
587 : : /* increase refcount for the snapshot builder */
3440 simon@2ndQuadrant.co 588 : 1 : SnapBuildSnapIncRefcount(builder->snapshot);
589 : : }
590 : :
591 : 8 : 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
3675 andres@anarazel.de 600 : 5060 : SnapBuildClearExportedSnapshot(void)
601 : : {
602 : : ResourceOwner tmpResOwner;
603 : :
604 : : /* nothing exported, that is the usual case */
4205 rhaas@postgresql.org 605 [ + + ]: 5060 : if (!ExportInProgress)
606 : 5059 : return;
607 : :
608 [ - + ]: 1 : if (!IsTransactionState())
4205 rhaas@postgresql.org 609 [ # # ]:UBC 0 : 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 : : */
1419 michael@paquier.xyz 615 :CBC 1 : tmpResOwner = SavedResourceOwnerDuringExport;
616 : :
617 : : /* make sure nothing could have ever happened */
4205 rhaas@postgresql.org 618 : 1 : AbortCurrentTransaction();
619 : :
1419 michael@paquier.xyz 620 : 1 : CurrentResourceOwner = tmpResOwner;
621 : : }
622 : :
623 : : /*
624 : : * Clear snapshot export state during transaction abort.
625 : : */
626 : : void
627 : 24951 : SnapBuildResetExportedSnapshotState(void)
628 : : {
4205 rhaas@postgresql.org 629 : 24951 : SavedResourceOwnerDuringExport = NULL;
630 : 24951 : ExportInProgress = false;
631 : 24951 : }
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 : 1256442 : 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 [ - + ]: 1256442 : if (builder->state < SNAPBUILD_FULL_SNAPSHOT)
4205 rhaas@postgresql.org 646 :UBC 0 : 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 : : */
4205 rhaas@postgresql.org 653 [ + + - + ]:CBC 1256445 : if (builder->state < SNAPBUILD_CONSISTENT &&
1664 andres@anarazel.de 654 : 3 : TransactionIdPrecedes(xid, builder->next_phase_at))
4205 rhaas@postgresql.org 655 :UBC 0 : 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 : : */
3472 andres@anarazel.de 661 [ + + ]:CBC 1256442 : if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, xid))
662 : : {
663 : : /* only build a new snapshot if we don't have a prebuilt one */
4205 rhaas@postgresql.org 664 [ + + ]: 3141 : if (builder->snapshot == NULL)
665 : : {
3006 andres@anarazel.de 666 : 381 : builder->snapshot = SnapBuildBuildSnapshot(builder);
667 : : /* increase refcount for the snapshot builder */
4205 rhaas@postgresql.org 668 : 381 : SnapBuildSnapIncRefcount(builder->snapshot);
669 : : }
670 : :
671 : : /*
672 : : * Increase refcount for the transaction we're handing the snapshot
673 : : * out to.
674 : : */
675 : 3141 : SnapBuildSnapIncRefcount(builder->snapshot);
676 : 3141 : ReorderBufferSetBaseSnapshot(builder->reorder, xid, lsn,
677 : : builder->snapshot);
678 : : }
679 : :
680 : 1256442 : 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 : 24222 : 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 : : */
4141 bruce@momjian.us 698 : 24222 : ReorderBufferXidSetCatalogChanges(builder->reorder, xid, lsn);
699 : :
4205 rhaas@postgresql.org 700 : 24222 : 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 [ + + ]: 24222 : if (xlrec->cmin != InvalidCommandId &&
707 [ + + ]: 20306 : xlrec->cmax != InvalidCommandId)
708 : 3204 : cid = Max(xlrec->cmin, xlrec->cmax);
709 [ + + ]: 21018 : else if (xlrec->cmax != InvalidCommandId)
710 : 3916 : cid = xlrec->cmax;
711 [ + - ]: 17102 : else if (xlrec->cmin != InvalidCommandId)
712 : 17102 : cid = xlrec->cmin;
713 : : else
714 : : {
4141 bruce@momjian.us 715 :UBC 0 : cid = InvalidCommandId; /* silence compiler */
4205 rhaas@postgresql.org 716 [ # # ]: 0 : elog(ERROR, "xl_heap_new_cid record without a valid CommandId");
717 : : }
718 : :
4205 rhaas@postgresql.org 719 :CBC 24222 : ReorderBufferAddNewCommandId(builder->reorder, xid, lsn, cid + 1);
720 : 24222 : }
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
149 akapila@postgresql.o 731 : 1191 : 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 : : */
4205 rhaas@postgresql.org 742 [ + - + + ]: 2418 : dlist_foreach(txn_i, &builder->reorder->toplevel_by_lsn)
743 : : {
744 : 1227 : txn = dlist_container(ReorderBufferTXN, node, txn_i.cur);
745 : :
746 [ - + ]: 1227 : 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 [ + + ]: 1227 : if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, txn->xid))
768 : 2 : 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 : : */
206 msawada@postgresql.o 774 [ + + ]: 1225 : if (rbtxn_is_prepared(txn))
1706 akapila@postgresql.o 775 : 28 : continue;
776 : :
61 alvherre@kurilemu.de 777 [ + + ]:GNC 1197 : elog(DEBUG2, "adding a new snapshot and invalidations to %u at %X/%08X",
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 : : */
4205 rhaas@postgresql.org 784 :CBC 1197 : SnapBuildSnapIncRefcount(builder->snapshot);
785 : 1197 : 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 : : */
149 akapila@postgresql.o 805 [ + + ]: 1197 : if (txn->xid != xid)
806 : : {
807 : : uint32 ninvalidations;
808 : 34 : SharedInvalidationMessage *msgs = NULL;
809 : :
810 : 34 : ninvalidations = ReorderBufferGetInvalidations(builder->reorder,
811 : : xid, &msgs);
812 : :
813 [ + + ]: 34 : if (ninvalidations > 0)
814 : : {
815 [ - + ]: 30 : Assert(msgs != NULL);
816 : :
82 msawada@postgresql.o 817 : 30 : ReorderBufferAddDistributedInvalidations(builder->reorder,
818 : : txn->xid, lsn,
819 : : ninvalidations, msgs);
820 : : }
821 : : }
822 : : }
4205 rhaas@postgresql.org 823 : 1191 : }
824 : :
825 : : /*
826 : : * Keep track of a new catalog changing transaction that has committed.
827 : : */
828 : : static void
829 : 1200 : SnapBuildAddCommittedTxn(SnapBuild *builder, TransactionId xid)
830 : : {
831 [ - + ]: 1200 : Assert(TransactionIdIsValid(xid));
832 : :
833 [ + + ]: 1200 : if (builder->committed.xcnt == builder->committed.xcnt_space)
834 : : {
4205 rhaas@postgresql.org 835 :GBC 1 : builder->committed.xcnt_space = builder->committed.xcnt_space * 2 + 1;
836 : :
837 [ - + ]: 1 : elog(DEBUG1, "increasing space for committed transactions to %u",
838 : : (uint32) builder->committed.xcnt_space);
839 : :
840 : 1 : builder->committed.xip = repalloc(builder->committed.xip,
2999 tgl@sss.pgh.pa.us 841 : 1 : 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 : : */
4205 rhaas@postgresql.org 849 :CBC 1200 : builder->committed.xip[builder->committed.xcnt++] = xid;
850 : 1200 : }
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
1122 akapila@postgresql.o 863 : 417 : SnapBuildPurgeOlderTxn(SnapBuild *builder)
864 : : {
865 : : int off;
866 : : TransactionId *workspace;
4205 rhaas@postgresql.org 867 : 417 : int surviving_xids = 0;
868 : :
869 : : /* not ready yet */
870 [ - + ]: 417 : if (!TransactionIdIsNormal(builder->xmin))
4205 rhaas@postgresql.org 871 :UBC 0 : return;
872 : :
873 : : /* TODO: Neater algorithm than just copying and iterating? */
874 : : workspace =
4205 rhaas@postgresql.org 875 :CBC 417 : MemoryContextAlloc(builder->context,
876 : 417 : builder->committed.xcnt * sizeof(TransactionId));
877 : :
878 : : /* copy xids that still are interesting to workspace */
879 [ + + ]: 627 : for (off = 0; off < builder->committed.xcnt; off++)
880 : : {
881 [ + - - + : 210 : if (NormalTransactionIdPrecedes(builder->committed.xip[off],
+ + ]
882 : : builder->xmin))
883 : : ; /* remove */
884 : : else
885 : 1 : workspace[surviving_xids++] = builder->committed.xip[off];
886 : : }
887 : :
888 : : /* copy workspace back to persistent state */
889 : 417 : memcpy(builder->committed.xip, workspace,
890 : : surviving_xids * sizeof(TransactionId));
891 : :
892 [ - + ]: 417 : 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 : 417 : builder->committed.xcnt = surviving_xids;
896 : :
897 : 417 : pfree(workspace);
898 : :
899 : : /*
900 : : * Purge xids in ->catchange as well. The purged array must also be sorted
901 : : * in xidComparator order.
902 : : */
1104 akapila@postgresql.o 903 [ + + ]: 417 : 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 [ + + ]: 9 : for (off = 0; off < builder->catchange.xcnt; off++)
910 : : {
911 [ + + ]: 6 : if (TransactionIdFollowsOrEquals(builder->catchange.xip[off],
912 : : builder->xmin))
913 : 1 : break;
914 : : }
915 : :
916 : 4 : surviving_xids = builder->catchange.xcnt - off;
917 : :
918 [ + + ]: 4 : if (surviving_xids > 0)
919 : : {
920 : 1 : memmove(builder->catchange.xip, &(builder->catchange.xip[off]),
921 : : surviving_xids * sizeof(TransactionId));
922 : : }
923 : : else
924 : : {
925 : 3 : pfree(builder->catchange.xip);
926 : 3 : builder->catchange.xip = NULL;
927 : : }
928 : :
929 [ - + ]: 4 : 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 : 4 : builder->catchange.xcnt = surviving_xids;
933 : : }
934 : : }
935 : :
936 : : /*
937 : : * Handle everything that needs to be done when a transaction commits
938 : : */
939 : : void
4205 rhaas@postgresql.org 940 : 3037 : SnapBuildCommitTxn(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid,
941 : : int nsubxacts, TransactionId *subxacts, uint32 xinfo)
942 : : {
943 : : int nxact;
944 : :
3038 andres@anarazel.de 945 : 3037 : bool needs_snapshot = false;
946 : 3037 : bool needs_timetravel = false;
4205 rhaas@postgresql.org 947 : 3037 : bool sub_needs_timetravel = false;
948 : :
949 : 3037 : 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 : : */
3038 andres@anarazel.de 955 [ + - ]: 3037 : if (builder->state == SNAPBUILD_START ||
956 [ - + - - ]: 3037 : (builder->state == SNAPBUILD_BUILDING_SNAPSHOT &&
1664 andres@anarazel.de 957 :UBC 0 : TransactionIdPrecedes(xid, builder->next_phase_at)))
958 : : {
959 : : /* ensure that only commits after this are getting replayed */
3038 960 [ # # ]: 0 : if (builder->start_decoding_at <= lsn)
961 : 0 : builder->start_decoding_at = lsn + 1;
962 : 0 : return;
963 : : }
964 : :
4205 rhaas@postgresql.org 965 [ + + ]:CBC 3037 : if (builder->state < SNAPBUILD_CONSISTENT)
966 : : {
967 : : /* ensure that only commits after this are getting replayed */
4111 andres@anarazel.de 968 [ + + ]: 5 : if (builder->start_decoding_at <= lsn)
969 : 2 : 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 : : */
3038 975 [ - + ]: 5 : if (builder->building_full_snapshot)
976 : : {
3038 andres@anarazel.de 977 :UBC 0 : needs_timetravel = true;
978 : : }
979 : : }
980 : :
4205 rhaas@postgresql.org 981 [ + + ]:CBC 4246 : for (nxact = 0; nxact < nsubxacts; nxact++)
982 : : {
983 : 1209 : 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 : : */
1122 akapila@postgresql.o 989 [ + + ]: 1209 : if (SnapBuildXidHasCatalogChanges(builder, subxid, xinfo))
990 : : {
3038 andres@anarazel.de 991 : 9 : sub_needs_timetravel = true;
992 : 9 : needs_snapshot = true;
993 : :
994 [ - + ]: 9 : elog(DEBUG1, "found subtransaction %u:%u with catalog changes",
995 : : xid, subxid);
996 : :
4205 rhaas@postgresql.org 997 : 9 : SnapBuildAddCommittedTxn(builder, subxid);
998 : :
999 [ + - - + : 9 : if (NormalTransactionIdFollows(subxid, xmax))
+ - ]
1000 : 9 : 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 : : */
3038 andres@anarazel.de 1009 [ - + ]: 1200 : else if (needs_timetravel)
1010 : : {
4205 rhaas@postgresql.org 1011 :UBC 0 : SnapBuildAddCommittedTxn(builder, subxid);
1012 [ # # # # : 0 : if (NormalTransactionIdFollows(subxid, xmax))
# # ]
1013 : 0 : xmax = subxid;
1014 : : }
1015 : : }
1016 : :
1017 : : /* if top-level modified catalog, it'll need a snapshot */
1122 akapila@postgresql.o 1018 [ + + ]:CBC 3037 : if (SnapBuildXidHasCatalogChanges(builder, xid, xinfo))
1019 : : {
3038 andres@anarazel.de 1020 [ + + ]: 1190 : elog(DEBUG2, "found top level transaction %u, with catalog changes",
1021 : : xid);
1022 : 1190 : needs_snapshot = true;
1023 : 1190 : needs_timetravel = true;
4205 rhaas@postgresql.org 1024 : 1190 : SnapBuildAddCommittedTxn(builder, xid);
1025 : : }
3038 andres@anarazel.de 1026 [ + + ]: 1847 : else if (sub_needs_timetravel)
1027 : : {
1028 : : /* track toplevel txn as well, subxact alone isn't meaningful */
1052 akapila@postgresql.o 1029 [ - + ]: 1 : elog(DEBUG2, "forced transaction %u to do timetravel due to one of its subtransactions",
1030 : : xid);
1031 : 1 : needs_timetravel = true;
4205 rhaas@postgresql.org 1032 : 1 : SnapBuildAddCommittedTxn(builder, xid);
1033 : : }
3038 andres@anarazel.de 1034 [ - + ]: 1846 : else if (needs_timetravel)
1035 : : {
3038 andres@anarazel.de 1036 [ # # ]:UBC 0 : elog(DEBUG2, "forced transaction %u to do timetravel", xid);
1037 : :
4205 rhaas@postgresql.org 1038 : 0 : SnapBuildAddCommittedTxn(builder, xid);
1039 : : }
1040 : :
3038 andres@anarazel.de 1041 [ + + ]:CBC 3037 : if (!needs_timetravel)
1042 : : {
1043 : : /* record that we cannot export a general snapshot anymore */
1044 : 1846 : builder->committed.includes_all_transactions = false;
1045 : : }
1046 : :
1047 [ + + - + ]: 3037 : 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 [ + + ]: 3037 : if (needs_timetravel &&
1055 [ + - + + ]: 2382 : (!TransactionIdIsValid(builder->xmax) ||
1056 : 1191 : TransactionIdFollowsOrEquals(xmax, builder->xmax)))
1057 : : {
1058 : 1190 : builder->xmax = xmax;
1059 [ - + ]: 1190 : TransactionIdAdvance(builder->xmax);
1060 : : }
1061 : :
1062 : : /* if there's any reason to build a historic snapshot, do so now */
1063 [ + + ]: 3037 : 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 : : */
4205 rhaas@postgresql.org 1069 [ - + ]: 1191 : if (builder->state < SNAPBUILD_FULL_SNAPSHOT)
4205 rhaas@postgresql.org 1070 :UBC 0 : 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 : : */
4205 rhaas@postgresql.org 1077 [ + - ]:CBC 1191 : if (builder->snapshot)
1078 : 1191 : SnapBuildSnapDecRefcount(builder->snapshot);
1079 : :
3006 andres@anarazel.de 1080 : 1191 : builder->snapshot = SnapBuildBuildSnapshot(builder);
1081 : :
1082 : : /* we might need to execute invalidations, add snapshot */
4205 rhaas@postgresql.org 1083 [ + + ]: 1191 : if (!ReorderBufferXidHasBaseSnapshot(builder->reorder, xid))
1084 : : {
1085 : 8 : SnapBuildSnapIncRefcount(builder->snapshot);
1086 : 8 : ReorderBufferSetBaseSnapshot(builder->reorder, xid, lsn,
1087 : : builder->snapshot);
1088 : : }
1089 : :
1090 : : /* refcount of the snapshot builder for the new snapshot */
1091 : 1191 : SnapBuildSnapIncRefcount(builder->snapshot);
1092 : :
1093 : : /*
1094 : : * Add a new catalog snapshot and invalidations messages to all
1095 : : * currently running transactions.
1096 : : */
149 akapila@postgresql.o 1097 : 1191 : 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
1122 1106 : 4246 : SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
1107 : : uint32 xinfo)
1108 : : {
1109 [ + + ]: 4246 : if (ReorderBufferXidHasCatalogChanges(builder->reorder, xid))
1110 : 1195 : return true;
1111 : :
1112 : : /*
1113 : : * The transactions that have changed catalogs must have invalidation
1114 : : * info.
1115 : : */
1116 [ + + ]: 3051 : if (!(xinfo & XACT_XINFO_HAS_INVALS))
1117 : 3043 : return false;
1118 : :
1119 : : /* Check the catchange XID array */
1120 [ + + + - ]: 12 : return ((builder->catchange.xcnt > 0) &&
1121 : 4 : (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
4205 rhaas@postgresql.org 1136 : 1419 : 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 [ + + ]: 1419 : if (builder->state < SNAPBUILD_CONSISTENT)
1147 : : {
1148 : : /* returns false if there's no point in performing cleanup just yet */
1149 [ + + ]: 1025 : if (!SnapBuildFindSnapshot(builder, lsn, running))
1150 : 1000 : return;
1151 : : }
1152 : : else
1153 : 394 : 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 : 417 : builder->xmin = running->oldestRunningXid;
1168 : :
1169 : : /* Remove transactions we don't need to keep track off anymore */
1122 akapila@postgresql.o 1170 : 417 : 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 : : */
2629 alvherre@alvh.no-ip. 1181 : 417 : xmin = ReorderBufferGetOldestXmin(builder->reorder);
1182 [ + + ]: 417 : if (xmin == InvalidTransactionId)
1183 : 371 : xmin = running->oldestRunningXid;
1184 [ - + ]: 417 : elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u",
1185 : : builder->xmin, builder->xmax, running->oldestRunningXid, xmin);
1186 : 417 : 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 : : */
4205 rhaas@postgresql.org 1204 [ + + ]: 417 : if (builder->state < SNAPBUILD_CONSISTENT)
1205 : 18 : return;
1206 : :
1207 : 399 : 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 [ + + + + ]: 399 : if (txn != NULL && txn->restart_decoding_lsn != InvalidXLogRecPtr)
1214 : 23 : 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 [ + + ]: 376 : else if (txn == NULL &&
2999 tgl@sss.pgh.pa.us 1221 [ + + ]: 344 : builder->reorder->current_restart_decoding_lsn != InvalidXLogRecPtr &&
4205 rhaas@postgresql.org 1222 [ + - ]: 342 : builder->last_serialized_snapshot != InvalidXLogRecPtr)
1223 : 342 : 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 : 1025 : 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 [ + + ]: 1025 : if (TransactionIdIsNormal(builder->initial_xmin_horizon) &&
1270 [ + - - + : 433 : NormalTransactionIdPrecedes(running->oldestRunningXid,
- + ]
1271 : : builder->initial_xmin_horizon))
1272 : : {
4205 rhaas@postgresql.org 1273 [ # # ]:UBC 0 : ereport(DEBUG1,
1274 : : errmsg_internal("skipping snapshot at %X/%08X 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 : :
3038 andres@anarazel.de 1280 : 0 : SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon);
1281 : :
4205 rhaas@postgresql.org 1282 : 0 : 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 : : */
3038 andres@anarazel.de 1294 [ + + ]:CBC 1025 : if (running->oldestRunningXid == running->nextXid)
1295 : : {
4111 1296 [ + + ]: 991 : if (builder->start_decoding_at == InvalidXLogRecPtr ||
1297 [ + + ]: 567 : builder->start_decoding_at <= lsn)
1298 : : /* can decode everything after this */
1299 : 425 : builder->start_decoding_at = lsn + 1;
1300 : :
1301 : : /* As no transactions were running xmin/xmax can be trivially set. */
2999 tgl@sss.pgh.pa.us 1302 : 991 : builder->xmin = running->nextXid; /* < are finished */
1303 : 991 : builder->xmax = running->nextXid; /* >= are running */
1304 : :
1305 : : /* so we can safely use the faster comparisons */
4205 rhaas@postgresql.org 1306 [ - + ]: 991 : Assert(TransactionIdIsNormal(builder->xmin));
1307 [ - + ]: 991 : Assert(TransactionIdIsNormal(builder->xmax));
1308 : :
1309 : 991 : builder->state = SNAPBUILD_CONSISTENT;
1664 andres@anarazel.de 1310 : 991 : builder->next_phase_at = InvalidTransactionId;
1311 : :
4205 rhaas@postgresql.org 1312 [ + - ]: 991 : ereport(LOG,
1313 : : errmsg("logical decoding found consistent point at %X/%08X",
1314 : : LSN_FORMAT_ARGS(lsn)),
1315 : : errdetail("There are no running transactions."));
1316 : :
1317 : 991 : return false;
1318 : : }
1319 : :
1320 : : /*
1321 : : * b) valid on disk state and while neither building full snapshot nor
1322 : : * creating a slot.
1323 : : */
3054 andres@anarazel.de 1324 [ + + ]: 34 : else if (!builder->building_full_snapshot &&
422 msawada@postgresql.o 1325 [ + + + + ]: 52 : !builder->in_slot_creation &&
3054 andres@anarazel.de 1326 : 20 : SnapBuildRestore(builder, lsn))
1327 : : {
1328 : : /* there won't be any state to cleanup */
4205 rhaas@postgresql.org 1329 : 9 : 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 : : */
3038 andres@anarazel.de 1344 [ + + ]: 25 : else if (builder->state == SNAPBUILD_START)
1345 : : {
1346 : 14 : builder->state = SNAPBUILD_BUILDING_SNAPSHOT;
1664 1347 : 14 : 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 : : */
2999 tgl@sss.pgh.pa.us 1354 : 14 : builder->xmin = running->nextXid; /* < are finished */
1355 : 14 : builder->xmax = running->nextXid; /* >= are running */
1356 : :
1357 : : /* so we can safely use the faster comparisons */
4205 rhaas@postgresql.org 1358 [ - + ]: 14 : Assert(TransactionIdIsNormal(builder->xmin));
1359 [ - + ]: 14 : Assert(TransactionIdIsNormal(builder->xmax));
1360 : :
1361 [ + - ]: 14 : ereport(LOG,
1362 : : errmsg("logical decoding found initial starting point at %X/%08X",
1363 : : LSN_FORMAT_ARGS(lsn)),
1364 : : errdetail("Waiting for transactions (approximately %d) older than %u to end.",
1365 : : running->xcnt, running->nextXid));
1366 : :
3038 andres@anarazel.de 1367 : 14 : 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 [ + + + + ]: 17 : else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT &&
1664 1379 : 6 : TransactionIdPrecedesOrEquals(builder->next_phase_at,
1380 : : running->oldestRunningXid))
1381 : : {
3038 1382 : 5 : builder->state = SNAPBUILD_FULL_SNAPSHOT;
1664 1383 : 5 : builder->next_phase_at = running->nextXid;
1384 : :
3038 1385 [ + - ]: 5 : ereport(LOG,
1386 : : errmsg("logical decoding found initial consistent point at %X/%08X",
1387 : : LSN_FORMAT_ARGS(lsn)),
1388 : : errdetail("Waiting for transactions (approximately %d) older than %u to end.",
1389 : : running->xcnt, running->nextXid));
1390 : :
1391 : 5 : 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 [ + + + - ]: 11 : else if (builder->state == SNAPBUILD_FULL_SNAPSHOT &&
1664 1403 : 5 : TransactionIdPrecedesOrEquals(builder->next_phase_at,
1404 : : running->oldestRunningXid))
1405 : : {
3038 1406 : 5 : builder->state = SNAPBUILD_CONSISTENT;
1664 1407 : 5 : builder->next_phase_at = InvalidTransactionId;
1408 : :
3038 1409 [ + - ]: 5 : ereport(LOG,
1410 : : errmsg("logical decoding found consistent point at %X/%08X",
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 : : */
4205 rhaas@postgresql.org 1420 : 23 : 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
3038 andres@anarazel.de 1435 : 19 : SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
1436 : : {
1437 : : int off;
1438 : :
1439 [ + + ]: 36 : for (off = 0; off < running->xcnt; off++)
1440 : : {
1441 : 19 : 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 [ - + ]: 19 : if (TransactionIdIsCurrentTransactionId(xid))
3038 andres@anarazel.de 1449 [ # # ]:UBC 0 : elog(ERROR, "waiting for ourselves");
1450 : :
3038 andres@anarazel.de 1451 [ - + ]:CBC 19 : if (TransactionIdFollows(xid, cutoff))
3038 andres@anarazel.de 1452 :UBC 0 : continue;
1453 : :
3038 andres@anarazel.de 1454 :CBC 19 : 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 [ + - ]: 17 : if (!RecoveryInProgress())
1464 : : {
1465 : 17 : LogStandbySnapshot();
1466 : : }
1467 : 17 : }
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
4205 rhaas@postgresql.org 1484 : 57 : SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn)
1485 : : {
1486 [ - + ]: 57 : if (builder->state < SNAPBUILD_CONSISTENT)
4205 rhaas@postgresql.org 1487 :UBC 0 : SnapBuildRestore(builder, lsn);
1488 : : else
4205 rhaas@postgresql.org 1489 :CBC 57 : SnapBuildSerialize(builder, lsn);
1490 : 57 : }
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 : 451 : SnapBuildSerialize(SnapBuild *builder, XLogRecPtr lsn)
1498 : : {
1499 : : Size needed_length;
1697 akapila@postgresql.o 1500 : 451 : SnapBuildOnDisk *ondisk = NULL;
1122 1501 : 451 : 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 : :
4205 rhaas@postgresql.org 1512 [ - + ]: 451 : Assert(lsn != InvalidXLogRecPtr);
1513 [ + + - + ]: 451 : 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 [ - + ]: 451 : if (builder->state < SNAPBUILD_CONSISTENT)
4205 rhaas@postgresql.org 1521 :UBC 0 : return;
1522 : :
1523 : : /* consistent snapshots have no next phase */
1664 andres@anarazel.de 1524 [ - + ]:CBC 451 : 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 : : */
372 michael@paquier.xyz 1532 : 451 : sprintf(path, "%s/%X-%X.snap",
1533 : : PG_LOGICAL_SNAPSHOTS_DIR,
1656 peter@eisentraut.org 1534 : 451 : 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 : : */
4205 rhaas@postgresql.org 1541 : 451 : ret = stat(path, &stat_buf);
1542 : :
1543 [ + + - + ]: 451 : if (ret != 0 && errno != ENOENT)
4205 rhaas@postgresql.org 1544 [ # # ]:UBC 0 : ereport(ERROR,
1545 : : (errcode_for_file_access(),
1546 : : errmsg("could not stat file \"%s\": %m", path)));
1547 : :
4205 rhaas@postgresql.org 1548 [ + + ]:CBC 451 : 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 : 169 : fsync_fname(path, false);
372 michael@paquier.xyz 1560 : 169 : fsync_fname(PG_LOGICAL_SNAPSHOTS_DIR, true);
1561 : :
4205 rhaas@postgresql.org 1562 : 169 : builder->last_serialized_snapshot = lsn;
1563 : 169 : 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 [ + + ]: 282 : elog(DEBUG1, "serializing snapshot to %s", path);
1574 : :
1575 : : /* to make sure only we will write to this tempfile, include pid */
372 michael@paquier.xyz 1576 : 282 : sprintf(tmppath, "%s/%X-%X.snap.%d.tmp",
1577 : : PG_LOGICAL_SNAPSHOTS_DIR,
1656 peter@eisentraut.org 1578 : 282 : 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 : : */
4205 rhaas@postgresql.org 1586 [ + - - + ]: 282 : if (unlink(tmppath) != 0 && errno != ENOENT)
4205 rhaas@postgresql.org 1587 [ # # ]:UBC 0 : ereport(ERROR,
1588 : : (errcode_for_file_access(),
1589 : : errmsg("could not remove file \"%s\": %m", tmppath)));
1590 : :
1122 akapila@postgresql.o 1591 :CBC 282 : old_ctx = MemoryContextSwitchTo(builder->context);
1592 : :
1593 : : /* Get the catalog modifying transactions that are yet not committed */
1594 : 282 : catchange_xip = ReorderBufferGetCatalogChangesXacts(builder->reorder);
1039 drowley@postgresql.o 1595 : 282 : catchange_xcnt = dclist_count(&builder->reorder->catchange_txns);
1596 : :
4205 rhaas@postgresql.org 1597 : 282 : needed_length = sizeof(SnapBuildOnDisk) +
1122 akapila@postgresql.o 1598 : 282 : sizeof(TransactionId) * (builder->committed.xcnt + catchange_xcnt);
1599 : :
1600 : 282 : ondisk_c = palloc0(needed_length);
4205 rhaas@postgresql.org 1601 : 282 : ondisk = (SnapBuildOnDisk *) ondisk_c;
1602 : 282 : ondisk->magic = SNAPBUILD_MAGIC;
1603 : 282 : ondisk->version = SNAPBUILD_VERSION;
1604 : 282 : ondisk->length = needed_length;
3959 heikki.linnakangas@i 1605 : 282 : INIT_CRC32C(ondisk->checksum);
1606 : 282 : COMP_CRC32C(ondisk->checksum,
1607 : : ((char *) ondisk) + SnapBuildOnDiskNotChecksummedSize,
1608 : : SnapBuildOnDiskConstantSize - SnapBuildOnDiskNotChecksummedSize);
4205 rhaas@postgresql.org 1609 : 282 : ondisk_c += sizeof(SnapBuildOnDisk);
1610 : :
1611 : 282 : memcpy(&ondisk->builder, builder, sizeof(SnapBuild));
1612 : : /* NULL-ify memory-only data */
1613 : 282 : ondisk->builder.context = NULL;
1614 : 282 : ondisk->builder.snapshot = NULL;
1615 : 282 : ondisk->builder.reorder = NULL;
1616 : 282 : ondisk->builder.committed.xip = NULL;
1122 akapila@postgresql.o 1617 : 282 : ondisk->builder.catchange.xip = NULL;
1618 : : /* update catchange only on disk data */
1619 : 282 : ondisk->builder.catchange.xcnt = catchange_xcnt;
1620 : :
3959 heikki.linnakangas@i 1621 : 282 : COMP_CRC32C(ondisk->checksum,
1622 : : &ondisk->builder,
1623 : : sizeof(SnapBuild));
1624 : :
1625 : : /* copy committed xacts */
1122 akapila@postgresql.o 1626 [ + + ]: 282 : if (builder->committed.xcnt > 0)
1627 : : {
1628 : 48 : sz = sizeof(TransactionId) * builder->committed.xcnt;
1629 : 48 : memcpy(ondisk_c, builder->committed.xip, sz);
1630 : 48 : COMP_CRC32C(ondisk->checksum, ondisk_c, sz);
1631 : 48 : ondisk_c += sz;
1632 : : }
1633 : :
1634 : : /* copy catalog modifying xacts */
1635 [ + + ]: 282 : if (catchange_xcnt > 0)
1636 : : {
1637 : 8 : sz = sizeof(TransactionId) * catchange_xcnt;
1638 : 8 : memcpy(ondisk_c, catchange_xip, sz);
1639 : 8 : COMP_CRC32C(ondisk->checksum, ondisk_c, sz);
1640 : 8 : ondisk_c += sz;
1641 : : }
1642 : :
3951 andres@anarazel.de 1643 : 282 : FIN_CRC32C(ondisk->checksum);
1644 : :
1645 : : /* we have valid data now, open tempfile and write it there */
4205 rhaas@postgresql.org 1646 : 282 : fd = OpenTransientFile(tmppath,
1647 : : O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
1648 [ - + ]: 282 : if (fd < 0)
4205 rhaas@postgresql.org 1649 [ # # ]:UBC 0 : ereport(ERROR,
1650 : : (errcode_for_file_access(),
1651 : : errmsg("could not open file \"%s\": %m", tmppath)));
1652 : :
2589 michael@paquier.xyz 1653 :CBC 282 : errno = 0;
3094 rhaas@postgresql.org 1654 : 282 : pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_WRITE);
4205 1655 [ - + ]: 282 : if ((write(fd, ondisk, needed_length)) != needed_length)
1656 : : {
2630 michael@paquier.xyz 1657 :UBC 0 : int save_errno = errno;
1658 : :
4205 rhaas@postgresql.org 1659 : 0 : CloseTransientFile(fd);
1660 : :
1661 : : /* if write didn't set errno, assume problem is no disk space */
2630 michael@paquier.xyz 1662 [ # # ]: 0 : errno = save_errno ? save_errno : ENOSPC;
4205 rhaas@postgresql.org 1663 [ # # ]: 0 : ereport(ERROR,
1664 : : (errcode_for_file_access(),
1665 : : errmsg("could not write to file \"%s\": %m", tmppath)));
1666 : : }
3094 rhaas@postgresql.org 1667 :CBC 282 : 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 : 282 : pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_SYNC);
4205 1681 [ - + ]: 282 : if (pg_fsync(fd) != 0)
1682 : : {
2630 michael@paquier.xyz 1683 :UBC 0 : int save_errno = errno;
1684 : :
4205 rhaas@postgresql.org 1685 : 0 : CloseTransientFile(fd);
2630 michael@paquier.xyz 1686 : 0 : errno = save_errno;
4205 rhaas@postgresql.org 1687 [ # # ]: 0 : ereport(ERROR,
1688 : : (errcode_for_file_access(),
1689 : : errmsg("could not fsync file \"%s\": %m", tmppath)));
1690 : : }
3094 rhaas@postgresql.org 1691 :CBC 282 : pgstat_report_wait_end();
1692 : :
2254 peter@eisentraut.org 1693 [ - + ]: 282 : if (CloseTransientFile(fd) != 0)
2373 michael@paquier.xyz 1694 [ # # ]:UBC 0 : ereport(ERROR,
1695 : : (errcode_for_file_access(),
1696 : : errmsg("could not close file \"%s\": %m", tmppath)));
1697 : :
372 michael@paquier.xyz 1698 :CBC 282 : 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 : : */
4205 rhaas@postgresql.org 1704 [ - + ]: 282 : if (rename(tmppath, path) != 0)
1705 : : {
4205 rhaas@postgresql.org 1706 [ # # ]:UBC 0 : 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 */
4205 rhaas@postgresql.org 1713 :CBC 282 : fsync_fname(path, false);
372 michael@paquier.xyz 1714 : 282 : 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 : : */
4205 rhaas@postgresql.org 1720 : 282 : builder->last_serialized_snapshot = lsn;
1721 : :
1122 akapila@postgresql.o 1722 : 282 : MemoryContextSwitchTo(old_ctx);
1723 : :
4205 rhaas@postgresql.org 1724 : 451 : out:
1725 : 451 : ReorderBufferSetRestartPoint(builder->reorder,
1726 : : builder->last_serialized_snapshot);
1727 : : /* be tidy */
1697 akapila@postgresql.o 1728 [ + + ]: 451 : if (ondisk)
1729 : 282 : pfree(ondisk);
1122 1730 [ + + ]: 451 : if (catchange_xip)
1731 : 8 : 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
179 msawada@postgresql.o 1742 : 22 : 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 : 22 : sprintf(path, "%s/%X-%X.snap",
1751 : : PG_LOGICAL_SNAPSHOTS_DIR,
1752 : 22 : LSN_FORMAT_ARGS(lsn));
1753 : :
2905 peter_e@gmx.net 1754 : 22 : fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
1755 : :
327 msawada@postgresql.o 1756 [ + + ]: 22 : if (fd < 0)
1757 : : {
1758 [ + - + - ]: 11 : if (missing_ok && errno == ENOENT)
1759 : 11 : return false;
1760 : :
4205 rhaas@postgresql.org 1761 [ # # ]:UBC 0 : 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 : : */
4205 rhaas@postgresql.org 1774 :CBC 11 : fsync_fname(path, false);
372 michael@paquier.xyz 1775 : 11 : fsync_fname(PG_LOGICAL_SNAPSHOTS_DIR, true);
1776 : :
1777 : : /* read statically sized portion of snapshot */
347 peter@eisentraut.org 1778 : 11 : SnapBuildRestoreContents(fd, ondisk, SnapBuildOnDiskConstantSize, path);
1779 : :
327 msawada@postgresql.o 1780 [ - + ]: 11 : if (ondisk->magic != SNAPBUILD_MAGIC)
4205 rhaas@postgresql.org 1781 [ # # ]:UBC 0 : 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 : :
327 msawada@postgresql.o 1786 [ - + ]:CBC 11 : if (ondisk->version != SNAPBUILD_VERSION)
4205 rhaas@postgresql.org 1787 [ # # ]:UBC 0 : 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 : :
3959 heikki.linnakangas@i 1792 :CBC 11 : INIT_CRC32C(checksum);
1793 : 11 : COMP_CRC32C(checksum,
1794 : : ((char *) ondisk) + SnapBuildOnDiskNotChecksummedSize,
1795 : : SnapBuildOnDiskConstantSize - SnapBuildOnDiskNotChecksummedSize);
1796 : :
1797 : : /* read SnapBuild */
347 peter@eisentraut.org 1798 : 11 : SnapBuildRestoreContents(fd, &ondisk->builder, sizeof(SnapBuild), path);
327 msawada@postgresql.o 1799 : 11 : COMP_CRC32C(checksum, &ondisk->builder, sizeof(SnapBuild));
1800 : :
1801 : : /* restore committed xacts information */
1802 [ + + ]: 11 : if (ondisk->builder.committed.xcnt > 0)
1803 : : {
1804 : 5 : sz = sizeof(TransactionId) * ondisk->builder.committed.xcnt;
1805 : 5 : ondisk->builder.committed.xip = MemoryContextAllocZero(context, sz);
347 peter@eisentraut.org 1806 : 5 : SnapBuildRestoreContents(fd, ondisk->builder.committed.xip, sz, path);
327 msawada@postgresql.o 1807 : 5 : COMP_CRC32C(checksum, ondisk->builder.committed.xip, sz);
1808 : : }
1809 : :
1810 : : /* restore catalog modifying xacts information */
1811 [ + + ]: 11 : if (ondisk->builder.catchange.xcnt > 0)
1812 : : {
1813 : 5 : sz = sizeof(TransactionId) * ondisk->builder.catchange.xcnt;
1814 : 5 : ondisk->builder.catchange.xip = MemoryContextAllocZero(context, sz);
347 peter@eisentraut.org 1815 : 5 : SnapBuildRestoreContents(fd, ondisk->builder.catchange.xip, sz, path);
327 msawada@postgresql.o 1816 : 5 : COMP_CRC32C(checksum, ondisk->builder.catchange.xip, sz);
1817 : : }
1818 : :
2254 peter@eisentraut.org 1819 [ - + ]: 11 : if (CloseTransientFile(fd) != 0)
2373 michael@paquier.xyz 1820 [ # # ]:UBC 0 : ereport(ERROR,
1821 : : (errcode_for_file_access(),
1822 : : errmsg("could not close file \"%s\": %m", path)));
1823 : :
3951 andres@anarazel.de 1824 :CBC 11 : FIN_CRC32C(checksum);
1825 : :
1826 : : /* verify checksum of what we've read */
327 msawada@postgresql.o 1827 [ - + ]: 11 : if (!EQ_CRC32C(checksum, ondisk->checksum))
4205 rhaas@postgresql.org 1828 [ # # ]:UBC 0 : 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 : :
327 msawada@postgresql.o 1833 :CBC 11 : 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 : 20 : SnapBuildRestore(SnapBuild *builder, XLogRecPtr lsn)
1842 : : {
1843 : : SnapBuildOnDisk ondisk;
1844 : :
1845 : : /* no point in loading a snapshot if we're already there */
1846 [ - + ]: 20 : if (builder->state == SNAPBUILD_CONSISTENT)
327 msawada@postgresql.o 1847 :UBC 0 : return false;
1848 : :
1849 : : /* validate and restore the snapshot to 'ondisk' */
179 msawada@postgresql.o 1850 [ + + ]:CBC 20 : if (!SnapBuildRestoreSnapshot(&ondisk, lsn, builder->context, true))
327 1851 : 11 : 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 : : */
4205 rhaas@postgresql.org 1863 [ - + ]: 9 : if (ondisk.builder.state < SNAPBUILD_CONSISTENT)
4205 rhaas@postgresql.org 1864 :UBC 0 : 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 : : */
4205 rhaas@postgresql.org 1870 [ - + ]:CBC 9 : if (TransactionIdPrecedes(ondisk.builder.xmin, builder->initial_xmin_horizon))
4205 rhaas@postgresql.org 1871 :UBC 0 : 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 : : */
1664 andres@anarazel.de 1877 [ - + ]:CBC 9 : Assert(ondisk.builder.next_phase_at == InvalidTransactionId);
795 dgustafsson@postgres 1878 : 9 : builder->next_phase_at = InvalidTransactionId;
1879 : :
1880 : : /* ok, we think the snapshot is sensible, copy over everything important */
4205 rhaas@postgresql.org 1881 : 9 : builder->xmin = ondisk.builder.xmin;
1882 : 9 : builder->xmax = ondisk.builder.xmax;
1883 : 9 : builder->state = ondisk.builder.state;
1884 : :
1885 : 9 : 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 [ + + ]: 9 : if (builder->committed.xcnt > 0)
1889 : : {
1890 : 3 : pfree(builder->committed.xip);
1891 : 3 : builder->committed.xcnt_space = ondisk.builder.committed.xcnt;
1892 : 3 : builder->committed.xip = ondisk.builder.committed.xip;
1893 : : }
1894 : 9 : ondisk.builder.committed.xip = NULL;
1895 : :
1896 : : /* set catalog modifying transactions */
1122 akapila@postgresql.o 1897 [ - + ]: 9 : if (builder->catchange.xip)
1122 akapila@postgresql.o 1898 :UBC 0 : pfree(builder->catchange.xip);
1122 akapila@postgresql.o 1899 :CBC 9 : builder->catchange.xcnt = ondisk.builder.catchange.xcnt;
1900 : 9 : builder->catchange.xip = ondisk.builder.catchange.xip;
1901 : 9 : ondisk.builder.catchange.xip = NULL;
1902 : :
1903 : : /* our snapshot is not interesting anymore, build a new one */
4205 rhaas@postgresql.org 1904 [ - + ]: 9 : if (builder->snapshot != NULL)
1905 : : {
4205 rhaas@postgresql.org 1906 :UBC 0 : SnapBuildSnapDecRefcount(builder->snapshot);
1907 : : }
3006 andres@anarazel.de 1908 :CBC 9 : builder->snapshot = SnapBuildBuildSnapshot(builder);
4205 rhaas@postgresql.org 1909 : 9 : SnapBuildSnapIncRefcount(builder->snapshot);
1910 : :
1911 : 9 : ReorderBufferSetRestartPoint(builder->reorder, lsn);
1912 : :
1913 [ - + ]: 9 : Assert(builder->state == SNAPBUILD_CONSISTENT);
1914 : :
1915 [ + - ]: 9 : ereport(LOG,
1916 : : errmsg("logical decoding found consistent point at %X/%08X",
1917 : : LSN_FORMAT_ARGS(lsn)),
1918 : : errdetail("Logical decoding will begin using saved snapshot."));
1919 : 9 : return true;
1920 : :
4205 rhaas@postgresql.org 1921 :UBC 0 : snapshot_not_interesting:
1922 [ # # ]: 0 : if (ondisk.builder.committed.xip != NULL)
1923 : 0 : pfree(ondisk.builder.committed.xip);
1122 akapila@postgresql.o 1924 [ # # ]: 0 : if (ondisk.builder.catchange.xip != NULL)
1925 : 0 : pfree(ondisk.builder.catchange.xip);
4205 rhaas@postgresql.org 1926 : 0 : return false;
1927 : : }
1928 : :
1929 : : /*
1930 : : * Read the contents of the serialized snapshot to 'dest'.
1931 : : */
1932 : : static void
347 peter@eisentraut.org 1933 :CBC 32 : SnapBuildRestoreContents(int fd, void *dest, Size size, const char *path)
1934 : : {
1935 : : int readBytes;
1936 : :
1122 akapila@postgresql.o 1937 : 32 : pgstat_report_wait_start(WAIT_EVENT_SNAPBUILD_READ);
1938 : 32 : readBytes = read(fd, dest, size);
1939 : 32 : pgstat_report_wait_end();
1940 [ - + ]: 32 : if (readBytes != size)
1941 : : {
1122 akapila@postgresql.o 1942 :UBC 0 : int save_errno = errno;
1943 : :
1944 : 0 : CloseTransientFile(fd);
1945 : :
1946 [ # # ]: 0 : if (readBytes < 0)
1947 : : {
1948 : 0 : errno = save_errno;
1949 [ # # ]: 0 : ereport(ERROR,
1950 : : (errcode_for_file_access(),
1951 : : errmsg("could not read file \"%s\": %m", path)));
1952 : : }
1953 : : else
1954 [ # # ]: 0 : ereport(ERROR,
1955 : : (errcode(ERRCODE_DATA_CORRUPTED),
1956 : : errmsg("could not read file \"%s\": read %d of %zu",
1957 : : path, readBytes, size)));
1958 : : }
1122 akapila@postgresql.o 1959 :CBC 32 : }
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
4205 rhaas@postgresql.org 1970 : 1677 : 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 : 1677 : redo = GetRedoRecPtr();
1984 : :
1985 : : /* now check for the restart ptrs from existing slots */
1986 : 1677 : cutoff = ReplicationSlotsComputeLogicalRestartLSN();
1987 : :
1988 : : /* don't start earlier than the restart lsn */
1989 [ + + ]: 1677 : if (redo < cutoff)
1990 : 1 : cutoff = redo;
1991 : :
372 michael@paquier.xyz 1992 : 1677 : snap_dir = AllocateDir(PG_LOGICAL_SNAPSHOTS_DIR);
1993 [ + + ]: 5296 : 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 : :
4205 rhaas@postgresql.org 2000 [ + + ]: 3619 : if (strcmp(snap_de->d_name, ".") == 0 ||
2001 [ + + ]: 1942 : strcmp(snap_de->d_name, "..") == 0)
2002 : 3354 : continue;
2003 : :
372 michael@paquier.xyz 2004 : 265 : snprintf(path, sizeof(path), "%s/%s", PG_LOGICAL_SNAPSHOTS_DIR, snap_de->d_name);
1100 2005 : 265 : de_type = get_dirent_type(path, snap_de, false, DEBUG1);
2006 : :
2007 [ + - - + ]: 265 : if (de_type != PGFILETYPE_ERROR && de_type != PGFILETYPE_REG)
2008 : : {
4205 rhaas@postgresql.org 2009 [ # # ]:UBC 0 : elog(DEBUG1, "only regular files expected: %s", path);
2010 : 0 : 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 : : */
4205 rhaas@postgresql.org 2022 [ - + ]:CBC 265 : if (sscanf(snap_de->d_name, "%X-%X.snap", &hi, &lo) != 2)
2023 : : {
4205 rhaas@postgresql.org 2024 [ # # ]:UBC 0 : ereport(LOG,
2025 : : (errmsg("could not parse file name \"%s\"", path)));
2026 : 0 : continue;
2027 : : }
2028 : :
4205 rhaas@postgresql.org 2029 :CBC 265 : lsn = ((uint64) hi) << 32 | lo;
2030 : :
2031 : : /* check whether we still need it */
2032 [ + + + + ]: 265 : if (lsn < cutoff || cutoff == InvalidXLogRecPtr)
2033 : : {
2034 [ + + ]: 177 : 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 [ - + ]: 177 : if (unlink(path) < 0)
2042 : : {
4205 rhaas@postgresql.org 2043 [ # # ]:UBC 0 : ereport(LOG,
2044 : : (errcode_for_file_access(),
2045 : : errmsg("could not remove file \"%s\": %m",
2046 : : path)));
2047 : 0 : continue;
2048 : : }
2049 : : }
2050 : : }
4205 rhaas@postgresql.org 2051 :CBC 1677 : FreeDir(snap_dir);
2052 : 1677 : }
2053 : :
2054 : : /*
2055 : : * Check if a logical snapshot at the specified point has been serialized.
2056 : : */
2057 : : bool
521 akapila@postgresql.o 2058 : 12 : SnapBuildSnapshotExists(XLogRecPtr lsn)
2059 : : {
2060 : : char path[MAXPGPATH];
2061 : : int ret;
2062 : : struct stat stat_buf;
2063 : :
23 fujii@postgresql.org 2064 : 12 : sprintf(path, "%s/%X-%X.snap",
2065 : : PG_LOGICAL_SNAPSHOTS_DIR,
521 akapila@postgresql.o 2066 : 12 : LSN_FORMAT_ARGS(lsn));
2067 : :
2068 : 12 : ret = stat(path, &stat_buf);
2069 : :
2070 [ + + - + ]: 12 : if (ret != 0 && errno != ENOENT)
521 akapila@postgresql.o 2071 [ # # ]:UBC 0 : ereport(ERROR,
2072 : : (errcode_for_file_access(),
2073 : : errmsg("could not stat file \"%s\": %m", path)));
2074 : :
521 akapila@postgresql.o 2075 :CBC 12 : return ret == 0;
2076 : : }
|
