Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * reorderbuffer.c
4 : * PostgreSQL logical replay/reorder buffer management
5 : *
6 : *
7 : * Copyright (c) 2012-2025, PostgreSQL Global Development Group
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/replication/logical/reorderbuffer.c
12 : *
13 : * NOTES
14 : * This module gets handed individual pieces of transactions in the order
15 : * they are written to the WAL and is responsible to reassemble them into
16 : * toplevel transaction sized pieces. When a transaction is completely
17 : * reassembled - signaled by reading the transaction commit record - it
18 : * will then call the output plugin (cf. ReorderBufferCommit()) with the
19 : * individual changes. The output plugins rely on snapshots built by
20 : * snapbuild.c which hands them to us.
21 : *
22 : * Transactions and subtransactions/savepoints in postgres are not
23 : * immediately linked to each other from outside the performing
24 : * backend. Only at commit/abort (or special xact_assignment records) they
25 : * are linked together. Which means that we will have to splice together a
26 : * toplevel transaction from its subtransactions. To do that efficiently we
27 : * build a binary heap indexed by the smallest current lsn of the individual
28 : * subtransactions' changestreams. As the individual streams are inherently
29 : * ordered by LSN - since that is where we build them from - the transaction
30 : * can easily be reassembled by always using the subtransaction with the
31 : * smallest current LSN from the heap.
32 : *
33 : * In order to cope with large transactions - which can be several times as
34 : * big as the available memory - this module supports spooling the contents
35 : * of large transactions to disk. When the transaction is replayed the
36 : * contents of individual (sub-)transactions will be read from disk in
37 : * chunks.
38 : *
39 : * This module also has to deal with reassembling toast records from the
40 : * individual chunks stored in WAL. When a new (or initial) version of a
41 : * tuple is stored in WAL it will always be preceded by the toast chunks
42 : * emitted for the columns stored out of line. Within a single toplevel
43 : * transaction there will be no other data carrying records between a row's
44 : * toast chunks and the row data itself. See ReorderBufferToast* for
45 : * details.
46 : *
47 : * ReorderBuffer uses two special memory context types - SlabContext for
48 : * allocations of fixed-length structures (changes and transactions), and
49 : * GenerationContext for the variable-length transaction data (allocated
50 : * and freed in groups with similar lifespans).
51 : *
52 : * To limit the amount of memory used by decoded changes, we track memory
53 : * used at the reorder buffer level (i.e. total amount of memory), and for
54 : * each transaction. When the total amount of used memory exceeds the
55 : * limit, the transaction consuming the most memory is then serialized to
56 : * disk.
57 : *
58 : * Only decoded changes are evicted from memory (spilled to disk), not the
59 : * transaction records. The number of toplevel transactions is limited,
60 : * but a transaction with many subtransactions may still consume significant
61 : * amounts of memory. However, the transaction records are fairly small and
62 : * are not included in the memory limit.
63 : *
64 : * The current eviction algorithm is very simple - the transaction is
65 : * picked merely by size, while it might be useful to also consider age
66 : * (LSN) of the changes for example. With the new Generational memory
67 : * allocator, evicting the oldest changes would make it more likely the
68 : * memory gets actually freed.
69 : *
70 : * We use a max-heap with transaction size as the key to efficiently find
71 : * the largest transaction. We update the max-heap whenever the memory
72 : * counter is updated; however transactions with size 0 are not stored in
73 : * the heap, because they have no changes to evict.
74 : *
75 : * We still rely on max_changes_in_memory when loading serialized changes
76 : * back into memory. At that point we can't use the memory limit directly
77 : * as we load the subxacts independently. One option to deal with this
78 : * would be to count the subxacts, and allow each to allocate 1/N of the
79 : * memory limit. That however does not seem very appealing, because with
80 : * many subtransactions it may easily cause thrashing (short cycles of
81 : * deserializing and applying very few changes). We probably should give
82 : * a bit more memory to the oldest subtransactions, because it's likely
83 : * they are the source for the next sequence of changes.
84 : *
85 : * -------------------------------------------------------------------------
86 : */
87 : #include "postgres.h"
88 :
89 : #include <unistd.h>
90 : #include <sys/stat.h>
91 :
92 : #include "access/detoast.h"
93 : #include "access/heapam.h"
94 : #include "access/rewriteheap.h"
95 : #include "access/transam.h"
96 : #include "access/xact.h"
97 : #include "access/xlog_internal.h"
98 : #include "catalog/catalog.h"
99 : #include "common/int.h"
100 : #include "lib/binaryheap.h"
101 : #include "miscadmin.h"
102 : #include "pgstat.h"
103 : #include "replication/logical.h"
104 : #include "replication/reorderbuffer.h"
105 : #include "replication/slot.h"
106 : #include "replication/snapbuild.h" /* just for SnapBuildSnapDecRefcount */
107 : #include "storage/bufmgr.h"
108 : #include "storage/fd.h"
109 : #include "storage/procarray.h"
110 : #include "storage/sinval.h"
111 : #include "utils/builtins.h"
112 : #include "utils/inval.h"
113 : #include "utils/memutils.h"
114 : #include "utils/rel.h"
115 : #include "utils/relfilenumbermap.h"
116 :
117 : /*
118 : * Each transaction has an 8MB limit for invalidation messages distributed from
119 : * other transactions. This limit is set considering scenarios with many
120 : * concurrent logical decoding operations. When the distributed invalidation
121 : * messages reach this threshold, the transaction is marked as
122 : * RBTXN_DISTR_INVAL_OVERFLOWED to invalidate the complete cache as we have lost
123 : * some inval messages and hence don't know what needs to be invalidated.
124 : */
125 : #define MAX_DISTR_INVAL_MSG_PER_TXN \
126 : ((8 * 1024 * 1024) / sizeof(SharedInvalidationMessage))
127 :
128 : /* entry for a hash table we use to map from xid to our transaction state */
129 : typedef struct ReorderBufferTXNByIdEnt
130 : {
131 : TransactionId xid;
132 : ReorderBufferTXN *txn;
133 : } ReorderBufferTXNByIdEnt;
134 :
135 : /* data structures for (relfilelocator, ctid) => (cmin, cmax) mapping */
136 : typedef struct ReorderBufferTupleCidKey
137 : {
138 : RelFileLocator rlocator;
139 : ItemPointerData tid;
140 : } ReorderBufferTupleCidKey;
141 :
142 : typedef struct ReorderBufferTupleCidEnt
143 : {
144 : ReorderBufferTupleCidKey key;
145 : CommandId cmin;
146 : CommandId cmax;
147 : CommandId combocid; /* just for debugging */
148 : } ReorderBufferTupleCidEnt;
149 :
150 : /* Virtual file descriptor with file offset tracking */
151 : typedef struct TXNEntryFile
152 : {
153 : File vfd; /* -1 when the file is closed */
154 : off_t curOffset; /* offset for next write or read. Reset to 0
155 : * when vfd is opened. */
156 : } TXNEntryFile;
157 :
158 : /* k-way in-order change iteration support structures */
159 : typedef struct ReorderBufferIterTXNEntry
160 : {
161 : XLogRecPtr lsn;
162 : ReorderBufferChange *change;
163 : ReorderBufferTXN *txn;
164 : TXNEntryFile file;
165 : XLogSegNo segno;
166 : } ReorderBufferIterTXNEntry;
167 :
168 : typedef struct ReorderBufferIterTXNState
169 : {
170 : binaryheap *heap;
171 : Size nr_txns;
172 : dlist_head old_change;
173 : ReorderBufferIterTXNEntry entries[FLEXIBLE_ARRAY_MEMBER];
174 : } ReorderBufferIterTXNState;
175 :
176 : /* toast datastructures */
177 : typedef struct ReorderBufferToastEnt
178 : {
179 : Oid chunk_id; /* toast_table.chunk_id */
180 : int32 last_chunk_seq; /* toast_table.chunk_seq of the last chunk we
181 : * have seen */
182 : Size num_chunks; /* number of chunks we've already seen */
183 : Size size; /* combined size of chunks seen */
184 : dlist_head chunks; /* linked list of chunks */
185 : struct varlena *reconstructed; /* reconstructed varlena now pointed to in
186 : * main tup */
187 : } ReorderBufferToastEnt;
188 :
189 : /* Disk serialization support datastructures */
190 : typedef struct ReorderBufferDiskChange
191 : {
192 : Size size;
193 : ReorderBufferChange change;
194 : /* data follows */
195 : } ReorderBufferDiskChange;
196 :
197 : #define IsSpecInsert(action) \
198 : ( \
199 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT) \
200 : )
201 : #define IsSpecConfirmOrAbort(action) \
202 : ( \
203 : (((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM) || \
204 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT)) \
205 : )
206 : #define IsInsertOrUpdate(action) \
207 : ( \
208 : (((action) == REORDER_BUFFER_CHANGE_INSERT) || \
209 : ((action) == REORDER_BUFFER_CHANGE_UPDATE) || \
210 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT)) \
211 : )
212 :
213 : /*
214 : * Maximum number of changes kept in memory, per transaction. After that,
215 : * changes are spooled to disk.
216 : *
217 : * The current value should be sufficient to decode the entire transaction
218 : * without hitting disk in OLTP workloads, while starting to spool to disk in
219 : * other workloads reasonably fast.
220 : *
221 : * At some point in the future it probably makes sense to have a more elaborate
222 : * resource management here, but it's not entirely clear what that would look
223 : * like.
224 : */
225 : int logical_decoding_work_mem;
226 : static const Size max_changes_in_memory = 4096; /* XXX for restore only */
227 :
228 : /* GUC variable */
229 : int debug_logical_replication_streaming = DEBUG_LOGICAL_REP_STREAMING_BUFFERED;
230 :
231 : /* ---------------------------------------
232 : * primary reorderbuffer support routines
233 : * ---------------------------------------
234 : */
235 : static ReorderBufferTXN *ReorderBufferAllocTXN(ReorderBuffer *rb);
236 : static void ReorderBufferFreeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
237 : static ReorderBufferTXN *ReorderBufferTXNByXid(ReorderBuffer *rb,
238 : TransactionId xid, bool create, bool *is_new,
239 : XLogRecPtr lsn, bool create_as_top);
240 : static void ReorderBufferTransferSnapToParent(ReorderBufferTXN *txn,
241 : ReorderBufferTXN *subtxn);
242 :
243 : static void AssertTXNLsnOrder(ReorderBuffer *rb);
244 :
245 : /* ---------------------------------------
246 : * support functions for lsn-order iterating over the ->changes of a
247 : * transaction and its subtransactions
248 : *
249 : * used for iteration over the k-way heap merge of a transaction and its
250 : * subtransactions
251 : * ---------------------------------------
252 : */
253 : static void ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn,
254 : ReorderBufferIterTXNState *volatile *iter_state);
255 : static ReorderBufferChange *ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state);
256 : static void ReorderBufferIterTXNFinish(ReorderBuffer *rb,
257 : ReorderBufferIterTXNState *state);
258 : static void ReorderBufferExecuteInvalidations(uint32 nmsgs, SharedInvalidationMessage *msgs);
259 :
260 : /*
261 : * ---------------------------------------
262 : * Disk serialization support functions
263 : * ---------------------------------------
264 : */
265 : static void ReorderBufferCheckMemoryLimit(ReorderBuffer *rb);
266 : static void ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
267 : static void ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
268 : int fd, ReorderBufferChange *change);
269 : static Size ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
270 : TXNEntryFile *file, XLogSegNo *segno);
271 : static void ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
272 : char *data);
273 : static void ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn);
274 : static void ReorderBufferTruncateTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
275 : bool txn_prepared);
276 : static void ReorderBufferMaybeMarkTXNStreamed(ReorderBuffer *rb, ReorderBufferTXN *txn);
277 : static bool ReorderBufferCheckAndTruncateAbortedTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
278 : static void ReorderBufferCleanupSerializedTXNs(const char *slotname);
279 : static void ReorderBufferSerializedPath(char *path, ReplicationSlot *slot,
280 : TransactionId xid, XLogSegNo segno);
281 : static int ReorderBufferTXNSizeCompare(const pairingheap_node *a, const pairingheap_node *b, void *arg);
282 :
283 : static void ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap);
284 : static Snapshot ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
285 : ReorderBufferTXN *txn, CommandId cid);
286 :
287 : /*
288 : * ---------------------------------------
289 : * Streaming support functions
290 : * ---------------------------------------
291 : */
292 : static inline bool ReorderBufferCanStream(ReorderBuffer *rb);
293 : static inline bool ReorderBufferCanStartStreaming(ReorderBuffer *rb);
294 : static void ReorderBufferStreamTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
295 : static void ReorderBufferStreamCommit(ReorderBuffer *rb, ReorderBufferTXN *txn);
296 :
297 : /* ---------------------------------------
298 : * toast reassembly support
299 : * ---------------------------------------
300 : */
301 : static void ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn);
302 : static void ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn);
303 : static void ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
304 : Relation relation, ReorderBufferChange *change);
305 : static void ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
306 : Relation relation, ReorderBufferChange *change);
307 :
308 : /*
309 : * ---------------------------------------
310 : * memory accounting
311 : * ---------------------------------------
312 : */
313 : static Size ReorderBufferChangeSize(ReorderBufferChange *change);
314 : static void ReorderBufferChangeMemoryUpdate(ReorderBuffer *rb,
315 : ReorderBufferChange *change,
316 : ReorderBufferTXN *txn,
317 : bool addition, Size sz);
318 :
319 : /*
320 : * Allocate a new ReorderBuffer and clean out any old serialized state from
321 : * prior ReorderBuffer instances for the same slot.
322 : */
323 : ReorderBuffer *
324 2166 : ReorderBufferAllocate(void)
325 : {
326 : ReorderBuffer *buffer;
327 : HASHCTL hash_ctl;
328 : MemoryContext new_ctx;
329 :
330 : Assert(MyReplicationSlot != NULL);
331 :
332 : /* allocate memory in own context, to have better accountability */
333 2166 : new_ctx = AllocSetContextCreate(CurrentMemoryContext,
334 : "ReorderBuffer",
335 : ALLOCSET_DEFAULT_SIZES);
336 :
337 : buffer =
338 2166 : (ReorderBuffer *) MemoryContextAlloc(new_ctx, sizeof(ReorderBuffer));
339 :
340 2166 : memset(&hash_ctl, 0, sizeof(hash_ctl));
341 :
342 2166 : buffer->context = new_ctx;
343 :
344 2166 : buffer->change_context = SlabContextCreate(new_ctx,
345 : "Change",
346 : SLAB_DEFAULT_BLOCK_SIZE,
347 : sizeof(ReorderBufferChange));
348 :
349 2166 : buffer->txn_context = SlabContextCreate(new_ctx,
350 : "TXN",
351 : SLAB_DEFAULT_BLOCK_SIZE,
352 : sizeof(ReorderBufferTXN));
353 :
354 : /*
355 : * To minimize memory fragmentation caused by long-running transactions
356 : * with changes spanning multiple memory blocks, we use a single
357 : * fixed-size memory block for decoded tuple storage. The performance
358 : * testing showed that the default memory block size maintains logical
359 : * decoding performance without causing fragmentation due to concurrent
360 : * transactions. One might think that we can use the max size as
361 : * SLAB_LARGE_BLOCK_SIZE but the test also showed it doesn't help resolve
362 : * the memory fragmentation.
363 : */
364 2166 : buffer->tup_context = GenerationContextCreate(new_ctx,
365 : "Tuples",
366 : SLAB_DEFAULT_BLOCK_SIZE,
367 : SLAB_DEFAULT_BLOCK_SIZE,
368 : SLAB_DEFAULT_BLOCK_SIZE);
369 :
370 2166 : hash_ctl.keysize = sizeof(TransactionId);
371 2166 : hash_ctl.entrysize = sizeof(ReorderBufferTXNByIdEnt);
372 2166 : hash_ctl.hcxt = buffer->context;
373 :
374 2166 : buffer->by_txn = hash_create("ReorderBufferByXid", 1000, &hash_ctl,
375 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
376 :
377 2166 : buffer->by_txn_last_xid = InvalidTransactionId;
378 2166 : buffer->by_txn_last_txn = NULL;
379 :
380 2166 : buffer->outbuf = NULL;
381 2166 : buffer->outbufsize = 0;
382 2166 : buffer->size = 0;
383 :
384 : /* txn_heap is ordered by transaction size */
385 2166 : buffer->txn_heap = pairingheap_allocate(ReorderBufferTXNSizeCompare, NULL);
386 :
387 2166 : buffer->spillTxns = 0;
388 2166 : buffer->spillCount = 0;
389 2166 : buffer->spillBytes = 0;
390 2166 : buffer->streamTxns = 0;
391 2166 : buffer->streamCount = 0;
392 2166 : buffer->streamBytes = 0;
393 2166 : buffer->totalTxns = 0;
394 2166 : buffer->totalBytes = 0;
395 :
396 2166 : buffer->current_restart_decoding_lsn = InvalidXLogRecPtr;
397 :
398 2166 : dlist_init(&buffer->toplevel_by_lsn);
399 2166 : dlist_init(&buffer->txns_by_base_snapshot_lsn);
400 2166 : dclist_init(&buffer->catchange_txns);
401 :
402 : /*
403 : * Ensure there's no stale data from prior uses of this slot, in case some
404 : * prior exit avoided calling ReorderBufferFree. Failure to do this can
405 : * produce duplicated txns, and it's very cheap if there's nothing there.
406 : */
407 2166 : ReorderBufferCleanupSerializedTXNs(NameStr(MyReplicationSlot->data.name));
408 :
409 2166 : return buffer;
410 : }
411 :
412 : /*
413 : * Free a ReorderBuffer
414 : */
415 : void
416 1740 : ReorderBufferFree(ReorderBuffer *rb)
417 : {
418 1740 : MemoryContext context = rb->context;
419 :
420 : /*
421 : * We free separately allocated data by entirely scrapping reorderbuffer's
422 : * memory context.
423 : */
424 1740 : MemoryContextDelete(context);
425 :
426 : /* Free disk space used by unconsumed reorder buffers */
427 1740 : ReorderBufferCleanupSerializedTXNs(NameStr(MyReplicationSlot->data.name));
428 1740 : }
429 :
430 : /*
431 : * Allocate a new ReorderBufferTXN.
432 : */
433 : static ReorderBufferTXN *
434 7686 : ReorderBufferAllocTXN(ReorderBuffer *rb)
435 : {
436 : ReorderBufferTXN *txn;
437 :
438 : txn = (ReorderBufferTXN *)
439 7686 : MemoryContextAlloc(rb->txn_context, sizeof(ReorderBufferTXN));
440 :
441 7686 : memset(txn, 0, sizeof(ReorderBufferTXN));
442 :
443 7686 : dlist_init(&txn->changes);
444 7686 : dlist_init(&txn->tuplecids);
445 7686 : dlist_init(&txn->subtxns);
446 :
447 : /* InvalidCommandId is not zero, so set it explicitly */
448 7686 : txn->command_id = InvalidCommandId;
449 7686 : txn->output_plugin_private = NULL;
450 :
451 7686 : return txn;
452 : }
453 :
454 : /*
455 : * Free a ReorderBufferTXN.
456 : */
457 : static void
458 7538 : ReorderBufferFreeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
459 : {
460 : /* clean the lookup cache if we were cached (quite likely) */
461 7538 : if (rb->by_txn_last_xid == txn->xid)
462 : {
463 7160 : rb->by_txn_last_xid = InvalidTransactionId;
464 7160 : rb->by_txn_last_txn = NULL;
465 : }
466 :
467 : /* free data that's contained */
468 :
469 7538 : if (txn->gid != NULL)
470 : {
471 72 : pfree(txn->gid);
472 72 : txn->gid = NULL;
473 : }
474 :
475 7538 : if (txn->tuplecid_hash != NULL)
476 : {
477 1064 : hash_destroy(txn->tuplecid_hash);
478 1064 : txn->tuplecid_hash = NULL;
479 : }
480 :
481 7538 : if (txn->invalidations)
482 : {
483 2246 : pfree(txn->invalidations);
484 2246 : txn->invalidations = NULL;
485 : }
486 :
487 7538 : if (txn->invalidations_distributed)
488 : {
489 42 : pfree(txn->invalidations_distributed);
490 42 : txn->invalidations_distributed = NULL;
491 : }
492 :
493 : /* Reset the toast hash */
494 7538 : ReorderBufferToastReset(rb, txn);
495 :
496 : /* All changes must be deallocated */
497 : Assert(txn->size == 0);
498 :
499 7538 : pfree(txn);
500 7538 : }
501 :
502 : /*
503 : * Allocate a ReorderBufferChange.
504 : */
505 : ReorderBufferChange *
506 3478902 : ReorderBufferAllocChange(ReorderBuffer *rb)
507 : {
508 : ReorderBufferChange *change;
509 :
510 : change = (ReorderBufferChange *)
511 3478902 : MemoryContextAlloc(rb->change_context, sizeof(ReorderBufferChange));
512 :
513 3478902 : memset(change, 0, sizeof(ReorderBufferChange));
514 3478902 : return change;
515 : }
516 :
517 : /*
518 : * Free a ReorderBufferChange and update memory accounting, if requested.
519 : */
520 : void
521 3478400 : ReorderBufferFreeChange(ReorderBuffer *rb, ReorderBufferChange *change,
522 : bool upd_mem)
523 : {
524 : /* update memory accounting info */
525 3478400 : if (upd_mem)
526 396362 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, false,
527 : ReorderBufferChangeSize(change));
528 :
529 : /* free contained data */
530 3478400 : switch (change->action)
531 : {
532 3327712 : case REORDER_BUFFER_CHANGE_INSERT:
533 : case REORDER_BUFFER_CHANGE_UPDATE:
534 : case REORDER_BUFFER_CHANGE_DELETE:
535 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
536 3327712 : if (change->data.tp.newtuple)
537 : {
538 2898454 : ReorderBufferFreeTupleBuf(change->data.tp.newtuple);
539 2898454 : change->data.tp.newtuple = NULL;
540 : }
541 :
542 3327712 : if (change->data.tp.oldtuple)
543 : {
544 292312 : ReorderBufferFreeTupleBuf(change->data.tp.oldtuple);
545 292312 : change->data.tp.oldtuple = NULL;
546 : }
547 3327712 : break;
548 80 : case REORDER_BUFFER_CHANGE_MESSAGE:
549 80 : if (change->data.msg.prefix != NULL)
550 80 : pfree(change->data.msg.prefix);
551 80 : change->data.msg.prefix = NULL;
552 80 : if (change->data.msg.message != NULL)
553 80 : pfree(change->data.msg.message);
554 80 : change->data.msg.message = NULL;
555 80 : break;
556 10358 : case REORDER_BUFFER_CHANGE_INVALIDATION:
557 10358 : if (change->data.inval.invalidations)
558 10358 : pfree(change->data.inval.invalidations);
559 10358 : change->data.inval.invalidations = NULL;
560 10358 : break;
561 2318 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
562 2318 : if (change->data.snapshot)
563 : {
564 2318 : ReorderBufferFreeSnap(rb, change->data.snapshot);
565 2318 : change->data.snapshot = NULL;
566 : }
567 2318 : break;
568 : /* no data in addition to the struct itself */
569 100 : case REORDER_BUFFER_CHANGE_TRUNCATE:
570 100 : if (change->data.truncate.relids != NULL)
571 : {
572 100 : ReorderBufferFreeRelids(rb, change->data.truncate.relids);
573 100 : change->data.truncate.relids = NULL;
574 : }
575 100 : break;
576 137832 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
577 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
578 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
579 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
580 137832 : break;
581 : }
582 :
583 3478400 : pfree(change);
584 3478400 : }
585 :
586 : /*
587 : * Allocate a HeapTuple fitting a tuple of size tuple_len (excluding header
588 : * overhead).
589 : */
590 : HeapTuple
591 3190890 : ReorderBufferAllocTupleBuf(ReorderBuffer *rb, Size tuple_len)
592 : {
593 : HeapTuple tuple;
594 : Size alloc_len;
595 :
596 3190890 : alloc_len = tuple_len + SizeofHeapTupleHeader;
597 :
598 3190890 : tuple = (HeapTuple) MemoryContextAlloc(rb->tup_context,
599 : HEAPTUPLESIZE + alloc_len);
600 3190890 : tuple->t_data = (HeapTupleHeader) ((char *) tuple + HEAPTUPLESIZE);
601 :
602 3190890 : return tuple;
603 : }
604 :
605 : /*
606 : * Free a HeapTuple returned by ReorderBufferAllocTupleBuf().
607 : */
608 : void
609 3190766 : ReorderBufferFreeTupleBuf(HeapTuple tuple)
610 : {
611 3190766 : pfree(tuple);
612 3190766 : }
613 :
614 : /*
615 : * Allocate an array for relids of truncated relations.
616 : *
617 : * We use the global memory context (for the whole reorder buffer), because
618 : * none of the existing ones seems like a good match (some are SLAB, so we
619 : * can't use those, and tup_context is meant for tuple data, not relids). We
620 : * could add yet another context, but it seems like an overkill - TRUNCATE is
621 : * not particularly common operation, so it does not seem worth it.
622 : */
623 : Oid *
624 110 : ReorderBufferAllocRelids(ReorderBuffer *rb, int nrelids)
625 : {
626 : Oid *relids;
627 : Size alloc_len;
628 :
629 110 : alloc_len = sizeof(Oid) * nrelids;
630 :
631 110 : relids = (Oid *) MemoryContextAlloc(rb->context, alloc_len);
632 :
633 110 : return relids;
634 : }
635 :
636 : /*
637 : * Free an array of relids.
638 : */
639 : void
640 100 : ReorderBufferFreeRelids(ReorderBuffer *rb, Oid *relids)
641 : {
642 100 : pfree(relids);
643 100 : }
644 :
645 : /*
646 : * Return the ReorderBufferTXN from the given buffer, specified by Xid.
647 : * If create is true, and a transaction doesn't already exist, create it
648 : * (with the given LSN, and as top transaction if that's specified);
649 : * when this happens, is_new is set to true.
650 : */
651 : static ReorderBufferTXN *
652 11586540 : ReorderBufferTXNByXid(ReorderBuffer *rb, TransactionId xid, bool create,
653 : bool *is_new, XLogRecPtr lsn, bool create_as_top)
654 : {
655 : ReorderBufferTXN *txn;
656 : ReorderBufferTXNByIdEnt *ent;
657 : bool found;
658 :
659 : Assert(TransactionIdIsValid(xid));
660 :
661 : /*
662 : * Check the one-entry lookup cache first
663 : */
664 11586540 : if (TransactionIdIsValid(rb->by_txn_last_xid) &&
665 11579274 : rb->by_txn_last_xid == xid)
666 : {
667 9599488 : txn = rb->by_txn_last_txn;
668 :
669 9599488 : if (txn != NULL)
670 : {
671 : /* found it, and it's valid */
672 9599462 : if (is_new)
673 6140 : *is_new = false;
674 9599462 : return txn;
675 : }
676 :
677 : /*
678 : * cached as non-existent, and asked not to create? Then nothing else
679 : * to do.
680 : */
681 26 : if (!create)
682 20 : return NULL;
683 : /* otherwise fall through to create it */
684 : }
685 :
686 : /*
687 : * If the cache wasn't hit or it yielded a "does-not-exist" and we want to
688 : * create an entry.
689 : */
690 :
691 : /* search the lookup table */
692 : ent = (ReorderBufferTXNByIdEnt *)
693 1987058 : hash_search(rb->by_txn,
694 : &xid,
695 : create ? HASH_ENTER : HASH_FIND,
696 : &found);
697 1987058 : if (found)
698 1976804 : txn = ent->txn;
699 10254 : else if (create)
700 : {
701 : /* initialize the new entry, if creation was requested */
702 : Assert(ent != NULL);
703 : Assert(lsn != InvalidXLogRecPtr);
704 :
705 7686 : ent->txn = ReorderBufferAllocTXN(rb);
706 7686 : ent->txn->xid = xid;
707 7686 : txn = ent->txn;
708 7686 : txn->first_lsn = lsn;
709 7686 : txn->restart_decoding_lsn = rb->current_restart_decoding_lsn;
710 :
711 7686 : if (create_as_top)
712 : {
713 6320 : dlist_push_tail(&rb->toplevel_by_lsn, &txn->node);
714 6320 : AssertTXNLsnOrder(rb);
715 : }
716 : }
717 : else
718 2568 : txn = NULL; /* not found and not asked to create */
719 :
720 : /* update cache */
721 1987058 : rb->by_txn_last_xid = xid;
722 1987058 : rb->by_txn_last_txn = txn;
723 :
724 1987058 : if (is_new)
725 3590 : *is_new = !found;
726 :
727 : Assert(!create || txn != NULL);
728 1987058 : return txn;
729 : }
730 :
731 : /*
732 : * Record the partial change for the streaming of in-progress transactions. We
733 : * can stream only complete changes so if we have a partial change like toast
734 : * table insert or speculative insert then we mark such a 'txn' so that it
735 : * can't be streamed. We also ensure that if the changes in such a 'txn' can
736 : * be streamed and are above logical_decoding_work_mem threshold then we stream
737 : * them as soon as we have a complete change.
738 : */
739 : static void
740 3063516 : ReorderBufferProcessPartialChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
741 : ReorderBufferChange *change,
742 : bool toast_insert)
743 : {
744 : ReorderBufferTXN *toptxn;
745 :
746 : /*
747 : * The partial changes need to be processed only while streaming
748 : * in-progress transactions.
749 : */
750 3063516 : if (!ReorderBufferCanStream(rb))
751 2420420 : return;
752 :
753 : /* Get the top transaction. */
754 643096 : toptxn = rbtxn_get_toptxn(txn);
755 :
756 : /*
757 : * Indicate a partial change for toast inserts. The change will be
758 : * considered as complete once we get the insert or update on the main
759 : * table and we are sure that the pending toast chunks are not required
760 : * anymore.
761 : *
762 : * If we allow streaming when there are pending toast chunks then such
763 : * chunks won't be released till the insert (multi_insert) is complete and
764 : * we expect the txn to have streamed all changes after streaming. This
765 : * restriction is mainly to ensure the correctness of streamed
766 : * transactions and it doesn't seem worth uplifting such a restriction
767 : * just to allow this case because anyway we will stream the transaction
768 : * once such an insert is complete.
769 : */
770 643096 : if (toast_insert)
771 3332 : toptxn->txn_flags |= RBTXN_HAS_PARTIAL_CHANGE;
772 639764 : else if (rbtxn_has_partial_change(toptxn) &&
773 126 : IsInsertOrUpdate(change->action) &&
774 126 : change->data.tp.clear_toast_afterwards)
775 86 : toptxn->txn_flags &= ~RBTXN_HAS_PARTIAL_CHANGE;
776 :
777 : /*
778 : * Indicate a partial change for speculative inserts. The change will be
779 : * considered as complete once we get the speculative confirm or abort
780 : * token.
781 : */
782 643096 : if (IsSpecInsert(change->action))
783 0 : toptxn->txn_flags |= RBTXN_HAS_PARTIAL_CHANGE;
784 643096 : else if (rbtxn_has_partial_change(toptxn) &&
785 3372 : IsSpecConfirmOrAbort(change->action))
786 0 : toptxn->txn_flags &= ~RBTXN_HAS_PARTIAL_CHANGE;
787 :
788 : /*
789 : * Stream the transaction if it is serialized before and the changes are
790 : * now complete in the top-level transaction.
791 : *
792 : * The reason for doing the streaming of such a transaction as soon as we
793 : * get the complete change for it is that previously it would have reached
794 : * the memory threshold and wouldn't get streamed because of incomplete
795 : * changes. Delaying such transactions would increase apply lag for them.
796 : */
797 643096 : if (ReorderBufferCanStartStreaming(rb) &&
798 359098 : !(rbtxn_has_partial_change(toptxn)) &&
799 356026 : rbtxn_is_serialized(txn) &&
800 76 : rbtxn_has_streamable_change(toptxn))
801 16 : ReorderBufferStreamTXN(rb, toptxn);
802 : }
803 :
804 : /*
805 : * Queue a change into a transaction so it can be replayed upon commit or will be
806 : * streamed when we reach logical_decoding_work_mem threshold.
807 : */
808 : void
809 3082334 : ReorderBufferQueueChange(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
810 : ReorderBufferChange *change, bool toast_insert)
811 : {
812 : ReorderBufferTXN *txn;
813 :
814 3082334 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
815 :
816 : /*
817 : * If we have detected that the transaction is aborted while streaming the
818 : * previous changes or by checking its CLOG, there is no point in
819 : * collecting further changes for it.
820 : */
821 3082334 : if (rbtxn_is_aborted(txn))
822 : {
823 : /*
824 : * We don't need to update memory accounting for this change as we
825 : * have not added it to the queue yet.
826 : */
827 18818 : ReorderBufferFreeChange(rb, change, false);
828 18818 : return;
829 : }
830 :
831 : /*
832 : * The changes that are sent downstream are considered streamable. We
833 : * remember such transactions so that only those will later be considered
834 : * for streaming.
835 : */
836 3063516 : if (change->action == REORDER_BUFFER_CHANGE_INSERT ||
837 858358 : change->action == REORDER_BUFFER_CHANGE_UPDATE ||
838 538794 : change->action == REORDER_BUFFER_CHANGE_DELETE ||
839 133540 : change->action == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT ||
840 97708 : change->action == REORDER_BUFFER_CHANGE_TRUNCATE ||
841 97618 : change->action == REORDER_BUFFER_CHANGE_MESSAGE)
842 : {
843 2965976 : ReorderBufferTXN *toptxn = rbtxn_get_toptxn(txn);
844 :
845 2965976 : toptxn->txn_flags |= RBTXN_HAS_STREAMABLE_CHANGE;
846 : }
847 :
848 3063516 : change->lsn = lsn;
849 3063516 : change->txn = txn;
850 :
851 : Assert(InvalidXLogRecPtr != lsn);
852 3063516 : dlist_push_tail(&txn->changes, &change->node);
853 3063516 : txn->nentries++;
854 3063516 : txn->nentries_mem++;
855 :
856 : /* update memory accounting information */
857 3063516 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
858 : ReorderBufferChangeSize(change));
859 :
860 : /* process partial change */
861 3063516 : ReorderBufferProcessPartialChange(rb, txn, change, toast_insert);
862 :
863 : /* check the memory limits and evict something if needed */
864 3063516 : ReorderBufferCheckMemoryLimit(rb);
865 : }
866 :
867 : /*
868 : * A transactional message is queued to be processed upon commit and a
869 : * non-transactional message gets processed immediately.
870 : */
871 : void
872 94 : ReorderBufferQueueMessage(ReorderBuffer *rb, TransactionId xid,
873 : Snapshot snap, XLogRecPtr lsn,
874 : bool transactional, const char *prefix,
875 : Size message_size, const char *message)
876 : {
877 94 : if (transactional)
878 : {
879 : MemoryContext oldcontext;
880 : ReorderBufferChange *change;
881 :
882 : Assert(xid != InvalidTransactionId);
883 :
884 : /*
885 : * We don't expect snapshots for transactional changes - we'll use the
886 : * snapshot derived later during apply (unless the change gets
887 : * skipped).
888 : */
889 : Assert(!snap);
890 :
891 78 : oldcontext = MemoryContextSwitchTo(rb->context);
892 :
893 78 : change = ReorderBufferAllocChange(rb);
894 78 : change->action = REORDER_BUFFER_CHANGE_MESSAGE;
895 78 : change->data.msg.prefix = pstrdup(prefix);
896 78 : change->data.msg.message_size = message_size;
897 78 : change->data.msg.message = palloc(message_size);
898 78 : memcpy(change->data.msg.message, message, message_size);
899 :
900 78 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
901 :
902 78 : MemoryContextSwitchTo(oldcontext);
903 : }
904 : else
905 : {
906 16 : ReorderBufferTXN *txn = NULL;
907 16 : volatile Snapshot snapshot_now = snap;
908 :
909 : /* Non-transactional changes require a valid snapshot. */
910 : Assert(snapshot_now);
911 :
912 16 : if (xid != InvalidTransactionId)
913 6 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
914 :
915 : /* setup snapshot to allow catalog access */
916 16 : SetupHistoricSnapshot(snapshot_now, NULL);
917 16 : PG_TRY();
918 : {
919 16 : rb->message(rb, txn, lsn, false, prefix, message_size, message);
920 :
921 16 : TeardownHistoricSnapshot(false);
922 : }
923 0 : PG_CATCH();
924 : {
925 0 : TeardownHistoricSnapshot(true);
926 0 : PG_RE_THROW();
927 : }
928 16 : PG_END_TRY();
929 : }
930 94 : }
931 :
932 : /*
933 : * AssertTXNLsnOrder
934 : * Verify LSN ordering of transaction lists in the reorderbuffer
935 : *
936 : * Other LSN-related invariants are checked too.
937 : *
938 : * No-op if assertions are not in use.
939 : */
940 : static void
941 15546 : AssertTXNLsnOrder(ReorderBuffer *rb)
942 : {
943 : #ifdef USE_ASSERT_CHECKING
944 : LogicalDecodingContext *ctx = rb->private_data;
945 : dlist_iter iter;
946 : XLogRecPtr prev_first_lsn = InvalidXLogRecPtr;
947 : XLogRecPtr prev_base_snap_lsn = InvalidXLogRecPtr;
948 :
949 : /*
950 : * Skip the verification if we don't reach the LSN at which we start
951 : * decoding the contents of transactions yet because until we reach the
952 : * LSN, we could have transactions that don't have the association between
953 : * the top-level transaction and subtransaction yet and consequently have
954 : * the same LSN. We don't guarantee this association until we try to
955 : * decode the actual contents of transaction. The ordering of the records
956 : * prior to the start_decoding_at LSN should have been checked before the
957 : * restart.
958 : */
959 : if (SnapBuildXactNeedsSkip(ctx->snapshot_builder, ctx->reader->EndRecPtr))
960 : return;
961 :
962 : dlist_foreach(iter, &rb->toplevel_by_lsn)
963 : {
964 : ReorderBufferTXN *cur_txn = dlist_container(ReorderBufferTXN, node,
965 : iter.cur);
966 :
967 : /* start LSN must be set */
968 : Assert(cur_txn->first_lsn != InvalidXLogRecPtr);
969 :
970 : /* If there is an end LSN, it must be higher than start LSN */
971 : if (cur_txn->end_lsn != InvalidXLogRecPtr)
972 : Assert(cur_txn->first_lsn <= cur_txn->end_lsn);
973 :
974 : /* Current initial LSN must be strictly higher than previous */
975 : if (prev_first_lsn != InvalidXLogRecPtr)
976 : Assert(prev_first_lsn < cur_txn->first_lsn);
977 :
978 : /* known-as-subtxn txns must not be listed */
979 : Assert(!rbtxn_is_known_subxact(cur_txn));
980 :
981 : prev_first_lsn = cur_txn->first_lsn;
982 : }
983 :
984 : dlist_foreach(iter, &rb->txns_by_base_snapshot_lsn)
985 : {
986 : ReorderBufferTXN *cur_txn = dlist_container(ReorderBufferTXN,
987 : base_snapshot_node,
988 : iter.cur);
989 :
990 : /* base snapshot (and its LSN) must be set */
991 : Assert(cur_txn->base_snapshot != NULL);
992 : Assert(cur_txn->base_snapshot_lsn != InvalidXLogRecPtr);
993 :
994 : /* current LSN must be strictly higher than previous */
995 : if (prev_base_snap_lsn != InvalidXLogRecPtr)
996 : Assert(prev_base_snap_lsn < cur_txn->base_snapshot_lsn);
997 :
998 : /* known-as-subtxn txns must not be listed */
999 : Assert(!rbtxn_is_known_subxact(cur_txn));
1000 :
1001 : prev_base_snap_lsn = cur_txn->base_snapshot_lsn;
1002 : }
1003 : #endif
1004 15546 : }
1005 :
1006 : /*
1007 : * AssertChangeLsnOrder
1008 : *
1009 : * Check ordering of changes in the (sub)transaction.
1010 : */
1011 : static void
1012 4832 : AssertChangeLsnOrder(ReorderBufferTXN *txn)
1013 : {
1014 : #ifdef USE_ASSERT_CHECKING
1015 : dlist_iter iter;
1016 : XLogRecPtr prev_lsn = txn->first_lsn;
1017 :
1018 : dlist_foreach(iter, &txn->changes)
1019 : {
1020 : ReorderBufferChange *cur_change;
1021 :
1022 : cur_change = dlist_container(ReorderBufferChange, node, iter.cur);
1023 :
1024 : Assert(txn->first_lsn != InvalidXLogRecPtr);
1025 : Assert(cur_change->lsn != InvalidXLogRecPtr);
1026 : Assert(txn->first_lsn <= cur_change->lsn);
1027 :
1028 : if (txn->end_lsn != InvalidXLogRecPtr)
1029 : Assert(cur_change->lsn <= txn->end_lsn);
1030 :
1031 : Assert(prev_lsn <= cur_change->lsn);
1032 :
1033 : prev_lsn = cur_change->lsn;
1034 : }
1035 : #endif
1036 4832 : }
1037 :
1038 : /*
1039 : * ReorderBufferGetOldestTXN
1040 : * Return oldest transaction in reorderbuffer
1041 : */
1042 : ReorderBufferTXN *
1043 784 : ReorderBufferGetOldestTXN(ReorderBuffer *rb)
1044 : {
1045 : ReorderBufferTXN *txn;
1046 :
1047 784 : AssertTXNLsnOrder(rb);
1048 :
1049 784 : if (dlist_is_empty(&rb->toplevel_by_lsn))
1050 666 : return NULL;
1051 :
1052 118 : txn = dlist_head_element(ReorderBufferTXN, node, &rb->toplevel_by_lsn);
1053 :
1054 : Assert(!rbtxn_is_known_subxact(txn));
1055 : Assert(txn->first_lsn != InvalidXLogRecPtr);
1056 118 : return txn;
1057 : }
1058 :
1059 : /*
1060 : * ReorderBufferGetOldestXmin
1061 : * Return oldest Xmin in reorderbuffer
1062 : *
1063 : * Returns oldest possibly running Xid from the point of view of snapshots
1064 : * used in the transactions kept by reorderbuffer, or InvalidTransactionId if
1065 : * there are none.
1066 : *
1067 : * Since snapshots are assigned monotonically, this equals the Xmin of the
1068 : * base snapshot with minimal base_snapshot_lsn.
1069 : */
1070 : TransactionId
1071 822 : ReorderBufferGetOldestXmin(ReorderBuffer *rb)
1072 : {
1073 : ReorderBufferTXN *txn;
1074 :
1075 822 : AssertTXNLsnOrder(rb);
1076 :
1077 822 : if (dlist_is_empty(&rb->txns_by_base_snapshot_lsn))
1078 724 : return InvalidTransactionId;
1079 :
1080 98 : txn = dlist_head_element(ReorderBufferTXN, base_snapshot_node,
1081 : &rb->txns_by_base_snapshot_lsn);
1082 98 : return txn->base_snapshot->xmin;
1083 : }
1084 :
1085 : void
1086 868 : ReorderBufferSetRestartPoint(ReorderBuffer *rb, XLogRecPtr ptr)
1087 : {
1088 868 : rb->current_restart_decoding_lsn = ptr;
1089 868 : }
1090 :
1091 : /*
1092 : * ReorderBufferAssignChild
1093 : *
1094 : * Make note that we know that subxid is a subtransaction of xid, seen as of
1095 : * the given lsn.
1096 : */
1097 : void
1098 1738 : ReorderBufferAssignChild(ReorderBuffer *rb, TransactionId xid,
1099 : TransactionId subxid, XLogRecPtr lsn)
1100 : {
1101 : ReorderBufferTXN *txn;
1102 : ReorderBufferTXN *subtxn;
1103 : bool new_top;
1104 : bool new_sub;
1105 :
1106 1738 : txn = ReorderBufferTXNByXid(rb, xid, true, &new_top, lsn, true);
1107 1738 : subtxn = ReorderBufferTXNByXid(rb, subxid, true, &new_sub, lsn, false);
1108 :
1109 1738 : if (!new_sub)
1110 : {
1111 372 : if (rbtxn_is_known_subxact(subtxn))
1112 : {
1113 : /* already associated, nothing to do */
1114 372 : return;
1115 : }
1116 : else
1117 : {
1118 : /*
1119 : * We already saw this transaction, but initially added it to the
1120 : * list of top-level txns. Now that we know it's not top-level,
1121 : * remove it from there.
1122 : */
1123 0 : dlist_delete(&subtxn->node);
1124 : }
1125 : }
1126 :
1127 1366 : subtxn->txn_flags |= RBTXN_IS_SUBXACT;
1128 1366 : subtxn->toplevel_xid = xid;
1129 : Assert(subtxn->nsubtxns == 0);
1130 :
1131 : /* set the reference to top-level transaction */
1132 1366 : subtxn->toptxn = txn;
1133 :
1134 : /* add to subtransaction list */
1135 1366 : dlist_push_tail(&txn->subtxns, &subtxn->node);
1136 1366 : txn->nsubtxns++;
1137 :
1138 : /* Possibly transfer the subtxn's snapshot to its top-level txn. */
1139 1366 : ReorderBufferTransferSnapToParent(txn, subtxn);
1140 :
1141 : /* Verify LSN-ordering invariant */
1142 1366 : AssertTXNLsnOrder(rb);
1143 : }
1144 :
1145 : /*
1146 : * ReorderBufferTransferSnapToParent
1147 : * Transfer base snapshot from subtxn to top-level txn, if needed
1148 : *
1149 : * This is done if the top-level txn doesn't have a base snapshot, or if the
1150 : * subtxn's base snapshot has an earlier LSN than the top-level txn's base
1151 : * snapshot's LSN. This can happen if there are no changes in the toplevel
1152 : * txn but there are some in the subtxn, or the first change in subtxn has
1153 : * earlier LSN than first change in the top-level txn and we learned about
1154 : * their kinship only now.
1155 : *
1156 : * The subtransaction's snapshot is cleared regardless of the transfer
1157 : * happening, since it's not needed anymore in either case.
1158 : *
1159 : * We do this as soon as we become aware of their kinship, to avoid queueing
1160 : * extra snapshots to txns known-as-subtxns -- only top-level txns will
1161 : * receive further snapshots.
1162 : */
1163 : static void
1164 1374 : ReorderBufferTransferSnapToParent(ReorderBufferTXN *txn,
1165 : ReorderBufferTXN *subtxn)
1166 : {
1167 : Assert(subtxn->toplevel_xid == txn->xid);
1168 :
1169 1374 : if (subtxn->base_snapshot != NULL)
1170 : {
1171 0 : if (txn->base_snapshot == NULL ||
1172 0 : subtxn->base_snapshot_lsn < txn->base_snapshot_lsn)
1173 : {
1174 : /*
1175 : * If the toplevel transaction already has a base snapshot but
1176 : * it's newer than the subxact's, purge it.
1177 : */
1178 0 : if (txn->base_snapshot != NULL)
1179 : {
1180 0 : SnapBuildSnapDecRefcount(txn->base_snapshot);
1181 0 : dlist_delete(&txn->base_snapshot_node);
1182 : }
1183 :
1184 : /*
1185 : * The snapshot is now the top transaction's; transfer it, and
1186 : * adjust the list position of the top transaction in the list by
1187 : * moving it to where the subtransaction is.
1188 : */
1189 0 : txn->base_snapshot = subtxn->base_snapshot;
1190 0 : txn->base_snapshot_lsn = subtxn->base_snapshot_lsn;
1191 0 : dlist_insert_before(&subtxn->base_snapshot_node,
1192 : &txn->base_snapshot_node);
1193 :
1194 : /*
1195 : * The subtransaction doesn't have a snapshot anymore (so it
1196 : * mustn't be in the list.)
1197 : */
1198 0 : subtxn->base_snapshot = NULL;
1199 0 : subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
1200 0 : dlist_delete(&subtxn->base_snapshot_node);
1201 : }
1202 : else
1203 : {
1204 : /* Base snap of toplevel is fine, so subxact's is not needed */
1205 0 : SnapBuildSnapDecRefcount(subtxn->base_snapshot);
1206 0 : dlist_delete(&subtxn->base_snapshot_node);
1207 0 : subtxn->base_snapshot = NULL;
1208 0 : subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
1209 : }
1210 : }
1211 1374 : }
1212 :
1213 : /*
1214 : * Associate a subtransaction with its toplevel transaction at commit
1215 : * time. There may be no further changes added after this.
1216 : */
1217 : void
1218 534 : ReorderBufferCommitChild(ReorderBuffer *rb, TransactionId xid,
1219 : TransactionId subxid, XLogRecPtr commit_lsn,
1220 : XLogRecPtr end_lsn)
1221 : {
1222 : ReorderBufferTXN *subtxn;
1223 :
1224 534 : subtxn = ReorderBufferTXNByXid(rb, subxid, false, NULL,
1225 : InvalidXLogRecPtr, false);
1226 :
1227 : /*
1228 : * No need to do anything if that subtxn didn't contain any changes
1229 : */
1230 534 : if (!subtxn)
1231 162 : return;
1232 :
1233 372 : subtxn->final_lsn = commit_lsn;
1234 372 : subtxn->end_lsn = end_lsn;
1235 :
1236 : /*
1237 : * Assign this subxact as a child of the toplevel xact (no-op if already
1238 : * done.)
1239 : */
1240 372 : ReorderBufferAssignChild(rb, xid, subxid, InvalidXLogRecPtr);
1241 : }
1242 :
1243 :
1244 : /*
1245 : * Support for efficiently iterating over a transaction's and its
1246 : * subtransactions' changes.
1247 : *
1248 : * We do by doing a k-way merge between transactions/subtransactions. For that
1249 : * we model the current heads of the different transactions as a binary heap
1250 : * so we easily know which (sub-)transaction has the change with the smallest
1251 : * lsn next.
1252 : *
1253 : * We assume the changes in individual transactions are already sorted by LSN.
1254 : */
1255 :
1256 : /*
1257 : * Binary heap comparison function.
1258 : */
1259 : static int
1260 103136 : ReorderBufferIterCompare(Datum a, Datum b, void *arg)
1261 : {
1262 103136 : ReorderBufferIterTXNState *state = (ReorderBufferIterTXNState *) arg;
1263 103136 : XLogRecPtr pos_a = state->entries[DatumGetInt32(a)].lsn;
1264 103136 : XLogRecPtr pos_b = state->entries[DatumGetInt32(b)].lsn;
1265 :
1266 103136 : if (pos_a < pos_b)
1267 101424 : return 1;
1268 1712 : else if (pos_a == pos_b)
1269 0 : return 0;
1270 1712 : return -1;
1271 : }
1272 :
1273 : /*
1274 : * Allocate & initialize an iterator which iterates in lsn order over a
1275 : * transaction and all its subtransactions.
1276 : *
1277 : * Note: The iterator state is returned through iter_state parameter rather
1278 : * than the function's return value. This is because the state gets cleaned up
1279 : * in a PG_CATCH block in the caller, so we want to make sure the caller gets
1280 : * back the state even if this function throws an exception.
1281 : */
1282 : static void
1283 3906 : ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn,
1284 : ReorderBufferIterTXNState *volatile *iter_state)
1285 : {
1286 3906 : Size nr_txns = 0;
1287 : ReorderBufferIterTXNState *state;
1288 : dlist_iter cur_txn_i;
1289 : int32 off;
1290 :
1291 3906 : *iter_state = NULL;
1292 :
1293 : /* Check ordering of changes in the toplevel transaction. */
1294 3906 : AssertChangeLsnOrder(txn);
1295 :
1296 : /*
1297 : * Calculate the size of our heap: one element for every transaction that
1298 : * contains changes. (Besides the transactions already in the reorder
1299 : * buffer, we count the one we were directly passed.)
1300 : */
1301 3906 : if (txn->nentries > 0)
1302 3554 : nr_txns++;
1303 :
1304 4832 : dlist_foreach(cur_txn_i, &txn->subtxns)
1305 : {
1306 : ReorderBufferTXN *cur_txn;
1307 :
1308 926 : cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
1309 :
1310 : /* Check ordering of changes in this subtransaction. */
1311 926 : AssertChangeLsnOrder(cur_txn);
1312 :
1313 926 : if (cur_txn->nentries > 0)
1314 602 : nr_txns++;
1315 : }
1316 :
1317 : /* allocate iteration state */
1318 : state = (ReorderBufferIterTXNState *)
1319 3906 : MemoryContextAllocZero(rb->context,
1320 : sizeof(ReorderBufferIterTXNState) +
1321 3906 : sizeof(ReorderBufferIterTXNEntry) * nr_txns);
1322 :
1323 3906 : state->nr_txns = nr_txns;
1324 3906 : dlist_init(&state->old_change);
1325 :
1326 8062 : for (off = 0; off < state->nr_txns; off++)
1327 : {
1328 4156 : state->entries[off].file.vfd = -1;
1329 4156 : state->entries[off].segno = 0;
1330 : }
1331 :
1332 : /* allocate heap */
1333 3906 : state->heap = binaryheap_allocate(state->nr_txns,
1334 : ReorderBufferIterCompare,
1335 : state);
1336 :
1337 : /* Now that the state fields are initialized, it is safe to return it. */
1338 3906 : *iter_state = state;
1339 :
1340 : /*
1341 : * Now insert items into the binary heap, in an unordered fashion. (We
1342 : * will run a heap assembly step at the end; this is more efficient.)
1343 : */
1344 :
1345 3906 : off = 0;
1346 :
1347 : /* add toplevel transaction if it contains changes */
1348 3906 : if (txn->nentries > 0)
1349 : {
1350 : ReorderBufferChange *cur_change;
1351 :
1352 3554 : if (rbtxn_is_serialized(txn))
1353 : {
1354 : /* serialize remaining changes */
1355 44 : ReorderBufferSerializeTXN(rb, txn);
1356 44 : ReorderBufferRestoreChanges(rb, txn, &state->entries[off].file,
1357 : &state->entries[off].segno);
1358 : }
1359 :
1360 3554 : cur_change = dlist_head_element(ReorderBufferChange, node,
1361 : &txn->changes);
1362 :
1363 3554 : state->entries[off].lsn = cur_change->lsn;
1364 3554 : state->entries[off].change = cur_change;
1365 3554 : state->entries[off].txn = txn;
1366 :
1367 3554 : binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
1368 : }
1369 :
1370 : /* add subtransactions if they contain changes */
1371 4832 : dlist_foreach(cur_txn_i, &txn->subtxns)
1372 : {
1373 : ReorderBufferTXN *cur_txn;
1374 :
1375 926 : cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
1376 :
1377 926 : if (cur_txn->nentries > 0)
1378 : {
1379 : ReorderBufferChange *cur_change;
1380 :
1381 602 : if (rbtxn_is_serialized(cur_txn))
1382 : {
1383 : /* serialize remaining changes */
1384 34 : ReorderBufferSerializeTXN(rb, cur_txn);
1385 34 : ReorderBufferRestoreChanges(rb, cur_txn,
1386 : &state->entries[off].file,
1387 : &state->entries[off].segno);
1388 : }
1389 602 : cur_change = dlist_head_element(ReorderBufferChange, node,
1390 : &cur_txn->changes);
1391 :
1392 602 : state->entries[off].lsn = cur_change->lsn;
1393 602 : state->entries[off].change = cur_change;
1394 602 : state->entries[off].txn = cur_txn;
1395 :
1396 602 : binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
1397 : }
1398 : }
1399 :
1400 : /* assemble a valid binary heap */
1401 3906 : binaryheap_build(state->heap);
1402 3906 : }
1403 :
1404 : /*
1405 : * Return the next change when iterating over a transaction and its
1406 : * subtransactions.
1407 : *
1408 : * Returns NULL when no further changes exist.
1409 : */
1410 : static ReorderBufferChange *
1411 728086 : ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state)
1412 : {
1413 : ReorderBufferChange *change;
1414 : ReorderBufferIterTXNEntry *entry;
1415 : int32 off;
1416 :
1417 : /* nothing there anymore */
1418 728086 : if (binaryheap_empty(state->heap))
1419 3888 : return NULL;
1420 :
1421 724198 : off = DatumGetInt32(binaryheap_first(state->heap));
1422 724198 : entry = &state->entries[off];
1423 :
1424 : /* free memory we might have "leaked" in the previous *Next call */
1425 724198 : if (!dlist_is_empty(&state->old_change))
1426 : {
1427 88 : change = dlist_container(ReorderBufferChange, node,
1428 : dlist_pop_head_node(&state->old_change));
1429 88 : ReorderBufferFreeChange(rb, change, true);
1430 : Assert(dlist_is_empty(&state->old_change));
1431 : }
1432 :
1433 724198 : change = entry->change;
1434 :
1435 : /*
1436 : * update heap with information about which transaction has the next
1437 : * relevant change in LSN order
1438 : */
1439 :
1440 : /* there are in-memory changes */
1441 724198 : if (dlist_has_next(&entry->txn->changes, &entry->change->node))
1442 : {
1443 719978 : dlist_node *next = dlist_next_node(&entry->txn->changes, &change->node);
1444 719978 : ReorderBufferChange *next_change =
1445 719978 : dlist_container(ReorderBufferChange, node, next);
1446 :
1447 : /* txn stays the same */
1448 719978 : state->entries[off].lsn = next_change->lsn;
1449 719978 : state->entries[off].change = next_change;
1450 :
1451 719978 : binaryheap_replace_first(state->heap, Int32GetDatum(off));
1452 719978 : return change;
1453 : }
1454 :
1455 : /* try to load changes from disk */
1456 4220 : if (entry->txn->nentries != entry->txn->nentries_mem)
1457 : {
1458 : /*
1459 : * Ugly: restoring changes will reuse *Change records, thus delete the
1460 : * current one from the per-tx list and only free in the next call.
1461 : */
1462 126 : dlist_delete(&change->node);
1463 126 : dlist_push_tail(&state->old_change, &change->node);
1464 :
1465 : /*
1466 : * Update the total bytes processed by the txn for which we are
1467 : * releasing the current set of changes and restoring the new set of
1468 : * changes.
1469 : */
1470 126 : rb->totalBytes += entry->txn->size;
1471 126 : if (ReorderBufferRestoreChanges(rb, entry->txn, &entry->file,
1472 : &state->entries[off].segno))
1473 : {
1474 : /* successfully restored changes from disk */
1475 : ReorderBufferChange *next_change =
1476 70 : dlist_head_element(ReorderBufferChange, node,
1477 : &entry->txn->changes);
1478 :
1479 70 : elog(DEBUG2, "restored %u/%u changes from disk",
1480 : (uint32) entry->txn->nentries_mem,
1481 : (uint32) entry->txn->nentries);
1482 :
1483 : Assert(entry->txn->nentries_mem);
1484 : /* txn stays the same */
1485 70 : state->entries[off].lsn = next_change->lsn;
1486 70 : state->entries[off].change = next_change;
1487 70 : binaryheap_replace_first(state->heap, Int32GetDatum(off));
1488 :
1489 70 : return change;
1490 : }
1491 : }
1492 :
1493 : /* ok, no changes there anymore, remove */
1494 4150 : binaryheap_remove_first(state->heap);
1495 :
1496 4150 : return change;
1497 : }
1498 :
1499 : /*
1500 : * Deallocate the iterator
1501 : */
1502 : static void
1503 3906 : ReorderBufferIterTXNFinish(ReorderBuffer *rb,
1504 : ReorderBufferIterTXNState *state)
1505 : {
1506 : int32 off;
1507 :
1508 8062 : for (off = 0; off < state->nr_txns; off++)
1509 : {
1510 4156 : if (state->entries[off].file.vfd != -1)
1511 0 : FileClose(state->entries[off].file.vfd);
1512 : }
1513 :
1514 : /* free memory we might have "leaked" in the last *Next call */
1515 3906 : if (!dlist_is_empty(&state->old_change))
1516 : {
1517 : ReorderBufferChange *change;
1518 :
1519 36 : change = dlist_container(ReorderBufferChange, node,
1520 : dlist_pop_head_node(&state->old_change));
1521 36 : ReorderBufferFreeChange(rb, change, true);
1522 : Assert(dlist_is_empty(&state->old_change));
1523 : }
1524 :
1525 3906 : binaryheap_free(state->heap);
1526 3906 : pfree(state);
1527 3906 : }
1528 :
1529 : /*
1530 : * Cleanup the contents of a transaction, usually after the transaction
1531 : * committed or aborted.
1532 : */
1533 : static void
1534 7538 : ReorderBufferCleanupTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
1535 : {
1536 : bool found;
1537 : dlist_mutable_iter iter;
1538 7538 : Size mem_freed = 0;
1539 :
1540 : /* cleanup subtransactions & their changes */
1541 7908 : dlist_foreach_modify(iter, &txn->subtxns)
1542 : {
1543 : ReorderBufferTXN *subtxn;
1544 :
1545 370 : subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1546 :
1547 : /*
1548 : * Subtransactions are always associated to the toplevel TXN, even if
1549 : * they originally were happening inside another subtxn, so we won't
1550 : * ever recurse more than one level deep here.
1551 : */
1552 : Assert(rbtxn_is_known_subxact(subtxn));
1553 : Assert(subtxn->nsubtxns == 0);
1554 :
1555 370 : ReorderBufferCleanupTXN(rb, subtxn);
1556 : }
1557 :
1558 : /* cleanup changes in the txn */
1559 164656 : dlist_foreach_modify(iter, &txn->changes)
1560 : {
1561 : ReorderBufferChange *change;
1562 :
1563 157118 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1564 :
1565 : /* Check we're not mixing changes from different transactions. */
1566 : Assert(change->txn == txn);
1567 :
1568 : /*
1569 : * Instead of updating the memory counter for individual changes, we
1570 : * sum up the size of memory to free so we can update the memory
1571 : * counter all together below. This saves costs of maintaining the
1572 : * max-heap.
1573 : */
1574 157118 : mem_freed += ReorderBufferChangeSize(change);
1575 :
1576 157118 : ReorderBufferFreeChange(rb, change, false);
1577 : }
1578 :
1579 : /* Update the memory counter */
1580 7538 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, mem_freed);
1581 :
1582 : /*
1583 : * Cleanup the tuplecids we stored for decoding catalog snapshot access.
1584 : * They are always stored in the toplevel transaction.
1585 : */
1586 56310 : dlist_foreach_modify(iter, &txn->tuplecids)
1587 : {
1588 : ReorderBufferChange *change;
1589 :
1590 48772 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1591 :
1592 : /* Check we're not mixing changes from different transactions. */
1593 : Assert(change->txn == txn);
1594 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1595 :
1596 48772 : ReorderBufferFreeChange(rb, change, true);
1597 : }
1598 :
1599 : /*
1600 : * Cleanup the base snapshot, if set.
1601 : */
1602 7538 : if (txn->base_snapshot != NULL)
1603 : {
1604 6126 : SnapBuildSnapDecRefcount(txn->base_snapshot);
1605 6126 : dlist_delete(&txn->base_snapshot_node);
1606 : }
1607 :
1608 : /*
1609 : * Cleanup the snapshot for the last streamed run.
1610 : */
1611 7538 : if (txn->snapshot_now != NULL)
1612 : {
1613 : Assert(rbtxn_is_streamed(txn));
1614 120 : ReorderBufferFreeSnap(rb, txn->snapshot_now);
1615 : }
1616 :
1617 : /*
1618 : * Remove TXN from its containing lists.
1619 : *
1620 : * Note: if txn is known as subxact, we are deleting the TXN from its
1621 : * parent's list of known subxacts; this leaves the parent's nsubxacts
1622 : * count too high, but we don't care. Otherwise, we are deleting the TXN
1623 : * from the LSN-ordered list of toplevel TXNs. We remove the TXN from the
1624 : * list of catalog modifying transactions as well.
1625 : */
1626 7538 : dlist_delete(&txn->node);
1627 7538 : if (rbtxn_has_catalog_changes(txn))
1628 2380 : dclist_delete_from(&rb->catchange_txns, &txn->catchange_node);
1629 :
1630 : /* now remove reference from buffer */
1631 7538 : hash_search(rb->by_txn, &txn->xid, HASH_REMOVE, &found);
1632 : Assert(found);
1633 :
1634 : /* remove entries spilled to disk */
1635 7538 : if (rbtxn_is_serialized(txn))
1636 530 : ReorderBufferRestoreCleanup(rb, txn);
1637 :
1638 : /* deallocate */
1639 7538 : ReorderBufferFreeTXN(rb, txn);
1640 7538 : }
1641 :
1642 : /*
1643 : * Discard changes from a transaction (and subtransactions), either after
1644 : * streaming, decoding them at PREPARE, or detecting the transaction abort.
1645 : * Keep the remaining info - transactions, tuplecids, invalidations and
1646 : * snapshots.
1647 : *
1648 : * We additionally remove tuplecids after decoding the transaction at prepare
1649 : * time as we only need to perform invalidation at rollback or commit prepared.
1650 : *
1651 : * 'txn_prepared' indicates that we have decoded the transaction at prepare
1652 : * time.
1653 : */
1654 : static void
1655 2036 : ReorderBufferTruncateTXN(ReorderBuffer *rb, ReorderBufferTXN *txn, bool txn_prepared)
1656 : {
1657 : dlist_mutable_iter iter;
1658 2036 : Size mem_freed = 0;
1659 :
1660 : /* cleanup subtransactions & their changes */
1661 2630 : dlist_foreach_modify(iter, &txn->subtxns)
1662 : {
1663 : ReorderBufferTXN *subtxn;
1664 :
1665 594 : subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1666 :
1667 : /*
1668 : * Subtransactions are always associated to the toplevel TXN, even if
1669 : * they originally were happening inside another subtxn, so we won't
1670 : * ever recurse more than one level deep here.
1671 : */
1672 : Assert(rbtxn_is_known_subxact(subtxn));
1673 : Assert(subtxn->nsubtxns == 0);
1674 :
1675 594 : ReorderBufferMaybeMarkTXNStreamed(rb, subtxn);
1676 594 : ReorderBufferTruncateTXN(rb, subtxn, txn_prepared);
1677 : }
1678 :
1679 : /* cleanup changes in the txn */
1680 326848 : dlist_foreach_modify(iter, &txn->changes)
1681 : {
1682 : ReorderBufferChange *change;
1683 :
1684 324812 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1685 :
1686 : /* Check we're not mixing changes from different transactions. */
1687 : Assert(change->txn == txn);
1688 :
1689 : /* remove the change from its containing list */
1690 324812 : dlist_delete(&change->node);
1691 :
1692 : /*
1693 : * Instead of updating the memory counter for individual changes, we
1694 : * sum up the size of memory to free so we can update the memory
1695 : * counter all together below. This saves costs of maintaining the
1696 : * max-heap.
1697 : */
1698 324812 : mem_freed += ReorderBufferChangeSize(change);
1699 :
1700 324812 : ReorderBufferFreeChange(rb, change, false);
1701 : }
1702 :
1703 : /* Update the memory counter */
1704 2036 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, mem_freed);
1705 :
1706 2036 : if (txn_prepared)
1707 : {
1708 : /*
1709 : * If this is a prepared txn, cleanup the tuplecids we stored for
1710 : * decoding catalog snapshot access. They are always stored in the
1711 : * toplevel transaction.
1712 : */
1713 338 : dlist_foreach_modify(iter, &txn->tuplecids)
1714 : {
1715 : ReorderBufferChange *change;
1716 :
1717 246 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1718 :
1719 : /* Check we're not mixing changes from different transactions. */
1720 : Assert(change->txn == txn);
1721 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1722 :
1723 : /* Remove the change from its containing list. */
1724 246 : dlist_delete(&change->node);
1725 :
1726 246 : ReorderBufferFreeChange(rb, change, true);
1727 : }
1728 : }
1729 :
1730 : /*
1731 : * Destroy the (relfilelocator, ctid) hashtable, so that we don't leak any
1732 : * memory. We could also keep the hash table and update it with new ctid
1733 : * values, but this seems simpler and good enough for now.
1734 : */
1735 2036 : if (txn->tuplecid_hash != NULL)
1736 : {
1737 102 : hash_destroy(txn->tuplecid_hash);
1738 102 : txn->tuplecid_hash = NULL;
1739 : }
1740 :
1741 : /* If this txn is serialized then clean the disk space. */
1742 2036 : if (rbtxn_is_serialized(txn))
1743 : {
1744 16 : ReorderBufferRestoreCleanup(rb, txn);
1745 16 : txn->txn_flags &= ~RBTXN_IS_SERIALIZED;
1746 :
1747 : /*
1748 : * We set this flag to indicate if the transaction is ever serialized.
1749 : * We need this to accurately update the stats as otherwise the same
1750 : * transaction can be counted as serialized multiple times.
1751 : */
1752 16 : txn->txn_flags |= RBTXN_IS_SERIALIZED_CLEAR;
1753 : }
1754 :
1755 : /* also reset the number of entries in the transaction */
1756 2036 : txn->nentries_mem = 0;
1757 2036 : txn->nentries = 0;
1758 2036 : }
1759 :
1760 : /*
1761 : * Check the transaction status by CLOG lookup and discard all changes if
1762 : * the transaction is aborted. The transaction status is cached in
1763 : * txn->txn_flags so we can skip future changes and avoid CLOG lookups on the
1764 : * next call.
1765 : *
1766 : * Return true if the transaction is aborted, otherwise return false.
1767 : *
1768 : * When the 'debug_logical_replication_streaming' is set to "immediate", we
1769 : * don't check the transaction status, meaning the caller will always process
1770 : * this transaction.
1771 : */
1772 : static bool
1773 8570 : ReorderBufferCheckAndTruncateAbortedTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
1774 : {
1775 : /* Quick return for regression tests */
1776 8570 : if (unlikely(debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_IMMEDIATE))
1777 1764 : return false;
1778 :
1779 : /*
1780 : * Quick return if the transaction status is already known.
1781 : */
1782 :
1783 6806 : if (rbtxn_is_committed(txn))
1784 5808 : return false;
1785 998 : if (rbtxn_is_aborted(txn))
1786 : {
1787 : /* Already-aborted transactions should not have any changes */
1788 : Assert(txn->size == 0);
1789 :
1790 0 : return true;
1791 : }
1792 :
1793 : /* Otherwise, check the transaction status using CLOG lookup */
1794 :
1795 998 : if (TransactionIdIsInProgress(txn->xid))
1796 518 : return false;
1797 :
1798 480 : if (TransactionIdDidCommit(txn->xid))
1799 : {
1800 : /*
1801 : * Remember the transaction is committed so that we can skip CLOG
1802 : * check next time, avoiding the pressure on CLOG lookup.
1803 : */
1804 : Assert(!rbtxn_is_aborted(txn));
1805 462 : txn->txn_flags |= RBTXN_IS_COMMITTED;
1806 462 : return false;
1807 : }
1808 :
1809 : /*
1810 : * The transaction aborted. We discard both the changes collected so far
1811 : * and the toast reconstruction data. The full cleanup will happen as part
1812 : * of decoding ABORT record of this transaction.
1813 : */
1814 18 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
1815 18 : ReorderBufferToastReset(rb, txn);
1816 :
1817 : /* All changes should be discarded */
1818 : Assert(txn->size == 0);
1819 :
1820 : /*
1821 : * Mark the transaction as aborted so we can ignore future changes of this
1822 : * transaction.
1823 : */
1824 : Assert(!rbtxn_is_committed(txn));
1825 18 : txn->txn_flags |= RBTXN_IS_ABORTED;
1826 :
1827 18 : return true;
1828 : }
1829 :
1830 : /*
1831 : * Build a hash with a (relfilelocator, ctid) -> (cmin, cmax) mapping for use by
1832 : * HeapTupleSatisfiesHistoricMVCC.
1833 : */
1834 : static void
1835 3906 : ReorderBufferBuildTupleCidHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
1836 : {
1837 : dlist_iter iter;
1838 : HASHCTL hash_ctl;
1839 :
1840 3906 : if (!rbtxn_has_catalog_changes(txn) || dlist_is_empty(&txn->tuplecids))
1841 2740 : return;
1842 :
1843 1166 : hash_ctl.keysize = sizeof(ReorderBufferTupleCidKey);
1844 1166 : hash_ctl.entrysize = sizeof(ReorderBufferTupleCidEnt);
1845 1166 : hash_ctl.hcxt = rb->context;
1846 :
1847 : /*
1848 : * create the hash with the exact number of to-be-stored tuplecids from
1849 : * the start
1850 : */
1851 1166 : txn->tuplecid_hash =
1852 1166 : hash_create("ReorderBufferTupleCid", txn->ntuplecids, &hash_ctl,
1853 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1854 :
1855 24798 : dlist_foreach(iter, &txn->tuplecids)
1856 : {
1857 : ReorderBufferTupleCidKey key;
1858 : ReorderBufferTupleCidEnt *ent;
1859 : bool found;
1860 : ReorderBufferChange *change;
1861 :
1862 23632 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1863 :
1864 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1865 :
1866 : /* be careful about padding */
1867 23632 : memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
1868 :
1869 23632 : key.rlocator = change->data.tuplecid.locator;
1870 :
1871 23632 : ItemPointerCopy(&change->data.tuplecid.tid,
1872 : &key.tid);
1873 :
1874 : ent = (ReorderBufferTupleCidEnt *)
1875 23632 : hash_search(txn->tuplecid_hash, &key, HASH_ENTER, &found);
1876 23632 : if (!found)
1877 : {
1878 20346 : ent->cmin = change->data.tuplecid.cmin;
1879 20346 : ent->cmax = change->data.tuplecid.cmax;
1880 20346 : ent->combocid = change->data.tuplecid.combocid;
1881 : }
1882 : else
1883 : {
1884 : /*
1885 : * Maybe we already saw this tuple before in this transaction, but
1886 : * if so it must have the same cmin.
1887 : */
1888 : Assert(ent->cmin == change->data.tuplecid.cmin);
1889 :
1890 : /*
1891 : * cmax may be initially invalid, but once set it can only grow,
1892 : * and never become invalid again.
1893 : */
1894 : Assert((ent->cmax == InvalidCommandId) ||
1895 : ((change->data.tuplecid.cmax != InvalidCommandId) &&
1896 : (change->data.tuplecid.cmax > ent->cmax)));
1897 3286 : ent->cmax = change->data.tuplecid.cmax;
1898 : }
1899 : }
1900 : }
1901 :
1902 : /*
1903 : * Copy a provided snapshot so we can modify it privately. This is needed so
1904 : * that catalog modifying transactions can look into intermediate catalog
1905 : * states.
1906 : */
1907 : static Snapshot
1908 3500 : ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
1909 : ReorderBufferTXN *txn, CommandId cid)
1910 : {
1911 : Snapshot snap;
1912 : dlist_iter iter;
1913 3500 : int i = 0;
1914 : Size size;
1915 :
1916 3500 : size = sizeof(SnapshotData) +
1917 3500 : sizeof(TransactionId) * orig_snap->xcnt +
1918 3500 : sizeof(TransactionId) * (txn->nsubtxns + 1);
1919 :
1920 3500 : snap = MemoryContextAllocZero(rb->context, size);
1921 3500 : memcpy(snap, orig_snap, sizeof(SnapshotData));
1922 :
1923 3500 : snap->copied = true;
1924 3500 : snap->active_count = 1; /* mark as active so nobody frees it */
1925 3500 : snap->regd_count = 0;
1926 3500 : snap->xip = (TransactionId *) (snap + 1);
1927 :
1928 3500 : memcpy(snap->xip, orig_snap->xip, sizeof(TransactionId) * snap->xcnt);
1929 :
1930 : /*
1931 : * snap->subxip contains all txids that belong to our transaction which we
1932 : * need to check via cmin/cmax. That's why we store the toplevel
1933 : * transaction in there as well.
1934 : */
1935 3500 : snap->subxip = snap->xip + snap->xcnt;
1936 3500 : snap->subxip[i++] = txn->xid;
1937 :
1938 : /*
1939 : * txn->nsubtxns isn't decreased when subtransactions abort, so count
1940 : * manually. Since it's an upper boundary it is safe to use it for the
1941 : * allocation above.
1942 : */
1943 3500 : snap->subxcnt = 1;
1944 :
1945 4118 : dlist_foreach(iter, &txn->subtxns)
1946 : {
1947 : ReorderBufferTXN *sub_txn;
1948 :
1949 618 : sub_txn = dlist_container(ReorderBufferTXN, node, iter.cur);
1950 618 : snap->subxip[i++] = sub_txn->xid;
1951 618 : snap->subxcnt++;
1952 : }
1953 :
1954 : /* sort so we can bsearch() later */
1955 3500 : qsort(snap->subxip, snap->subxcnt, sizeof(TransactionId), xidComparator);
1956 :
1957 : /* store the specified current CommandId */
1958 3500 : snap->curcid = cid;
1959 :
1960 3500 : return snap;
1961 : }
1962 :
1963 : /*
1964 : * Free a previously ReorderBufferCopySnap'ed snapshot
1965 : */
1966 : static void
1967 5810 : ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap)
1968 : {
1969 5810 : if (snap->copied)
1970 3496 : pfree(snap);
1971 : else
1972 2314 : SnapBuildSnapDecRefcount(snap);
1973 5810 : }
1974 :
1975 : /*
1976 : * If the transaction was (partially) streamed, we need to prepare or commit
1977 : * it in a 'streamed' way. That is, we first stream the remaining part of the
1978 : * transaction, and then invoke stream_prepare or stream_commit message as per
1979 : * the case.
1980 : */
1981 : static void
1982 116 : ReorderBufferStreamCommit(ReorderBuffer *rb, ReorderBufferTXN *txn)
1983 : {
1984 : /* we should only call this for previously streamed transactions */
1985 : Assert(rbtxn_is_streamed(txn));
1986 :
1987 116 : ReorderBufferStreamTXN(rb, txn);
1988 :
1989 116 : if (rbtxn_is_prepared(txn))
1990 : {
1991 : /*
1992 : * Note, we send stream prepare even if a concurrent abort is
1993 : * detected. See DecodePrepare for more information.
1994 : */
1995 : Assert(!rbtxn_sent_prepare(txn));
1996 18 : rb->stream_prepare(rb, txn, txn->final_lsn);
1997 18 : txn->txn_flags |= RBTXN_SENT_PREPARE;
1998 :
1999 : /*
2000 : * This is a PREPARED transaction, part of a two-phase commit. The
2001 : * full cleanup will happen as part of the COMMIT PREPAREDs, so now
2002 : * just truncate txn by removing changes and tuplecids.
2003 : */
2004 18 : ReorderBufferTruncateTXN(rb, txn, true);
2005 : /* Reset the CheckXidAlive */
2006 18 : CheckXidAlive = InvalidTransactionId;
2007 : }
2008 : else
2009 : {
2010 98 : rb->stream_commit(rb, txn, txn->final_lsn);
2011 98 : ReorderBufferCleanupTXN(rb, txn);
2012 : }
2013 116 : }
2014 :
2015 : /*
2016 : * Set xid to detect concurrent aborts.
2017 : *
2018 : * While streaming an in-progress transaction or decoding a prepared
2019 : * transaction there is a possibility that the (sub)transaction might get
2020 : * aborted concurrently. In such case if the (sub)transaction has catalog
2021 : * update then we might decode the tuple using wrong catalog version. For
2022 : * example, suppose there is one catalog tuple with (xmin: 500, xmax: 0). Now,
2023 : * the transaction 501 updates the catalog tuple and after that we will have
2024 : * two tuples (xmin: 500, xmax: 501) and (xmin: 501, xmax: 0). Now, if 501 is
2025 : * aborted and some other transaction say 502 updates the same catalog tuple
2026 : * then the first tuple will be changed to (xmin: 500, xmax: 502). So, the
2027 : * problem is that when we try to decode the tuple inserted/updated in 501
2028 : * after the catalog update, we will see the catalog tuple with (xmin: 500,
2029 : * xmax: 502) as visible because it will consider that the tuple is deleted by
2030 : * xid 502 which is not visible to our snapshot. And when we will try to
2031 : * decode with that catalog tuple, it can lead to a wrong result or a crash.
2032 : * So, it is necessary to detect concurrent aborts to allow streaming of
2033 : * in-progress transactions or decoding of prepared transactions.
2034 : *
2035 : * For detecting the concurrent abort we set CheckXidAlive to the current
2036 : * (sub)transaction's xid for which this change belongs to. And, during
2037 : * catalog scan we can check the status of the xid and if it is aborted we will
2038 : * report a specific error so that we can stop streaming current transaction
2039 : * and discard the already streamed changes on such an error. We might have
2040 : * already streamed some of the changes for the aborted (sub)transaction, but
2041 : * that is fine because when we decode the abort we will stream abort message
2042 : * to truncate the changes in the subscriber. Similarly, for prepared
2043 : * transactions, we stop decoding if concurrent abort is detected and then
2044 : * rollback the changes when rollback prepared is encountered. See
2045 : * DecodePrepare.
2046 : */
2047 : static inline void
2048 355656 : SetupCheckXidLive(TransactionId xid)
2049 : {
2050 : /*
2051 : * If the input transaction id is already set as a CheckXidAlive then
2052 : * nothing to do.
2053 : */
2054 355656 : if (TransactionIdEquals(CheckXidAlive, xid))
2055 210136 : return;
2056 :
2057 : /*
2058 : * setup CheckXidAlive if it's not committed yet. We don't check if the
2059 : * xid is aborted. That will happen during catalog access.
2060 : */
2061 145520 : if (!TransactionIdDidCommit(xid))
2062 894 : CheckXidAlive = xid;
2063 : else
2064 144626 : CheckXidAlive = InvalidTransactionId;
2065 : }
2066 :
2067 : /*
2068 : * Helper function for ReorderBufferProcessTXN for applying change.
2069 : */
2070 : static inline void
2071 679992 : ReorderBufferApplyChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
2072 : Relation relation, ReorderBufferChange *change,
2073 : bool streaming)
2074 : {
2075 679992 : if (streaming)
2076 351932 : rb->stream_change(rb, txn, relation, change);
2077 : else
2078 328060 : rb->apply_change(rb, txn, relation, change);
2079 679990 : }
2080 :
2081 : /*
2082 : * Helper function for ReorderBufferProcessTXN for applying the truncate.
2083 : */
2084 : static inline void
2085 44 : ReorderBufferApplyTruncate(ReorderBuffer *rb, ReorderBufferTXN *txn,
2086 : int nrelations, Relation *relations,
2087 : ReorderBufferChange *change, bool streaming)
2088 : {
2089 44 : if (streaming)
2090 0 : rb->stream_truncate(rb, txn, nrelations, relations, change);
2091 : else
2092 44 : rb->apply_truncate(rb, txn, nrelations, relations, change);
2093 44 : }
2094 :
2095 : /*
2096 : * Helper function for ReorderBufferProcessTXN for applying the message.
2097 : */
2098 : static inline void
2099 22 : ReorderBufferApplyMessage(ReorderBuffer *rb, ReorderBufferTXN *txn,
2100 : ReorderBufferChange *change, bool streaming)
2101 : {
2102 22 : if (streaming)
2103 6 : rb->stream_message(rb, txn, change->lsn, true,
2104 6 : change->data.msg.prefix,
2105 : change->data.msg.message_size,
2106 6 : change->data.msg.message);
2107 : else
2108 16 : rb->message(rb, txn, change->lsn, true,
2109 16 : change->data.msg.prefix,
2110 : change->data.msg.message_size,
2111 16 : change->data.msg.message);
2112 22 : }
2113 :
2114 : /*
2115 : * Function to store the command id and snapshot at the end of the current
2116 : * stream so that we can reuse the same while sending the next stream.
2117 : */
2118 : static inline void
2119 1354 : ReorderBufferSaveTXNSnapshot(ReorderBuffer *rb, ReorderBufferTXN *txn,
2120 : Snapshot snapshot_now, CommandId command_id)
2121 : {
2122 1354 : txn->command_id = command_id;
2123 :
2124 : /* Avoid copying if it's already copied. */
2125 1354 : if (snapshot_now->copied)
2126 1354 : txn->snapshot_now = snapshot_now;
2127 : else
2128 0 : txn->snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
2129 : txn, command_id);
2130 1354 : }
2131 :
2132 : /*
2133 : * Mark the given transaction as streamed if it's a top-level transaction
2134 : * or has changes.
2135 : */
2136 : static void
2137 1948 : ReorderBufferMaybeMarkTXNStreamed(ReorderBuffer *rb, ReorderBufferTXN *txn)
2138 : {
2139 : /*
2140 : * The top-level transaction, is marked as streamed always, even if it
2141 : * does not contain any changes (that is, when all the changes are in
2142 : * subtransactions).
2143 : *
2144 : * For subtransactions, we only mark them as streamed when there are
2145 : * changes in them.
2146 : *
2147 : * We do it this way because of aborts - we don't want to send aborts for
2148 : * XIDs the downstream is not aware of. And of course, it always knows
2149 : * about the top-level xact (we send the XID in all messages), but we
2150 : * never stream XIDs of empty subxacts.
2151 : */
2152 1948 : if (rbtxn_is_toptxn(txn) || (txn->nentries_mem != 0))
2153 1624 : txn->txn_flags |= RBTXN_IS_STREAMED;
2154 1948 : }
2155 :
2156 : /*
2157 : * Helper function for ReorderBufferProcessTXN to handle the concurrent
2158 : * abort of the streaming transaction. This resets the TXN such that it
2159 : * can be used to stream the remaining data of transaction being processed.
2160 : * This can happen when the subtransaction is aborted and we still want to
2161 : * continue processing the main or other subtransactions data.
2162 : */
2163 : static void
2164 16 : ReorderBufferResetTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
2165 : Snapshot snapshot_now,
2166 : CommandId command_id,
2167 : XLogRecPtr last_lsn,
2168 : ReorderBufferChange *specinsert)
2169 : {
2170 : /* Discard the changes that we just streamed */
2171 16 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
2172 :
2173 : /* Free all resources allocated for toast reconstruction */
2174 16 : ReorderBufferToastReset(rb, txn);
2175 :
2176 : /* Return the spec insert change if it is not NULL */
2177 16 : if (specinsert != NULL)
2178 : {
2179 0 : ReorderBufferFreeChange(rb, specinsert, true);
2180 0 : specinsert = NULL;
2181 : }
2182 :
2183 : /*
2184 : * For the streaming case, stop the stream and remember the command ID and
2185 : * snapshot for the streaming run.
2186 : */
2187 16 : if (rbtxn_is_streamed(txn))
2188 : {
2189 16 : rb->stream_stop(rb, txn, last_lsn);
2190 16 : ReorderBufferSaveTXNSnapshot(rb, txn, snapshot_now, command_id);
2191 : }
2192 :
2193 : /* All changes must be deallocated */
2194 : Assert(txn->size == 0);
2195 16 : }
2196 :
2197 : /*
2198 : * Helper function for ReorderBufferReplay and ReorderBufferStreamTXN.
2199 : *
2200 : * Send data of a transaction (and its subtransactions) to the
2201 : * output plugin. We iterate over the top and subtransactions (using a k-way
2202 : * merge) and replay the changes in lsn order.
2203 : *
2204 : * If streaming is true then data will be sent using stream API.
2205 : *
2206 : * Note: "volatile" markers on some parameters are to avoid trouble with
2207 : * PG_TRY inside the function.
2208 : */
2209 : static void
2210 3906 : ReorderBufferProcessTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
2211 : XLogRecPtr commit_lsn,
2212 : volatile Snapshot snapshot_now,
2213 : volatile CommandId command_id,
2214 : bool streaming)
2215 : {
2216 : bool using_subtxn;
2217 3906 : MemoryContext ccxt = CurrentMemoryContext;
2218 3906 : ReorderBufferIterTXNState *volatile iterstate = NULL;
2219 3906 : volatile XLogRecPtr prev_lsn = InvalidXLogRecPtr;
2220 3906 : ReorderBufferChange *volatile specinsert = NULL;
2221 3906 : volatile bool stream_started = false;
2222 3906 : ReorderBufferTXN *volatile curtxn = NULL;
2223 :
2224 : /* build data to be able to lookup the CommandIds of catalog tuples */
2225 3906 : ReorderBufferBuildTupleCidHash(rb, txn);
2226 :
2227 : /* setup the initial snapshot */
2228 3906 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2229 :
2230 : /*
2231 : * Decoding needs access to syscaches et al., which in turn use
2232 : * heavyweight locks and such. Thus we need to have enough state around to
2233 : * keep track of those. The easiest way is to simply use a transaction
2234 : * internally. That also allows us to easily enforce that nothing writes
2235 : * to the database by checking for xid assignments.
2236 : *
2237 : * When we're called via the SQL SRF there's already a transaction
2238 : * started, so start an explicit subtransaction there.
2239 : */
2240 3906 : using_subtxn = IsTransactionOrTransactionBlock();
2241 :
2242 3906 : PG_TRY();
2243 : {
2244 : ReorderBufferChange *change;
2245 3906 : int changes_count = 0; /* used to accumulate the number of
2246 : * changes */
2247 :
2248 3906 : if (using_subtxn)
2249 984 : BeginInternalSubTransaction(streaming ? "stream" : "replay");
2250 : else
2251 2922 : StartTransactionCommand();
2252 :
2253 : /*
2254 : * We only need to send begin/begin-prepare for non-streamed
2255 : * transactions.
2256 : */
2257 3906 : if (!streaming)
2258 : {
2259 2552 : if (rbtxn_is_prepared(txn))
2260 52 : rb->begin_prepare(rb, txn);
2261 : else
2262 2500 : rb->begin(rb, txn);
2263 : }
2264 :
2265 3906 : ReorderBufferIterTXNInit(rb, txn, &iterstate);
2266 731992 : while ((change = ReorderBufferIterTXNNext(rb, iterstate)) != NULL)
2267 : {
2268 724198 : Relation relation = NULL;
2269 : Oid reloid;
2270 :
2271 724198 : CHECK_FOR_INTERRUPTS();
2272 :
2273 : /*
2274 : * We can't call start stream callback before processing first
2275 : * change.
2276 : */
2277 724198 : if (prev_lsn == InvalidXLogRecPtr)
2278 : {
2279 3840 : if (streaming)
2280 : {
2281 1290 : txn->origin_id = change->origin_id;
2282 1290 : rb->stream_start(rb, txn, change->lsn);
2283 1290 : stream_started = true;
2284 : }
2285 : }
2286 :
2287 : /*
2288 : * Enforce correct ordering of changes, merged from multiple
2289 : * subtransactions. The changes may have the same LSN due to
2290 : * MULTI_INSERT xlog records.
2291 : */
2292 : Assert(prev_lsn == InvalidXLogRecPtr || prev_lsn <= change->lsn);
2293 :
2294 724198 : prev_lsn = change->lsn;
2295 :
2296 : /*
2297 : * Set the current xid to detect concurrent aborts. This is
2298 : * required for the cases when we decode the changes before the
2299 : * COMMIT record is processed.
2300 : */
2301 724198 : if (streaming || rbtxn_is_prepared(change->txn))
2302 : {
2303 355656 : curtxn = change->txn;
2304 355656 : SetupCheckXidLive(curtxn->xid);
2305 : }
2306 :
2307 724198 : switch (change->action)
2308 : {
2309 3564 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
2310 :
2311 : /*
2312 : * Confirmation for speculative insertion arrived. Simply
2313 : * use as a normal record. It'll be cleaned up at the end
2314 : * of INSERT processing.
2315 : */
2316 3564 : if (specinsert == NULL)
2317 0 : elog(ERROR, "invalid ordering of speculative insertion changes");
2318 : Assert(specinsert->data.tp.oldtuple == NULL);
2319 3564 : change = specinsert;
2320 3564 : change->action = REORDER_BUFFER_CHANGE_INSERT;
2321 :
2322 : /* intentionally fall through */
2323 693086 : case REORDER_BUFFER_CHANGE_INSERT:
2324 : case REORDER_BUFFER_CHANGE_UPDATE:
2325 : case REORDER_BUFFER_CHANGE_DELETE:
2326 : Assert(snapshot_now);
2327 :
2328 693086 : reloid = RelidByRelfilenumber(change->data.tp.rlocator.spcOid,
2329 : change->data.tp.rlocator.relNumber);
2330 :
2331 : /*
2332 : * Mapped catalog tuple without data, emitted while
2333 : * catalog table was in the process of being rewritten. We
2334 : * can fail to look up the relfilenumber, because the
2335 : * relmapper has no "historic" view, in contrast to the
2336 : * normal catalog during decoding. Thus repeated rewrites
2337 : * can cause a lookup failure. That's OK because we do not
2338 : * decode catalog changes anyway. Normally such tuples
2339 : * would be skipped over below, but we can't identify
2340 : * whether the table should be logically logged without
2341 : * mapping the relfilenumber to the oid.
2342 : */
2343 693070 : if (reloid == InvalidOid &&
2344 166 : change->data.tp.newtuple == NULL &&
2345 166 : change->data.tp.oldtuple == NULL)
2346 166 : goto change_done;
2347 692904 : else if (reloid == InvalidOid)
2348 0 : elog(ERROR, "could not map filenumber \"%s\" to relation OID",
2349 : relpathperm(change->data.tp.rlocator,
2350 : MAIN_FORKNUM).str);
2351 :
2352 692904 : relation = RelationIdGetRelation(reloid);
2353 :
2354 692904 : if (!RelationIsValid(relation))
2355 0 : elog(ERROR, "could not open relation with OID %u (for filenumber \"%s\")",
2356 : reloid,
2357 : relpathperm(change->data.tp.rlocator,
2358 : MAIN_FORKNUM).str);
2359 :
2360 692904 : if (!RelationIsLogicallyLogged(relation))
2361 8738 : goto change_done;
2362 :
2363 : /*
2364 : * Ignore temporary heaps created during DDL unless the
2365 : * plugin has asked for them.
2366 : */
2367 684166 : if (relation->rd_rel->relrewrite && !rb->output_rewrites)
2368 52 : goto change_done;
2369 :
2370 : /*
2371 : * For now ignore sequence changes entirely. Most of the
2372 : * time they don't log changes using records we
2373 : * understand, so it doesn't make sense to handle the few
2374 : * cases we do.
2375 : */
2376 684114 : if (relation->rd_rel->relkind == RELKIND_SEQUENCE)
2377 0 : goto change_done;
2378 :
2379 : /* user-triggered change */
2380 684114 : if (!IsToastRelation(relation))
2381 : {
2382 679992 : ReorderBufferToastReplace(rb, txn, relation, change);
2383 679992 : ReorderBufferApplyChange(rb, txn, relation, change,
2384 : streaming);
2385 :
2386 : /*
2387 : * Only clear reassembled toast chunks if we're sure
2388 : * they're not required anymore. The creator of the
2389 : * tuple tells us.
2390 : */
2391 679990 : if (change->data.tp.clear_toast_afterwards)
2392 679548 : ReorderBufferToastReset(rb, txn);
2393 : }
2394 : /* we're not interested in toast deletions */
2395 4122 : else if (change->action == REORDER_BUFFER_CHANGE_INSERT)
2396 : {
2397 : /*
2398 : * Need to reassemble the full toasted Datum in
2399 : * memory, to ensure the chunks don't get reused till
2400 : * we're done remove it from the list of this
2401 : * transaction's changes. Otherwise it will get
2402 : * freed/reused while restoring spooled data from
2403 : * disk.
2404 : */
2405 : Assert(change->data.tp.newtuple != NULL);
2406 :
2407 3660 : dlist_delete(&change->node);
2408 3660 : ReorderBufferToastAppendChunk(rb, txn, relation,
2409 : change);
2410 : }
2411 :
2412 462 : change_done:
2413 :
2414 : /*
2415 : * If speculative insertion was confirmed, the record
2416 : * isn't needed anymore.
2417 : */
2418 693068 : if (specinsert != NULL)
2419 : {
2420 3564 : ReorderBufferFreeChange(rb, specinsert, true);
2421 3564 : specinsert = NULL;
2422 : }
2423 :
2424 693068 : if (RelationIsValid(relation))
2425 : {
2426 692902 : RelationClose(relation);
2427 692902 : relation = NULL;
2428 : }
2429 693068 : break;
2430 :
2431 3564 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
2432 :
2433 : /*
2434 : * Speculative insertions are dealt with by delaying the
2435 : * processing of the insert until the confirmation record
2436 : * arrives. For that we simply unlink the record from the
2437 : * chain, so it does not get freed/reused while restoring
2438 : * spooled data from disk.
2439 : *
2440 : * This is safe in the face of concurrent catalog changes
2441 : * because the relevant relation can't be changed between
2442 : * speculative insertion and confirmation due to
2443 : * CheckTableNotInUse() and locking.
2444 : */
2445 :
2446 : /* clear out a pending (and thus failed) speculation */
2447 3564 : if (specinsert != NULL)
2448 : {
2449 0 : ReorderBufferFreeChange(rb, specinsert, true);
2450 0 : specinsert = NULL;
2451 : }
2452 :
2453 : /* and memorize the pending insertion */
2454 3564 : dlist_delete(&change->node);
2455 3564 : specinsert = change;
2456 3564 : break;
2457 :
2458 0 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
2459 :
2460 : /*
2461 : * Abort for speculative insertion arrived. So cleanup the
2462 : * specinsert tuple and toast hash.
2463 : *
2464 : * Note that we get the spec abort change for each toast
2465 : * entry but we need to perform the cleanup only the first
2466 : * time we get it for the main table.
2467 : */
2468 0 : if (specinsert != NULL)
2469 : {
2470 : /*
2471 : * We must clean the toast hash before processing a
2472 : * completely new tuple to avoid confusion about the
2473 : * previous tuple's toast chunks.
2474 : */
2475 : Assert(change->data.tp.clear_toast_afterwards);
2476 0 : ReorderBufferToastReset(rb, txn);
2477 :
2478 : /* We don't need this record anymore. */
2479 0 : ReorderBufferFreeChange(rb, specinsert, true);
2480 0 : specinsert = NULL;
2481 : }
2482 0 : break;
2483 :
2484 44 : case REORDER_BUFFER_CHANGE_TRUNCATE:
2485 : {
2486 : int i;
2487 44 : int nrelids = change->data.truncate.nrelids;
2488 44 : int nrelations = 0;
2489 : Relation *relations;
2490 :
2491 44 : relations = palloc0(nrelids * sizeof(Relation));
2492 128 : for (i = 0; i < nrelids; i++)
2493 : {
2494 84 : Oid relid = change->data.truncate.relids[i];
2495 : Relation rel;
2496 :
2497 84 : rel = RelationIdGetRelation(relid);
2498 :
2499 84 : if (!RelationIsValid(rel))
2500 0 : elog(ERROR, "could not open relation with OID %u", relid);
2501 :
2502 84 : if (!RelationIsLogicallyLogged(rel))
2503 0 : continue;
2504 :
2505 84 : relations[nrelations++] = rel;
2506 : }
2507 :
2508 : /* Apply the truncate. */
2509 44 : ReorderBufferApplyTruncate(rb, txn, nrelations,
2510 : relations, change,
2511 : streaming);
2512 :
2513 128 : for (i = 0; i < nrelations; i++)
2514 84 : RelationClose(relations[i]);
2515 :
2516 44 : break;
2517 : }
2518 :
2519 22 : case REORDER_BUFFER_CHANGE_MESSAGE:
2520 22 : ReorderBufferApplyMessage(rb, txn, change, streaming);
2521 22 : break;
2522 :
2523 4532 : case REORDER_BUFFER_CHANGE_INVALIDATION:
2524 : /* Execute the invalidation messages locally */
2525 4532 : ReorderBufferExecuteInvalidations(change->data.inval.ninvalidations,
2526 : change->data.inval.invalidations);
2527 4532 : break;
2528 :
2529 1128 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
2530 : /* get rid of the old */
2531 1128 : TeardownHistoricSnapshot(false);
2532 :
2533 1128 : if (snapshot_now->copied)
2534 : {
2535 1078 : ReorderBufferFreeSnap(rb, snapshot_now);
2536 1078 : snapshot_now =
2537 1078 : ReorderBufferCopySnap(rb, change->data.snapshot,
2538 : txn, command_id);
2539 : }
2540 :
2541 : /*
2542 : * Restored from disk, need to be careful not to double
2543 : * free. We could introduce refcounting for that, but for
2544 : * now this seems infrequent enough not to care.
2545 : */
2546 50 : else if (change->data.snapshot->copied)
2547 : {
2548 0 : snapshot_now =
2549 0 : ReorderBufferCopySnap(rb, change->data.snapshot,
2550 : txn, command_id);
2551 : }
2552 : else
2553 : {
2554 50 : snapshot_now = change->data.snapshot;
2555 : }
2556 :
2557 : /* and continue with the new one */
2558 1128 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2559 1128 : break;
2560 :
2561 21822 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
2562 : Assert(change->data.command_id != InvalidCommandId);
2563 :
2564 21822 : if (command_id < change->data.command_id)
2565 : {
2566 3898 : command_id = change->data.command_id;
2567 :
2568 3898 : if (!snapshot_now->copied)
2569 : {
2570 : /* we don't use the global one anymore */
2571 1068 : snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
2572 : txn, command_id);
2573 : }
2574 :
2575 3898 : snapshot_now->curcid = command_id;
2576 :
2577 3898 : TeardownHistoricSnapshot(false);
2578 3898 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2579 : }
2580 :
2581 21822 : break;
2582 :
2583 0 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
2584 0 : elog(ERROR, "tuplecid value in changequeue");
2585 : break;
2586 : }
2587 :
2588 : /*
2589 : * It is possible that the data is not sent to downstream for a
2590 : * long time either because the output plugin filtered it or there
2591 : * is a DDL that generates a lot of data that is not processed by
2592 : * the plugin. So, in such cases, the downstream can timeout. To
2593 : * avoid that we try to send a keepalive message if required.
2594 : * Trying to send a keepalive message after every change has some
2595 : * overhead, but testing showed there is no noticeable overhead if
2596 : * we do it after every ~100 changes.
2597 : */
2598 : #define CHANGES_THRESHOLD 100
2599 :
2600 724180 : if (++changes_count >= CHANGES_THRESHOLD)
2601 : {
2602 6320 : rb->update_progress_txn(rb, txn, change->lsn);
2603 6320 : changes_count = 0;
2604 : }
2605 : }
2606 :
2607 : /* speculative insertion record must be freed by now */
2608 : Assert(!specinsert);
2609 :
2610 : /* clean up the iterator */
2611 3888 : ReorderBufferIterTXNFinish(rb, iterstate);
2612 3888 : iterstate = NULL;
2613 :
2614 : /*
2615 : * Update total transaction count and total bytes processed by the
2616 : * transaction and its subtransactions. Ensure to not count the
2617 : * streamed transaction multiple times.
2618 : *
2619 : * Note that the statistics computation has to be done after
2620 : * ReorderBufferIterTXNFinish as it releases the serialized change
2621 : * which we have already accounted in ReorderBufferIterTXNNext.
2622 : */
2623 3888 : if (!rbtxn_is_streamed(txn))
2624 2670 : rb->totalTxns++;
2625 :
2626 3888 : rb->totalBytes += txn->total_size;
2627 :
2628 : /*
2629 : * Done with current changes, send the last message for this set of
2630 : * changes depending upon streaming mode.
2631 : */
2632 3888 : if (streaming)
2633 : {
2634 1338 : if (stream_started)
2635 : {
2636 1274 : rb->stream_stop(rb, txn, prev_lsn);
2637 1274 : stream_started = false;
2638 : }
2639 : }
2640 : else
2641 : {
2642 : /*
2643 : * Call either PREPARE (for two-phase transactions) or COMMIT (for
2644 : * regular ones).
2645 : */
2646 2550 : if (rbtxn_is_prepared(txn))
2647 : {
2648 : Assert(!rbtxn_sent_prepare(txn));
2649 52 : rb->prepare(rb, txn, commit_lsn);
2650 52 : txn->txn_flags |= RBTXN_SENT_PREPARE;
2651 : }
2652 : else
2653 2498 : rb->commit(rb, txn, commit_lsn);
2654 : }
2655 :
2656 : /* this is just a sanity check against bad output plugin behaviour */
2657 3844 : if (GetCurrentTransactionIdIfAny() != InvalidTransactionId)
2658 0 : elog(ERROR, "output plugin used XID %u",
2659 : GetCurrentTransactionId());
2660 :
2661 : /*
2662 : * Remember the command ID and snapshot for the next set of changes in
2663 : * streaming mode.
2664 : */
2665 3844 : if (streaming)
2666 1338 : ReorderBufferSaveTXNSnapshot(rb, txn, snapshot_now, command_id);
2667 2506 : else if (snapshot_now->copied)
2668 1068 : ReorderBufferFreeSnap(rb, snapshot_now);
2669 :
2670 : /* cleanup */
2671 3844 : TeardownHistoricSnapshot(false);
2672 :
2673 : /*
2674 : * Aborting the current (sub-)transaction as a whole has the right
2675 : * semantics. We want all locks acquired in here to be released, not
2676 : * reassigned to the parent and we do not want any database access
2677 : * have persistent effects.
2678 : */
2679 3844 : AbortCurrentTransaction();
2680 :
2681 : /* make sure there's no cache pollution */
2682 3844 : if (rbtxn_distr_inval_overflowed(txn))
2683 : {
2684 : Assert(txn->ninvalidations_distributed == 0);
2685 0 : InvalidateSystemCaches();
2686 : }
2687 : else
2688 : {
2689 3844 : ReorderBufferExecuteInvalidations(txn->ninvalidations, txn->invalidations);
2690 3844 : ReorderBufferExecuteInvalidations(txn->ninvalidations_distributed,
2691 : txn->invalidations_distributed);
2692 : }
2693 :
2694 3844 : if (using_subtxn)
2695 976 : RollbackAndReleaseCurrentSubTransaction();
2696 :
2697 : /*
2698 : * We are here due to one of the four reasons: 1. Decoding an
2699 : * in-progress txn. 2. Decoding a prepared txn. 3. Decoding of a
2700 : * prepared txn that was (partially) streamed. 4. Decoding a committed
2701 : * txn.
2702 : *
2703 : * For 1, we allow truncation of txn data by removing the changes
2704 : * already streamed but still keeping other things like invalidations,
2705 : * snapshot, and tuplecids. For 2 and 3, we indicate
2706 : * ReorderBufferTruncateTXN to do more elaborate truncation of txn
2707 : * data as the entire transaction has been decoded except for commit.
2708 : * For 4, as the entire txn has been decoded, we can fully clean up
2709 : * the TXN reorder buffer.
2710 : */
2711 3844 : if (streaming || rbtxn_is_prepared(txn))
2712 : {
2713 1390 : if (streaming)
2714 1338 : ReorderBufferMaybeMarkTXNStreamed(rb, txn);
2715 :
2716 1390 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
2717 : /* Reset the CheckXidAlive */
2718 1390 : CheckXidAlive = InvalidTransactionId;
2719 : }
2720 : else
2721 2454 : ReorderBufferCleanupTXN(rb, txn);
2722 : }
2723 18 : PG_CATCH();
2724 : {
2725 18 : MemoryContext ecxt = MemoryContextSwitchTo(ccxt);
2726 18 : ErrorData *errdata = CopyErrorData();
2727 :
2728 : /* TODO: Encapsulate cleanup from the PG_TRY and PG_CATCH blocks */
2729 18 : if (iterstate)
2730 18 : ReorderBufferIterTXNFinish(rb, iterstate);
2731 :
2732 18 : TeardownHistoricSnapshot(true);
2733 :
2734 : /*
2735 : * Force cache invalidation to happen outside of a valid transaction
2736 : * to prevent catalog access as we just caught an error.
2737 : */
2738 18 : AbortCurrentTransaction();
2739 :
2740 : /* make sure there's no cache pollution */
2741 18 : if (rbtxn_distr_inval_overflowed(txn))
2742 : {
2743 : Assert(txn->ninvalidations_distributed == 0);
2744 0 : InvalidateSystemCaches();
2745 : }
2746 : else
2747 : {
2748 18 : ReorderBufferExecuteInvalidations(txn->ninvalidations, txn->invalidations);
2749 18 : ReorderBufferExecuteInvalidations(txn->ninvalidations_distributed,
2750 : txn->invalidations_distributed);
2751 : }
2752 :
2753 18 : if (using_subtxn)
2754 8 : RollbackAndReleaseCurrentSubTransaction();
2755 :
2756 : /*
2757 : * The error code ERRCODE_TRANSACTION_ROLLBACK indicates a concurrent
2758 : * abort of the (sub)transaction we are streaming or preparing. We
2759 : * need to do the cleanup and return gracefully on this error, see
2760 : * SetupCheckXidLive.
2761 : *
2762 : * This error code can be thrown by one of the callbacks we call
2763 : * during decoding so we need to ensure that we return gracefully only
2764 : * when we are sending the data in streaming mode and the streaming is
2765 : * not finished yet or when we are sending the data out on a PREPARE
2766 : * during a two-phase commit.
2767 : */
2768 18 : if (errdata->sqlerrcode == ERRCODE_TRANSACTION_ROLLBACK &&
2769 16 : (stream_started || rbtxn_is_prepared(txn)))
2770 : {
2771 : /* curtxn must be set for streaming or prepared transactions */
2772 : Assert(curtxn);
2773 :
2774 : /* Cleanup the temporary error state. */
2775 16 : FlushErrorState();
2776 16 : FreeErrorData(errdata);
2777 16 : errdata = NULL;
2778 :
2779 : /* Remember the transaction is aborted. */
2780 : Assert(!rbtxn_is_committed(curtxn));
2781 16 : curtxn->txn_flags |= RBTXN_IS_ABORTED;
2782 :
2783 : /* Mark the transaction is streamed if appropriate */
2784 16 : if (stream_started)
2785 16 : ReorderBufferMaybeMarkTXNStreamed(rb, txn);
2786 :
2787 : /* Reset the TXN so that it is allowed to stream remaining data. */
2788 16 : ReorderBufferResetTXN(rb, txn, snapshot_now,
2789 : command_id, prev_lsn,
2790 : specinsert);
2791 : }
2792 : else
2793 : {
2794 2 : ReorderBufferCleanupTXN(rb, txn);
2795 2 : MemoryContextSwitchTo(ecxt);
2796 2 : PG_RE_THROW();
2797 : }
2798 : }
2799 3860 : PG_END_TRY();
2800 3860 : }
2801 :
2802 : /*
2803 : * Perform the replay of a transaction and its non-aborted subtransactions.
2804 : *
2805 : * Subtransactions previously have to be processed by
2806 : * ReorderBufferCommitChild(), even if previously assigned to the toplevel
2807 : * transaction with ReorderBufferAssignChild.
2808 : *
2809 : * This interface is called once a prepare or toplevel commit is read for both
2810 : * streamed as well as non-streamed transactions.
2811 : */
2812 : static void
2813 2674 : ReorderBufferReplay(ReorderBufferTXN *txn,
2814 : ReorderBuffer *rb, TransactionId xid,
2815 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2816 : TimestampTz commit_time,
2817 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2818 : {
2819 : Snapshot snapshot_now;
2820 2674 : CommandId command_id = FirstCommandId;
2821 :
2822 2674 : txn->final_lsn = commit_lsn;
2823 2674 : txn->end_lsn = end_lsn;
2824 2674 : txn->xact_time.commit_time = commit_time;
2825 2674 : txn->origin_id = origin_id;
2826 2674 : txn->origin_lsn = origin_lsn;
2827 :
2828 : /*
2829 : * If the transaction was (partially) streamed, we need to commit it in a
2830 : * 'streamed' way. That is, we first stream the remaining part of the
2831 : * transaction, and then invoke stream_commit message.
2832 : *
2833 : * Called after everything (origin ID, LSN, ...) is stored in the
2834 : * transaction to avoid passing that information directly.
2835 : */
2836 2674 : if (rbtxn_is_streamed(txn))
2837 : {
2838 116 : ReorderBufferStreamCommit(rb, txn);
2839 116 : return;
2840 : }
2841 :
2842 : /*
2843 : * If this transaction has no snapshot, it didn't make any changes to the
2844 : * database, so there's nothing to decode. Note that
2845 : * ReorderBufferCommitChild will have transferred any snapshots from
2846 : * subtransactions if there were any.
2847 : */
2848 2558 : if (txn->base_snapshot == NULL)
2849 : {
2850 : Assert(txn->ninvalidations == 0);
2851 :
2852 : /*
2853 : * Removing this txn before a commit might result in the computation
2854 : * of an incorrect restart_lsn. See SnapBuildProcessRunningXacts.
2855 : */
2856 6 : if (!rbtxn_is_prepared(txn))
2857 6 : ReorderBufferCleanupTXN(rb, txn);
2858 6 : return;
2859 : }
2860 :
2861 2552 : snapshot_now = txn->base_snapshot;
2862 :
2863 : /* Process and send the changes to output plugin. */
2864 2552 : ReorderBufferProcessTXN(rb, txn, commit_lsn, snapshot_now,
2865 : command_id, false);
2866 : }
2867 :
2868 : /*
2869 : * Commit a transaction.
2870 : *
2871 : * See comments for ReorderBufferReplay().
2872 : */
2873 : void
2874 2608 : ReorderBufferCommit(ReorderBuffer *rb, TransactionId xid,
2875 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2876 : TimestampTz commit_time,
2877 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2878 : {
2879 : ReorderBufferTXN *txn;
2880 :
2881 2608 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2882 : false);
2883 :
2884 : /* unknown transaction, nothing to replay */
2885 2608 : if (txn == NULL)
2886 4 : return;
2887 :
2888 2604 : ReorderBufferReplay(txn, rb, xid, commit_lsn, end_lsn, commit_time,
2889 : origin_id, origin_lsn);
2890 : }
2891 :
2892 : /*
2893 : * Record the prepare information for a transaction. Also, mark the transaction
2894 : * as a prepared transaction.
2895 : */
2896 : bool
2897 256 : ReorderBufferRememberPrepareInfo(ReorderBuffer *rb, TransactionId xid,
2898 : XLogRecPtr prepare_lsn, XLogRecPtr end_lsn,
2899 : TimestampTz prepare_time,
2900 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2901 : {
2902 : ReorderBufferTXN *txn;
2903 :
2904 256 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr, false);
2905 :
2906 : /* unknown transaction, nothing to do */
2907 256 : if (txn == NULL)
2908 0 : return false;
2909 :
2910 : /*
2911 : * Remember the prepare information to be later used by commit prepared in
2912 : * case we skip doing prepare.
2913 : */
2914 256 : txn->final_lsn = prepare_lsn;
2915 256 : txn->end_lsn = end_lsn;
2916 256 : txn->xact_time.prepare_time = prepare_time;
2917 256 : txn->origin_id = origin_id;
2918 256 : txn->origin_lsn = origin_lsn;
2919 :
2920 : /* Mark this transaction as a prepared transaction */
2921 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == 0);
2922 256 : txn->txn_flags |= RBTXN_IS_PREPARED;
2923 :
2924 256 : return true;
2925 : }
2926 :
2927 : /* Remember that we have skipped prepare */
2928 : void
2929 192 : ReorderBufferSkipPrepare(ReorderBuffer *rb, TransactionId xid)
2930 : {
2931 : ReorderBufferTXN *txn;
2932 :
2933 192 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr, false);
2934 :
2935 : /* unknown transaction, nothing to do */
2936 192 : if (txn == NULL)
2937 0 : return;
2938 :
2939 : /* txn must have been marked as a prepared transaction */
2940 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == RBTXN_IS_PREPARED);
2941 192 : txn->txn_flags |= RBTXN_SKIPPED_PREPARE;
2942 : }
2943 :
2944 : /*
2945 : * Prepare a two-phase transaction.
2946 : *
2947 : * See comments for ReorderBufferReplay().
2948 : */
2949 : void
2950 64 : ReorderBufferPrepare(ReorderBuffer *rb, TransactionId xid,
2951 : char *gid)
2952 : {
2953 : ReorderBufferTXN *txn;
2954 :
2955 64 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2956 : false);
2957 :
2958 : /* unknown transaction, nothing to replay */
2959 64 : if (txn == NULL)
2960 0 : return;
2961 :
2962 : /*
2963 : * txn must have been marked as a prepared transaction and must have
2964 : * neither been skipped nor sent a prepare. Also, the prepare info must
2965 : * have been updated in it by now.
2966 : */
2967 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == RBTXN_IS_PREPARED);
2968 : Assert(txn->final_lsn != InvalidXLogRecPtr);
2969 :
2970 64 : txn->gid = pstrdup(gid);
2971 :
2972 64 : ReorderBufferReplay(txn, rb, xid, txn->final_lsn, txn->end_lsn,
2973 64 : txn->xact_time.prepare_time, txn->origin_id, txn->origin_lsn);
2974 :
2975 : /*
2976 : * Send a prepare if not already done so. This might occur if we have
2977 : * detected a concurrent abort while replaying the non-streaming
2978 : * transaction.
2979 : */
2980 64 : if (!rbtxn_sent_prepare(txn))
2981 : {
2982 0 : rb->prepare(rb, txn, txn->final_lsn);
2983 0 : txn->txn_flags |= RBTXN_SENT_PREPARE;
2984 : }
2985 : }
2986 :
2987 : /*
2988 : * This is used to handle COMMIT/ROLLBACK PREPARED.
2989 : */
2990 : void
2991 72 : ReorderBufferFinishPrepared(ReorderBuffer *rb, TransactionId xid,
2992 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2993 : XLogRecPtr two_phase_at,
2994 : TimestampTz commit_time, RepOriginId origin_id,
2995 : XLogRecPtr origin_lsn, char *gid, bool is_commit)
2996 : {
2997 : ReorderBufferTXN *txn;
2998 : XLogRecPtr prepare_end_lsn;
2999 : TimestampTz prepare_time;
3000 :
3001 72 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, commit_lsn, false);
3002 :
3003 : /* unknown transaction, nothing to do */
3004 72 : if (txn == NULL)
3005 0 : return;
3006 :
3007 : /*
3008 : * By this time the txn has the prepare record information, remember it to
3009 : * be later used for rollback.
3010 : */
3011 72 : prepare_end_lsn = txn->end_lsn;
3012 72 : prepare_time = txn->xact_time.prepare_time;
3013 :
3014 : /* add the gid in the txn */
3015 72 : txn->gid = pstrdup(gid);
3016 :
3017 : /*
3018 : * It is possible that this transaction is not decoded at prepare time
3019 : * either because by that time we didn't have a consistent snapshot, or
3020 : * two_phase was not enabled, or it was decoded earlier but we have
3021 : * restarted. We only need to send the prepare if it was not decoded
3022 : * earlier. We don't need to decode the xact for aborts if it is not done
3023 : * already.
3024 : */
3025 72 : if ((txn->final_lsn < two_phase_at) && is_commit)
3026 : {
3027 : /*
3028 : * txn must have been marked as a prepared transaction and skipped but
3029 : * not sent a prepare. Also, the prepare info must have been updated
3030 : * in txn even if we skip prepare.
3031 : */
3032 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) ==
3033 : (RBTXN_IS_PREPARED | RBTXN_SKIPPED_PREPARE));
3034 : Assert(txn->final_lsn != InvalidXLogRecPtr);
3035 :
3036 : /*
3037 : * By this time the txn has the prepare record information and it is
3038 : * important to use that so that downstream gets the accurate
3039 : * information. If instead, we have passed commit information here
3040 : * then downstream can behave as it has already replayed commit
3041 : * prepared after the restart.
3042 : */
3043 6 : ReorderBufferReplay(txn, rb, xid, txn->final_lsn, txn->end_lsn,
3044 6 : txn->xact_time.prepare_time, txn->origin_id, txn->origin_lsn);
3045 : }
3046 :
3047 72 : txn->final_lsn = commit_lsn;
3048 72 : txn->end_lsn = end_lsn;
3049 72 : txn->xact_time.commit_time = commit_time;
3050 72 : txn->origin_id = origin_id;
3051 72 : txn->origin_lsn = origin_lsn;
3052 :
3053 72 : if (is_commit)
3054 58 : rb->commit_prepared(rb, txn, commit_lsn);
3055 : else
3056 14 : rb->rollback_prepared(rb, txn, prepare_end_lsn, prepare_time);
3057 :
3058 : /* cleanup: make sure there's no cache pollution */
3059 72 : ReorderBufferExecuteInvalidations(txn->ninvalidations,
3060 : txn->invalidations);
3061 72 : ReorderBufferCleanupTXN(rb, txn);
3062 : }
3063 :
3064 : /*
3065 : * Abort a transaction that possibly has previous changes. Needs to be first
3066 : * called for subtransactions and then for the toplevel xid.
3067 : *
3068 : * NB: Transactions handled here have to have actively aborted (i.e. have
3069 : * produced an abort record). Implicitly aborted transactions are handled via
3070 : * ReorderBufferAbortOld(); transactions we're just not interested in, but
3071 : * which have committed are handled in ReorderBufferForget().
3072 : *
3073 : * This function purges this transaction and its contents from memory and
3074 : * disk.
3075 : */
3076 : void
3077 248 : ReorderBufferAbort(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
3078 : TimestampTz abort_time)
3079 : {
3080 : ReorderBufferTXN *txn;
3081 :
3082 248 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3083 : false);
3084 :
3085 : /* unknown, nothing to remove */
3086 248 : if (txn == NULL)
3087 0 : return;
3088 :
3089 248 : txn->xact_time.abort_time = abort_time;
3090 :
3091 : /* For streamed transactions notify the remote node about the abort. */
3092 248 : if (rbtxn_is_streamed(txn))
3093 : {
3094 60 : rb->stream_abort(rb, txn, lsn);
3095 :
3096 : /*
3097 : * We might have decoded changes for this transaction that could load
3098 : * the cache as per the current transaction's view (consider DDL's
3099 : * happened in this transaction). We don't want the decoding of future
3100 : * transactions to use those cache entries so execute only the inval
3101 : * messages in this transaction.
3102 : */
3103 60 : if (txn->ninvalidations > 0)
3104 0 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3105 : txn->invalidations);
3106 : }
3107 :
3108 : /* cosmetic... */
3109 248 : txn->final_lsn = lsn;
3110 :
3111 : /* remove potential on-disk data, and deallocate */
3112 248 : ReorderBufferCleanupTXN(rb, txn);
3113 : }
3114 :
3115 : /*
3116 : * Abort all transactions that aren't actually running anymore because the
3117 : * server restarted.
3118 : *
3119 : * NB: These really have to be transactions that have aborted due to a server
3120 : * crash/immediate restart, as we don't deal with invalidations here.
3121 : */
3122 : void
3123 2704 : ReorderBufferAbortOld(ReorderBuffer *rb, TransactionId oldestRunningXid)
3124 : {
3125 : dlist_mutable_iter it;
3126 :
3127 : /*
3128 : * Iterate through all (potential) toplevel TXNs and abort all that are
3129 : * older than what possibly can be running. Once we've found the first
3130 : * that is alive we stop, there might be some that acquired an xid earlier
3131 : * but started writing later, but it's unlikely and they will be cleaned
3132 : * up in a later call to this function.
3133 : */
3134 2724 : dlist_foreach_modify(it, &rb->toplevel_by_lsn)
3135 : {
3136 : ReorderBufferTXN *txn;
3137 :
3138 142 : txn = dlist_container(ReorderBufferTXN, node, it.cur);
3139 :
3140 142 : if (TransactionIdPrecedes(txn->xid, oldestRunningXid))
3141 : {
3142 20 : elog(DEBUG2, "aborting old transaction %u", txn->xid);
3143 :
3144 : /* Notify the remote node about the crash/immediate restart. */
3145 20 : if (rbtxn_is_streamed(txn))
3146 0 : rb->stream_abort(rb, txn, InvalidXLogRecPtr);
3147 :
3148 : /* remove potential on-disk data, and deallocate this tx */
3149 20 : ReorderBufferCleanupTXN(rb, txn);
3150 : }
3151 : else
3152 122 : return;
3153 : }
3154 : }
3155 :
3156 : /*
3157 : * Forget the contents of a transaction if we aren't interested in its
3158 : * contents. Needs to be first called for subtransactions and then for the
3159 : * toplevel xid.
3160 : *
3161 : * This is significantly different to ReorderBufferAbort() because
3162 : * transactions that have committed need to be treated differently from aborted
3163 : * ones since they may have modified the catalog.
3164 : *
3165 : * Note that this is only allowed to be called in the moment a transaction
3166 : * commit has just been read, not earlier; otherwise later records referring
3167 : * to this xid might re-create the transaction incompletely.
3168 : */
3169 : void
3170 5390 : ReorderBufferForget(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3171 : {
3172 : ReorderBufferTXN *txn;
3173 :
3174 5390 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3175 : false);
3176 :
3177 : /* unknown, nothing to forget */
3178 5390 : if (txn == NULL)
3179 1122 : return;
3180 :
3181 : /* this transaction mustn't be streamed */
3182 : Assert(!rbtxn_is_streamed(txn));
3183 :
3184 : /* cosmetic... */
3185 4268 : txn->final_lsn = lsn;
3186 :
3187 : /*
3188 : * Process only cache invalidation messages in this transaction if there
3189 : * are any. Even if we're not interested in the transaction's contents, it
3190 : * could have manipulated the catalog and we need to update the caches
3191 : * according to that.
3192 : */
3193 4268 : if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
3194 1158 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3195 : txn->invalidations);
3196 : else
3197 : Assert(txn->ninvalidations == 0);
3198 :
3199 : /* remove potential on-disk data, and deallocate */
3200 4268 : ReorderBufferCleanupTXN(rb, txn);
3201 : }
3202 :
3203 : /*
3204 : * Invalidate cache for those transactions that need to be skipped just in case
3205 : * catalogs were manipulated as part of the transaction.
3206 : *
3207 : * Note that this is a special-purpose function for prepared transactions where
3208 : * we don't want to clean up the TXN even when we decide to skip it. See
3209 : * DecodePrepare.
3210 : */
3211 : void
3212 186 : ReorderBufferInvalidate(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3213 : {
3214 : ReorderBufferTXN *txn;
3215 :
3216 186 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3217 : false);
3218 :
3219 : /* unknown, nothing to do */
3220 186 : if (txn == NULL)
3221 0 : return;
3222 :
3223 : /*
3224 : * Process cache invalidation messages if there are any. Even if we're not
3225 : * interested in the transaction's contents, it could have manipulated the
3226 : * catalog and we need to update the caches according to that.
3227 : */
3228 186 : if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
3229 58 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3230 : txn->invalidations);
3231 : else
3232 : Assert(txn->ninvalidations == 0);
3233 : }
3234 :
3235 :
3236 : /*
3237 : * Execute invalidations happening outside the context of a decoded
3238 : * transaction. That currently happens either for xid-less commits
3239 : * (cf. RecordTransactionCommit()) or for invalidations in uninteresting
3240 : * transactions (via ReorderBufferForget()).
3241 : */
3242 : void
3243 1222 : ReorderBufferImmediateInvalidation(ReorderBuffer *rb, uint32 ninvalidations,
3244 : SharedInvalidationMessage *invalidations)
3245 : {
3246 1222 : bool use_subtxn = IsTransactionOrTransactionBlock();
3247 : int i;
3248 :
3249 1222 : if (use_subtxn)
3250 870 : BeginInternalSubTransaction("replay");
3251 :
3252 : /*
3253 : * Force invalidations to happen outside of a valid transaction - that way
3254 : * entries will just be marked as invalid without accessing the catalog.
3255 : * That's advantageous because we don't need to setup the full state
3256 : * necessary for catalog access.
3257 : */
3258 1222 : if (use_subtxn)
3259 870 : AbortCurrentTransaction();
3260 :
3261 52440 : for (i = 0; i < ninvalidations; i++)
3262 51218 : LocalExecuteInvalidationMessage(&invalidations[i]);
3263 :
3264 1222 : if (use_subtxn)
3265 870 : RollbackAndReleaseCurrentSubTransaction();
3266 1222 : }
3267 :
3268 : /*
3269 : * Tell reorderbuffer about an xid seen in the WAL stream. Has to be called at
3270 : * least once for every xid in XLogRecord->xl_xid (other places in records
3271 : * may, but do not have to be passed through here).
3272 : *
3273 : * Reorderbuffer keeps some data structures about transactions in LSN order,
3274 : * for efficiency. To do that it has to know about when transactions are seen
3275 : * first in the WAL. As many types of records are not actually interesting for
3276 : * logical decoding, they do not necessarily pass through here.
3277 : */
3278 : void
3279 4333034 : ReorderBufferProcessXid(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3280 : {
3281 : /* many records won't have an xid assigned, centralize check here */
3282 4333034 : if (xid != InvalidTransactionId)
3283 4329256 : ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3284 4333034 : }
3285 :
3286 : /*
3287 : * Add a new snapshot to this transaction that may only used after lsn 'lsn'
3288 : * because the previous snapshot doesn't describe the catalog correctly for
3289 : * following rows.
3290 : */
3291 : void
3292 2330 : ReorderBufferAddSnapshot(ReorderBuffer *rb, TransactionId xid,
3293 : XLogRecPtr lsn, Snapshot snap)
3294 : {
3295 2330 : ReorderBufferChange *change = ReorderBufferAllocChange(rb);
3296 :
3297 2330 : change->data.snapshot = snap;
3298 2330 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT;
3299 :
3300 2330 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3301 2330 : }
3302 :
3303 : /*
3304 : * Set up the transaction's base snapshot.
3305 : *
3306 : * If we know that xid is a subtransaction, set the base snapshot on the
3307 : * top-level transaction instead.
3308 : */
3309 : void
3310 6254 : ReorderBufferSetBaseSnapshot(ReorderBuffer *rb, TransactionId xid,
3311 : XLogRecPtr lsn, Snapshot snap)
3312 : {
3313 : ReorderBufferTXN *txn;
3314 : bool is_new;
3315 :
3316 : Assert(snap != NULL);
3317 :
3318 : /*
3319 : * Fetch the transaction to operate on. If we know it's a subtransaction,
3320 : * operate on its top-level transaction instead.
3321 : */
3322 6254 : txn = ReorderBufferTXNByXid(rb, xid, true, &is_new, lsn, true);
3323 6254 : if (rbtxn_is_known_subxact(txn))
3324 244 : txn = ReorderBufferTXNByXid(rb, txn->toplevel_xid, false,
3325 : NULL, InvalidXLogRecPtr, false);
3326 : Assert(txn->base_snapshot == NULL);
3327 :
3328 6254 : txn->base_snapshot = snap;
3329 6254 : txn->base_snapshot_lsn = lsn;
3330 6254 : dlist_push_tail(&rb->txns_by_base_snapshot_lsn, &txn->base_snapshot_node);
3331 :
3332 6254 : AssertTXNLsnOrder(rb);
3333 6254 : }
3334 :
3335 : /*
3336 : * Access the catalog with this CommandId at this point in the changestream.
3337 : *
3338 : * May only be called for command ids > 1
3339 : */
3340 : void
3341 49250 : ReorderBufferAddNewCommandId(ReorderBuffer *rb, TransactionId xid,
3342 : XLogRecPtr lsn, CommandId cid)
3343 : {
3344 49250 : ReorderBufferChange *change = ReorderBufferAllocChange(rb);
3345 :
3346 49250 : change->data.command_id = cid;
3347 49250 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID;
3348 :
3349 49250 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3350 49250 : }
3351 :
3352 : /*
3353 : * Update memory counters to account for the new or removed change.
3354 : *
3355 : * We update two counters - in the reorder buffer, and in the transaction
3356 : * containing the change. The reorder buffer counter allows us to quickly
3357 : * decide if we reached the memory limit, the transaction counter allows
3358 : * us to quickly pick the largest transaction for eviction.
3359 : *
3360 : * Either txn or change must be non-NULL at least. We update the memory
3361 : * counter of txn if it's non-NULL, otherwise change->txn.
3362 : *
3363 : * When streaming is enabled, we need to update the toplevel transaction
3364 : * counters instead - we don't really care about subtransactions as we
3365 : * can't stream them individually anyway, and we only pick toplevel
3366 : * transactions for eviction. So only toplevel transactions matter.
3367 : */
3368 : static void
3369 3825700 : ReorderBufferChangeMemoryUpdate(ReorderBuffer *rb,
3370 : ReorderBufferChange *change,
3371 : ReorderBufferTXN *txn,
3372 : bool addition, Size sz)
3373 : {
3374 : ReorderBufferTXN *toptxn;
3375 :
3376 : Assert(txn || change);
3377 :
3378 : /*
3379 : * Ignore tuple CID changes, because those are not evicted when reaching
3380 : * memory limit. So we just don't count them, because it might easily
3381 : * trigger a pointless attempt to spill.
3382 : */
3383 3825700 : if (change && change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID)
3384 49018 : return;
3385 :
3386 3776682 : if (sz == 0)
3387 1932 : return;
3388 :
3389 3774750 : if (txn == NULL)
3390 3759162 : txn = change->txn;
3391 : Assert(txn != NULL);
3392 :
3393 : /*
3394 : * Update the total size in top level as well. This is later used to
3395 : * compute the decoding stats.
3396 : */
3397 3774750 : toptxn = rbtxn_get_toptxn(txn);
3398 :
3399 3774750 : if (addition)
3400 : {
3401 3411326 : Size oldsize = txn->size;
3402 :
3403 3411326 : txn->size += sz;
3404 3411326 : rb->size += sz;
3405 :
3406 : /* Update the total size in the top transaction. */
3407 3411326 : toptxn->total_size += sz;
3408 :
3409 : /* Update the max-heap */
3410 3411326 : if (oldsize != 0)
3411 3395576 : pairingheap_remove(rb->txn_heap, &txn->txn_node);
3412 3411326 : pairingheap_add(rb->txn_heap, &txn->txn_node);
3413 : }
3414 : else
3415 : {
3416 : Assert((rb->size >= sz) && (txn->size >= sz));
3417 363424 : txn->size -= sz;
3418 363424 : rb->size -= sz;
3419 :
3420 : /* Update the total size in the top transaction. */
3421 363424 : toptxn->total_size -= sz;
3422 :
3423 : /* Update the max-heap */
3424 363424 : pairingheap_remove(rb->txn_heap, &txn->txn_node);
3425 363424 : if (txn->size != 0)
3426 347780 : pairingheap_add(rb->txn_heap, &txn->txn_node);
3427 : }
3428 :
3429 : Assert(txn->size <= rb->size);
3430 : }
3431 :
3432 : /*
3433 : * Add new (relfilelocator, tid) -> (cmin, cmax) mappings.
3434 : *
3435 : * We do not include this change type in memory accounting, because we
3436 : * keep CIDs in a separate list and do not evict them when reaching
3437 : * the memory limit.
3438 : */
3439 : void
3440 49250 : ReorderBufferAddNewTupleCids(ReorderBuffer *rb, TransactionId xid,
3441 : XLogRecPtr lsn, RelFileLocator locator,
3442 : ItemPointerData tid, CommandId cmin,
3443 : CommandId cmax, CommandId combocid)
3444 : {
3445 49250 : ReorderBufferChange *change = ReorderBufferAllocChange(rb);
3446 : ReorderBufferTXN *txn;
3447 :
3448 49250 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3449 :
3450 49250 : change->data.tuplecid.locator = locator;
3451 49250 : change->data.tuplecid.tid = tid;
3452 49250 : change->data.tuplecid.cmin = cmin;
3453 49250 : change->data.tuplecid.cmax = cmax;
3454 49250 : change->data.tuplecid.combocid = combocid;
3455 49250 : change->lsn = lsn;
3456 49250 : change->txn = txn;
3457 49250 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID;
3458 :
3459 49250 : dlist_push_tail(&txn->tuplecids, &change->node);
3460 49250 : txn->ntuplecids++;
3461 49250 : }
3462 :
3463 : /*
3464 : * Add new invalidation messages to the reorder buffer queue.
3465 : */
3466 : static void
3467 10350 : ReorderBufferQueueInvalidations(ReorderBuffer *rb, TransactionId xid,
3468 : XLogRecPtr lsn, Size nmsgs,
3469 : SharedInvalidationMessage *msgs)
3470 : {
3471 : ReorderBufferChange *change;
3472 :
3473 10350 : change = ReorderBufferAllocChange(rb);
3474 10350 : change->action = REORDER_BUFFER_CHANGE_INVALIDATION;
3475 10350 : change->data.inval.ninvalidations = nmsgs;
3476 10350 : change->data.inval.invalidations = (SharedInvalidationMessage *)
3477 10350 : palloc(sizeof(SharedInvalidationMessage) * nmsgs);
3478 10350 : memcpy(change->data.inval.invalidations, msgs,
3479 : sizeof(SharedInvalidationMessage) * nmsgs);
3480 :
3481 10350 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3482 10350 : }
3483 :
3484 : /*
3485 : * A helper function for ReorderBufferAddInvalidations() and
3486 : * ReorderBufferAddDistributedInvalidations() to accumulate the invalidation
3487 : * messages to the **invals_out.
3488 : */
3489 : static void
3490 10350 : ReorderBufferAccumulateInvalidations(SharedInvalidationMessage **invals_out,
3491 : uint32 *ninvals_out,
3492 : SharedInvalidationMessage *msgs_new,
3493 : Size nmsgs_new)
3494 : {
3495 10350 : if (*ninvals_out == 0)
3496 : {
3497 2332 : *ninvals_out = nmsgs_new;
3498 2332 : *invals_out = (SharedInvalidationMessage *)
3499 2332 : palloc(sizeof(SharedInvalidationMessage) * nmsgs_new);
3500 2332 : memcpy(*invals_out, msgs_new, sizeof(SharedInvalidationMessage) * nmsgs_new);
3501 : }
3502 : else
3503 : {
3504 : /* Enlarge the array of inval messages */
3505 8018 : *invals_out = (SharedInvalidationMessage *)
3506 8018 : repalloc(*invals_out, sizeof(SharedInvalidationMessage) *
3507 8018 : (*ninvals_out + nmsgs_new));
3508 8018 : memcpy(*invals_out + *ninvals_out, msgs_new,
3509 : nmsgs_new * sizeof(SharedInvalidationMessage));
3510 8018 : *ninvals_out += nmsgs_new;
3511 : }
3512 10350 : }
3513 :
3514 : /*
3515 : * Accumulate the invalidations for executing them later.
3516 : *
3517 : * This needs to be called for each XLOG_XACT_INVALIDATIONS message and
3518 : * accumulates all the invalidation messages in the toplevel transaction, if
3519 : * available, otherwise in the current transaction, as well as in the form of
3520 : * change in reorder buffer. We require to record it in form of the change
3521 : * so that we can execute only the required invalidations instead of executing
3522 : * all the invalidations on each CommandId increment. We also need to
3523 : * accumulate these in the txn buffer because in some cases where we skip
3524 : * processing the transaction (see ReorderBufferForget), we need to execute
3525 : * all the invalidations together.
3526 : */
3527 : void
3528 10294 : ReorderBufferAddInvalidations(ReorderBuffer *rb, TransactionId xid,
3529 : XLogRecPtr lsn, Size nmsgs,
3530 : SharedInvalidationMessage *msgs)
3531 : {
3532 : ReorderBufferTXN *txn;
3533 : MemoryContext oldcontext;
3534 :
3535 10294 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3536 :
3537 10294 : oldcontext = MemoryContextSwitchTo(rb->context);
3538 :
3539 : /*
3540 : * Collect all the invalidations under the top transaction, if available,
3541 : * so that we can execute them all together. See comments atop this
3542 : * function.
3543 : */
3544 10294 : txn = rbtxn_get_toptxn(txn);
3545 :
3546 : Assert(nmsgs > 0);
3547 :
3548 10294 : ReorderBufferAccumulateInvalidations(&txn->invalidations,
3549 : &txn->ninvalidations,
3550 : msgs, nmsgs);
3551 :
3552 10294 : ReorderBufferQueueInvalidations(rb, xid, lsn, nmsgs, msgs);
3553 :
3554 10294 : MemoryContextSwitchTo(oldcontext);
3555 10294 : }
3556 :
3557 : /*
3558 : * Accumulate the invalidations distributed by other committed transactions
3559 : * for executing them later.
3560 : *
3561 : * This function is similar to ReorderBufferAddInvalidations() but stores
3562 : * the given inval messages to the txn->invalidations_distributed with the
3563 : * overflow check.
3564 : *
3565 : * This needs to be called by committed transactions to distribute their
3566 : * inval messages to in-progress transactions.
3567 : */
3568 : void
3569 56 : ReorderBufferAddDistributedInvalidations(ReorderBuffer *rb, TransactionId xid,
3570 : XLogRecPtr lsn, Size nmsgs,
3571 : SharedInvalidationMessage *msgs)
3572 : {
3573 : ReorderBufferTXN *txn;
3574 : MemoryContext oldcontext;
3575 :
3576 56 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3577 :
3578 56 : oldcontext = MemoryContextSwitchTo(rb->context);
3579 :
3580 : /*
3581 : * Collect all the invalidations under the top transaction, if available,
3582 : * so that we can execute them all together. See comments
3583 : * ReorderBufferAddInvalidations.
3584 : */
3585 56 : txn = rbtxn_get_toptxn(txn);
3586 :
3587 : Assert(nmsgs > 0);
3588 :
3589 56 : if (!rbtxn_distr_inval_overflowed(txn))
3590 : {
3591 : /*
3592 : * Check the transaction has enough space for storing distributed
3593 : * invalidation messages.
3594 : */
3595 56 : if (txn->ninvalidations_distributed + nmsgs >= MAX_DISTR_INVAL_MSG_PER_TXN)
3596 : {
3597 : /*
3598 : * Mark the invalidation message as overflowed and free up the
3599 : * messages accumulated so far.
3600 : */
3601 0 : txn->txn_flags |= RBTXN_DISTR_INVAL_OVERFLOWED;
3602 :
3603 0 : if (txn->invalidations_distributed)
3604 : {
3605 0 : pfree(txn->invalidations_distributed);
3606 0 : txn->invalidations_distributed = NULL;
3607 0 : txn->ninvalidations_distributed = 0;
3608 : }
3609 : }
3610 : else
3611 56 : ReorderBufferAccumulateInvalidations(&txn->invalidations_distributed,
3612 : &txn->ninvalidations_distributed,
3613 : msgs, nmsgs);
3614 : }
3615 :
3616 : /* Queue the invalidation messages into the transaction */
3617 56 : ReorderBufferQueueInvalidations(rb, xid, lsn, nmsgs, msgs);
3618 :
3619 56 : MemoryContextSwitchTo(oldcontext);
3620 56 : }
3621 :
3622 : /*
3623 : * Apply all invalidations we know. Possibly we only need parts at this point
3624 : * in the changestream but we don't know which those are.
3625 : */
3626 : static void
3627 12328 : ReorderBufferExecuteInvalidations(uint32 nmsgs, SharedInvalidationMessage *msgs)
3628 : {
3629 : int i;
3630 :
3631 96564 : for (i = 0; i < nmsgs; i++)
3632 84236 : LocalExecuteInvalidationMessage(&msgs[i]);
3633 12328 : }
3634 :
3635 : /*
3636 : * Mark a transaction as containing catalog changes
3637 : */
3638 : void
3639 59634 : ReorderBufferXidSetCatalogChanges(ReorderBuffer *rb, TransactionId xid,
3640 : XLogRecPtr lsn)
3641 : {
3642 : ReorderBufferTXN *txn;
3643 :
3644 59634 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3645 :
3646 59634 : if (!rbtxn_has_catalog_changes(txn))
3647 : {
3648 2378 : txn->txn_flags |= RBTXN_HAS_CATALOG_CHANGES;
3649 2378 : dclist_push_tail(&rb->catchange_txns, &txn->catchange_node);
3650 : }
3651 :
3652 : /*
3653 : * Mark top-level transaction as having catalog changes too if one of its
3654 : * children has so that the ReorderBufferBuildTupleCidHash can
3655 : * conveniently check just top-level transaction and decide whether to
3656 : * build the hash table or not.
3657 : */
3658 59634 : if (rbtxn_is_subtxn(txn))
3659 : {
3660 1792 : ReorderBufferTXN *toptxn = rbtxn_get_toptxn(txn);
3661 :
3662 1792 : if (!rbtxn_has_catalog_changes(toptxn))
3663 : {
3664 40 : toptxn->txn_flags |= RBTXN_HAS_CATALOG_CHANGES;
3665 40 : dclist_push_tail(&rb->catchange_txns, &toptxn->catchange_node);
3666 : }
3667 : }
3668 59634 : }
3669 :
3670 : /*
3671 : * Return palloc'ed array of the transactions that have changed catalogs.
3672 : * The returned array is sorted in xidComparator order.
3673 : *
3674 : * The caller must free the returned array when done with it.
3675 : */
3676 : TransactionId *
3677 584 : ReorderBufferGetCatalogChangesXacts(ReorderBuffer *rb)
3678 : {
3679 : dlist_iter iter;
3680 584 : TransactionId *xids = NULL;
3681 584 : size_t xcnt = 0;
3682 :
3683 : /* Quick return if the list is empty */
3684 584 : if (dclist_count(&rb->catchange_txns) == 0)
3685 566 : return NULL;
3686 :
3687 : /* Initialize XID array */
3688 18 : xids = (TransactionId *) palloc(sizeof(TransactionId) *
3689 18 : dclist_count(&rb->catchange_txns));
3690 42 : dclist_foreach(iter, &rb->catchange_txns)
3691 : {
3692 24 : ReorderBufferTXN *txn = dclist_container(ReorderBufferTXN,
3693 : catchange_node,
3694 : iter.cur);
3695 :
3696 : Assert(rbtxn_has_catalog_changes(txn));
3697 :
3698 24 : xids[xcnt++] = txn->xid;
3699 : }
3700 :
3701 18 : qsort(xids, xcnt, sizeof(TransactionId), xidComparator);
3702 :
3703 : Assert(xcnt == dclist_count(&rb->catchange_txns));
3704 18 : return xids;
3705 : }
3706 :
3707 : /*
3708 : * Query whether a transaction is already *known* to contain catalog
3709 : * changes. This can be wrong until directly before the commit!
3710 : */
3711 : bool
3712 8588 : ReorderBufferXidHasCatalogChanges(ReorderBuffer *rb, TransactionId xid)
3713 : {
3714 : ReorderBufferTXN *txn;
3715 :
3716 8588 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3717 : false);
3718 8588 : if (txn == NULL)
3719 1294 : return false;
3720 :
3721 7294 : return rbtxn_has_catalog_changes(txn);
3722 : }
3723 :
3724 : /*
3725 : * ReorderBufferXidHasBaseSnapshot
3726 : * Have we already set the base snapshot for the given txn/subtxn?
3727 : */
3728 : bool
3729 3043460 : ReorderBufferXidHasBaseSnapshot(ReorderBuffer *rb, TransactionId xid)
3730 : {
3731 : ReorderBufferTXN *txn;
3732 :
3733 3043460 : txn = ReorderBufferTXNByXid(rb, xid, false,
3734 : NULL, InvalidXLogRecPtr, false);
3735 :
3736 : /* transaction isn't known yet, ergo no snapshot */
3737 3043460 : if (txn == NULL)
3738 6 : return false;
3739 :
3740 : /* a known subtxn? operate on top-level txn instead */
3741 3043454 : if (rbtxn_is_known_subxact(txn))
3742 984074 : txn = ReorderBufferTXNByXid(rb, txn->toplevel_xid, false,
3743 : NULL, InvalidXLogRecPtr, false);
3744 :
3745 3043454 : return txn->base_snapshot != NULL;
3746 : }
3747 :
3748 :
3749 : /*
3750 : * ---------------------------------------
3751 : * Disk serialization support
3752 : * ---------------------------------------
3753 : */
3754 :
3755 : /*
3756 : * Ensure the IO buffer is >= sz.
3757 : */
3758 : static void
3759 5822676 : ReorderBufferSerializeReserve(ReorderBuffer *rb, Size sz)
3760 : {
3761 5822676 : if (!rb->outbufsize)
3762 : {
3763 90 : rb->outbuf = MemoryContextAlloc(rb->context, sz);
3764 90 : rb->outbufsize = sz;
3765 : }
3766 5822586 : else if (rb->outbufsize < sz)
3767 : {
3768 574 : rb->outbuf = repalloc(rb->outbuf, sz);
3769 574 : rb->outbufsize = sz;
3770 : }
3771 5822676 : }
3772 :
3773 :
3774 : /* Compare two transactions by size */
3775 : static int
3776 648930 : ReorderBufferTXNSizeCompare(const pairingheap_node *a, const pairingheap_node *b, void *arg)
3777 : {
3778 648930 : const ReorderBufferTXN *ta = pairingheap_const_container(ReorderBufferTXN, txn_node, a);
3779 648930 : const ReorderBufferTXN *tb = pairingheap_const_container(ReorderBufferTXN, txn_node, b);
3780 :
3781 648930 : if (ta->size < tb->size)
3782 465316 : return -1;
3783 183614 : if (ta->size > tb->size)
3784 181820 : return 1;
3785 1794 : return 0;
3786 : }
3787 :
3788 : /*
3789 : * Find the largest transaction (toplevel or subxact) to evict (spill to disk).
3790 : */
3791 : static ReorderBufferTXN *
3792 7348 : ReorderBufferLargestTXN(ReorderBuffer *rb)
3793 : {
3794 : ReorderBufferTXN *largest;
3795 :
3796 : /* Get the largest transaction from the max-heap */
3797 7348 : largest = pairingheap_container(ReorderBufferTXN, txn_node,
3798 : pairingheap_first(rb->txn_heap));
3799 :
3800 : Assert(largest);
3801 : Assert(largest->size > 0);
3802 : Assert(largest->size <= rb->size);
3803 :
3804 7348 : return largest;
3805 : }
3806 :
3807 : /*
3808 : * Find the largest streamable (and non-aborted) toplevel transaction to evict
3809 : * (by streaming).
3810 : *
3811 : * This can be seen as an optimized version of ReorderBufferLargestTXN, which
3812 : * should give us the same transaction (because we don't update memory account
3813 : * for subtransaction with streaming, so it's always 0). But we can simply
3814 : * iterate over the limited number of toplevel transactions that have a base
3815 : * snapshot. There is no use of selecting a transaction that doesn't have base
3816 : * snapshot because we don't decode such transactions. Also, we do not select
3817 : * the transaction which doesn't have any streamable change.
3818 : *
3819 : * Note that, we skip transactions that contain incomplete changes. There
3820 : * is a scope of optimization here such that we can select the largest
3821 : * transaction which has incomplete changes. But that will make the code and
3822 : * design quite complex and that might not be worth the benefit. If we plan to
3823 : * stream the transactions that contain incomplete changes then we need to
3824 : * find a way to partially stream/truncate the transaction changes in-memory
3825 : * and build a mechanism to partially truncate the spilled files.
3826 : * Additionally, whenever we partially stream the transaction we need to
3827 : * maintain the last streamed lsn and next time we need to restore from that
3828 : * segment and the offset in WAL. As we stream the changes from the top
3829 : * transaction and restore them subtransaction wise, we need to even remember
3830 : * the subxact from where we streamed the last change.
3831 : */
3832 : static ReorderBufferTXN *
3833 1576 : ReorderBufferLargestStreamableTopTXN(ReorderBuffer *rb)
3834 : {
3835 : dlist_iter iter;
3836 1576 : Size largest_size = 0;
3837 1576 : ReorderBufferTXN *largest = NULL;
3838 :
3839 : /* Find the largest top-level transaction having a base snapshot. */
3840 3376 : dlist_foreach(iter, &rb->txns_by_base_snapshot_lsn)
3841 : {
3842 : ReorderBufferTXN *txn;
3843 :
3844 1800 : txn = dlist_container(ReorderBufferTXN, base_snapshot_node, iter.cur);
3845 :
3846 : /* must not be a subtxn */
3847 : Assert(!rbtxn_is_known_subxact(txn));
3848 : /* base_snapshot must be set */
3849 : Assert(txn->base_snapshot != NULL);
3850 :
3851 : /* Don't consider these kinds of transactions for eviction. */
3852 1800 : if (rbtxn_has_partial_change(txn) ||
3853 1506 : !rbtxn_has_streamable_change(txn) ||
3854 1446 : rbtxn_is_aborted(txn))
3855 354 : continue;
3856 :
3857 : /* Find the largest of the eviction candidates. */
3858 1446 : if ((largest == NULL || txn->total_size > largest_size) &&
3859 1446 : (txn->total_size > 0))
3860 : {
3861 1354 : largest = txn;
3862 1354 : largest_size = txn->total_size;
3863 : }
3864 : }
3865 :
3866 1576 : return largest;
3867 : }
3868 :
3869 : /*
3870 : * Check whether the logical_decoding_work_mem limit was reached, and if yes
3871 : * pick the largest (sub)transaction at-a-time to evict and spill its changes to
3872 : * disk or send to the output plugin until we reach under the memory limit.
3873 : *
3874 : * If debug_logical_replication_streaming is set to "immediate", stream or
3875 : * serialize the changes immediately.
3876 : *
3877 : * XXX At this point we select the transactions until we reach under the memory
3878 : * limit, but we might also adapt a more elaborate eviction strategy - for example
3879 : * evicting enough transactions to free certain fraction (e.g. 50%) of the memory
3880 : * limit.
3881 : */
3882 : static void
3883 3063516 : ReorderBufferCheckMemoryLimit(ReorderBuffer *rb)
3884 : {
3885 : ReorderBufferTXN *txn;
3886 :
3887 : /*
3888 : * Bail out if debug_logical_replication_streaming is buffered and we
3889 : * haven't exceeded the memory limit.
3890 : */
3891 3063516 : if (debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_BUFFERED &&
3892 3061746 : rb->size < logical_decoding_work_mem * (Size) 1024)
3893 3054940 : return;
3894 :
3895 : /*
3896 : * If debug_logical_replication_streaming is immediate, loop until there's
3897 : * no change. Otherwise, loop until we reach under the memory limit. One
3898 : * might think that just by evicting the largest (sub)transaction we will
3899 : * come under the memory limit based on assumption that the selected
3900 : * transaction is at least as large as the most recent change (which
3901 : * caused us to go over the memory limit). However, that is not true
3902 : * because a user can reduce the logical_decoding_work_mem to a smaller
3903 : * value before the most recent change.
3904 : */
3905 17146 : while (rb->size >= logical_decoding_work_mem * (Size) 1024 ||
3906 10340 : (debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_IMMEDIATE &&
3907 3534 : rb->size > 0))
3908 : {
3909 : /*
3910 : * Pick the largest non-aborted transaction and evict it from memory
3911 : * by streaming, if possible. Otherwise, spill to disk.
3912 : */
3913 10146 : if (ReorderBufferCanStartStreaming(rb) &&
3914 1576 : (txn = ReorderBufferLargestStreamableTopTXN(rb)) != NULL)
3915 : {
3916 : /* we know there has to be one, because the size is not zero */
3917 : Assert(txn && rbtxn_is_toptxn(txn));
3918 : Assert(txn->total_size > 0);
3919 : Assert(rb->size >= txn->total_size);
3920 :
3921 : /* skip the transaction if aborted */
3922 1222 : if (ReorderBufferCheckAndTruncateAbortedTXN(rb, txn))
3923 0 : continue;
3924 :
3925 1222 : ReorderBufferStreamTXN(rb, txn);
3926 : }
3927 : else
3928 : {
3929 : /*
3930 : * Pick the largest transaction (or subtransaction) and evict it
3931 : * from memory by serializing it to disk.
3932 : */
3933 7348 : txn = ReorderBufferLargestTXN(rb);
3934 :
3935 : /* we know there has to be one, because the size is not zero */
3936 : Assert(txn);
3937 : Assert(txn->size > 0);
3938 : Assert(rb->size >= txn->size);
3939 :
3940 : /* skip the transaction if aborted */
3941 7348 : if (ReorderBufferCheckAndTruncateAbortedTXN(rb, txn))
3942 18 : continue;
3943 :
3944 7330 : ReorderBufferSerializeTXN(rb, txn);
3945 : }
3946 :
3947 : /*
3948 : * After eviction, the transaction should have no entries in memory,
3949 : * and should use 0 bytes for changes.
3950 : */
3951 : Assert(txn->size == 0);
3952 : Assert(txn->nentries_mem == 0);
3953 : }
3954 :
3955 : /* We must be under the memory limit now. */
3956 : Assert(rb->size < logical_decoding_work_mem * (Size) 1024);
3957 : }
3958 :
3959 : /*
3960 : * Spill data of a large transaction (and its subtransactions) to disk.
3961 : */
3962 : static void
3963 7946 : ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
3964 : {
3965 : dlist_iter subtxn_i;
3966 : dlist_mutable_iter change_i;
3967 7946 : int fd = -1;
3968 7946 : XLogSegNo curOpenSegNo = 0;
3969 7946 : Size spilled = 0;
3970 7946 : Size size = txn->size;
3971 :
3972 7946 : elog(DEBUG2, "spill %u changes in XID %u to disk",
3973 : (uint32) txn->nentries_mem, txn->xid);
3974 :
3975 : /* do the same to all child TXs */
3976 8484 : dlist_foreach(subtxn_i, &txn->subtxns)
3977 : {
3978 : ReorderBufferTXN *subtxn;
3979 :
3980 538 : subtxn = dlist_container(ReorderBufferTXN, node, subtxn_i.cur);
3981 538 : ReorderBufferSerializeTXN(rb, subtxn);
3982 : }
3983 :
3984 : /* serialize changestream */
3985 2589236 : dlist_foreach_modify(change_i, &txn->changes)
3986 : {
3987 : ReorderBufferChange *change;
3988 :
3989 2581290 : change = dlist_container(ReorderBufferChange, node, change_i.cur);
3990 :
3991 : /*
3992 : * store in segment in which it belongs by start lsn, don't split over
3993 : * multiple segments tho
3994 : */
3995 2581290 : if (fd == -1 ||
3996 2573848 : !XLByteInSeg(change->lsn, curOpenSegNo, wal_segment_size))
3997 : {
3998 : char path[MAXPGPATH];
3999 :
4000 7460 : if (fd != -1)
4001 18 : CloseTransientFile(fd);
4002 :
4003 7460 : XLByteToSeg(change->lsn, curOpenSegNo, wal_segment_size);
4004 :
4005 : /*
4006 : * No need to care about TLIs here, only used during a single run,
4007 : * so each LSN only maps to a specific WAL record.
4008 : */
4009 7460 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid,
4010 : curOpenSegNo);
4011 :
4012 : /* open segment, create it if necessary */
4013 7460 : fd = OpenTransientFile(path,
4014 : O_CREAT | O_WRONLY | O_APPEND | PG_BINARY);
4015 :
4016 7460 : if (fd < 0)
4017 0 : ereport(ERROR,
4018 : (errcode_for_file_access(),
4019 : errmsg("could not open file \"%s\": %m", path)));
4020 : }
4021 :
4022 2581290 : ReorderBufferSerializeChange(rb, txn, fd, change);
4023 2581290 : dlist_delete(&change->node);
4024 2581290 : ReorderBufferFreeChange(rb, change, false);
4025 :
4026 2581290 : spilled++;
4027 : }
4028 :
4029 : /* Update the memory counter */
4030 7946 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, size);
4031 :
4032 : /* update the statistics iff we have spilled anything */
4033 7946 : if (spilled)
4034 : {
4035 7442 : rb->spillCount += 1;
4036 7442 : rb->spillBytes += size;
4037 :
4038 : /* don't consider already serialized transactions */
4039 7442 : rb->spillTxns += (rbtxn_is_serialized(txn) || rbtxn_is_serialized_clear(txn)) ? 0 : 1;
4040 :
4041 : /* update the decoding stats */
4042 7442 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
4043 : }
4044 :
4045 : Assert(spilled == txn->nentries_mem);
4046 : Assert(dlist_is_empty(&txn->changes));
4047 7946 : txn->nentries_mem = 0;
4048 7946 : txn->txn_flags |= RBTXN_IS_SERIALIZED;
4049 :
4050 7946 : if (fd != -1)
4051 7442 : CloseTransientFile(fd);
4052 7946 : }
4053 :
4054 : /*
4055 : * Serialize individual change to disk.
4056 : */
4057 : static void
4058 2581290 : ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
4059 : int fd, ReorderBufferChange *change)
4060 : {
4061 : ReorderBufferDiskChange *ondisk;
4062 2581290 : Size sz = sizeof(ReorderBufferDiskChange);
4063 :
4064 2581290 : ReorderBufferSerializeReserve(rb, sz);
4065 :
4066 2581290 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4067 2581290 : memcpy(&ondisk->change, change, sizeof(ReorderBufferChange));
4068 :
4069 2581290 : switch (change->action)
4070 : {
4071 : /* fall through these, they're all similar enough */
4072 2546314 : case REORDER_BUFFER_CHANGE_INSERT:
4073 : case REORDER_BUFFER_CHANGE_UPDATE:
4074 : case REORDER_BUFFER_CHANGE_DELETE:
4075 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4076 : {
4077 : char *data;
4078 : HeapTuple oldtup,
4079 : newtup;
4080 2546314 : Size oldlen = 0;
4081 2546314 : Size newlen = 0;
4082 :
4083 2546314 : oldtup = change->data.tp.oldtuple;
4084 2546314 : newtup = change->data.tp.newtuple;
4085 :
4086 2546314 : if (oldtup)
4087 : {
4088 193976 : sz += sizeof(HeapTupleData);
4089 193976 : oldlen = oldtup->t_len;
4090 193976 : sz += oldlen;
4091 : }
4092 :
4093 2546314 : if (newtup)
4094 : {
4095 2244908 : sz += sizeof(HeapTupleData);
4096 2244908 : newlen = newtup->t_len;
4097 2244908 : sz += newlen;
4098 : }
4099 :
4100 : /* make sure we have enough space */
4101 2546314 : ReorderBufferSerializeReserve(rb, sz);
4102 :
4103 2546314 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4104 : /* might have been reallocated above */
4105 2546314 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4106 :
4107 2546314 : if (oldlen)
4108 : {
4109 193976 : memcpy(data, oldtup, sizeof(HeapTupleData));
4110 193976 : data += sizeof(HeapTupleData);
4111 :
4112 193976 : memcpy(data, oldtup->t_data, oldlen);
4113 193976 : data += oldlen;
4114 : }
4115 :
4116 2546314 : if (newlen)
4117 : {
4118 2244908 : memcpy(data, newtup, sizeof(HeapTupleData));
4119 2244908 : data += sizeof(HeapTupleData);
4120 :
4121 2244908 : memcpy(data, newtup->t_data, newlen);
4122 2244908 : data += newlen;
4123 : }
4124 2546314 : break;
4125 : }
4126 26 : case REORDER_BUFFER_CHANGE_MESSAGE:
4127 : {
4128 : char *data;
4129 26 : Size prefix_size = strlen(change->data.msg.prefix) + 1;
4130 :
4131 26 : sz += prefix_size + change->data.msg.message_size +
4132 : sizeof(Size) + sizeof(Size);
4133 26 : ReorderBufferSerializeReserve(rb, sz);
4134 :
4135 26 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4136 :
4137 : /* might have been reallocated above */
4138 26 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4139 :
4140 : /* write the prefix including the size */
4141 26 : memcpy(data, &prefix_size, sizeof(Size));
4142 26 : data += sizeof(Size);
4143 26 : memcpy(data, change->data.msg.prefix,
4144 : prefix_size);
4145 26 : data += prefix_size;
4146 :
4147 : /* write the message including the size */
4148 26 : memcpy(data, &change->data.msg.message_size, sizeof(Size));
4149 26 : data += sizeof(Size);
4150 26 : memcpy(data, change->data.msg.message,
4151 : change->data.msg.message_size);
4152 26 : data += change->data.msg.message_size;
4153 :
4154 26 : break;
4155 : }
4156 308 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4157 : {
4158 : char *data;
4159 308 : Size inval_size = sizeof(SharedInvalidationMessage) *
4160 308 : change->data.inval.ninvalidations;
4161 :
4162 308 : sz += inval_size;
4163 :
4164 308 : ReorderBufferSerializeReserve(rb, sz);
4165 308 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4166 :
4167 : /* might have been reallocated above */
4168 308 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4169 308 : memcpy(data, change->data.inval.invalidations, inval_size);
4170 308 : data += inval_size;
4171 :
4172 308 : break;
4173 : }
4174 16 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4175 : {
4176 : Snapshot snap;
4177 : char *data;
4178 :
4179 16 : snap = change->data.snapshot;
4180 :
4181 16 : sz += sizeof(SnapshotData) +
4182 16 : sizeof(TransactionId) * snap->xcnt +
4183 16 : sizeof(TransactionId) * snap->subxcnt;
4184 :
4185 : /* make sure we have enough space */
4186 16 : ReorderBufferSerializeReserve(rb, sz);
4187 16 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4188 : /* might have been reallocated above */
4189 16 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4190 :
4191 16 : memcpy(data, snap, sizeof(SnapshotData));
4192 16 : data += sizeof(SnapshotData);
4193 :
4194 16 : if (snap->xcnt)
4195 : {
4196 16 : memcpy(data, snap->xip,
4197 16 : sizeof(TransactionId) * snap->xcnt);
4198 16 : data += sizeof(TransactionId) * snap->xcnt;
4199 : }
4200 :
4201 16 : if (snap->subxcnt)
4202 : {
4203 0 : memcpy(data, snap->subxip,
4204 0 : sizeof(TransactionId) * snap->subxcnt);
4205 0 : data += sizeof(TransactionId) * snap->subxcnt;
4206 : }
4207 16 : break;
4208 : }
4209 4 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4210 : {
4211 : Size size;
4212 : char *data;
4213 :
4214 : /* account for the OIDs of truncated relations */
4215 4 : size = sizeof(Oid) * change->data.truncate.nrelids;
4216 4 : sz += size;
4217 :
4218 : /* make sure we have enough space */
4219 4 : ReorderBufferSerializeReserve(rb, sz);
4220 :
4221 4 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4222 : /* might have been reallocated above */
4223 4 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4224 :
4225 4 : memcpy(data, change->data.truncate.relids, size);
4226 4 : data += size;
4227 :
4228 4 : break;
4229 : }
4230 34622 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4231 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4232 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4233 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4234 : /* ReorderBufferChange contains everything important */
4235 34622 : break;
4236 : }
4237 :
4238 2581290 : ondisk->size = sz;
4239 :
4240 2581290 : errno = 0;
4241 2581290 : pgstat_report_wait_start(WAIT_EVENT_REORDER_BUFFER_WRITE);
4242 2581290 : if (write(fd, rb->outbuf, ondisk->size) != ondisk->size)
4243 : {
4244 0 : int save_errno = errno;
4245 :
4246 0 : CloseTransientFile(fd);
4247 :
4248 : /* if write didn't set errno, assume problem is no disk space */
4249 0 : errno = save_errno ? save_errno : ENOSPC;
4250 0 : ereport(ERROR,
4251 : (errcode_for_file_access(),
4252 : errmsg("could not write to data file for XID %u: %m",
4253 : txn->xid)));
4254 : }
4255 2581290 : pgstat_report_wait_end();
4256 :
4257 : /*
4258 : * Keep the transaction's final_lsn up to date with each change we send to
4259 : * disk, so that ReorderBufferRestoreCleanup works correctly. (We used to
4260 : * only do this on commit and abort records, but that doesn't work if a
4261 : * system crash leaves a transaction without its abort record).
4262 : *
4263 : * Make sure not to move it backwards.
4264 : */
4265 2581290 : if (txn->final_lsn < change->lsn)
4266 2572324 : txn->final_lsn = change->lsn;
4267 :
4268 : Assert(ondisk->change.action == change->action);
4269 2581290 : }
4270 :
4271 : /* Returns true, if the output plugin supports streaming, false, otherwise. */
4272 : static inline bool
4273 3715182 : ReorderBufferCanStream(ReorderBuffer *rb)
4274 : {
4275 3715182 : LogicalDecodingContext *ctx = rb->private_data;
4276 :
4277 3715182 : return ctx->streaming;
4278 : }
4279 :
4280 : /* Returns true, if the streaming can be started now, false, otherwise. */
4281 : static inline bool
4282 651666 : ReorderBufferCanStartStreaming(ReorderBuffer *rb)
4283 : {
4284 651666 : LogicalDecodingContext *ctx = rb->private_data;
4285 651666 : SnapBuild *builder = ctx->snapshot_builder;
4286 :
4287 : /* We can't start streaming unless a consistent state is reached. */
4288 651666 : if (SnapBuildCurrentState(builder) < SNAPBUILD_CONSISTENT)
4289 0 : return false;
4290 :
4291 : /*
4292 : * We can't start streaming immediately even if the streaming is enabled
4293 : * because we previously decoded this transaction and now just are
4294 : * restarting.
4295 : */
4296 651666 : if (ReorderBufferCanStream(rb) &&
4297 646370 : !SnapBuildXactNeedsSkip(builder, ctx->reader->ReadRecPtr))
4298 360674 : return true;
4299 :
4300 290992 : return false;
4301 : }
4302 :
4303 : /*
4304 : * Send data of a large transaction (and its subtransactions) to the
4305 : * output plugin, but using the stream API.
4306 : */
4307 : static void
4308 1354 : ReorderBufferStreamTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
4309 : {
4310 : Snapshot snapshot_now;
4311 : CommandId command_id;
4312 : Size stream_bytes;
4313 : bool txn_is_streamed;
4314 :
4315 : /* We can never reach here for a subtransaction. */
4316 : Assert(rbtxn_is_toptxn(txn));
4317 :
4318 : /*
4319 : * We can't make any assumptions about base snapshot here, similar to what
4320 : * ReorderBufferCommit() does. That relies on base_snapshot getting
4321 : * transferred from subxact in ReorderBufferCommitChild(), but that was
4322 : * not yet called as the transaction is in-progress.
4323 : *
4324 : * So just walk the subxacts and use the same logic here. But we only need
4325 : * to do that once, when the transaction is streamed for the first time.
4326 : * After that we need to reuse the snapshot from the previous run.
4327 : *
4328 : * Unlike DecodeCommit which adds xids of all the subtransactions in
4329 : * snapshot's xip array via SnapBuildCommitTxn, we can't do that here but
4330 : * we do add them to subxip array instead via ReorderBufferCopySnap. This
4331 : * allows the catalog changes made in subtransactions decoded till now to
4332 : * be visible.
4333 : */
4334 1354 : if (txn->snapshot_now == NULL)
4335 : {
4336 : dlist_iter subxact_i;
4337 :
4338 : /* make sure this transaction is streamed for the first time */
4339 : Assert(!rbtxn_is_streamed(txn));
4340 :
4341 : /* at the beginning we should have invalid command ID */
4342 : Assert(txn->command_id == InvalidCommandId);
4343 :
4344 136 : dlist_foreach(subxact_i, &txn->subtxns)
4345 : {
4346 : ReorderBufferTXN *subtxn;
4347 :
4348 8 : subtxn = dlist_container(ReorderBufferTXN, node, subxact_i.cur);
4349 8 : ReorderBufferTransferSnapToParent(txn, subtxn);
4350 : }
4351 :
4352 : /*
4353 : * If this transaction has no snapshot, it didn't make any changes to
4354 : * the database till now, so there's nothing to decode.
4355 : */
4356 128 : if (txn->base_snapshot == NULL)
4357 : {
4358 : Assert(txn->ninvalidations == 0);
4359 0 : return;
4360 : }
4361 :
4362 128 : command_id = FirstCommandId;
4363 128 : snapshot_now = ReorderBufferCopySnap(rb, txn->base_snapshot,
4364 : txn, command_id);
4365 : }
4366 : else
4367 : {
4368 : /* the transaction must have been already streamed */
4369 : Assert(rbtxn_is_streamed(txn));
4370 :
4371 : /*
4372 : * Nah, we already have snapshot from the previous streaming run. We
4373 : * assume new subxacts can't move the LSN backwards, and so can't beat
4374 : * the LSN condition in the previous branch (so no need to walk
4375 : * through subxacts again). In fact, we must not do that as we may be
4376 : * using snapshot half-way through the subxact.
4377 : */
4378 1226 : command_id = txn->command_id;
4379 :
4380 : /*
4381 : * We can't use txn->snapshot_now directly because after the last
4382 : * streaming run, we might have got some new sub-transactions. So we
4383 : * need to add them to the snapshot.
4384 : */
4385 1226 : snapshot_now = ReorderBufferCopySnap(rb, txn->snapshot_now,
4386 : txn, command_id);
4387 :
4388 : /* Free the previously copied snapshot. */
4389 : Assert(txn->snapshot_now->copied);
4390 1226 : ReorderBufferFreeSnap(rb, txn->snapshot_now);
4391 1226 : txn->snapshot_now = NULL;
4392 : }
4393 :
4394 : /*
4395 : * Remember this information to be used later to update stats. We can't
4396 : * update the stats here as an error while processing the changes would
4397 : * lead to the accumulation of stats even though we haven't streamed all
4398 : * the changes.
4399 : */
4400 1354 : txn_is_streamed = rbtxn_is_streamed(txn);
4401 1354 : stream_bytes = txn->total_size;
4402 :
4403 : /* Process and send the changes to output plugin. */
4404 1354 : ReorderBufferProcessTXN(rb, txn, InvalidXLogRecPtr, snapshot_now,
4405 : command_id, true);
4406 :
4407 1354 : rb->streamCount += 1;
4408 1354 : rb->streamBytes += stream_bytes;
4409 :
4410 : /* Don't consider already streamed transaction. */
4411 1354 : rb->streamTxns += (txn_is_streamed) ? 0 : 1;
4412 :
4413 : /* update the decoding stats */
4414 1354 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
4415 :
4416 : Assert(dlist_is_empty(&txn->changes));
4417 : Assert(txn->nentries == 0);
4418 : Assert(txn->nentries_mem == 0);
4419 : }
4420 :
4421 : /*
4422 : * Size of a change in memory.
4423 : */
4424 : static Size
4425 4290110 : ReorderBufferChangeSize(ReorderBufferChange *change)
4426 : {
4427 4290110 : Size sz = sizeof(ReorderBufferChange);
4428 :
4429 4290110 : switch (change->action)
4430 : {
4431 : /* fall through these, they're all similar enough */
4432 4073074 : case REORDER_BUFFER_CHANGE_INSERT:
4433 : case REORDER_BUFFER_CHANGE_UPDATE:
4434 : case REORDER_BUFFER_CHANGE_DELETE:
4435 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4436 : {
4437 : HeapTuple oldtup,
4438 : newtup;
4439 4073074 : Size oldlen = 0;
4440 4073074 : Size newlen = 0;
4441 :
4442 4073074 : oldtup = change->data.tp.oldtuple;
4443 4073074 : newtup = change->data.tp.newtuple;
4444 :
4445 4073074 : if (oldtup)
4446 : {
4447 390670 : sz += sizeof(HeapTupleData);
4448 390670 : oldlen = oldtup->t_len;
4449 390670 : sz += oldlen;
4450 : }
4451 :
4452 4073074 : if (newtup)
4453 : {
4454 3516082 : sz += sizeof(HeapTupleData);
4455 3516082 : newlen = newtup->t_len;
4456 3516082 : sz += newlen;
4457 : }
4458 :
4459 4073074 : break;
4460 : }
4461 134 : case REORDER_BUFFER_CHANGE_MESSAGE:
4462 : {
4463 134 : Size prefix_size = strlen(change->data.msg.prefix) + 1;
4464 :
4465 134 : sz += prefix_size + change->data.msg.message_size +
4466 : sizeof(Size) + sizeof(Size);
4467 :
4468 134 : break;
4469 : }
4470 20274 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4471 : {
4472 20274 : sz += sizeof(SharedInvalidationMessage) *
4473 20274 : change->data.inval.ninvalidations;
4474 20274 : break;
4475 : }
4476 4636 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4477 : {
4478 : Snapshot snap;
4479 :
4480 4636 : snap = change->data.snapshot;
4481 :
4482 4636 : sz += sizeof(SnapshotData) +
4483 4636 : sizeof(TransactionId) * snap->xcnt +
4484 4636 : sizeof(TransactionId) * snap->subxcnt;
4485 :
4486 4636 : break;
4487 : }
4488 166 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4489 : {
4490 166 : sz += sizeof(Oid) * change->data.truncate.nrelids;
4491 :
4492 166 : break;
4493 : }
4494 191826 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4495 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4496 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4497 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4498 : /* ReorderBufferChange contains everything important */
4499 191826 : break;
4500 : }
4501 :
4502 4290110 : return sz;
4503 : }
4504 :
4505 :
4506 : /*
4507 : * Restore a number of changes spilled to disk back into memory.
4508 : */
4509 : static Size
4510 204 : ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
4511 : TXNEntryFile *file, XLogSegNo *segno)
4512 : {
4513 204 : Size restored = 0;
4514 : XLogSegNo last_segno;
4515 : dlist_mutable_iter cleanup_iter;
4516 204 : File *fd = &file->vfd;
4517 :
4518 : Assert(txn->first_lsn != InvalidXLogRecPtr);
4519 : Assert(txn->final_lsn != InvalidXLogRecPtr);
4520 :
4521 : /* free current entries, so we have memory for more */
4522 340200 : dlist_foreach_modify(cleanup_iter, &txn->changes)
4523 : {
4524 339996 : ReorderBufferChange *cleanup =
4525 339996 : dlist_container(ReorderBufferChange, node, cleanup_iter.cur);
4526 :
4527 339996 : dlist_delete(&cleanup->node);
4528 339996 : ReorderBufferFreeChange(rb, cleanup, true);
4529 : }
4530 204 : txn->nentries_mem = 0;
4531 : Assert(dlist_is_empty(&txn->changes));
4532 :
4533 204 : XLByteToSeg(txn->final_lsn, last_segno, wal_segment_size);
4534 :
4535 347604 : while (restored < max_changes_in_memory && *segno <= last_segno)
4536 : {
4537 : int readBytes;
4538 : ReorderBufferDiskChange *ondisk;
4539 :
4540 347400 : CHECK_FOR_INTERRUPTS();
4541 :
4542 347400 : if (*fd == -1)
4543 : {
4544 : char path[MAXPGPATH];
4545 :
4546 : /* first time in */
4547 82 : if (*segno == 0)
4548 78 : XLByteToSeg(txn->first_lsn, *segno, wal_segment_size);
4549 :
4550 : Assert(*segno != 0 || dlist_is_empty(&txn->changes));
4551 :
4552 : /*
4553 : * No need to care about TLIs here, only used during a single run,
4554 : * so each LSN only maps to a specific WAL record.
4555 : */
4556 82 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid,
4557 : *segno);
4558 :
4559 82 : *fd = PathNameOpenFile(path, O_RDONLY | PG_BINARY);
4560 :
4561 : /* No harm in resetting the offset even in case of failure */
4562 82 : file->curOffset = 0;
4563 :
4564 82 : if (*fd < 0 && errno == ENOENT)
4565 : {
4566 0 : *fd = -1;
4567 0 : (*segno)++;
4568 0 : continue;
4569 : }
4570 82 : else if (*fd < 0)
4571 0 : ereport(ERROR,
4572 : (errcode_for_file_access(),
4573 : errmsg("could not open file \"%s\": %m",
4574 : path)));
4575 : }
4576 :
4577 : /*
4578 : * Read the statically sized part of a change which has information
4579 : * about the total size. If we couldn't read a record, we're at the
4580 : * end of this file.
4581 : */
4582 347400 : ReorderBufferSerializeReserve(rb, sizeof(ReorderBufferDiskChange));
4583 347400 : readBytes = FileRead(file->vfd, rb->outbuf,
4584 : sizeof(ReorderBufferDiskChange),
4585 : file->curOffset, WAIT_EVENT_REORDER_BUFFER_READ);
4586 :
4587 : /* eof */
4588 347400 : if (readBytes == 0)
4589 : {
4590 82 : FileClose(*fd);
4591 82 : *fd = -1;
4592 82 : (*segno)++;
4593 82 : continue;
4594 : }
4595 347318 : else if (readBytes < 0)
4596 0 : ereport(ERROR,
4597 : (errcode_for_file_access(),
4598 : errmsg("could not read from reorderbuffer spill file: %m")));
4599 347318 : else if (readBytes != sizeof(ReorderBufferDiskChange))
4600 0 : ereport(ERROR,
4601 : (errcode_for_file_access(),
4602 : errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
4603 : readBytes,
4604 : (uint32) sizeof(ReorderBufferDiskChange))));
4605 :
4606 347318 : file->curOffset += readBytes;
4607 :
4608 347318 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4609 :
4610 347318 : ReorderBufferSerializeReserve(rb,
4611 347318 : sizeof(ReorderBufferDiskChange) + ondisk->size);
4612 347318 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4613 :
4614 694636 : readBytes = FileRead(file->vfd,
4615 347318 : rb->outbuf + sizeof(ReorderBufferDiskChange),
4616 347318 : ondisk->size - sizeof(ReorderBufferDiskChange),
4617 : file->curOffset,
4618 : WAIT_EVENT_REORDER_BUFFER_READ);
4619 :
4620 347318 : if (readBytes < 0)
4621 0 : ereport(ERROR,
4622 : (errcode_for_file_access(),
4623 : errmsg("could not read from reorderbuffer spill file: %m")));
4624 347318 : else if (readBytes != ondisk->size - sizeof(ReorderBufferDiskChange))
4625 0 : ereport(ERROR,
4626 : (errcode_for_file_access(),
4627 : errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
4628 : readBytes,
4629 : (uint32) (ondisk->size - sizeof(ReorderBufferDiskChange)))));
4630 :
4631 347318 : file->curOffset += readBytes;
4632 :
4633 : /*
4634 : * ok, read a full change from disk, now restore it into proper
4635 : * in-memory format
4636 : */
4637 347318 : ReorderBufferRestoreChange(rb, txn, rb->outbuf);
4638 347318 : restored++;
4639 : }
4640 :
4641 204 : return restored;
4642 : }
4643 :
4644 : /*
4645 : * Convert change from its on-disk format to in-memory format and queue it onto
4646 : * the TXN's ->changes list.
4647 : *
4648 : * Note: although "data" is declared char*, at entry it points to a
4649 : * maxalign'd buffer, making it safe in most of this function to assume
4650 : * that the pointed-to data is suitably aligned for direct access.
4651 : */
4652 : static void
4653 347318 : ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
4654 : char *data)
4655 : {
4656 : ReorderBufferDiskChange *ondisk;
4657 : ReorderBufferChange *change;
4658 :
4659 347318 : ondisk = (ReorderBufferDiskChange *) data;
4660 :
4661 347318 : change = ReorderBufferAllocChange(rb);
4662 :
4663 : /* copy static part */
4664 347318 : memcpy(change, &ondisk->change, sizeof(ReorderBufferChange));
4665 :
4666 347318 : data += sizeof(ReorderBufferDiskChange);
4667 :
4668 : /* restore individual stuff */
4669 347318 : switch (change->action)
4670 : {
4671 : /* fall through these, they're all similar enough */
4672 343460 : case REORDER_BUFFER_CHANGE_INSERT:
4673 : case REORDER_BUFFER_CHANGE_UPDATE:
4674 : case REORDER_BUFFER_CHANGE_DELETE:
4675 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4676 343460 : if (change->data.tp.oldtuple)
4677 : {
4678 10012 : uint32 tuplelen = ((HeapTuple) data)->t_len;
4679 :
4680 10012 : change->data.tp.oldtuple =
4681 10012 : ReorderBufferAllocTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
4682 :
4683 : /* restore ->tuple */
4684 10012 : memcpy(change->data.tp.oldtuple, data,
4685 : sizeof(HeapTupleData));
4686 10012 : data += sizeof(HeapTupleData);
4687 :
4688 : /* reset t_data pointer into the new tuplebuf */
4689 10012 : change->data.tp.oldtuple->t_data =
4690 10012 : (HeapTupleHeader) ((char *) change->data.tp.oldtuple + HEAPTUPLESIZE);
4691 :
4692 : /* restore tuple data itself */
4693 10012 : memcpy(change->data.tp.oldtuple->t_data, data, tuplelen);
4694 10012 : data += tuplelen;
4695 : }
4696 :
4697 343460 : if (change->data.tp.newtuple)
4698 : {
4699 : /* here, data might not be suitably aligned! */
4700 : uint32 tuplelen;
4701 :
4702 323018 : memcpy(&tuplelen, data + offsetof(HeapTupleData, t_len),
4703 : sizeof(uint32));
4704 :
4705 323018 : change->data.tp.newtuple =
4706 323018 : ReorderBufferAllocTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
4707 :
4708 : /* restore ->tuple */
4709 323018 : memcpy(change->data.tp.newtuple, data,
4710 : sizeof(HeapTupleData));
4711 323018 : data += sizeof(HeapTupleData);
4712 :
4713 : /* reset t_data pointer into the new tuplebuf */
4714 323018 : change->data.tp.newtuple->t_data =
4715 323018 : (HeapTupleHeader) ((char *) change->data.tp.newtuple + HEAPTUPLESIZE);
4716 :
4717 : /* restore tuple data itself */
4718 323018 : memcpy(change->data.tp.newtuple->t_data, data, tuplelen);
4719 323018 : data += tuplelen;
4720 : }
4721 :
4722 343460 : break;
4723 2 : case REORDER_BUFFER_CHANGE_MESSAGE:
4724 : {
4725 : Size prefix_size;
4726 :
4727 : /* read prefix */
4728 2 : memcpy(&prefix_size, data, sizeof(Size));
4729 2 : data += sizeof(Size);
4730 2 : change->data.msg.prefix = MemoryContextAlloc(rb->context,
4731 : prefix_size);
4732 2 : memcpy(change->data.msg.prefix, data, prefix_size);
4733 : Assert(change->data.msg.prefix[prefix_size - 1] == '\0');
4734 2 : data += prefix_size;
4735 :
4736 : /* read the message */
4737 2 : memcpy(&change->data.msg.message_size, data, sizeof(Size));
4738 2 : data += sizeof(Size);
4739 2 : change->data.msg.message = MemoryContextAlloc(rb->context,
4740 : change->data.msg.message_size);
4741 2 : memcpy(change->data.msg.message, data,
4742 : change->data.msg.message_size);
4743 2 : data += change->data.msg.message_size;
4744 :
4745 2 : break;
4746 : }
4747 46 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4748 : {
4749 46 : Size inval_size = sizeof(SharedInvalidationMessage) *
4750 46 : change->data.inval.ninvalidations;
4751 :
4752 46 : change->data.inval.invalidations =
4753 46 : MemoryContextAlloc(rb->context, inval_size);
4754 :
4755 : /* read the message */
4756 46 : memcpy(change->data.inval.invalidations, data, inval_size);
4757 :
4758 46 : break;
4759 : }
4760 4 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4761 : {
4762 : Snapshot oldsnap;
4763 : Snapshot newsnap;
4764 : Size size;
4765 :
4766 4 : oldsnap = (Snapshot) data;
4767 :
4768 4 : size = sizeof(SnapshotData) +
4769 4 : sizeof(TransactionId) * oldsnap->xcnt +
4770 4 : sizeof(TransactionId) * (oldsnap->subxcnt + 0);
4771 :
4772 4 : change->data.snapshot = MemoryContextAllocZero(rb->context, size);
4773 :
4774 4 : newsnap = change->data.snapshot;
4775 :
4776 4 : memcpy(newsnap, data, size);
4777 4 : newsnap->xip = (TransactionId *)
4778 : (((char *) newsnap) + sizeof(SnapshotData));
4779 4 : newsnap->subxip = newsnap->xip + newsnap->xcnt;
4780 4 : newsnap->copied = true;
4781 4 : break;
4782 : }
4783 : /* the base struct contains all the data, easy peasy */
4784 0 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4785 : {
4786 : Oid *relids;
4787 :
4788 0 : relids = ReorderBufferAllocRelids(rb, change->data.truncate.nrelids);
4789 0 : memcpy(relids, data, change->data.truncate.nrelids * sizeof(Oid));
4790 0 : change->data.truncate.relids = relids;
4791 :
4792 0 : break;
4793 : }
4794 3806 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4795 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4796 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4797 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4798 3806 : break;
4799 : }
4800 :
4801 347318 : dlist_push_tail(&txn->changes, &change->node);
4802 347318 : txn->nentries_mem++;
4803 :
4804 : /*
4805 : * Update memory accounting for the restored change. We need to do this
4806 : * although we don't check the memory limit when restoring the changes in
4807 : * this branch (we only do that when initially queueing the changes after
4808 : * decoding), because we will release the changes later, and that will
4809 : * update the accounting too (subtracting the size from the counters). And
4810 : * we don't want to underflow there.
4811 : */
4812 347318 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
4813 : ReorderBufferChangeSize(change));
4814 347318 : }
4815 :
4816 : /*
4817 : * Remove all on-disk stored for the passed in transaction.
4818 : */
4819 : static void
4820 546 : ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn)
4821 : {
4822 : XLogSegNo first;
4823 : XLogSegNo cur;
4824 : XLogSegNo last;
4825 :
4826 : Assert(txn->first_lsn != InvalidXLogRecPtr);
4827 : Assert(txn->final_lsn != InvalidXLogRecPtr);
4828 :
4829 546 : XLByteToSeg(txn->first_lsn, first, wal_segment_size);
4830 546 : XLByteToSeg(txn->final_lsn, last, wal_segment_size);
4831 :
4832 : /* iterate over all possible filenames, and delete them */
4833 1110 : for (cur = first; cur <= last; cur++)
4834 : {
4835 : char path[MAXPGPATH];
4836 :
4837 564 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid, cur);
4838 564 : if (unlink(path) != 0 && errno != ENOENT)
4839 0 : ereport(ERROR,
4840 : (errcode_for_file_access(),
4841 : errmsg("could not remove file \"%s\": %m", path)));
4842 : }
4843 546 : }
4844 :
4845 : /*
4846 : * Remove any leftover serialized reorder buffers from a slot directory after a
4847 : * prior crash or decoding session exit.
4848 : */
4849 : static void
4850 4070 : ReorderBufferCleanupSerializedTXNs(const char *slotname)
4851 : {
4852 : DIR *spill_dir;
4853 : struct dirent *spill_de;
4854 : struct stat statbuf;
4855 : char path[MAXPGPATH * 2 + sizeof(PG_REPLSLOT_DIR)];
4856 :
4857 4070 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, slotname);
4858 :
4859 : /* we're only handling directories here, skip if it's not ours */
4860 4070 : if (lstat(path, &statbuf) == 0 && !S_ISDIR(statbuf.st_mode))
4861 0 : return;
4862 :
4863 4070 : spill_dir = AllocateDir(path);
4864 20350 : while ((spill_de = ReadDirExtended(spill_dir, path, INFO)) != NULL)
4865 : {
4866 : /* only look at names that can be ours */
4867 12210 : if (strncmp(spill_de->d_name, "xid", 3) == 0)
4868 : {
4869 0 : snprintf(path, sizeof(path),
4870 : "%s/%s/%s", PG_REPLSLOT_DIR, slotname,
4871 0 : spill_de->d_name);
4872 :
4873 0 : if (unlink(path) != 0)
4874 0 : ereport(ERROR,
4875 : (errcode_for_file_access(),
4876 : errmsg("could not remove file \"%s\" during removal of %s/%s/xid*: %m",
4877 : path, PG_REPLSLOT_DIR, slotname)));
4878 : }
4879 : }
4880 4070 : FreeDir(spill_dir);
4881 : }
4882 :
4883 : /*
4884 : * Given a replication slot, transaction ID and segment number, fill in the
4885 : * corresponding spill file into 'path', which is a caller-owned buffer of size
4886 : * at least MAXPGPATH.
4887 : */
4888 : static void
4889 8106 : ReorderBufferSerializedPath(char *path, ReplicationSlot *slot, TransactionId xid,
4890 : XLogSegNo segno)
4891 : {
4892 : XLogRecPtr recptr;
4893 :
4894 8106 : XLogSegNoOffsetToRecPtr(segno, 0, wal_segment_size, recptr);
4895 :
4896 8106 : snprintf(path, MAXPGPATH, "%s/%s/xid-%u-lsn-%X-%X.spill",
4897 : PG_REPLSLOT_DIR,
4898 8106 : NameStr(MyReplicationSlot->data.name),
4899 8106 : xid, LSN_FORMAT_ARGS(recptr));
4900 8106 : }
4901 :
4902 : /*
4903 : * Delete all data spilled to disk after we've restarted/crashed. It will be
4904 : * recreated when the respective slots are reused.
4905 : */
4906 : void
4907 1864 : StartupReorderBuffer(void)
4908 : {
4909 : DIR *logical_dir;
4910 : struct dirent *logical_de;
4911 :
4912 1864 : logical_dir = AllocateDir(PG_REPLSLOT_DIR);
4913 5756 : while ((logical_de = ReadDir(logical_dir, PG_REPLSLOT_DIR)) != NULL)
4914 : {
4915 3892 : if (strcmp(logical_de->d_name, ".") == 0 ||
4916 2028 : strcmp(logical_de->d_name, "..") == 0)
4917 3728 : continue;
4918 :
4919 : /* if it cannot be a slot, skip the directory */
4920 164 : if (!ReplicationSlotValidateName(logical_de->d_name, true, DEBUG2))
4921 0 : continue;
4922 :
4923 : /*
4924 : * ok, has to be a surviving logical slot, iterate and delete
4925 : * everything starting with xid-*
4926 : */
4927 164 : ReorderBufferCleanupSerializedTXNs(logical_de->d_name);
4928 : }
4929 1864 : FreeDir(logical_dir);
4930 1864 : }
4931 :
4932 : /* ---------------------------------------
4933 : * toast reassembly support
4934 : * ---------------------------------------
4935 : */
4936 :
4937 : /*
4938 : * Initialize per tuple toast reconstruction support.
4939 : */
4940 : static void
4941 70 : ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
4942 : {
4943 : HASHCTL hash_ctl;
4944 :
4945 : Assert(txn->toast_hash == NULL);
4946 :
4947 70 : hash_ctl.keysize = sizeof(Oid);
4948 70 : hash_ctl.entrysize = sizeof(ReorderBufferToastEnt);
4949 70 : hash_ctl.hcxt = rb->context;
4950 70 : txn->toast_hash = hash_create("ReorderBufferToastHash", 5, &hash_ctl,
4951 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
4952 70 : }
4953 :
4954 : /*
4955 : * Per toast-chunk handling for toast reconstruction
4956 : *
4957 : * Appends a toast chunk so we can reconstruct it when the tuple "owning" the
4958 : * toasted Datum comes along.
4959 : */
4960 : static void
4961 3660 : ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
4962 : Relation relation, ReorderBufferChange *change)
4963 : {
4964 : ReorderBufferToastEnt *ent;
4965 : HeapTuple newtup;
4966 : bool found;
4967 : int32 chunksize;
4968 : bool isnull;
4969 : Pointer chunk;
4970 3660 : TupleDesc desc = RelationGetDescr(relation);
4971 : Oid chunk_id;
4972 : int32 chunk_seq;
4973 :
4974 3660 : if (txn->toast_hash == NULL)
4975 70 : ReorderBufferToastInitHash(rb, txn);
4976 :
4977 : Assert(IsToastRelation(relation));
4978 :
4979 3660 : newtup = change->data.tp.newtuple;
4980 3660 : chunk_id = DatumGetObjectId(fastgetattr(newtup, 1, desc, &isnull));
4981 : Assert(!isnull);
4982 3660 : chunk_seq = DatumGetInt32(fastgetattr(newtup, 2, desc, &isnull));
4983 : Assert(!isnull);
4984 :
4985 : ent = (ReorderBufferToastEnt *)
4986 3660 : hash_search(txn->toast_hash, &chunk_id, HASH_ENTER, &found);
4987 :
4988 3660 : if (!found)
4989 : {
4990 : Assert(ent->chunk_id == chunk_id);
4991 98 : ent->num_chunks = 0;
4992 98 : ent->last_chunk_seq = 0;
4993 98 : ent->size = 0;
4994 98 : ent->reconstructed = NULL;
4995 98 : dlist_init(&ent->chunks);
4996 :
4997 98 : if (chunk_seq != 0)
4998 0 : elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq 0",
4999 : chunk_seq, chunk_id);
5000 : }
5001 3562 : else if (found && chunk_seq != ent->last_chunk_seq + 1)
5002 0 : elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq %d",
5003 : chunk_seq, chunk_id, ent->last_chunk_seq + 1);
5004 :
5005 3660 : chunk = DatumGetPointer(fastgetattr(newtup, 3, desc, &isnull));
5006 : Assert(!isnull);
5007 :
5008 : /* calculate size so we can allocate the right size at once later */
5009 3660 : if (!VARATT_IS_EXTENDED(chunk))
5010 3660 : chunksize = VARSIZE(chunk) - VARHDRSZ;
5011 0 : else if (VARATT_IS_SHORT(chunk))
5012 : /* could happen due to heap_form_tuple doing its thing */
5013 0 : chunksize = VARSIZE_SHORT(chunk) - VARHDRSZ_SHORT;
5014 : else
5015 0 : elog(ERROR, "unexpected type of toast chunk");
5016 :
5017 3660 : ent->size += chunksize;
5018 3660 : ent->last_chunk_seq = chunk_seq;
5019 3660 : ent->num_chunks++;
5020 3660 : dlist_push_tail(&ent->chunks, &change->node);
5021 3660 : }
5022 :
5023 : /*
5024 : * Rejigger change->newtuple to point to in-memory toast tuples instead of
5025 : * on-disk toast tuples that may no longer exist (think DROP TABLE or VACUUM).
5026 : *
5027 : * We cannot replace unchanged toast tuples though, so those will still point
5028 : * to on-disk toast data.
5029 : *
5030 : * While updating the existing change with detoasted tuple data, we need to
5031 : * update the memory accounting info, because the change size will differ.
5032 : * Otherwise the accounting may get out of sync, triggering serialization
5033 : * at unexpected times.
5034 : *
5035 : * We simply subtract size of the change before rejiggering the tuple, and
5036 : * then add the new size. This makes it look like the change was removed
5037 : * and then added back, except it only tweaks the accounting info.
5038 : *
5039 : * In particular it can't trigger serialization, which would be pointless
5040 : * anyway as it happens during commit processing right before handing
5041 : * the change to the output plugin.
5042 : */
5043 : static void
5044 679992 : ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
5045 : Relation relation, ReorderBufferChange *change)
5046 : {
5047 : TupleDesc desc;
5048 : int natt;
5049 : Datum *attrs;
5050 : bool *isnull;
5051 : bool *free;
5052 : HeapTuple tmphtup;
5053 : Relation toast_rel;
5054 : TupleDesc toast_desc;
5055 : MemoryContext oldcontext;
5056 : HeapTuple newtup;
5057 : Size old_size;
5058 :
5059 : /* no toast tuples changed */
5060 679992 : if (txn->toast_hash == NULL)
5061 679500 : return;
5062 :
5063 : /*
5064 : * We're going to modify the size of the change. So, to make sure the
5065 : * accounting is correct we record the current change size and then after
5066 : * re-computing the change we'll subtract the recorded size and then
5067 : * re-add the new change size at the end. We don't immediately subtract
5068 : * the old size because if there is any error before we add the new size,
5069 : * we will release the changes and that will update the accounting info
5070 : * (subtracting the size from the counters). And we don't want to
5071 : * underflow there.
5072 : */
5073 492 : old_size = ReorderBufferChangeSize(change);
5074 :
5075 492 : oldcontext = MemoryContextSwitchTo(rb->context);
5076 :
5077 : /* we should only have toast tuples in an INSERT or UPDATE */
5078 : Assert(change->data.tp.newtuple);
5079 :
5080 492 : desc = RelationGetDescr(relation);
5081 :
5082 492 : toast_rel = RelationIdGetRelation(relation->rd_rel->reltoastrelid);
5083 492 : if (!RelationIsValid(toast_rel))
5084 0 : elog(ERROR, "could not open toast relation with OID %u (base relation \"%s\")",
5085 : relation->rd_rel->reltoastrelid, RelationGetRelationName(relation));
5086 :
5087 492 : toast_desc = RelationGetDescr(toast_rel);
5088 :
5089 : /* should we allocate from stack instead? */
5090 492 : attrs = palloc0(sizeof(Datum) * desc->natts);
5091 492 : isnull = palloc0(sizeof(bool) * desc->natts);
5092 492 : free = palloc0(sizeof(bool) * desc->natts);
5093 :
5094 492 : newtup = change->data.tp.newtuple;
5095 :
5096 492 : heap_deform_tuple(newtup, desc, attrs, isnull);
5097 :
5098 1514 : for (natt = 0; natt < desc->natts; natt++)
5099 : {
5100 1022 : Form_pg_attribute attr = TupleDescAttr(desc, natt);
5101 : ReorderBufferToastEnt *ent;
5102 : struct varlena *varlena;
5103 :
5104 : /* va_rawsize is the size of the original datum -- including header */
5105 : struct varatt_external toast_pointer;
5106 : struct varatt_indirect redirect_pointer;
5107 1022 : struct varlena *new_datum = NULL;
5108 : struct varlena *reconstructed;
5109 : dlist_iter it;
5110 1022 : Size data_done = 0;
5111 :
5112 : /* system columns aren't toasted */
5113 1022 : if (attr->attnum < 0)
5114 926 : continue;
5115 :
5116 1022 : if (attr->attisdropped)
5117 0 : continue;
5118 :
5119 : /* not a varlena datatype */
5120 1022 : if (attr->attlen != -1)
5121 482 : continue;
5122 :
5123 : /* no data */
5124 540 : if (isnull[natt])
5125 24 : continue;
5126 :
5127 : /* ok, we know we have a toast datum */
5128 516 : varlena = (struct varlena *) DatumGetPointer(attrs[natt]);
5129 :
5130 : /* no need to do anything if the tuple isn't external */
5131 516 : if (!VARATT_IS_EXTERNAL(varlena))
5132 404 : continue;
5133 :
5134 112 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, varlena);
5135 :
5136 : /*
5137 : * Check whether the toast tuple changed, replace if so.
5138 : */
5139 : ent = (ReorderBufferToastEnt *)
5140 112 : hash_search(txn->toast_hash,
5141 : &toast_pointer.va_valueid,
5142 : HASH_FIND,
5143 : NULL);
5144 112 : if (ent == NULL)
5145 16 : continue;
5146 :
5147 : new_datum =
5148 96 : (struct varlena *) palloc0(INDIRECT_POINTER_SIZE);
5149 :
5150 96 : free[natt] = true;
5151 :
5152 96 : reconstructed = palloc0(toast_pointer.va_rawsize);
5153 :
5154 96 : ent->reconstructed = reconstructed;
5155 :
5156 : /* stitch toast tuple back together from its parts */
5157 3654 : dlist_foreach(it, &ent->chunks)
5158 : {
5159 : bool cisnull;
5160 : ReorderBufferChange *cchange;
5161 : HeapTuple ctup;
5162 : Pointer chunk;
5163 :
5164 3558 : cchange = dlist_container(ReorderBufferChange, node, it.cur);
5165 3558 : ctup = cchange->data.tp.newtuple;
5166 3558 : chunk = DatumGetPointer(fastgetattr(ctup, 3, toast_desc, &cisnull));
5167 :
5168 : Assert(!cisnull);
5169 : Assert(!VARATT_IS_EXTERNAL(chunk));
5170 : Assert(!VARATT_IS_SHORT(chunk));
5171 :
5172 3558 : memcpy(VARDATA(reconstructed) + data_done,
5173 3558 : VARDATA(chunk),
5174 3558 : VARSIZE(chunk) - VARHDRSZ);
5175 3558 : data_done += VARSIZE(chunk) - VARHDRSZ;
5176 : }
5177 : Assert(data_done == VARATT_EXTERNAL_GET_EXTSIZE(toast_pointer));
5178 :
5179 : /* make sure its marked as compressed or not */
5180 96 : if (VARATT_EXTERNAL_IS_COMPRESSED(toast_pointer))
5181 10 : SET_VARSIZE_COMPRESSED(reconstructed, data_done + VARHDRSZ);
5182 : else
5183 86 : SET_VARSIZE(reconstructed, data_done + VARHDRSZ);
5184 :
5185 96 : memset(&redirect_pointer, 0, sizeof(redirect_pointer));
5186 96 : redirect_pointer.pointer = reconstructed;
5187 :
5188 96 : SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
5189 96 : memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
5190 : sizeof(redirect_pointer));
5191 :
5192 96 : attrs[natt] = PointerGetDatum(new_datum);
5193 : }
5194 :
5195 : /*
5196 : * Build tuple in separate memory & copy tuple back into the tuplebuf
5197 : * passed to the output plugin. We can't directly heap_fill_tuple() into
5198 : * the tuplebuf because attrs[] will point back into the current content.
5199 : */
5200 492 : tmphtup = heap_form_tuple(desc, attrs, isnull);
5201 : Assert(newtup->t_len <= MaxHeapTupleSize);
5202 : Assert(newtup->t_data == (HeapTupleHeader) ((char *) newtup + HEAPTUPLESIZE));
5203 :
5204 492 : memcpy(newtup->t_data, tmphtup->t_data, tmphtup->t_len);
5205 492 : newtup->t_len = tmphtup->t_len;
5206 :
5207 : /*
5208 : * free resources we won't further need, more persistent stuff will be
5209 : * free'd in ReorderBufferToastReset().
5210 : */
5211 492 : RelationClose(toast_rel);
5212 492 : pfree(tmphtup);
5213 1514 : for (natt = 0; natt < desc->natts; natt++)
5214 : {
5215 1022 : if (free[natt])
5216 96 : pfree(DatumGetPointer(attrs[natt]));
5217 : }
5218 492 : pfree(attrs);
5219 492 : pfree(free);
5220 492 : pfree(isnull);
5221 :
5222 492 : MemoryContextSwitchTo(oldcontext);
5223 :
5224 : /* subtract the old change size */
5225 492 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, false, old_size);
5226 : /* now add the change back, with the correct size */
5227 492 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
5228 : ReorderBufferChangeSize(change));
5229 : }
5230 :
5231 : /*
5232 : * Free all resources allocated for toast reconstruction.
5233 : */
5234 : static void
5235 687120 : ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn)
5236 : {
5237 : HASH_SEQ_STATUS hstat;
5238 : ReorderBufferToastEnt *ent;
5239 :
5240 687120 : if (txn->toast_hash == NULL)
5241 687050 : return;
5242 :
5243 : /* sequentially walk over the hash and free everything */
5244 70 : hash_seq_init(&hstat, txn->toast_hash);
5245 168 : while ((ent = (ReorderBufferToastEnt *) hash_seq_search(&hstat)) != NULL)
5246 : {
5247 : dlist_mutable_iter it;
5248 :
5249 98 : if (ent->reconstructed != NULL)
5250 96 : pfree(ent->reconstructed);
5251 :
5252 3758 : dlist_foreach_modify(it, &ent->chunks)
5253 : {
5254 3660 : ReorderBufferChange *change =
5255 3660 : dlist_container(ReorderBufferChange, node, it.cur);
5256 :
5257 3660 : dlist_delete(&change->node);
5258 3660 : ReorderBufferFreeChange(rb, change, true);
5259 : }
5260 : }
5261 :
5262 70 : hash_destroy(txn->toast_hash);
5263 70 : txn->toast_hash = NULL;
5264 : }
5265 :
5266 :
5267 : /* ---------------------------------------
5268 : * Visibility support for logical decoding
5269 : *
5270 : *
5271 : * Lookup actual cmin/cmax values when using decoding snapshot. We can't
5272 : * always rely on stored cmin/cmax values because of two scenarios:
5273 : *
5274 : * * A tuple got changed multiple times during a single transaction and thus
5275 : * has got a combo CID. Combo CIDs are only valid for the duration of a
5276 : * single transaction.
5277 : * * A tuple with a cmin but no cmax (and thus no combo CID) got
5278 : * deleted/updated in another transaction than the one which created it
5279 : * which we are looking at right now. As only one of cmin, cmax or combo CID
5280 : * is actually stored in the heap we don't have access to the value we
5281 : * need anymore.
5282 : *
5283 : * To resolve those problems we have a per-transaction hash of (cmin,
5284 : * cmax) tuples keyed by (relfilelocator, ctid) which contains the actual
5285 : * (cmin, cmax) values. That also takes care of combo CIDs by simply
5286 : * not caring about them at all. As we have the real cmin/cmax values
5287 : * combo CIDs aren't interesting.
5288 : *
5289 : * As we only care about catalog tuples here the overhead of this
5290 : * hashtable should be acceptable.
5291 : *
5292 : * Heap rewrites complicate this a bit, check rewriteheap.c for
5293 : * details.
5294 : * -------------------------------------------------------------------------
5295 : */
5296 :
5297 : /* struct for sorting mapping files by LSN efficiently */
5298 : typedef struct RewriteMappingFile
5299 : {
5300 : XLogRecPtr lsn;
5301 : char fname[MAXPGPATH];
5302 : } RewriteMappingFile;
5303 :
5304 : #ifdef NOT_USED
5305 : static void
5306 : DisplayMapping(HTAB *tuplecid_data)
5307 : {
5308 : HASH_SEQ_STATUS hstat;
5309 : ReorderBufferTupleCidEnt *ent;
5310 :
5311 : hash_seq_init(&hstat, tuplecid_data);
5312 : while ((ent = (ReorderBufferTupleCidEnt *) hash_seq_search(&hstat)) != NULL)
5313 : {
5314 : elog(DEBUG3, "mapping: node: %u/%u/%u tid: %u/%u cmin: %u, cmax: %u",
5315 : ent->key.rlocator.dbOid,
5316 : ent->key.rlocator.spcOid,
5317 : ent->key.rlocator.relNumber,
5318 : ItemPointerGetBlockNumber(&ent->key.tid),
5319 : ItemPointerGetOffsetNumber(&ent->key.tid),
5320 : ent->cmin,
5321 : ent->cmax
5322 : );
5323 : }
5324 : }
5325 : #endif
5326 :
5327 : /*
5328 : * Apply a single mapping file to tuplecid_data.
5329 : *
5330 : * The mapping file has to have been verified to be a) committed b) for our
5331 : * transaction c) applied in LSN order.
5332 : */
5333 : static void
5334 54 : ApplyLogicalMappingFile(HTAB *tuplecid_data, Oid relid, const char *fname)
5335 : {
5336 : char path[MAXPGPATH];
5337 : int fd;
5338 : int readBytes;
5339 : LogicalRewriteMappingData map;
5340 :
5341 54 : sprintf(path, "%s/%s", PG_LOGICAL_MAPPINGS_DIR, fname);
5342 54 : fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
5343 54 : if (fd < 0)
5344 0 : ereport(ERROR,
5345 : (errcode_for_file_access(),
5346 : errmsg("could not open file \"%s\": %m", path)));
5347 :
5348 : while (true)
5349 418 : {
5350 : ReorderBufferTupleCidKey key;
5351 : ReorderBufferTupleCidEnt *ent;
5352 : ReorderBufferTupleCidEnt *new_ent;
5353 : bool found;
5354 :
5355 : /* be careful about padding */
5356 472 : memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
5357 :
5358 : /* read all mappings till the end of the file */
5359 472 : pgstat_report_wait_start(WAIT_EVENT_REORDER_LOGICAL_MAPPING_READ);
5360 472 : readBytes = read(fd, &map, sizeof(LogicalRewriteMappingData));
5361 472 : pgstat_report_wait_end();
5362 :
5363 472 : if (readBytes < 0)
5364 0 : ereport(ERROR,
5365 : (errcode_for_file_access(),
5366 : errmsg("could not read file \"%s\": %m",
5367 : path)));
5368 472 : else if (readBytes == 0) /* EOF */
5369 54 : break;
5370 418 : else if (readBytes != sizeof(LogicalRewriteMappingData))
5371 0 : ereport(ERROR,
5372 : (errcode_for_file_access(),
5373 : errmsg("could not read from file \"%s\": read %d instead of %d bytes",
5374 : path, readBytes,
5375 : (int32) sizeof(LogicalRewriteMappingData))));
5376 :
5377 418 : key.rlocator = map.old_locator;
5378 418 : ItemPointerCopy(&map.old_tid,
5379 : &key.tid);
5380 :
5381 :
5382 : ent = (ReorderBufferTupleCidEnt *)
5383 418 : hash_search(tuplecid_data, &key, HASH_FIND, NULL);
5384 :
5385 : /* no existing mapping, no need to update */
5386 418 : if (!ent)
5387 0 : continue;
5388 :
5389 418 : key.rlocator = map.new_locator;
5390 418 : ItemPointerCopy(&map.new_tid,
5391 : &key.tid);
5392 :
5393 : new_ent = (ReorderBufferTupleCidEnt *)
5394 418 : hash_search(tuplecid_data, &key, HASH_ENTER, &found);
5395 :
5396 418 : if (found)
5397 : {
5398 : /*
5399 : * Make sure the existing mapping makes sense. We sometime update
5400 : * old records that did not yet have a cmax (e.g. pg_class' own
5401 : * entry while rewriting it) during rewrites, so allow that.
5402 : */
5403 : Assert(ent->cmin == InvalidCommandId || ent->cmin == new_ent->cmin);
5404 : Assert(ent->cmax == InvalidCommandId || ent->cmax == new_ent->cmax);
5405 : }
5406 : else
5407 : {
5408 : /* update mapping */
5409 406 : new_ent->cmin = ent->cmin;
5410 406 : new_ent->cmax = ent->cmax;
5411 406 : new_ent->combocid = ent->combocid;
5412 : }
5413 : }
5414 :
5415 54 : if (CloseTransientFile(fd) != 0)
5416 0 : ereport(ERROR,
5417 : (errcode_for_file_access(),
5418 : errmsg("could not close file \"%s\": %m", path)));
5419 54 : }
5420 :
5421 :
5422 : /*
5423 : * Check whether the TransactionId 'xid' is in the pre-sorted array 'xip'.
5424 : */
5425 : static bool
5426 696 : TransactionIdInArray(TransactionId xid, TransactionId *xip, Size num)
5427 : {
5428 696 : return bsearch(&xid, xip, num,
5429 696 : sizeof(TransactionId), xidComparator) != NULL;
5430 : }
5431 :
5432 : /*
5433 : * list_sort() comparator for sorting RewriteMappingFiles in LSN order.
5434 : */
5435 : static int
5436 68 : file_sort_by_lsn(const ListCell *a_p, const ListCell *b_p)
5437 : {
5438 68 : RewriteMappingFile *a = (RewriteMappingFile *) lfirst(a_p);
5439 68 : RewriteMappingFile *b = (RewriteMappingFile *) lfirst(b_p);
5440 :
5441 68 : return pg_cmp_u64(a->lsn, b->lsn);
5442 : }
5443 :
5444 : /*
5445 : * Apply any existing logical remapping files if there are any targeted at our
5446 : * transaction for relid.
5447 : */
5448 : static void
5449 22 : UpdateLogicalMappings(HTAB *tuplecid_data, Oid relid, Snapshot snapshot)
5450 : {
5451 : DIR *mapping_dir;
5452 : struct dirent *mapping_de;
5453 22 : List *files = NIL;
5454 : ListCell *file;
5455 22 : Oid dboid = IsSharedRelation(relid) ? InvalidOid : MyDatabaseId;
5456 :
5457 22 : mapping_dir = AllocateDir(PG_LOGICAL_MAPPINGS_DIR);
5458 1146 : while ((mapping_de = ReadDir(mapping_dir, PG_LOGICAL_MAPPINGS_DIR)) != NULL)
5459 : {
5460 : Oid f_dboid;
5461 : Oid f_relid;
5462 : TransactionId f_mapped_xid;
5463 : TransactionId f_create_xid;
5464 : XLogRecPtr f_lsn;
5465 : uint32 f_hi,
5466 : f_lo;
5467 : RewriteMappingFile *f;
5468 :
5469 1124 : if (strcmp(mapping_de->d_name, ".") == 0 ||
5470 1102 : strcmp(mapping_de->d_name, "..") == 0)
5471 1070 : continue;
5472 :
5473 : /* Ignore files that aren't ours */
5474 1080 : if (strncmp(mapping_de->d_name, "map-", 4) != 0)
5475 0 : continue;
5476 :
5477 1080 : if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
5478 : &f_dboid, &f_relid, &f_hi, &f_lo,
5479 : &f_mapped_xid, &f_create_xid) != 6)
5480 0 : elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
5481 :
5482 1080 : f_lsn = ((uint64) f_hi) << 32 | f_lo;
5483 :
5484 : /* mapping for another database */
5485 1080 : if (f_dboid != dboid)
5486 0 : continue;
5487 :
5488 : /* mapping for another relation */
5489 1080 : if (f_relid != relid)
5490 120 : continue;
5491 :
5492 : /* did the creating transaction abort? */
5493 960 : if (!TransactionIdDidCommit(f_create_xid))
5494 264 : continue;
5495 :
5496 : /* not for our transaction */
5497 696 : if (!TransactionIdInArray(f_mapped_xid, snapshot->subxip, snapshot->subxcnt))
5498 642 : continue;
5499 :
5500 : /* ok, relevant, queue for apply */
5501 54 : f = palloc(sizeof(RewriteMappingFile));
5502 54 : f->lsn = f_lsn;
5503 54 : strcpy(f->fname, mapping_de->d_name);
5504 54 : files = lappend(files, f);
5505 : }
5506 22 : FreeDir(mapping_dir);
5507 :
5508 : /* sort files so we apply them in LSN order */
5509 22 : list_sort(files, file_sort_by_lsn);
5510 :
5511 76 : foreach(file, files)
5512 : {
5513 54 : RewriteMappingFile *f = (RewriteMappingFile *) lfirst(file);
5514 :
5515 54 : elog(DEBUG1, "applying mapping: \"%s\" in %u", f->fname,
5516 : snapshot->subxip[0]);
5517 54 : ApplyLogicalMappingFile(tuplecid_data, relid, f->fname);
5518 54 : pfree(f);
5519 : }
5520 22 : }
5521 :
5522 : /*
5523 : * Lookup cmin/cmax of a tuple, during logical decoding where we can't rely on
5524 : * combo CIDs.
5525 : */
5526 : bool
5527 1538 : ResolveCminCmaxDuringDecoding(HTAB *tuplecid_data,
5528 : Snapshot snapshot,
5529 : HeapTuple htup, Buffer buffer,
5530 : CommandId *cmin, CommandId *cmax)
5531 : {
5532 : ReorderBufferTupleCidKey key;
5533 : ReorderBufferTupleCidEnt *ent;
5534 : ForkNumber forkno;
5535 : BlockNumber blockno;
5536 1538 : bool updated_mapping = false;
5537 :
5538 : /*
5539 : * Return unresolved if tuplecid_data is not valid. That's because when
5540 : * streaming in-progress transactions we may run into tuples with the CID
5541 : * before actually decoding them. Think e.g. about INSERT followed by
5542 : * TRUNCATE, where the TRUNCATE may not be decoded yet when applying the
5543 : * INSERT. So in such cases, we assume the CID is from the future
5544 : * command.
5545 : */
5546 1538 : if (tuplecid_data == NULL)
5547 28 : return false;
5548 :
5549 : /* be careful about padding */
5550 1510 : memset(&key, 0, sizeof(key));
5551 :
5552 : Assert(!BufferIsLocal(buffer));
5553 :
5554 : /*
5555 : * get relfilelocator from the buffer, no convenient way to access it
5556 : * other than that.
5557 : */
5558 1510 : BufferGetTag(buffer, &key.rlocator, &forkno, &blockno);
5559 :
5560 : /* tuples can only be in the main fork */
5561 : Assert(forkno == MAIN_FORKNUM);
5562 : Assert(blockno == ItemPointerGetBlockNumber(&htup->t_self));
5563 :
5564 1510 : ItemPointerCopy(&htup->t_self,
5565 : &key.tid);
5566 :
5567 1532 : restart:
5568 : ent = (ReorderBufferTupleCidEnt *)
5569 1532 : hash_search(tuplecid_data, &key, HASH_FIND, NULL);
5570 :
5571 : /*
5572 : * failed to find a mapping, check whether the table was rewritten and
5573 : * apply mapping if so, but only do that once - there can be no new
5574 : * mappings while we are in here since we have to hold a lock on the
5575 : * relation.
5576 : */
5577 1532 : if (ent == NULL && !updated_mapping)
5578 : {
5579 22 : UpdateLogicalMappings(tuplecid_data, htup->t_tableOid, snapshot);
5580 : /* now check but don't update for a mapping again */
5581 22 : updated_mapping = true;
5582 22 : goto restart;
5583 : }
5584 1510 : else if (ent == NULL)
5585 10 : return false;
5586 :
5587 1500 : if (cmin)
5588 1500 : *cmin = ent->cmin;
5589 1500 : if (cmax)
5590 1500 : *cmax = ent->cmax;
5591 1500 : return true;
5592 : }
5593 :
5594 : /*
5595 : * Count invalidation messages of specified transaction.
5596 : *
5597 : * Returns number of messages, and msgs is set to the pointer of the linked
5598 : * list for the messages.
5599 : */
5600 : uint32
5601 64 : ReorderBufferGetInvalidations(ReorderBuffer *rb, TransactionId xid,
5602 : SharedInvalidationMessage **msgs)
5603 : {
5604 : ReorderBufferTXN *txn;
5605 :
5606 64 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
5607 : false);
5608 :
5609 64 : if (txn == NULL)
5610 0 : return 0;
5611 :
5612 64 : *msgs = txn->invalidations;
5613 :
5614 64 : return txn->ninvalidations;
5615 : }
|
