Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * reorderbuffer.c
4 : * PostgreSQL logical replay/reorder buffer management
5 : *
6 : *
7 : * Copyright (c) 2012-2026, 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 : #include "utils/wait_event.h"
117 :
118 : /*
119 : * Each transaction has an 8MB limit for invalidation messages distributed from
120 : * other transactions. This limit is set considering scenarios with many
121 : * concurrent logical decoding operations. When the distributed invalidation
122 : * messages reach this threshold, the transaction is marked as
123 : * RBTXN_DISTR_INVAL_OVERFLOWED to invalidate the complete cache as we have lost
124 : * some inval messages and hence don't know what needs to be invalidated.
125 : */
126 : #define MAX_DISTR_INVAL_MSG_PER_TXN \
127 : ((8 * 1024 * 1024) / sizeof(SharedInvalidationMessage))
128 :
129 : /* entry for a hash table we use to map from xid to our transaction state */
130 : typedef struct ReorderBufferTXNByIdEnt
131 : {
132 : TransactionId xid;
133 : ReorderBufferTXN *txn;
134 : } ReorderBufferTXNByIdEnt;
135 :
136 : /* data structures for (relfilelocator, ctid) => (cmin, cmax) mapping */
137 : typedef struct ReorderBufferTupleCidKey
138 : {
139 : RelFileLocator rlocator;
140 : ItemPointerData tid;
141 : } ReorderBufferTupleCidKey;
142 :
143 : typedef struct ReorderBufferTupleCidEnt
144 : {
145 : ReorderBufferTupleCidKey key;
146 : CommandId cmin;
147 : CommandId cmax;
148 : CommandId combocid; /* just for debugging */
149 : } ReorderBufferTupleCidEnt;
150 :
151 : /* Virtual file descriptor with file offset tracking */
152 : typedef struct TXNEntryFile
153 : {
154 : File vfd; /* -1 when the file is closed */
155 : off_t curOffset; /* offset for next write or read. Reset to 0
156 : * when vfd is opened. */
157 : } TXNEntryFile;
158 :
159 : /* k-way in-order change iteration support structures */
160 : typedef struct ReorderBufferIterTXNEntry
161 : {
162 : XLogRecPtr lsn;
163 : ReorderBufferChange *change;
164 : ReorderBufferTXN *txn;
165 : TXNEntryFile file;
166 : XLogSegNo segno;
167 : } ReorderBufferIterTXNEntry;
168 :
169 : typedef struct ReorderBufferIterTXNState
170 : {
171 : binaryheap *heap;
172 : Size nr_txns;
173 : dlist_head old_change;
174 : ReorderBufferIterTXNEntry entries[FLEXIBLE_ARRAY_MEMBER];
175 : } ReorderBufferIterTXNState;
176 :
177 : /* toast datastructures */
178 : typedef struct ReorderBufferToastEnt
179 : {
180 : Oid chunk_id; /* toast_table.chunk_id */
181 : int32 last_chunk_seq; /* toast_table.chunk_seq of the last chunk we
182 : * have seen */
183 : Size num_chunks; /* number of chunks we've already seen */
184 : Size size; /* combined size of chunks seen */
185 : dlist_head chunks; /* linked list of chunks */
186 : varlena *reconstructed; /* reconstructed varlena now pointed to in
187 : * main tup */
188 : } ReorderBufferToastEnt;
189 :
190 : /* Disk serialization support datastructures */
191 : typedef struct ReorderBufferDiskChange
192 : {
193 : Size size;
194 : ReorderBufferChange change;
195 : /* data follows */
196 : } ReorderBufferDiskChange;
197 :
198 : #define IsSpecInsert(action) \
199 : ( \
200 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT) \
201 : )
202 : #define IsSpecConfirmOrAbort(action) \
203 : ( \
204 : (((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM) || \
205 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT)) \
206 : )
207 : #define IsInsertOrUpdate(action) \
208 : ( \
209 : (((action) == REORDER_BUFFER_CHANGE_INSERT) || \
210 : ((action) == REORDER_BUFFER_CHANGE_UPDATE) || \
211 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT)) \
212 : )
213 :
214 : /*
215 : * Maximum number of changes kept in memory, per transaction. After that,
216 : * changes are spooled to disk.
217 : *
218 : * The current value should be sufficient to decode the entire transaction
219 : * without hitting disk in OLTP workloads, while starting to spool to disk in
220 : * other workloads reasonably fast.
221 : *
222 : * At some point in the future it probably makes sense to have a more elaborate
223 : * resource management here, but it's not entirely clear what that would look
224 : * like.
225 : */
226 : int logical_decoding_work_mem;
227 : static const Size max_changes_in_memory = 4096; /* XXX for restore only */
228 :
229 : /* GUC variable */
230 : int debug_logical_replication_streaming = DEBUG_LOGICAL_REP_STREAMING_BUFFERED;
231 :
232 : /* ---------------------------------------
233 : * primary reorderbuffer support routines
234 : * ---------------------------------------
235 : */
236 : static ReorderBufferTXN *ReorderBufferAllocTXN(ReorderBuffer *rb);
237 : static void ReorderBufferFreeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
238 : static ReorderBufferTXN *ReorderBufferTXNByXid(ReorderBuffer *rb,
239 : TransactionId xid, bool create, bool *is_new,
240 : XLogRecPtr lsn, bool create_as_top);
241 : static void ReorderBufferTransferSnapToParent(ReorderBufferTXN *txn,
242 : ReorderBufferTXN *subtxn);
243 :
244 : static void AssertTXNLsnOrder(ReorderBuffer *rb);
245 :
246 : /* ---------------------------------------
247 : * support functions for lsn-order iterating over the ->changes of a
248 : * transaction and its subtransactions
249 : *
250 : * used for iteration over the k-way heap merge of a transaction and its
251 : * subtransactions
252 : * ---------------------------------------
253 : */
254 : static void ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn,
255 : ReorderBufferIterTXNState *volatile *iter_state);
256 : static ReorderBufferChange *ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state);
257 : static void ReorderBufferIterTXNFinish(ReorderBuffer *rb,
258 : ReorderBufferIterTXNState *state);
259 : static void ReorderBufferExecuteInvalidations(uint32 nmsgs, SharedInvalidationMessage *msgs);
260 :
261 : /*
262 : * ---------------------------------------
263 : * Disk serialization support functions
264 : * ---------------------------------------
265 : */
266 : static void ReorderBufferCheckMemoryLimit(ReorderBuffer *rb);
267 : static void ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
268 : static void ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
269 : int fd, ReorderBufferChange *change);
270 : static Size ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
271 : TXNEntryFile *file, XLogSegNo *segno);
272 : static void ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
273 : char *data);
274 : static void ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn);
275 : static void ReorderBufferTruncateTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
276 : bool txn_prepared);
277 : static void ReorderBufferMaybeMarkTXNStreamed(ReorderBuffer *rb, ReorderBufferTXN *txn);
278 : static bool ReorderBufferCheckAndTruncateAbortedTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
279 : static void ReorderBufferCleanupSerializedTXNs(const char *slotname);
280 : static void ReorderBufferSerializedPath(char *path, ReplicationSlot *slot,
281 : TransactionId xid, XLogSegNo segno);
282 : static int ReorderBufferTXNSizeCompare(const pairingheap_node *a, const pairingheap_node *b, void *arg);
283 :
284 : static void ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap);
285 : static Snapshot ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
286 : ReorderBufferTXN *txn, CommandId cid);
287 :
288 : /*
289 : * ---------------------------------------
290 : * Streaming support functions
291 : * ---------------------------------------
292 : */
293 : static inline bool ReorderBufferCanStream(ReorderBuffer *rb);
294 : static inline bool ReorderBufferCanStartStreaming(ReorderBuffer *rb);
295 : static void ReorderBufferStreamTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
296 : static void ReorderBufferStreamCommit(ReorderBuffer *rb, ReorderBufferTXN *txn);
297 :
298 : /* ---------------------------------------
299 : * toast reassembly support
300 : * ---------------------------------------
301 : */
302 : static void ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn);
303 : static void ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn);
304 : static void ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
305 : Relation relation, ReorderBufferChange *change);
306 : static void ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
307 : Relation relation, ReorderBufferChange *change);
308 :
309 : /*
310 : * ---------------------------------------
311 : * memory accounting
312 : * ---------------------------------------
313 : */
314 : static Size ReorderBufferChangeSize(ReorderBufferChange *change);
315 : static void ReorderBufferChangeMemoryUpdate(ReorderBuffer *rb,
316 : ReorderBufferChange *change,
317 : ReorderBufferTXN *txn,
318 : bool addition, Size sz);
319 :
320 : /*
321 : * Allocate a new ReorderBuffer and clean out any old serialized state from
322 : * prior ReorderBuffer instances for the same slot.
323 : */
324 : ReorderBuffer *
325 1223 : ReorderBufferAllocate(void)
326 : {
327 : ReorderBuffer *buffer;
328 : HASHCTL hash_ctl;
329 : MemoryContext new_ctx;
330 :
331 : Assert(MyReplicationSlot != NULL);
332 :
333 : /* allocate memory in own context, to have better accountability */
334 1223 : new_ctx = AllocSetContextCreate(CurrentMemoryContext,
335 : "ReorderBuffer",
336 : ALLOCSET_DEFAULT_SIZES);
337 :
338 : buffer =
339 1223 : (ReorderBuffer *) MemoryContextAlloc(new_ctx, sizeof(ReorderBuffer));
340 :
341 1223 : memset(&hash_ctl, 0, sizeof(hash_ctl));
342 :
343 1223 : buffer->context = new_ctx;
344 :
345 1223 : buffer->change_context = SlabContextCreate(new_ctx,
346 : "Change",
347 : SLAB_DEFAULT_BLOCK_SIZE,
348 : sizeof(ReorderBufferChange));
349 :
350 1223 : buffer->txn_context = SlabContextCreate(new_ctx,
351 : "TXN",
352 : SLAB_DEFAULT_BLOCK_SIZE,
353 : sizeof(ReorderBufferTXN));
354 :
355 : /*
356 : * To minimize memory fragmentation caused by long-running transactions
357 : * with changes spanning multiple memory blocks, we use a single
358 : * fixed-size memory block for decoded tuple storage. The performance
359 : * testing showed that the default memory block size maintains logical
360 : * decoding performance without causing fragmentation due to concurrent
361 : * transactions. One might think that we can use the max size as
362 : * SLAB_LARGE_BLOCK_SIZE but the test also showed it doesn't help resolve
363 : * the memory fragmentation.
364 : */
365 1223 : buffer->tup_context = GenerationContextCreate(new_ctx,
366 : "Tuples",
367 : SLAB_DEFAULT_BLOCK_SIZE,
368 : SLAB_DEFAULT_BLOCK_SIZE,
369 : SLAB_DEFAULT_BLOCK_SIZE);
370 :
371 1223 : hash_ctl.keysize = sizeof(TransactionId);
372 1223 : hash_ctl.entrysize = sizeof(ReorderBufferTXNByIdEnt);
373 1223 : hash_ctl.hcxt = buffer->context;
374 :
375 1223 : buffer->by_txn = hash_create("ReorderBufferByXid", 1000, &hash_ctl,
376 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
377 :
378 1223 : buffer->by_txn_last_xid = InvalidTransactionId;
379 1223 : buffer->by_txn_last_txn = NULL;
380 :
381 1223 : buffer->outbuf = NULL;
382 1223 : buffer->outbufsize = 0;
383 1223 : buffer->size = 0;
384 :
385 : /* txn_heap is ordered by transaction size */
386 1223 : buffer->txn_heap = pairingheap_allocate(ReorderBufferTXNSizeCompare, NULL);
387 :
388 1223 : buffer->spillTxns = 0;
389 1223 : buffer->spillCount = 0;
390 1223 : buffer->spillBytes = 0;
391 1223 : buffer->streamTxns = 0;
392 1223 : buffer->streamCount = 0;
393 1223 : buffer->streamBytes = 0;
394 1223 : buffer->memExceededCount = 0;
395 1223 : buffer->totalTxns = 0;
396 1223 : buffer->totalBytes = 0;
397 :
398 1223 : buffer->current_restart_decoding_lsn = InvalidXLogRecPtr;
399 :
400 1223 : dlist_init(&buffer->toplevel_by_lsn);
401 1223 : dlist_init(&buffer->txns_by_base_snapshot_lsn);
402 1223 : dclist_init(&buffer->catchange_txns);
403 :
404 : /*
405 : * Ensure there's no stale data from prior uses of this slot, in case some
406 : * prior exit avoided calling ReorderBufferFree. Failure to do this can
407 : * produce duplicated txns, and it's very cheap if there's nothing there.
408 : */
409 1223 : ReorderBufferCleanupSerializedTXNs(NameStr(MyReplicationSlot->data.name));
410 :
411 1223 : return buffer;
412 : }
413 :
414 : /*
415 : * Free a ReorderBuffer
416 : */
417 : void
418 953 : ReorderBufferFree(ReorderBuffer *rb)
419 : {
420 953 : MemoryContext context = rb->context;
421 :
422 : /*
423 : * We free separately allocated data by entirely scrapping reorderbuffer's
424 : * memory context.
425 : */
426 953 : MemoryContextDelete(context);
427 :
428 : /* Free disk space used by unconsumed reorder buffers */
429 953 : ReorderBufferCleanupSerializedTXNs(NameStr(MyReplicationSlot->data.name));
430 953 : }
431 :
432 : /*
433 : * Allocate a new ReorderBufferTXN.
434 : */
435 : static ReorderBufferTXN *
436 4848 : ReorderBufferAllocTXN(ReorderBuffer *rb)
437 : {
438 : ReorderBufferTXN *txn;
439 :
440 : txn = (ReorderBufferTXN *)
441 4848 : MemoryContextAlloc(rb->txn_context, sizeof(ReorderBufferTXN));
442 :
443 4848 : memset(txn, 0, sizeof(ReorderBufferTXN));
444 :
445 4848 : dlist_init(&txn->changes);
446 4848 : dlist_init(&txn->tuplecids);
447 4848 : dlist_init(&txn->subtxns);
448 :
449 : /* InvalidCommandId is not zero, so set it explicitly */
450 4848 : txn->command_id = InvalidCommandId;
451 4848 : txn->output_plugin_private = NULL;
452 :
453 4848 : return txn;
454 : }
455 :
456 : /*
457 : * Free a ReorderBufferTXN.
458 : */
459 : static void
460 4778 : ReorderBufferFreeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
461 : {
462 : /* clean the lookup cache if we were cached (quite likely) */
463 4778 : if (rb->by_txn_last_xid == txn->xid)
464 : {
465 4590 : rb->by_txn_last_xid = InvalidTransactionId;
466 4590 : rb->by_txn_last_txn = NULL;
467 : }
468 :
469 : /* free data that's contained */
470 :
471 4778 : if (txn->gid != NULL)
472 : {
473 46 : pfree(txn->gid);
474 46 : txn->gid = NULL;
475 : }
476 :
477 4778 : if (txn->tuplecid_hash != NULL)
478 : {
479 939 : hash_destroy(txn->tuplecid_hash);
480 939 : txn->tuplecid_hash = NULL;
481 : }
482 :
483 4778 : if (txn->invalidations)
484 : {
485 1572 : pfree(txn->invalidations);
486 1572 : txn->invalidations = NULL;
487 : }
488 :
489 4778 : if (txn->invalidations_distributed)
490 : {
491 21 : pfree(txn->invalidations_distributed);
492 21 : txn->invalidations_distributed = NULL;
493 : }
494 :
495 : /* Reset the toast hash */
496 4778 : ReorderBufferToastReset(rb, txn);
497 :
498 : /* All changes must be deallocated */
499 : Assert(txn->size == 0);
500 :
501 4778 : pfree(txn);
502 4778 : }
503 :
504 : /*
505 : * Allocate a ReorderBufferChange.
506 : */
507 : ReorderBufferChange *
508 1798586 : ReorderBufferAllocChange(ReorderBuffer *rb)
509 : {
510 : ReorderBufferChange *change;
511 :
512 : change = (ReorderBufferChange *)
513 1798586 : MemoryContextAlloc(rb->change_context, sizeof(ReorderBufferChange));
514 :
515 1798586 : memset(change, 0, sizeof(ReorderBufferChange));
516 1798586 : return change;
517 : }
518 :
519 : /*
520 : * Free a ReorderBufferChange and update memory accounting, if requested.
521 : */
522 : void
523 1777851 : ReorderBufferFreeChange(ReorderBuffer *rb, ReorderBufferChange *change,
524 : bool upd_mem)
525 : {
526 : /* update memory accounting info */
527 1777851 : if (upd_mem)
528 205293 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, false,
529 : ReorderBufferChangeSize(change));
530 :
531 : /* free contained data */
532 1777851 : switch (change->action)
533 : {
534 1695793 : case REORDER_BUFFER_CHANGE_INSERT:
535 : case REORDER_BUFFER_CHANGE_UPDATE:
536 : case REORDER_BUFFER_CHANGE_DELETE:
537 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
538 1695793 : if (change->data.tp.newtuple)
539 : {
540 1445835 : ReorderBufferFreeTupleBuf(change->data.tp.newtuple);
541 1445835 : change->data.tp.newtuple = NULL;
542 : }
543 :
544 1695793 : if (change->data.tp.oldtuple)
545 : {
546 181262 : ReorderBufferFreeTupleBuf(change->data.tp.oldtuple);
547 181262 : change->data.tp.oldtuple = NULL;
548 : }
549 1695793 : break;
550 40 : case REORDER_BUFFER_CHANGE_MESSAGE:
551 40 : if (change->data.msg.prefix != NULL)
552 40 : pfree(change->data.msg.prefix);
553 40 : change->data.msg.prefix = NULL;
554 40 : if (change->data.msg.message != NULL)
555 40 : pfree(change->data.msg.message);
556 40 : change->data.msg.message = NULL;
557 40 : break;
558 5890 : case REORDER_BUFFER_CHANGE_INVALIDATION:
559 5890 : if (change->data.inval.invalidations)
560 5890 : pfree(change->data.inval.invalidations);
561 5890 : change->data.inval.invalidations = NULL;
562 5890 : break;
563 1613 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
564 1613 : if (change->data.snapshot)
565 : {
566 1613 : ReorderBufferFreeSnap(rb, change->data.snapshot);
567 1613 : change->data.snapshot = NULL;
568 : }
569 1613 : break;
570 : /* no data in addition to the struct itself */
571 56 : case REORDER_BUFFER_CHANGE_TRUNCATE:
572 56 : if (change->data.truncate.relids != NULL)
573 : {
574 56 : ReorderBufferFreeRelids(rb, change->data.truncate.relids);
575 56 : change->data.truncate.relids = NULL;
576 : }
577 56 : break;
578 74459 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
579 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
580 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
581 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
582 74459 : break;
583 : }
584 :
585 1777851 : pfree(change);
586 1777851 : }
587 :
588 : /*
589 : * Allocate a HeapTuple fitting a tuple of size tuple_len (excluding header
590 : * overhead).
591 : */
592 : HeapTuple
593 1647143 : ReorderBufferAllocTupleBuf(ReorderBuffer *rb, Size tuple_len)
594 : {
595 : HeapTuple tuple;
596 : Size alloc_len;
597 :
598 1647143 : alloc_len = tuple_len + SizeofHeapTupleHeader;
599 :
600 1647143 : tuple = (HeapTuple) MemoryContextAlloc(rb->tup_context,
601 : HEAPTUPLESIZE + alloc_len);
602 1647143 : tuple->t_data = (HeapTupleHeader) ((char *) tuple + HEAPTUPLESIZE);
603 :
604 1647143 : return tuple;
605 : }
606 :
607 : /*
608 : * Free a HeapTuple returned by ReorderBufferAllocTupleBuf().
609 : */
610 : void
611 1627097 : ReorderBufferFreeTupleBuf(HeapTuple tuple)
612 : {
613 1627097 : pfree(tuple);
614 1627097 : }
615 :
616 : /*
617 : * Allocate an array for relids of truncated relations.
618 : *
619 : * We use the global memory context (for the whole reorder buffer), because
620 : * none of the existing ones seems like a good match (some are SLAB, so we
621 : * can't use those, and tup_context is meant for tuple data, not relids). We
622 : * could add yet another context, but it seems like an overkill - TRUNCATE is
623 : * not particularly common operation, so it does not seem worth it.
624 : */
625 : Oid *
626 61 : ReorderBufferAllocRelids(ReorderBuffer *rb, int nrelids)
627 : {
628 : Oid *relids;
629 : Size alloc_len;
630 :
631 61 : alloc_len = sizeof(Oid) * nrelids;
632 :
633 61 : relids = (Oid *) MemoryContextAlloc(rb->context, alloc_len);
634 :
635 61 : return relids;
636 : }
637 :
638 : /*
639 : * Free an array of relids.
640 : */
641 : void
642 56 : ReorderBufferFreeRelids(ReorderBuffer *rb, Oid *relids)
643 : {
644 56 : pfree(relids);
645 56 : }
646 :
647 : /*
648 : * Return the ReorderBufferTXN from the given buffer, specified by Xid.
649 : * If create is true, and a transaction doesn't already exist, create it
650 : * (with the given LSN, and as top transaction if that's specified);
651 : * when this happens, is_new is set to true.
652 : */
653 : static ReorderBufferTXN *
654 6023046 : ReorderBufferTXNByXid(ReorderBuffer *rb, TransactionId xid, bool create,
655 : bool *is_new, XLogRecPtr lsn, bool create_as_top)
656 : {
657 : ReorderBufferTXN *txn;
658 : ReorderBufferTXNByIdEnt *ent;
659 : bool found;
660 :
661 : Assert(TransactionIdIsValid(xid));
662 :
663 : /*
664 : * Check the one-entry lookup cache first
665 : */
666 6023046 : if (TransactionIdIsValid(rb->by_txn_last_xid) &&
667 6018405 : rb->by_txn_last_xid == xid)
668 : {
669 5148514 : txn = rb->by_txn_last_txn;
670 :
671 5148514 : if (txn != NULL)
672 : {
673 : /* found it, and it's valid */
674 5148458 : if (is_new)
675 4102 : *is_new = false;
676 5148458 : return txn;
677 : }
678 :
679 : /*
680 : * cached as non-existent, and asked not to create? Then nothing else
681 : * to do.
682 : */
683 56 : if (!create)
684 53 : return NULL;
685 : /* otherwise fall through to create it */
686 : }
687 :
688 : /*
689 : * If the cache wasn't hit or it yielded a "does-not-exist" and we want to
690 : * create an entry.
691 : */
692 :
693 : /* search the lookup table */
694 : ent = (ReorderBufferTXNByIdEnt *)
695 874535 : hash_search(rb->by_txn,
696 : &xid,
697 : create ? HASH_ENTER : HASH_FIND,
698 : &found);
699 874535 : if (found)
700 868360 : txn = ent->txn;
701 6175 : else if (create)
702 : {
703 : /* initialize the new entry, if creation was requested */
704 : Assert(ent != NULL);
705 : Assert(XLogRecPtrIsValid(lsn));
706 :
707 4848 : ent->txn = ReorderBufferAllocTXN(rb);
708 4848 : ent->txn->xid = xid;
709 4848 : txn = ent->txn;
710 4848 : txn->first_lsn = lsn;
711 4848 : txn->restart_decoding_lsn = rb->current_restart_decoding_lsn;
712 :
713 4848 : if (create_as_top)
714 : {
715 4177 : dlist_push_tail(&rb->toplevel_by_lsn, &txn->node);
716 4177 : AssertTXNLsnOrder(rb);
717 : }
718 : }
719 : else
720 1327 : txn = NULL; /* not found and not asked to create */
721 :
722 : /* update cache */
723 874535 : rb->by_txn_last_xid = xid;
724 874535 : rb->by_txn_last_txn = txn;
725 :
726 874535 : if (is_new)
727 1770 : *is_new = !found;
728 :
729 : Assert(!create || txn != NULL);
730 874535 : return txn;
731 : }
732 :
733 : /*
734 : * Record the partial change for the streaming of in-progress transactions. We
735 : * can stream only complete changes so if we have a partial change like toast
736 : * table insert or speculative insert then we mark such a 'txn' so that it
737 : * can't be streamed. We also ensure that if the changes in such a 'txn' can
738 : * be streamed and are above logical_decoding_work_mem threshold then we stream
739 : * them as soon as we have a complete change.
740 : */
741 : static void
742 1583573 : ReorderBufferProcessPartialChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
743 : ReorderBufferChange *change,
744 : bool toast_insert)
745 : {
746 : ReorderBufferTXN *toptxn;
747 :
748 : /*
749 : * The partial changes need to be processed only while streaming
750 : * in-progress transactions.
751 : */
752 1583573 : if (!ReorderBufferCanStream(rb))
753 1127739 : return;
754 :
755 : /* Get the top transaction. */
756 455834 : toptxn = rbtxn_get_toptxn(txn);
757 :
758 : /*
759 : * Indicate a partial change for toast inserts. The change will be
760 : * considered as complete once we get the insert or update on the main
761 : * table and we are sure that the pending toast chunks are not required
762 : * anymore.
763 : *
764 : * If we allow streaming when there are pending toast chunks then such
765 : * chunks won't be released till the insert (multi_insert) is complete and
766 : * we expect the txn to have streamed all changes after streaming. This
767 : * restriction is mainly to ensure the correctness of streamed
768 : * transactions and it doesn't seem worth uplifting such a restriction
769 : * just to allow this case because anyway we will stream the transaction
770 : * once such an insert is complete.
771 : */
772 455834 : if (toast_insert)
773 1649 : toptxn->txn_flags |= RBTXN_HAS_PARTIAL_CHANGE;
774 454185 : else if (rbtxn_has_partial_change(toptxn) &&
775 57 : IsInsertOrUpdate(change->action) &&
776 57 : change->data.tp.clear_toast_afterwards)
777 37 : toptxn->txn_flags &= ~RBTXN_HAS_PARTIAL_CHANGE;
778 :
779 : /*
780 : * Indicate a partial change for speculative inserts. The change will be
781 : * considered as complete once we get the speculative confirm or abort
782 : * token.
783 : */
784 455834 : if (IsSpecInsert(change->action))
785 0 : toptxn->txn_flags |= RBTXN_HAS_PARTIAL_CHANGE;
786 455834 : else if (rbtxn_has_partial_change(toptxn) &&
787 1669 : IsSpecConfirmOrAbort(change->action))
788 0 : toptxn->txn_flags &= ~RBTXN_HAS_PARTIAL_CHANGE;
789 :
790 : /*
791 : * Stream the transaction if it is serialized before and the changes are
792 : * now complete in the top-level transaction.
793 : *
794 : * The reason for doing the streaming of such a transaction as soon as we
795 : * get the complete change for it is that previously it would have reached
796 : * the memory threshold and wouldn't get streamed because of incomplete
797 : * changes. Delaying such transactions would increase apply lag for them.
798 : */
799 455834 : if (ReorderBufferCanStartStreaming(rb) &&
800 197427 : !(rbtxn_has_partial_change(toptxn)) &&
801 195896 : rbtxn_is_serialized(txn) &&
802 39 : rbtxn_has_streamable_change(toptxn))
803 9 : ReorderBufferStreamTXN(rb, toptxn);
804 : }
805 :
806 : /*
807 : * Queue a change into a transaction so it can be replayed upon commit or will be
808 : * streamed when we reach logical_decoding_work_mem threshold.
809 : */
810 : void
811 1592982 : ReorderBufferQueueChange(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
812 : ReorderBufferChange *change, bool toast_insert)
813 : {
814 : ReorderBufferTXN *txn;
815 :
816 1592982 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
817 :
818 : /*
819 : * If we have detected that the transaction is aborted while streaming the
820 : * previous changes or by checking its CLOG, there is no point in
821 : * collecting further changes for it.
822 : */
823 1592982 : if (rbtxn_is_aborted(txn))
824 : {
825 : /*
826 : * We don't need to update memory accounting for this change as we
827 : * have not added it to the queue yet.
828 : */
829 9409 : ReorderBufferFreeChange(rb, change, false);
830 9409 : return;
831 : }
832 :
833 : /*
834 : * The changes that are sent downstream are considered streamable. We
835 : * remember such transactions so that only those will later be considered
836 : * for streaming.
837 : */
838 1583573 : if (change->action == REORDER_BUFFER_CHANGE_INSERT ||
839 489735 : change->action == REORDER_BUFFER_CHANGE_UPDATE ||
840 308999 : change->action == REORDER_BUFFER_CHANGE_DELETE ||
841 71058 : change->action == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT ||
842 53093 : change->action == REORDER_BUFFER_CHANGE_TRUNCATE ||
843 53042 : change->action == REORDER_BUFFER_CHANGE_MESSAGE)
844 : {
845 1530570 : ReorderBufferTXN *toptxn = rbtxn_get_toptxn(txn);
846 :
847 1530570 : toptxn->txn_flags |= RBTXN_HAS_STREAMABLE_CHANGE;
848 : }
849 :
850 1583573 : change->lsn = lsn;
851 1583573 : change->txn = txn;
852 :
853 : Assert(XLogRecPtrIsValid(lsn));
854 1583573 : dlist_push_tail(&txn->changes, &change->node);
855 1583573 : txn->nentries++;
856 1583573 : txn->nentries_mem++;
857 :
858 : /* update memory accounting information */
859 1583573 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
860 : ReorderBufferChangeSize(change));
861 :
862 : /* process partial change */
863 1583573 : ReorderBufferProcessPartialChange(rb, txn, change, toast_insert);
864 :
865 : /* check the memory limits and evict something if needed */
866 1583573 : ReorderBufferCheckMemoryLimit(rb);
867 : }
868 :
869 : /*
870 : * A transactional message is queued to be processed upon commit and a
871 : * non-transactional message gets processed immediately.
872 : */
873 : void
874 47 : ReorderBufferQueueMessage(ReorderBuffer *rb, TransactionId xid,
875 : Snapshot snap, XLogRecPtr lsn,
876 : bool transactional, const char *prefix,
877 : Size message_size, const char *message)
878 : {
879 47 : if (transactional)
880 : {
881 : MemoryContext oldcontext;
882 : ReorderBufferChange *change;
883 :
884 : Assert(xid != InvalidTransactionId);
885 :
886 : /*
887 : * We don't expect snapshots for transactional changes - we'll use the
888 : * snapshot derived later during apply (unless the change gets
889 : * skipped).
890 : */
891 : Assert(!snap);
892 :
893 39 : oldcontext = MemoryContextSwitchTo(rb->context);
894 :
895 39 : change = ReorderBufferAllocChange(rb);
896 39 : change->action = REORDER_BUFFER_CHANGE_MESSAGE;
897 39 : change->data.msg.prefix = pstrdup(prefix);
898 39 : change->data.msg.message_size = message_size;
899 39 : change->data.msg.message = palloc(message_size);
900 39 : memcpy(change->data.msg.message, message, message_size);
901 :
902 39 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
903 :
904 39 : MemoryContextSwitchTo(oldcontext);
905 : }
906 : else
907 : {
908 8 : ReorderBufferTXN *txn = NULL;
909 8 : volatile Snapshot snapshot_now = snap;
910 :
911 : /* Non-transactional changes require a valid snapshot. */
912 : Assert(snapshot_now);
913 :
914 8 : if (xid != InvalidTransactionId)
915 3 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
916 :
917 : /* setup snapshot to allow catalog access */
918 8 : SetupHistoricSnapshot(snapshot_now, NULL);
919 8 : PG_TRY();
920 : {
921 8 : rb->message(rb, txn, lsn, false, prefix, message_size, message);
922 :
923 8 : TeardownHistoricSnapshot(false);
924 : }
925 0 : PG_CATCH();
926 : {
927 0 : TeardownHistoricSnapshot(true);
928 0 : PG_RE_THROW();
929 : }
930 8 : PG_END_TRY();
931 : }
932 47 : }
933 :
934 : /*
935 : * AssertTXNLsnOrder
936 : * Verify LSN ordering of transaction lists in the reorderbuffer
937 : *
938 : * Other LSN-related invariants are checked too.
939 : *
940 : * No-op if assertions are not in use.
941 : */
942 : static void
943 10063 : AssertTXNLsnOrder(ReorderBuffer *rb)
944 : {
945 : #ifdef USE_ASSERT_CHECKING
946 : LogicalDecodingContext *ctx = rb->private_data;
947 : dlist_iter iter;
948 : XLogRecPtr prev_first_lsn = InvalidXLogRecPtr;
949 : XLogRecPtr prev_base_snap_lsn = InvalidXLogRecPtr;
950 :
951 : /*
952 : * Skip the verification if we don't reach the LSN at which we start
953 : * decoding the contents of transactions yet because until we reach the
954 : * LSN, we could have transactions that don't have the association between
955 : * the top-level transaction and subtransaction yet and consequently have
956 : * the same LSN. We don't guarantee this association until we try to
957 : * decode the actual contents of transaction. The ordering of the records
958 : * prior to the start_decoding_at LSN should have been checked before the
959 : * restart.
960 : */
961 : if (SnapBuildXactNeedsSkip(ctx->snapshot_builder, ctx->reader->EndRecPtr))
962 : return;
963 :
964 : dlist_foreach(iter, &rb->toplevel_by_lsn)
965 : {
966 : ReorderBufferTXN *cur_txn = dlist_container(ReorderBufferTXN, node,
967 : iter.cur);
968 :
969 : /* start LSN must be set */
970 : Assert(XLogRecPtrIsValid(cur_txn->first_lsn));
971 :
972 : /* If there is an end LSN, it must be higher than start LSN */
973 : if (XLogRecPtrIsValid(cur_txn->end_lsn))
974 : Assert(cur_txn->first_lsn <= cur_txn->end_lsn);
975 :
976 : /* Current initial LSN must be strictly higher than previous */
977 : if (XLogRecPtrIsValid(prev_first_lsn))
978 : Assert(prev_first_lsn < cur_txn->first_lsn);
979 :
980 : /* known-as-subtxn txns must not be listed */
981 : Assert(!rbtxn_is_known_subxact(cur_txn));
982 :
983 : prev_first_lsn = cur_txn->first_lsn;
984 : }
985 :
986 : dlist_foreach(iter, &rb->txns_by_base_snapshot_lsn)
987 : {
988 : ReorderBufferTXN *cur_txn = dlist_container(ReorderBufferTXN,
989 : base_snapshot_node,
990 : iter.cur);
991 :
992 : /* base snapshot (and its LSN) must be set */
993 : Assert(cur_txn->base_snapshot != NULL);
994 : Assert(XLogRecPtrIsValid(cur_txn->base_snapshot_lsn));
995 :
996 : /* current LSN must be strictly higher than previous */
997 : if (XLogRecPtrIsValid(prev_base_snap_lsn))
998 : Assert(prev_base_snap_lsn < cur_txn->base_snapshot_lsn);
999 :
1000 : /* known-as-subtxn txns must not be listed */
1001 : Assert(!rbtxn_is_known_subxact(cur_txn));
1002 :
1003 : prev_base_snap_lsn = cur_txn->base_snapshot_lsn;
1004 : }
1005 : #endif
1006 10063 : }
1007 :
1008 : /*
1009 : * AssertChangeLsnOrder
1010 : *
1011 : * Check ordering of changes in the (sub)transaction.
1012 : */
1013 : static void
1014 3076 : AssertChangeLsnOrder(ReorderBufferTXN *txn)
1015 : {
1016 : #ifdef USE_ASSERT_CHECKING
1017 : dlist_iter iter;
1018 : XLogRecPtr prev_lsn = txn->first_lsn;
1019 :
1020 : dlist_foreach(iter, &txn->changes)
1021 : {
1022 : ReorderBufferChange *cur_change;
1023 :
1024 : cur_change = dlist_container(ReorderBufferChange, node, iter.cur);
1025 :
1026 : Assert(XLogRecPtrIsValid(txn->first_lsn));
1027 : Assert(XLogRecPtrIsValid(cur_change->lsn));
1028 : Assert(txn->first_lsn <= cur_change->lsn);
1029 :
1030 : if (XLogRecPtrIsValid(txn->end_lsn))
1031 : Assert(cur_change->lsn <= txn->end_lsn);
1032 :
1033 : Assert(prev_lsn <= cur_change->lsn);
1034 :
1035 : prev_lsn = cur_change->lsn;
1036 : }
1037 : #endif
1038 3076 : }
1039 :
1040 : /*
1041 : * ReorderBufferGetOldestTXN
1042 : * Return oldest transaction in reorderbuffer
1043 : */
1044 : ReorderBufferTXN *
1045 523 : ReorderBufferGetOldestTXN(ReorderBuffer *rb)
1046 : {
1047 : ReorderBufferTXN *txn;
1048 :
1049 523 : AssertTXNLsnOrder(rb);
1050 :
1051 523 : if (dlist_is_empty(&rb->toplevel_by_lsn))
1052 465 : return NULL;
1053 :
1054 58 : txn = dlist_head_element(ReorderBufferTXN, node, &rb->toplevel_by_lsn);
1055 :
1056 : Assert(!rbtxn_is_known_subxact(txn));
1057 : Assert(XLogRecPtrIsValid(txn->first_lsn));
1058 58 : return txn;
1059 : }
1060 :
1061 : /*
1062 : * ReorderBufferGetOldestXmin
1063 : * Return oldest Xmin in reorderbuffer
1064 : *
1065 : * Returns oldest possibly running Xid from the point of view of snapshots
1066 : * used in the transactions kept by reorderbuffer, or InvalidTransactionId if
1067 : * there are none.
1068 : *
1069 : * Since snapshots are assigned monotonically, this equals the Xmin of the
1070 : * base snapshot with minimal base_snapshot_lsn.
1071 : */
1072 : TransactionId
1073 540 : ReorderBufferGetOldestXmin(ReorderBuffer *rb)
1074 : {
1075 : ReorderBufferTXN *txn;
1076 :
1077 540 : AssertTXNLsnOrder(rb);
1078 :
1079 540 : if (dlist_is_empty(&rb->txns_by_base_snapshot_lsn))
1080 491 : return InvalidTransactionId;
1081 :
1082 49 : txn = dlist_head_element(ReorderBufferTXN, base_snapshot_node,
1083 : &rb->txns_by_base_snapshot_lsn);
1084 49 : return txn->base_snapshot->xmin;
1085 : }
1086 :
1087 : void
1088 619 : ReorderBufferSetRestartPoint(ReorderBuffer *rb, XLogRecPtr ptr)
1089 : {
1090 619 : rb->current_restart_decoding_lsn = ptr;
1091 619 : }
1092 :
1093 : /*
1094 : * ReorderBufferAssignChild
1095 : *
1096 : * Make note that we know that subxid is a subtransaction of xid, seen as of
1097 : * the given lsn.
1098 : */
1099 : void
1100 860 : ReorderBufferAssignChild(ReorderBuffer *rb, TransactionId xid,
1101 : TransactionId subxid, XLogRecPtr lsn)
1102 : {
1103 : ReorderBufferTXN *txn;
1104 : ReorderBufferTXN *subtxn;
1105 : bool new_top;
1106 : bool new_sub;
1107 :
1108 860 : txn = ReorderBufferTXNByXid(rb, xid, true, &new_top, lsn, true);
1109 860 : subtxn = ReorderBufferTXNByXid(rb, subxid, true, &new_sub, lsn, false);
1110 :
1111 860 : if (!new_sub)
1112 : {
1113 189 : if (rbtxn_is_known_subxact(subtxn))
1114 : {
1115 : /* already associated, nothing to do */
1116 189 : return;
1117 : }
1118 : else
1119 : {
1120 : /*
1121 : * We already saw this transaction, but initially added it to the
1122 : * list of top-level txns. Now that we know it's not top-level,
1123 : * remove it from there.
1124 : */
1125 0 : dlist_delete(&subtxn->node);
1126 : }
1127 : }
1128 :
1129 671 : subtxn->txn_flags |= RBTXN_IS_SUBXACT;
1130 671 : subtxn->toplevel_xid = xid;
1131 : Assert(subtxn->nsubtxns == 0);
1132 :
1133 : /* set the reference to top-level transaction */
1134 671 : subtxn->toptxn = txn;
1135 :
1136 : /* add to subtransaction list */
1137 671 : dlist_push_tail(&txn->subtxns, &subtxn->node);
1138 671 : txn->nsubtxns++;
1139 :
1140 : /* Possibly transfer the subtxn's snapshot to its top-level txn. */
1141 671 : ReorderBufferTransferSnapToParent(txn, subtxn);
1142 :
1143 : /* Verify LSN-ordering invariant */
1144 671 : AssertTXNLsnOrder(rb);
1145 : }
1146 :
1147 : /*
1148 : * ReorderBufferTransferSnapToParent
1149 : * Transfer base snapshot from subtxn to top-level txn, if needed
1150 : *
1151 : * This is done if the top-level txn doesn't have a base snapshot, or if the
1152 : * subtxn's base snapshot has an earlier LSN than the top-level txn's base
1153 : * snapshot's LSN. This can happen if there are no changes in the toplevel
1154 : * txn but there are some in the subtxn, or the first change in subtxn has
1155 : * earlier LSN than first change in the top-level txn and we learned about
1156 : * their kinship only now.
1157 : *
1158 : * The subtransaction's snapshot is cleared regardless of the transfer
1159 : * happening, since it's not needed anymore in either case.
1160 : *
1161 : * We do this as soon as we become aware of their kinship, to avoid queueing
1162 : * extra snapshots to txns known-as-subtxns -- only top-level txns will
1163 : * receive further snapshots.
1164 : */
1165 : static void
1166 675 : ReorderBufferTransferSnapToParent(ReorderBufferTXN *txn,
1167 : ReorderBufferTXN *subtxn)
1168 : {
1169 : Assert(subtxn->toplevel_xid == txn->xid);
1170 :
1171 675 : if (subtxn->base_snapshot != NULL)
1172 : {
1173 0 : if (txn->base_snapshot == NULL ||
1174 0 : subtxn->base_snapshot_lsn < txn->base_snapshot_lsn)
1175 : {
1176 : /*
1177 : * If the toplevel transaction already has a base snapshot but
1178 : * it's newer than the subxact's, purge it.
1179 : */
1180 0 : if (txn->base_snapshot != NULL)
1181 : {
1182 0 : SnapBuildSnapDecRefcount(txn->base_snapshot);
1183 0 : dlist_delete(&txn->base_snapshot_node);
1184 : }
1185 :
1186 : /*
1187 : * The snapshot is now the top transaction's; transfer it, and
1188 : * adjust the list position of the top transaction in the list by
1189 : * moving it to where the subtransaction is.
1190 : */
1191 0 : txn->base_snapshot = subtxn->base_snapshot;
1192 0 : txn->base_snapshot_lsn = subtxn->base_snapshot_lsn;
1193 0 : dlist_insert_before(&subtxn->base_snapshot_node,
1194 : &txn->base_snapshot_node);
1195 :
1196 : /*
1197 : * The subtransaction doesn't have a snapshot anymore (so it
1198 : * mustn't be in the list.)
1199 : */
1200 0 : subtxn->base_snapshot = NULL;
1201 0 : subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
1202 0 : dlist_delete(&subtxn->base_snapshot_node);
1203 : }
1204 : else
1205 : {
1206 : /* Base snap of toplevel is fine, so subxact's is not needed */
1207 0 : SnapBuildSnapDecRefcount(subtxn->base_snapshot);
1208 0 : dlist_delete(&subtxn->base_snapshot_node);
1209 0 : subtxn->base_snapshot = NULL;
1210 0 : subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
1211 : }
1212 : }
1213 675 : }
1214 :
1215 : /*
1216 : * Associate a subtransaction with its toplevel transaction at commit
1217 : * time. There may be no further changes added after this.
1218 : */
1219 : void
1220 270 : ReorderBufferCommitChild(ReorderBuffer *rb, TransactionId xid,
1221 : TransactionId subxid, XLogRecPtr commit_lsn,
1222 : XLogRecPtr end_lsn)
1223 : {
1224 : ReorderBufferTXN *subtxn;
1225 :
1226 270 : subtxn = ReorderBufferTXNByXid(rb, subxid, false, NULL,
1227 : InvalidXLogRecPtr, false);
1228 :
1229 : /*
1230 : * No need to do anything if that subtxn didn't contain any changes
1231 : */
1232 270 : if (!subtxn)
1233 81 : return;
1234 :
1235 189 : subtxn->final_lsn = commit_lsn;
1236 189 : subtxn->end_lsn = end_lsn;
1237 :
1238 : /*
1239 : * Assign this subxact as a child of the toplevel xact (no-op if already
1240 : * done.)
1241 : */
1242 189 : ReorderBufferAssignChild(rb, xid, subxid, InvalidXLogRecPtr);
1243 : }
1244 :
1245 :
1246 : /*
1247 : * Support for efficiently iterating over a transaction's and its
1248 : * subtransactions' changes.
1249 : *
1250 : * We do by doing a k-way merge between transactions/subtransactions. For that
1251 : * we model the current heads of the different transactions as a binary heap
1252 : * so we easily know which (sub-)transaction has the change with the smallest
1253 : * lsn next.
1254 : *
1255 : * We assume the changes in individual transactions are already sorted by LSN.
1256 : */
1257 :
1258 : /*
1259 : * Binary heap comparison function.
1260 : */
1261 : static int
1262 51571 : ReorderBufferIterCompare(Datum a, Datum b, void *arg)
1263 : {
1264 51571 : ReorderBufferIterTXNState *state = (ReorderBufferIterTXNState *) arg;
1265 51571 : XLogRecPtr pos_a = state->entries[DatumGetInt32(a)].lsn;
1266 51571 : XLogRecPtr pos_b = state->entries[DatumGetInt32(b)].lsn;
1267 :
1268 51571 : if (pos_a < pos_b)
1269 50714 : return 1;
1270 857 : else if (pos_a == pos_b)
1271 0 : return 0;
1272 857 : return -1;
1273 : }
1274 :
1275 : /*
1276 : * Allocate & initialize an iterator which iterates in lsn order over a
1277 : * transaction and all its subtransactions.
1278 : *
1279 : * Note: The iterator state is returned through iter_state parameter rather
1280 : * than the function's return value. This is because the state gets cleaned up
1281 : * in a PG_CATCH block in the caller, so we want to make sure the caller gets
1282 : * back the state even if this function throws an exception.
1283 : */
1284 : static void
1285 2610 : ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn,
1286 : ReorderBufferIterTXNState *volatile *iter_state)
1287 : {
1288 2610 : Size nr_txns = 0;
1289 : ReorderBufferIterTXNState *state;
1290 : dlist_iter cur_txn_i;
1291 : int32 off;
1292 :
1293 2610 : *iter_state = NULL;
1294 :
1295 : /* Check ordering of changes in the toplevel transaction. */
1296 2610 : AssertChangeLsnOrder(txn);
1297 :
1298 : /*
1299 : * Calculate the size of our heap: one element for every transaction that
1300 : * contains changes. (Besides the transactions already in the reorder
1301 : * buffer, we count the one we were directly passed.)
1302 : */
1303 2610 : if (txn->nentries > 0)
1304 2420 : nr_txns++;
1305 :
1306 3076 : dlist_foreach(cur_txn_i, &txn->subtxns)
1307 : {
1308 : ReorderBufferTXN *cur_txn;
1309 :
1310 466 : cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
1311 :
1312 : /* Check ordering of changes in this subtransaction. */
1313 466 : AssertChangeLsnOrder(cur_txn);
1314 :
1315 466 : if (cur_txn->nentries > 0)
1316 304 : nr_txns++;
1317 : }
1318 :
1319 : /* allocate iteration state */
1320 : state = (ReorderBufferIterTXNState *)
1321 2610 : MemoryContextAllocZero(rb->context,
1322 : sizeof(ReorderBufferIterTXNState) +
1323 2610 : sizeof(ReorderBufferIterTXNEntry) * nr_txns);
1324 :
1325 2610 : state->nr_txns = nr_txns;
1326 2610 : dlist_init(&state->old_change);
1327 :
1328 5334 : for (off = 0; off < state->nr_txns; off++)
1329 : {
1330 2724 : state->entries[off].file.vfd = -1;
1331 2724 : state->entries[off].segno = 0;
1332 : }
1333 :
1334 : /* allocate heap */
1335 2610 : state->heap = binaryheap_allocate(state->nr_txns,
1336 : ReorderBufferIterCompare,
1337 : state);
1338 :
1339 : /* Now that the state fields are initialized, it is safe to return it. */
1340 2610 : *iter_state = state;
1341 :
1342 : /*
1343 : * Now insert items into the binary heap, in an unordered fashion. (We
1344 : * will run a heap assembly step at the end; this is more efficient.)
1345 : */
1346 :
1347 2610 : off = 0;
1348 :
1349 : /* add toplevel transaction if it contains changes */
1350 2610 : if (txn->nentries > 0)
1351 : {
1352 : ReorderBufferChange *cur_change;
1353 :
1354 2420 : if (rbtxn_is_serialized(txn))
1355 : {
1356 : /* serialize remaining changes */
1357 23 : ReorderBufferSerializeTXN(rb, txn);
1358 23 : ReorderBufferRestoreChanges(rb, txn, &state->entries[off].file,
1359 : &state->entries[off].segno);
1360 : }
1361 :
1362 2420 : cur_change = dlist_head_element(ReorderBufferChange, node,
1363 : &txn->changes);
1364 :
1365 2420 : state->entries[off].lsn = cur_change->lsn;
1366 2420 : state->entries[off].change = cur_change;
1367 2420 : state->entries[off].txn = txn;
1368 :
1369 2420 : binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
1370 : }
1371 :
1372 : /* add subtransactions if they contain changes */
1373 3076 : dlist_foreach(cur_txn_i, &txn->subtxns)
1374 : {
1375 : ReorderBufferTXN *cur_txn;
1376 :
1377 466 : cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
1378 :
1379 466 : if (cur_txn->nentries > 0)
1380 : {
1381 : ReorderBufferChange *cur_change;
1382 :
1383 304 : if (rbtxn_is_serialized(cur_txn))
1384 : {
1385 : /* serialize remaining changes */
1386 17 : ReorderBufferSerializeTXN(rb, cur_txn);
1387 17 : ReorderBufferRestoreChanges(rb, cur_txn,
1388 : &state->entries[off].file,
1389 : &state->entries[off].segno);
1390 : }
1391 304 : cur_change = dlist_head_element(ReorderBufferChange, node,
1392 : &cur_txn->changes);
1393 :
1394 304 : state->entries[off].lsn = cur_change->lsn;
1395 304 : state->entries[off].change = cur_change;
1396 304 : state->entries[off].txn = cur_txn;
1397 :
1398 304 : binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
1399 : }
1400 : }
1401 :
1402 : /* assemble a valid binary heap */
1403 2610 : binaryheap_build(state->heap);
1404 2610 : }
1405 :
1406 : /*
1407 : * Return the next change when iterating over a transaction and its
1408 : * subtransactions.
1409 : *
1410 : * Returns NULL when no further changes exist.
1411 : */
1412 : static ReorderBufferChange *
1413 383200 : ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state)
1414 : {
1415 : ReorderBufferChange *change;
1416 : ReorderBufferIterTXNEntry *entry;
1417 : int32 off;
1418 :
1419 : /* nothing there anymore */
1420 383200 : if (binaryheap_empty(state->heap))
1421 2599 : return NULL;
1422 :
1423 380601 : off = DatumGetInt32(binaryheap_first(state->heap));
1424 380601 : entry = &state->entries[off];
1425 :
1426 : /* free memory we might have "leaked" in the previous *Next call */
1427 380601 : if (!dlist_is_empty(&state->old_change))
1428 : {
1429 45 : change = dlist_container(ReorderBufferChange, node,
1430 : dlist_pop_head_node(&state->old_change));
1431 45 : ReorderBufferFreeChange(rb, change, true);
1432 : Assert(dlist_is_empty(&state->old_change));
1433 : }
1434 :
1435 380601 : change = entry->change;
1436 :
1437 : /*
1438 : * update heap with information about which transaction has the next
1439 : * relevant change in LSN order
1440 : */
1441 :
1442 : /* there are in-memory changes */
1443 380601 : if (dlist_has_next(&entry->txn->changes, &entry->change->node))
1444 : {
1445 377845 : dlist_node *next = dlist_next_node(&entry->txn->changes, &change->node);
1446 377845 : ReorderBufferChange *next_change =
1447 : dlist_container(ReorderBufferChange, node, next);
1448 :
1449 : /* txn stays the same */
1450 377845 : state->entries[off].lsn = next_change->lsn;
1451 377845 : state->entries[off].change = next_change;
1452 :
1453 377845 : binaryheap_replace_first(state->heap, Int32GetDatum(off));
1454 377845 : return change;
1455 : }
1456 :
1457 : /* try to load changes from disk */
1458 2756 : if (entry->txn->nentries != entry->txn->nentries_mem)
1459 : {
1460 : /*
1461 : * Ugly: restoring changes will reuse *Change records, thus delete the
1462 : * current one from the per-tx list and only free in the next call.
1463 : */
1464 65 : dlist_delete(&change->node);
1465 65 : dlist_push_tail(&state->old_change, &change->node);
1466 :
1467 : /*
1468 : * Update the total bytes processed by the txn for which we are
1469 : * releasing the current set of changes and restoring the new set of
1470 : * changes.
1471 : */
1472 65 : rb->totalBytes += entry->txn->size;
1473 65 : if (ReorderBufferRestoreChanges(rb, entry->txn, &entry->file,
1474 : &state->entries[off].segno))
1475 : {
1476 : /* successfully restored changes from disk */
1477 : ReorderBufferChange *next_change =
1478 36 : dlist_head_element(ReorderBufferChange, node,
1479 : &entry->txn->changes);
1480 :
1481 36 : elog(DEBUG2, "restored %u/%u changes from disk",
1482 : (uint32) entry->txn->nentries_mem,
1483 : (uint32) entry->txn->nentries);
1484 :
1485 : Assert(entry->txn->nentries_mem);
1486 : /* txn stays the same */
1487 36 : state->entries[off].lsn = next_change->lsn;
1488 36 : state->entries[off].change = next_change;
1489 36 : binaryheap_replace_first(state->heap, Int32GetDatum(off));
1490 :
1491 36 : return change;
1492 : }
1493 : }
1494 :
1495 : /* ok, no changes there anymore, remove */
1496 2720 : binaryheap_remove_first(state->heap);
1497 :
1498 2720 : return change;
1499 : }
1500 :
1501 : /*
1502 : * Deallocate the iterator
1503 : */
1504 : static void
1505 2609 : ReorderBufferIterTXNFinish(ReorderBuffer *rb,
1506 : ReorderBufferIterTXNState *state)
1507 : {
1508 : int32 off;
1509 :
1510 5332 : for (off = 0; off < state->nr_txns; off++)
1511 : {
1512 2723 : if (state->entries[off].file.vfd != -1)
1513 0 : FileClose(state->entries[off].file.vfd);
1514 : }
1515 :
1516 : /* free memory we might have "leaked" in the last *Next call */
1517 2609 : if (!dlist_is_empty(&state->old_change))
1518 : {
1519 : ReorderBufferChange *change;
1520 :
1521 19 : change = dlist_container(ReorderBufferChange, node,
1522 : dlist_pop_head_node(&state->old_change));
1523 19 : ReorderBufferFreeChange(rb, change, true);
1524 : Assert(dlist_is_empty(&state->old_change));
1525 : }
1526 :
1527 2609 : binaryheap_free(state->heap);
1528 2609 : pfree(state);
1529 2609 : }
1530 :
1531 : /*
1532 : * Cleanup the contents of a transaction, usually after the transaction
1533 : * committed or aborted.
1534 : */
1535 : static void
1536 4778 : ReorderBufferCleanupTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
1537 : {
1538 : bool found;
1539 : dlist_mutable_iter iter;
1540 4778 : Size mem_freed = 0;
1541 :
1542 : /* cleanup subtransactions & their changes */
1543 4966 : dlist_foreach_modify(iter, &txn->subtxns)
1544 : {
1545 : ReorderBufferTXN *subtxn;
1546 :
1547 188 : subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1548 :
1549 : /*
1550 : * Subtransactions are always associated to the toplevel TXN, even if
1551 : * they originally were happening inside another subtxn, so we won't
1552 : * ever recurse more than one level deep here.
1553 : */
1554 : Assert(rbtxn_is_known_subxact(subtxn));
1555 : Assert(subtxn->nsubtxns == 0);
1556 :
1557 188 : ReorderBufferCleanupTXN(rb, subtxn);
1558 : }
1559 :
1560 : /* cleanup changes in the txn */
1561 102115 : dlist_foreach_modify(iter, &txn->changes)
1562 : {
1563 : ReorderBufferChange *change;
1564 :
1565 97337 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1566 :
1567 : /* Check we're not mixing changes from different transactions. */
1568 : Assert(change->txn == txn);
1569 :
1570 : /*
1571 : * Instead of updating the memory counter for individual changes, we
1572 : * sum up the size of memory to free so we can update the memory
1573 : * counter all together below. This saves costs of maintaining the
1574 : * max-heap.
1575 : */
1576 97337 : mem_freed += ReorderBufferChangeSize(change);
1577 :
1578 97337 : ReorderBufferFreeChange(rb, change, false);
1579 : }
1580 :
1581 : /* Update the memory counter */
1582 4778 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, mem_freed);
1583 :
1584 : /*
1585 : * Cleanup the tuplecids we stored for decoding catalog snapshot access.
1586 : * They are always stored in the toplevel transaction.
1587 : */
1588 31911 : dlist_foreach_modify(iter, &txn->tuplecids)
1589 : {
1590 : ReorderBufferChange *change;
1591 :
1592 27133 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1593 :
1594 : /* Check we're not mixing changes from different transactions. */
1595 : Assert(change->txn == txn);
1596 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1597 :
1598 27133 : ReorderBufferFreeChange(rb, change, true);
1599 : }
1600 :
1601 : /*
1602 : * Cleanup the base snapshot, if set.
1603 : */
1604 4778 : if (txn->base_snapshot != NULL)
1605 : {
1606 4092 : SnapBuildSnapDecRefcount(txn->base_snapshot);
1607 4092 : dlist_delete(&txn->base_snapshot_node);
1608 : }
1609 :
1610 : /*
1611 : * Cleanup the snapshot for the last streamed run.
1612 : */
1613 4778 : if (txn->snapshot_now != NULL)
1614 : {
1615 : Assert(rbtxn_is_streamed(txn));
1616 69 : ReorderBufferFreeSnap(rb, txn->snapshot_now);
1617 : }
1618 :
1619 : /*
1620 : * Remove TXN from its containing lists.
1621 : *
1622 : * Note: if txn is known as subxact, we are deleting the TXN from its
1623 : * parent's list of known subxacts; this leaves the parent's nsubxacts
1624 : * count too high, but we don't care. Otherwise, we are deleting the TXN
1625 : * from the LSN-ordered list of toplevel TXNs. We remove the TXN from the
1626 : * list of catalog modifying transactions as well.
1627 : */
1628 4778 : dlist_delete(&txn->node);
1629 4778 : if (rbtxn_has_catalog_changes(txn))
1630 1643 : dclist_delete_from(&rb->catchange_txns, &txn->catchange_node);
1631 :
1632 : /* now remove reference from buffer */
1633 4778 : hash_search(rb->by_txn, &txn->xid, HASH_REMOVE, &found);
1634 : Assert(found);
1635 :
1636 : /* remove entries spilled to disk */
1637 4778 : if (rbtxn_is_serialized(txn))
1638 338 : ReorderBufferRestoreCleanup(rb, txn);
1639 :
1640 : /* deallocate */
1641 4778 : ReorderBufferFreeTXN(rb, txn);
1642 4778 : }
1643 :
1644 : /*
1645 : * Discard changes from a transaction (and subtransactions), either after
1646 : * streaming, decoding them at PREPARE, or detecting the transaction abort.
1647 : * Keep the remaining info - transactions, tuplecids, invalidations and
1648 : * snapshots.
1649 : *
1650 : * We additionally remove tuplecids after decoding the transaction at prepare
1651 : * time as we only need to perform invalidation at rollback or commit prepared.
1652 : *
1653 : * 'txn_prepared' indicates that we have decoded the transaction at prepare
1654 : * time.
1655 : */
1656 : static void
1657 1076 : ReorderBufferTruncateTXN(ReorderBuffer *rb, ReorderBufferTXN *txn, bool txn_prepared)
1658 : {
1659 : dlist_mutable_iter iter;
1660 1076 : Size mem_freed = 0;
1661 :
1662 : /* cleanup subtransactions & their changes */
1663 1373 : dlist_foreach_modify(iter, &txn->subtxns)
1664 : {
1665 : ReorderBufferTXN *subtxn;
1666 :
1667 297 : subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1668 :
1669 : /*
1670 : * Subtransactions are always associated to the toplevel TXN, even if
1671 : * they originally were happening inside another subtxn, so we won't
1672 : * ever recurse more than one level deep here.
1673 : */
1674 : Assert(rbtxn_is_known_subxact(subtxn));
1675 : Assert(subtxn->nsubtxns == 0);
1676 :
1677 297 : ReorderBufferMaybeMarkTXNStreamed(rb, subtxn);
1678 297 : ReorderBufferTruncateTXN(rb, subtxn, txn_prepared);
1679 : }
1680 :
1681 : /* cleanup changes in the txn */
1682 159184 : dlist_foreach_modify(iter, &txn->changes)
1683 : {
1684 : ReorderBufferChange *change;
1685 :
1686 158108 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1687 :
1688 : /* Check we're not mixing changes from different transactions. */
1689 : Assert(change->txn == txn);
1690 :
1691 : /* remove the change from its containing list */
1692 158108 : dlist_delete(&change->node);
1693 :
1694 : /*
1695 : * Instead of updating the memory counter for individual changes, we
1696 : * sum up the size of memory to free so we can update the memory
1697 : * counter all together below. This saves costs of maintaining the
1698 : * max-heap.
1699 : */
1700 158108 : mem_freed += ReorderBufferChangeSize(change);
1701 :
1702 158108 : ReorderBufferFreeChange(rb, change, false);
1703 : }
1704 :
1705 : /* Update the memory counter */
1706 1076 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, mem_freed);
1707 :
1708 1076 : if (txn_prepared)
1709 : {
1710 : /*
1711 : * If this is a prepared txn, cleanup the tuplecids we stored for
1712 : * decoding catalog snapshot access. They are always stored in the
1713 : * toplevel transaction.
1714 : */
1715 191 : dlist_foreach_modify(iter, &txn->tuplecids)
1716 : {
1717 : ReorderBufferChange *change;
1718 :
1719 123 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1720 :
1721 : /* Check we're not mixing changes from different transactions. */
1722 : Assert(change->txn == txn);
1723 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1724 :
1725 : /* Remove the change from its containing list. */
1726 123 : dlist_delete(&change->node);
1727 :
1728 123 : ReorderBufferFreeChange(rb, change, true);
1729 : }
1730 : }
1731 :
1732 : /*
1733 : * Destroy the (relfilelocator, ctid) hashtable, so that we don't leak any
1734 : * memory. We could also keep the hash table and update it with new ctid
1735 : * values, but this seems simpler and good enough for now.
1736 : */
1737 1076 : if (txn->tuplecid_hash != NULL)
1738 : {
1739 51 : hash_destroy(txn->tuplecid_hash);
1740 51 : txn->tuplecid_hash = NULL;
1741 : }
1742 :
1743 : /* If this txn is serialized then clean the disk space. */
1744 1076 : if (rbtxn_is_serialized(txn))
1745 : {
1746 9 : ReorderBufferRestoreCleanup(rb, txn);
1747 9 : txn->txn_flags &= ~RBTXN_IS_SERIALIZED;
1748 :
1749 : /*
1750 : * We set this flag to indicate if the transaction is ever serialized.
1751 : * We need this to accurately update the stats as otherwise the same
1752 : * transaction can be counted as serialized multiple times.
1753 : */
1754 9 : txn->txn_flags |= RBTXN_IS_SERIALIZED_CLEAR;
1755 : }
1756 :
1757 : /* also reset the number of entries in the transaction */
1758 1076 : txn->nentries_mem = 0;
1759 1076 : txn->nentries = 0;
1760 1076 : }
1761 :
1762 : /*
1763 : * Check the transaction status by CLOG lookup and discard all changes if
1764 : * the transaction is aborted. The transaction status is cached in
1765 : * txn->txn_flags so we can skip future changes and avoid CLOG lookups on the
1766 : * next call.
1767 : *
1768 : * Return true if the transaction is aborted, otherwise return false.
1769 : *
1770 : * When the 'debug_logical_replication_streaming' is set to "immediate", we
1771 : * don't check the transaction status, meaning the caller will always process
1772 : * this transaction.
1773 : */
1774 : static bool
1775 4646 : ReorderBufferCheckAndTruncateAbortedTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
1776 : {
1777 : /* Quick return for regression tests */
1778 4646 : if (unlikely(debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_IMMEDIATE))
1779 1208 : return false;
1780 :
1781 : /*
1782 : * Quick return if the transaction status is already known.
1783 : */
1784 :
1785 3438 : if (rbtxn_is_committed(txn))
1786 2985 : return false;
1787 453 : if (rbtxn_is_aborted(txn))
1788 : {
1789 : /* Already-aborted transactions should not have any changes */
1790 : Assert(txn->size == 0);
1791 :
1792 0 : return true;
1793 : }
1794 :
1795 : /* Otherwise, check the transaction status using CLOG lookup */
1796 :
1797 453 : if (TransactionIdIsInProgress(txn->xid))
1798 220 : return false;
1799 :
1800 233 : if (TransactionIdDidCommit(txn->xid))
1801 : {
1802 : /*
1803 : * Remember the transaction is committed so that we can skip CLOG
1804 : * check next time, avoiding the pressure on CLOG lookup.
1805 : */
1806 : Assert(!rbtxn_is_aborted(txn));
1807 224 : txn->txn_flags |= RBTXN_IS_COMMITTED;
1808 224 : return false;
1809 : }
1810 :
1811 : /*
1812 : * The transaction aborted. We discard both the changes collected so far
1813 : * and the toast reconstruction data. The full cleanup will happen as part
1814 : * of decoding ABORT record of this transaction.
1815 : */
1816 9 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
1817 9 : ReorderBufferToastReset(rb, txn);
1818 :
1819 : /* All changes should be discarded */
1820 : Assert(txn->size == 0);
1821 :
1822 : /*
1823 : * Mark the transaction as aborted so we can ignore future changes of this
1824 : * transaction.
1825 : */
1826 : Assert(!rbtxn_is_committed(txn));
1827 9 : txn->txn_flags |= RBTXN_IS_ABORTED;
1828 :
1829 9 : return true;
1830 : }
1831 :
1832 : /*
1833 : * Build a hash with a (relfilelocator, ctid) -> (cmin, cmax) mapping for use by
1834 : * HeapTupleSatisfiesHistoricMVCC.
1835 : */
1836 : static void
1837 2610 : ReorderBufferBuildTupleCidHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
1838 : {
1839 : dlist_iter iter;
1840 : HASHCTL hash_ctl;
1841 :
1842 2610 : if (!rbtxn_has_catalog_changes(txn) || dlist_is_empty(&txn->tuplecids))
1843 1620 : return;
1844 :
1845 990 : hash_ctl.keysize = sizeof(ReorderBufferTupleCidKey);
1846 990 : hash_ctl.entrysize = sizeof(ReorderBufferTupleCidEnt);
1847 990 : hash_ctl.hcxt = rb->context;
1848 :
1849 : /*
1850 : * create the hash with the exact number of to-be-stored tuplecids from
1851 : * the start
1852 : */
1853 990 : txn->tuplecid_hash =
1854 990 : hash_create("ReorderBufferTupleCid", txn->ntuplecids, &hash_ctl,
1855 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1856 :
1857 14540 : dlist_foreach(iter, &txn->tuplecids)
1858 : {
1859 : ReorderBufferTupleCidKey key;
1860 : ReorderBufferTupleCidEnt *ent;
1861 : bool found;
1862 : ReorderBufferChange *change;
1863 :
1864 13550 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1865 :
1866 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1867 :
1868 : /* be careful about padding */
1869 13550 : memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
1870 :
1871 13550 : key.rlocator = change->data.tuplecid.locator;
1872 :
1873 13550 : ItemPointerCopy(&change->data.tuplecid.tid,
1874 : &key.tid);
1875 :
1876 : ent = (ReorderBufferTupleCidEnt *)
1877 13550 : hash_search(txn->tuplecid_hash, &key, HASH_ENTER, &found);
1878 13550 : if (!found)
1879 : {
1880 11907 : ent->cmin = change->data.tuplecid.cmin;
1881 11907 : ent->cmax = change->data.tuplecid.cmax;
1882 11907 : ent->combocid = change->data.tuplecid.combocid;
1883 : }
1884 : else
1885 : {
1886 : /*
1887 : * Maybe we already saw this tuple before in this transaction, but
1888 : * if so it must have the same cmin.
1889 : */
1890 : Assert(ent->cmin == change->data.tuplecid.cmin);
1891 :
1892 : /*
1893 : * cmax may be initially invalid, but once set it can only grow,
1894 : * and never become invalid again.
1895 : */
1896 : Assert((ent->cmax == InvalidCommandId) ||
1897 : ((change->data.tuplecid.cmax != InvalidCommandId) &&
1898 : (change->data.tuplecid.cmax > ent->cmax)));
1899 1643 : ent->cmax = change->data.tuplecid.cmax;
1900 : }
1901 : }
1902 : }
1903 :
1904 : /*
1905 : * Copy a provided snapshot so we can modify it privately. This is needed so
1906 : * that catalog modifying transactions can look into intermediate catalog
1907 : * states.
1908 : */
1909 : static Snapshot
1910 2609 : ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
1911 : ReorderBufferTXN *txn, CommandId cid)
1912 : {
1913 : Snapshot snap;
1914 : dlist_iter iter;
1915 2609 : int i = 0;
1916 : Size size;
1917 :
1918 2609 : size = sizeof(SnapshotData) +
1919 2609 : sizeof(TransactionId) * orig_snap->xcnt +
1920 2609 : sizeof(TransactionId) * (txn->nsubtxns + 1);
1921 :
1922 2609 : snap = MemoryContextAllocZero(rb->context, size);
1923 2609 : memcpy(snap, orig_snap, sizeof(SnapshotData));
1924 :
1925 2609 : snap->copied = true;
1926 2609 : snap->active_count = 1; /* mark as active so nobody frees it */
1927 2609 : snap->regd_count = 0;
1928 2609 : snap->xip = (TransactionId *) (snap + 1);
1929 :
1930 2609 : memcpy(snap->xip, orig_snap->xip, sizeof(TransactionId) * snap->xcnt);
1931 :
1932 : /*
1933 : * snap->subxip contains all txids that belong to our transaction which we
1934 : * need to check via cmin/cmax. That's why we store the toplevel
1935 : * transaction in there as well.
1936 : */
1937 2609 : snap->subxip = snap->xip + snap->xcnt;
1938 2609 : snap->subxip[i++] = txn->xid;
1939 :
1940 : /*
1941 : * txn->nsubtxns isn't decreased when subtransactions abort, so count
1942 : * manually. Since it's an upper boundary it is safe to use it for the
1943 : * allocation above.
1944 : */
1945 2609 : snap->subxcnt = 1;
1946 :
1947 2918 : dlist_foreach(iter, &txn->subtxns)
1948 : {
1949 : ReorderBufferTXN *sub_txn;
1950 :
1951 309 : sub_txn = dlist_container(ReorderBufferTXN, node, iter.cur);
1952 309 : snap->subxip[i++] = sub_txn->xid;
1953 309 : snap->subxcnt++;
1954 : }
1955 :
1956 : /* sort so we can bsearch() later */
1957 2609 : qsort(snap->subxip, snap->subxcnt, sizeof(TransactionId), xidComparator);
1958 :
1959 : /* store the specified current CommandId */
1960 2609 : snap->curcid = cid;
1961 :
1962 2609 : return snap;
1963 : }
1964 :
1965 : /*
1966 : * Free a previously ReorderBufferCopySnap'ed snapshot
1967 : */
1968 : static void
1969 4216 : ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap)
1970 : {
1971 4216 : if (snap->copied)
1972 2605 : pfree(snap);
1973 : else
1974 1611 : SnapBuildSnapDecRefcount(snap);
1975 4216 : }
1976 :
1977 : /*
1978 : * If the transaction was (partially) streamed, we need to prepare or commit
1979 : * it in a 'streamed' way. That is, we first stream the remaining part of the
1980 : * transaction, and then invoke stream_prepare or stream_commit message as per
1981 : * the case.
1982 : */
1983 : static void
1984 69 : ReorderBufferStreamCommit(ReorderBuffer *rb, ReorderBufferTXN *txn)
1985 : {
1986 : /* we should only call this for previously streamed transactions */
1987 : Assert(rbtxn_is_streamed(txn));
1988 :
1989 69 : ReorderBufferStreamTXN(rb, txn);
1990 :
1991 69 : if (rbtxn_is_prepared(txn))
1992 : {
1993 : /*
1994 : * Note, we send stream prepare even if a concurrent abort is
1995 : * detected. See DecodePrepare for more information.
1996 : */
1997 : Assert(!rbtxn_sent_prepare(txn));
1998 18 : rb->stream_prepare(rb, txn, txn->final_lsn);
1999 18 : txn->txn_flags |= RBTXN_SENT_PREPARE;
2000 :
2001 : /*
2002 : * This is a PREPARED transaction, part of a two-phase commit. The
2003 : * full cleanup will happen as part of the COMMIT PREPAREDs, so now
2004 : * just truncate txn by removing changes and tuplecids.
2005 : */
2006 18 : ReorderBufferTruncateTXN(rb, txn, true);
2007 : /* Reset the CheckXidAlive */
2008 18 : CheckXidAlive = InvalidTransactionId;
2009 : }
2010 : else
2011 : {
2012 51 : rb->stream_commit(rb, txn, txn->final_lsn);
2013 51 : ReorderBufferCleanupTXN(rb, txn);
2014 : }
2015 69 : }
2016 :
2017 : /*
2018 : * Set xid to detect concurrent aborts.
2019 : *
2020 : * While streaming an in-progress transaction or decoding a prepared
2021 : * transaction there is a possibility that the (sub)transaction might get
2022 : * aborted concurrently. In such case if the (sub)transaction has catalog
2023 : * update then we might decode the tuple using wrong catalog version. For
2024 : * example, suppose there is one catalog tuple with (xmin: 500, xmax: 0). Now,
2025 : * the transaction 501 updates the catalog tuple and after that we will have
2026 : * two tuples (xmin: 500, xmax: 501) and (xmin: 501, xmax: 0). Now, if 501 is
2027 : * aborted and some other transaction say 502 updates the same catalog tuple
2028 : * then the first tuple will be changed to (xmin: 500, xmax: 502). So, the
2029 : * problem is that when we try to decode the tuple inserted/updated in 501
2030 : * after the catalog update, we will see the catalog tuple with (xmin: 500,
2031 : * xmax: 502) as visible because it will consider that the tuple is deleted by
2032 : * xid 502 which is not visible to our snapshot. And when we will try to
2033 : * decode with that catalog tuple, it can lead to a wrong result or a crash.
2034 : * So, it is necessary to detect concurrent aborts to allow streaming of
2035 : * in-progress transactions or decoding of prepared transactions.
2036 : *
2037 : * For detecting the concurrent abort we set CheckXidAlive to the current
2038 : * (sub)transaction's xid for which this change belongs to. And, during
2039 : * catalog scan we can check the status of the xid and if it is aborted we will
2040 : * report a specific error so that we can stop streaming current transaction
2041 : * and discard the already streamed changes on such an error. We might have
2042 : * already streamed some of the changes for the aborted (sub)transaction, but
2043 : * that is fine because when we decode the abort we will stream abort message
2044 : * to truncate the changes in the subscriber. Similarly, for prepared
2045 : * transactions, we stop decoding if concurrent abort is detected and then
2046 : * rollback the changes when rollback prepared is encountered. See
2047 : * DecodePrepare.
2048 : */
2049 : static inline void
2050 177877 : SetupCheckXidLive(TransactionId xid)
2051 : {
2052 : /*
2053 : * If the input transaction id is already set as a CheckXidAlive then
2054 : * nothing to do.
2055 : */
2056 177877 : if (TransactionIdEquals(CheckXidAlive, xid))
2057 89621 : return;
2058 :
2059 : /*
2060 : * setup CheckXidAlive if it's not committed yet. We don't check if the
2061 : * xid is aborted. That will happen during catalog access.
2062 : */
2063 88256 : if (!TransactionIdDidCommit(xid))
2064 416 : CheckXidAlive = xid;
2065 : else
2066 87840 : CheckXidAlive = InvalidTransactionId;
2067 : }
2068 :
2069 : /*
2070 : * Helper function for ReorderBufferProcessTXN for applying change.
2071 : */
2072 : static inline void
2073 355505 : ReorderBufferApplyChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
2074 : Relation relation, ReorderBufferChange *change,
2075 : bool streaming)
2076 : {
2077 355505 : if (streaming)
2078 176009 : rb->stream_change(rb, txn, relation, change);
2079 : else
2080 179496 : rb->apply_change(rb, txn, relation, change);
2081 355502 : }
2082 :
2083 : /*
2084 : * Helper function for ReorderBufferProcessTXN for applying the truncate.
2085 : */
2086 : static inline void
2087 36 : ReorderBufferApplyTruncate(ReorderBuffer *rb, ReorderBufferTXN *txn,
2088 : int nrelations, Relation *relations,
2089 : ReorderBufferChange *change, bool streaming)
2090 : {
2091 36 : if (streaming)
2092 0 : rb->stream_truncate(rb, txn, nrelations, relations, change);
2093 : else
2094 36 : rb->apply_truncate(rb, txn, nrelations, relations, change);
2095 36 : }
2096 :
2097 : /*
2098 : * Helper function for ReorderBufferProcessTXN for applying the message.
2099 : */
2100 : static inline void
2101 11 : ReorderBufferApplyMessage(ReorderBuffer *rb, ReorderBufferTXN *txn,
2102 : ReorderBufferChange *change, bool streaming)
2103 : {
2104 11 : if (streaming)
2105 3 : rb->stream_message(rb, txn, change->lsn, true,
2106 3 : change->data.msg.prefix,
2107 : change->data.msg.message_size,
2108 3 : change->data.msg.message);
2109 : else
2110 8 : rb->message(rb, txn, change->lsn, true,
2111 8 : change->data.msg.prefix,
2112 : change->data.msg.message_size,
2113 8 : change->data.msg.message);
2114 11 : }
2115 :
2116 : /*
2117 : * Function to store the command id and snapshot at the end of the current
2118 : * stream so that we can reuse the same while sending the next stream.
2119 : */
2120 : static inline void
2121 722 : ReorderBufferSaveTXNSnapshot(ReorderBuffer *rb, ReorderBufferTXN *txn,
2122 : Snapshot snapshot_now, CommandId command_id)
2123 : {
2124 722 : txn->command_id = command_id;
2125 :
2126 : /* Avoid copying if it's already copied. */
2127 722 : if (snapshot_now->copied)
2128 722 : txn->snapshot_now = snapshot_now;
2129 : else
2130 0 : txn->snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
2131 : txn, command_id);
2132 722 : }
2133 :
2134 : /*
2135 : * Mark the given transaction as streamed if it's a top-level transaction
2136 : * or has changes.
2137 : */
2138 : static void
2139 1019 : ReorderBufferMaybeMarkTXNStreamed(ReorderBuffer *rb, ReorderBufferTXN *txn)
2140 : {
2141 : /*
2142 : * The top-level transaction, is marked as streamed always, even if it
2143 : * does not contain any changes (that is, when all the changes are in
2144 : * subtransactions).
2145 : *
2146 : * For subtransactions, we only mark them as streamed when there are
2147 : * changes in them.
2148 : *
2149 : * We do it this way because of aborts - we don't want to send aborts for
2150 : * XIDs the downstream is not aware of. And of course, it always knows
2151 : * about the top-level xact (we send the XID in all messages), but we
2152 : * never stream XIDs of empty subxacts.
2153 : */
2154 1019 : if (rbtxn_is_toptxn(txn) || (txn->nentries_mem != 0))
2155 857 : txn->txn_flags |= RBTXN_IS_STREAMED;
2156 1019 : }
2157 :
2158 : /*
2159 : * Helper function for ReorderBufferProcessTXN to handle the concurrent
2160 : * abort of the streaming transaction. This resets the TXN such that it
2161 : * can be used to stream the remaining data of transaction being processed.
2162 : * This can happen when the subtransaction is aborted and we still want to
2163 : * continue processing the main or other subtransactions data.
2164 : */
2165 : static void
2166 8 : ReorderBufferResetTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
2167 : Snapshot snapshot_now,
2168 : CommandId command_id,
2169 : XLogRecPtr last_lsn)
2170 : {
2171 : /* Discard the changes that we just streamed */
2172 8 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
2173 :
2174 : /* Free all resources allocated for toast reconstruction */
2175 8 : ReorderBufferToastReset(rb, txn);
2176 :
2177 : /*
2178 : * For the streaming case, stop the stream and remember the command ID and
2179 : * snapshot for the streaming run.
2180 : */
2181 8 : if (rbtxn_is_streamed(txn))
2182 : {
2183 8 : rb->stream_stop(rb, txn, last_lsn);
2184 8 : ReorderBufferSaveTXNSnapshot(rb, txn, snapshot_now, command_id);
2185 : }
2186 :
2187 : /* All changes must be deallocated */
2188 : Assert(txn->size == 0);
2189 8 : }
2190 :
2191 : /*
2192 : * Helper function for ReorderBufferReplay and ReorderBufferStreamTXN.
2193 : *
2194 : * Send data of a transaction (and its subtransactions) to the
2195 : * output plugin. We iterate over the top and subtransactions (using a k-way
2196 : * merge) and replay the changes in lsn order.
2197 : *
2198 : * If streaming is true then data will be sent using stream API.
2199 : *
2200 : * Note: "volatile" markers on some parameters are to avoid trouble with
2201 : * PG_TRY inside the function.
2202 : */
2203 : static void
2204 2610 : ReorderBufferProcessTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
2205 : XLogRecPtr commit_lsn,
2206 : volatile Snapshot snapshot_now,
2207 : volatile CommandId command_id,
2208 : bool streaming)
2209 : {
2210 : bool using_subtxn;
2211 2610 : MemoryContext ccxt = CurrentMemoryContext;
2212 2610 : ResourceOwner cowner = CurrentResourceOwner;
2213 2610 : ReorderBufferIterTXNState *volatile iterstate = NULL;
2214 2610 : volatile XLogRecPtr prev_lsn = InvalidXLogRecPtr;
2215 2610 : ReorderBufferChange *volatile specinsert = NULL;
2216 2610 : volatile bool stream_started = false;
2217 2610 : ReorderBufferTXN *volatile curtxn = NULL;
2218 :
2219 : /* build data to be able to lookup the CommandIds of catalog tuples */
2220 2610 : ReorderBufferBuildTupleCidHash(rb, txn);
2221 :
2222 : /* setup the initial snapshot */
2223 2610 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2224 :
2225 : /*
2226 : * Decoding needs access to syscaches et al., which in turn use
2227 : * heavyweight locks and such. Thus we need to have enough state around to
2228 : * keep track of those. The easiest way is to simply use a transaction
2229 : * internally. That also allows us to easily enforce that nothing writes
2230 : * to the database by checking for xid assignments.
2231 : *
2232 : * When we're called via the SQL SRF there's already a transaction
2233 : * started, so start an explicit subtransaction there.
2234 : */
2235 2610 : using_subtxn = IsTransactionOrTransactionBlock();
2236 :
2237 2610 : PG_TRY();
2238 : {
2239 : ReorderBufferChange *change;
2240 2610 : int changes_count = 0; /* used to accumulate the number of
2241 : * changes */
2242 :
2243 2610 : if (using_subtxn)
2244 514 : BeginInternalSubTransaction(streaming ? "stream" : "replay");
2245 : else
2246 2096 : StartTransactionCommand();
2247 :
2248 : /*
2249 : * We only need to send begin/begin-prepare for non-streamed
2250 : * transactions.
2251 : */
2252 2610 : if (!streaming)
2253 : {
2254 1888 : if (rbtxn_is_prepared(txn))
2255 30 : rb->begin_prepare(rb, txn);
2256 : else
2257 1858 : rb->begin(rb, txn);
2258 : }
2259 :
2260 2610 : ReorderBufferIterTXNInit(rb, txn, &iterstate);
2261 385810 : while ((change = ReorderBufferIterTXNNext(rb, iterstate)) != NULL)
2262 : {
2263 380601 : Relation relation = NULL;
2264 : Oid reloid;
2265 :
2266 380601 : CHECK_FOR_INTERRUPTS();
2267 :
2268 : /*
2269 : * We can't call start stream callback before processing first
2270 : * change.
2271 : */
2272 380601 : if (!XLogRecPtrIsValid(prev_lsn))
2273 : {
2274 2564 : if (streaming)
2275 : {
2276 681 : txn->origin_id = change->origin_id;
2277 681 : rb->stream_start(rb, txn, change->lsn);
2278 681 : stream_started = true;
2279 : }
2280 : }
2281 :
2282 : /*
2283 : * Enforce correct ordering of changes, merged from multiple
2284 : * subtransactions. The changes may have the same LSN due to
2285 : * MULTI_INSERT xlog records.
2286 : */
2287 : Assert(!XLogRecPtrIsValid(prev_lsn) || prev_lsn <= change->lsn);
2288 :
2289 380601 : prev_lsn = change->lsn;
2290 :
2291 : /*
2292 : * Set the current xid to detect concurrent aborts. This is
2293 : * required for the cases when we decode the changes before the
2294 : * COMMIT record is processed.
2295 : */
2296 380601 : if (streaming || rbtxn_is_prepared(change->txn))
2297 : {
2298 177877 : curtxn = change->txn;
2299 177877 : SetupCheckXidLive(curtxn->xid);
2300 : }
2301 :
2302 380601 : switch (change->action)
2303 : {
2304 1783 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
2305 :
2306 : /*
2307 : * Confirmation for speculative insertion arrived. Simply
2308 : * use as a normal record. It'll be cleaned up at the end
2309 : * of INSERT processing.
2310 : */
2311 1783 : if (specinsert == NULL)
2312 0 : elog(ERROR, "invalid ordering of speculative insertion changes");
2313 : Assert(specinsert->data.tp.oldtuple == NULL);
2314 1783 : change = specinsert;
2315 1783 : change->action = REORDER_BUFFER_CHANGE_INSERT;
2316 :
2317 : /* intentionally fall through */
2318 : pg_fallthrough;
2319 362283 : case REORDER_BUFFER_CHANGE_INSERT:
2320 : case REORDER_BUFFER_CHANGE_UPDATE:
2321 : case REORDER_BUFFER_CHANGE_DELETE:
2322 : Assert(snapshot_now);
2323 :
2324 362283 : reloid = RelidByRelfilenumber(change->data.tp.rlocator.spcOid,
2325 : change->data.tp.rlocator.relNumber);
2326 :
2327 : /*
2328 : * Mapped catalog tuple without data, emitted while
2329 : * catalog table was in the process of being rewritten. We
2330 : * can fail to look up the relfilenumber, because the
2331 : * relmapper has no "historic" view, in contrast to the
2332 : * normal catalog during decoding. Thus repeated rewrites
2333 : * can cause a lookup failure. That's OK because we do not
2334 : * decode catalog changes anyway. Normally such tuples
2335 : * would be skipped over below, but we can't identify
2336 : * whether the table should be logically logged without
2337 : * mapping the relfilenumber to the oid.
2338 : */
2339 362275 : if (reloid == InvalidOid &&
2340 83 : change->data.tp.newtuple == NULL &&
2341 83 : change->data.tp.oldtuple == NULL)
2342 83 : goto change_done;
2343 362192 : else if (reloid == InvalidOid)
2344 0 : elog(ERROR, "could not map filenumber \"%s\" to relation OID",
2345 : relpathperm(change->data.tp.rlocator,
2346 : MAIN_FORKNUM).str);
2347 :
2348 362192 : relation = RelationIdGetRelation(reloid);
2349 :
2350 362192 : if (!RelationIsValid(relation))
2351 0 : elog(ERROR, "could not open relation with OID %u (for filenumber \"%s\")",
2352 : reloid,
2353 : relpathperm(change->data.tp.rlocator,
2354 : MAIN_FORKNUM).str);
2355 :
2356 362192 : if (!RelationIsLogicallyLogged(relation))
2357 4564 : goto change_done;
2358 :
2359 : /*
2360 : * Ignore temporary heaps created during DDL unless the
2361 : * plugin has asked for them.
2362 : */
2363 357628 : if (relation->rd_rel->relrewrite && !rb->output_rewrites)
2364 26 : goto change_done;
2365 :
2366 : /*
2367 : * For now ignore sequence changes entirely. Most of the
2368 : * time they don't log changes using records we
2369 : * understand, so it doesn't make sense to handle the few
2370 : * cases we do.
2371 : */
2372 357602 : if (relation->rd_rel->relkind == RELKIND_SEQUENCE)
2373 0 : goto change_done;
2374 :
2375 : /* user-triggered change */
2376 357602 : if (!IsToastRelation(relation))
2377 : {
2378 355505 : ReorderBufferToastReplace(rb, txn, relation, change);
2379 355505 : ReorderBufferApplyChange(rb, txn, relation, change,
2380 : streaming);
2381 :
2382 : /*
2383 : * Only clear reassembled toast chunks if we're sure
2384 : * they're not required anymore. The creator of the
2385 : * tuple tells us.
2386 : */
2387 355502 : if (change->data.tp.clear_toast_afterwards)
2388 355281 : ReorderBufferToastReset(rb, txn);
2389 : }
2390 : /* we're not interested in toast deletions */
2391 2097 : else if (change->action == REORDER_BUFFER_CHANGE_INSERT)
2392 : {
2393 : /*
2394 : * Need to reassemble the full toasted Datum in
2395 : * memory, to ensure the chunks don't get reused till
2396 : * we're done remove it from the list of this
2397 : * transaction's changes. Otherwise it will get
2398 : * freed/reused while restoring spooled data from
2399 : * disk.
2400 : */
2401 : Assert(change->data.tp.newtuple != NULL);
2402 :
2403 1847 : dlist_delete(&change->node);
2404 1847 : ReorderBufferToastAppendChunk(rb, txn, relation,
2405 : change);
2406 : }
2407 :
2408 250 : change_done:
2409 :
2410 : /*
2411 : * If speculative insertion was confirmed, the record
2412 : * isn't needed anymore.
2413 : */
2414 362272 : if (specinsert != NULL)
2415 : {
2416 1782 : ReorderBufferFreeChange(rb, specinsert, true);
2417 1782 : specinsert = NULL;
2418 : }
2419 :
2420 362272 : if (RelationIsValid(relation))
2421 : {
2422 362189 : RelationClose(relation);
2423 362189 : relation = NULL;
2424 : }
2425 362272 : break;
2426 :
2427 1783 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
2428 :
2429 : /*
2430 : * Speculative insertions are dealt with by delaying the
2431 : * processing of the insert until the confirmation record
2432 : * arrives. For that we simply unlink the record from the
2433 : * chain, so it does not get freed/reused while restoring
2434 : * spooled data from disk.
2435 : *
2436 : * This is safe in the face of concurrent catalog changes
2437 : * because the relevant relation can't be changed between
2438 : * speculative insertion and confirmation due to
2439 : * CheckTableNotInUse() and locking.
2440 : */
2441 :
2442 : /* Previous speculative insertion must be aborted */
2443 : Assert(specinsert == NULL);
2444 :
2445 : /* and memorize the pending insertion */
2446 1783 : dlist_delete(&change->node);
2447 1783 : specinsert = change;
2448 1783 : break;
2449 :
2450 0 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
2451 :
2452 : /*
2453 : * Abort for speculative insertion arrived. So cleanup the
2454 : * specinsert tuple and toast hash.
2455 : *
2456 : * Note that we get the spec abort change for each toast
2457 : * entry but we need to perform the cleanup only the first
2458 : * time we get it for the main table.
2459 : */
2460 0 : if (specinsert != NULL)
2461 : {
2462 : /*
2463 : * We must clean the toast hash before processing a
2464 : * completely new tuple to avoid confusion about the
2465 : * previous tuple's toast chunks.
2466 : */
2467 : Assert(change->data.tp.clear_toast_afterwards);
2468 0 : ReorderBufferToastReset(rb, txn);
2469 :
2470 : /* We don't need this record anymore. */
2471 0 : ReorderBufferFreeChange(rb, specinsert, true);
2472 0 : specinsert = NULL;
2473 : }
2474 0 : break;
2475 :
2476 36 : case REORDER_BUFFER_CHANGE_TRUNCATE:
2477 : {
2478 : int i;
2479 36 : int nrelids = change->data.truncate.nrelids;
2480 36 : int nrelations = 0;
2481 : Relation *relations;
2482 :
2483 36 : relations = palloc0_array(Relation, nrelids);
2484 92 : for (i = 0; i < nrelids; i++)
2485 : {
2486 56 : Oid relid = change->data.truncate.relids[i];
2487 : Relation rel;
2488 :
2489 56 : rel = RelationIdGetRelation(relid);
2490 :
2491 56 : if (!RelationIsValid(rel))
2492 0 : elog(ERROR, "could not open relation with OID %u", relid);
2493 :
2494 56 : if (!RelationIsLogicallyLogged(rel))
2495 0 : continue;
2496 :
2497 56 : relations[nrelations++] = rel;
2498 : }
2499 :
2500 : /* Apply the truncate. */
2501 36 : ReorderBufferApplyTruncate(rb, txn, nrelations,
2502 : relations, change,
2503 : streaming);
2504 :
2505 92 : for (i = 0; i < nrelations; i++)
2506 56 : RelationClose(relations[i]);
2507 :
2508 36 : break;
2509 : }
2510 :
2511 11 : case REORDER_BUFFER_CHANGE_MESSAGE:
2512 11 : ReorderBufferApplyMessage(rb, txn, change, streaming);
2513 11 : break;
2514 :
2515 2872 : case REORDER_BUFFER_CHANGE_INVALIDATION:
2516 : /* Execute the invalidation messages locally */
2517 2872 : ReorderBufferExecuteInvalidations(change->data.inval.ninvalidations,
2518 : change->data.inval.invalidations);
2519 2872 : break;
2520 :
2521 971 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
2522 : /* get rid of the old */
2523 971 : TeardownHistoricSnapshot(false);
2524 :
2525 971 : if (snapshot_now->copied)
2526 : {
2527 946 : ReorderBufferFreeSnap(rb, snapshot_now);
2528 946 : snapshot_now =
2529 946 : ReorderBufferCopySnap(rb, change->data.snapshot,
2530 : txn, command_id);
2531 : }
2532 :
2533 : /*
2534 : * Restored from disk, need to be careful not to double
2535 : * free. We could introduce refcounting for that, but for
2536 : * now this seems infrequent enough not to care.
2537 : */
2538 25 : else if (change->data.snapshot->copied)
2539 : {
2540 0 : snapshot_now =
2541 0 : ReorderBufferCopySnap(rb, change->data.snapshot,
2542 : txn, command_id);
2543 : }
2544 : else
2545 : {
2546 25 : snapshot_now = change->data.snapshot;
2547 : }
2548 :
2549 : /* and continue with the new one */
2550 971 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2551 971 : break;
2552 :
2553 12645 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
2554 : Assert(change->data.command_id != InvalidCommandId);
2555 :
2556 12645 : if (command_id < change->data.command_id)
2557 : {
2558 2521 : command_id = change->data.command_id;
2559 :
2560 2521 : if (!snapshot_now->copied)
2561 : {
2562 : /* we don't use the global one anymore */
2563 941 : snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
2564 : txn, command_id);
2565 : }
2566 :
2567 2521 : snapshot_now->curcid = command_id;
2568 :
2569 2521 : TeardownHistoricSnapshot(false);
2570 2521 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2571 : }
2572 :
2573 12645 : break;
2574 :
2575 0 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
2576 0 : elog(ERROR, "tuplecid value in changequeue");
2577 : break;
2578 : }
2579 :
2580 : /*
2581 : * It is possible that the data is not sent to downstream for a
2582 : * long time either because the output plugin filtered it or there
2583 : * is a DDL that generates a lot of data that is not processed by
2584 : * the plugin. So, in such cases, the downstream can timeout. To
2585 : * avoid that we try to send a keepalive message if required.
2586 : * Trying to send a keepalive message after every change has some
2587 : * overhead, but testing showed there is no noticeable overhead if
2588 : * we do it after every ~100 changes.
2589 : */
2590 : #define CHANGES_THRESHOLD 100
2591 :
2592 380590 : if (++changes_count >= CHANGES_THRESHOLD)
2593 : {
2594 3316 : rb->update_progress_txn(rb, txn, prev_lsn);
2595 3316 : changes_count = 0;
2596 : }
2597 : }
2598 :
2599 : /* speculative insertion record must be freed by now */
2600 : Assert(!specinsert);
2601 :
2602 : /* clean up the iterator */
2603 2599 : ReorderBufferIterTXNFinish(rb, iterstate);
2604 2599 : iterstate = NULL;
2605 :
2606 : /*
2607 : * Update total transaction count and total bytes processed by the
2608 : * transaction and its subtransactions. Ensure to not count the
2609 : * streamed transaction multiple times.
2610 : *
2611 : * Note that the statistics computation has to be done after
2612 : * ReorderBufferIterTXNFinish as it releases the serialized change
2613 : * which we have already accounted in ReorderBufferIterTXNNext.
2614 : */
2615 2599 : if (!rbtxn_is_streamed(txn))
2616 1956 : rb->totalTxns++;
2617 :
2618 2599 : rb->totalBytes += txn->total_size;
2619 :
2620 : /*
2621 : * Done with current changes, send the last message for this set of
2622 : * changes depending upon streaming mode.
2623 : */
2624 2599 : if (streaming)
2625 : {
2626 714 : if (stream_started)
2627 : {
2628 673 : rb->stream_stop(rb, txn, prev_lsn);
2629 673 : stream_started = false;
2630 : }
2631 : }
2632 : else
2633 : {
2634 : /*
2635 : * Call either PREPARE (for two-phase transactions) or COMMIT (for
2636 : * regular ones).
2637 : */
2638 1885 : if (rbtxn_is_prepared(txn))
2639 : {
2640 : Assert(!rbtxn_sent_prepare(txn));
2641 30 : rb->prepare(rb, txn, commit_lsn);
2642 30 : txn->txn_flags |= RBTXN_SENT_PREPARE;
2643 : }
2644 : else
2645 1855 : rb->commit(rb, txn, commit_lsn);
2646 : }
2647 :
2648 : /* this is just a sanity check against bad output plugin behaviour */
2649 2592 : if (GetCurrentTransactionIdIfAny() != InvalidTransactionId)
2650 0 : elog(ERROR, "output plugin used XID %u",
2651 : GetCurrentTransactionId());
2652 :
2653 : /*
2654 : * Remember the command ID and snapshot for the next set of changes in
2655 : * streaming mode.
2656 : */
2657 2592 : if (streaming)
2658 714 : ReorderBufferSaveTXNSnapshot(rb, txn, snapshot_now, command_id);
2659 1878 : else if (snapshot_now->copied)
2660 941 : ReorderBufferFreeSnap(rb, snapshot_now);
2661 :
2662 : /* cleanup */
2663 2592 : TeardownHistoricSnapshot(false);
2664 :
2665 : /*
2666 : * Aborting the current (sub-)transaction as a whole has the right
2667 : * semantics. We want all locks acquired in here to be released, not
2668 : * reassigned to the parent and we do not want any database access
2669 : * have persistent effects.
2670 : */
2671 2592 : AbortCurrentTransaction();
2672 :
2673 : /* make sure there's no cache pollution */
2674 2592 : if (rbtxn_distr_inval_overflowed(txn))
2675 : {
2676 : Assert(txn->ninvalidations_distributed == 0);
2677 0 : InvalidateSystemCaches();
2678 : }
2679 : else
2680 : {
2681 2592 : ReorderBufferExecuteInvalidations(txn->ninvalidations, txn->invalidations);
2682 2592 : ReorderBufferExecuteInvalidations(txn->ninvalidations_distributed,
2683 : txn->invalidations_distributed);
2684 : }
2685 :
2686 2592 : if (using_subtxn)
2687 : {
2688 509 : RollbackAndReleaseCurrentSubTransaction();
2689 509 : MemoryContextSwitchTo(ccxt);
2690 509 : CurrentResourceOwner = cowner;
2691 : }
2692 :
2693 : /*
2694 : * We are here due to one of the four reasons: 1. Decoding an
2695 : * in-progress txn. 2. Decoding a prepared txn. 3. Decoding of a
2696 : * prepared txn that was (partially) streamed. 4. Decoding a committed
2697 : * txn.
2698 : *
2699 : * For 1, we allow truncation of txn data by removing the changes
2700 : * already streamed but still keeping other things like invalidations,
2701 : * snapshot, and tuplecids. For 2 and 3, we indicate
2702 : * ReorderBufferTruncateTXN to do more elaborate truncation of txn
2703 : * data as the entire transaction has been decoded except for commit.
2704 : * For 4, as the entire txn has been decoded, we can fully clean up
2705 : * the TXN reorder buffer.
2706 : */
2707 2592 : if (streaming || rbtxn_is_prepared(txn))
2708 : {
2709 744 : if (streaming)
2710 714 : ReorderBufferMaybeMarkTXNStreamed(rb, txn);
2711 :
2712 744 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
2713 : /* Reset the CheckXidAlive */
2714 744 : CheckXidAlive = InvalidTransactionId;
2715 : }
2716 : else
2717 1848 : ReorderBufferCleanupTXN(rb, txn);
2718 : }
2719 10 : PG_CATCH();
2720 : {
2721 10 : MemoryContext ecxt = MemoryContextSwitchTo(ccxt);
2722 10 : ErrorData *errdata = CopyErrorData();
2723 :
2724 : /* TODO: Encapsulate cleanup from the PG_TRY and PG_CATCH blocks */
2725 10 : if (iterstate)
2726 10 : ReorderBufferIterTXNFinish(rb, iterstate);
2727 :
2728 10 : TeardownHistoricSnapshot(true);
2729 :
2730 : /*
2731 : * Force cache invalidation to happen outside of a valid transaction
2732 : * to prevent catalog access as we just caught an error.
2733 : */
2734 10 : AbortCurrentTransaction();
2735 :
2736 : /* make sure there's no cache pollution */
2737 10 : if (rbtxn_distr_inval_overflowed(txn))
2738 : {
2739 : Assert(txn->ninvalidations_distributed == 0);
2740 0 : InvalidateSystemCaches();
2741 : }
2742 : else
2743 : {
2744 10 : ReorderBufferExecuteInvalidations(txn->ninvalidations, txn->invalidations);
2745 10 : ReorderBufferExecuteInvalidations(txn->ninvalidations_distributed,
2746 : txn->invalidations_distributed);
2747 : }
2748 :
2749 10 : if (using_subtxn)
2750 : {
2751 5 : RollbackAndReleaseCurrentSubTransaction();
2752 5 : MemoryContextSwitchTo(ccxt);
2753 5 : CurrentResourceOwner = cowner;
2754 : }
2755 :
2756 : /* Free the specinsert change before freeing the ReorderBufferTXN */
2757 10 : if (specinsert != NULL)
2758 : {
2759 1 : ReorderBufferFreeChange(rb, specinsert, true);
2760 1 : specinsert = NULL;
2761 : }
2762 :
2763 : /*
2764 : * The error code ERRCODE_TRANSACTION_ROLLBACK indicates a concurrent
2765 : * abort of the (sub)transaction we are streaming or preparing. We
2766 : * need to do the cleanup and return gracefully on this error, see
2767 : * SetupCheckXidLive.
2768 : *
2769 : * This error code can be thrown by one of the callbacks we call
2770 : * during decoding so we need to ensure that we return gracefully only
2771 : * when we are sending the data in streaming mode and the streaming is
2772 : * not finished yet or when we are sending the data out on a PREPARE
2773 : * during a two-phase commit.
2774 : */
2775 10 : if (errdata->sqlerrcode == ERRCODE_TRANSACTION_ROLLBACK &&
2776 8 : (stream_started || rbtxn_is_prepared(txn)))
2777 : {
2778 : /* curtxn must be set for streaming or prepared transactions */
2779 : Assert(curtxn);
2780 :
2781 : /* Cleanup the temporary error state. */
2782 8 : FlushErrorState();
2783 8 : FreeErrorData(errdata);
2784 8 : errdata = NULL;
2785 :
2786 : /* Remember the transaction is aborted. */
2787 : Assert(!rbtxn_is_committed(curtxn));
2788 8 : curtxn->txn_flags |= RBTXN_IS_ABORTED;
2789 :
2790 : /* Mark the transaction is streamed if appropriate */
2791 8 : if (stream_started)
2792 8 : ReorderBufferMaybeMarkTXNStreamed(rb, txn);
2793 :
2794 : /* Reset the TXN so that it is allowed to stream remaining data. */
2795 8 : ReorderBufferResetTXN(rb, txn, snapshot_now,
2796 : command_id, prev_lsn);
2797 : }
2798 : else
2799 : {
2800 2 : ReorderBufferCleanupTXN(rb, txn);
2801 2 : MemoryContextSwitchTo(ecxt);
2802 2 : PG_RE_THROW();
2803 : }
2804 : }
2805 2600 : PG_END_TRY();
2806 2600 : }
2807 :
2808 : /*
2809 : * Perform the replay of a transaction and its non-aborted subtransactions.
2810 : *
2811 : * Subtransactions previously have to be processed by
2812 : * ReorderBufferCommitChild(), even if previously assigned to the toplevel
2813 : * transaction with ReorderBufferAssignChild.
2814 : *
2815 : * This interface is called once a prepare or toplevel commit is read for both
2816 : * streamed as well as non-streamed transactions.
2817 : */
2818 : static void
2819 1960 : ReorderBufferReplay(ReorderBufferTXN *txn,
2820 : ReorderBuffer *rb, TransactionId xid,
2821 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2822 : TimestampTz commit_time,
2823 : ReplOriginId origin_id, XLogRecPtr origin_lsn)
2824 : {
2825 : Snapshot snapshot_now;
2826 1960 : CommandId command_id = FirstCommandId;
2827 :
2828 1960 : txn->final_lsn = commit_lsn;
2829 1960 : txn->end_lsn = end_lsn;
2830 1960 : txn->commit_time = commit_time;
2831 1960 : txn->origin_id = origin_id;
2832 1960 : txn->origin_lsn = origin_lsn;
2833 :
2834 : /*
2835 : * If the transaction was (partially) streamed, we need to commit it in a
2836 : * 'streamed' way. That is, we first stream the remaining part of the
2837 : * transaction, and then invoke stream_commit message.
2838 : *
2839 : * Called after everything (origin ID, LSN, ...) is stored in the
2840 : * transaction to avoid passing that information directly.
2841 : */
2842 1960 : if (rbtxn_is_streamed(txn))
2843 : {
2844 69 : ReorderBufferStreamCommit(rb, txn);
2845 69 : return;
2846 : }
2847 :
2848 : /*
2849 : * If this transaction has no snapshot, it didn't make any changes to the
2850 : * database, so there's nothing to decode. Note that
2851 : * ReorderBufferCommitChild will have transferred any snapshots from
2852 : * subtransactions if there were any.
2853 : */
2854 1891 : if (txn->base_snapshot == NULL)
2855 : {
2856 : Assert(txn->ninvalidations == 0);
2857 :
2858 : /*
2859 : * Removing this txn before a commit might result in the computation
2860 : * of an incorrect restart_lsn. See SnapBuildProcessRunningXacts.
2861 : */
2862 3 : if (!rbtxn_is_prepared(txn))
2863 3 : ReorderBufferCleanupTXN(rb, txn);
2864 3 : return;
2865 : }
2866 :
2867 1888 : snapshot_now = txn->base_snapshot;
2868 :
2869 : /* Process and send the changes to output plugin. */
2870 1888 : ReorderBufferProcessTXN(rb, txn, commit_lsn, snapshot_now,
2871 : command_id, false);
2872 : }
2873 :
2874 : /*
2875 : * Commit a transaction.
2876 : *
2877 : * See comments for ReorderBufferReplay().
2878 : */
2879 : void
2880 1953 : ReorderBufferCommit(ReorderBuffer *rb, TransactionId xid,
2881 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2882 : TimestampTz commit_time,
2883 : ReplOriginId origin_id, XLogRecPtr origin_lsn)
2884 : {
2885 : ReorderBufferTXN *txn;
2886 :
2887 1953 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2888 : false);
2889 :
2890 : /* unknown transaction, nothing to replay */
2891 1953 : if (txn == NULL)
2892 41 : return;
2893 :
2894 1912 : ReorderBufferReplay(txn, rb, xid, commit_lsn, end_lsn, commit_time,
2895 : origin_id, origin_lsn);
2896 : }
2897 :
2898 : /*
2899 : * Record the prepare information for a transaction. Also, mark the transaction
2900 : * as a prepared transaction.
2901 : */
2902 : bool
2903 182 : ReorderBufferRememberPrepareInfo(ReorderBuffer *rb, TransactionId xid,
2904 : XLogRecPtr prepare_lsn, XLogRecPtr end_lsn,
2905 : TimestampTz prepare_time,
2906 : ReplOriginId origin_id, XLogRecPtr origin_lsn)
2907 : {
2908 : ReorderBufferTXN *txn;
2909 :
2910 182 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr, false);
2911 :
2912 : /* unknown transaction, nothing to do */
2913 182 : if (txn == NULL)
2914 0 : return false;
2915 :
2916 : /*
2917 : * Remember the prepare information to be later used by commit prepared in
2918 : * case we skip doing prepare.
2919 : */
2920 182 : txn->final_lsn = prepare_lsn;
2921 182 : txn->end_lsn = end_lsn;
2922 182 : txn->prepare_time = prepare_time;
2923 182 : txn->origin_id = origin_id;
2924 182 : txn->origin_lsn = origin_lsn;
2925 :
2926 : /* Mark this transaction as a prepared transaction */
2927 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == 0);
2928 182 : txn->txn_flags |= RBTXN_IS_PREPARED;
2929 :
2930 182 : return true;
2931 : }
2932 :
2933 : /* Remember that we have skipped prepare */
2934 : void
2935 137 : ReorderBufferSkipPrepare(ReorderBuffer *rb, TransactionId xid)
2936 : {
2937 : ReorderBufferTXN *txn;
2938 :
2939 137 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr, false);
2940 :
2941 : /* unknown transaction, nothing to do */
2942 137 : if (txn == NULL)
2943 0 : return;
2944 :
2945 : /* txn must have been marked as a prepared transaction */
2946 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == RBTXN_IS_PREPARED);
2947 137 : txn->txn_flags |= RBTXN_SKIPPED_PREPARE;
2948 : }
2949 :
2950 : /*
2951 : * Prepare a two-phase transaction.
2952 : *
2953 : * See comments for ReorderBufferReplay().
2954 : */
2955 : void
2956 45 : ReorderBufferPrepare(ReorderBuffer *rb, TransactionId xid,
2957 : char *gid)
2958 : {
2959 : ReorderBufferTXN *txn;
2960 :
2961 45 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2962 : false);
2963 :
2964 : /* unknown transaction, nothing to replay */
2965 45 : if (txn == NULL)
2966 0 : return;
2967 :
2968 : /*
2969 : * txn must have been marked as a prepared transaction and must have
2970 : * neither been skipped nor sent a prepare. Also, the prepare info must
2971 : * have been updated in it by now.
2972 : */
2973 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == RBTXN_IS_PREPARED);
2974 : Assert(XLogRecPtrIsValid(txn->final_lsn));
2975 :
2976 45 : txn->gid = pstrdup(gid);
2977 :
2978 45 : ReorderBufferReplay(txn, rb, xid, txn->final_lsn, txn->end_lsn,
2979 45 : txn->prepare_time, txn->origin_id, txn->origin_lsn);
2980 :
2981 : /*
2982 : * Send a prepare if not already done so. This might occur if we have
2983 : * detected a concurrent abort while replaying the non-streaming
2984 : * transaction.
2985 : */
2986 45 : if (!rbtxn_sent_prepare(txn))
2987 : {
2988 0 : rb->prepare(rb, txn, txn->final_lsn);
2989 0 : txn->txn_flags |= RBTXN_SENT_PREPARE;
2990 : }
2991 : }
2992 :
2993 : /*
2994 : * This is used to handle COMMIT/ROLLBACK PREPARED.
2995 : */
2996 : void
2997 46 : ReorderBufferFinishPrepared(ReorderBuffer *rb, TransactionId xid,
2998 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2999 : XLogRecPtr two_phase_at,
3000 : TimestampTz commit_time, ReplOriginId origin_id,
3001 : XLogRecPtr origin_lsn, char *gid, bool is_commit)
3002 : {
3003 : ReorderBufferTXN *txn;
3004 : XLogRecPtr prepare_end_lsn;
3005 : TimestampTz prepare_time;
3006 :
3007 46 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, commit_lsn, false);
3008 :
3009 : /* unknown transaction, nothing to do */
3010 46 : if (txn == NULL)
3011 0 : return;
3012 :
3013 : /*
3014 : * By this time the txn has the prepare record information, remember it to
3015 : * be later used for rollback.
3016 : */
3017 46 : prepare_end_lsn = txn->end_lsn;
3018 46 : prepare_time = txn->prepare_time;
3019 :
3020 : /* add the gid in the txn */
3021 46 : txn->gid = pstrdup(gid);
3022 :
3023 : /*
3024 : * It is possible that this transaction is not decoded at prepare time
3025 : * either because by that time we didn't have a consistent snapshot, or
3026 : * two_phase was not enabled, or it was decoded earlier but we have
3027 : * restarted. We only need to send the prepare if it was not decoded
3028 : * earlier. We don't need to decode the xact for aborts if it is not done
3029 : * already.
3030 : */
3031 46 : if ((txn->final_lsn < two_phase_at) && is_commit)
3032 : {
3033 : /*
3034 : * txn must have been marked as a prepared transaction and skipped but
3035 : * not sent a prepare. Also, the prepare info must have been updated
3036 : * in txn even if we skip prepare.
3037 : */
3038 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) ==
3039 : (RBTXN_IS_PREPARED | RBTXN_SKIPPED_PREPARE));
3040 : Assert(XLogRecPtrIsValid(txn->final_lsn));
3041 :
3042 : /*
3043 : * By this time the txn has the prepare record information and it is
3044 : * important to use that so that downstream gets the accurate
3045 : * information. If instead, we have passed commit information here
3046 : * then downstream can behave as it has already replayed commit
3047 : * prepared after the restart.
3048 : */
3049 3 : ReorderBufferReplay(txn, rb, xid, txn->final_lsn, txn->end_lsn,
3050 3 : txn->prepare_time, txn->origin_id, txn->origin_lsn);
3051 : }
3052 :
3053 46 : txn->final_lsn = commit_lsn;
3054 46 : txn->end_lsn = end_lsn;
3055 46 : txn->commit_time = commit_time;
3056 46 : txn->origin_id = origin_id;
3057 46 : txn->origin_lsn = origin_lsn;
3058 :
3059 46 : if (is_commit)
3060 35 : rb->commit_prepared(rb, txn, commit_lsn);
3061 : else
3062 11 : rb->rollback_prepared(rb, txn, prepare_end_lsn, prepare_time);
3063 :
3064 : /* cleanup: make sure there's no cache pollution */
3065 46 : ReorderBufferExecuteInvalidations(txn->ninvalidations,
3066 : txn->invalidations);
3067 46 : ReorderBufferCleanupTXN(rb, txn);
3068 : }
3069 :
3070 : /*
3071 : * Abort a transaction that possibly has previous changes. Needs to be first
3072 : * called for subtransactions and then for the toplevel xid.
3073 : *
3074 : * NB: Transactions handled here have to have actively aborted (i.e. have
3075 : * produced an abort record). Implicitly aborted transactions are handled via
3076 : * ReorderBufferAbortOld(); transactions we're just not interested in, but
3077 : * which have committed are handled in ReorderBufferForget().
3078 : *
3079 : * This function purges this transaction and its contents from memory and
3080 : * disk.
3081 : */
3082 : void
3083 255 : ReorderBufferAbort(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
3084 : TimestampTz abort_time)
3085 : {
3086 : ReorderBufferTXN *txn;
3087 :
3088 255 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3089 : false);
3090 :
3091 : /* unknown, nothing to remove */
3092 255 : if (txn == NULL)
3093 0 : return;
3094 :
3095 255 : txn->abort_time = abort_time;
3096 :
3097 : /* For streamed transactions notify the remote node about the abort. */
3098 255 : if (rbtxn_is_streamed(txn))
3099 : {
3100 30 : rb->stream_abort(rb, txn, lsn);
3101 :
3102 : /*
3103 : * We might have decoded changes for this transaction that could load
3104 : * the cache as per the current transaction's view (consider DDL's
3105 : * happened in this transaction). We don't want the decoding of future
3106 : * transactions to use those cache entries so execute only the inval
3107 : * messages in this transaction.
3108 : */
3109 30 : if (txn->ninvalidations > 0)
3110 0 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3111 : txn->invalidations);
3112 : }
3113 :
3114 : /* cosmetic... */
3115 255 : txn->final_lsn = lsn;
3116 :
3117 : /* remove potential on-disk data, and deallocate */
3118 255 : ReorderBufferCleanupTXN(rb, txn);
3119 : }
3120 :
3121 : /*
3122 : * Abort all transactions that aren't actually running anymore because the
3123 : * server restarted.
3124 : *
3125 : * NB: These really have to be transactions that have aborted due to a server
3126 : * crash/immediate restart, as we don't deal with invalidations here.
3127 : */
3128 : void
3129 1701 : ReorderBufferAbortOld(ReorderBuffer *rb, TransactionId oldestRunningXid)
3130 : {
3131 : dlist_mutable_iter it;
3132 :
3133 : /*
3134 : * Iterate through all (potential) toplevel TXNs and abort all that are
3135 : * older than what possibly can be running. Once we've found the first
3136 : * that is alive we stop, there might be some that acquired an xid earlier
3137 : * but started writing later, but it's unlikely and they will be cleaned
3138 : * up in a later call to this function.
3139 : */
3140 1705 : dlist_foreach_modify(it, &rb->toplevel_by_lsn)
3141 : {
3142 : ReorderBufferTXN *txn;
3143 :
3144 63 : txn = dlist_container(ReorderBufferTXN, node, it.cur);
3145 :
3146 63 : if (TransactionIdPrecedes(txn->xid, oldestRunningXid))
3147 : {
3148 4 : elog(DEBUG2, "aborting old transaction %u", txn->xid);
3149 :
3150 : /* Notify the remote node about the crash/immediate restart. */
3151 4 : if (rbtxn_is_streamed(txn))
3152 0 : rb->stream_abort(rb, txn, InvalidXLogRecPtr);
3153 :
3154 : /* remove potential on-disk data, and deallocate this tx */
3155 4 : ReorderBufferCleanupTXN(rb, txn);
3156 : }
3157 : else
3158 59 : return;
3159 : }
3160 : }
3161 :
3162 : /*
3163 : * Forget the contents of a transaction if we aren't interested in its
3164 : * contents. Needs to be first called for subtransactions and then for the
3165 : * toplevel xid.
3166 : *
3167 : * This is significantly different to ReorderBufferAbort() because
3168 : * transactions that have committed need to be treated differently from aborted
3169 : * ones since they may have modified the catalog.
3170 : *
3171 : * Note that this is only allowed to be called in the moment a transaction
3172 : * commit has just been read, not earlier; otherwise later records referring
3173 : * to this xid might re-create the transaction incompletely.
3174 : */
3175 : void
3176 2946 : ReorderBufferForget(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3177 : {
3178 : ReorderBufferTXN *txn;
3179 :
3180 2946 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3181 : false);
3182 :
3183 : /* unknown, nothing to forget */
3184 2946 : if (txn == NULL)
3185 565 : return;
3186 :
3187 : /* this transaction mustn't be streamed */
3188 : Assert(!rbtxn_is_streamed(txn));
3189 :
3190 : /* cosmetic... */
3191 2381 : txn->final_lsn = lsn;
3192 :
3193 : /*
3194 : * Process only cache invalidation messages in this transaction if there
3195 : * are any. Even if we're not interested in the transaction's contents, it
3196 : * could have manipulated the catalog and we need to update the caches
3197 : * according to that.
3198 : */
3199 2381 : if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
3200 622 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3201 : txn->invalidations);
3202 : else
3203 : Assert(txn->ninvalidations == 0);
3204 :
3205 : /* remove potential on-disk data, and deallocate */
3206 2381 : ReorderBufferCleanupTXN(rb, txn);
3207 : }
3208 :
3209 : /*
3210 : * Invalidate cache for those transactions that need to be skipped just in case
3211 : * catalogs were manipulated as part of the transaction.
3212 : *
3213 : * Note that this is a special-purpose function for prepared transactions where
3214 : * we don't want to clean up the TXN even when we decide to skip it. See
3215 : * DecodePrepare.
3216 : */
3217 : void
3218 134 : ReorderBufferInvalidate(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3219 : {
3220 : ReorderBufferTXN *txn;
3221 :
3222 134 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3223 : false);
3224 :
3225 : /* unknown, nothing to do */
3226 134 : if (txn == NULL)
3227 0 : return;
3228 :
3229 : /*
3230 : * Process cache invalidation messages if there are any. Even if we're not
3231 : * interested in the transaction's contents, it could have manipulated the
3232 : * catalog and we need to update the caches according to that.
3233 : */
3234 134 : if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
3235 29 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3236 : txn->invalidations);
3237 : else
3238 : Assert(txn->ninvalidations == 0);
3239 : }
3240 :
3241 :
3242 : /*
3243 : * Execute invalidations happening outside the context of a decoded
3244 : * transaction. That currently happens either for xid-less commits
3245 : * (cf. RecordTransactionCommit()) or for invalidations in uninteresting
3246 : * transactions (via ReorderBufferForget()).
3247 : */
3248 : void
3249 683 : ReorderBufferImmediateInvalidation(ReorderBuffer *rb, uint32 ninvalidations,
3250 : SharedInvalidationMessage *invalidations)
3251 : {
3252 683 : bool use_subtxn = IsTransactionOrTransactionBlock();
3253 683 : MemoryContext ccxt = CurrentMemoryContext;
3254 683 : ResourceOwner cowner = CurrentResourceOwner;
3255 : int i;
3256 :
3257 683 : if (use_subtxn)
3258 477 : BeginInternalSubTransaction("replay");
3259 :
3260 : /*
3261 : * Force invalidations to happen outside of a valid transaction - that way
3262 : * entries will just be marked as invalid without accessing the catalog.
3263 : * That's advantageous because we don't need to setup the full state
3264 : * necessary for catalog access.
3265 : */
3266 683 : if (use_subtxn)
3267 477 : AbortCurrentTransaction();
3268 :
3269 28155 : for (i = 0; i < ninvalidations; i++)
3270 27472 : LocalExecuteInvalidationMessage(&invalidations[i]);
3271 :
3272 683 : if (use_subtxn)
3273 : {
3274 477 : RollbackAndReleaseCurrentSubTransaction();
3275 477 : MemoryContextSwitchTo(ccxt);
3276 477 : CurrentResourceOwner = cowner;
3277 : }
3278 683 : }
3279 :
3280 : /*
3281 : * Tell reorderbuffer about an xid seen in the WAL stream. Has to be called at
3282 : * least once for every xid in XLogRecord->xl_xid (other places in records
3283 : * may, but do not have to be passed through here).
3284 : *
3285 : * Reorderbuffer keeps some data structures about transactions in LSN order,
3286 : * for efficiency. To do that it has to know about when transactions are seen
3287 : * first in the WAL. As many types of records are not actually interesting for
3288 : * logical decoding, they do not necessarily pass through here.
3289 : */
3290 : void
3291 2343728 : ReorderBufferProcessXid(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3292 : {
3293 : /* many records won't have an xid assigned, centralize check here */
3294 2343728 : if (xid != InvalidTransactionId)
3295 2341087 : ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3296 2343728 : }
3297 :
3298 : /*
3299 : * Add a new snapshot to this transaction that may only used after lsn 'lsn'
3300 : * because the previous snapshot doesn't describe the catalog correctly for
3301 : * following rows.
3302 : */
3303 : void
3304 1619 : ReorderBufferAddSnapshot(ReorderBuffer *rb, TransactionId xid,
3305 : XLogRecPtr lsn, Snapshot snap)
3306 : {
3307 1619 : ReorderBufferChange *change = ReorderBufferAllocChange(rb);
3308 :
3309 1619 : change->data.snapshot = snap;
3310 1619 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT;
3311 :
3312 1619 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3313 1619 : }
3314 :
3315 : /*
3316 : * Set up the transaction's base snapshot.
3317 : *
3318 : * If we know that xid is a subtransaction, set the base snapshot on the
3319 : * top-level transaction instead.
3320 : */
3321 : void
3322 4152 : ReorderBufferSetBaseSnapshot(ReorderBuffer *rb, TransactionId xid,
3323 : XLogRecPtr lsn, Snapshot snap)
3324 : {
3325 : ReorderBufferTXN *txn;
3326 : bool is_new;
3327 :
3328 : Assert(snap != NULL);
3329 :
3330 : /*
3331 : * Fetch the transaction to operate on. If we know it's a subtransaction,
3332 : * operate on its top-level transaction instead.
3333 : */
3334 4152 : txn = ReorderBufferTXNByXid(rb, xid, true, &is_new, lsn, true);
3335 4152 : if (rbtxn_is_known_subxact(txn))
3336 114 : txn = ReorderBufferTXNByXid(rb, txn->toplevel_xid, false,
3337 : NULL, InvalidXLogRecPtr, false);
3338 : Assert(txn->base_snapshot == NULL);
3339 :
3340 4152 : txn->base_snapshot = snap;
3341 4152 : txn->base_snapshot_lsn = lsn;
3342 4152 : dlist_push_tail(&rb->txns_by_base_snapshot_lsn, &txn->base_snapshot_node);
3343 :
3344 4152 : AssertTXNLsnOrder(rb);
3345 4152 : }
3346 :
3347 : /*
3348 : * Access the catalog with this CommandId at this point in the changestream.
3349 : *
3350 : * May only be called for command ids > 1
3351 : */
3352 : void
3353 27598 : ReorderBufferAddNewCommandId(ReorderBuffer *rb, TransactionId xid,
3354 : XLogRecPtr lsn, CommandId cid)
3355 : {
3356 27598 : ReorderBufferChange *change = ReorderBufferAllocChange(rb);
3357 :
3358 27598 : change->data.command_id = cid;
3359 27598 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID;
3360 :
3361 27598 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3362 27598 : }
3363 :
3364 : /*
3365 : * Update memory counters to account for the new or removed change.
3366 : *
3367 : * We update two counters - in the reorder buffer, and in the transaction
3368 : * containing the change. The reorder buffer counter allows us to quickly
3369 : * decide if we reached the memory limit, the transaction counter allows
3370 : * us to quickly pick the largest transaction for eviction.
3371 : *
3372 : * Either txn or change must be non-NULL at least. We update the memory
3373 : * counter of txn if it's non-NULL, otherwise change->txn.
3374 : *
3375 : * When streaming is enabled, we need to update the toplevel transaction
3376 : * counters instead - we don't really care about subtransactions as we
3377 : * can't stream them individually anyway, and we only pick toplevel
3378 : * transactions for eviction. So only toplevel transactions matter.
3379 : */
3380 : static void
3381 1977514 : ReorderBufferChangeMemoryUpdate(ReorderBuffer *rb,
3382 : ReorderBufferChange *change,
3383 : ReorderBufferTXN *txn,
3384 : bool addition, Size sz)
3385 : {
3386 : ReorderBufferTXN *toptxn;
3387 :
3388 : Assert(txn || change);
3389 :
3390 : /*
3391 : * Ignore tuple CID changes, because those are not evicted when reaching
3392 : * memory limit. So we just don't count them, because it might easily
3393 : * trigger a pointless attempt to spill.
3394 : */
3395 1977514 : if (change && change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID)
3396 27256 : return;
3397 :
3398 1950258 : if (sz == 0)
3399 1175 : return;
3400 :
3401 1949083 : if (txn == NULL)
3402 1940126 : txn = change->txn;
3403 : Assert(txn != NULL);
3404 :
3405 : /*
3406 : * Update the total size in top level as well. This is later used to
3407 : * compute the decoding stats.
3408 : */
3409 1949083 : toptxn = rbtxn_get_toptxn(txn);
3410 :
3411 1949083 : if (addition)
3412 : {
3413 1761834 : Size oldsize = txn->size;
3414 :
3415 1761834 : txn->size += sz;
3416 1761834 : rb->size += sz;
3417 :
3418 : /* Update the total size in the top transaction. */
3419 1761834 : toptxn->total_size += sz;
3420 :
3421 : /* Update the max-heap */
3422 1761834 : if (oldsize != 0)
3423 1752803 : pairingheap_remove(rb->txn_heap, &txn->txn_node);
3424 1761834 : pairingheap_add(rb->txn_heap, &txn->txn_node);
3425 : }
3426 : else
3427 : {
3428 : Assert((rb->size >= sz) && (txn->size >= sz));
3429 187249 : txn->size -= sz;
3430 187249 : rb->size -= sz;
3431 :
3432 : /* Update the total size in the top transaction. */
3433 187249 : toptxn->total_size -= sz;
3434 :
3435 : /* Update the max-heap */
3436 187249 : pairingheap_remove(rb->txn_heap, &txn->txn_node);
3437 187249 : if (txn->size != 0)
3438 178263 : pairingheap_add(rb->txn_heap, &txn->txn_node);
3439 : }
3440 :
3441 : Assert(txn->size <= rb->size);
3442 : }
3443 :
3444 : /*
3445 : * Add new (relfilelocator, tid) -> (cmin, cmax) mappings.
3446 : *
3447 : * We do not include this change type in memory accounting, because we
3448 : * keep CIDs in a separate list and do not evict them when reaching
3449 : * the memory limit.
3450 : */
3451 : void
3452 27598 : ReorderBufferAddNewTupleCids(ReorderBuffer *rb, TransactionId xid,
3453 : XLogRecPtr lsn, RelFileLocator locator,
3454 : ItemPointerData tid, CommandId cmin,
3455 : CommandId cmax, CommandId combocid)
3456 : {
3457 27598 : ReorderBufferChange *change = ReorderBufferAllocChange(rb);
3458 : ReorderBufferTXN *txn;
3459 :
3460 27598 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3461 :
3462 27598 : change->data.tuplecid.locator = locator;
3463 27598 : change->data.tuplecid.tid = tid;
3464 27598 : change->data.tuplecid.cmin = cmin;
3465 27598 : change->data.tuplecid.cmax = cmax;
3466 27598 : change->data.tuplecid.combocid = combocid;
3467 27598 : change->lsn = lsn;
3468 27598 : change->txn = txn;
3469 27598 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID;
3470 :
3471 27598 : dlist_push_tail(&txn->tuplecids, &change->node);
3472 27598 : txn->ntuplecids++;
3473 27598 : }
3474 :
3475 : /*
3476 : * Add new invalidation messages to the reorder buffer queue.
3477 : */
3478 : static void
3479 5932 : ReorderBufferQueueInvalidations(ReorderBuffer *rb, TransactionId xid,
3480 : XLogRecPtr lsn, Size nmsgs,
3481 : SharedInvalidationMessage *msgs)
3482 : {
3483 : ReorderBufferChange *change;
3484 :
3485 5932 : change = ReorderBufferAllocChange(rb);
3486 5932 : change->action = REORDER_BUFFER_CHANGE_INVALIDATION;
3487 5932 : change->data.inval.ninvalidations = nmsgs;
3488 5932 : change->data.inval.invalidations = palloc_array(SharedInvalidationMessage, nmsgs);
3489 5932 : memcpy(change->data.inval.invalidations, msgs,
3490 : sizeof(SharedInvalidationMessage) * nmsgs);
3491 :
3492 5932 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3493 5932 : }
3494 :
3495 : /*
3496 : * A helper function for ReorderBufferAddInvalidations() and
3497 : * ReorderBufferAddDistributedInvalidations() to accumulate the invalidation
3498 : * messages to the **invals_out.
3499 : */
3500 : static void
3501 5932 : ReorderBufferAccumulateInvalidations(SharedInvalidationMessage **invals_out,
3502 : uint32 *ninvals_out,
3503 : SharedInvalidationMessage *msgs_new,
3504 : Size nmsgs_new)
3505 : {
3506 5932 : if (*ninvals_out == 0)
3507 : {
3508 1619 : *ninvals_out = nmsgs_new;
3509 1619 : *invals_out = palloc_array(SharedInvalidationMessage, nmsgs_new);
3510 1619 : memcpy(*invals_out, msgs_new, sizeof(SharedInvalidationMessage) * nmsgs_new);
3511 : }
3512 : else
3513 : {
3514 : /* Enlarge the array of inval messages */
3515 4313 : *invals_out =
3516 4313 : repalloc_array(*invals_out, SharedInvalidationMessage,
3517 : (*ninvals_out + nmsgs_new));
3518 4313 : memcpy(*invals_out + *ninvals_out, msgs_new,
3519 : nmsgs_new * sizeof(SharedInvalidationMessage));
3520 4313 : *ninvals_out += nmsgs_new;
3521 : }
3522 5932 : }
3523 :
3524 : /*
3525 : * Accumulate the invalidations for executing them later.
3526 : *
3527 : * This needs to be called for each XLOG_XACT_INVALIDATIONS message and
3528 : * accumulates all the invalidation messages in the toplevel transaction, if
3529 : * available, otherwise in the current transaction, as well as in the form of
3530 : * change in reorder buffer. We require to record it in form of the change
3531 : * so that we can execute only the required invalidations instead of executing
3532 : * all the invalidations on each CommandId increment. We also need to
3533 : * accumulate these in the txn buffer because in some cases where we skip
3534 : * processing the transaction (see ReorderBufferForget), we need to execute
3535 : * all the invalidations together.
3536 : */
3537 : void
3538 5904 : ReorderBufferAddInvalidations(ReorderBuffer *rb, TransactionId xid,
3539 : XLogRecPtr lsn, Size nmsgs,
3540 : SharedInvalidationMessage *msgs)
3541 : {
3542 : ReorderBufferTXN *txn;
3543 : MemoryContext oldcontext;
3544 :
3545 5904 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3546 :
3547 5904 : oldcontext = MemoryContextSwitchTo(rb->context);
3548 :
3549 : /*
3550 : * Collect all the invalidations under the top transaction, if available,
3551 : * so that we can execute them all together. See comments atop this
3552 : * function.
3553 : */
3554 5904 : txn = rbtxn_get_toptxn(txn);
3555 :
3556 : Assert(nmsgs > 0);
3557 :
3558 5904 : ReorderBufferAccumulateInvalidations(&txn->invalidations,
3559 : &txn->ninvalidations,
3560 : msgs, nmsgs);
3561 :
3562 5904 : ReorderBufferQueueInvalidations(rb, xid, lsn, nmsgs, msgs);
3563 :
3564 5904 : MemoryContextSwitchTo(oldcontext);
3565 5904 : }
3566 :
3567 : /*
3568 : * Accumulate the invalidations distributed by other committed transactions
3569 : * for executing them later.
3570 : *
3571 : * This function is similar to ReorderBufferAddInvalidations() but stores
3572 : * the given inval messages to the txn->invalidations_distributed with the
3573 : * overflow check.
3574 : *
3575 : * This needs to be called by committed transactions to distribute their
3576 : * inval messages to in-progress transactions.
3577 : */
3578 : void
3579 28 : ReorderBufferAddDistributedInvalidations(ReorderBuffer *rb, TransactionId xid,
3580 : XLogRecPtr lsn, Size nmsgs,
3581 : SharedInvalidationMessage *msgs)
3582 : {
3583 : ReorderBufferTXN *txn;
3584 : MemoryContext oldcontext;
3585 :
3586 28 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3587 :
3588 28 : oldcontext = MemoryContextSwitchTo(rb->context);
3589 :
3590 : /*
3591 : * Collect all the invalidations under the top transaction, if available,
3592 : * so that we can execute them all together. See comments
3593 : * ReorderBufferAddInvalidations.
3594 : */
3595 28 : txn = rbtxn_get_toptxn(txn);
3596 :
3597 : Assert(nmsgs > 0);
3598 :
3599 28 : if (!rbtxn_distr_inval_overflowed(txn))
3600 : {
3601 : /*
3602 : * Check the transaction has enough space for storing distributed
3603 : * invalidation messages.
3604 : */
3605 28 : if (txn->ninvalidations_distributed + nmsgs >= MAX_DISTR_INVAL_MSG_PER_TXN)
3606 : {
3607 : /*
3608 : * Mark the invalidation message as overflowed and free up the
3609 : * messages accumulated so far.
3610 : */
3611 0 : txn->txn_flags |= RBTXN_DISTR_INVAL_OVERFLOWED;
3612 :
3613 0 : if (txn->invalidations_distributed)
3614 : {
3615 0 : pfree(txn->invalidations_distributed);
3616 0 : txn->invalidations_distributed = NULL;
3617 0 : txn->ninvalidations_distributed = 0;
3618 : }
3619 : }
3620 : else
3621 28 : ReorderBufferAccumulateInvalidations(&txn->invalidations_distributed,
3622 : &txn->ninvalidations_distributed,
3623 : msgs, nmsgs);
3624 : }
3625 :
3626 : /* Queue the invalidation messages into the transaction */
3627 28 : ReorderBufferQueueInvalidations(rb, xid, lsn, nmsgs, msgs);
3628 :
3629 28 : MemoryContextSwitchTo(oldcontext);
3630 28 : }
3631 :
3632 : /*
3633 : * Apply all invalidations we know. Possibly we only need parts at this point
3634 : * in the changestream but we don't know which those are.
3635 : */
3636 : static void
3637 8122 : ReorderBufferExecuteInvalidations(uint32 nmsgs, SharedInvalidationMessage *msgs)
3638 : {
3639 : int i;
3640 :
3641 55666 : for (i = 0; i < nmsgs; i++)
3642 47544 : LocalExecuteInvalidationMessage(&msgs[i]);
3643 8122 : }
3644 :
3645 : /*
3646 : * Mark a transaction as containing catalog changes
3647 : */
3648 : void
3649 33577 : ReorderBufferXidSetCatalogChanges(ReorderBuffer *rb, TransactionId xid,
3650 : XLogRecPtr lsn)
3651 : {
3652 : ReorderBufferTXN *txn;
3653 :
3654 33577 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3655 :
3656 33577 : if (!rbtxn_has_catalog_changes(txn))
3657 : {
3658 1647 : txn->txn_flags |= RBTXN_HAS_CATALOG_CHANGES;
3659 1647 : dclist_push_tail(&rb->catchange_txns, &txn->catchange_node);
3660 : }
3661 :
3662 : /*
3663 : * Mark top-level transaction as having catalog changes too if one of its
3664 : * children has so that the ReorderBufferBuildTupleCidHash can
3665 : * conveniently check just top-level transaction and decide whether to
3666 : * build the hash table or not.
3667 : */
3668 33577 : if (rbtxn_is_subtxn(txn))
3669 : {
3670 896 : ReorderBufferTXN *toptxn = rbtxn_get_toptxn(txn);
3671 :
3672 896 : if (!rbtxn_has_catalog_changes(toptxn))
3673 : {
3674 20 : toptxn->txn_flags |= RBTXN_HAS_CATALOG_CHANGES;
3675 20 : dclist_push_tail(&rb->catchange_txns, &toptxn->catchange_node);
3676 : }
3677 : }
3678 33577 : }
3679 :
3680 : /*
3681 : * Return palloc'ed array of the transactions that have changed catalogs.
3682 : * The returned array is sorted in xidComparator order.
3683 : *
3684 : * The caller must free the returned array when done with it.
3685 : */
3686 : TransactionId *
3687 329 : ReorderBufferGetCatalogChangesXacts(ReorderBuffer *rb)
3688 : {
3689 : dlist_iter iter;
3690 329 : TransactionId *xids = NULL;
3691 329 : size_t xcnt = 0;
3692 :
3693 : /* Quick return if the list is empty */
3694 329 : if (dclist_count(&rb->catchange_txns) == 0)
3695 321 : return NULL;
3696 :
3697 : /* Initialize XID array */
3698 8 : xids = palloc_array(TransactionId, dclist_count(&rb->catchange_txns));
3699 19 : dclist_foreach(iter, &rb->catchange_txns)
3700 : {
3701 11 : ReorderBufferTXN *txn = dclist_container(ReorderBufferTXN,
3702 : catchange_node,
3703 : iter.cur);
3704 :
3705 : Assert(rbtxn_has_catalog_changes(txn));
3706 :
3707 11 : xids[xcnt++] = txn->xid;
3708 : }
3709 :
3710 8 : qsort(xids, xcnt, sizeof(TransactionId), xidComparator);
3711 :
3712 : Assert(xcnt == dclist_count(&rb->catchange_txns));
3713 8 : return xids;
3714 : }
3715 :
3716 : /*
3717 : * Query whether a transaction is already *known* to contain catalog
3718 : * changes. This can be wrong until directly before the commit!
3719 : */
3720 : bool
3721 5203 : ReorderBufferXidHasCatalogChanges(ReorderBuffer *rb, TransactionId xid)
3722 : {
3723 : ReorderBufferTXN *txn;
3724 :
3725 5203 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3726 : false);
3727 5203 : if (txn == NULL)
3728 690 : return false;
3729 :
3730 4513 : return rbtxn_has_catalog_changes(txn);
3731 : }
3732 :
3733 : /*
3734 : * ReorderBufferXidHasBaseSnapshot
3735 : * Have we already set the base snapshot for the given txn/subtxn?
3736 : */
3737 : bool
3738 1572641 : ReorderBufferXidHasBaseSnapshot(ReorderBuffer *rb, TransactionId xid)
3739 : {
3740 : ReorderBufferTXN *txn;
3741 :
3742 1572641 : txn = ReorderBufferTXNByXid(rb, xid, false,
3743 : NULL, InvalidXLogRecPtr, false);
3744 :
3745 : /* transaction isn't known yet, ergo no snapshot */
3746 1572641 : if (txn == NULL)
3747 3 : return false;
3748 :
3749 : /* a known subtxn? operate on top-level txn instead */
3750 1572638 : if (rbtxn_is_known_subxact(txn))
3751 432037 : txn = ReorderBufferTXNByXid(rb, txn->toplevel_xid, false,
3752 : NULL, InvalidXLogRecPtr, false);
3753 :
3754 1572638 : return txn->base_snapshot != NULL;
3755 : }
3756 :
3757 :
3758 : /*
3759 : * ---------------------------------------
3760 : * Disk serialization support
3761 : * ---------------------------------------
3762 : */
3763 :
3764 : /*
3765 : * Ensure the IO buffer is >= sz.
3766 : */
3767 : static void
3768 2954152 : ReorderBufferSerializeReserve(ReorderBuffer *rb, Size sz)
3769 : {
3770 2954152 : if (!rb->outbufsize)
3771 : {
3772 50 : rb->outbuf = MemoryContextAlloc(rb->context, sz);
3773 50 : rb->outbufsize = sz;
3774 : }
3775 2954102 : else if (rb->outbufsize < sz)
3776 : {
3777 281 : rb->outbuf = repalloc(rb->outbuf, sz);
3778 281 : rb->outbufsize = sz;
3779 : }
3780 2954152 : }
3781 :
3782 :
3783 : /* Compare two transactions by size */
3784 : static int
3785 342033 : ReorderBufferTXNSizeCompare(const pairingheap_node *a, const pairingheap_node *b, void *arg)
3786 : {
3787 342033 : const ReorderBufferTXN *ta = pairingheap_const_container(ReorderBufferTXN, txn_node, a);
3788 342033 : const ReorderBufferTXN *tb = pairingheap_const_container(ReorderBufferTXN, txn_node, b);
3789 :
3790 342033 : if (ta->size < tb->size)
3791 243118 : return -1;
3792 98915 : if (ta->size > tb->size)
3793 98003 : return 1;
3794 912 : return 0;
3795 : }
3796 :
3797 : /*
3798 : * Find the largest transaction (toplevel or subxact) to evict (spill to disk).
3799 : */
3800 : static ReorderBufferTXN *
3801 4002 : ReorderBufferLargestTXN(ReorderBuffer *rb)
3802 : {
3803 : ReorderBufferTXN *largest;
3804 :
3805 : /* Get the largest transaction from the max-heap */
3806 4002 : largest = pairingheap_container(ReorderBufferTXN, txn_node,
3807 : pairingheap_first(rb->txn_heap));
3808 :
3809 : Assert(largest);
3810 : Assert(largest->size > 0);
3811 : Assert(largest->size <= rb->size);
3812 :
3813 4002 : return largest;
3814 : }
3815 :
3816 : /*
3817 : * Find the largest streamable (and non-aborted) toplevel transaction to evict
3818 : * (by streaming).
3819 : *
3820 : * This can be seen as an optimized version of ReorderBufferLargestTXN, which
3821 : * should give us the same transaction (because we don't update memory account
3822 : * for subtransaction with streaming, so it's always 0). But we can simply
3823 : * iterate over the limited number of toplevel transactions that have a base
3824 : * snapshot. There is no use of selecting a transaction that doesn't have base
3825 : * snapshot because we don't decode such transactions. Also, we do not select
3826 : * the transaction which doesn't have any streamable change.
3827 : *
3828 : * Note that, we skip transactions that contain incomplete changes. There
3829 : * is a scope of optimization here such that we can select the largest
3830 : * transaction which has incomplete changes. But that will make the code and
3831 : * design quite complex and that might not be worth the benefit. If we plan to
3832 : * stream the transactions that contain incomplete changes then we need to
3833 : * find a way to partially stream/truncate the transaction changes in-memory
3834 : * and build a mechanism to partially truncate the spilled files.
3835 : * Additionally, whenever we partially stream the transaction we need to
3836 : * maintain the last streamed lsn and next time we need to restore from that
3837 : * segment and the offset in WAL. As we stream the changes from the top
3838 : * transaction and restore them subtransaction wise, we need to even remember
3839 : * the subxact from where we streamed the last change.
3840 : */
3841 : static ReorderBufferTXN *
3842 821 : ReorderBufferLargestStreamableTopTXN(ReorderBuffer *rb)
3843 : {
3844 : dlist_iter iter;
3845 821 : Size largest_size = 0;
3846 821 : ReorderBufferTXN *largest = NULL;
3847 :
3848 : /* Find the largest top-level transaction having a base snapshot. */
3849 1754 : dlist_foreach(iter, &rb->txns_by_base_snapshot_lsn)
3850 : {
3851 : ReorderBufferTXN *txn;
3852 :
3853 933 : txn = dlist_container(ReorderBufferTXN, base_snapshot_node, iter.cur);
3854 :
3855 : /* must not be a subtxn */
3856 : Assert(!rbtxn_is_known_subxact(txn));
3857 : /* base_snapshot must be set */
3858 : Assert(txn->base_snapshot != NULL);
3859 :
3860 : /* Don't consider these kinds of transactions for eviction. */
3861 933 : if (rbtxn_has_partial_change(txn) ||
3862 786 : !rbtxn_has_streamable_change(txn) ||
3863 756 : rbtxn_is_aborted(txn))
3864 177 : continue;
3865 :
3866 : /* Find the largest of the eviction candidates. */
3867 756 : if ((largest == NULL || txn->total_size > largest_size) &&
3868 756 : (txn->total_size > 0))
3869 : {
3870 710 : largest = txn;
3871 710 : largest_size = txn->total_size;
3872 : }
3873 : }
3874 :
3875 821 : return largest;
3876 : }
3877 :
3878 : /*
3879 : * Check whether the logical_decoding_work_mem limit was reached, and if yes
3880 : * pick the largest (sub)transaction at-a-time to evict and spill its changes to
3881 : * disk or send to the output plugin until we reach under the memory limit.
3882 : *
3883 : * If debug_logical_replication_streaming is set to "immediate", stream or
3884 : * serialize the changes immediately.
3885 : *
3886 : * XXX At this point we select the transactions until we reach under the memory
3887 : * limit, but we might also adapt a more elaborate eviction strategy - for example
3888 : * evicting enough transactions to free certain fraction (e.g. 50%) of the memory
3889 : * limit.
3890 : */
3891 : static void
3892 1583573 : ReorderBufferCheckMemoryLimit(ReorderBuffer *rb)
3893 : {
3894 : ReorderBufferTXN *txn;
3895 1583573 : bool update_stats = true;
3896 :
3897 1583573 : if (rb->size >= logical_decoding_work_mem * (Size) 1024)
3898 : {
3899 : /*
3900 : * Update the statistics as the memory usage has reached the limit. We
3901 : * report the statistics update later in this function since we can
3902 : * update the slot statistics altogether while streaming or
3903 : * serializing transactions in most cases.
3904 : */
3905 3438 : rb->memExceededCount += 1;
3906 : }
3907 1580135 : else if (debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_BUFFERED)
3908 : {
3909 : /*
3910 : * Bail out if debug_logical_replication_streaming is buffered and we
3911 : * haven't exceeded the memory limit.
3912 : */
3913 1578924 : return;
3914 : }
3915 :
3916 : /*
3917 : * If debug_logical_replication_streaming is immediate, loop until there's
3918 : * no change. Otherwise, loop until we reach under the memory limit. One
3919 : * might think that just by evicting the largest (sub)transaction we will
3920 : * come under the memory limit based on assumption that the selected
3921 : * transaction is at least as large as the most recent change (which
3922 : * caused us to go over the memory limit). However, that is not true
3923 : * because a user can reduce the logical_decoding_work_mem to a smaller
3924 : * value before the most recent change.
3925 : */
3926 9295 : while (rb->size >= logical_decoding_work_mem * (Size) 1024 ||
3927 5857 : (debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_IMMEDIATE &&
3928 2419 : rb->size > 0))
3929 : {
3930 : /*
3931 : * Pick the largest non-aborted transaction and evict it from memory
3932 : * by streaming, if possible. Otherwise, spill to disk.
3933 : */
3934 5467 : if (ReorderBufferCanStartStreaming(rb) &&
3935 821 : (txn = ReorderBufferLargestStreamableTopTXN(rb)) != NULL)
3936 : {
3937 : /* we know there has to be one, because the size is not zero */
3938 : Assert(txn && rbtxn_is_toptxn(txn));
3939 : Assert(txn->total_size > 0);
3940 : Assert(rb->size >= txn->total_size);
3941 :
3942 : /* skip the transaction if aborted */
3943 644 : if (ReorderBufferCheckAndTruncateAbortedTXN(rb, txn))
3944 0 : continue;
3945 :
3946 644 : ReorderBufferStreamTXN(rb, txn);
3947 : }
3948 : else
3949 : {
3950 : /*
3951 : * Pick the largest transaction (or subtransaction) and evict it
3952 : * from memory by serializing it to disk.
3953 : */
3954 4002 : txn = ReorderBufferLargestTXN(rb);
3955 :
3956 : /* we know there has to be one, because the size is not zero */
3957 : Assert(txn);
3958 : Assert(txn->size > 0);
3959 : Assert(rb->size >= txn->size);
3960 :
3961 : /* skip the transaction if aborted */
3962 4002 : if (ReorderBufferCheckAndTruncateAbortedTXN(rb, txn))
3963 9 : continue;
3964 :
3965 3993 : ReorderBufferSerializeTXN(rb, txn);
3966 : }
3967 :
3968 : /*
3969 : * After eviction, the transaction should have no entries in memory,
3970 : * and should use 0 bytes for changes.
3971 : */
3972 : Assert(txn->size == 0);
3973 : Assert(txn->nentries_mem == 0);
3974 :
3975 : /*
3976 : * We've reported the memExceededCount update while streaming or
3977 : * serializing the transaction.
3978 : */
3979 4637 : update_stats = false;
3980 : }
3981 :
3982 4649 : if (update_stats)
3983 12 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
3984 :
3985 : /* We must be under the memory limit now. */
3986 : Assert(rb->size < logical_decoding_work_mem * (Size) 1024);
3987 : }
3988 :
3989 : /*
3990 : * Spill data of a large transaction (and its subtransactions) to disk.
3991 : */
3992 : static void
3993 4278 : ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
3994 : {
3995 : dlist_iter subtxn_i;
3996 : dlist_mutable_iter change_i;
3997 4278 : int fd = -1;
3998 4278 : XLogSegNo curOpenSegNo = 0;
3999 4278 : Size spilled = 0;
4000 4278 : Size size = txn->size;
4001 :
4002 4278 : elog(DEBUG2, "spill %u changes in XID %u to disk",
4003 : (uint32) txn->nentries_mem, txn->xid);
4004 :
4005 : /* do the same to all child TXs */
4006 4523 : dlist_foreach(subtxn_i, &txn->subtxns)
4007 : {
4008 : ReorderBufferTXN *subtxn;
4009 :
4010 245 : subtxn = dlist_container(ReorderBufferTXN, node, subtxn_i.cur);
4011 245 : ReorderBufferSerializeTXN(rb, subtxn);
4012 : }
4013 :
4014 : /* serialize changestream */
4015 1311982 : dlist_foreach_modify(change_i, &txn->changes)
4016 : {
4017 : ReorderBufferChange *change;
4018 :
4019 1307704 : change = dlist_container(ReorderBufferChange, node, change_i.cur);
4020 :
4021 : /*
4022 : * store in segment in which it belongs by start lsn, don't split over
4023 : * multiple segments tho
4024 : */
4025 1307704 : if (fd == -1 ||
4026 1303656 : !XLByteInSeg(change->lsn, curOpenSegNo, wal_segment_size))
4027 : {
4028 : char path[MAXPGPATH];
4029 :
4030 4066 : if (fd != -1)
4031 18 : CloseTransientFile(fd);
4032 :
4033 4066 : XLByteToSeg(change->lsn, curOpenSegNo, wal_segment_size);
4034 :
4035 : /*
4036 : * No need to care about TLIs here, only used during a single run,
4037 : * so each LSN only maps to a specific WAL record.
4038 : */
4039 4066 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid,
4040 : curOpenSegNo);
4041 :
4042 : /* open segment, create it if necessary */
4043 4066 : fd = OpenTransientFile(path,
4044 : O_CREAT | O_WRONLY | O_APPEND | PG_BINARY);
4045 :
4046 4066 : if (fd < 0)
4047 0 : ereport(ERROR,
4048 : (errcode_for_file_access(),
4049 : errmsg("could not open file \"%s\": %m", path)));
4050 : }
4051 :
4052 1307704 : ReorderBufferSerializeChange(rb, txn, fd, change);
4053 1307704 : dlist_delete(&change->node);
4054 1307704 : ReorderBufferFreeChange(rb, change, false);
4055 :
4056 1307704 : spilled++;
4057 : }
4058 :
4059 : /* Update the memory counter */
4060 4278 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, size);
4061 :
4062 : /* update the statistics iff we have spilled anything */
4063 4278 : if (spilled)
4064 : {
4065 4048 : rb->spillCount += 1;
4066 4048 : rb->spillBytes += size;
4067 :
4068 : /* don't consider already serialized transactions */
4069 4048 : rb->spillTxns += (rbtxn_is_serialized(txn) || rbtxn_is_serialized_clear(txn)) ? 0 : 1;
4070 :
4071 : /* update the decoding stats */
4072 4048 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
4073 : }
4074 :
4075 : Assert(spilled == txn->nentries_mem);
4076 : Assert(dlist_is_empty(&txn->changes));
4077 4278 : txn->nentries_mem = 0;
4078 4278 : txn->txn_flags |= RBTXN_IS_SERIALIZED;
4079 :
4080 4278 : if (fd != -1)
4081 4048 : CloseTransientFile(fd);
4082 4278 : }
4083 :
4084 : /*
4085 : * Serialize individual change to disk.
4086 : */
4087 : static void
4088 1307704 : ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
4089 : int fd, ReorderBufferChange *change)
4090 : {
4091 : ReorderBufferDiskChange *ondisk;
4092 1307704 : Size sz = sizeof(ReorderBufferDiskChange);
4093 :
4094 1307704 : ReorderBufferSerializeReserve(rb, sz);
4095 :
4096 1307704 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4097 1307704 : memcpy(&ondisk->change, change, sizeof(ReorderBufferChange));
4098 :
4099 1307704 : switch (change->action)
4100 : {
4101 : /* fall through these, they're all similar enough */
4102 1290216 : case REORDER_BUFFER_CHANGE_INSERT:
4103 : case REORDER_BUFFER_CHANGE_UPDATE:
4104 : case REORDER_BUFFER_CHANGE_DELETE:
4105 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4106 : {
4107 : char *data;
4108 : HeapTuple oldtup,
4109 : newtup;
4110 1290216 : Size oldlen = 0;
4111 1290216 : Size newlen = 0;
4112 :
4113 1290216 : oldtup = change->data.tp.oldtuple;
4114 1290216 : newtup = change->data.tp.newtuple;
4115 :
4116 1290216 : if (oldtup)
4117 : {
4118 131227 : sz += sizeof(HeapTupleData);
4119 131227 : oldlen = oldtup->t_len;
4120 131227 : sz += oldlen;
4121 : }
4122 :
4123 1290216 : if (newtup)
4124 : {
4125 1105274 : sz += sizeof(HeapTupleData);
4126 1105274 : newlen = newtup->t_len;
4127 1105274 : sz += newlen;
4128 : }
4129 :
4130 : /* make sure we have enough space */
4131 1290216 : ReorderBufferSerializeReserve(rb, sz);
4132 :
4133 1290216 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4134 : /* might have been reallocated above */
4135 1290216 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4136 :
4137 1290216 : if (oldlen)
4138 : {
4139 131227 : memcpy(data, oldtup, sizeof(HeapTupleData));
4140 131227 : data += sizeof(HeapTupleData);
4141 :
4142 131227 : memcpy(data, oldtup->t_data, oldlen);
4143 131227 : data += oldlen;
4144 : }
4145 :
4146 1290216 : if (newlen)
4147 : {
4148 1105274 : memcpy(data, newtup, sizeof(HeapTupleData));
4149 1105274 : data += sizeof(HeapTupleData);
4150 :
4151 1105274 : memcpy(data, newtup->t_data, newlen);
4152 1105274 : data += newlen;
4153 : }
4154 1290216 : break;
4155 : }
4156 13 : case REORDER_BUFFER_CHANGE_MESSAGE:
4157 : {
4158 : char *data;
4159 13 : Size prefix_size = strlen(change->data.msg.prefix) + 1;
4160 :
4161 13 : sz += prefix_size + change->data.msg.message_size +
4162 : sizeof(Size) + sizeof(Size);
4163 13 : ReorderBufferSerializeReserve(rb, sz);
4164 :
4165 13 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4166 :
4167 : /* might have been reallocated above */
4168 13 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4169 :
4170 : /* write the prefix including the size */
4171 13 : memcpy(data, &prefix_size, sizeof(Size));
4172 13 : data += sizeof(Size);
4173 13 : memcpy(data, change->data.msg.prefix,
4174 : prefix_size);
4175 13 : data += prefix_size;
4176 :
4177 : /* write the message including the size */
4178 13 : memcpy(data, &change->data.msg.message_size, sizeof(Size));
4179 13 : data += sizeof(Size);
4180 13 : memcpy(data, change->data.msg.message,
4181 : change->data.msg.message_size);
4182 13 : data += change->data.msg.message_size;
4183 :
4184 13 : break;
4185 : }
4186 154 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4187 : {
4188 : char *data;
4189 154 : Size inval_size = sizeof(SharedInvalidationMessage) *
4190 154 : change->data.inval.ninvalidations;
4191 :
4192 154 : sz += inval_size;
4193 :
4194 154 : ReorderBufferSerializeReserve(rb, sz);
4195 154 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4196 :
4197 : /* might have been reallocated above */
4198 154 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4199 154 : memcpy(data, change->data.inval.invalidations, inval_size);
4200 154 : data += inval_size;
4201 :
4202 154 : break;
4203 : }
4204 8 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4205 : {
4206 : Snapshot snap;
4207 : char *data;
4208 :
4209 8 : snap = change->data.snapshot;
4210 :
4211 8 : sz += sizeof(SnapshotData) +
4212 8 : sizeof(TransactionId) * snap->xcnt +
4213 8 : sizeof(TransactionId) * snap->subxcnt;
4214 :
4215 : /* make sure we have enough space */
4216 8 : ReorderBufferSerializeReserve(rb, sz);
4217 8 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4218 : /* might have been reallocated above */
4219 8 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4220 :
4221 8 : memcpy(data, snap, sizeof(SnapshotData));
4222 8 : data += sizeof(SnapshotData);
4223 :
4224 8 : if (snap->xcnt)
4225 : {
4226 8 : memcpy(data, snap->xip,
4227 8 : sizeof(TransactionId) * snap->xcnt);
4228 8 : data += sizeof(TransactionId) * snap->xcnt;
4229 : }
4230 :
4231 8 : if (snap->subxcnt)
4232 : {
4233 0 : memcpy(data, snap->subxip,
4234 0 : sizeof(TransactionId) * snap->subxcnt);
4235 0 : data += sizeof(TransactionId) * snap->subxcnt;
4236 : }
4237 8 : break;
4238 : }
4239 2 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4240 : {
4241 : Size size;
4242 : char *data;
4243 :
4244 : /* account for the OIDs of truncated relations */
4245 2 : size = sizeof(Oid) * change->data.truncate.nrelids;
4246 2 : sz += size;
4247 :
4248 : /* make sure we have enough space */
4249 2 : ReorderBufferSerializeReserve(rb, sz);
4250 :
4251 2 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4252 : /* might have been reallocated above */
4253 2 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4254 :
4255 2 : memcpy(data, change->data.truncate.relids, size);
4256 2 : data += size;
4257 :
4258 2 : break;
4259 : }
4260 17311 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4261 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4262 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4263 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4264 : /* ReorderBufferChange contains everything important */
4265 17311 : break;
4266 : }
4267 :
4268 1307704 : ondisk->size = sz;
4269 :
4270 1307704 : errno = 0;
4271 1307704 : pgstat_report_wait_start(WAIT_EVENT_REORDER_BUFFER_WRITE);
4272 1307704 : if (write(fd, rb->outbuf, ondisk->size) != ondisk->size)
4273 : {
4274 0 : int save_errno = errno;
4275 :
4276 0 : CloseTransientFile(fd);
4277 :
4278 : /* if write didn't set errno, assume problem is no disk space */
4279 0 : errno = save_errno ? save_errno : ENOSPC;
4280 0 : ereport(ERROR,
4281 : (errcode_for_file_access(),
4282 : errmsg("could not write to data file for XID %u: %m",
4283 : txn->xid)));
4284 : }
4285 1307704 : pgstat_report_wait_end();
4286 :
4287 : /*
4288 : * Keep the transaction's final_lsn up to date with each change we send to
4289 : * disk, so that ReorderBufferRestoreCleanup works correctly. (We used to
4290 : * only do this on commit and abort records, but that doesn't work if a
4291 : * system crash leaves a transaction without its abort record).
4292 : *
4293 : * Make sure not to move it backwards.
4294 : */
4295 1307704 : if (txn->final_lsn < change->lsn)
4296 1303221 : txn->final_lsn = change->lsn;
4297 :
4298 : Assert(ondisk->change.action == change->action);
4299 1307704 : }
4300 :
4301 : /* Returns true, if the output plugin supports streaming, false, otherwise. */
4302 : static inline bool
4303 2044053 : ReorderBufferCanStream(ReorderBuffer *rb)
4304 : {
4305 2044053 : LogicalDecodingContext *ctx = rb->private_data;
4306 :
4307 2044053 : return ctx->streaming;
4308 : }
4309 :
4310 : /* Returns true, if the streaming can be started now, false, otherwise. */
4311 : static inline bool
4312 460480 : ReorderBufferCanStartStreaming(ReorderBuffer *rb)
4313 : {
4314 460480 : LogicalDecodingContext *ctx = rb->private_data;
4315 460480 : SnapBuild *builder = ctx->snapshot_builder;
4316 :
4317 : /* We can't start streaming unless a consistent state is reached. */
4318 460480 : if (SnapBuildCurrentState(builder) < SNAPBUILD_CONSISTENT)
4319 0 : return false;
4320 :
4321 : /*
4322 : * We can't start streaming immediately even if the streaming is enabled
4323 : * because we previously decoded this transaction and now just are
4324 : * restarting.
4325 : */
4326 460480 : if (ReorderBufferCanStream(rb) &&
4327 458039 : !SnapBuildXactNeedsSkip(builder, ctx->reader->ReadRecPtr))
4328 198248 : return true;
4329 :
4330 262232 : return false;
4331 : }
4332 :
4333 : /*
4334 : * Send data of a large transaction (and its subtransactions) to the
4335 : * output plugin, but using the stream API.
4336 : */
4337 : static void
4338 722 : ReorderBufferStreamTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
4339 : {
4340 : Snapshot snapshot_now;
4341 : CommandId command_id;
4342 : Size stream_bytes;
4343 : bool txn_is_streamed;
4344 :
4345 : /* We can never reach here for a subtransaction. */
4346 : Assert(rbtxn_is_toptxn(txn));
4347 :
4348 : /*
4349 : * We can't make any assumptions about base snapshot here, similar to what
4350 : * ReorderBufferCommit() does. That relies on base_snapshot getting
4351 : * transferred from subxact in ReorderBufferCommitChild(), but that was
4352 : * not yet called as the transaction is in-progress.
4353 : *
4354 : * So just walk the subxacts and use the same logic here. But we only need
4355 : * to do that once, when the transaction is streamed for the first time.
4356 : * After that we need to reuse the snapshot from the previous run.
4357 : *
4358 : * Unlike DecodeCommit which adds xids of all the subtransactions in
4359 : * snapshot's xip array via SnapBuildCommitTxn, we can't do that here but
4360 : * we do add them to subxip array instead via ReorderBufferCopySnap. This
4361 : * allows the catalog changes made in subtransactions decoded till now to
4362 : * be visible.
4363 : */
4364 722 : if (txn->snapshot_now == NULL)
4365 : {
4366 : dlist_iter subxact_i;
4367 :
4368 : /* make sure this transaction is streamed for the first time */
4369 : Assert(!rbtxn_is_streamed(txn));
4370 :
4371 : /* at the beginning we should have invalid command ID */
4372 : Assert(txn->command_id == InvalidCommandId);
4373 :
4374 79 : dlist_foreach(subxact_i, &txn->subtxns)
4375 : {
4376 : ReorderBufferTXN *subtxn;
4377 :
4378 4 : subtxn = dlist_container(ReorderBufferTXN, node, subxact_i.cur);
4379 4 : ReorderBufferTransferSnapToParent(txn, subtxn);
4380 : }
4381 :
4382 : /*
4383 : * If this transaction has no snapshot, it didn't make any changes to
4384 : * the database till now, so there's nothing to decode.
4385 : */
4386 75 : if (txn->base_snapshot == NULL)
4387 : {
4388 : Assert(txn->ninvalidations == 0);
4389 0 : return;
4390 : }
4391 :
4392 75 : command_id = FirstCommandId;
4393 75 : snapshot_now = ReorderBufferCopySnap(rb, txn->base_snapshot,
4394 : txn, command_id);
4395 : }
4396 : else
4397 : {
4398 : /* the transaction must have been already streamed */
4399 : Assert(rbtxn_is_streamed(txn));
4400 :
4401 : /*
4402 : * Nah, we already have snapshot from the previous streaming run. We
4403 : * assume new subxacts can't move the LSN backwards, and so can't beat
4404 : * the LSN condition in the previous branch (so no need to walk
4405 : * through subxacts again). In fact, we must not do that as we may be
4406 : * using snapshot half-way through the subxact.
4407 : */
4408 647 : command_id = txn->command_id;
4409 :
4410 : /*
4411 : * We can't use txn->snapshot_now directly because after the last
4412 : * streaming run, we might have got some new sub-transactions. So we
4413 : * need to add them to the snapshot.
4414 : */
4415 647 : snapshot_now = ReorderBufferCopySnap(rb, txn->snapshot_now,
4416 : txn, command_id);
4417 :
4418 : /* Free the previously copied snapshot. */
4419 : Assert(txn->snapshot_now->copied);
4420 647 : ReorderBufferFreeSnap(rb, txn->snapshot_now);
4421 647 : txn->snapshot_now = NULL;
4422 : }
4423 :
4424 : /*
4425 : * Remember this information to be used later to update stats. We can't
4426 : * update the stats here as an error while processing the changes would
4427 : * lead to the accumulation of stats even though we haven't streamed all
4428 : * the changes.
4429 : */
4430 722 : txn_is_streamed = rbtxn_is_streamed(txn);
4431 722 : stream_bytes = txn->total_size;
4432 :
4433 : /* Process and send the changes to output plugin. */
4434 722 : ReorderBufferProcessTXN(rb, txn, InvalidXLogRecPtr, snapshot_now,
4435 : command_id, true);
4436 :
4437 722 : rb->streamCount += 1;
4438 722 : rb->streamBytes += stream_bytes;
4439 :
4440 : /* Don't consider already streamed transaction. */
4441 722 : rb->streamTxns += (txn_is_streamed) ? 0 : 1;
4442 :
4443 : /* update the decoding stats */
4444 722 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
4445 :
4446 : Assert(dlist_is_empty(&txn->changes));
4447 : Assert(txn->nentries == 0);
4448 : Assert(txn->nentries_mem == 0);
4449 : }
4450 :
4451 : /*
4452 : * Size of a change in memory.
4453 : */
4454 : static Size
4455 2222827 : ReorderBufferChangeSize(ReorderBufferChange *change)
4456 : {
4457 2222827 : Size sz = sizeof(ReorderBufferChange);
4458 :
4459 2222827 : switch (change->action)
4460 : {
4461 : /* fall through these, they're all similar enough */
4462 2103356 : case REORDER_BUFFER_CHANGE_INSERT:
4463 : case REORDER_BUFFER_CHANGE_UPDATE:
4464 : case REORDER_BUFFER_CHANGE_DELETE:
4465 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4466 : {
4467 : HeapTuple oldtup,
4468 : newtup;
4469 2103356 : Size oldlen = 0;
4470 2103356 : Size newlen = 0;
4471 :
4472 2103356 : oldtup = change->data.tp.oldtuple;
4473 2103356 : newtup = change->data.tp.newtuple;
4474 :
4475 2103356 : if (oldtup)
4476 : {
4477 231306 : sz += sizeof(HeapTupleData);
4478 231306 : oldlen = oldtup->t_len;
4479 231306 : sz += oldlen;
4480 : }
4481 :
4482 2103356 : if (newtup)
4483 : {
4484 1788441 : sz += sizeof(HeapTupleData);
4485 1788441 : newlen = newtup->t_len;
4486 1788441 : sz += newlen;
4487 : }
4488 :
4489 2103356 : break;
4490 : }
4491 67 : case REORDER_BUFFER_CHANGE_MESSAGE:
4492 : {
4493 67 : Size prefix_size = strlen(change->data.msg.prefix) + 1;
4494 :
4495 67 : sz += prefix_size + change->data.msg.message_size +
4496 : sizeof(Size) + sizeof(Size);
4497 :
4498 67 : break;
4499 : }
4500 11605 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4501 : {
4502 11605 : sz += sizeof(SharedInvalidationMessage) *
4503 11605 : change->data.inval.ninvalidations;
4504 11605 : break;
4505 : }
4506 3226 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4507 : {
4508 : Snapshot snap;
4509 :
4510 3226 : snap = change->data.snapshot;
4511 :
4512 3226 : sz += sizeof(SnapshotData) +
4513 3226 : sizeof(TransactionId) * snap->xcnt +
4514 3226 : sizeof(TransactionId) * snap->subxcnt;
4515 :
4516 3226 : break;
4517 : }
4518 95 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4519 : {
4520 95 : sz += sizeof(Oid) * change->data.truncate.nrelids;
4521 :
4522 95 : break;
4523 : }
4524 104478 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4525 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4526 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4527 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4528 : /* ReorderBufferChange contains everything important */
4529 104478 : break;
4530 : }
4531 :
4532 2222827 : return sz;
4533 : }
4534 :
4535 :
4536 : /*
4537 : * Restore a number of changes spilled to disk back into memory.
4538 : */
4539 : static Size
4540 105 : ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
4541 : TXNEntryFile *file, XLogSegNo *segno)
4542 : {
4543 105 : Size restored = 0;
4544 : XLogSegNo last_segno;
4545 : dlist_mutable_iter cleanup_iter;
4546 105 : File *fd = &file->vfd;
4547 :
4548 : Assert(XLogRecPtrIsValid(txn->first_lsn));
4549 : Assert(XLogRecPtrIsValid(txn->final_lsn));
4550 :
4551 : /* free current entries, so we have memory for more */
4552 174448 : dlist_foreach_modify(cleanup_iter, &txn->changes)
4553 : {
4554 174343 : ReorderBufferChange *cleanup =
4555 174343 : dlist_container(ReorderBufferChange, node, cleanup_iter.cur);
4556 :
4557 174343 : dlist_delete(&cleanup->node);
4558 174343 : ReorderBufferFreeChange(rb, cleanup, true);
4559 : }
4560 105 : txn->nentries_mem = 0;
4561 : Assert(dlist_is_empty(&txn->changes));
4562 :
4563 105 : XLByteToSeg(txn->final_lsn, last_segno, wal_segment_size);
4564 :
4565 178154 : while (restored < max_changes_in_memory && *segno <= last_segno)
4566 : {
4567 : int readBytes;
4568 : ReorderBufferDiskChange *ondisk;
4569 :
4570 178049 : CHECK_FOR_INTERRUPTS();
4571 :
4572 178049 : if (*fd == -1)
4573 : {
4574 : char path[MAXPGPATH];
4575 :
4576 : /* first time in */
4577 43 : if (*segno == 0)
4578 40 : XLByteToSeg(txn->first_lsn, *segno, wal_segment_size);
4579 :
4580 : Assert(*segno != 0 || dlist_is_empty(&txn->changes));
4581 :
4582 : /*
4583 : * No need to care about TLIs here, only used during a single run,
4584 : * so each LSN only maps to a specific WAL record.
4585 : */
4586 43 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid,
4587 : *segno);
4588 :
4589 43 : *fd = PathNameOpenFile(path, O_RDONLY | PG_BINARY);
4590 :
4591 : /* No harm in resetting the offset even in case of failure */
4592 43 : file->curOffset = 0;
4593 :
4594 43 : if (*fd < 0 && errno == ENOENT)
4595 : {
4596 0 : *fd = -1;
4597 0 : (*segno)++;
4598 0 : continue;
4599 : }
4600 43 : else if (*fd < 0)
4601 0 : ereport(ERROR,
4602 : (errcode_for_file_access(),
4603 : errmsg("could not open file \"%s\": %m",
4604 : path)));
4605 : }
4606 :
4607 : /*
4608 : * Read the statically sized part of a change which has information
4609 : * about the total size. If we couldn't read a record, we're at the
4610 : * end of this file.
4611 : */
4612 178049 : ReorderBufferSerializeReserve(rb, sizeof(ReorderBufferDiskChange));
4613 178049 : readBytes = FileRead(file->vfd, rb->outbuf,
4614 : sizeof(ReorderBufferDiskChange),
4615 : file->curOffset, WAIT_EVENT_REORDER_BUFFER_READ);
4616 :
4617 : /* eof */
4618 178049 : if (readBytes == 0)
4619 : {
4620 43 : FileClose(*fd);
4621 43 : *fd = -1;
4622 43 : (*segno)++;
4623 43 : continue;
4624 : }
4625 178006 : else if (readBytes < 0)
4626 0 : ereport(ERROR,
4627 : (errcode_for_file_access(),
4628 : errmsg("could not read from reorderbuffer spill file: %m")));
4629 178006 : else if (readBytes != sizeof(ReorderBufferDiskChange))
4630 0 : ereport(ERROR,
4631 : (errcode_for_file_access(),
4632 : errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
4633 : readBytes,
4634 : (uint32) sizeof(ReorderBufferDiskChange))));
4635 :
4636 178006 : file->curOffset += readBytes;
4637 :
4638 178006 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4639 :
4640 178006 : ReorderBufferSerializeReserve(rb,
4641 178006 : sizeof(ReorderBufferDiskChange) + ondisk->size);
4642 178006 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4643 :
4644 356012 : readBytes = FileRead(file->vfd,
4645 178006 : rb->outbuf + sizeof(ReorderBufferDiskChange),
4646 178006 : ondisk->size - sizeof(ReorderBufferDiskChange),
4647 : file->curOffset,
4648 : WAIT_EVENT_REORDER_BUFFER_READ);
4649 :
4650 178006 : if (readBytes < 0)
4651 0 : ereport(ERROR,
4652 : (errcode_for_file_access(),
4653 : errmsg("could not read from reorderbuffer spill file: %m")));
4654 178006 : else if (readBytes != ondisk->size - sizeof(ReorderBufferDiskChange))
4655 0 : ereport(ERROR,
4656 : (errcode_for_file_access(),
4657 : errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
4658 : readBytes,
4659 : (uint32) (ondisk->size - sizeof(ReorderBufferDiskChange)))));
4660 :
4661 178006 : file->curOffset += readBytes;
4662 :
4663 : /*
4664 : * ok, read a full change from disk, now restore it into proper
4665 : * in-memory format
4666 : */
4667 178006 : ReorderBufferRestoreChange(rb, txn, rb->outbuf);
4668 178006 : restored++;
4669 : }
4670 :
4671 105 : return restored;
4672 : }
4673 :
4674 : /*
4675 : * Convert change from its on-disk format to in-memory format and queue it onto
4676 : * the TXN's ->changes list.
4677 : *
4678 : * Note: although "data" is declared char*, at entry it points to a
4679 : * maxalign'd buffer, making it safe in most of this function to assume
4680 : * that the pointed-to data is suitably aligned for direct access.
4681 : */
4682 : static void
4683 178006 : ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
4684 : char *data)
4685 : {
4686 : ReorderBufferDiskChange *ondisk;
4687 : ReorderBufferChange *change;
4688 :
4689 178006 : ondisk = (ReorderBufferDiskChange *) data;
4690 :
4691 178006 : change = ReorderBufferAllocChange(rb);
4692 :
4693 : /* copy static part */
4694 178006 : memcpy(change, &ondisk->change, sizeof(ReorderBufferChange));
4695 :
4696 178006 : data += sizeof(ReorderBufferDiskChange);
4697 :
4698 : /* restore individual stuff */
4699 178006 : switch (change->action)
4700 : {
4701 : /* fall through these, they're all similar enough */
4702 176077 : case REORDER_BUFFER_CHANGE_INSERT:
4703 : case REORDER_BUFFER_CHANGE_UPDATE:
4704 : case REORDER_BUFFER_CHANGE_DELETE:
4705 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4706 176077 : if (change->data.tp.oldtuple)
4707 : {
4708 5006 : uint32 tuplelen = ((HeapTuple) data)->t_len;
4709 :
4710 5006 : change->data.tp.oldtuple =
4711 5006 : ReorderBufferAllocTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
4712 :
4713 : /* restore ->tuple */
4714 5006 : memcpy(change->data.tp.oldtuple, data,
4715 : sizeof(HeapTupleData));
4716 5006 : data += sizeof(HeapTupleData);
4717 :
4718 : /* reset t_data pointer into the new tuplebuf */
4719 5006 : change->data.tp.oldtuple->t_data =
4720 5006 : (HeapTupleHeader) ((char *) change->data.tp.oldtuple + HEAPTUPLESIZE);
4721 :
4722 : /* restore tuple data itself */
4723 5006 : memcpy(change->data.tp.oldtuple->t_data, data, tuplelen);
4724 5006 : data += tuplelen;
4725 : }
4726 :
4727 176077 : if (change->data.tp.newtuple)
4728 : {
4729 : /* here, data might not be suitably aligned! */
4730 : uint32 tuplelen;
4731 :
4732 165856 : memcpy(&tuplelen, data + offsetof(HeapTupleData, t_len),
4733 : sizeof(uint32));
4734 :
4735 165856 : change->data.tp.newtuple =
4736 165856 : ReorderBufferAllocTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
4737 :
4738 : /* restore ->tuple */
4739 165856 : memcpy(change->data.tp.newtuple, data,
4740 : sizeof(HeapTupleData));
4741 165856 : data += sizeof(HeapTupleData);
4742 :
4743 : /* reset t_data pointer into the new tuplebuf */
4744 165856 : change->data.tp.newtuple->t_data =
4745 165856 : (HeapTupleHeader) ((char *) change->data.tp.newtuple + HEAPTUPLESIZE);
4746 :
4747 : /* restore tuple data itself */
4748 165856 : memcpy(change->data.tp.newtuple->t_data, data, tuplelen);
4749 165856 : data += tuplelen;
4750 : }
4751 :
4752 176077 : break;
4753 1 : case REORDER_BUFFER_CHANGE_MESSAGE:
4754 : {
4755 : Size prefix_size;
4756 :
4757 : /* read prefix */
4758 1 : memcpy(&prefix_size, data, sizeof(Size));
4759 1 : data += sizeof(Size);
4760 1 : change->data.msg.prefix = MemoryContextAlloc(rb->context,
4761 : prefix_size);
4762 1 : memcpy(change->data.msg.prefix, data, prefix_size);
4763 : Assert(change->data.msg.prefix[prefix_size - 1] == '\0');
4764 1 : data += prefix_size;
4765 :
4766 : /* read the message */
4767 1 : memcpy(&change->data.msg.message_size, data, sizeof(Size));
4768 1 : data += sizeof(Size);
4769 1 : change->data.msg.message = MemoryContextAlloc(rb->context,
4770 : change->data.msg.message_size);
4771 1 : memcpy(change->data.msg.message, data,
4772 : change->data.msg.message_size);
4773 1 : data += change->data.msg.message_size;
4774 :
4775 1 : break;
4776 : }
4777 23 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4778 : {
4779 23 : Size inval_size = sizeof(SharedInvalidationMessage) *
4780 23 : change->data.inval.ninvalidations;
4781 :
4782 23 : change->data.inval.invalidations =
4783 23 : MemoryContextAlloc(rb->context, inval_size);
4784 :
4785 : /* read the message */
4786 23 : memcpy(change->data.inval.invalidations, data, inval_size);
4787 :
4788 23 : break;
4789 : }
4790 2 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4791 : {
4792 : Snapshot oldsnap;
4793 : Snapshot newsnap;
4794 : Size size;
4795 :
4796 2 : oldsnap = (Snapshot) data;
4797 :
4798 2 : size = sizeof(SnapshotData) +
4799 2 : sizeof(TransactionId) * oldsnap->xcnt +
4800 2 : sizeof(TransactionId) * (oldsnap->subxcnt + 0);
4801 :
4802 2 : change->data.snapshot = MemoryContextAllocZero(rb->context, size);
4803 :
4804 2 : newsnap = change->data.snapshot;
4805 :
4806 2 : memcpy(newsnap, data, size);
4807 2 : newsnap->xip = (TransactionId *)
4808 : (((char *) newsnap) + sizeof(SnapshotData));
4809 2 : newsnap->subxip = newsnap->xip + newsnap->xcnt;
4810 2 : newsnap->copied = true;
4811 2 : break;
4812 : }
4813 : /* the base struct contains all the data, easy peasy */
4814 0 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4815 : {
4816 : Oid *relids;
4817 :
4818 0 : relids = ReorderBufferAllocRelids(rb, change->data.truncate.nrelids);
4819 0 : memcpy(relids, data, change->data.truncate.nrelids * sizeof(Oid));
4820 0 : change->data.truncate.relids = relids;
4821 :
4822 0 : break;
4823 : }
4824 1903 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4825 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4826 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4827 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4828 1903 : break;
4829 : }
4830 :
4831 178006 : dlist_push_tail(&txn->changes, &change->node);
4832 178006 : txn->nentries_mem++;
4833 :
4834 : /*
4835 : * Update memory accounting for the restored change. We need to do this
4836 : * although we don't check the memory limit when restoring the changes in
4837 : * this branch (we only do that when initially queueing the changes after
4838 : * decoding), because we will release the changes later, and that will
4839 : * update the accounting too (subtracting the size from the counters). And
4840 : * we don't want to underflow there.
4841 : */
4842 178006 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
4843 : ReorderBufferChangeSize(change));
4844 178006 : }
4845 :
4846 : /*
4847 : * Remove all on-disk stored for the passed in transaction.
4848 : */
4849 : static void
4850 347 : ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn)
4851 : {
4852 : XLogSegNo first;
4853 : XLogSegNo cur;
4854 : XLogSegNo last;
4855 :
4856 : Assert(XLogRecPtrIsValid(txn->first_lsn));
4857 : Assert(XLogRecPtrIsValid(txn->final_lsn));
4858 :
4859 347 : XLByteToSeg(txn->first_lsn, first, wal_segment_size);
4860 347 : XLByteToSeg(txn->final_lsn, last, wal_segment_size);
4861 :
4862 : /* iterate over all possible filenames, and delete them */
4863 712 : for (cur = first; cur <= last; cur++)
4864 : {
4865 : char path[MAXPGPATH];
4866 :
4867 365 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid, cur);
4868 365 : if (unlink(path) != 0 && errno != ENOENT)
4869 0 : ereport(ERROR,
4870 : (errcode_for_file_access(),
4871 : errmsg("could not remove file \"%s\": %m", path)));
4872 : }
4873 347 : }
4874 :
4875 : /*
4876 : * Remove any leftover serialized reorder buffers from a slot directory after a
4877 : * prior crash or decoding session exit.
4878 : */
4879 : static void
4880 2303 : ReorderBufferCleanupSerializedTXNs(const char *slotname)
4881 : {
4882 : DIR *spill_dir;
4883 : struct dirent *spill_de;
4884 : struct stat statbuf;
4885 : char path[MAXPGPATH * 2 + sizeof(PG_REPLSLOT_DIR)];
4886 :
4887 2303 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, slotname);
4888 :
4889 : /* we're only handling directories here, skip if it's not ours */
4890 2303 : if (lstat(path, &statbuf) == 0 && !S_ISDIR(statbuf.st_mode))
4891 0 : return;
4892 :
4893 2303 : spill_dir = AllocateDir(path);
4894 11515 : while ((spill_de = ReadDirExtended(spill_dir, path, INFO)) != NULL)
4895 : {
4896 : /* only look at names that can be ours */
4897 6909 : if (strncmp(spill_de->d_name, "xid", 3) == 0)
4898 : {
4899 0 : snprintf(path, sizeof(path),
4900 : "%s/%s/%s", PG_REPLSLOT_DIR, slotname,
4901 0 : spill_de->d_name);
4902 :
4903 0 : if (unlink(path) != 0)
4904 0 : ereport(ERROR,
4905 : (errcode_for_file_access(),
4906 : errmsg("could not remove file \"%s\" during removal of %s/%s/xid*: %m",
4907 : path, PG_REPLSLOT_DIR, slotname)));
4908 : }
4909 : }
4910 2303 : FreeDir(spill_dir);
4911 : }
4912 :
4913 : /*
4914 : * Given a replication slot, transaction ID and segment number, fill in the
4915 : * corresponding spill file into 'path', which is a caller-owned buffer of size
4916 : * at least MAXPGPATH.
4917 : */
4918 : static void
4919 4474 : ReorderBufferSerializedPath(char *path, ReplicationSlot *slot, TransactionId xid,
4920 : XLogSegNo segno)
4921 : {
4922 : XLogRecPtr recptr;
4923 :
4924 4474 : XLogSegNoOffsetToRecPtr(segno, 0, wal_segment_size, recptr);
4925 :
4926 4474 : snprintf(path, MAXPGPATH, "%s/%s/xid-%u-lsn-%X-%X.spill",
4927 : PG_REPLSLOT_DIR,
4928 4474 : NameStr(MyReplicationSlot->data.name),
4929 4474 : xid, LSN_FORMAT_ARGS(recptr));
4930 4474 : }
4931 :
4932 : /*
4933 : * Delete all data spilled to disk after we've restarted/crashed. It will be
4934 : * recreated when the respective slots are reused.
4935 : */
4936 : void
4937 1082 : StartupReorderBuffer(void)
4938 : {
4939 : DIR *logical_dir;
4940 : struct dirent *logical_de;
4941 :
4942 1082 : logical_dir = AllocateDir(PG_REPLSLOT_DIR);
4943 3373 : while ((logical_de = ReadDir(logical_dir, PG_REPLSLOT_DIR)) != NULL)
4944 : {
4945 2291 : if (strcmp(logical_de->d_name, ".") == 0 ||
4946 1209 : strcmp(logical_de->d_name, "..") == 0)
4947 2164 : continue;
4948 :
4949 : /* if it cannot be a slot, skip the directory */
4950 127 : if (!ReplicationSlotValidateName(logical_de->d_name, true, DEBUG2))
4951 0 : continue;
4952 :
4953 : /*
4954 : * ok, has to be a surviving logical slot, iterate and delete
4955 : * everything starting with xid-*
4956 : */
4957 127 : ReorderBufferCleanupSerializedTXNs(logical_de->d_name);
4958 : }
4959 1082 : FreeDir(logical_dir);
4960 1082 : }
4961 :
4962 : /* ---------------------------------------
4963 : * toast reassembly support
4964 : * ---------------------------------------
4965 : */
4966 :
4967 : /*
4968 : * Initialize per tuple toast reconstruction support.
4969 : */
4970 : static void
4971 44 : ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
4972 : {
4973 : HASHCTL hash_ctl;
4974 :
4975 : Assert(txn->toast_hash == NULL);
4976 :
4977 44 : hash_ctl.keysize = sizeof(Oid);
4978 44 : hash_ctl.entrysize = sizeof(ReorderBufferToastEnt);
4979 44 : hash_ctl.hcxt = rb->context;
4980 44 : txn->toast_hash = hash_create("ReorderBufferToastHash", 5, &hash_ctl,
4981 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
4982 44 : }
4983 :
4984 : /*
4985 : * Per toast-chunk handling for toast reconstruction
4986 : *
4987 : * Appends a toast chunk so we can reconstruct it when the tuple "owning" the
4988 : * toasted Datum comes along.
4989 : */
4990 : static void
4991 1847 : ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
4992 : Relation relation, ReorderBufferChange *change)
4993 : {
4994 : ReorderBufferToastEnt *ent;
4995 : HeapTuple newtup;
4996 : bool found;
4997 : int32 chunksize;
4998 : bool isnull;
4999 : Pointer chunk;
5000 1847 : TupleDesc desc = RelationGetDescr(relation);
5001 : Oid chunk_id;
5002 : int32 chunk_seq;
5003 :
5004 1847 : if (txn->toast_hash == NULL)
5005 44 : ReorderBufferToastInitHash(rb, txn);
5006 :
5007 : Assert(IsToastRelation(relation));
5008 :
5009 1847 : newtup = change->data.tp.newtuple;
5010 1847 : chunk_id = DatumGetObjectId(fastgetattr(newtup, 1, desc, &isnull));
5011 : Assert(!isnull);
5012 1847 : chunk_seq = DatumGetInt32(fastgetattr(newtup, 2, desc, &isnull));
5013 : Assert(!isnull);
5014 :
5015 : ent = (ReorderBufferToastEnt *)
5016 1847 : hash_search(txn->toast_hash, &chunk_id, HASH_ENTER, &found);
5017 :
5018 1847 : if (!found)
5019 : {
5020 : Assert(ent->chunk_id == chunk_id);
5021 62 : ent->num_chunks = 0;
5022 62 : ent->last_chunk_seq = 0;
5023 62 : ent->size = 0;
5024 62 : ent->reconstructed = NULL;
5025 62 : dlist_init(&ent->chunks);
5026 :
5027 62 : if (chunk_seq != 0)
5028 0 : elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq 0",
5029 : chunk_seq, chunk_id);
5030 : }
5031 1785 : else if (found && chunk_seq != ent->last_chunk_seq + 1)
5032 0 : elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq %d",
5033 : chunk_seq, chunk_id, ent->last_chunk_seq + 1);
5034 :
5035 1847 : chunk = DatumGetPointer(fastgetattr(newtup, 3, desc, &isnull));
5036 : Assert(!isnull);
5037 :
5038 : /* calculate size so we can allocate the right size at once later */
5039 1847 : if (!VARATT_IS_EXTENDED(chunk))
5040 1847 : chunksize = VARSIZE(chunk) - VARHDRSZ;
5041 0 : else if (VARATT_IS_SHORT(chunk))
5042 : /* could happen due to heap_form_tuple doing its thing */
5043 0 : chunksize = VARSIZE_SHORT(chunk) - VARHDRSZ_SHORT;
5044 : else
5045 0 : elog(ERROR, "unexpected type of toast chunk");
5046 :
5047 1847 : ent->size += chunksize;
5048 1847 : ent->last_chunk_seq = chunk_seq;
5049 1847 : ent->num_chunks++;
5050 1847 : dlist_push_tail(&ent->chunks, &change->node);
5051 1847 : }
5052 :
5053 : /*
5054 : * Rejigger change->newtuple to point to in-memory toast tuples instead of
5055 : * on-disk toast tuples that may no longer exist (think DROP TABLE or VACUUM).
5056 : *
5057 : * We cannot replace unchanged toast tuples though, so those will still point
5058 : * to on-disk toast data.
5059 : *
5060 : * While updating the existing change with detoasted tuple data, we need to
5061 : * update the memory accounting info, because the change size will differ.
5062 : * Otherwise the accounting may get out of sync, triggering serialization
5063 : * at unexpected times.
5064 : *
5065 : * We simply subtract size of the change before rejiggering the tuple, and
5066 : * then add the new size. This makes it look like the change was removed
5067 : * and then added back, except it only tweaks the accounting info.
5068 : *
5069 : * In particular it can't trigger serialization, which would be pointless
5070 : * anyway as it happens during commit processing right before handing
5071 : * the change to the output plugin.
5072 : */
5073 : static void
5074 355505 : ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
5075 : Relation relation, ReorderBufferChange *change)
5076 : {
5077 : TupleDesc desc;
5078 : int natt;
5079 : Datum *attrs;
5080 : bool *isnull;
5081 : bool *free;
5082 : HeapTuple tmphtup;
5083 : Relation toast_rel;
5084 : TupleDesc toast_desc;
5085 : MemoryContext oldcontext;
5086 : HeapTuple newtup;
5087 : Size old_size;
5088 :
5089 : /* no toast tuples changed */
5090 355505 : if (txn->toast_hash == NULL)
5091 355250 : return;
5092 :
5093 : /*
5094 : * We're going to modify the size of the change. So, to make sure the
5095 : * accounting is correct we record the current change size and then after
5096 : * re-computing the change we'll subtract the recorded size and then
5097 : * re-add the new change size at the end. We don't immediately subtract
5098 : * the old size because if there is any error before we add the new size,
5099 : * we will release the changes and that will update the accounting info
5100 : * (subtracting the size from the counters). And we don't want to
5101 : * underflow there.
5102 : */
5103 255 : old_size = ReorderBufferChangeSize(change);
5104 :
5105 255 : oldcontext = MemoryContextSwitchTo(rb->context);
5106 :
5107 : /* we should only have toast tuples in an INSERT or UPDATE */
5108 : Assert(change->data.tp.newtuple);
5109 :
5110 255 : desc = RelationGetDescr(relation);
5111 :
5112 255 : toast_rel = RelationIdGetRelation(relation->rd_rel->reltoastrelid);
5113 255 : if (!RelationIsValid(toast_rel))
5114 0 : elog(ERROR, "could not open toast relation with OID %u (base relation \"%s\")",
5115 : relation->rd_rel->reltoastrelid, RelationGetRelationName(relation));
5116 :
5117 255 : toast_desc = RelationGetDescr(toast_rel);
5118 :
5119 : /* should we allocate from stack instead? */
5120 255 : attrs = palloc0_array(Datum, desc->natts);
5121 255 : isnull = palloc0_array(bool, desc->natts);
5122 255 : free = palloc0_array(bool, desc->natts);
5123 :
5124 255 : newtup = change->data.tp.newtuple;
5125 :
5126 255 : heap_deform_tuple(newtup, desc, attrs, isnull);
5127 :
5128 815 : for (natt = 0; natt < desc->natts; natt++)
5129 : {
5130 560 : CompactAttribute *attr = TupleDescCompactAttr(desc, natt);
5131 : ReorderBufferToastEnt *ent;
5132 : varlena *varlena_pointer;
5133 :
5134 : /* va_rawsize is the size of the original datum -- including header */
5135 : varatt_external toast_pointer;
5136 : varatt_indirect redirect_pointer;
5137 560 : varlena *new_datum = NULL;
5138 : varlena *reconstructed;
5139 : dlist_iter it;
5140 560 : Size data_done = 0;
5141 :
5142 560 : if (attr->attisdropped)
5143 499 : continue;
5144 :
5145 : /* not a varlena datatype */
5146 538 : if (attr->attlen != -1)
5147 249 : continue;
5148 :
5149 : /* no data */
5150 289 : if (isnull[natt])
5151 12 : continue;
5152 :
5153 : /* ok, we know we have a toast datum */
5154 277 : varlena_pointer = (varlena *) DatumGetPointer(attrs[natt]);
5155 :
5156 : /* no need to do anything if the tuple isn't external */
5157 277 : if (!VARATT_IS_EXTERNAL(varlena_pointer))
5158 208 : continue;
5159 :
5160 69 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, varlena_pointer);
5161 :
5162 : /*
5163 : * Check whether the toast tuple changed, replace if so.
5164 : */
5165 : ent = (ReorderBufferToastEnt *)
5166 69 : hash_search(txn->toast_hash,
5167 : &toast_pointer.va_valueid,
5168 : HASH_FIND,
5169 : NULL);
5170 69 : if (ent == NULL)
5171 8 : continue;
5172 :
5173 : new_datum =
5174 61 : (varlena *) palloc0(INDIRECT_POINTER_SIZE);
5175 :
5176 61 : free[natt] = true;
5177 :
5178 61 : reconstructed = palloc0(toast_pointer.va_rawsize);
5179 :
5180 61 : ent->reconstructed = reconstructed;
5181 :
5182 : /* stitch toast tuple back together from its parts */
5183 1857 : dlist_foreach(it, &ent->chunks)
5184 : {
5185 : bool cisnull;
5186 : ReorderBufferChange *cchange;
5187 : HeapTuple ctup;
5188 : Pointer chunk;
5189 :
5190 1796 : cchange = dlist_container(ReorderBufferChange, node, it.cur);
5191 1796 : ctup = cchange->data.tp.newtuple;
5192 1796 : chunk = DatumGetPointer(fastgetattr(ctup, 3, toast_desc, &cisnull));
5193 :
5194 : Assert(!cisnull);
5195 : Assert(!VARATT_IS_EXTERNAL(chunk));
5196 : Assert(!VARATT_IS_SHORT(chunk));
5197 :
5198 1796 : memcpy(VARDATA(reconstructed) + data_done,
5199 1796 : VARDATA(chunk),
5200 1796 : VARSIZE(chunk) - VARHDRSZ);
5201 1796 : data_done += VARSIZE(chunk) - VARHDRSZ;
5202 : }
5203 : Assert(data_done == VARATT_EXTERNAL_GET_EXTSIZE(toast_pointer));
5204 :
5205 : /* make sure its marked as compressed or not */
5206 61 : if (VARATT_EXTERNAL_IS_COMPRESSED(toast_pointer))
5207 16 : SET_VARSIZE_COMPRESSED(reconstructed, data_done + VARHDRSZ);
5208 : else
5209 45 : SET_VARSIZE(reconstructed, data_done + VARHDRSZ);
5210 :
5211 61 : memset(&redirect_pointer, 0, sizeof(redirect_pointer));
5212 61 : redirect_pointer.pointer = reconstructed;
5213 :
5214 61 : SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
5215 61 : memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
5216 : sizeof(redirect_pointer));
5217 :
5218 61 : attrs[natt] = PointerGetDatum(new_datum);
5219 : }
5220 :
5221 : /*
5222 : * Build tuple in separate memory & copy tuple back into the tuplebuf
5223 : * passed to the output plugin. We can't directly heap_fill_tuple() into
5224 : * the tuplebuf because attrs[] will point back into the current content.
5225 : */
5226 255 : tmphtup = heap_form_tuple(desc, attrs, isnull);
5227 : Assert(newtup->t_len <= MaxHeapTupleSize);
5228 : Assert(newtup->t_data == (HeapTupleHeader) ((char *) newtup + HEAPTUPLESIZE));
5229 :
5230 255 : memcpy(newtup->t_data, tmphtup->t_data, tmphtup->t_len);
5231 255 : newtup->t_len = tmphtup->t_len;
5232 :
5233 : /*
5234 : * free resources we won't further need, more persistent stuff will be
5235 : * free'd in ReorderBufferToastReset().
5236 : */
5237 255 : RelationClose(toast_rel);
5238 255 : pfree(tmphtup);
5239 815 : for (natt = 0; natt < desc->natts; natt++)
5240 : {
5241 560 : if (free[natt])
5242 61 : pfree(DatumGetPointer(attrs[natt]));
5243 : }
5244 255 : pfree(attrs);
5245 255 : pfree(free);
5246 255 : pfree(isnull);
5247 :
5248 255 : MemoryContextSwitchTo(oldcontext);
5249 :
5250 : /* subtract the old change size */
5251 255 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, false, old_size);
5252 : /* now add the change back, with the correct size */
5253 255 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
5254 : ReorderBufferChangeSize(change));
5255 : }
5256 :
5257 : /*
5258 : * Free all resources allocated for toast reconstruction.
5259 : */
5260 : static void
5261 360076 : ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn)
5262 : {
5263 : HASH_SEQ_STATUS hstat;
5264 : ReorderBufferToastEnt *ent;
5265 :
5266 360076 : if (txn->toast_hash == NULL)
5267 360032 : return;
5268 :
5269 : /* sequentially walk over the hash and free everything */
5270 44 : hash_seq_init(&hstat, txn->toast_hash);
5271 106 : while ((ent = (ReorderBufferToastEnt *) hash_seq_search(&hstat)) != NULL)
5272 : {
5273 : dlist_mutable_iter it;
5274 :
5275 62 : if (ent->reconstructed != NULL)
5276 61 : pfree(ent->reconstructed);
5277 :
5278 1909 : dlist_foreach_modify(it, &ent->chunks)
5279 : {
5280 1847 : ReorderBufferChange *change =
5281 1847 : dlist_container(ReorderBufferChange, node, it.cur);
5282 :
5283 1847 : dlist_delete(&change->node);
5284 1847 : ReorderBufferFreeChange(rb, change, true);
5285 : }
5286 : }
5287 :
5288 44 : hash_destroy(txn->toast_hash);
5289 44 : txn->toast_hash = NULL;
5290 : }
5291 :
5292 :
5293 : /* ---------------------------------------
5294 : * Visibility support for logical decoding
5295 : *
5296 : *
5297 : * Lookup actual cmin/cmax values when using decoding snapshot. We can't
5298 : * always rely on stored cmin/cmax values because of two scenarios:
5299 : *
5300 : * * A tuple got changed multiple times during a single transaction and thus
5301 : * has got a combo CID. Combo CIDs are only valid for the duration of a
5302 : * single transaction.
5303 : * * A tuple with a cmin but no cmax (and thus no combo CID) got
5304 : * deleted/updated in another transaction than the one which created it
5305 : * which we are looking at right now. As only one of cmin, cmax or combo CID
5306 : * is actually stored in the heap we don't have access to the value we
5307 : * need anymore.
5308 : *
5309 : * To resolve those problems we have a per-transaction hash of (cmin,
5310 : * cmax) tuples keyed by (relfilelocator, ctid) which contains the actual
5311 : * (cmin, cmax) values. That also takes care of combo CIDs by simply
5312 : * not caring about them at all. As we have the real cmin/cmax values
5313 : * combo CIDs aren't interesting.
5314 : *
5315 : * As we only care about catalog tuples here the overhead of this
5316 : * hashtable should be acceptable.
5317 : *
5318 : * Heap rewrites complicate this a bit, check rewriteheap.c for
5319 : * details.
5320 : * -------------------------------------------------------------------------
5321 : */
5322 :
5323 : /* struct for sorting mapping files by LSN efficiently */
5324 : typedef struct RewriteMappingFile
5325 : {
5326 : XLogRecPtr lsn;
5327 : char fname[MAXPGPATH];
5328 : } RewriteMappingFile;
5329 :
5330 : #ifdef NOT_USED
5331 : static void
5332 : DisplayMapping(HTAB *tuplecid_data)
5333 : {
5334 : HASH_SEQ_STATUS hstat;
5335 : ReorderBufferTupleCidEnt *ent;
5336 :
5337 : hash_seq_init(&hstat, tuplecid_data);
5338 : while ((ent = (ReorderBufferTupleCidEnt *) hash_seq_search(&hstat)) != NULL)
5339 : {
5340 : elog(DEBUG3, "mapping: node: %u/%u/%u tid: %u/%u cmin: %u, cmax: %u",
5341 : ent->key.rlocator.dbOid,
5342 : ent->key.rlocator.spcOid,
5343 : ent->key.rlocator.relNumber,
5344 : ItemPointerGetBlockNumber(&ent->key.tid),
5345 : ItemPointerGetOffsetNumber(&ent->key.tid),
5346 : ent->cmin,
5347 : ent->cmax
5348 : );
5349 : }
5350 : }
5351 : #endif
5352 :
5353 : /*
5354 : * Apply a single mapping file to tuplecid_data.
5355 : *
5356 : * The mapping file has to have been verified to be a) committed b) for our
5357 : * transaction c) applied in LSN order.
5358 : */
5359 : static void
5360 27 : ApplyLogicalMappingFile(HTAB *tuplecid_data, const char *fname)
5361 : {
5362 : char path[MAXPGPATH];
5363 : int fd;
5364 : int readBytes;
5365 : LogicalRewriteMappingData map;
5366 :
5367 27 : sprintf(path, "%s/%s", PG_LOGICAL_MAPPINGS_DIR, fname);
5368 27 : fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
5369 27 : if (fd < 0)
5370 0 : ereport(ERROR,
5371 : (errcode_for_file_access(),
5372 : errmsg("could not open file \"%s\": %m", path)));
5373 :
5374 : while (true)
5375 209 : {
5376 : ReorderBufferTupleCidKey key;
5377 : ReorderBufferTupleCidEnt *ent;
5378 : ReorderBufferTupleCidEnt *new_ent;
5379 : bool found;
5380 :
5381 : /* be careful about padding */
5382 236 : memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
5383 :
5384 : /* read all mappings till the end of the file */
5385 236 : pgstat_report_wait_start(WAIT_EVENT_REORDER_LOGICAL_MAPPING_READ);
5386 236 : readBytes = read(fd, &map, sizeof(LogicalRewriteMappingData));
5387 236 : pgstat_report_wait_end();
5388 :
5389 236 : if (readBytes < 0)
5390 0 : ereport(ERROR,
5391 : (errcode_for_file_access(),
5392 : errmsg("could not read file \"%s\": %m",
5393 : path)));
5394 236 : else if (readBytes == 0) /* EOF */
5395 27 : break;
5396 209 : else if (readBytes != sizeof(LogicalRewriteMappingData))
5397 0 : ereport(ERROR,
5398 : (errcode_for_file_access(),
5399 : errmsg("could not read from file \"%s\": read %d instead of %d bytes",
5400 : path, readBytes,
5401 : (int32) sizeof(LogicalRewriteMappingData))));
5402 :
5403 209 : key.rlocator = map.old_locator;
5404 209 : ItemPointerCopy(&map.old_tid,
5405 : &key.tid);
5406 :
5407 :
5408 : ent = (ReorderBufferTupleCidEnt *)
5409 209 : hash_search(tuplecid_data, &key, HASH_FIND, NULL);
5410 :
5411 : /* no existing mapping, no need to update */
5412 209 : if (!ent)
5413 0 : continue;
5414 :
5415 209 : key.rlocator = map.new_locator;
5416 209 : ItemPointerCopy(&map.new_tid,
5417 : &key.tid);
5418 :
5419 : new_ent = (ReorderBufferTupleCidEnt *)
5420 209 : hash_search(tuplecid_data, &key, HASH_ENTER, &found);
5421 :
5422 209 : if (found)
5423 : {
5424 : /*
5425 : * Make sure the existing mapping makes sense. We sometime update
5426 : * old records that did not yet have a cmax (e.g. pg_class' own
5427 : * entry while rewriting it) during rewrites, so allow that.
5428 : */
5429 : Assert(ent->cmin == InvalidCommandId || ent->cmin == new_ent->cmin);
5430 : Assert(ent->cmax == InvalidCommandId || ent->cmax == new_ent->cmax);
5431 : }
5432 : else
5433 : {
5434 : /* update mapping */
5435 203 : new_ent->cmin = ent->cmin;
5436 203 : new_ent->cmax = ent->cmax;
5437 203 : new_ent->combocid = ent->combocid;
5438 : }
5439 : }
5440 :
5441 27 : if (CloseTransientFile(fd) != 0)
5442 0 : ereport(ERROR,
5443 : (errcode_for_file_access(),
5444 : errmsg("could not close file \"%s\": %m", path)));
5445 27 : }
5446 :
5447 :
5448 : /*
5449 : * Check whether the TransactionId 'xid' is in the pre-sorted array 'xip'.
5450 : */
5451 : static bool
5452 348 : TransactionIdInArray(TransactionId xid, TransactionId *xip, Size num)
5453 : {
5454 348 : return bsearch(&xid, xip, num,
5455 348 : sizeof(TransactionId), xidComparator) != NULL;
5456 : }
5457 :
5458 : /*
5459 : * list_sort() comparator for sorting RewriteMappingFiles in LSN order.
5460 : */
5461 : static int
5462 40 : file_sort_by_lsn(const ListCell *a_p, const ListCell *b_p)
5463 : {
5464 40 : RewriteMappingFile *a = (RewriteMappingFile *) lfirst(a_p);
5465 40 : RewriteMappingFile *b = (RewriteMappingFile *) lfirst(b_p);
5466 :
5467 40 : return pg_cmp_u64(a->lsn, b->lsn);
5468 : }
5469 :
5470 : /*
5471 : * Apply any existing logical remapping files if there are any targeted at our
5472 : * transaction for relid.
5473 : */
5474 : static void
5475 11 : UpdateLogicalMappings(HTAB *tuplecid_data, Oid relid, Snapshot snapshot)
5476 : {
5477 : DIR *mapping_dir;
5478 : struct dirent *mapping_de;
5479 11 : List *files = NIL;
5480 : ListCell *file;
5481 11 : Oid dboid = IsSharedRelation(relid) ? InvalidOid : MyDatabaseId;
5482 :
5483 11 : mapping_dir = AllocateDir(PG_LOGICAL_MAPPINGS_DIR);
5484 573 : while ((mapping_de = ReadDir(mapping_dir, PG_LOGICAL_MAPPINGS_DIR)) != NULL)
5485 : {
5486 : Oid f_dboid;
5487 : Oid f_relid;
5488 : TransactionId f_mapped_xid;
5489 : TransactionId f_create_xid;
5490 : XLogRecPtr f_lsn;
5491 : uint32 f_hi,
5492 : f_lo;
5493 : RewriteMappingFile *f;
5494 :
5495 562 : if (strcmp(mapping_de->d_name, ".") == 0 ||
5496 551 : strcmp(mapping_de->d_name, "..") == 0)
5497 535 : continue;
5498 :
5499 : /* Ignore files that aren't ours */
5500 540 : if (strncmp(mapping_de->d_name, "map-", 4) != 0)
5501 0 : continue;
5502 :
5503 540 : if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
5504 : &f_dboid, &f_relid, &f_hi, &f_lo,
5505 : &f_mapped_xid, &f_create_xid) != 6)
5506 0 : elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
5507 :
5508 540 : f_lsn = ((uint64) f_hi) << 32 | f_lo;
5509 :
5510 : /* mapping for another database */
5511 540 : if (f_dboid != dboid)
5512 0 : continue;
5513 :
5514 : /* mapping for another relation */
5515 540 : if (f_relid != relid)
5516 60 : continue;
5517 :
5518 : /* did the creating transaction abort? */
5519 480 : if (!TransactionIdDidCommit(f_create_xid))
5520 132 : continue;
5521 :
5522 : /* not for our transaction */
5523 348 : if (!TransactionIdInArray(f_mapped_xid, snapshot->subxip, snapshot->subxcnt))
5524 321 : continue;
5525 :
5526 : /* ok, relevant, queue for apply */
5527 27 : f = palloc_object(RewriteMappingFile);
5528 27 : f->lsn = f_lsn;
5529 27 : strcpy(f->fname, mapping_de->d_name);
5530 27 : files = lappend(files, f);
5531 : }
5532 11 : FreeDir(mapping_dir);
5533 :
5534 : /* sort files so we apply them in LSN order */
5535 11 : list_sort(files, file_sort_by_lsn);
5536 :
5537 38 : foreach(file, files)
5538 : {
5539 27 : RewriteMappingFile *f = (RewriteMappingFile *) lfirst(file);
5540 :
5541 27 : elog(DEBUG1, "applying mapping: \"%s\" in %u", f->fname,
5542 : snapshot->subxip[0]);
5543 27 : ApplyLogicalMappingFile(tuplecid_data, f->fname);
5544 27 : pfree(f);
5545 : }
5546 11 : }
5547 :
5548 : /*
5549 : * Lookup cmin/cmax of a tuple, during logical decoding where we can't rely on
5550 : * combo CIDs.
5551 : */
5552 : bool
5553 861 : ResolveCminCmaxDuringDecoding(HTAB *tuplecid_data,
5554 : Snapshot snapshot,
5555 : HeapTuple htup, Buffer buffer,
5556 : CommandId *cmin, CommandId *cmax)
5557 : {
5558 : ReorderBufferTupleCidKey key;
5559 : ReorderBufferTupleCidEnt *ent;
5560 : ForkNumber forkno;
5561 : BlockNumber blockno;
5562 861 : bool updated_mapping = false;
5563 :
5564 : /*
5565 : * Return unresolved if tuplecid_data is not valid. That's because when
5566 : * streaming in-progress transactions we may run into tuples with the CID
5567 : * before actually decoding them. Think e.g. about INSERT followed by
5568 : * TRUNCATE, where the TRUNCATE may not be decoded yet when applying the
5569 : * INSERT. So in such cases, we assume the CID is from the future
5570 : * command.
5571 : */
5572 861 : if (tuplecid_data == NULL)
5573 11 : return false;
5574 :
5575 : /* be careful about padding */
5576 850 : memset(&key, 0, sizeof(key));
5577 :
5578 : Assert(!BufferIsLocal(buffer));
5579 :
5580 : /*
5581 : * get relfilelocator from the buffer, no convenient way to access it
5582 : * other than that.
5583 : */
5584 850 : BufferGetTag(buffer, &key.rlocator, &forkno, &blockno);
5585 :
5586 : /* tuples can only be in the main fork */
5587 : Assert(forkno == MAIN_FORKNUM);
5588 : Assert(blockno == ItemPointerGetBlockNumber(&htup->t_self));
5589 :
5590 850 : ItemPointerCopy(&htup->t_self,
5591 : &key.tid);
5592 :
5593 861 : restart:
5594 : ent = (ReorderBufferTupleCidEnt *)
5595 861 : hash_search(tuplecid_data, &key, HASH_FIND, NULL);
5596 :
5597 : /*
5598 : * failed to find a mapping, check whether the table was rewritten and
5599 : * apply mapping if so, but only do that once - there can be no new
5600 : * mappings while we are in here since we have to hold a lock on the
5601 : * relation.
5602 : */
5603 861 : if (ent == NULL && !updated_mapping)
5604 : {
5605 11 : UpdateLogicalMappings(tuplecid_data, htup->t_tableOid, snapshot);
5606 : /* now check but don't update for a mapping again */
5607 11 : updated_mapping = true;
5608 11 : goto restart;
5609 : }
5610 850 : else if (ent == NULL)
5611 5 : return false;
5612 :
5613 845 : if (cmin)
5614 845 : *cmin = ent->cmin;
5615 845 : if (cmax)
5616 845 : *cmax = ent->cmax;
5617 845 : return true;
5618 : }
5619 :
5620 : /*
5621 : * Count invalidation messages of specified transaction.
5622 : *
5623 : * Returns number of messages, and msgs is set to the pointer of the linked
5624 : * list for the messages.
5625 : */
5626 : uint32
5627 32 : ReorderBufferGetInvalidations(ReorderBuffer *rb, TransactionId xid,
5628 : SharedInvalidationMessage **msgs)
5629 : {
5630 : ReorderBufferTXN *txn;
5631 :
5632 32 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
5633 : false);
5634 :
5635 32 : if (txn == NULL)
5636 0 : return 0;
5637 :
5638 32 : *msgs = txn->invalidations;
5639 :
5640 32 : return txn->ninvalidations;
5641 : }
|
