Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * reorderbuffer.c
4 : * PostgreSQL logical replay/reorder buffer management
5 : *
6 : *
7 : * Copyright (c) 2012-2025, PostgreSQL Global Development Group
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/replication/logical/reorderbuffer.c
12 : *
13 : * NOTES
14 : * This module gets handed individual pieces of transactions in the order
15 : * they are written to the WAL and is responsible to reassemble them into
16 : * toplevel transaction sized pieces. When a transaction is completely
17 : * reassembled - signaled by reading the transaction commit record - it
18 : * will then call the output plugin (cf. ReorderBufferCommit()) with the
19 : * individual changes. The output plugins rely on snapshots built by
20 : * snapbuild.c which hands them to us.
21 : *
22 : * Transactions and subtransactions/savepoints in postgres are not
23 : * immediately linked to each other from outside the performing
24 : * backend. Only at commit/abort (or special xact_assignment records) they
25 : * are linked together. Which means that we will have to splice together a
26 : * toplevel transaction from its subtransactions. To do that efficiently we
27 : * build a binary heap indexed by the smallest current lsn of the individual
28 : * subtransactions' changestreams. As the individual streams are inherently
29 : * ordered by LSN - since that is where we build them from - the transaction
30 : * can easily be reassembled by always using the subtransaction with the
31 : * smallest current LSN from the heap.
32 : *
33 : * In order to cope with large transactions - which can be several times as
34 : * big as the available memory - this module supports spooling the contents
35 : * of large transactions to disk. When the transaction is replayed the
36 : * contents of individual (sub-)transactions will be read from disk in
37 : * chunks.
38 : *
39 : * This module also has to deal with reassembling toast records from the
40 : * individual chunks stored in WAL. When a new (or initial) version of a
41 : * tuple is stored in WAL it will always be preceded by the toast chunks
42 : * emitted for the columns stored out of line. Within a single toplevel
43 : * transaction there will be no other data carrying records between a row's
44 : * toast chunks and the row data itself. See ReorderBufferToast* for
45 : * details.
46 : *
47 : * ReorderBuffer uses two special memory context types - SlabContext for
48 : * allocations of fixed-length structures (changes and transactions), and
49 : * GenerationContext for the variable-length transaction data (allocated
50 : * and freed in groups with similar lifespans).
51 : *
52 : * To limit the amount of memory used by decoded changes, we track memory
53 : * used at the reorder buffer level (i.e. total amount of memory), and for
54 : * each transaction. When the total amount of used memory exceeds the
55 : * limit, the transaction consuming the most memory is then serialized to
56 : * disk.
57 : *
58 : * Only decoded changes are evicted from memory (spilled to disk), not the
59 : * transaction records. The number of toplevel transactions is limited,
60 : * but a transaction with many subtransactions may still consume significant
61 : * amounts of memory. However, the transaction records are fairly small and
62 : * are not included in the memory limit.
63 : *
64 : * The current eviction algorithm is very simple - the transaction is
65 : * picked merely by size, while it might be useful to also consider age
66 : * (LSN) of the changes for example. With the new Generational memory
67 : * allocator, evicting the oldest changes would make it more likely the
68 : * memory gets actually freed.
69 : *
70 : * We use a max-heap with transaction size as the key to efficiently find
71 : * the largest transaction. We update the max-heap whenever the memory
72 : * counter is updated; however transactions with size 0 are not stored in
73 : * the heap, because they have no changes to evict.
74 : *
75 : * We still rely on max_changes_in_memory when loading serialized changes
76 : * back into memory. At that point we can't use the memory limit directly
77 : * as we load the subxacts independently. One option to deal with this
78 : * would be to count the subxacts, and allow each to allocate 1/N of the
79 : * memory limit. That however does not seem very appealing, because with
80 : * many subtransactions it may easily cause thrashing (short cycles of
81 : * deserializing and applying very few changes). We probably should give
82 : * a bit more memory to the oldest subtransactions, because it's likely
83 : * they are the source for the next sequence of changes.
84 : *
85 : * -------------------------------------------------------------------------
86 : */
87 : #include "postgres.h"
88 :
89 : #include <unistd.h>
90 : #include <sys/stat.h>
91 :
92 : #include "access/detoast.h"
93 : #include "access/heapam.h"
94 : #include "access/rewriteheap.h"
95 : #include "access/transam.h"
96 : #include "access/xact.h"
97 : #include "access/xlog_internal.h"
98 : #include "catalog/catalog.h"
99 : #include "common/int.h"
100 : #include "lib/binaryheap.h"
101 : #include "miscadmin.h"
102 : #include "pgstat.h"
103 : #include "replication/logical.h"
104 : #include "replication/reorderbuffer.h"
105 : #include "replication/slot.h"
106 : #include "replication/snapbuild.h" /* just for SnapBuildSnapDecRefcount */
107 : #include "storage/bufmgr.h"
108 : #include "storage/fd.h"
109 : #include "storage/procarray.h"
110 : #include "storage/sinval.h"
111 : #include "utils/builtins.h"
112 : #include "utils/memutils.h"
113 : #include "utils/rel.h"
114 : #include "utils/relfilenumbermap.h"
115 :
116 : /* entry for a hash table we use to map from xid to our transaction state */
117 : typedef struct ReorderBufferTXNByIdEnt
118 : {
119 : TransactionId xid;
120 : ReorderBufferTXN *txn;
121 : } ReorderBufferTXNByIdEnt;
122 :
123 : /* data structures for (relfilelocator, ctid) => (cmin, cmax) mapping */
124 : typedef struct ReorderBufferTupleCidKey
125 : {
126 : RelFileLocator rlocator;
127 : ItemPointerData tid;
128 : } ReorderBufferTupleCidKey;
129 :
130 : typedef struct ReorderBufferTupleCidEnt
131 : {
132 : ReorderBufferTupleCidKey key;
133 : CommandId cmin;
134 : CommandId cmax;
135 : CommandId combocid; /* just for debugging */
136 : } ReorderBufferTupleCidEnt;
137 :
138 : /* Virtual file descriptor with file offset tracking */
139 : typedef struct TXNEntryFile
140 : {
141 : File vfd; /* -1 when the file is closed */
142 : off_t curOffset; /* offset for next write or read. Reset to 0
143 : * when vfd is opened. */
144 : } TXNEntryFile;
145 :
146 : /* k-way in-order change iteration support structures */
147 : typedef struct ReorderBufferIterTXNEntry
148 : {
149 : XLogRecPtr lsn;
150 : ReorderBufferChange *change;
151 : ReorderBufferTXN *txn;
152 : TXNEntryFile file;
153 : XLogSegNo segno;
154 : } ReorderBufferIterTXNEntry;
155 :
156 : typedef struct ReorderBufferIterTXNState
157 : {
158 : binaryheap *heap;
159 : Size nr_txns;
160 : dlist_head old_change;
161 : ReorderBufferIterTXNEntry entries[FLEXIBLE_ARRAY_MEMBER];
162 : } ReorderBufferIterTXNState;
163 :
164 : /* toast datastructures */
165 : typedef struct ReorderBufferToastEnt
166 : {
167 : Oid chunk_id; /* toast_table.chunk_id */
168 : int32 last_chunk_seq; /* toast_table.chunk_seq of the last chunk we
169 : * have seen */
170 : Size num_chunks; /* number of chunks we've already seen */
171 : Size size; /* combined size of chunks seen */
172 : dlist_head chunks; /* linked list of chunks */
173 : struct varlena *reconstructed; /* reconstructed varlena now pointed to in
174 : * main tup */
175 : } ReorderBufferToastEnt;
176 :
177 : /* Disk serialization support datastructures */
178 : typedef struct ReorderBufferDiskChange
179 : {
180 : Size size;
181 : ReorderBufferChange change;
182 : /* data follows */
183 : } ReorderBufferDiskChange;
184 :
185 : #define IsSpecInsert(action) \
186 : ( \
187 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT) \
188 : )
189 : #define IsSpecConfirmOrAbort(action) \
190 : ( \
191 : (((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM) || \
192 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT)) \
193 : )
194 : #define IsInsertOrUpdate(action) \
195 : ( \
196 : (((action) == REORDER_BUFFER_CHANGE_INSERT) || \
197 : ((action) == REORDER_BUFFER_CHANGE_UPDATE) || \
198 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT)) \
199 : )
200 :
201 : /*
202 : * Maximum number of changes kept in memory, per transaction. After that,
203 : * changes are spooled to disk.
204 : *
205 : * The current value should be sufficient to decode the entire transaction
206 : * without hitting disk in OLTP workloads, while starting to spool to disk in
207 : * other workloads reasonably fast.
208 : *
209 : * At some point in the future it probably makes sense to have a more elaborate
210 : * resource management here, but it's not entirely clear what that would look
211 : * like.
212 : */
213 : int logical_decoding_work_mem;
214 : static const Size max_changes_in_memory = 4096; /* XXX for restore only */
215 :
216 : /* GUC variable */
217 : int debug_logical_replication_streaming = DEBUG_LOGICAL_REP_STREAMING_BUFFERED;
218 :
219 : /* ---------------------------------------
220 : * primary reorderbuffer support routines
221 : * ---------------------------------------
222 : */
223 : static ReorderBufferTXN *ReorderBufferGetTXN(ReorderBuffer *rb);
224 : static void ReorderBufferReturnTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
225 : static ReorderBufferTXN *ReorderBufferTXNByXid(ReorderBuffer *rb,
226 : TransactionId xid, bool create, bool *is_new,
227 : XLogRecPtr lsn, bool create_as_top);
228 : static void ReorderBufferTransferSnapToParent(ReorderBufferTXN *txn,
229 : ReorderBufferTXN *subtxn);
230 :
231 : static void AssertTXNLsnOrder(ReorderBuffer *rb);
232 :
233 : /* ---------------------------------------
234 : * support functions for lsn-order iterating over the ->changes of a
235 : * transaction and its subtransactions
236 : *
237 : * used for iteration over the k-way heap merge of a transaction and its
238 : * subtransactions
239 : * ---------------------------------------
240 : */
241 : static void ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn,
242 : ReorderBufferIterTXNState *volatile *iter_state);
243 : static ReorderBufferChange *ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state);
244 : static void ReorderBufferIterTXNFinish(ReorderBuffer *rb,
245 : ReorderBufferIterTXNState *state);
246 : static void ReorderBufferExecuteInvalidations(uint32 nmsgs, SharedInvalidationMessage *msgs);
247 :
248 : /*
249 : * ---------------------------------------
250 : * Disk serialization support functions
251 : * ---------------------------------------
252 : */
253 : static void ReorderBufferCheckMemoryLimit(ReorderBuffer *rb);
254 : static void ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
255 : static void ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
256 : int fd, ReorderBufferChange *change);
257 : static Size ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
258 : TXNEntryFile *file, XLogSegNo *segno);
259 : static void ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
260 : char *data);
261 : static void ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn);
262 : static void ReorderBufferTruncateTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
263 : bool txn_prepared);
264 : static void ReorderBufferMaybeMarkTXNStreamed(ReorderBuffer *rb, ReorderBufferTXN *txn);
265 : static bool ReorderBufferCheckAndTruncateAbortedTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
266 : static void ReorderBufferCleanupSerializedTXNs(const char *slotname);
267 : static void ReorderBufferSerializedPath(char *path, ReplicationSlot *slot,
268 : TransactionId xid, XLogSegNo segno);
269 : static int ReorderBufferTXNSizeCompare(const pairingheap_node *a, const pairingheap_node *b, void *arg);
270 :
271 : static void ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap);
272 : static Snapshot ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
273 : ReorderBufferTXN *txn, CommandId cid);
274 :
275 : /*
276 : * ---------------------------------------
277 : * Streaming support functions
278 : * ---------------------------------------
279 : */
280 : static inline bool ReorderBufferCanStream(ReorderBuffer *rb);
281 : static inline bool ReorderBufferCanStartStreaming(ReorderBuffer *rb);
282 : static void ReorderBufferStreamTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
283 : static void ReorderBufferStreamCommit(ReorderBuffer *rb, ReorderBufferTXN *txn);
284 :
285 : /* ---------------------------------------
286 : * toast reassembly support
287 : * ---------------------------------------
288 : */
289 : static void ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn);
290 : static void ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn);
291 : static void ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
292 : Relation relation, ReorderBufferChange *change);
293 : static void ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
294 : Relation relation, ReorderBufferChange *change);
295 :
296 : /*
297 : * ---------------------------------------
298 : * memory accounting
299 : * ---------------------------------------
300 : */
301 : static Size ReorderBufferChangeSize(ReorderBufferChange *change);
302 : static void ReorderBufferChangeMemoryUpdate(ReorderBuffer *rb,
303 : ReorderBufferChange *change,
304 : ReorderBufferTXN *txn,
305 : bool addition, Size sz);
306 :
307 : /*
308 : * Allocate a new ReorderBuffer and clean out any old serialized state from
309 : * prior ReorderBuffer instances for the same slot.
310 : */
311 : ReorderBuffer *
312 2034 : ReorderBufferAllocate(void)
313 : {
314 : ReorderBuffer *buffer;
315 : HASHCTL hash_ctl;
316 : MemoryContext new_ctx;
317 :
318 : Assert(MyReplicationSlot != NULL);
319 :
320 : /* allocate memory in own context, to have better accountability */
321 2034 : new_ctx = AllocSetContextCreate(CurrentMemoryContext,
322 : "ReorderBuffer",
323 : ALLOCSET_DEFAULT_SIZES);
324 :
325 : buffer =
326 2034 : (ReorderBuffer *) MemoryContextAlloc(new_ctx, sizeof(ReorderBuffer));
327 :
328 2034 : memset(&hash_ctl, 0, sizeof(hash_ctl));
329 :
330 2034 : buffer->context = new_ctx;
331 :
332 2034 : buffer->change_context = SlabContextCreate(new_ctx,
333 : "Change",
334 : SLAB_DEFAULT_BLOCK_SIZE,
335 : sizeof(ReorderBufferChange));
336 :
337 2034 : buffer->txn_context = SlabContextCreate(new_ctx,
338 : "TXN",
339 : SLAB_DEFAULT_BLOCK_SIZE,
340 : sizeof(ReorderBufferTXN));
341 :
342 : /*
343 : * To minimize memory fragmentation caused by long-running transactions
344 : * with changes spanning multiple memory blocks, we use a single
345 : * fixed-size memory block for decoded tuple storage. The performance
346 : * testing showed that the default memory block size maintains logical
347 : * decoding performance without causing fragmentation due to concurrent
348 : * transactions. One might think that we can use the max size as
349 : * SLAB_LARGE_BLOCK_SIZE but the test also showed it doesn't help resolve
350 : * the memory fragmentation.
351 : */
352 2034 : buffer->tup_context = GenerationContextCreate(new_ctx,
353 : "Tuples",
354 : SLAB_DEFAULT_BLOCK_SIZE,
355 : SLAB_DEFAULT_BLOCK_SIZE,
356 : SLAB_DEFAULT_BLOCK_SIZE);
357 :
358 2034 : hash_ctl.keysize = sizeof(TransactionId);
359 2034 : hash_ctl.entrysize = sizeof(ReorderBufferTXNByIdEnt);
360 2034 : hash_ctl.hcxt = buffer->context;
361 :
362 2034 : buffer->by_txn = hash_create("ReorderBufferByXid", 1000, &hash_ctl,
363 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
364 :
365 2034 : buffer->by_txn_last_xid = InvalidTransactionId;
366 2034 : buffer->by_txn_last_txn = NULL;
367 :
368 2034 : buffer->outbuf = NULL;
369 2034 : buffer->outbufsize = 0;
370 2034 : buffer->size = 0;
371 :
372 : /* txn_heap is ordered by transaction size */
373 2034 : buffer->txn_heap = pairingheap_allocate(ReorderBufferTXNSizeCompare, NULL);
374 :
375 2034 : buffer->spillTxns = 0;
376 2034 : buffer->spillCount = 0;
377 2034 : buffer->spillBytes = 0;
378 2034 : buffer->streamTxns = 0;
379 2034 : buffer->streamCount = 0;
380 2034 : buffer->streamBytes = 0;
381 2034 : buffer->totalTxns = 0;
382 2034 : buffer->totalBytes = 0;
383 :
384 2034 : buffer->current_restart_decoding_lsn = InvalidXLogRecPtr;
385 :
386 2034 : dlist_init(&buffer->toplevel_by_lsn);
387 2034 : dlist_init(&buffer->txns_by_base_snapshot_lsn);
388 2034 : dclist_init(&buffer->catchange_txns);
389 :
390 : /*
391 : * Ensure there's no stale data from prior uses of this slot, in case some
392 : * prior exit avoided calling ReorderBufferFree. Failure to do this can
393 : * produce duplicated txns, and it's very cheap if there's nothing there.
394 : */
395 2034 : ReorderBufferCleanupSerializedTXNs(NameStr(MyReplicationSlot->data.name));
396 :
397 2034 : return buffer;
398 : }
399 :
400 : /*
401 : * Free a ReorderBuffer
402 : */
403 : void
404 1668 : ReorderBufferFree(ReorderBuffer *rb)
405 : {
406 1668 : MemoryContext context = rb->context;
407 :
408 : /*
409 : * We free separately allocated data by entirely scrapping reorderbuffer's
410 : * memory context.
411 : */
412 1668 : MemoryContextDelete(context);
413 :
414 : /* Free disk space used by unconsumed reorder buffers */
415 1668 : ReorderBufferCleanupSerializedTXNs(NameStr(MyReplicationSlot->data.name));
416 1668 : }
417 :
418 : /*
419 : * Get an unused, possibly preallocated, ReorderBufferTXN.
420 : */
421 : static ReorderBufferTXN *
422 7022 : ReorderBufferGetTXN(ReorderBuffer *rb)
423 : {
424 : ReorderBufferTXN *txn;
425 :
426 : txn = (ReorderBufferTXN *)
427 7022 : MemoryContextAlloc(rb->txn_context, sizeof(ReorderBufferTXN));
428 :
429 7022 : memset(txn, 0, sizeof(ReorderBufferTXN));
430 :
431 7022 : dlist_init(&txn->changes);
432 7022 : dlist_init(&txn->tuplecids);
433 7022 : dlist_init(&txn->subtxns);
434 :
435 : /* InvalidCommandId is not zero, so set it explicitly */
436 7022 : txn->command_id = InvalidCommandId;
437 7022 : txn->output_plugin_private = NULL;
438 :
439 7022 : return txn;
440 : }
441 :
442 : /*
443 : * Free a ReorderBufferTXN.
444 : */
445 : static void
446 6914 : ReorderBufferReturnTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
447 : {
448 : /* clean the lookup cache if we were cached (quite likely) */
449 6914 : if (rb->by_txn_last_xid == txn->xid)
450 : {
451 6542 : rb->by_txn_last_xid = InvalidTransactionId;
452 6542 : rb->by_txn_last_txn = NULL;
453 : }
454 :
455 : /* free data that's contained */
456 :
457 6914 : if (txn->gid != NULL)
458 : {
459 84 : pfree(txn->gid);
460 84 : txn->gid = NULL;
461 : }
462 :
463 6914 : if (txn->tuplecid_hash != NULL)
464 : {
465 972 : hash_destroy(txn->tuplecid_hash);
466 972 : txn->tuplecid_hash = NULL;
467 : }
468 :
469 6914 : if (txn->invalidations)
470 : {
471 2052 : pfree(txn->invalidations);
472 2052 : txn->invalidations = NULL;
473 : }
474 :
475 : /* Reset the toast hash */
476 6914 : ReorderBufferToastReset(rb, txn);
477 :
478 : /* All changes must be deallocated */
479 : Assert(txn->size == 0);
480 :
481 6914 : pfree(txn);
482 6914 : }
483 :
484 : /*
485 : * Get a fresh ReorderBufferChange.
486 : */
487 : ReorderBufferChange *
488 3427668 : ReorderBufferGetChange(ReorderBuffer *rb)
489 : {
490 : ReorderBufferChange *change;
491 :
492 : change = (ReorderBufferChange *)
493 3427668 : MemoryContextAlloc(rb->change_context, sizeof(ReorderBufferChange));
494 :
495 3427668 : memset(change, 0, sizeof(ReorderBufferChange));
496 3427668 : return change;
497 : }
498 :
499 : /*
500 : * Free a ReorderBufferChange and update memory accounting, if requested.
501 : */
502 : void
503 3427234 : ReorderBufferReturnChange(ReorderBuffer *rb, ReorderBufferChange *change,
504 : bool upd_mem)
505 : {
506 : /* update memory accounting info */
507 3427234 : if (upd_mem)
508 405028 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, false,
509 : ReorderBufferChangeSize(change));
510 :
511 : /* free contained data */
512 3427234 : switch (change->action)
513 : {
514 3279680 : case REORDER_BUFFER_CHANGE_INSERT:
515 : case REORDER_BUFFER_CHANGE_UPDATE:
516 : case REORDER_BUFFER_CHANGE_DELETE:
517 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
518 3279680 : if (change->data.tp.newtuple)
519 : {
520 2720994 : ReorderBufferReturnTupleBuf(change->data.tp.newtuple);
521 2720994 : change->data.tp.newtuple = NULL;
522 : }
523 :
524 3279680 : if (change->data.tp.oldtuple)
525 : {
526 422070 : ReorderBufferReturnTupleBuf(change->data.tp.oldtuple);
527 422070 : change->data.tp.oldtuple = NULL;
528 : }
529 3279680 : break;
530 80 : case REORDER_BUFFER_CHANGE_MESSAGE:
531 80 : if (change->data.msg.prefix != NULL)
532 80 : pfree(change->data.msg.prefix);
533 80 : change->data.msg.prefix = NULL;
534 80 : if (change->data.msg.message != NULL)
535 80 : pfree(change->data.msg.message);
536 80 : change->data.msg.message = NULL;
537 80 : break;
538 9986 : case REORDER_BUFFER_CHANGE_INVALIDATION:
539 9986 : if (change->data.inval.invalidations)
540 9986 : pfree(change->data.inval.invalidations);
541 9986 : change->data.inval.invalidations = NULL;
542 9986 : break;
543 2092 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
544 2092 : if (change->data.snapshot)
545 : {
546 2092 : ReorderBufferFreeSnap(rb, change->data.snapshot);
547 2092 : change->data.snapshot = NULL;
548 : }
549 2092 : break;
550 : /* no data in addition to the struct itself */
551 104 : case REORDER_BUFFER_CHANGE_TRUNCATE:
552 104 : if (change->data.truncate.relids != NULL)
553 : {
554 104 : ReorderBufferReturnRelids(rb, change->data.truncate.relids);
555 104 : change->data.truncate.relids = NULL;
556 : }
557 104 : break;
558 135292 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
559 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
560 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
561 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
562 135292 : break;
563 : }
564 :
565 3427234 : pfree(change);
566 3427234 : }
567 :
568 : /*
569 : * Get a fresh HeapTuple fitting a tuple of size tuple_len (excluding header
570 : * overhead).
571 : */
572 : HeapTuple
573 3143142 : ReorderBufferGetTupleBuf(ReorderBuffer *rb, Size tuple_len)
574 : {
575 : HeapTuple tuple;
576 : Size alloc_len;
577 :
578 3143142 : alloc_len = tuple_len + SizeofHeapTupleHeader;
579 :
580 3143142 : tuple = (HeapTuple) MemoryContextAlloc(rb->tup_context,
581 : HEAPTUPLESIZE + alloc_len);
582 3143142 : tuple->t_data = (HeapTupleHeader) ((char *) tuple + HEAPTUPLESIZE);
583 :
584 3143142 : return tuple;
585 : }
586 :
587 : /*
588 : * Free a HeapTuple returned by ReorderBufferGetTupleBuf().
589 : */
590 : void
591 3143064 : ReorderBufferReturnTupleBuf(HeapTuple tuple)
592 : {
593 3143064 : pfree(tuple);
594 3143064 : }
595 :
596 : /*
597 : * Get an array for relids of truncated relations.
598 : *
599 : * We use the global memory context (for the whole reorder buffer), because
600 : * none of the existing ones seems like a good match (some are SLAB, so we
601 : * can't use those, and tup_context is meant for tuple data, not relids). We
602 : * could add yet another context, but it seems like an overkill - TRUNCATE is
603 : * not particularly common operation, so it does not seem worth it.
604 : */
605 : Oid *
606 114 : ReorderBufferGetRelids(ReorderBuffer *rb, int nrelids)
607 : {
608 : Oid *relids;
609 : Size alloc_len;
610 :
611 114 : alloc_len = sizeof(Oid) * nrelids;
612 :
613 114 : relids = (Oid *) MemoryContextAlloc(rb->context, alloc_len);
614 :
615 114 : return relids;
616 : }
617 :
618 : /*
619 : * Free an array of relids.
620 : */
621 : void
622 104 : ReorderBufferReturnRelids(ReorderBuffer *rb, Oid *relids)
623 : {
624 104 : pfree(relids);
625 104 : }
626 :
627 : /*
628 : * Return the ReorderBufferTXN from the given buffer, specified by Xid.
629 : * If create is true, and a transaction doesn't already exist, create it
630 : * (with the given LSN, and as top transaction if that's specified);
631 : * when this happens, is_new is set to true.
632 : */
633 : static ReorderBufferTXN *
634 11402474 : ReorderBufferTXNByXid(ReorderBuffer *rb, TransactionId xid, bool create,
635 : bool *is_new, XLogRecPtr lsn, bool create_as_top)
636 : {
637 : ReorderBufferTXN *txn;
638 : ReorderBufferTXNByIdEnt *ent;
639 : bool found;
640 :
641 : Assert(TransactionIdIsValid(xid));
642 :
643 : /*
644 : * Check the one-entry lookup cache first
645 : */
646 11402474 : if (TransactionIdIsValid(rb->by_txn_last_xid) &&
647 11395860 : rb->by_txn_last_xid == xid)
648 : {
649 9956608 : txn = rb->by_txn_last_txn;
650 :
651 9956608 : if (txn != NULL)
652 : {
653 : /* found it, and it's valid */
654 9956580 : if (is_new)
655 5566 : *is_new = false;
656 9956580 : return txn;
657 : }
658 :
659 : /*
660 : * cached as non-existent, and asked not to create? Then nothing else
661 : * to do.
662 : */
663 28 : if (!create)
664 20 : return NULL;
665 : /* otherwise fall through to create it */
666 : }
667 :
668 : /*
669 : * If the cache wasn't hit or it yielded a "does-not-exist" and we want to
670 : * create an entry.
671 : */
672 :
673 : /* search the lookup table */
674 : ent = (ReorderBufferTXNByIdEnt *)
675 1445874 : hash_search(rb->by_txn,
676 : &xid,
677 : create ? HASH_ENTER : HASH_FIND,
678 : &found);
679 1445874 : if (found)
680 1436280 : txn = ent->txn;
681 9594 : else if (create)
682 : {
683 : /* initialize the new entry, if creation was requested */
684 : Assert(ent != NULL);
685 : Assert(lsn != InvalidXLogRecPtr);
686 :
687 7022 : ent->txn = ReorderBufferGetTXN(rb);
688 7022 : ent->txn->xid = xid;
689 7022 : txn = ent->txn;
690 7022 : txn->first_lsn = lsn;
691 7022 : txn->restart_decoding_lsn = rb->current_restart_decoding_lsn;
692 :
693 7022 : if (create_as_top)
694 : {
695 5726 : dlist_push_tail(&rb->toplevel_by_lsn, &txn->node);
696 5726 : AssertTXNLsnOrder(rb);
697 : }
698 : }
699 : else
700 2572 : txn = NULL; /* not found and not asked to create */
701 :
702 : /* update cache */
703 1445874 : rb->by_txn_last_xid = xid;
704 1445874 : rb->by_txn_last_txn = txn;
705 :
706 1445874 : if (is_new)
707 3428 : *is_new = !found;
708 :
709 : Assert(!create || txn != NULL);
710 1445874 : return txn;
711 : }
712 :
713 : /*
714 : * Record the partial change for the streaming of in-progress transactions. We
715 : * can stream only complete changes so if we have a partial change like toast
716 : * table insert or speculative insert then we mark such a 'txn' so that it
717 : * can't be streamed. We also ensure that if the changes in such a 'txn' can
718 : * be streamed and are above logical_decoding_work_mem threshold then we stream
719 : * them as soon as we have a complete change.
720 : */
721 : static void
722 3003620 : ReorderBufferProcessPartialChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
723 : ReorderBufferChange *change,
724 : bool toast_insert)
725 : {
726 : ReorderBufferTXN *toptxn;
727 :
728 : /*
729 : * The partial changes need to be processed only while streaming
730 : * in-progress transactions.
731 : */
732 3003620 : if (!ReorderBufferCanStream(rb))
733 1999928 : return;
734 :
735 : /* Get the top transaction. */
736 1003692 : toptxn = rbtxn_get_toptxn(txn);
737 :
738 : /*
739 : * Indicate a partial change for toast inserts. The change will be
740 : * considered as complete once we get the insert or update on the main
741 : * table and we are sure that the pending toast chunks are not required
742 : * anymore.
743 : *
744 : * If we allow streaming when there are pending toast chunks then such
745 : * chunks won't be released till the insert (multi_insert) is complete and
746 : * we expect the txn to have streamed all changes after streaming. This
747 : * restriction is mainly to ensure the correctness of streamed
748 : * transactions and it doesn't seem worth uplifting such a restriction
749 : * just to allow this case because anyway we will stream the transaction
750 : * once such an insert is complete.
751 : */
752 1003692 : if (toast_insert)
753 3332 : toptxn->txn_flags |= RBTXN_HAS_PARTIAL_CHANGE;
754 1000360 : else if (rbtxn_has_partial_change(toptxn) &&
755 126 : IsInsertOrUpdate(change->action) &&
756 126 : change->data.tp.clear_toast_afterwards)
757 86 : toptxn->txn_flags &= ~RBTXN_HAS_PARTIAL_CHANGE;
758 :
759 : /*
760 : * Indicate a partial change for speculative inserts. The change will be
761 : * considered as complete once we get the speculative confirm or abort
762 : * token.
763 : */
764 1003692 : if (IsSpecInsert(change->action))
765 0 : toptxn->txn_flags |= RBTXN_HAS_PARTIAL_CHANGE;
766 1003692 : else if (rbtxn_has_partial_change(toptxn) &&
767 3372 : IsSpecConfirmOrAbort(change->action))
768 0 : toptxn->txn_flags &= ~RBTXN_HAS_PARTIAL_CHANGE;
769 :
770 : /*
771 : * Stream the transaction if it is serialized before and the changes are
772 : * now complete in the top-level transaction.
773 : *
774 : * The reason for doing the streaming of such a transaction as soon as we
775 : * get the complete change for it is that previously it would have reached
776 : * the memory threshold and wouldn't get streamed because of incomplete
777 : * changes. Delaying such transactions would increase apply lag for them.
778 : */
779 1003692 : if (ReorderBufferCanStartStreaming(rb) &&
780 336092 : !(rbtxn_has_partial_change(toptxn)) &&
781 333020 : rbtxn_is_serialized(txn) &&
782 78 : rbtxn_has_streamable_change(toptxn))
783 18 : ReorderBufferStreamTXN(rb, toptxn);
784 : }
785 :
786 : /*
787 : * Queue a change into a transaction so it can be replayed upon commit or will be
788 : * streamed when we reach logical_decoding_work_mem threshold.
789 : */
790 : void
791 3022438 : ReorderBufferQueueChange(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
792 : ReorderBufferChange *change, bool toast_insert)
793 : {
794 : ReorderBufferTXN *txn;
795 :
796 3022438 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
797 :
798 : /*
799 : * If we have detected that the transaction is aborted while streaming the
800 : * previous changes or by checking its CLOG, there is no point in
801 : * collecting further changes for it.
802 : */
803 3022438 : if (rbtxn_is_aborted(txn))
804 : {
805 : /*
806 : * We don't need to update memory accounting for this change as we
807 : * have not added it to the queue yet.
808 : */
809 18818 : ReorderBufferReturnChange(rb, change, false);
810 18818 : return;
811 : }
812 :
813 : /*
814 : * The changes that are sent downstream are considered streamable. We
815 : * remember such transactions so that only those will later be considered
816 : * for streaming.
817 : */
818 3003620 : if (change->action == REORDER_BUFFER_CHANGE_INSERT ||
819 1080698 : change->action == REORDER_BUFFER_CHANGE_UPDATE ||
820 666400 : change->action == REORDER_BUFFER_CHANGE_DELETE ||
821 131652 : change->action == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT ||
822 95820 : change->action == REORDER_BUFFER_CHANGE_TRUNCATE ||
823 95726 : change->action == REORDER_BUFFER_CHANGE_MESSAGE)
824 : {
825 2907972 : ReorderBufferTXN *toptxn = rbtxn_get_toptxn(txn);
826 :
827 2907972 : toptxn->txn_flags |= RBTXN_HAS_STREAMABLE_CHANGE;
828 : }
829 :
830 3003620 : change->lsn = lsn;
831 3003620 : change->txn = txn;
832 :
833 : Assert(InvalidXLogRecPtr != lsn);
834 3003620 : dlist_push_tail(&txn->changes, &change->node);
835 3003620 : txn->nentries++;
836 3003620 : txn->nentries_mem++;
837 :
838 : /* update memory accounting information */
839 3003620 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
840 : ReorderBufferChangeSize(change));
841 :
842 : /* process partial change */
843 3003620 : ReorderBufferProcessPartialChange(rb, txn, change, toast_insert);
844 :
845 : /* check the memory limits and evict something if needed */
846 3003620 : ReorderBufferCheckMemoryLimit(rb);
847 : }
848 :
849 : /*
850 : * A transactional message is queued to be processed upon commit and a
851 : * non-transactional message gets processed immediately.
852 : */
853 : void
854 94 : ReorderBufferQueueMessage(ReorderBuffer *rb, TransactionId xid,
855 : Snapshot snap, XLogRecPtr lsn,
856 : bool transactional, const char *prefix,
857 : Size message_size, const char *message)
858 : {
859 94 : if (transactional)
860 : {
861 : MemoryContext oldcontext;
862 : ReorderBufferChange *change;
863 :
864 : Assert(xid != InvalidTransactionId);
865 :
866 : /*
867 : * We don't expect snapshots for transactional changes - we'll use the
868 : * snapshot derived later during apply (unless the change gets
869 : * skipped).
870 : */
871 : Assert(!snap);
872 :
873 78 : oldcontext = MemoryContextSwitchTo(rb->context);
874 :
875 78 : change = ReorderBufferGetChange(rb);
876 78 : change->action = REORDER_BUFFER_CHANGE_MESSAGE;
877 78 : change->data.msg.prefix = pstrdup(prefix);
878 78 : change->data.msg.message_size = message_size;
879 78 : change->data.msg.message = palloc(message_size);
880 78 : memcpy(change->data.msg.message, message, message_size);
881 :
882 78 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
883 :
884 78 : MemoryContextSwitchTo(oldcontext);
885 : }
886 : else
887 : {
888 16 : ReorderBufferTXN *txn = NULL;
889 16 : volatile Snapshot snapshot_now = snap;
890 :
891 : /* Non-transactional changes require a valid snapshot. */
892 : Assert(snapshot_now);
893 :
894 16 : if (xid != InvalidTransactionId)
895 6 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
896 :
897 : /* setup snapshot to allow catalog access */
898 16 : SetupHistoricSnapshot(snapshot_now, NULL);
899 16 : PG_TRY();
900 : {
901 16 : rb->message(rb, txn, lsn, false, prefix, message_size, message);
902 :
903 16 : TeardownHistoricSnapshot(false);
904 : }
905 0 : PG_CATCH();
906 : {
907 0 : TeardownHistoricSnapshot(true);
908 0 : PG_RE_THROW();
909 : }
910 16 : PG_END_TRY();
911 : }
912 94 : }
913 :
914 : /*
915 : * AssertTXNLsnOrder
916 : * Verify LSN ordering of transaction lists in the reorderbuffer
917 : *
918 : * Other LSN-related invariants are checked too.
919 : *
920 : * No-op if assertions are not in use.
921 : */
922 : static void
923 14092 : AssertTXNLsnOrder(ReorderBuffer *rb)
924 : {
925 : #ifdef USE_ASSERT_CHECKING
926 : LogicalDecodingContext *ctx = rb->private_data;
927 : dlist_iter iter;
928 : XLogRecPtr prev_first_lsn = InvalidXLogRecPtr;
929 : XLogRecPtr prev_base_snap_lsn = InvalidXLogRecPtr;
930 :
931 : /*
932 : * Skip the verification if we don't reach the LSN at which we start
933 : * decoding the contents of transactions yet because until we reach the
934 : * LSN, we could have transactions that don't have the association between
935 : * the top-level transaction and subtransaction yet and consequently have
936 : * the same LSN. We don't guarantee this association until we try to
937 : * decode the actual contents of transaction. The ordering of the records
938 : * prior to the start_decoding_at LSN should have been checked before the
939 : * restart.
940 : */
941 : if (SnapBuildXactNeedsSkip(ctx->snapshot_builder, ctx->reader->EndRecPtr))
942 : return;
943 :
944 : dlist_foreach(iter, &rb->toplevel_by_lsn)
945 : {
946 : ReorderBufferTXN *cur_txn = dlist_container(ReorderBufferTXN, node,
947 : iter.cur);
948 :
949 : /* start LSN must be set */
950 : Assert(cur_txn->first_lsn != InvalidXLogRecPtr);
951 :
952 : /* If there is an end LSN, it must be higher than start LSN */
953 : if (cur_txn->end_lsn != InvalidXLogRecPtr)
954 : Assert(cur_txn->first_lsn <= cur_txn->end_lsn);
955 :
956 : /* Current initial LSN must be strictly higher than previous */
957 : if (prev_first_lsn != InvalidXLogRecPtr)
958 : Assert(prev_first_lsn < cur_txn->first_lsn);
959 :
960 : /* known-as-subtxn txns must not be listed */
961 : Assert(!rbtxn_is_known_subxact(cur_txn));
962 :
963 : prev_first_lsn = cur_txn->first_lsn;
964 : }
965 :
966 : dlist_foreach(iter, &rb->txns_by_base_snapshot_lsn)
967 : {
968 : ReorderBufferTXN *cur_txn = dlist_container(ReorderBufferTXN,
969 : base_snapshot_node,
970 : iter.cur);
971 :
972 : /* base snapshot (and its LSN) must be set */
973 : Assert(cur_txn->base_snapshot != NULL);
974 : Assert(cur_txn->base_snapshot_lsn != InvalidXLogRecPtr);
975 :
976 : /* current LSN must be strictly higher than previous */
977 : if (prev_base_snap_lsn != InvalidXLogRecPtr)
978 : Assert(prev_base_snap_lsn < cur_txn->base_snapshot_lsn);
979 :
980 : /* known-as-subtxn txns must not be listed */
981 : Assert(!rbtxn_is_known_subxact(cur_txn));
982 :
983 : prev_base_snap_lsn = cur_txn->base_snapshot_lsn;
984 : }
985 : #endif
986 14092 : }
987 :
988 : /*
989 : * AssertChangeLsnOrder
990 : *
991 : * Check ordering of changes in the (sub)transaction.
992 : */
993 : static void
994 4732 : AssertChangeLsnOrder(ReorderBufferTXN *txn)
995 : {
996 : #ifdef USE_ASSERT_CHECKING
997 : dlist_iter iter;
998 : XLogRecPtr prev_lsn = txn->first_lsn;
999 :
1000 : dlist_foreach(iter, &txn->changes)
1001 : {
1002 : ReorderBufferChange *cur_change;
1003 :
1004 : cur_change = dlist_container(ReorderBufferChange, node, iter.cur);
1005 :
1006 : Assert(txn->first_lsn != InvalidXLogRecPtr);
1007 : Assert(cur_change->lsn != InvalidXLogRecPtr);
1008 : Assert(txn->first_lsn <= cur_change->lsn);
1009 :
1010 : if (txn->end_lsn != InvalidXLogRecPtr)
1011 : Assert(cur_change->lsn <= txn->end_lsn);
1012 :
1013 : Assert(prev_lsn <= cur_change->lsn);
1014 :
1015 : prev_lsn = cur_change->lsn;
1016 : }
1017 : #endif
1018 4732 : }
1019 :
1020 : /*
1021 : * ReorderBufferGetOldestTXN
1022 : * Return oldest transaction in reorderbuffer
1023 : */
1024 : ReorderBufferTXN *
1025 690 : ReorderBufferGetOldestTXN(ReorderBuffer *rb)
1026 : {
1027 : ReorderBufferTXN *txn;
1028 :
1029 690 : AssertTXNLsnOrder(rb);
1030 :
1031 690 : if (dlist_is_empty(&rb->toplevel_by_lsn))
1032 580 : return NULL;
1033 :
1034 110 : txn = dlist_head_element(ReorderBufferTXN, node, &rb->toplevel_by_lsn);
1035 :
1036 : Assert(!rbtxn_is_known_subxact(txn));
1037 : Assert(txn->first_lsn != InvalidXLogRecPtr);
1038 110 : return txn;
1039 : }
1040 :
1041 : /*
1042 : * ReorderBufferGetOldestXmin
1043 : * Return oldest Xmin in reorderbuffer
1044 : *
1045 : * Returns oldest possibly running Xid from the point of view of snapshots
1046 : * used in the transactions kept by reorderbuffer, or InvalidTransactionId if
1047 : * there are none.
1048 : *
1049 : * Since snapshots are assigned monotonically, this equals the Xmin of the
1050 : * base snapshot with minimal base_snapshot_lsn.
1051 : */
1052 : TransactionId
1053 726 : ReorderBufferGetOldestXmin(ReorderBuffer *rb)
1054 : {
1055 : ReorderBufferTXN *txn;
1056 :
1057 726 : AssertTXNLsnOrder(rb);
1058 :
1059 726 : if (dlist_is_empty(&rb->txns_by_base_snapshot_lsn))
1060 636 : return InvalidTransactionId;
1061 :
1062 90 : txn = dlist_head_element(ReorderBufferTXN, base_snapshot_node,
1063 : &rb->txns_by_base_snapshot_lsn);
1064 90 : return txn->base_snapshot->xmin;
1065 : }
1066 :
1067 : void
1068 778 : ReorderBufferSetRestartPoint(ReorderBuffer *rb, XLogRecPtr ptr)
1069 : {
1070 778 : rb->current_restart_decoding_lsn = ptr;
1071 778 : }
1072 :
1073 : /*
1074 : * ReorderBufferAssignChild
1075 : *
1076 : * Make note that we know that subxid is a subtransaction of xid, seen as of
1077 : * the given lsn.
1078 : */
1079 : void
1080 1670 : ReorderBufferAssignChild(ReorderBuffer *rb, TransactionId xid,
1081 : TransactionId subxid, XLogRecPtr lsn)
1082 : {
1083 : ReorderBufferTXN *txn;
1084 : ReorderBufferTXN *subtxn;
1085 : bool new_top;
1086 : bool new_sub;
1087 :
1088 1670 : txn = ReorderBufferTXNByXid(rb, xid, true, &new_top, lsn, true);
1089 1670 : subtxn = ReorderBufferTXNByXid(rb, subxid, true, &new_sub, lsn, false);
1090 :
1091 1670 : if (!new_sub)
1092 : {
1093 374 : if (rbtxn_is_known_subxact(subtxn))
1094 : {
1095 : /* already associated, nothing to do */
1096 374 : return;
1097 : }
1098 : else
1099 : {
1100 : /*
1101 : * We already saw this transaction, but initially added it to the
1102 : * list of top-level txns. Now that we know it's not top-level,
1103 : * remove it from there.
1104 : */
1105 0 : dlist_delete(&subtxn->node);
1106 : }
1107 : }
1108 :
1109 1296 : subtxn->txn_flags |= RBTXN_IS_SUBXACT;
1110 1296 : subtxn->toplevel_xid = xid;
1111 : Assert(subtxn->nsubtxns == 0);
1112 :
1113 : /* set the reference to top-level transaction */
1114 1296 : subtxn->toptxn = txn;
1115 :
1116 : /* add to subtransaction list */
1117 1296 : dlist_push_tail(&txn->subtxns, &subtxn->node);
1118 1296 : txn->nsubtxns++;
1119 :
1120 : /* Possibly transfer the subtxn's snapshot to its top-level txn. */
1121 1296 : ReorderBufferTransferSnapToParent(txn, subtxn);
1122 :
1123 : /* Verify LSN-ordering invariant */
1124 1296 : AssertTXNLsnOrder(rb);
1125 : }
1126 :
1127 : /*
1128 : * ReorderBufferTransferSnapToParent
1129 : * Transfer base snapshot from subtxn to top-level txn, if needed
1130 : *
1131 : * This is done if the top-level txn doesn't have a base snapshot, or if the
1132 : * subtxn's base snapshot has an earlier LSN than the top-level txn's base
1133 : * snapshot's LSN. This can happen if there are no changes in the toplevel
1134 : * txn but there are some in the subtxn, or the first change in subtxn has
1135 : * earlier LSN than first change in the top-level txn and we learned about
1136 : * their kinship only now.
1137 : *
1138 : * The subtransaction's snapshot is cleared regardless of the transfer
1139 : * happening, since it's not needed anymore in either case.
1140 : *
1141 : * We do this as soon as we become aware of their kinship, to avoid queueing
1142 : * extra snapshots to txns known-as-subtxns -- only top-level txns will
1143 : * receive further snapshots.
1144 : */
1145 : static void
1146 1304 : ReorderBufferTransferSnapToParent(ReorderBufferTXN *txn,
1147 : ReorderBufferTXN *subtxn)
1148 : {
1149 : Assert(subtxn->toplevel_xid == txn->xid);
1150 :
1151 1304 : if (subtxn->base_snapshot != NULL)
1152 : {
1153 0 : if (txn->base_snapshot == NULL ||
1154 0 : subtxn->base_snapshot_lsn < txn->base_snapshot_lsn)
1155 : {
1156 : /*
1157 : * If the toplevel transaction already has a base snapshot but
1158 : * it's newer than the subxact's, purge it.
1159 : */
1160 0 : if (txn->base_snapshot != NULL)
1161 : {
1162 0 : SnapBuildSnapDecRefcount(txn->base_snapshot);
1163 0 : dlist_delete(&txn->base_snapshot_node);
1164 : }
1165 :
1166 : /*
1167 : * The snapshot is now the top transaction's; transfer it, and
1168 : * adjust the list position of the top transaction in the list by
1169 : * moving it to where the subtransaction is.
1170 : */
1171 0 : txn->base_snapshot = subtxn->base_snapshot;
1172 0 : txn->base_snapshot_lsn = subtxn->base_snapshot_lsn;
1173 0 : dlist_insert_before(&subtxn->base_snapshot_node,
1174 : &txn->base_snapshot_node);
1175 :
1176 : /*
1177 : * The subtransaction doesn't have a snapshot anymore (so it
1178 : * mustn't be in the list.)
1179 : */
1180 0 : subtxn->base_snapshot = NULL;
1181 0 : subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
1182 0 : dlist_delete(&subtxn->base_snapshot_node);
1183 : }
1184 : else
1185 : {
1186 : /* Base snap of toplevel is fine, so subxact's is not needed */
1187 0 : SnapBuildSnapDecRefcount(subtxn->base_snapshot);
1188 0 : dlist_delete(&subtxn->base_snapshot_node);
1189 0 : subtxn->base_snapshot = NULL;
1190 0 : subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
1191 : }
1192 : }
1193 1304 : }
1194 :
1195 : /*
1196 : * Associate a subtransaction with its toplevel transaction at commit
1197 : * time. There may be no further changes added after this.
1198 : */
1199 : void
1200 536 : ReorderBufferCommitChild(ReorderBuffer *rb, TransactionId xid,
1201 : TransactionId subxid, XLogRecPtr commit_lsn,
1202 : XLogRecPtr end_lsn)
1203 : {
1204 : ReorderBufferTXN *subtxn;
1205 :
1206 536 : subtxn = ReorderBufferTXNByXid(rb, subxid, false, NULL,
1207 : InvalidXLogRecPtr, false);
1208 :
1209 : /*
1210 : * No need to do anything if that subtxn didn't contain any changes
1211 : */
1212 536 : if (!subtxn)
1213 162 : return;
1214 :
1215 374 : subtxn->final_lsn = commit_lsn;
1216 374 : subtxn->end_lsn = end_lsn;
1217 :
1218 : /*
1219 : * Assign this subxact as a child of the toplevel xact (no-op if already
1220 : * done.)
1221 : */
1222 374 : ReorderBufferAssignChild(rb, xid, subxid, InvalidXLogRecPtr);
1223 : }
1224 :
1225 :
1226 : /*
1227 : * Support for efficiently iterating over a transaction's and its
1228 : * subtransactions' changes.
1229 : *
1230 : * We do by doing a k-way merge between transactions/subtransactions. For that
1231 : * we model the current heads of the different transactions as a binary heap
1232 : * so we easily know which (sub-)transaction has the change with the smallest
1233 : * lsn next.
1234 : *
1235 : * We assume the changes in individual transactions are already sorted by LSN.
1236 : */
1237 :
1238 : /*
1239 : * Binary heap comparison function.
1240 : */
1241 : static int
1242 103142 : ReorderBufferIterCompare(Datum a, Datum b, void *arg)
1243 : {
1244 103142 : ReorderBufferIterTXNState *state = (ReorderBufferIterTXNState *) arg;
1245 103142 : XLogRecPtr pos_a = state->entries[DatumGetInt32(a)].lsn;
1246 103142 : XLogRecPtr pos_b = state->entries[DatumGetInt32(b)].lsn;
1247 :
1248 103142 : if (pos_a < pos_b)
1249 101428 : return 1;
1250 1714 : else if (pos_a == pos_b)
1251 0 : return 0;
1252 1714 : return -1;
1253 : }
1254 :
1255 : /*
1256 : * Allocate & initialize an iterator which iterates in lsn order over a
1257 : * transaction and all its subtransactions.
1258 : *
1259 : * Note: The iterator state is returned through iter_state parameter rather
1260 : * than the function's return value. This is because the state gets cleaned up
1261 : * in a PG_CATCH block in the caller, so we want to make sure the caller gets
1262 : * back the state even if this function throws an exception.
1263 : */
1264 : static void
1265 3804 : ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn,
1266 : ReorderBufferIterTXNState *volatile *iter_state)
1267 : {
1268 3804 : Size nr_txns = 0;
1269 : ReorderBufferIterTXNState *state;
1270 : dlist_iter cur_txn_i;
1271 : int32 off;
1272 :
1273 3804 : *iter_state = NULL;
1274 :
1275 : /* Check ordering of changes in the toplevel transaction. */
1276 3804 : AssertChangeLsnOrder(txn);
1277 :
1278 : /*
1279 : * Calculate the size of our heap: one element for every transaction that
1280 : * contains changes. (Besides the transactions already in the reorder
1281 : * buffer, we count the one we were directly passed.)
1282 : */
1283 3804 : if (txn->nentries > 0)
1284 3438 : nr_txns++;
1285 :
1286 4732 : dlist_foreach(cur_txn_i, &txn->subtxns)
1287 : {
1288 : ReorderBufferTXN *cur_txn;
1289 :
1290 928 : cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
1291 :
1292 : /* Check ordering of changes in this subtransaction. */
1293 928 : AssertChangeLsnOrder(cur_txn);
1294 :
1295 928 : if (cur_txn->nentries > 0)
1296 604 : nr_txns++;
1297 : }
1298 :
1299 : /* allocate iteration state */
1300 : state = (ReorderBufferIterTXNState *)
1301 3804 : MemoryContextAllocZero(rb->context,
1302 : sizeof(ReorderBufferIterTXNState) +
1303 3804 : sizeof(ReorderBufferIterTXNEntry) * nr_txns);
1304 :
1305 3804 : state->nr_txns = nr_txns;
1306 3804 : dlist_init(&state->old_change);
1307 :
1308 7846 : for (off = 0; off < state->nr_txns; off++)
1309 : {
1310 4042 : state->entries[off].file.vfd = -1;
1311 4042 : state->entries[off].segno = 0;
1312 : }
1313 :
1314 : /* allocate heap */
1315 3804 : state->heap = binaryheap_allocate(state->nr_txns,
1316 : ReorderBufferIterCompare,
1317 : state);
1318 :
1319 : /* Now that the state fields are initialized, it is safe to return it. */
1320 3804 : *iter_state = state;
1321 :
1322 : /*
1323 : * Now insert items into the binary heap, in an unordered fashion. (We
1324 : * will run a heap assembly step at the end; this is more efficient.)
1325 : */
1326 :
1327 3804 : off = 0;
1328 :
1329 : /* add toplevel transaction if it contains changes */
1330 3804 : if (txn->nentries > 0)
1331 : {
1332 : ReorderBufferChange *cur_change;
1333 :
1334 3438 : if (rbtxn_is_serialized(txn))
1335 : {
1336 : /* serialize remaining changes */
1337 46 : ReorderBufferSerializeTXN(rb, txn);
1338 46 : ReorderBufferRestoreChanges(rb, txn, &state->entries[off].file,
1339 : &state->entries[off].segno);
1340 : }
1341 :
1342 3438 : cur_change = dlist_head_element(ReorderBufferChange, node,
1343 : &txn->changes);
1344 :
1345 3438 : state->entries[off].lsn = cur_change->lsn;
1346 3438 : state->entries[off].change = cur_change;
1347 3438 : state->entries[off].txn = txn;
1348 :
1349 3438 : binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
1350 : }
1351 :
1352 : /* add subtransactions if they contain changes */
1353 4732 : dlist_foreach(cur_txn_i, &txn->subtxns)
1354 : {
1355 : ReorderBufferTXN *cur_txn;
1356 :
1357 928 : cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
1358 :
1359 928 : if (cur_txn->nentries > 0)
1360 : {
1361 : ReorderBufferChange *cur_change;
1362 :
1363 604 : if (rbtxn_is_serialized(cur_txn))
1364 : {
1365 : /* serialize remaining changes */
1366 34 : ReorderBufferSerializeTXN(rb, cur_txn);
1367 34 : ReorderBufferRestoreChanges(rb, cur_txn,
1368 : &state->entries[off].file,
1369 : &state->entries[off].segno);
1370 : }
1371 604 : cur_change = dlist_head_element(ReorderBufferChange, node,
1372 : &cur_txn->changes);
1373 :
1374 604 : state->entries[off].lsn = cur_change->lsn;
1375 604 : state->entries[off].change = cur_change;
1376 604 : state->entries[off].txn = cur_txn;
1377 :
1378 604 : binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
1379 : }
1380 : }
1381 :
1382 : /* assemble a valid binary heap */
1383 3804 : binaryheap_build(state->heap);
1384 3804 : }
1385 :
1386 : /*
1387 : * Return the next change when iterating over a transaction and its
1388 : * subtransactions.
1389 : *
1390 : * Returns NULL when no further changes exist.
1391 : */
1392 : static ReorderBufferChange *
1393 714844 : ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state)
1394 : {
1395 : ReorderBufferChange *change;
1396 : ReorderBufferIterTXNEntry *entry;
1397 : int32 off;
1398 :
1399 : /* nothing there anymore */
1400 714844 : if (state->heap->bh_size == 0)
1401 3782 : return NULL;
1402 :
1403 711062 : off = DatumGetInt32(binaryheap_first(state->heap));
1404 711062 : entry = &state->entries[off];
1405 :
1406 : /* free memory we might have "leaked" in the previous *Next call */
1407 711062 : if (!dlist_is_empty(&state->old_change))
1408 : {
1409 90 : change = dlist_container(ReorderBufferChange, node,
1410 : dlist_pop_head_node(&state->old_change));
1411 90 : ReorderBufferReturnChange(rb, change, true);
1412 : Assert(dlist_is_empty(&state->old_change));
1413 : }
1414 :
1415 711062 : change = entry->change;
1416 :
1417 : /*
1418 : * update heap with information about which transaction has the next
1419 : * relevant change in LSN order
1420 : */
1421 :
1422 : /* there are in-memory changes */
1423 711062 : if (dlist_has_next(&entry->txn->changes, &entry->change->node))
1424 : {
1425 706954 : dlist_node *next = dlist_next_node(&entry->txn->changes, &change->node);
1426 706954 : ReorderBufferChange *next_change =
1427 706954 : dlist_container(ReorderBufferChange, node, next);
1428 :
1429 : /* txn stays the same */
1430 706954 : state->entries[off].lsn = next_change->lsn;
1431 706954 : state->entries[off].change = next_change;
1432 :
1433 706954 : binaryheap_replace_first(state->heap, Int32GetDatum(off));
1434 706954 : return change;
1435 : }
1436 :
1437 : /* try to load changes from disk */
1438 4108 : if (entry->txn->nentries != entry->txn->nentries_mem)
1439 : {
1440 : /*
1441 : * Ugly: restoring changes will reuse *Change records, thus delete the
1442 : * current one from the per-tx list and only free in the next call.
1443 : */
1444 130 : dlist_delete(&change->node);
1445 130 : dlist_push_tail(&state->old_change, &change->node);
1446 :
1447 : /*
1448 : * Update the total bytes processed by the txn for which we are
1449 : * releasing the current set of changes and restoring the new set of
1450 : * changes.
1451 : */
1452 130 : rb->totalBytes += entry->txn->size;
1453 130 : if (ReorderBufferRestoreChanges(rb, entry->txn, &entry->file,
1454 : &state->entries[off].segno))
1455 : {
1456 : /* successfully restored changes from disk */
1457 : ReorderBufferChange *next_change =
1458 72 : dlist_head_element(ReorderBufferChange, node,
1459 : &entry->txn->changes);
1460 :
1461 72 : elog(DEBUG2, "restored %u/%u changes from disk",
1462 : (uint32) entry->txn->nentries_mem,
1463 : (uint32) entry->txn->nentries);
1464 :
1465 : Assert(entry->txn->nentries_mem);
1466 : /* txn stays the same */
1467 72 : state->entries[off].lsn = next_change->lsn;
1468 72 : state->entries[off].change = next_change;
1469 72 : binaryheap_replace_first(state->heap, Int32GetDatum(off));
1470 :
1471 72 : return change;
1472 : }
1473 : }
1474 :
1475 : /* ok, no changes there anymore, remove */
1476 4036 : binaryheap_remove_first(state->heap);
1477 :
1478 4036 : return change;
1479 : }
1480 :
1481 : /*
1482 : * Deallocate the iterator
1483 : */
1484 : static void
1485 3804 : ReorderBufferIterTXNFinish(ReorderBuffer *rb,
1486 : ReorderBufferIterTXNState *state)
1487 : {
1488 : int32 off;
1489 :
1490 7846 : for (off = 0; off < state->nr_txns; off++)
1491 : {
1492 4042 : if (state->entries[off].file.vfd != -1)
1493 0 : FileClose(state->entries[off].file.vfd);
1494 : }
1495 :
1496 : /* free memory we might have "leaked" in the last *Next call */
1497 3804 : if (!dlist_is_empty(&state->old_change))
1498 : {
1499 : ReorderBufferChange *change;
1500 :
1501 38 : change = dlist_container(ReorderBufferChange, node,
1502 : dlist_pop_head_node(&state->old_change));
1503 38 : ReorderBufferReturnChange(rb, change, true);
1504 : Assert(dlist_is_empty(&state->old_change));
1505 : }
1506 :
1507 3804 : binaryheap_free(state->heap);
1508 3804 : pfree(state);
1509 3804 : }
1510 :
1511 : /*
1512 : * Cleanup the contents of a transaction, usually after the transaction
1513 : * committed or aborted.
1514 : */
1515 : static void
1516 6914 : ReorderBufferCleanupTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
1517 : {
1518 : bool found;
1519 : dlist_mutable_iter iter;
1520 6914 : Size mem_freed = 0;
1521 :
1522 : /* cleanup subtransactions & their changes */
1523 7286 : dlist_foreach_modify(iter, &txn->subtxns)
1524 : {
1525 : ReorderBufferTXN *subtxn;
1526 :
1527 372 : subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1528 :
1529 : /*
1530 : * Subtransactions are always associated to the toplevel TXN, even if
1531 : * they originally were happening inside another subtxn, so we won't
1532 : * ever recurse more than one level deep here.
1533 : */
1534 : Assert(rbtxn_is_known_subxact(subtxn));
1535 : Assert(subtxn->nsubtxns == 0);
1536 :
1537 372 : ReorderBufferCleanupTXN(rb, subtxn);
1538 : }
1539 :
1540 : /* cleanup changes in the txn */
1541 141940 : dlist_foreach_modify(iter, &txn->changes)
1542 : {
1543 : ReorderBufferChange *change;
1544 :
1545 135026 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1546 :
1547 : /* Check we're not mixing changes from different transactions. */
1548 : Assert(change->txn == txn);
1549 :
1550 : /*
1551 : * Instead of updating the memory counter for individual changes, we
1552 : * sum up the size of memory to free so we can update the memory
1553 : * counter all together below. This saves costs of maintaining the
1554 : * max-heap.
1555 : */
1556 135026 : mem_freed += ReorderBufferChangeSize(change);
1557 :
1558 135026 : ReorderBufferReturnChange(rb, change, false);
1559 : }
1560 :
1561 : /* Update the memory counter */
1562 6914 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, mem_freed);
1563 :
1564 : /*
1565 : * Cleanup the tuplecids we stored for decoding catalog snapshot access.
1566 : * They are always stored in the toplevel transaction.
1567 : */
1568 54416 : dlist_foreach_modify(iter, &txn->tuplecids)
1569 : {
1570 : ReorderBufferChange *change;
1571 :
1572 47502 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1573 :
1574 : /* Check we're not mixing changes from different transactions. */
1575 : Assert(change->txn == txn);
1576 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1577 :
1578 47502 : ReorderBufferReturnChange(rb, change, true);
1579 : }
1580 :
1581 : /*
1582 : * Cleanup the base snapshot, if set.
1583 : */
1584 6914 : if (txn->base_snapshot != NULL)
1585 : {
1586 5568 : SnapBuildSnapDecRefcount(txn->base_snapshot);
1587 5568 : dlist_delete(&txn->base_snapshot_node);
1588 : }
1589 :
1590 : /*
1591 : * Cleanup the snapshot for the last streamed run.
1592 : */
1593 6914 : if (txn->snapshot_now != NULL)
1594 : {
1595 : Assert(rbtxn_is_streamed(txn));
1596 132 : ReorderBufferFreeSnap(rb, txn->snapshot_now);
1597 : }
1598 :
1599 : /*
1600 : * Remove TXN from its containing lists.
1601 : *
1602 : * Note: if txn is known as subxact, we are deleting the TXN from its
1603 : * parent's list of known subxacts; this leaves the parent's nsubxacts
1604 : * count too high, but we don't care. Otherwise, we are deleting the TXN
1605 : * from the LSN-ordered list of toplevel TXNs. We remove the TXN from the
1606 : * list of catalog modifying transactions as well.
1607 : */
1608 6914 : dlist_delete(&txn->node);
1609 6914 : if (rbtxn_has_catalog_changes(txn))
1610 2160 : dclist_delete_from(&rb->catchange_txns, &txn->catchange_node);
1611 :
1612 : /* now remove reference from buffer */
1613 6914 : hash_search(rb->by_txn, &txn->xid, HASH_REMOVE, &found);
1614 : Assert(found);
1615 :
1616 : /* remove entries spilled to disk */
1617 6914 : if (rbtxn_is_serialized(txn))
1618 510 : ReorderBufferRestoreCleanup(rb, txn);
1619 :
1620 : /* deallocate */
1621 6914 : ReorderBufferReturnTXN(rb, txn);
1622 6914 : }
1623 :
1624 : /*
1625 : * Discard changes from a transaction (and subtransactions), either after
1626 : * streaming, decoding them at PREPARE, or detecting the transaction abort.
1627 : * Keep the remaining info - transactions, tuplecids, invalidations and
1628 : * snapshots.
1629 : *
1630 : * We additionally remove tuplecids after decoding the transaction at prepare
1631 : * time as we only need to perform invalidation at rollback or commit prepared.
1632 : *
1633 : * 'txn_prepared' indicates that we have decoded the transaction at prepare
1634 : * time.
1635 : */
1636 : static void
1637 2130 : ReorderBufferTruncateTXN(ReorderBuffer *rb, ReorderBufferTXN *txn, bool txn_prepared)
1638 : {
1639 : dlist_mutable_iter iter;
1640 2130 : Size mem_freed = 0;
1641 :
1642 : /* cleanup subtransactions & their changes */
1643 2724 : dlist_foreach_modify(iter, &txn->subtxns)
1644 : {
1645 : ReorderBufferTXN *subtxn;
1646 :
1647 594 : subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1648 :
1649 : /*
1650 : * Subtransactions are always associated to the toplevel TXN, even if
1651 : * they originally were happening inside another subtxn, so we won't
1652 : * ever recurse more than one level deep here.
1653 : */
1654 : Assert(rbtxn_is_known_subxact(subtxn));
1655 : Assert(subtxn->nsubtxns == 0);
1656 :
1657 594 : ReorderBufferMaybeMarkTXNStreamed(rb, subtxn);
1658 594 : ReorderBufferTruncateTXN(rb, subtxn, txn_prepared);
1659 : }
1660 :
1661 : /* cleanup changes in the txn */
1662 317090 : dlist_foreach_modify(iter, &txn->changes)
1663 : {
1664 : ReorderBufferChange *change;
1665 :
1666 314960 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1667 :
1668 : /* Check we're not mixing changes from different transactions. */
1669 : Assert(change->txn == txn);
1670 :
1671 : /* remove the change from its containing list */
1672 314960 : dlist_delete(&change->node);
1673 :
1674 : /*
1675 : * Instead of updating the memory counter for individual changes, we
1676 : * sum up the size of memory to free so we can update the memory
1677 : * counter all together below. This saves costs of maintaining the
1678 : * max-heap.
1679 : */
1680 314960 : mem_freed += ReorderBufferChangeSize(change);
1681 :
1682 314960 : ReorderBufferReturnChange(rb, change, false);
1683 : }
1684 :
1685 : /* Update the memory counter */
1686 2130 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, mem_freed);
1687 :
1688 2130 : if (txn_prepared)
1689 : {
1690 : /*
1691 : * If this is a prepared txn, cleanup the tuplecids we stored for
1692 : * decoding catalog snapshot access. They are always stored in the
1693 : * toplevel transaction.
1694 : */
1695 366 : dlist_foreach_modify(iter, &txn->tuplecids)
1696 : {
1697 : ReorderBufferChange *change;
1698 :
1699 246 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1700 :
1701 : /* Check we're not mixing changes from different transactions. */
1702 : Assert(change->txn == txn);
1703 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1704 :
1705 : /* Remove the change from its containing list. */
1706 246 : dlist_delete(&change->node);
1707 :
1708 246 : ReorderBufferReturnChange(rb, change, true);
1709 : }
1710 : }
1711 :
1712 : /*
1713 : * Destroy the (relfilelocator, ctid) hashtable, so that we don't leak any
1714 : * memory. We could also keep the hash table and update it with new ctid
1715 : * values, but this seems simpler and good enough for now.
1716 : */
1717 2130 : if (txn->tuplecid_hash != NULL)
1718 : {
1719 102 : hash_destroy(txn->tuplecid_hash);
1720 102 : txn->tuplecid_hash = NULL;
1721 : }
1722 :
1723 : /* If this txn is serialized then clean the disk space. */
1724 2130 : if (rbtxn_is_serialized(txn))
1725 : {
1726 18 : ReorderBufferRestoreCleanup(rb, txn);
1727 18 : txn->txn_flags &= ~RBTXN_IS_SERIALIZED;
1728 :
1729 : /*
1730 : * We set this flag to indicate if the transaction is ever serialized.
1731 : * We need this to accurately update the stats as otherwise the same
1732 : * transaction can be counted as serialized multiple times.
1733 : */
1734 18 : txn->txn_flags |= RBTXN_IS_SERIALIZED_CLEAR;
1735 : }
1736 :
1737 : /* also reset the number of entries in the transaction */
1738 2130 : txn->nentries_mem = 0;
1739 2130 : txn->nentries = 0;
1740 2130 : }
1741 :
1742 : /*
1743 : * Check the transaction status by CLOG lookup and discard all changes if
1744 : * the transaction is aborted. The transaction status is cached in
1745 : * txn->txn_flags so we can skip future changes and avoid CLOG lookups on the
1746 : * next call.
1747 : *
1748 : * Return true if the transaction is aborted, otherwise return false.
1749 : *
1750 : * When the 'debug_logical_replication_streaming' is set to "immediate", we
1751 : * don't check the transaction status, meaning the caller will always process
1752 : * this transaction.
1753 : */
1754 : static bool
1755 8638 : ReorderBufferCheckAndTruncateAbortedTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
1756 : {
1757 : /* Quick return for regression tests */
1758 8638 : if (unlikely(debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_IMMEDIATE))
1759 1924 : return false;
1760 :
1761 : /*
1762 : * Quick return if the transaction status is already known.
1763 : */
1764 :
1765 6714 : if (rbtxn_is_committed(txn))
1766 5898 : return false;
1767 816 : if (rbtxn_is_aborted(txn))
1768 : {
1769 : /* Already-aborted transactions should not have any changes */
1770 : Assert(txn->size == 0);
1771 :
1772 0 : return true;
1773 : }
1774 :
1775 : /* Otherwise, check the transaction status using CLOG lookup */
1776 :
1777 816 : if (TransactionIdIsInProgress(txn->xid))
1778 376 : return false;
1779 :
1780 440 : if (TransactionIdDidCommit(txn->xid))
1781 : {
1782 : /*
1783 : * Remember the transaction is committed so that we can skip CLOG
1784 : * check next time, avoiding the pressure on CLOG lookup.
1785 : */
1786 : Assert(!rbtxn_is_aborted(txn));
1787 422 : txn->txn_flags |= RBTXN_IS_COMMITTED;
1788 422 : return false;
1789 : }
1790 :
1791 : /*
1792 : * The transaction aborted. We discard both the changes collected so far
1793 : * and the toast reconstruction data. The full cleanup will happen as part
1794 : * of decoding ABORT record of this transaction.
1795 : */
1796 18 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
1797 18 : ReorderBufferToastReset(rb, txn);
1798 :
1799 : /* All changes should be discarded */
1800 : Assert(txn->size == 0);
1801 :
1802 : /*
1803 : * Mark the transaction as aborted so we can ignore future changes of this
1804 : * transaction.
1805 : */
1806 : Assert(!rbtxn_is_committed(txn));
1807 18 : txn->txn_flags |= RBTXN_IS_ABORTED;
1808 :
1809 18 : return true;
1810 : }
1811 :
1812 : /*
1813 : * Build a hash with a (relfilelocator, ctid) -> (cmin, cmax) mapping for use by
1814 : * HeapTupleSatisfiesHistoricMVCC.
1815 : */
1816 : static void
1817 3804 : ReorderBufferBuildTupleCidHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
1818 : {
1819 : dlist_iter iter;
1820 : HASHCTL hash_ctl;
1821 :
1822 3804 : if (!rbtxn_has_catalog_changes(txn) || dlist_is_empty(&txn->tuplecids))
1823 2730 : return;
1824 :
1825 1074 : hash_ctl.keysize = sizeof(ReorderBufferTupleCidKey);
1826 1074 : hash_ctl.entrysize = sizeof(ReorderBufferTupleCidEnt);
1827 1074 : hash_ctl.hcxt = rb->context;
1828 :
1829 : /*
1830 : * create the hash with the exact number of to-be-stored tuplecids from
1831 : * the start
1832 : */
1833 1074 : txn->tuplecid_hash =
1834 1074 : hash_create("ReorderBufferTupleCid", txn->ntuplecids, &hash_ctl,
1835 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1836 :
1837 24124 : dlist_foreach(iter, &txn->tuplecids)
1838 : {
1839 : ReorderBufferTupleCidKey key;
1840 : ReorderBufferTupleCidEnt *ent;
1841 : bool found;
1842 : ReorderBufferChange *change;
1843 :
1844 23050 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1845 :
1846 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1847 :
1848 : /* be careful about padding */
1849 23050 : memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
1850 :
1851 23050 : key.rlocator = change->data.tuplecid.locator;
1852 :
1853 23050 : ItemPointerCopy(&change->data.tuplecid.tid,
1854 : &key.tid);
1855 :
1856 : ent = (ReorderBufferTupleCidEnt *)
1857 23050 : hash_search(txn->tuplecid_hash, &key, HASH_ENTER, &found);
1858 23050 : if (!found)
1859 : {
1860 19764 : ent->cmin = change->data.tuplecid.cmin;
1861 19764 : ent->cmax = change->data.tuplecid.cmax;
1862 19764 : ent->combocid = change->data.tuplecid.combocid;
1863 : }
1864 : else
1865 : {
1866 : /*
1867 : * Maybe we already saw this tuple before in this transaction, but
1868 : * if so it must have the same cmin.
1869 : */
1870 : Assert(ent->cmin == change->data.tuplecid.cmin);
1871 :
1872 : /*
1873 : * cmax may be initially invalid, but once set it can only grow,
1874 : * and never become invalid again.
1875 : */
1876 : Assert((ent->cmax == InvalidCommandId) ||
1877 : ((change->data.tuplecid.cmax != InvalidCommandId) &&
1878 : (change->data.tuplecid.cmax > ent->cmax)));
1879 3286 : ent->cmax = change->data.tuplecid.cmax;
1880 : }
1881 : }
1882 : }
1883 :
1884 : /*
1885 : * Copy a provided snapshot so we can modify it privately. This is needed so
1886 : * that catalog modifying transactions can look into intermediate catalog
1887 : * states.
1888 : */
1889 : static Snapshot
1890 3394 : ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
1891 : ReorderBufferTXN *txn, CommandId cid)
1892 : {
1893 : Snapshot snap;
1894 : dlist_iter iter;
1895 3394 : int i = 0;
1896 : Size size;
1897 :
1898 3394 : size = sizeof(SnapshotData) +
1899 3394 : sizeof(TransactionId) * orig_snap->xcnt +
1900 3394 : sizeof(TransactionId) * (txn->nsubtxns + 1);
1901 :
1902 3394 : snap = MemoryContextAllocZero(rb->context, size);
1903 3394 : memcpy(snap, orig_snap, sizeof(SnapshotData));
1904 :
1905 3394 : snap->copied = true;
1906 3394 : snap->active_count = 1; /* mark as active so nobody frees it */
1907 3394 : snap->regd_count = 0;
1908 3394 : snap->xip = (TransactionId *) (snap + 1);
1909 :
1910 3394 : memcpy(snap->xip, orig_snap->xip, sizeof(TransactionId) * snap->xcnt);
1911 :
1912 : /*
1913 : * snap->subxip contains all txids that belong to our transaction which we
1914 : * need to check via cmin/cmax. That's why we store the toplevel
1915 : * transaction in there as well.
1916 : */
1917 3394 : snap->subxip = snap->xip + snap->xcnt;
1918 3394 : snap->subxip[i++] = txn->xid;
1919 :
1920 : /*
1921 : * txn->nsubtxns isn't decreased when subtransactions abort, so count
1922 : * manually. Since it's an upper boundary it is safe to use it for the
1923 : * allocation above.
1924 : */
1925 3394 : snap->subxcnt = 1;
1926 :
1927 4016 : dlist_foreach(iter, &txn->subtxns)
1928 : {
1929 : ReorderBufferTXN *sub_txn;
1930 :
1931 622 : sub_txn = dlist_container(ReorderBufferTXN, node, iter.cur);
1932 622 : snap->subxip[i++] = sub_txn->xid;
1933 622 : snap->subxcnt++;
1934 : }
1935 :
1936 : /* sort so we can bsearch() later */
1937 3394 : qsort(snap->subxip, snap->subxcnt, sizeof(TransactionId), xidComparator);
1938 :
1939 : /* store the specified current CommandId */
1940 3394 : snap->curcid = cid;
1941 :
1942 3394 : return snap;
1943 : }
1944 :
1945 : /*
1946 : * Free a previously ReorderBufferCopySnap'ed snapshot
1947 : */
1948 : static void
1949 5474 : ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap)
1950 : {
1951 5474 : if (snap->copied)
1952 3386 : pfree(snap);
1953 : else
1954 2088 : SnapBuildSnapDecRefcount(snap);
1955 5474 : }
1956 :
1957 : /*
1958 : * If the transaction was (partially) streamed, we need to prepare or commit
1959 : * it in a 'streamed' way. That is, we first stream the remaining part of the
1960 : * transaction, and then invoke stream_prepare or stream_commit message as per
1961 : * the case.
1962 : */
1963 : static void
1964 132 : ReorderBufferStreamCommit(ReorderBuffer *rb, ReorderBufferTXN *txn)
1965 : {
1966 : /* we should only call this for previously streamed transactions */
1967 : Assert(rbtxn_is_streamed(txn));
1968 :
1969 132 : ReorderBufferStreamTXN(rb, txn);
1970 :
1971 132 : if (rbtxn_is_prepared(txn))
1972 : {
1973 : /*
1974 : * Note, we send stream prepare even if a concurrent abort is
1975 : * detected. See DecodePrepare for more information.
1976 : */
1977 : Assert(!rbtxn_sent_prepare(txn));
1978 30 : rb->stream_prepare(rb, txn, txn->final_lsn);
1979 30 : txn->txn_flags |= RBTXN_SENT_PREPARE;
1980 :
1981 : /*
1982 : * This is a PREPARED transaction, part of a two-phase commit. The
1983 : * full cleanup will happen as part of the COMMIT PREPAREDs, so now
1984 : * just truncate txn by removing changes and tuplecids.
1985 : */
1986 30 : ReorderBufferTruncateTXN(rb, txn, true);
1987 : /* Reset the CheckXidAlive */
1988 30 : CheckXidAlive = InvalidTransactionId;
1989 : }
1990 : else
1991 : {
1992 102 : rb->stream_commit(rb, txn, txn->final_lsn);
1993 102 : ReorderBufferCleanupTXN(rb, txn);
1994 : }
1995 132 : }
1996 :
1997 : /*
1998 : * Set xid to detect concurrent aborts.
1999 : *
2000 : * While streaming an in-progress transaction or decoding a prepared
2001 : * transaction there is a possibility that the (sub)transaction might get
2002 : * aborted concurrently. In such case if the (sub)transaction has catalog
2003 : * update then we might decode the tuple using wrong catalog version. For
2004 : * example, suppose there is one catalog tuple with (xmin: 500, xmax: 0). Now,
2005 : * the transaction 501 updates the catalog tuple and after that we will have
2006 : * two tuples (xmin: 500, xmax: 501) and (xmin: 501, xmax: 0). Now, if 501 is
2007 : * aborted and some other transaction say 502 updates the same catalog tuple
2008 : * then the first tuple will be changed to (xmin: 500, xmax: 502). So, the
2009 : * problem is that when we try to decode the tuple inserted/updated in 501
2010 : * after the catalog update, we will see the catalog tuple with (xmin: 500,
2011 : * xmax: 502) as visible because it will consider that the tuple is deleted by
2012 : * xid 502 which is not visible to our snapshot. And when we will try to
2013 : * decode with that catalog tuple, it can lead to a wrong result or a crash.
2014 : * So, it is necessary to detect concurrent aborts to allow streaming of
2015 : * in-progress transactions or decoding of prepared transactions.
2016 : *
2017 : * For detecting the concurrent abort we set CheckXidAlive to the current
2018 : * (sub)transaction's xid for which this change belongs to. And, during
2019 : * catalog scan we can check the status of the xid and if it is aborted we will
2020 : * report a specific error so that we can stop streaming current transaction
2021 : * and discard the already streamed changes on such an error. We might have
2022 : * already streamed some of the changes for the aborted (sub)transaction, but
2023 : * that is fine because when we decode the abort we will stream abort message
2024 : * to truncate the changes in the subscriber. Similarly, for prepared
2025 : * transactions, we stop decoding if concurrent abort is detected and then
2026 : * rollback the changes when rollback prepared is encountered. See
2027 : * DecodePrepare.
2028 : */
2029 : static inline void
2030 355740 : SetupCheckXidLive(TransactionId xid)
2031 : {
2032 : /*
2033 : * If the input transaction id is already set as a CheckXidAlive then
2034 : * nothing to do.
2035 : */
2036 355740 : if (TransactionIdEquals(CheckXidAlive, xid))
2037 152728 : return;
2038 :
2039 : /*
2040 : * setup CheckXidAlive if it's not committed yet. We don't check if the
2041 : * xid is aborted. That will happen during catalog access.
2042 : */
2043 203012 : if (!TransactionIdDidCommit(xid))
2044 662 : CheckXidAlive = xid;
2045 : else
2046 202350 : CheckXidAlive = InvalidTransactionId;
2047 : }
2048 :
2049 : /*
2050 : * Helper function for ReorderBufferProcessTXN for applying change.
2051 : */
2052 : static inline void
2053 667972 : ReorderBufferApplyChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
2054 : Relation relation, ReorderBufferChange *change,
2055 : bool streaming)
2056 : {
2057 667972 : if (streaming)
2058 352012 : rb->stream_change(rb, txn, relation, change);
2059 : else
2060 315960 : rb->apply_change(rb, txn, relation, change);
2061 667966 : }
2062 :
2063 : /*
2064 : * Helper function for ReorderBufferProcessTXN for applying the truncate.
2065 : */
2066 : static inline void
2067 46 : ReorderBufferApplyTruncate(ReorderBuffer *rb, ReorderBufferTXN *txn,
2068 : int nrelations, Relation *relations,
2069 : ReorderBufferChange *change, bool streaming)
2070 : {
2071 46 : if (streaming)
2072 0 : rb->stream_truncate(rb, txn, nrelations, relations, change);
2073 : else
2074 46 : rb->apply_truncate(rb, txn, nrelations, relations, change);
2075 46 : }
2076 :
2077 : /*
2078 : * Helper function for ReorderBufferProcessTXN for applying the message.
2079 : */
2080 : static inline void
2081 22 : ReorderBufferApplyMessage(ReorderBuffer *rb, ReorderBufferTXN *txn,
2082 : ReorderBufferChange *change, bool streaming)
2083 : {
2084 22 : if (streaming)
2085 6 : rb->stream_message(rb, txn, change->lsn, true,
2086 6 : change->data.msg.prefix,
2087 : change->data.msg.message_size,
2088 6 : change->data.msg.message);
2089 : else
2090 16 : rb->message(rb, txn, change->lsn, true,
2091 16 : change->data.msg.prefix,
2092 : change->data.msg.message_size,
2093 16 : change->data.msg.message);
2094 22 : }
2095 :
2096 : /*
2097 : * Function to store the command id and snapshot at the end of the current
2098 : * stream so that we can reuse the same while sending the next stream.
2099 : */
2100 : static inline void
2101 1432 : ReorderBufferSaveTXNSnapshot(ReorderBuffer *rb, ReorderBufferTXN *txn,
2102 : Snapshot snapshot_now, CommandId command_id)
2103 : {
2104 1432 : txn->command_id = command_id;
2105 :
2106 : /* Avoid copying if it's already copied. */
2107 1432 : if (snapshot_now->copied)
2108 1432 : txn->snapshot_now = snapshot_now;
2109 : else
2110 0 : txn->snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
2111 : txn, command_id);
2112 1432 : }
2113 :
2114 : /*
2115 : * Mark the given transaction as streamed if it's a top-level transaction
2116 : * or has changes.
2117 : */
2118 : static void
2119 2026 : ReorderBufferMaybeMarkTXNStreamed(ReorderBuffer *rb, ReorderBufferTXN *txn)
2120 : {
2121 : /*
2122 : * The top-level transaction, is marked as streamed always, even if it
2123 : * does not contain any changes (that is, when all the changes are in
2124 : * subtransactions).
2125 : *
2126 : * For subtransactions, we only mark them as streamed when there are
2127 : * changes in them.
2128 : *
2129 : * We do it this way because of aborts - we don't want to send aborts for
2130 : * XIDs the downstream is not aware of. And of course, it always knows
2131 : * about the top-level xact (we send the XID in all messages), but we
2132 : * never stream XIDs of empty subxacts.
2133 : */
2134 2026 : if (rbtxn_is_toptxn(txn) || (txn->nentries_mem != 0))
2135 1702 : txn->txn_flags |= RBTXN_IS_STREAMED;
2136 2026 : }
2137 :
2138 : /*
2139 : * Helper function for ReorderBufferProcessTXN to handle the concurrent
2140 : * abort of the streaming transaction. This resets the TXN such that it
2141 : * can be used to stream the remaining data of transaction being processed.
2142 : * This can happen when the subtransaction is aborted and we still want to
2143 : * continue processing the main or other subtransactions data.
2144 : */
2145 : static void
2146 16 : ReorderBufferResetTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
2147 : Snapshot snapshot_now,
2148 : CommandId command_id,
2149 : XLogRecPtr last_lsn,
2150 : ReorderBufferChange *specinsert)
2151 : {
2152 : /* Discard the changes that we just streamed */
2153 16 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
2154 :
2155 : /* Free all resources allocated for toast reconstruction */
2156 16 : ReorderBufferToastReset(rb, txn);
2157 :
2158 : /* Return the spec insert change if it is not NULL */
2159 16 : if (specinsert != NULL)
2160 : {
2161 0 : ReorderBufferReturnChange(rb, specinsert, true);
2162 0 : specinsert = NULL;
2163 : }
2164 :
2165 : /*
2166 : * For the streaming case, stop the stream and remember the command ID and
2167 : * snapshot for the streaming run.
2168 : */
2169 16 : if (rbtxn_is_streamed(txn))
2170 : {
2171 16 : rb->stream_stop(rb, txn, last_lsn);
2172 16 : ReorderBufferSaveTXNSnapshot(rb, txn, snapshot_now, command_id);
2173 : }
2174 :
2175 : /* All changes must be deallocated */
2176 : Assert(txn->size == 0);
2177 16 : }
2178 :
2179 : /*
2180 : * Helper function for ReorderBufferReplay and ReorderBufferStreamTXN.
2181 : *
2182 : * Send data of a transaction (and its subtransactions) to the
2183 : * output plugin. We iterate over the top and subtransactions (using a k-way
2184 : * merge) and replay the changes in lsn order.
2185 : *
2186 : * If streaming is true then data will be sent using stream API.
2187 : *
2188 : * Note: "volatile" markers on some parameters are to avoid trouble with
2189 : * PG_TRY inside the function.
2190 : */
2191 : static void
2192 3804 : ReorderBufferProcessTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
2193 : XLogRecPtr commit_lsn,
2194 : volatile Snapshot snapshot_now,
2195 : volatile CommandId command_id,
2196 : bool streaming)
2197 : {
2198 : bool using_subtxn;
2199 3804 : MemoryContext ccxt = CurrentMemoryContext;
2200 3804 : ReorderBufferIterTXNState *volatile iterstate = NULL;
2201 3804 : volatile XLogRecPtr prev_lsn = InvalidXLogRecPtr;
2202 3804 : ReorderBufferChange *volatile specinsert = NULL;
2203 3804 : volatile bool stream_started = false;
2204 3804 : ReorderBufferTXN *volatile curtxn = NULL;
2205 :
2206 : /* build data to be able to lookup the CommandIds of catalog tuples */
2207 3804 : ReorderBufferBuildTupleCidHash(rb, txn);
2208 :
2209 : /* setup the initial snapshot */
2210 3804 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2211 :
2212 : /*
2213 : * Decoding needs access to syscaches et al., which in turn use
2214 : * heavyweight locks and such. Thus we need to have enough state around to
2215 : * keep track of those. The easiest way is to simply use a transaction
2216 : * internally. That also allows us to easily enforce that nothing writes
2217 : * to the database by checking for xid assignments.
2218 : *
2219 : * When we're called via the SQL SRF there's already a transaction
2220 : * started, so start an explicit subtransaction there.
2221 : */
2222 3804 : using_subtxn = IsTransactionOrTransactionBlock();
2223 :
2224 3804 : PG_TRY();
2225 : {
2226 : ReorderBufferChange *change;
2227 3804 : int changes_count = 0; /* used to accumulate the number of
2228 : * changes */
2229 :
2230 3804 : if (using_subtxn)
2231 962 : BeginInternalSubTransaction(streaming ? "stream" : "replay");
2232 : else
2233 2842 : StartTransactionCommand();
2234 :
2235 : /*
2236 : * We only need to send begin/begin-prepare for non-streamed
2237 : * transactions.
2238 : */
2239 3804 : if (!streaming)
2240 : {
2241 2372 : if (rbtxn_is_prepared(txn))
2242 56 : rb->begin_prepare(rb, txn);
2243 : else
2244 2316 : rb->begin(rb, txn);
2245 : }
2246 :
2247 3804 : ReorderBufferIterTXNInit(rb, txn, &iterstate);
2248 714844 : while ((change = ReorderBufferIterTXNNext(rb, iterstate)) != NULL)
2249 : {
2250 711062 : Relation relation = NULL;
2251 : Oid reloid;
2252 :
2253 711062 : CHECK_FOR_INTERRUPTS();
2254 :
2255 : /*
2256 : * We can't call start stream callback before processing first
2257 : * change.
2258 : */
2259 711062 : if (prev_lsn == InvalidXLogRecPtr)
2260 : {
2261 3724 : if (streaming)
2262 : {
2263 1354 : txn->origin_id = change->origin_id;
2264 1354 : rb->stream_start(rb, txn, change->lsn);
2265 1354 : stream_started = true;
2266 : }
2267 : }
2268 :
2269 : /*
2270 : * Enforce correct ordering of changes, merged from multiple
2271 : * subtransactions. The changes may have the same LSN due to
2272 : * MULTI_INSERT xlog records.
2273 : */
2274 : Assert(prev_lsn == InvalidXLogRecPtr || prev_lsn <= change->lsn);
2275 :
2276 711062 : prev_lsn = change->lsn;
2277 :
2278 : /*
2279 : * Set the current xid to detect concurrent aborts. This is
2280 : * required for the cases when we decode the changes before the
2281 : * COMMIT record is processed.
2282 : */
2283 711062 : if (streaming || rbtxn_is_prepared(change->txn))
2284 : {
2285 355740 : curtxn = change->txn;
2286 355740 : SetupCheckXidLive(curtxn->xid);
2287 : }
2288 :
2289 711062 : switch (change->action)
2290 : {
2291 3564 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
2292 :
2293 : /*
2294 : * Confirmation for speculative insertion arrived. Simply
2295 : * use as a normal record. It'll be cleaned up at the end
2296 : * of INSERT processing.
2297 : */
2298 3564 : if (specinsert == NULL)
2299 0 : elog(ERROR, "invalid ordering of speculative insertion changes");
2300 : Assert(specinsert->data.tp.oldtuple == NULL);
2301 3564 : change = specinsert;
2302 3564 : change->action = REORDER_BUFFER_CHANGE_INSERT;
2303 :
2304 : /* intentionally fall through */
2305 680818 : case REORDER_BUFFER_CHANGE_INSERT:
2306 : case REORDER_BUFFER_CHANGE_UPDATE:
2307 : case REORDER_BUFFER_CHANGE_DELETE:
2308 : Assert(snapshot_now);
2309 :
2310 680818 : reloid = RelidByRelfilenumber(change->data.tp.rlocator.spcOid,
2311 : change->data.tp.rlocator.relNumber);
2312 :
2313 : /*
2314 : * Mapped catalog tuple without data, emitted while
2315 : * catalog table was in the process of being rewritten. We
2316 : * can fail to look up the relfilenumber, because the
2317 : * relmapper has no "historic" view, in contrast to the
2318 : * normal catalog during decoding. Thus repeated rewrites
2319 : * can cause a lookup failure. That's OK because we do not
2320 : * decode catalog changes anyway. Normally such tuples
2321 : * would be skipped over below, but we can't identify
2322 : * whether the table should be logically logged without
2323 : * mapping the relfilenumber to the oid.
2324 : */
2325 680802 : if (reloid == InvalidOid &&
2326 166 : change->data.tp.newtuple == NULL &&
2327 166 : change->data.tp.oldtuple == NULL)
2328 166 : goto change_done;
2329 680636 : else if (reloid == InvalidOid)
2330 0 : elog(ERROR, "could not map filenumber \"%s\" to relation OID",
2331 : relpathperm(change->data.tp.rlocator,
2332 : MAIN_FORKNUM).str);
2333 :
2334 680636 : relation = RelationIdGetRelation(reloid);
2335 :
2336 680636 : if (!RelationIsValid(relation))
2337 0 : elog(ERROR, "could not open relation with OID %u (for filenumber \"%s\")",
2338 : reloid,
2339 : relpathperm(change->data.tp.rlocator,
2340 : MAIN_FORKNUM).str);
2341 :
2342 680636 : if (!RelationIsLogicallyLogged(relation))
2343 8490 : goto change_done;
2344 :
2345 : /*
2346 : * Ignore temporary heaps created during DDL unless the
2347 : * plugin has asked for them.
2348 : */
2349 672146 : if (relation->rd_rel->relrewrite && !rb->output_rewrites)
2350 52 : goto change_done;
2351 :
2352 : /*
2353 : * For now ignore sequence changes entirely. Most of the
2354 : * time they don't log changes using records we
2355 : * understand, so it doesn't make sense to handle the few
2356 : * cases we do.
2357 : */
2358 672094 : if (relation->rd_rel->relkind == RELKIND_SEQUENCE)
2359 0 : goto change_done;
2360 :
2361 : /* user-triggered change */
2362 672094 : if (!IsToastRelation(relation))
2363 : {
2364 667972 : ReorderBufferToastReplace(rb, txn, relation, change);
2365 667972 : ReorderBufferApplyChange(rb, txn, relation, change,
2366 : streaming);
2367 :
2368 : /*
2369 : * Only clear reassembled toast chunks if we're sure
2370 : * they're not required anymore. The creator of the
2371 : * tuple tells us.
2372 : */
2373 667966 : if (change->data.tp.clear_toast_afterwards)
2374 667522 : ReorderBufferToastReset(rb, txn);
2375 : }
2376 : /* we're not interested in toast deletions */
2377 4122 : else if (change->action == REORDER_BUFFER_CHANGE_INSERT)
2378 : {
2379 : /*
2380 : * Need to reassemble the full toasted Datum in
2381 : * memory, to ensure the chunks don't get reused till
2382 : * we're done remove it from the list of this
2383 : * transaction's changes. Otherwise it will get
2384 : * freed/reused while restoring spooled data from
2385 : * disk.
2386 : */
2387 : Assert(change->data.tp.newtuple != NULL);
2388 :
2389 3660 : dlist_delete(&change->node);
2390 3660 : ReorderBufferToastAppendChunk(rb, txn, relation,
2391 : change);
2392 : }
2393 :
2394 462 : change_done:
2395 :
2396 : /*
2397 : * If speculative insertion was confirmed, the record
2398 : * isn't needed anymore.
2399 : */
2400 680796 : if (specinsert != NULL)
2401 : {
2402 3564 : ReorderBufferReturnChange(rb, specinsert, true);
2403 3564 : specinsert = NULL;
2404 : }
2405 :
2406 680796 : if (RelationIsValid(relation))
2407 : {
2408 680630 : RelationClose(relation);
2409 680630 : relation = NULL;
2410 : }
2411 680796 : break;
2412 :
2413 3564 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
2414 :
2415 : /*
2416 : * Speculative insertions are dealt with by delaying the
2417 : * processing of the insert until the confirmation record
2418 : * arrives. For that we simply unlink the record from the
2419 : * chain, so it does not get freed/reused while restoring
2420 : * spooled data from disk.
2421 : *
2422 : * This is safe in the face of concurrent catalog changes
2423 : * because the relevant relation can't be changed between
2424 : * speculative insertion and confirmation due to
2425 : * CheckTableNotInUse() and locking.
2426 : */
2427 :
2428 : /* clear out a pending (and thus failed) speculation */
2429 3564 : if (specinsert != NULL)
2430 : {
2431 0 : ReorderBufferReturnChange(rb, specinsert, true);
2432 0 : specinsert = NULL;
2433 : }
2434 :
2435 : /* and memorize the pending insertion */
2436 3564 : dlist_delete(&change->node);
2437 3564 : specinsert = change;
2438 3564 : break;
2439 :
2440 0 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
2441 :
2442 : /*
2443 : * Abort for speculative insertion arrived. So cleanup the
2444 : * specinsert tuple and toast hash.
2445 : *
2446 : * Note that we get the spec abort change for each toast
2447 : * entry but we need to perform the cleanup only the first
2448 : * time we get it for the main table.
2449 : */
2450 0 : if (specinsert != NULL)
2451 : {
2452 : /*
2453 : * We must clean the toast hash before processing a
2454 : * completely new tuple to avoid confusion about the
2455 : * previous tuple's toast chunks.
2456 : */
2457 : Assert(change->data.tp.clear_toast_afterwards);
2458 0 : ReorderBufferToastReset(rb, txn);
2459 :
2460 : /* We don't need this record anymore. */
2461 0 : ReorderBufferReturnChange(rb, specinsert, true);
2462 0 : specinsert = NULL;
2463 : }
2464 0 : break;
2465 :
2466 46 : case REORDER_BUFFER_CHANGE_TRUNCATE:
2467 : {
2468 : int i;
2469 46 : int nrelids = change->data.truncate.nrelids;
2470 46 : int nrelations = 0;
2471 : Relation *relations;
2472 :
2473 46 : relations = palloc0(nrelids * sizeof(Relation));
2474 132 : for (i = 0; i < nrelids; i++)
2475 : {
2476 86 : Oid relid = change->data.truncate.relids[i];
2477 : Relation rel;
2478 :
2479 86 : rel = RelationIdGetRelation(relid);
2480 :
2481 86 : if (!RelationIsValid(rel))
2482 0 : elog(ERROR, "could not open relation with OID %u", relid);
2483 :
2484 86 : if (!RelationIsLogicallyLogged(rel))
2485 0 : continue;
2486 :
2487 86 : relations[nrelations++] = rel;
2488 : }
2489 :
2490 : /* Apply the truncate. */
2491 46 : ReorderBufferApplyTruncate(rb, txn, nrelations,
2492 : relations, change,
2493 : streaming);
2494 :
2495 132 : for (i = 0; i < nrelations; i++)
2496 86 : RelationClose(relations[i]);
2497 :
2498 46 : break;
2499 : }
2500 :
2501 22 : case REORDER_BUFFER_CHANGE_MESSAGE:
2502 22 : ReorderBufferApplyMessage(rb, txn, change, streaming);
2503 22 : break;
2504 :
2505 4346 : case REORDER_BUFFER_CHANGE_INVALIDATION:
2506 : /* Execute the invalidation messages locally */
2507 4346 : ReorderBufferExecuteInvalidations(change->data.inval.ninvalidations,
2508 : change->data.inval.invalidations);
2509 4346 : break;
2510 :
2511 1026 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
2512 : /* get rid of the old */
2513 1026 : TeardownHistoricSnapshot(false);
2514 :
2515 1026 : if (snapshot_now->copied)
2516 : {
2517 986 : ReorderBufferFreeSnap(rb, snapshot_now);
2518 986 : snapshot_now =
2519 986 : ReorderBufferCopySnap(rb, change->data.snapshot,
2520 : txn, command_id);
2521 : }
2522 :
2523 : /*
2524 : * Restored from disk, need to be careful not to double
2525 : * free. We could introduce refcounting for that, but for
2526 : * now this seems infrequent enough not to care.
2527 : */
2528 40 : else if (change->data.snapshot->copied)
2529 : {
2530 0 : snapshot_now =
2531 0 : ReorderBufferCopySnap(rb, change->data.snapshot,
2532 : txn, command_id);
2533 : }
2534 : else
2535 : {
2536 40 : snapshot_now = change->data.snapshot;
2537 : }
2538 :
2539 : /* and continue with the new one */
2540 1026 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2541 1026 : break;
2542 :
2543 21240 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
2544 : Assert(change->data.command_id != InvalidCommandId);
2545 :
2546 21240 : if (command_id < change->data.command_id)
2547 : {
2548 3752 : command_id = change->data.command_id;
2549 :
2550 3752 : if (!snapshot_now->copied)
2551 : {
2552 : /* we don't use the global one anymore */
2553 976 : snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
2554 : txn, command_id);
2555 : }
2556 :
2557 3752 : snapshot_now->curcid = command_id;
2558 :
2559 3752 : TeardownHistoricSnapshot(false);
2560 3752 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2561 : }
2562 :
2563 21240 : break;
2564 :
2565 0 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
2566 0 : elog(ERROR, "tuplecid value in changequeue");
2567 : break;
2568 : }
2569 :
2570 : /*
2571 : * It is possible that the data is not sent to downstream for a
2572 : * long time either because the output plugin filtered it or there
2573 : * is a DDL that generates a lot of data that is not processed by
2574 : * the plugin. So, in such cases, the downstream can timeout. To
2575 : * avoid that we try to send a keepalive message if required.
2576 : * Trying to send a keepalive message after every change has some
2577 : * overhead, but testing showed there is no noticeable overhead if
2578 : * we do it after every ~100 changes.
2579 : */
2580 : #define CHANGES_THRESHOLD 100
2581 :
2582 711040 : if (++changes_count >= CHANGES_THRESHOLD)
2583 : {
2584 6214 : rb->update_progress_txn(rb, txn, change->lsn);
2585 6214 : changes_count = 0;
2586 : }
2587 : }
2588 :
2589 : /* speculative insertion record must be freed by now */
2590 : Assert(!specinsert);
2591 :
2592 : /* clean up the iterator */
2593 3782 : ReorderBufferIterTXNFinish(rb, iterstate);
2594 3782 : iterstate = NULL;
2595 :
2596 : /*
2597 : * Update total transaction count and total bytes processed by the
2598 : * transaction and its subtransactions. Ensure to not count the
2599 : * streamed transaction multiple times.
2600 : *
2601 : * Note that the statistics computation has to be done after
2602 : * ReorderBufferIterTXNFinish as it releases the serialized change
2603 : * which we have already accounted in ReorderBufferIterTXNNext.
2604 : */
2605 3782 : if (!rbtxn_is_streamed(txn))
2606 2502 : rb->totalTxns++;
2607 :
2608 3782 : rb->totalBytes += txn->total_size;
2609 :
2610 : /*
2611 : * Done with current changes, send the last message for this set of
2612 : * changes depending upon streaming mode.
2613 : */
2614 3782 : if (streaming)
2615 : {
2616 1416 : if (stream_started)
2617 : {
2618 1338 : rb->stream_stop(rb, txn, prev_lsn);
2619 1338 : stream_started = false;
2620 : }
2621 : }
2622 : else
2623 : {
2624 : /*
2625 : * Call either PREPARE (for two-phase transactions) or COMMIT (for
2626 : * regular ones).
2627 : */
2628 2366 : if (rbtxn_is_prepared(txn))
2629 : {
2630 : Assert(!rbtxn_sent_prepare(txn));
2631 56 : rb->prepare(rb, txn, commit_lsn);
2632 56 : txn->txn_flags |= RBTXN_SENT_PREPARE;
2633 : }
2634 : else
2635 2310 : rb->commit(rb, txn, commit_lsn);
2636 : }
2637 :
2638 : /* this is just a sanity check against bad output plugin behaviour */
2639 3776 : if (GetCurrentTransactionIdIfAny() != InvalidTransactionId)
2640 0 : elog(ERROR, "output plugin used XID %u",
2641 : GetCurrentTransactionId());
2642 :
2643 : /*
2644 : * Remember the command ID and snapshot for the next set of changes in
2645 : * streaming mode.
2646 : */
2647 3776 : if (streaming)
2648 1416 : ReorderBufferSaveTXNSnapshot(rb, txn, snapshot_now, command_id);
2649 2360 : else if (snapshot_now->copied)
2650 976 : ReorderBufferFreeSnap(rb, snapshot_now);
2651 :
2652 : /* cleanup */
2653 3776 : TeardownHistoricSnapshot(false);
2654 :
2655 : /*
2656 : * Aborting the current (sub-)transaction as a whole has the right
2657 : * semantics. We want all locks acquired in here to be released, not
2658 : * reassigned to the parent and we do not want any database access
2659 : * have persistent effects.
2660 : */
2661 3776 : AbortCurrentTransaction();
2662 :
2663 : /* make sure there's no cache pollution */
2664 3776 : ReorderBufferExecuteInvalidations(txn->ninvalidations, txn->invalidations);
2665 :
2666 3776 : if (using_subtxn)
2667 954 : RollbackAndReleaseCurrentSubTransaction();
2668 :
2669 : /*
2670 : * We are here due to one of the four reasons: 1. Decoding an
2671 : * in-progress txn. 2. Decoding a prepared txn. 3. Decoding of a
2672 : * prepared txn that was (partially) streamed. 4. Decoding a committed
2673 : * txn.
2674 : *
2675 : * For 1, we allow truncation of txn data by removing the changes
2676 : * already streamed but still keeping other things like invalidations,
2677 : * snapshot, and tuplecids. For 2 and 3, we indicate
2678 : * ReorderBufferTruncateTXN to do more elaborate truncation of txn
2679 : * data as the entire transaction has been decoded except for commit.
2680 : * For 4, as the entire txn has been decoded, we can fully clean up
2681 : * the TXN reorder buffer.
2682 : */
2683 3776 : if (streaming || rbtxn_is_prepared(txn))
2684 : {
2685 1472 : if (streaming)
2686 1416 : ReorderBufferMaybeMarkTXNStreamed(rb, txn);
2687 :
2688 1472 : ReorderBufferTruncateTXN(rb, txn, rbtxn_is_prepared(txn));
2689 : /* Reset the CheckXidAlive */
2690 1472 : CheckXidAlive = InvalidTransactionId;
2691 : }
2692 : else
2693 2304 : ReorderBufferCleanupTXN(rb, txn);
2694 : }
2695 22 : PG_CATCH();
2696 : {
2697 22 : MemoryContext ecxt = MemoryContextSwitchTo(ccxt);
2698 22 : ErrorData *errdata = CopyErrorData();
2699 :
2700 : /* TODO: Encapsulate cleanup from the PG_TRY and PG_CATCH blocks */
2701 22 : if (iterstate)
2702 22 : ReorderBufferIterTXNFinish(rb, iterstate);
2703 :
2704 22 : TeardownHistoricSnapshot(true);
2705 :
2706 : /*
2707 : * Force cache invalidation to happen outside of a valid transaction
2708 : * to prevent catalog access as we just caught an error.
2709 : */
2710 22 : AbortCurrentTransaction();
2711 :
2712 : /* make sure there's no cache pollution */
2713 22 : ReorderBufferExecuteInvalidations(txn->ninvalidations,
2714 : txn->invalidations);
2715 :
2716 22 : if (using_subtxn)
2717 8 : RollbackAndReleaseCurrentSubTransaction();
2718 :
2719 : /*
2720 : * The error code ERRCODE_TRANSACTION_ROLLBACK indicates a concurrent
2721 : * abort of the (sub)transaction we are streaming or preparing. We
2722 : * need to do the cleanup and return gracefully on this error, see
2723 : * SetupCheckXidLive.
2724 : *
2725 : * This error code can be thrown by one of the callbacks we call
2726 : * during decoding so we need to ensure that we return gracefully only
2727 : * when we are sending the data in streaming mode and the streaming is
2728 : * not finished yet or when we are sending the data out on a PREPARE
2729 : * during a two-phase commit.
2730 : */
2731 22 : if (errdata->sqlerrcode == ERRCODE_TRANSACTION_ROLLBACK &&
2732 16 : (stream_started || rbtxn_is_prepared(txn)))
2733 : {
2734 : /* curtxn must be set for streaming or prepared transactions */
2735 : Assert(curtxn);
2736 :
2737 : /* Cleanup the temporary error state. */
2738 16 : FlushErrorState();
2739 16 : FreeErrorData(errdata);
2740 16 : errdata = NULL;
2741 :
2742 : /* Remember the transaction is aborted. */
2743 : Assert(!rbtxn_is_committed(curtxn));
2744 16 : curtxn->txn_flags |= RBTXN_IS_ABORTED;
2745 :
2746 : /* Mark the transaction is streamed if appropriate */
2747 16 : if (stream_started)
2748 16 : ReorderBufferMaybeMarkTXNStreamed(rb, txn);
2749 :
2750 : /* Reset the TXN so that it is allowed to stream remaining data. */
2751 16 : ReorderBufferResetTXN(rb, txn, snapshot_now,
2752 : command_id, prev_lsn,
2753 : specinsert);
2754 : }
2755 : else
2756 : {
2757 6 : ReorderBufferCleanupTXN(rb, txn);
2758 6 : MemoryContextSwitchTo(ecxt);
2759 6 : PG_RE_THROW();
2760 : }
2761 : }
2762 3792 : PG_END_TRY();
2763 3792 : }
2764 :
2765 : /*
2766 : * Perform the replay of a transaction and its non-aborted subtransactions.
2767 : *
2768 : * Subtransactions previously have to be processed by
2769 : * ReorderBufferCommitChild(), even if previously assigned to the toplevel
2770 : * transaction with ReorderBufferAssignChild.
2771 : *
2772 : * This interface is called once a prepare or toplevel commit is read for both
2773 : * streamed as well as non-streamed transactions.
2774 : */
2775 : static void
2776 2510 : ReorderBufferReplay(ReorderBufferTXN *txn,
2777 : ReorderBuffer *rb, TransactionId xid,
2778 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2779 : TimestampTz commit_time,
2780 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2781 : {
2782 : Snapshot snapshot_now;
2783 2510 : CommandId command_id = FirstCommandId;
2784 :
2785 2510 : txn->final_lsn = commit_lsn;
2786 2510 : txn->end_lsn = end_lsn;
2787 2510 : txn->xact_time.commit_time = commit_time;
2788 2510 : txn->origin_id = origin_id;
2789 2510 : txn->origin_lsn = origin_lsn;
2790 :
2791 : /*
2792 : * If the transaction was (partially) streamed, we need to commit it in a
2793 : * 'streamed' way. That is, we first stream the remaining part of the
2794 : * transaction, and then invoke stream_commit message.
2795 : *
2796 : * Called after everything (origin ID, LSN, ...) is stored in the
2797 : * transaction to avoid passing that information directly.
2798 : */
2799 2510 : if (rbtxn_is_streamed(txn))
2800 : {
2801 132 : ReorderBufferStreamCommit(rb, txn);
2802 132 : return;
2803 : }
2804 :
2805 : /*
2806 : * If this transaction has no snapshot, it didn't make any changes to the
2807 : * database, so there's nothing to decode. Note that
2808 : * ReorderBufferCommitChild will have transferred any snapshots from
2809 : * subtransactions if there were any.
2810 : */
2811 2378 : if (txn->base_snapshot == NULL)
2812 : {
2813 : Assert(txn->ninvalidations == 0);
2814 :
2815 : /*
2816 : * Removing this txn before a commit might result in the computation
2817 : * of an incorrect restart_lsn. See SnapBuildProcessRunningXacts.
2818 : */
2819 6 : if (!rbtxn_is_prepared(txn))
2820 6 : ReorderBufferCleanupTXN(rb, txn);
2821 6 : return;
2822 : }
2823 :
2824 2372 : snapshot_now = txn->base_snapshot;
2825 :
2826 : /* Process and send the changes to output plugin. */
2827 2372 : ReorderBufferProcessTXN(rb, txn, commit_lsn, snapshot_now,
2828 : command_id, false);
2829 : }
2830 :
2831 : /*
2832 : * Commit a transaction.
2833 : *
2834 : * See comments for ReorderBufferReplay().
2835 : */
2836 : void
2837 2426 : ReorderBufferCommit(ReorderBuffer *rb, TransactionId xid,
2838 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2839 : TimestampTz commit_time,
2840 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2841 : {
2842 : ReorderBufferTXN *txn;
2843 :
2844 2426 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2845 : false);
2846 :
2847 : /* unknown transaction, nothing to replay */
2848 2426 : if (txn == NULL)
2849 2 : return;
2850 :
2851 2424 : ReorderBufferReplay(txn, rb, xid, commit_lsn, end_lsn, commit_time,
2852 : origin_id, origin_lsn);
2853 : }
2854 :
2855 : /*
2856 : * Record the prepare information for a transaction. Also, mark the transaction
2857 : * as a prepared transaction.
2858 : */
2859 : bool
2860 286 : ReorderBufferRememberPrepareInfo(ReorderBuffer *rb, TransactionId xid,
2861 : XLogRecPtr prepare_lsn, XLogRecPtr end_lsn,
2862 : TimestampTz prepare_time,
2863 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2864 : {
2865 : ReorderBufferTXN *txn;
2866 :
2867 286 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr, false);
2868 :
2869 : /* unknown transaction, nothing to do */
2870 286 : if (txn == NULL)
2871 0 : return false;
2872 :
2873 : /*
2874 : * Remember the prepare information to be later used by commit prepared in
2875 : * case we skip doing prepare.
2876 : */
2877 286 : txn->final_lsn = prepare_lsn;
2878 286 : txn->end_lsn = end_lsn;
2879 286 : txn->xact_time.prepare_time = prepare_time;
2880 286 : txn->origin_id = origin_id;
2881 286 : txn->origin_lsn = origin_lsn;
2882 :
2883 : /* Mark this transaction as a prepared transaction */
2884 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == 0);
2885 286 : txn->txn_flags |= RBTXN_IS_PREPARED;
2886 :
2887 286 : return true;
2888 : }
2889 :
2890 : /* Remember that we have skipped prepare */
2891 : void
2892 204 : ReorderBufferSkipPrepare(ReorderBuffer *rb, TransactionId xid)
2893 : {
2894 : ReorderBufferTXN *txn;
2895 :
2896 204 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr, false);
2897 :
2898 : /* unknown transaction, nothing to do */
2899 204 : if (txn == NULL)
2900 0 : return;
2901 :
2902 : /* txn must have been marked as a prepared transaction */
2903 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == RBTXN_IS_PREPARED);
2904 204 : txn->txn_flags |= RBTXN_SKIPPED_PREPARE;
2905 : }
2906 :
2907 : /*
2908 : * Prepare a two-phase transaction.
2909 : *
2910 : * See comments for ReorderBufferReplay().
2911 : */
2912 : void
2913 82 : ReorderBufferPrepare(ReorderBuffer *rb, TransactionId xid,
2914 : char *gid)
2915 : {
2916 : ReorderBufferTXN *txn;
2917 :
2918 82 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2919 : false);
2920 :
2921 : /* unknown transaction, nothing to replay */
2922 82 : if (txn == NULL)
2923 0 : return;
2924 :
2925 : /*
2926 : * txn must have been marked as a prepared transaction and must have
2927 : * neither been skipped nor sent a prepare. Also, the prepare info must
2928 : * have been updated in it by now.
2929 : */
2930 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) == RBTXN_IS_PREPARED);
2931 : Assert(txn->final_lsn != InvalidXLogRecPtr);
2932 :
2933 82 : txn->gid = pstrdup(gid);
2934 :
2935 82 : ReorderBufferReplay(txn, rb, xid, txn->final_lsn, txn->end_lsn,
2936 82 : txn->xact_time.prepare_time, txn->origin_id, txn->origin_lsn);
2937 :
2938 : /*
2939 : * Send a prepare if not already done so. This might occur if we have
2940 : * detected a concurrent abort while replaying the non-streaming
2941 : * transaction.
2942 : */
2943 82 : if (!rbtxn_sent_prepare(txn))
2944 : {
2945 0 : rb->prepare(rb, txn, txn->final_lsn);
2946 0 : txn->txn_flags |= RBTXN_SENT_PREPARE;
2947 : }
2948 : }
2949 :
2950 : /*
2951 : * This is used to handle COMMIT/ROLLBACK PREPARED.
2952 : */
2953 : void
2954 84 : ReorderBufferFinishPrepared(ReorderBuffer *rb, TransactionId xid,
2955 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2956 : XLogRecPtr two_phase_at,
2957 : TimestampTz commit_time, RepOriginId origin_id,
2958 : XLogRecPtr origin_lsn, char *gid, bool is_commit)
2959 : {
2960 : ReorderBufferTXN *txn;
2961 : XLogRecPtr prepare_end_lsn;
2962 : TimestampTz prepare_time;
2963 :
2964 84 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, commit_lsn, false);
2965 :
2966 : /* unknown transaction, nothing to do */
2967 84 : if (txn == NULL)
2968 0 : return;
2969 :
2970 : /*
2971 : * By this time the txn has the prepare record information, remember it to
2972 : * be later used for rollback.
2973 : */
2974 84 : prepare_end_lsn = txn->end_lsn;
2975 84 : prepare_time = txn->xact_time.prepare_time;
2976 :
2977 : /* add the gid in the txn */
2978 84 : txn->gid = pstrdup(gid);
2979 :
2980 : /*
2981 : * It is possible that this transaction is not decoded at prepare time
2982 : * either because by that time we didn't have a consistent snapshot, or
2983 : * two_phase was not enabled, or it was decoded earlier but we have
2984 : * restarted. We only need to send the prepare if it was not decoded
2985 : * earlier. We don't need to decode the xact for aborts if it is not done
2986 : * already.
2987 : */
2988 84 : if ((txn->final_lsn < two_phase_at) && is_commit)
2989 : {
2990 : /*
2991 : * txn must have been marked as a prepared transaction and skipped but
2992 : * not sent a prepare. Also, the prepare info must have been updated
2993 : * in txn even if we skip prepare.
2994 : */
2995 : Assert((txn->txn_flags & RBTXN_PREPARE_STATUS_MASK) ==
2996 : (RBTXN_IS_PREPARED | RBTXN_SKIPPED_PREPARE));
2997 : Assert(txn->final_lsn != InvalidXLogRecPtr);
2998 :
2999 : /*
3000 : * By this time the txn has the prepare record information and it is
3001 : * important to use that so that downstream gets the accurate
3002 : * information. If instead, we have passed commit information here
3003 : * then downstream can behave as it has already replayed commit
3004 : * prepared after the restart.
3005 : */
3006 4 : ReorderBufferReplay(txn, rb, xid, txn->final_lsn, txn->end_lsn,
3007 4 : txn->xact_time.prepare_time, txn->origin_id, txn->origin_lsn);
3008 : }
3009 :
3010 84 : txn->final_lsn = commit_lsn;
3011 84 : txn->end_lsn = end_lsn;
3012 84 : txn->xact_time.commit_time = commit_time;
3013 84 : txn->origin_id = origin_id;
3014 84 : txn->origin_lsn = origin_lsn;
3015 :
3016 84 : if (is_commit)
3017 62 : rb->commit_prepared(rb, txn, commit_lsn);
3018 : else
3019 22 : rb->rollback_prepared(rb, txn, prepare_end_lsn, prepare_time);
3020 :
3021 : /* cleanup: make sure there's no cache pollution */
3022 84 : ReorderBufferExecuteInvalidations(txn->ninvalidations,
3023 : txn->invalidations);
3024 84 : ReorderBufferCleanupTXN(rb, txn);
3025 : }
3026 :
3027 : /*
3028 : * Abort a transaction that possibly has previous changes. Needs to be first
3029 : * called for subtransactions and then for the toplevel xid.
3030 : *
3031 : * NB: Transactions handled here have to have actively aborted (i.e. have
3032 : * produced an abort record). Implicitly aborted transactions are handled via
3033 : * ReorderBufferAbortOld(); transactions we're just not interested in, but
3034 : * which have committed are handled in ReorderBufferForget().
3035 : *
3036 : * This function purges this transaction and its contents from memory and
3037 : * disk.
3038 : */
3039 : void
3040 220 : ReorderBufferAbort(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
3041 : TimestampTz abort_time)
3042 : {
3043 : ReorderBufferTXN *txn;
3044 :
3045 220 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3046 : false);
3047 :
3048 : /* unknown, nothing to remove */
3049 220 : if (txn == NULL)
3050 0 : return;
3051 :
3052 220 : txn->xact_time.abort_time = abort_time;
3053 :
3054 : /* For streamed transactions notify the remote node about the abort. */
3055 220 : if (rbtxn_is_streamed(txn))
3056 : {
3057 60 : rb->stream_abort(rb, txn, lsn);
3058 :
3059 : /*
3060 : * We might have decoded changes for this transaction that could load
3061 : * the cache as per the current transaction's view (consider DDL's
3062 : * happened in this transaction). We don't want the decoding of future
3063 : * transactions to use those cache entries so execute invalidations.
3064 : */
3065 60 : if (txn->ninvalidations > 0)
3066 0 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3067 : txn->invalidations);
3068 : }
3069 :
3070 : /* cosmetic... */
3071 220 : txn->final_lsn = lsn;
3072 :
3073 : /* remove potential on-disk data, and deallocate */
3074 220 : ReorderBufferCleanupTXN(rb, txn);
3075 : }
3076 :
3077 : /*
3078 : * Abort all transactions that aren't actually running anymore because the
3079 : * server restarted.
3080 : *
3081 : * NB: These really have to be transactions that have aborted due to a server
3082 : * crash/immediate restart, as we don't deal with invalidations here.
3083 : */
3084 : void
3085 2618 : ReorderBufferAbortOld(ReorderBuffer *rb, TransactionId oldestRunningXid)
3086 : {
3087 : dlist_mutable_iter it;
3088 :
3089 : /*
3090 : * Iterate through all (potential) toplevel TXNs and abort all that are
3091 : * older than what possibly can be running. Once we've found the first
3092 : * that is alive we stop, there might be some that acquired an xid earlier
3093 : * but started writing later, but it's unlikely and they will be cleaned
3094 : * up in a later call to this function.
3095 : */
3096 2628 : dlist_foreach_modify(it, &rb->toplevel_by_lsn)
3097 : {
3098 : ReorderBufferTXN *txn;
3099 :
3100 122 : txn = dlist_container(ReorderBufferTXN, node, it.cur);
3101 :
3102 122 : if (TransactionIdPrecedes(txn->xid, oldestRunningXid))
3103 : {
3104 10 : elog(DEBUG2, "aborting old transaction %u", txn->xid);
3105 :
3106 : /* Notify the remote node about the crash/immediate restart. */
3107 10 : if (rbtxn_is_streamed(txn))
3108 0 : rb->stream_abort(rb, txn, InvalidXLogRecPtr);
3109 :
3110 : /* remove potential on-disk data, and deallocate this tx */
3111 10 : ReorderBufferCleanupTXN(rb, txn);
3112 : }
3113 : else
3114 112 : return;
3115 : }
3116 : }
3117 :
3118 : /*
3119 : * Forget the contents of a transaction if we aren't interested in its
3120 : * contents. Needs to be first called for subtransactions and then for the
3121 : * toplevel xid.
3122 : *
3123 : * This is significantly different to ReorderBufferAbort() because
3124 : * transactions that have committed need to be treated differently from aborted
3125 : * ones since they may have modified the catalog.
3126 : *
3127 : * Note that this is only allowed to be called in the moment a transaction
3128 : * commit has just been read, not earlier; otherwise later records referring
3129 : * to this xid might re-create the transaction incompletely.
3130 : */
3131 : void
3132 4934 : ReorderBufferForget(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3133 : {
3134 : ReorderBufferTXN *txn;
3135 :
3136 4934 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3137 : false);
3138 :
3139 : /* unknown, nothing to forget */
3140 4934 : if (txn == NULL)
3141 1124 : return;
3142 :
3143 : /* this transaction mustn't be streamed */
3144 : Assert(!rbtxn_is_streamed(txn));
3145 :
3146 : /* cosmetic... */
3147 3810 : txn->final_lsn = lsn;
3148 :
3149 : /*
3150 : * Process cache invalidation messages if there are any. Even if we're not
3151 : * interested in the transaction's contents, it could have manipulated the
3152 : * catalog and we need to update the caches according to that.
3153 : */
3154 3810 : if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
3155 1060 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3156 : txn->invalidations);
3157 : else
3158 : Assert(txn->ninvalidations == 0);
3159 :
3160 : /* remove potential on-disk data, and deallocate */
3161 3810 : ReorderBufferCleanupTXN(rb, txn);
3162 : }
3163 :
3164 : /*
3165 : * Invalidate cache for those transactions that need to be skipped just in case
3166 : * catalogs were manipulated as part of the transaction.
3167 : *
3168 : * Note that this is a special-purpose function for prepared transactions where
3169 : * we don't want to clean up the TXN even when we decide to skip it. See
3170 : * DecodePrepare.
3171 : */
3172 : void
3173 198 : ReorderBufferInvalidate(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3174 : {
3175 : ReorderBufferTXN *txn;
3176 :
3177 198 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3178 : false);
3179 :
3180 : /* unknown, nothing to do */
3181 198 : if (txn == NULL)
3182 0 : return;
3183 :
3184 : /*
3185 : * Process cache invalidation messages if there are any. Even if we're not
3186 : * interested in the transaction's contents, it could have manipulated the
3187 : * catalog and we need to update the caches according to that.
3188 : */
3189 198 : if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
3190 58 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3191 : txn->invalidations);
3192 : else
3193 : Assert(txn->ninvalidations == 0);
3194 : }
3195 :
3196 :
3197 : /*
3198 : * Execute invalidations happening outside the context of a decoded
3199 : * transaction. That currently happens either for xid-less commits
3200 : * (cf. RecordTransactionCommit()) or for invalidations in uninteresting
3201 : * transactions (via ReorderBufferForget()).
3202 : */
3203 : void
3204 1122 : ReorderBufferImmediateInvalidation(ReorderBuffer *rb, uint32 ninvalidations,
3205 : SharedInvalidationMessage *invalidations)
3206 : {
3207 1122 : bool use_subtxn = IsTransactionOrTransactionBlock();
3208 : int i;
3209 :
3210 1122 : if (use_subtxn)
3211 860 : BeginInternalSubTransaction("replay");
3212 :
3213 : /*
3214 : * Force invalidations to happen outside of a valid transaction - that way
3215 : * entries will just be marked as invalid without accessing the catalog.
3216 : * That's advantageous because we don't need to setup the full state
3217 : * necessary for catalog access.
3218 : */
3219 1122 : if (use_subtxn)
3220 860 : AbortCurrentTransaction();
3221 :
3222 50774 : for (i = 0; i < ninvalidations; i++)
3223 49652 : LocalExecuteInvalidationMessage(&invalidations[i]);
3224 :
3225 1122 : if (use_subtxn)
3226 860 : RollbackAndReleaseCurrentSubTransaction();
3227 1122 : }
3228 :
3229 : /*
3230 : * Tell reorderbuffer about an xid seen in the WAL stream. Has to be called at
3231 : * least once for every xid in XLogRecord->xl_xid (other places in records
3232 : * may, but do not have to be passed through here).
3233 : *
3234 : * Reorderbuffer keeps some data structures about transactions in LSN order,
3235 : * for efficiency. To do that it has to know about when transactions are seen
3236 : * first in the WAL. As many types of records are not actually interesting for
3237 : * logical decoding, they do not necessarily pass through here.
3238 : */
3239 : void
3240 4543524 : ReorderBufferProcessXid(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3241 : {
3242 : /* many records won't have an xid assigned, centralize check here */
3243 4543524 : if (xid != InvalidTransactionId)
3244 4539842 : ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3245 4543524 : }
3246 :
3247 : /*
3248 : * Add a new snapshot to this transaction that may only used after lsn 'lsn'
3249 : * because the previous snapshot doesn't describe the catalog correctly for
3250 : * following rows.
3251 : */
3252 : void
3253 2098 : ReorderBufferAddSnapshot(ReorderBuffer *rb, TransactionId xid,
3254 : XLogRecPtr lsn, Snapshot snap)
3255 : {
3256 2098 : ReorderBufferChange *change = ReorderBufferGetChange(rb);
3257 :
3258 2098 : change->data.snapshot = snap;
3259 2098 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT;
3260 :
3261 2098 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3262 2098 : }
3263 :
3264 : /*
3265 : * Set up the transaction's base snapshot.
3266 : *
3267 : * If we know that xid is a subtransaction, set the base snapshot on the
3268 : * top-level transaction instead.
3269 : */
3270 : void
3271 5654 : ReorderBufferSetBaseSnapshot(ReorderBuffer *rb, TransactionId xid,
3272 : XLogRecPtr lsn, Snapshot snap)
3273 : {
3274 : ReorderBufferTXN *txn;
3275 : bool is_new;
3276 :
3277 : Assert(snap != NULL);
3278 :
3279 : /*
3280 : * Fetch the transaction to operate on. If we know it's a subtransaction,
3281 : * operate on its top-level transaction instead.
3282 : */
3283 5654 : txn = ReorderBufferTXNByXid(rb, xid, true, &is_new, lsn, true);
3284 5654 : if (rbtxn_is_known_subxact(txn))
3285 204 : txn = ReorderBufferTXNByXid(rb, txn->toplevel_xid, false,
3286 : NULL, InvalidXLogRecPtr, false);
3287 : Assert(txn->base_snapshot == NULL);
3288 :
3289 5654 : txn->base_snapshot = snap;
3290 5654 : txn->base_snapshot_lsn = lsn;
3291 5654 : dlist_push_tail(&rb->txns_by_base_snapshot_lsn, &txn->base_snapshot_node);
3292 :
3293 5654 : AssertTXNLsnOrder(rb);
3294 5654 : }
3295 :
3296 : /*
3297 : * Access the catalog with this CommandId at this point in the changestream.
3298 : *
3299 : * May only be called for command ids > 1
3300 : */
3301 : void
3302 47976 : ReorderBufferAddNewCommandId(ReorderBuffer *rb, TransactionId xid,
3303 : XLogRecPtr lsn, CommandId cid)
3304 : {
3305 47976 : ReorderBufferChange *change = ReorderBufferGetChange(rb);
3306 :
3307 47976 : change->data.command_id = cid;
3308 47976 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID;
3309 :
3310 47976 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3311 47976 : }
3312 :
3313 : /*
3314 : * Update memory counters to account for the new or removed change.
3315 : *
3316 : * We update two counters - in the reorder buffer, and in the transaction
3317 : * containing the change. The reorder buffer counter allows us to quickly
3318 : * decide if we reached the memory limit, the transaction counter allows
3319 : * us to quickly pick the largest transaction for eviction.
3320 : *
3321 : * Either txn or change must be non-NULL at least. We update the memory
3322 : * counter of txn if it's non-NULL, otherwise change->txn.
3323 : *
3324 : * When streaming is enabled, we need to update the toplevel transaction
3325 : * counters instead - we don't really care about subtransactions as we
3326 : * can't stream them individually anyway, and we only pick toplevel
3327 : * transactions for eviction. So only toplevel transactions matter.
3328 : */
3329 : static void
3330 3783742 : ReorderBufferChangeMemoryUpdate(ReorderBuffer *rb,
3331 : ReorderBufferChange *change,
3332 : ReorderBufferTXN *txn,
3333 : bool addition, Size sz)
3334 : {
3335 : ReorderBufferTXN *toptxn;
3336 :
3337 : Assert(txn || change);
3338 :
3339 : /*
3340 : * Ignore tuple CID changes, because those are not evicted when reaching
3341 : * memory limit. So we just don't count them, because it might easily
3342 : * trigger a pointless attempt to spill.
3343 : */
3344 3783742 : if (change && change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID)
3345 47748 : return;
3346 :
3347 3735994 : if (sz == 0)
3348 1766 : return;
3349 :
3350 3734228 : if (txn == NULL)
3351 3719138 : txn = change->txn;
3352 : Assert(txn != NULL);
3353 :
3354 : /*
3355 : * Update the total size in top level as well. This is later used to
3356 : * compute the decoding stats.
3357 : */
3358 3734228 : toptxn = rbtxn_get_toptxn(txn);
3359 :
3360 3734228 : if (addition)
3361 : {
3362 3361366 : Size oldsize = txn->size;
3363 :
3364 3361366 : txn->size += sz;
3365 3361366 : rb->size += sz;
3366 :
3367 : /* Update the total size in the top transaction. */
3368 3361366 : toptxn->total_size += sz;
3369 :
3370 : /* Update the max-heap */
3371 3361366 : if (oldsize != 0)
3372 3346160 : pairingheap_remove(rb->txn_heap, &txn->txn_node);
3373 3361366 : pairingheap_add(rb->txn_heap, &txn->txn_node);
3374 : }
3375 : else
3376 : {
3377 : Assert((rb->size >= sz) && (txn->size >= sz));
3378 372862 : txn->size -= sz;
3379 372862 : rb->size -= sz;
3380 :
3381 : /* Update the total size in the top transaction. */
3382 372862 : toptxn->total_size -= sz;
3383 :
3384 : /* Update the max-heap */
3385 372862 : pairingheap_remove(rb->txn_heap, &txn->txn_node);
3386 372862 : if (txn->size != 0)
3387 357714 : pairingheap_add(rb->txn_heap, &txn->txn_node);
3388 : }
3389 :
3390 : Assert(txn->size <= rb->size);
3391 : }
3392 :
3393 : /*
3394 : * Add new (relfilelocator, tid) -> (cmin, cmax) mappings.
3395 : *
3396 : * We do not include this change type in memory accounting, because we
3397 : * keep CIDs in a separate list and do not evict them when reaching
3398 : * the memory limit.
3399 : */
3400 : void
3401 47976 : ReorderBufferAddNewTupleCids(ReorderBuffer *rb, TransactionId xid,
3402 : XLogRecPtr lsn, RelFileLocator locator,
3403 : ItemPointerData tid, CommandId cmin,
3404 : CommandId cmax, CommandId combocid)
3405 : {
3406 47976 : ReorderBufferChange *change = ReorderBufferGetChange(rb);
3407 : ReorderBufferTXN *txn;
3408 :
3409 47976 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3410 :
3411 47976 : change->data.tuplecid.locator = locator;
3412 47976 : change->data.tuplecid.tid = tid;
3413 47976 : change->data.tuplecid.cmin = cmin;
3414 47976 : change->data.tuplecid.cmax = cmax;
3415 47976 : change->data.tuplecid.combocid = combocid;
3416 47976 : change->lsn = lsn;
3417 47976 : change->txn = txn;
3418 47976 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID;
3419 :
3420 47976 : dlist_push_tail(&txn->tuplecids, &change->node);
3421 47976 : txn->ntuplecids++;
3422 47976 : }
3423 :
3424 : /*
3425 : * Accumulate the invalidations for executing them later.
3426 : *
3427 : * This needs to be called for each XLOG_XACT_INVALIDATIONS message and
3428 : * accumulates all the invalidation messages in the toplevel transaction, if
3429 : * available, otherwise in the current transaction, as well as in the form of
3430 : * change in reorder buffer. We require to record it in form of the change
3431 : * so that we can execute only the required invalidations instead of executing
3432 : * all the invalidations on each CommandId increment. We also need to
3433 : * accumulate these in the txn buffer because in some cases where we skip
3434 : * processing the transaction (see ReorderBufferForget), we need to execute
3435 : * all the invalidations together.
3436 : */
3437 : void
3438 9964 : ReorderBufferAddInvalidations(ReorderBuffer *rb, TransactionId xid,
3439 : XLogRecPtr lsn, Size nmsgs,
3440 : SharedInvalidationMessage *msgs)
3441 : {
3442 : ReorderBufferTXN *txn;
3443 : MemoryContext oldcontext;
3444 : ReorderBufferChange *change;
3445 :
3446 9964 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3447 :
3448 9964 : oldcontext = MemoryContextSwitchTo(rb->context);
3449 :
3450 : /*
3451 : * Collect all the invalidations under the top transaction, if available,
3452 : * so that we can execute them all together. See comments atop this
3453 : * function.
3454 : */
3455 9964 : txn = rbtxn_get_toptxn(txn);
3456 :
3457 : Assert(nmsgs > 0);
3458 :
3459 : /* Accumulate invalidations. */
3460 9964 : if (txn->ninvalidations == 0)
3461 : {
3462 2082 : txn->ninvalidations = nmsgs;
3463 2082 : txn->invalidations = (SharedInvalidationMessage *)
3464 2082 : palloc(sizeof(SharedInvalidationMessage) * nmsgs);
3465 2082 : memcpy(txn->invalidations, msgs,
3466 : sizeof(SharedInvalidationMessage) * nmsgs);
3467 : }
3468 : else
3469 : {
3470 7882 : txn->invalidations = (SharedInvalidationMessage *)
3471 7882 : repalloc(txn->invalidations, sizeof(SharedInvalidationMessage) *
3472 7882 : (txn->ninvalidations + nmsgs));
3473 :
3474 7882 : memcpy(txn->invalidations + txn->ninvalidations, msgs,
3475 : nmsgs * sizeof(SharedInvalidationMessage));
3476 7882 : txn->ninvalidations += nmsgs;
3477 : }
3478 :
3479 9964 : change = ReorderBufferGetChange(rb);
3480 9964 : change->action = REORDER_BUFFER_CHANGE_INVALIDATION;
3481 9964 : change->data.inval.ninvalidations = nmsgs;
3482 9964 : change->data.inval.invalidations = (SharedInvalidationMessage *)
3483 9964 : palloc(sizeof(SharedInvalidationMessage) * nmsgs);
3484 9964 : memcpy(change->data.inval.invalidations, msgs,
3485 : sizeof(SharedInvalidationMessage) * nmsgs);
3486 :
3487 9964 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3488 :
3489 9964 : MemoryContextSwitchTo(oldcontext);
3490 9964 : }
3491 :
3492 : /*
3493 : * Apply all invalidations we know. Possibly we only need parts at this point
3494 : * in the changestream but we don't know which those are.
3495 : */
3496 : static void
3497 8228 : ReorderBufferExecuteInvalidations(uint32 nmsgs, SharedInvalidationMessage *msgs)
3498 : {
3499 : int i;
3500 :
3501 87420 : for (i = 0; i < nmsgs; i++)
3502 79192 : LocalExecuteInvalidationMessage(&msgs[i]);
3503 8228 : }
3504 :
3505 : /*
3506 : * Mark a transaction as containing catalog changes
3507 : */
3508 : void
3509 57972 : ReorderBufferXidSetCatalogChanges(ReorderBuffer *rb, TransactionId xid,
3510 : XLogRecPtr lsn)
3511 : {
3512 : ReorderBufferTXN *txn;
3513 :
3514 57972 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3515 :
3516 57972 : if (!rbtxn_has_catalog_changes(txn))
3517 : {
3518 2154 : txn->txn_flags |= RBTXN_HAS_CATALOG_CHANGES;
3519 2154 : dclist_push_tail(&rb->catchange_txns, &txn->catchange_node);
3520 : }
3521 :
3522 : /*
3523 : * Mark top-level transaction as having catalog changes too if one of its
3524 : * children has so that the ReorderBufferBuildTupleCidHash can
3525 : * conveniently check just top-level transaction and decide whether to
3526 : * build the hash table or not.
3527 : */
3528 57972 : if (rbtxn_is_subtxn(txn))
3529 : {
3530 1798 : ReorderBufferTXN *toptxn = rbtxn_get_toptxn(txn);
3531 :
3532 1798 : if (!rbtxn_has_catalog_changes(toptxn))
3533 : {
3534 42 : toptxn->txn_flags |= RBTXN_HAS_CATALOG_CHANGES;
3535 42 : dclist_push_tail(&rb->catchange_txns, &toptxn->catchange_node);
3536 : }
3537 : }
3538 57972 : }
3539 :
3540 : /*
3541 : * Return palloc'ed array of the transactions that have changed catalogs.
3542 : * The returned array is sorted in xidComparator order.
3543 : *
3544 : * The caller must free the returned array when done with it.
3545 : */
3546 : TransactionId *
3547 540 : ReorderBufferGetCatalogChangesXacts(ReorderBuffer *rb)
3548 : {
3549 : dlist_iter iter;
3550 540 : TransactionId *xids = NULL;
3551 540 : size_t xcnt = 0;
3552 :
3553 : /* Quick return if the list is empty */
3554 540 : if (dclist_count(&rb->catchange_txns) == 0)
3555 522 : return NULL;
3556 :
3557 : /* Initialize XID array */
3558 18 : xids = (TransactionId *) palloc(sizeof(TransactionId) *
3559 18 : dclist_count(&rb->catchange_txns));
3560 42 : dclist_foreach(iter, &rb->catchange_txns)
3561 : {
3562 24 : ReorderBufferTXN *txn = dclist_container(ReorderBufferTXN,
3563 : catchange_node,
3564 : iter.cur);
3565 :
3566 : Assert(rbtxn_has_catalog_changes(txn));
3567 :
3568 24 : xids[xcnt++] = txn->xid;
3569 : }
3570 :
3571 18 : qsort(xids, xcnt, sizeof(TransactionId), xidComparator);
3572 :
3573 : Assert(xcnt == dclist_count(&rb->catchange_txns));
3574 18 : return xids;
3575 : }
3576 :
3577 : /*
3578 : * Query whether a transaction is already *known* to contain catalog
3579 : * changes. This can be wrong until directly before the commit!
3580 : */
3581 : bool
3582 7956 : ReorderBufferXidHasCatalogChanges(ReorderBuffer *rb, TransactionId xid)
3583 : {
3584 : ReorderBufferTXN *txn;
3585 :
3586 7956 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3587 : false);
3588 7956 : if (txn == NULL)
3589 1296 : return false;
3590 :
3591 6660 : return rbtxn_has_catalog_changes(txn);
3592 : }
3593 :
3594 : /*
3595 : * ReorderBufferXidHasBaseSnapshot
3596 : * Have we already set the base snapshot for the given txn/subtxn?
3597 : */
3598 : bool
3599 2984102 : ReorderBufferXidHasBaseSnapshot(ReorderBuffer *rb, TransactionId xid)
3600 : {
3601 : ReorderBufferTXN *txn;
3602 :
3603 2984102 : txn = ReorderBufferTXNByXid(rb, xid, false,
3604 : NULL, InvalidXLogRecPtr, false);
3605 :
3606 : /* transaction isn't known yet, ergo no snapshot */
3607 2984102 : if (txn == NULL)
3608 8 : return false;
3609 :
3610 : /* a known subtxn? operate on top-level txn instead */
3611 2984094 : if (rbtxn_is_known_subxact(txn))
3612 714050 : txn = ReorderBufferTXNByXid(rb, txn->toplevel_xid, false,
3613 : NULL, InvalidXLogRecPtr, false);
3614 :
3615 2984094 : return txn->base_snapshot != NULL;
3616 : }
3617 :
3618 :
3619 : /*
3620 : * ---------------------------------------
3621 : * Disk serialization support
3622 : * ---------------------------------------
3623 : */
3624 :
3625 : /*
3626 : * Ensure the IO buffer is >= sz.
3627 : */
3628 : static void
3629 5786774 : ReorderBufferSerializeReserve(ReorderBuffer *rb, Size sz)
3630 : {
3631 5786774 : if (!rb->outbufsize)
3632 : {
3633 94 : rb->outbuf = MemoryContextAlloc(rb->context, sz);
3634 94 : rb->outbufsize = sz;
3635 : }
3636 5786680 : else if (rb->outbufsize < sz)
3637 : {
3638 520 : rb->outbuf = repalloc(rb->outbuf, sz);
3639 520 : rb->outbufsize = sz;
3640 : }
3641 5786774 : }
3642 :
3643 :
3644 : /* Compare two transactions by size */
3645 : static int
3646 662706 : ReorderBufferTXNSizeCompare(const pairingheap_node *a, const pairingheap_node *b, void *arg)
3647 : {
3648 662706 : const ReorderBufferTXN *ta = pairingheap_const_container(ReorderBufferTXN, txn_node, a);
3649 662706 : const ReorderBufferTXN *tb = pairingheap_const_container(ReorderBufferTXN, txn_node, b);
3650 :
3651 662706 : if (ta->size < tb->size)
3652 464594 : return -1;
3653 198112 : if (ta->size > tb->size)
3654 196378 : return 1;
3655 1734 : return 0;
3656 : }
3657 :
3658 : /*
3659 : * Find the largest transaction (toplevel or subxact) to evict (spill to disk).
3660 : */
3661 : static ReorderBufferTXN *
3662 7356 : ReorderBufferLargestTXN(ReorderBuffer *rb)
3663 : {
3664 : ReorderBufferTXN *largest;
3665 :
3666 : /* Get the largest transaction from the max-heap */
3667 7356 : largest = pairingheap_container(ReorderBufferTXN, txn_node,
3668 : pairingheap_first(rb->txn_heap));
3669 :
3670 : Assert(largest);
3671 : Assert(largest->size > 0);
3672 : Assert(largest->size <= rb->size);
3673 :
3674 7356 : return largest;
3675 : }
3676 :
3677 : /*
3678 : * Find the largest streamable (and non-aborted) toplevel transaction to evict
3679 : * (by streaming).
3680 : *
3681 : * This can be seen as an optimized version of ReorderBufferLargestTXN, which
3682 : * should give us the same transaction (because we don't update memory account
3683 : * for subtransaction with streaming, so it's always 0). But we can simply
3684 : * iterate over the limited number of toplevel transactions that have a base
3685 : * snapshot. There is no use of selecting a transaction that doesn't have base
3686 : * snapshot because we don't decode such transactions. Also, we do not select
3687 : * the transaction which doesn't have any streamable change.
3688 : *
3689 : * Note that, we skip transactions that contain incomplete changes. There
3690 : * is a scope of optimization here such that we can select the largest
3691 : * transaction which has incomplete changes. But that will make the code and
3692 : * design quite complex and that might not be worth the benefit. If we plan to
3693 : * stream the transactions that contain incomplete changes then we need to
3694 : * find a way to partially stream/truncate the transaction changes in-memory
3695 : * and build a mechanism to partially truncate the spilled files.
3696 : * Additionally, whenever we partially stream the transaction we need to
3697 : * maintain the last streamed lsn and next time we need to restore from that
3698 : * segment and the offset in WAL. As we stream the changes from the top
3699 : * transaction and restore them subtransaction wise, we need to even remember
3700 : * the subxact from where we streamed the last change.
3701 : */
3702 : static ReorderBufferTXN *
3703 1636 : ReorderBufferLargestStreamableTopTXN(ReorderBuffer *rb)
3704 : {
3705 : dlist_iter iter;
3706 1636 : Size largest_size = 0;
3707 1636 : ReorderBufferTXN *largest = NULL;
3708 :
3709 : /* Find the largest top-level transaction having a base snapshot. */
3710 3496 : dlist_foreach(iter, &rb->txns_by_base_snapshot_lsn)
3711 : {
3712 : ReorderBufferTXN *txn;
3713 :
3714 1860 : txn = dlist_container(ReorderBufferTXN, base_snapshot_node, iter.cur);
3715 :
3716 : /* must not be a subtxn */
3717 : Assert(!rbtxn_is_known_subxact(txn));
3718 : /* base_snapshot must be set */
3719 : Assert(txn->base_snapshot != NULL);
3720 :
3721 : /* Don't consider these kinds of transactions for eviction. */
3722 1860 : if (rbtxn_has_partial_change(txn) ||
3723 1566 : !rbtxn_has_streamable_change(txn) ||
3724 1506 : rbtxn_is_aborted(txn))
3725 354 : continue;
3726 :
3727 : /* Find the largest of the eviction candidates. */
3728 1506 : if ((largest == NULL || txn->total_size > largest_size) &&
3729 1506 : (txn->total_size > 0))
3730 : {
3731 1414 : largest = txn;
3732 1414 : largest_size = txn->total_size;
3733 : }
3734 : }
3735 :
3736 1636 : return largest;
3737 : }
3738 :
3739 : /*
3740 : * Check whether the logical_decoding_work_mem limit was reached, and if yes
3741 : * pick the largest (sub)transaction at-a-time to evict and spill its changes to
3742 : * disk or send to the output plugin until we reach under the memory limit.
3743 : *
3744 : * If debug_logical_replication_streaming is set to "immediate", stream or
3745 : * serialize the changes immediately.
3746 : *
3747 : * XXX At this point we select the transactions until we reach under the memory
3748 : * limit, but we might also adapt a more elaborate eviction strategy - for example
3749 : * evicting enough transactions to free certain fraction (e.g. 50%) of the memory
3750 : * limit.
3751 : */
3752 : static void
3753 3003620 : ReorderBufferCheckMemoryLimit(ReorderBuffer *rb)
3754 : {
3755 : ReorderBufferTXN *txn;
3756 :
3757 : /*
3758 : * Bail out if debug_logical_replication_streaming is buffered and we
3759 : * haven't exceeded the memory limit.
3760 : */
3761 3003620 : if (debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_BUFFERED &&
3762 3001690 : rb->size < logical_decoding_work_mem * (Size) 1024)
3763 2994976 : return;
3764 :
3765 : /*
3766 : * If debug_logical_replication_streaming is immediate, loop until there's
3767 : * no change. Otherwise, loop until we reach under the memory limit. One
3768 : * might think that just by evicting the largest (sub)transaction we will
3769 : * come under the memory limit based on assumption that the selected
3770 : * transaction is at least as large as the most recent change (which
3771 : * caused us to go over the memory limit). However, that is not true
3772 : * because a user can reduce the logical_decoding_work_mem to a smaller
3773 : * value before the most recent change.
3774 : */
3775 17282 : while (rb->size >= logical_decoding_work_mem * (Size) 1024 ||
3776 10568 : (debug_logical_replication_streaming == DEBUG_LOGICAL_REP_STREAMING_IMMEDIATE &&
3777 3854 : rb->size > 0))
3778 : {
3779 : /*
3780 : * Pick the largest non-aborted transaction and evict it from memory
3781 : * by streaming, if possible. Otherwise, spill to disk.
3782 : */
3783 10274 : if (ReorderBufferCanStartStreaming(rb) &&
3784 1636 : (txn = ReorderBufferLargestStreamableTopTXN(rb)) != NULL)
3785 : {
3786 : /* we know there has to be one, because the size is not zero */
3787 : Assert(txn && rbtxn_is_toptxn(txn));
3788 : Assert(txn->total_size > 0);
3789 : Assert(rb->size >= txn->total_size);
3790 :
3791 : /* skip the transaction if aborted */
3792 1282 : if (ReorderBufferCheckAndTruncateAbortedTXN(rb, txn))
3793 0 : continue;
3794 :
3795 1282 : ReorderBufferStreamTXN(rb, txn);
3796 : }
3797 : else
3798 : {
3799 : /*
3800 : * Pick the largest transaction (or subtransaction) and evict it
3801 : * from memory by serializing it to disk.
3802 : */
3803 7356 : txn = ReorderBufferLargestTXN(rb);
3804 :
3805 : /* we know there has to be one, because the size is not zero */
3806 : Assert(txn);
3807 : Assert(txn->size > 0);
3808 : Assert(rb->size >= txn->size);
3809 :
3810 : /* skip the transaction if aborted */
3811 7356 : if (ReorderBufferCheckAndTruncateAbortedTXN(rb, txn))
3812 18 : continue;
3813 :
3814 7338 : ReorderBufferSerializeTXN(rb, txn);
3815 : }
3816 :
3817 : /*
3818 : * After eviction, the transaction should have no entries in memory,
3819 : * and should use 0 bytes for changes.
3820 : */
3821 : Assert(txn->size == 0);
3822 : Assert(txn->nentries_mem == 0);
3823 : }
3824 :
3825 : /* We must be under the memory limit now. */
3826 : Assert(rb->size < logical_decoding_work_mem * (Size) 1024);
3827 : }
3828 :
3829 : /*
3830 : * Spill data of a large transaction (and its subtransactions) to disk.
3831 : */
3832 : static void
3833 7812 : ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
3834 : {
3835 : dlist_iter subtxn_i;
3836 : dlist_mutable_iter change_i;
3837 7812 : int fd = -1;
3838 7812 : XLogSegNo curOpenSegNo = 0;
3839 7812 : Size spilled = 0;
3840 7812 : Size size = txn->size;
3841 :
3842 7812 : elog(DEBUG2, "spill %u changes in XID %u to disk",
3843 : (uint32) txn->nentries_mem, txn->xid);
3844 :
3845 : /* do the same to all child TXs */
3846 8206 : dlist_foreach(subtxn_i, &txn->subtxns)
3847 : {
3848 : ReorderBufferTXN *subtxn;
3849 :
3850 394 : subtxn = dlist_container(ReorderBufferTXN, node, subtxn_i.cur);
3851 394 : ReorderBufferSerializeTXN(rb, subtxn);
3852 : }
3853 :
3854 : /* serialize changestream */
3855 2561214 : dlist_foreach_modify(change_i, &txn->changes)
3856 : {
3857 : ReorderBufferChange *change;
3858 :
3859 2553402 : change = dlist_container(ReorderBufferChange, node, change_i.cur);
3860 :
3861 : /*
3862 : * store in segment in which it belongs by start lsn, don't split over
3863 : * multiple segments tho
3864 : */
3865 2553402 : if (fd == -1 ||
3866 2545962 : !XLByteInSeg(change->lsn, curOpenSegNo, wal_segment_size))
3867 : {
3868 : char path[MAXPGPATH];
3869 :
3870 7452 : if (fd != -1)
3871 12 : CloseTransientFile(fd);
3872 :
3873 7452 : XLByteToSeg(change->lsn, curOpenSegNo, wal_segment_size);
3874 :
3875 : /*
3876 : * No need to care about TLIs here, only used during a single run,
3877 : * so each LSN only maps to a specific WAL record.
3878 : */
3879 7452 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid,
3880 : curOpenSegNo);
3881 :
3882 : /* open segment, create it if necessary */
3883 7452 : fd = OpenTransientFile(path,
3884 : O_CREAT | O_WRONLY | O_APPEND | PG_BINARY);
3885 :
3886 7452 : if (fd < 0)
3887 0 : ereport(ERROR,
3888 : (errcode_for_file_access(),
3889 : errmsg("could not open file \"%s\": %m", path)));
3890 : }
3891 :
3892 2553402 : ReorderBufferSerializeChange(rb, txn, fd, change);
3893 2553402 : dlist_delete(&change->node);
3894 2553402 : ReorderBufferReturnChange(rb, change, false);
3895 :
3896 2553402 : spilled++;
3897 : }
3898 :
3899 : /* Update the memory counter */
3900 7812 : ReorderBufferChangeMemoryUpdate(rb, NULL, txn, false, size);
3901 :
3902 : /* update the statistics iff we have spilled anything */
3903 7812 : if (spilled)
3904 : {
3905 7440 : rb->spillCount += 1;
3906 7440 : rb->spillBytes += size;
3907 :
3908 : /* don't consider already serialized transactions */
3909 7440 : rb->spillTxns += (rbtxn_is_serialized(txn) || rbtxn_is_serialized_clear(txn)) ? 0 : 1;
3910 :
3911 : /* update the decoding stats */
3912 7440 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
3913 : }
3914 :
3915 : Assert(spilled == txn->nentries_mem);
3916 : Assert(dlist_is_empty(&txn->changes));
3917 7812 : txn->nentries_mem = 0;
3918 7812 : txn->txn_flags |= RBTXN_IS_SERIALIZED;
3919 :
3920 7812 : if (fd != -1)
3921 7440 : CloseTransientFile(fd);
3922 7812 : }
3923 :
3924 : /*
3925 : * Serialize individual change to disk.
3926 : */
3927 : static void
3928 2553402 : ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
3929 : int fd, ReorderBufferChange *change)
3930 : {
3931 : ReorderBufferDiskChange *ondisk;
3932 2553402 : Size sz = sizeof(ReorderBufferDiskChange);
3933 :
3934 2553402 : ReorderBufferSerializeReserve(rb, sz);
3935 :
3936 2553402 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
3937 2553402 : memcpy(&ondisk->change, change, sizeof(ReorderBufferChange));
3938 :
3939 2553402 : switch (change->action)
3940 : {
3941 : /* fall through these, they're all similar enough */
3942 2518426 : case REORDER_BUFFER_CHANGE_INSERT:
3943 : case REORDER_BUFFER_CHANGE_UPDATE:
3944 : case REORDER_BUFFER_CHANGE_DELETE:
3945 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
3946 : {
3947 : char *data;
3948 : HeapTuple oldtup,
3949 : newtup;
3950 2518426 : Size oldlen = 0;
3951 2518426 : Size newlen = 0;
3952 :
3953 2518426 : oldtup = change->data.tp.oldtuple;
3954 2518426 : newtup = change->data.tp.newtuple;
3955 :
3956 2518426 : if (oldtup)
3957 : {
3958 320254 : sz += sizeof(HeapTupleData);
3959 320254 : oldlen = oldtup->t_len;
3960 320254 : sz += oldlen;
3961 : }
3962 :
3963 2518426 : if (newtup)
3964 : {
3965 2090742 : sz += sizeof(HeapTupleData);
3966 2090742 : newlen = newtup->t_len;
3967 2090742 : sz += newlen;
3968 : }
3969 :
3970 : /* make sure we have enough space */
3971 2518426 : ReorderBufferSerializeReserve(rb, sz);
3972 :
3973 2518426 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
3974 : /* might have been reallocated above */
3975 2518426 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
3976 :
3977 2518426 : if (oldlen)
3978 : {
3979 320254 : memcpy(data, oldtup, sizeof(HeapTupleData));
3980 320254 : data += sizeof(HeapTupleData);
3981 :
3982 320254 : memcpy(data, oldtup->t_data, oldlen);
3983 320254 : data += oldlen;
3984 : }
3985 :
3986 2518426 : if (newlen)
3987 : {
3988 2090742 : memcpy(data, newtup, sizeof(HeapTupleData));
3989 2090742 : data += sizeof(HeapTupleData);
3990 :
3991 2090742 : memcpy(data, newtup->t_data, newlen);
3992 2090742 : data += newlen;
3993 : }
3994 2518426 : break;
3995 : }
3996 26 : case REORDER_BUFFER_CHANGE_MESSAGE:
3997 : {
3998 : char *data;
3999 26 : Size prefix_size = strlen(change->data.msg.prefix) + 1;
4000 :
4001 26 : sz += prefix_size + change->data.msg.message_size +
4002 : sizeof(Size) + sizeof(Size);
4003 26 : ReorderBufferSerializeReserve(rb, sz);
4004 :
4005 26 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4006 :
4007 : /* might have been reallocated above */
4008 26 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4009 :
4010 : /* write the prefix including the size */
4011 26 : memcpy(data, &prefix_size, sizeof(Size));
4012 26 : data += sizeof(Size);
4013 26 : memcpy(data, change->data.msg.prefix,
4014 : prefix_size);
4015 26 : data += prefix_size;
4016 :
4017 : /* write the message including the size */
4018 26 : memcpy(data, &change->data.msg.message_size, sizeof(Size));
4019 26 : data += sizeof(Size);
4020 26 : memcpy(data, change->data.msg.message,
4021 : change->data.msg.message_size);
4022 26 : data += change->data.msg.message_size;
4023 :
4024 26 : break;
4025 : }
4026 308 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4027 : {
4028 : char *data;
4029 308 : Size inval_size = sizeof(SharedInvalidationMessage) *
4030 308 : change->data.inval.ninvalidations;
4031 :
4032 308 : sz += inval_size;
4033 :
4034 308 : ReorderBufferSerializeReserve(rb, sz);
4035 308 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4036 :
4037 : /* might have been reallocated above */
4038 308 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4039 308 : memcpy(data, change->data.inval.invalidations, inval_size);
4040 308 : data += inval_size;
4041 :
4042 308 : break;
4043 : }
4044 16 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4045 : {
4046 : Snapshot snap;
4047 : char *data;
4048 :
4049 16 : snap = change->data.snapshot;
4050 :
4051 16 : sz += sizeof(SnapshotData) +
4052 16 : sizeof(TransactionId) * snap->xcnt +
4053 16 : sizeof(TransactionId) * snap->subxcnt;
4054 :
4055 : /* make sure we have enough space */
4056 16 : ReorderBufferSerializeReserve(rb, sz);
4057 16 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4058 : /* might have been reallocated above */
4059 16 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4060 :
4061 16 : memcpy(data, snap, sizeof(SnapshotData));
4062 16 : data += sizeof(SnapshotData);
4063 :
4064 16 : if (snap->xcnt)
4065 : {
4066 16 : memcpy(data, snap->xip,
4067 16 : sizeof(TransactionId) * snap->xcnt);
4068 16 : data += sizeof(TransactionId) * snap->xcnt;
4069 : }
4070 :
4071 16 : if (snap->subxcnt)
4072 : {
4073 0 : memcpy(data, snap->subxip,
4074 0 : sizeof(TransactionId) * snap->subxcnt);
4075 0 : data += sizeof(TransactionId) * snap->subxcnt;
4076 : }
4077 16 : break;
4078 : }
4079 4 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4080 : {
4081 : Size size;
4082 : char *data;
4083 :
4084 : /* account for the OIDs of truncated relations */
4085 4 : size = sizeof(Oid) * change->data.truncate.nrelids;
4086 4 : sz += size;
4087 :
4088 : /* make sure we have enough space */
4089 4 : ReorderBufferSerializeReserve(rb, sz);
4090 :
4091 4 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
4092 : /* might have been reallocated above */
4093 4 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4094 :
4095 4 : memcpy(data, change->data.truncate.relids, size);
4096 4 : data += size;
4097 :
4098 4 : break;
4099 : }
4100 34622 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4101 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4102 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4103 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4104 : /* ReorderBufferChange contains everything important */
4105 34622 : break;
4106 : }
4107 :
4108 2553402 : ondisk->size = sz;
4109 :
4110 2553402 : errno = 0;
4111 2553402 : pgstat_report_wait_start(WAIT_EVENT_REORDER_BUFFER_WRITE);
4112 2553402 : if (write(fd, rb->outbuf, ondisk->size) != ondisk->size)
4113 : {
4114 0 : int save_errno = errno;
4115 :
4116 0 : CloseTransientFile(fd);
4117 :
4118 : /* if write didn't set errno, assume problem is no disk space */
4119 0 : errno = save_errno ? save_errno : ENOSPC;
4120 0 : ereport(ERROR,
4121 : (errcode_for_file_access(),
4122 : errmsg("could not write to data file for XID %u: %m",
4123 : txn->xid)));
4124 : }
4125 2553402 : pgstat_report_wait_end();
4126 :
4127 : /*
4128 : * Keep the transaction's final_lsn up to date with each change we send to
4129 : * disk, so that ReorderBufferRestoreCleanup works correctly. (We used to
4130 : * only do this on commit and abort records, but that doesn't work if a
4131 : * system crash leaves a transaction without its abort record).
4132 : *
4133 : * Make sure not to move it backwards.
4134 : */
4135 2553402 : if (txn->final_lsn < change->lsn)
4136 2544438 : txn->final_lsn = change->lsn;
4137 :
4138 : Assert(ondisk->change.action == change->action);
4139 2553402 : }
4140 :
4141 : /* Returns true, if the output plugin supports streaming, false, otherwise. */
4142 : static inline bool
4143 4015950 : ReorderBufferCanStream(ReorderBuffer *rb)
4144 : {
4145 4015950 : LogicalDecodingContext *ctx = rb->private_data;
4146 :
4147 4015950 : return ctx->streaming;
4148 : }
4149 :
4150 : /* Returns true, if the streaming can be started now, false, otherwise. */
4151 : static inline bool
4152 1012330 : ReorderBufferCanStartStreaming(ReorderBuffer *rb)
4153 : {
4154 1012330 : LogicalDecodingContext *ctx = rb->private_data;
4155 1012330 : SnapBuild *builder = ctx->snapshot_builder;
4156 :
4157 : /* We can't start streaming unless a consistent state is reached. */
4158 1012330 : if (SnapBuildCurrentState(builder) < SNAPBUILD_CONSISTENT)
4159 0 : return false;
4160 :
4161 : /*
4162 : * We can't start streaming immediately even if the streaming is enabled
4163 : * because we previously decoded this transaction and now just are
4164 : * restarting.
4165 : */
4166 1012330 : if (ReorderBufferCanStream(rb) &&
4167 1008054 : !SnapBuildXactNeedsSkip(builder, ctx->reader->ReadRecPtr))
4168 337728 : return true;
4169 :
4170 674602 : return false;
4171 : }
4172 :
4173 : /*
4174 : * Send data of a large transaction (and its subtransactions) to the
4175 : * output plugin, but using the stream API.
4176 : */
4177 : static void
4178 1432 : ReorderBufferStreamTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
4179 : {
4180 : Snapshot snapshot_now;
4181 : CommandId command_id;
4182 : Size stream_bytes;
4183 : bool txn_is_streamed;
4184 :
4185 : /* We can never reach here for a subtransaction. */
4186 : Assert(rbtxn_is_toptxn(txn));
4187 :
4188 : /*
4189 : * We can't make any assumptions about base snapshot here, similar to what
4190 : * ReorderBufferCommit() does. That relies on base_snapshot getting
4191 : * transferred from subxact in ReorderBufferCommitChild(), but that was
4192 : * not yet called as the transaction is in-progress.
4193 : *
4194 : * So just walk the subxacts and use the same logic here. But we only need
4195 : * to do that once, when the transaction is streamed for the first time.
4196 : * After that we need to reuse the snapshot from the previous run.
4197 : *
4198 : * Unlike DecodeCommit which adds xids of all the subtransactions in
4199 : * snapshot's xip array via SnapBuildCommitTxn, we can't do that here but
4200 : * we do add them to subxip array instead via ReorderBufferCopySnap. This
4201 : * allows the catalog changes made in subtransactions decoded till now to
4202 : * be visible.
4203 : */
4204 1432 : if (txn->snapshot_now == NULL)
4205 : {
4206 : dlist_iter subxact_i;
4207 :
4208 : /* make sure this transaction is streamed for the first time */
4209 : Assert(!rbtxn_is_streamed(txn));
4210 :
4211 : /* at the beginning we should have invalid command ID */
4212 : Assert(txn->command_id == InvalidCommandId);
4213 :
4214 152 : dlist_foreach(subxact_i, &txn->subtxns)
4215 : {
4216 : ReorderBufferTXN *subtxn;
4217 :
4218 8 : subtxn = dlist_container(ReorderBufferTXN, node, subxact_i.cur);
4219 8 : ReorderBufferTransferSnapToParent(txn, subtxn);
4220 : }
4221 :
4222 : /*
4223 : * If this transaction has no snapshot, it didn't make any changes to
4224 : * the database till now, so there's nothing to decode.
4225 : */
4226 144 : if (txn->base_snapshot == NULL)
4227 : {
4228 : Assert(txn->ninvalidations == 0);
4229 0 : return;
4230 : }
4231 :
4232 144 : command_id = FirstCommandId;
4233 144 : snapshot_now = ReorderBufferCopySnap(rb, txn->base_snapshot,
4234 : txn, command_id);
4235 : }
4236 : else
4237 : {
4238 : /* the transaction must have been already streamed */
4239 : Assert(rbtxn_is_streamed(txn));
4240 :
4241 : /*
4242 : * Nah, we already have snapshot from the previous streaming run. We
4243 : * assume new subxacts can't move the LSN backwards, and so can't beat
4244 : * the LSN condition in the previous branch (so no need to walk
4245 : * through subxacts again). In fact, we must not do that as we may be
4246 : * using snapshot half-way through the subxact.
4247 : */
4248 1288 : command_id = txn->command_id;
4249 :
4250 : /*
4251 : * We can't use txn->snapshot_now directly because after the last
4252 : * streaming run, we might have got some new sub-transactions. So we
4253 : * need to add them to the snapshot.
4254 : */
4255 1288 : snapshot_now = ReorderBufferCopySnap(rb, txn->snapshot_now,
4256 : txn, command_id);
4257 :
4258 : /* Free the previously copied snapshot. */
4259 : Assert(txn->snapshot_now->copied);
4260 1288 : ReorderBufferFreeSnap(rb, txn->snapshot_now);
4261 1288 : txn->snapshot_now = NULL;
4262 : }
4263 :
4264 : /*
4265 : * Remember this information to be used later to update stats. We can't
4266 : * update the stats here as an error while processing the changes would
4267 : * lead to the accumulation of stats even though we haven't streamed all
4268 : * the changes.
4269 : */
4270 1432 : txn_is_streamed = rbtxn_is_streamed(txn);
4271 1432 : stream_bytes = txn->total_size;
4272 :
4273 : /* Process and send the changes to output plugin. */
4274 1432 : ReorderBufferProcessTXN(rb, txn, InvalidXLogRecPtr, snapshot_now,
4275 : command_id, true);
4276 :
4277 1432 : rb->streamCount += 1;
4278 1432 : rb->streamBytes += stream_bytes;
4279 :
4280 : /* Don't consider already streamed transaction. */
4281 1432 : rb->streamTxns += (txn_is_streamed) ? 0 : 1;
4282 :
4283 : /* update the decoding stats */
4284 1432 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
4285 :
4286 : Assert(dlist_is_empty(&txn->changes));
4287 : Assert(txn->nentries == 0);
4288 : Assert(txn->nentries_mem == 0);
4289 : }
4290 :
4291 : /*
4292 : * Size of a change in memory.
4293 : */
4294 : static Size
4295 4216872 : ReorderBufferChangeSize(ReorderBufferChange *change)
4296 : {
4297 4216872 : Size sz = sizeof(ReorderBufferChange);
4298 :
4299 4216872 : switch (change->action)
4300 : {
4301 : /* fall through these, they're all similar enough */
4302 4004858 : case REORDER_BUFFER_CHANGE_INSERT:
4303 : case REORDER_BUFFER_CHANGE_UPDATE:
4304 : case REORDER_BUFFER_CHANGE_DELETE:
4305 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4306 : {
4307 : HeapTuple oldtup,
4308 : newtup;
4309 4004858 : Size oldlen = 0;
4310 4004858 : Size newlen = 0;
4311 :
4312 4004858 : oldtup = change->data.tp.oldtuple;
4313 4004858 : newtup = change->data.tp.newtuple;
4314 :
4315 4004858 : if (oldtup)
4316 : {
4317 523902 : sz += sizeof(HeapTupleData);
4318 523902 : oldlen = oldtup->t_len;
4319 523902 : sz += oldlen;
4320 : }
4321 :
4322 4004858 : if (newtup)
4323 : {
4324 3315288 : sz += sizeof(HeapTupleData);
4325 3315288 : newlen = newtup->t_len;
4326 3315288 : sz += newlen;
4327 : }
4328 :
4329 4004858 : break;
4330 : }
4331 134 : case REORDER_BUFFER_CHANGE_MESSAGE:
4332 : {
4333 134 : Size prefix_size = strlen(change->data.msg.prefix) + 1;
4334 :
4335 134 : sz += prefix_size + change->data.msg.message_size +
4336 : sizeof(Size) + sizeof(Size);
4337 :
4338 134 : break;
4339 : }
4340 19516 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4341 : {
4342 19516 : sz += sizeof(SharedInvalidationMessage) *
4343 19516 : change->data.inval.ninvalidations;
4344 19516 : break;
4345 : }
4346 4178 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4347 : {
4348 : Snapshot snap;
4349 :
4350 4178 : snap = change->data.snapshot;
4351 :
4352 4178 : sz += sizeof(SnapshotData) +
4353 4178 : sizeof(TransactionId) * snap->xcnt +
4354 4178 : sizeof(TransactionId) * snap->subxcnt;
4355 :
4356 4178 : break;
4357 : }
4358 174 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4359 : {
4360 174 : sz += sizeof(Oid) * change->data.truncate.nrelids;
4361 :
4362 174 : break;
4363 : }
4364 188012 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4365 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4366 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4367 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4368 : /* ReorderBufferChange contains everything important */
4369 188012 : break;
4370 : }
4371 :
4372 4216872 : return sz;
4373 : }
4374 :
4375 :
4376 : /*
4377 : * Restore a number of changes spilled to disk back into memory.
4378 : */
4379 : static Size
4380 210 : ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
4381 : TXNEntryFile *file, XLogSegNo *segno)
4382 : {
4383 210 : Size restored = 0;
4384 : XLogSegNo last_segno;
4385 : dlist_mutable_iter cleanup_iter;
4386 210 : File *fd = &file->vfd;
4387 :
4388 : Assert(txn->first_lsn != InvalidXLogRecPtr);
4389 : Assert(txn->final_lsn != InvalidXLogRecPtr);
4390 :
4391 : /* free current entries, so we have memory for more */
4392 350138 : dlist_foreach_modify(cleanup_iter, &txn->changes)
4393 : {
4394 349928 : ReorderBufferChange *cleanup =
4395 349928 : dlist_container(ReorderBufferChange, node, cleanup_iter.cur);
4396 :
4397 349928 : dlist_delete(&cleanup->node);
4398 349928 : ReorderBufferReturnChange(rb, cleanup, true);
4399 : }
4400 210 : txn->nentries_mem = 0;
4401 : Assert(dlist_is_empty(&txn->changes));
4402 :
4403 210 : XLByteToSeg(txn->final_lsn, last_segno, wal_segment_size);
4404 :
4405 357548 : while (restored < max_changes_in_memory && *segno <= last_segno)
4406 : {
4407 : int readBytes;
4408 : ReorderBufferDiskChange *ondisk;
4409 :
4410 357338 : CHECK_FOR_INTERRUPTS();
4411 :
4412 357338 : if (*fd == -1)
4413 : {
4414 : char path[MAXPGPATH];
4415 :
4416 : /* first time in */
4417 84 : if (*segno == 0)
4418 80 : XLByteToSeg(txn->first_lsn, *segno, wal_segment_size);
4419 :
4420 : Assert(*segno != 0 || dlist_is_empty(&txn->changes));
4421 :
4422 : /*
4423 : * No need to care about TLIs here, only used during a single run,
4424 : * so each LSN only maps to a specific WAL record.
4425 : */
4426 84 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid,
4427 : *segno);
4428 :
4429 84 : *fd = PathNameOpenFile(path, O_RDONLY | PG_BINARY);
4430 :
4431 : /* No harm in resetting the offset even in case of failure */
4432 84 : file->curOffset = 0;
4433 :
4434 84 : if (*fd < 0 && errno == ENOENT)
4435 : {
4436 0 : *fd = -1;
4437 0 : (*segno)++;
4438 0 : continue;
4439 : }
4440 84 : else if (*fd < 0)
4441 0 : ereport(ERROR,
4442 : (errcode_for_file_access(),
4443 : errmsg("could not open file \"%s\": %m",
4444 : path)));
4445 : }
4446 :
4447 : /*
4448 : * Read the statically sized part of a change which has information
4449 : * about the total size. If we couldn't read a record, we're at the
4450 : * end of this file.
4451 : */
4452 357338 : ReorderBufferSerializeReserve(rb, sizeof(ReorderBufferDiskChange));
4453 357338 : readBytes = FileRead(file->vfd, rb->outbuf,
4454 : sizeof(ReorderBufferDiskChange),
4455 : file->curOffset, WAIT_EVENT_REORDER_BUFFER_READ);
4456 :
4457 : /* eof */
4458 357338 : if (readBytes == 0)
4459 : {
4460 84 : FileClose(*fd);
4461 84 : *fd = -1;
4462 84 : (*segno)++;
4463 84 : continue;
4464 : }
4465 357254 : else if (readBytes < 0)
4466 0 : ereport(ERROR,
4467 : (errcode_for_file_access(),
4468 : errmsg("could not read from reorderbuffer spill file: %m")));
4469 357254 : else if (readBytes != sizeof(ReorderBufferDiskChange))
4470 0 : ereport(ERROR,
4471 : (errcode_for_file_access(),
4472 : errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
4473 : readBytes,
4474 : (uint32) sizeof(ReorderBufferDiskChange))));
4475 :
4476 357254 : file->curOffset += readBytes;
4477 :
4478 357254 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4479 :
4480 357254 : ReorderBufferSerializeReserve(rb,
4481 357254 : sizeof(ReorderBufferDiskChange) + ondisk->size);
4482 357254 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4483 :
4484 714508 : readBytes = FileRead(file->vfd,
4485 357254 : rb->outbuf + sizeof(ReorderBufferDiskChange),
4486 357254 : ondisk->size - sizeof(ReorderBufferDiskChange),
4487 : file->curOffset,
4488 : WAIT_EVENT_REORDER_BUFFER_READ);
4489 :
4490 357254 : if (readBytes < 0)
4491 0 : ereport(ERROR,
4492 : (errcode_for_file_access(),
4493 : errmsg("could not read from reorderbuffer spill file: %m")));
4494 357254 : else if (readBytes != ondisk->size - sizeof(ReorderBufferDiskChange))
4495 0 : ereport(ERROR,
4496 : (errcode_for_file_access(),
4497 : errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
4498 : readBytes,
4499 : (uint32) (ondisk->size - sizeof(ReorderBufferDiskChange)))));
4500 :
4501 357254 : file->curOffset += readBytes;
4502 :
4503 : /*
4504 : * ok, read a full change from disk, now restore it into proper
4505 : * in-memory format
4506 : */
4507 357254 : ReorderBufferRestoreChange(rb, txn, rb->outbuf);
4508 357254 : restored++;
4509 : }
4510 :
4511 210 : return restored;
4512 : }
4513 :
4514 : /*
4515 : * Convert change from its on-disk format to in-memory format and queue it onto
4516 : * the TXN's ->changes list.
4517 : *
4518 : * Note: although "data" is declared char*, at entry it points to a
4519 : * maxalign'd buffer, making it safe in most of this function to assume
4520 : * that the pointed-to data is suitably aligned for direct access.
4521 : */
4522 : static void
4523 357254 : ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
4524 : char *data)
4525 : {
4526 : ReorderBufferDiskChange *ondisk;
4527 : ReorderBufferChange *change;
4528 :
4529 357254 : ondisk = (ReorderBufferDiskChange *) data;
4530 :
4531 357254 : change = ReorderBufferGetChange(rb);
4532 :
4533 : /* copy static part */
4534 357254 : memcpy(change, &ondisk->change, sizeof(ReorderBufferChange));
4535 :
4536 357254 : data += sizeof(ReorderBufferDiskChange);
4537 :
4538 : /* restore individual stuff */
4539 357254 : switch (change->action)
4540 : {
4541 : /* fall through these, they're all similar enough */
4542 353396 : case REORDER_BUFFER_CHANGE_INSERT:
4543 : case REORDER_BUFFER_CHANGE_UPDATE:
4544 : case REORDER_BUFFER_CHANGE_DELETE:
4545 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4546 353396 : if (change->data.tp.oldtuple)
4547 : {
4548 10012 : uint32 tuplelen = ((HeapTuple) data)->t_len;
4549 :
4550 10012 : change->data.tp.oldtuple =
4551 10012 : ReorderBufferGetTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
4552 :
4553 : /* restore ->tuple */
4554 10012 : memcpy(change->data.tp.oldtuple, data,
4555 : sizeof(HeapTupleData));
4556 10012 : data += sizeof(HeapTupleData);
4557 :
4558 : /* reset t_data pointer into the new tuplebuf */
4559 10012 : change->data.tp.oldtuple->t_data =
4560 10012 : (HeapTupleHeader) ((char *) change->data.tp.oldtuple + HEAPTUPLESIZE);
4561 :
4562 : /* restore tuple data itself */
4563 10012 : memcpy(change->data.tp.oldtuple->t_data, data, tuplelen);
4564 10012 : data += tuplelen;
4565 : }
4566 :
4567 353396 : if (change->data.tp.newtuple)
4568 : {
4569 : /* here, data might not be suitably aligned! */
4570 : uint32 tuplelen;
4571 :
4572 332954 : memcpy(&tuplelen, data + offsetof(HeapTupleData, t_len),
4573 : sizeof(uint32));
4574 :
4575 332954 : change->data.tp.newtuple =
4576 332954 : ReorderBufferGetTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
4577 :
4578 : /* restore ->tuple */
4579 332954 : memcpy(change->data.tp.newtuple, data,
4580 : sizeof(HeapTupleData));
4581 332954 : data += sizeof(HeapTupleData);
4582 :
4583 : /* reset t_data pointer into the new tuplebuf */
4584 332954 : change->data.tp.newtuple->t_data =
4585 332954 : (HeapTupleHeader) ((char *) change->data.tp.newtuple + HEAPTUPLESIZE);
4586 :
4587 : /* restore tuple data itself */
4588 332954 : memcpy(change->data.tp.newtuple->t_data, data, tuplelen);
4589 332954 : data += tuplelen;
4590 : }
4591 :
4592 353396 : break;
4593 2 : case REORDER_BUFFER_CHANGE_MESSAGE:
4594 : {
4595 : Size prefix_size;
4596 :
4597 : /* read prefix */
4598 2 : memcpy(&prefix_size, data, sizeof(Size));
4599 2 : data += sizeof(Size);
4600 2 : change->data.msg.prefix = MemoryContextAlloc(rb->context,
4601 : prefix_size);
4602 2 : memcpy(change->data.msg.prefix, data, prefix_size);
4603 : Assert(change->data.msg.prefix[prefix_size - 1] == '\0');
4604 2 : data += prefix_size;
4605 :
4606 : /* read the message */
4607 2 : memcpy(&change->data.msg.message_size, data, sizeof(Size));
4608 2 : data += sizeof(Size);
4609 2 : change->data.msg.message = MemoryContextAlloc(rb->context,
4610 : change->data.msg.message_size);
4611 2 : memcpy(change->data.msg.message, data,
4612 : change->data.msg.message_size);
4613 2 : data += change->data.msg.message_size;
4614 :
4615 2 : break;
4616 : }
4617 46 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4618 : {
4619 46 : Size inval_size = sizeof(SharedInvalidationMessage) *
4620 46 : change->data.inval.ninvalidations;
4621 :
4622 46 : change->data.inval.invalidations =
4623 46 : MemoryContextAlloc(rb->context, inval_size);
4624 :
4625 : /* read the message */
4626 46 : memcpy(change->data.inval.invalidations, data, inval_size);
4627 :
4628 46 : break;
4629 : }
4630 4 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4631 : {
4632 : Snapshot oldsnap;
4633 : Snapshot newsnap;
4634 : Size size;
4635 :
4636 4 : oldsnap = (Snapshot) data;
4637 :
4638 4 : size = sizeof(SnapshotData) +
4639 4 : sizeof(TransactionId) * oldsnap->xcnt +
4640 4 : sizeof(TransactionId) * (oldsnap->subxcnt + 0);
4641 :
4642 4 : change->data.snapshot = MemoryContextAllocZero(rb->context, size);
4643 :
4644 4 : newsnap = change->data.snapshot;
4645 :
4646 4 : memcpy(newsnap, data, size);
4647 4 : newsnap->xip = (TransactionId *)
4648 : (((char *) newsnap) + sizeof(SnapshotData));
4649 4 : newsnap->subxip = newsnap->xip + newsnap->xcnt;
4650 4 : newsnap->copied = true;
4651 4 : break;
4652 : }
4653 : /* the base struct contains all the data, easy peasy */
4654 0 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4655 : {
4656 : Oid *relids;
4657 :
4658 0 : relids = ReorderBufferGetRelids(rb,
4659 0 : change->data.truncate.nrelids);
4660 0 : memcpy(relids, data, change->data.truncate.nrelids * sizeof(Oid));
4661 0 : change->data.truncate.relids = relids;
4662 :
4663 0 : break;
4664 : }
4665 3806 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4666 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4667 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4668 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4669 3806 : break;
4670 : }
4671 :
4672 357254 : dlist_push_tail(&txn->changes, &change->node);
4673 357254 : txn->nentries_mem++;
4674 :
4675 : /*
4676 : * Update memory accounting for the restored change. We need to do this
4677 : * although we don't check the memory limit when restoring the changes in
4678 : * this branch (we only do that when initially queueing the changes after
4679 : * decoding), because we will release the changes later, and that will
4680 : * update the accounting too (subtracting the size from the counters). And
4681 : * we don't want to underflow there.
4682 : */
4683 357254 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
4684 : ReorderBufferChangeSize(change));
4685 357254 : }
4686 :
4687 : /*
4688 : * Remove all on-disk stored for the passed in transaction.
4689 : */
4690 : static void
4691 528 : ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn)
4692 : {
4693 : XLogSegNo first;
4694 : XLogSegNo cur;
4695 : XLogSegNo last;
4696 :
4697 : Assert(txn->first_lsn != InvalidXLogRecPtr);
4698 : Assert(txn->final_lsn != InvalidXLogRecPtr);
4699 :
4700 528 : XLByteToSeg(txn->first_lsn, first, wal_segment_size);
4701 528 : XLByteToSeg(txn->final_lsn, last, wal_segment_size);
4702 :
4703 : /* iterate over all possible filenames, and delete them */
4704 1068 : for (cur = first; cur <= last; cur++)
4705 : {
4706 : char path[MAXPGPATH];
4707 :
4708 540 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid, cur);
4709 540 : if (unlink(path) != 0 && errno != ENOENT)
4710 0 : ereport(ERROR,
4711 : (errcode_for_file_access(),
4712 : errmsg("could not remove file \"%s\": %m", path)));
4713 : }
4714 528 : }
4715 :
4716 : /*
4717 : * Remove any leftover serialized reorder buffers from a slot directory after a
4718 : * prior crash or decoding session exit.
4719 : */
4720 : static void
4721 3840 : ReorderBufferCleanupSerializedTXNs(const char *slotname)
4722 : {
4723 : DIR *spill_dir;
4724 : struct dirent *spill_de;
4725 : struct stat statbuf;
4726 : char path[MAXPGPATH * 2 + sizeof(PG_REPLSLOT_DIR)];
4727 :
4728 3840 : sprintf(path, "%s/%s", PG_REPLSLOT_DIR, slotname);
4729 :
4730 : /* we're only handling directories here, skip if it's not ours */
4731 3840 : if (lstat(path, &statbuf) == 0 && !S_ISDIR(statbuf.st_mode))
4732 0 : return;
4733 :
4734 3840 : spill_dir = AllocateDir(path);
4735 15360 : while ((spill_de = ReadDirExtended(spill_dir, path, INFO)) != NULL)
4736 : {
4737 : /* only look at names that can be ours */
4738 11520 : if (strncmp(spill_de->d_name, "xid", 3) == 0)
4739 : {
4740 0 : snprintf(path, sizeof(path),
4741 : "%s/%s/%s", PG_REPLSLOT_DIR, slotname,
4742 0 : spill_de->d_name);
4743 :
4744 0 : if (unlink(path) != 0)
4745 0 : ereport(ERROR,
4746 : (errcode_for_file_access(),
4747 : errmsg("could not remove file \"%s\" during removal of %s/%s/xid*: %m",
4748 : path, PG_REPLSLOT_DIR, slotname)));
4749 : }
4750 : }
4751 3840 : FreeDir(spill_dir);
4752 : }
4753 :
4754 : /*
4755 : * Given a replication slot, transaction ID and segment number, fill in the
4756 : * corresponding spill file into 'path', which is a caller-owned buffer of size
4757 : * at least MAXPGPATH.
4758 : */
4759 : static void
4760 8076 : ReorderBufferSerializedPath(char *path, ReplicationSlot *slot, TransactionId xid,
4761 : XLogSegNo segno)
4762 : {
4763 : XLogRecPtr recptr;
4764 :
4765 8076 : XLogSegNoOffsetToRecPtr(segno, 0, wal_segment_size, recptr);
4766 :
4767 8076 : snprintf(path, MAXPGPATH, "%s/%s/xid-%u-lsn-%X-%X.spill",
4768 : PG_REPLSLOT_DIR,
4769 8076 : NameStr(MyReplicationSlot->data.name),
4770 8076 : xid, LSN_FORMAT_ARGS(recptr));
4771 8076 : }
4772 :
4773 : /*
4774 : * Delete all data spilled to disk after we've restarted/crashed. It will be
4775 : * recreated when the respective slots are reused.
4776 : */
4777 : void
4778 1670 : StartupReorderBuffer(void)
4779 : {
4780 : DIR *logical_dir;
4781 : struct dirent *logical_de;
4782 :
4783 1670 : logical_dir = AllocateDir(PG_REPLSLOT_DIR);
4784 5148 : while ((logical_de = ReadDir(logical_dir, PG_REPLSLOT_DIR)) != NULL)
4785 : {
4786 3478 : if (strcmp(logical_de->d_name, ".") == 0 ||
4787 1808 : strcmp(logical_de->d_name, "..") == 0)
4788 3340 : continue;
4789 :
4790 : /* if it cannot be a slot, skip the directory */
4791 138 : if (!ReplicationSlotValidateName(logical_de->d_name, DEBUG2))
4792 0 : continue;
4793 :
4794 : /*
4795 : * ok, has to be a surviving logical slot, iterate and delete
4796 : * everything starting with xid-*
4797 : */
4798 138 : ReorderBufferCleanupSerializedTXNs(logical_de->d_name);
4799 : }
4800 1670 : FreeDir(logical_dir);
4801 1670 : }
4802 :
4803 : /* ---------------------------------------
4804 : * toast reassembly support
4805 : * ---------------------------------------
4806 : */
4807 :
4808 : /*
4809 : * Initialize per tuple toast reconstruction support.
4810 : */
4811 : static void
4812 70 : ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
4813 : {
4814 : HASHCTL hash_ctl;
4815 :
4816 : Assert(txn->toast_hash == NULL);
4817 :
4818 70 : hash_ctl.keysize = sizeof(Oid);
4819 70 : hash_ctl.entrysize = sizeof(ReorderBufferToastEnt);
4820 70 : hash_ctl.hcxt = rb->context;
4821 70 : txn->toast_hash = hash_create("ReorderBufferToastHash", 5, &hash_ctl,
4822 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
4823 70 : }
4824 :
4825 : /*
4826 : * Per toast-chunk handling for toast reconstruction
4827 : *
4828 : * Appends a toast chunk so we can reconstruct it when the tuple "owning" the
4829 : * toasted Datum comes along.
4830 : */
4831 : static void
4832 3660 : ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
4833 : Relation relation, ReorderBufferChange *change)
4834 : {
4835 : ReorderBufferToastEnt *ent;
4836 : HeapTuple newtup;
4837 : bool found;
4838 : int32 chunksize;
4839 : bool isnull;
4840 : Pointer chunk;
4841 3660 : TupleDesc desc = RelationGetDescr(relation);
4842 : Oid chunk_id;
4843 : int32 chunk_seq;
4844 :
4845 3660 : if (txn->toast_hash == NULL)
4846 70 : ReorderBufferToastInitHash(rb, txn);
4847 :
4848 : Assert(IsToastRelation(relation));
4849 :
4850 3660 : newtup = change->data.tp.newtuple;
4851 3660 : chunk_id = DatumGetObjectId(fastgetattr(newtup, 1, desc, &isnull));
4852 : Assert(!isnull);
4853 3660 : chunk_seq = DatumGetInt32(fastgetattr(newtup, 2, desc, &isnull));
4854 : Assert(!isnull);
4855 :
4856 : ent = (ReorderBufferToastEnt *)
4857 3660 : hash_search(txn->toast_hash, &chunk_id, HASH_ENTER, &found);
4858 :
4859 3660 : if (!found)
4860 : {
4861 : Assert(ent->chunk_id == chunk_id);
4862 98 : ent->num_chunks = 0;
4863 98 : ent->last_chunk_seq = 0;
4864 98 : ent->size = 0;
4865 98 : ent->reconstructed = NULL;
4866 98 : dlist_init(&ent->chunks);
4867 :
4868 98 : if (chunk_seq != 0)
4869 0 : elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq 0",
4870 : chunk_seq, chunk_id);
4871 : }
4872 3562 : else if (found && chunk_seq != ent->last_chunk_seq + 1)
4873 0 : elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq %d",
4874 : chunk_seq, chunk_id, ent->last_chunk_seq + 1);
4875 :
4876 3660 : chunk = DatumGetPointer(fastgetattr(newtup, 3, desc, &isnull));
4877 : Assert(!isnull);
4878 :
4879 : /* calculate size so we can allocate the right size at once later */
4880 3660 : if (!VARATT_IS_EXTENDED(chunk))
4881 3660 : chunksize = VARSIZE(chunk) - VARHDRSZ;
4882 0 : else if (VARATT_IS_SHORT(chunk))
4883 : /* could happen due to heap_form_tuple doing its thing */
4884 0 : chunksize = VARSIZE_SHORT(chunk) - VARHDRSZ_SHORT;
4885 : else
4886 0 : elog(ERROR, "unexpected type of toast chunk");
4887 :
4888 3660 : ent->size += chunksize;
4889 3660 : ent->last_chunk_seq = chunk_seq;
4890 3660 : ent->num_chunks++;
4891 3660 : dlist_push_tail(&ent->chunks, &change->node);
4892 3660 : }
4893 :
4894 : /*
4895 : * Rejigger change->newtuple to point to in-memory toast tuples instead of
4896 : * on-disk toast tuples that may no longer exist (think DROP TABLE or VACUUM).
4897 : *
4898 : * We cannot replace unchanged toast tuples though, so those will still point
4899 : * to on-disk toast data.
4900 : *
4901 : * While updating the existing change with detoasted tuple data, we need to
4902 : * update the memory accounting info, because the change size will differ.
4903 : * Otherwise the accounting may get out of sync, triggering serialization
4904 : * at unexpected times.
4905 : *
4906 : * We simply subtract size of the change before rejiggering the tuple, and
4907 : * then add the new size. This makes it look like the change was removed
4908 : * and then added back, except it only tweaks the accounting info.
4909 : *
4910 : * In particular it can't trigger serialization, which would be pointless
4911 : * anyway as it happens during commit processing right before handing
4912 : * the change to the output plugin.
4913 : */
4914 : static void
4915 667972 : ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
4916 : Relation relation, ReorderBufferChange *change)
4917 : {
4918 : TupleDesc desc;
4919 : int natt;
4920 : Datum *attrs;
4921 : bool *isnull;
4922 : bool *free;
4923 : HeapTuple tmphtup;
4924 : Relation toast_rel;
4925 : TupleDesc toast_desc;
4926 : MemoryContext oldcontext;
4927 : HeapTuple newtup;
4928 : Size old_size;
4929 :
4930 : /* no toast tuples changed */
4931 667972 : if (txn->toast_hash == NULL)
4932 667480 : return;
4933 :
4934 : /*
4935 : * We're going to modify the size of the change. So, to make sure the
4936 : * accounting is correct we record the current change size and then after
4937 : * re-computing the change we'll subtract the recorded size and then
4938 : * re-add the new change size at the end. We don't immediately subtract
4939 : * the old size because if there is any error before we add the new size,
4940 : * we will release the changes and that will update the accounting info
4941 : * (subtracting the size from the counters). And we don't want to
4942 : * underflow there.
4943 : */
4944 492 : old_size = ReorderBufferChangeSize(change);
4945 :
4946 492 : oldcontext = MemoryContextSwitchTo(rb->context);
4947 :
4948 : /* we should only have toast tuples in an INSERT or UPDATE */
4949 : Assert(change->data.tp.newtuple);
4950 :
4951 492 : desc = RelationGetDescr(relation);
4952 :
4953 492 : toast_rel = RelationIdGetRelation(relation->rd_rel->reltoastrelid);
4954 492 : if (!RelationIsValid(toast_rel))
4955 0 : elog(ERROR, "could not open toast relation with OID %u (base relation \"%s\")",
4956 : relation->rd_rel->reltoastrelid, RelationGetRelationName(relation));
4957 :
4958 492 : toast_desc = RelationGetDescr(toast_rel);
4959 :
4960 : /* should we allocate from stack instead? */
4961 492 : attrs = palloc0(sizeof(Datum) * desc->natts);
4962 492 : isnull = palloc0(sizeof(bool) * desc->natts);
4963 492 : free = palloc0(sizeof(bool) * desc->natts);
4964 :
4965 492 : newtup = change->data.tp.newtuple;
4966 :
4967 492 : heap_deform_tuple(newtup, desc, attrs, isnull);
4968 :
4969 1514 : for (natt = 0; natt < desc->natts; natt++)
4970 : {
4971 1022 : Form_pg_attribute attr = TupleDescAttr(desc, natt);
4972 : ReorderBufferToastEnt *ent;
4973 : struct varlena *varlena;
4974 :
4975 : /* va_rawsize is the size of the original datum -- including header */
4976 : struct varatt_external toast_pointer;
4977 : struct varatt_indirect redirect_pointer;
4978 1022 : struct varlena *new_datum = NULL;
4979 : struct varlena *reconstructed;
4980 : dlist_iter it;
4981 1022 : Size data_done = 0;
4982 :
4983 : /* system columns aren't toasted */
4984 1022 : if (attr->attnum < 0)
4985 926 : continue;
4986 :
4987 1022 : if (attr->attisdropped)
4988 0 : continue;
4989 :
4990 : /* not a varlena datatype */
4991 1022 : if (attr->attlen != -1)
4992 482 : continue;
4993 :
4994 : /* no data */
4995 540 : if (isnull[natt])
4996 24 : continue;
4997 :
4998 : /* ok, we know we have a toast datum */
4999 516 : varlena = (struct varlena *) DatumGetPointer(attrs[natt]);
5000 :
5001 : /* no need to do anything if the tuple isn't external */
5002 516 : if (!VARATT_IS_EXTERNAL(varlena))
5003 404 : continue;
5004 :
5005 112 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, varlena);
5006 :
5007 : /*
5008 : * Check whether the toast tuple changed, replace if so.
5009 : */
5010 : ent = (ReorderBufferToastEnt *)
5011 112 : hash_search(txn->toast_hash,
5012 : &toast_pointer.va_valueid,
5013 : HASH_FIND,
5014 : NULL);
5015 112 : if (ent == NULL)
5016 16 : continue;
5017 :
5018 : new_datum =
5019 96 : (struct varlena *) palloc0(INDIRECT_POINTER_SIZE);
5020 :
5021 96 : free[natt] = true;
5022 :
5023 96 : reconstructed = palloc0(toast_pointer.va_rawsize);
5024 :
5025 96 : ent->reconstructed = reconstructed;
5026 :
5027 : /* stitch toast tuple back together from its parts */
5028 3654 : dlist_foreach(it, &ent->chunks)
5029 : {
5030 : bool cisnull;
5031 : ReorderBufferChange *cchange;
5032 : HeapTuple ctup;
5033 : Pointer chunk;
5034 :
5035 3558 : cchange = dlist_container(ReorderBufferChange, node, it.cur);
5036 3558 : ctup = cchange->data.tp.newtuple;
5037 3558 : chunk = DatumGetPointer(fastgetattr(ctup, 3, toast_desc, &cisnull));
5038 :
5039 : Assert(!cisnull);
5040 : Assert(!VARATT_IS_EXTERNAL(chunk));
5041 : Assert(!VARATT_IS_SHORT(chunk));
5042 :
5043 3558 : memcpy(VARDATA(reconstructed) + data_done,
5044 3558 : VARDATA(chunk),
5045 3558 : VARSIZE(chunk) - VARHDRSZ);
5046 3558 : data_done += VARSIZE(chunk) - VARHDRSZ;
5047 : }
5048 : Assert(data_done == VARATT_EXTERNAL_GET_EXTSIZE(toast_pointer));
5049 :
5050 : /* make sure its marked as compressed or not */
5051 96 : if (VARATT_EXTERNAL_IS_COMPRESSED(toast_pointer))
5052 10 : SET_VARSIZE_COMPRESSED(reconstructed, data_done + VARHDRSZ);
5053 : else
5054 86 : SET_VARSIZE(reconstructed, data_done + VARHDRSZ);
5055 :
5056 96 : memset(&redirect_pointer, 0, sizeof(redirect_pointer));
5057 96 : redirect_pointer.pointer = reconstructed;
5058 :
5059 96 : SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
5060 96 : memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
5061 : sizeof(redirect_pointer));
5062 :
5063 96 : attrs[natt] = PointerGetDatum(new_datum);
5064 : }
5065 :
5066 : /*
5067 : * Build tuple in separate memory & copy tuple back into the tuplebuf
5068 : * passed to the output plugin. We can't directly heap_fill_tuple() into
5069 : * the tuplebuf because attrs[] will point back into the current content.
5070 : */
5071 492 : tmphtup = heap_form_tuple(desc, attrs, isnull);
5072 : Assert(newtup->t_len <= MaxHeapTupleSize);
5073 : Assert(newtup->t_data == (HeapTupleHeader) ((char *) newtup + HEAPTUPLESIZE));
5074 :
5075 492 : memcpy(newtup->t_data, tmphtup->t_data, tmphtup->t_len);
5076 492 : newtup->t_len = tmphtup->t_len;
5077 :
5078 : /*
5079 : * free resources we won't further need, more persistent stuff will be
5080 : * free'd in ReorderBufferToastReset().
5081 : */
5082 492 : RelationClose(toast_rel);
5083 492 : pfree(tmphtup);
5084 1514 : for (natt = 0; natt < desc->natts; natt++)
5085 : {
5086 1022 : if (free[natt])
5087 96 : pfree(DatumGetPointer(attrs[natt]));
5088 : }
5089 492 : pfree(attrs);
5090 492 : pfree(free);
5091 492 : pfree(isnull);
5092 :
5093 492 : MemoryContextSwitchTo(oldcontext);
5094 :
5095 : /* subtract the old change size */
5096 492 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, false, old_size);
5097 : /* now add the change back, with the correct size */
5098 492 : ReorderBufferChangeMemoryUpdate(rb, change, NULL, true,
5099 : ReorderBufferChangeSize(change));
5100 : }
5101 :
5102 : /*
5103 : * Free all resources allocated for toast reconstruction.
5104 : */
5105 : static void
5106 674470 : ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn)
5107 : {
5108 : HASH_SEQ_STATUS hstat;
5109 : ReorderBufferToastEnt *ent;
5110 :
5111 674470 : if (txn->toast_hash == NULL)
5112 674400 : return;
5113 :
5114 : /* sequentially walk over the hash and free everything */
5115 70 : hash_seq_init(&hstat, txn->toast_hash);
5116 168 : while ((ent = (ReorderBufferToastEnt *) hash_seq_search(&hstat)) != NULL)
5117 : {
5118 : dlist_mutable_iter it;
5119 :
5120 98 : if (ent->reconstructed != NULL)
5121 96 : pfree(ent->reconstructed);
5122 :
5123 3758 : dlist_foreach_modify(it, &ent->chunks)
5124 : {
5125 3660 : ReorderBufferChange *change =
5126 3660 : dlist_container(ReorderBufferChange, node, it.cur);
5127 :
5128 3660 : dlist_delete(&change->node);
5129 3660 : ReorderBufferReturnChange(rb, change, true);
5130 : }
5131 : }
5132 :
5133 70 : hash_destroy(txn->toast_hash);
5134 70 : txn->toast_hash = NULL;
5135 : }
5136 :
5137 :
5138 : /* ---------------------------------------
5139 : * Visibility support for logical decoding
5140 : *
5141 : *
5142 : * Lookup actual cmin/cmax values when using decoding snapshot. We can't
5143 : * always rely on stored cmin/cmax values because of two scenarios:
5144 : *
5145 : * * A tuple got changed multiple times during a single transaction and thus
5146 : * has got a combo CID. Combo CIDs are only valid for the duration of a
5147 : * single transaction.
5148 : * * A tuple with a cmin but no cmax (and thus no combo CID) got
5149 : * deleted/updated in another transaction than the one which created it
5150 : * which we are looking at right now. As only one of cmin, cmax or combo CID
5151 : * is actually stored in the heap we don't have access to the value we
5152 : * need anymore.
5153 : *
5154 : * To resolve those problems we have a per-transaction hash of (cmin,
5155 : * cmax) tuples keyed by (relfilelocator, ctid) which contains the actual
5156 : * (cmin, cmax) values. That also takes care of combo CIDs by simply
5157 : * not caring about them at all. As we have the real cmin/cmax values
5158 : * combo CIDs aren't interesting.
5159 : *
5160 : * As we only care about catalog tuples here the overhead of this
5161 : * hashtable should be acceptable.
5162 : *
5163 : * Heap rewrites complicate this a bit, check rewriteheap.c for
5164 : * details.
5165 : * -------------------------------------------------------------------------
5166 : */
5167 :
5168 : /* struct for sorting mapping files by LSN efficiently */
5169 : typedef struct RewriteMappingFile
5170 : {
5171 : XLogRecPtr lsn;
5172 : char fname[MAXPGPATH];
5173 : } RewriteMappingFile;
5174 :
5175 : #ifdef NOT_USED
5176 : static void
5177 : DisplayMapping(HTAB *tuplecid_data)
5178 : {
5179 : HASH_SEQ_STATUS hstat;
5180 : ReorderBufferTupleCidEnt *ent;
5181 :
5182 : hash_seq_init(&hstat, tuplecid_data);
5183 : while ((ent = (ReorderBufferTupleCidEnt *) hash_seq_search(&hstat)) != NULL)
5184 : {
5185 : elog(DEBUG3, "mapping: node: %u/%u/%u tid: %u/%u cmin: %u, cmax: %u",
5186 : ent->key.rlocator.dbOid,
5187 : ent->key.rlocator.spcOid,
5188 : ent->key.rlocator.relNumber,
5189 : ItemPointerGetBlockNumber(&ent->key.tid),
5190 : ItemPointerGetOffsetNumber(&ent->key.tid),
5191 : ent->cmin,
5192 : ent->cmax
5193 : );
5194 : }
5195 : }
5196 : #endif
5197 :
5198 : /*
5199 : * Apply a single mapping file to tuplecid_data.
5200 : *
5201 : * The mapping file has to have been verified to be a) committed b) for our
5202 : * transaction c) applied in LSN order.
5203 : */
5204 : static void
5205 54 : ApplyLogicalMappingFile(HTAB *tuplecid_data, Oid relid, const char *fname)
5206 : {
5207 : char path[MAXPGPATH];
5208 : int fd;
5209 : int readBytes;
5210 : LogicalRewriteMappingData map;
5211 :
5212 54 : sprintf(path, "%s/%s", PG_LOGICAL_MAPPINGS_DIR, fname);
5213 54 : fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
5214 54 : if (fd < 0)
5215 0 : ereport(ERROR,
5216 : (errcode_for_file_access(),
5217 : errmsg("could not open file \"%s\": %m", path)));
5218 :
5219 : while (true)
5220 418 : {
5221 : ReorderBufferTupleCidKey key;
5222 : ReorderBufferTupleCidEnt *ent;
5223 : ReorderBufferTupleCidEnt *new_ent;
5224 : bool found;
5225 :
5226 : /* be careful about padding */
5227 472 : memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
5228 :
5229 : /* read all mappings till the end of the file */
5230 472 : pgstat_report_wait_start(WAIT_EVENT_REORDER_LOGICAL_MAPPING_READ);
5231 472 : readBytes = read(fd, &map, sizeof(LogicalRewriteMappingData));
5232 472 : pgstat_report_wait_end();
5233 :
5234 472 : if (readBytes < 0)
5235 0 : ereport(ERROR,
5236 : (errcode_for_file_access(),
5237 : errmsg("could not read file \"%s\": %m",
5238 : path)));
5239 472 : else if (readBytes == 0) /* EOF */
5240 54 : break;
5241 418 : else if (readBytes != sizeof(LogicalRewriteMappingData))
5242 0 : ereport(ERROR,
5243 : (errcode_for_file_access(),
5244 : errmsg("could not read from file \"%s\": read %d instead of %d bytes",
5245 : path, readBytes,
5246 : (int32) sizeof(LogicalRewriteMappingData))));
5247 :
5248 418 : key.rlocator = map.old_locator;
5249 418 : ItemPointerCopy(&map.old_tid,
5250 : &key.tid);
5251 :
5252 :
5253 : ent = (ReorderBufferTupleCidEnt *)
5254 418 : hash_search(tuplecid_data, &key, HASH_FIND, NULL);
5255 :
5256 : /* no existing mapping, no need to update */
5257 418 : if (!ent)
5258 0 : continue;
5259 :
5260 418 : key.rlocator = map.new_locator;
5261 418 : ItemPointerCopy(&map.new_tid,
5262 : &key.tid);
5263 :
5264 : new_ent = (ReorderBufferTupleCidEnt *)
5265 418 : hash_search(tuplecid_data, &key, HASH_ENTER, &found);
5266 :
5267 418 : if (found)
5268 : {
5269 : /*
5270 : * Make sure the existing mapping makes sense. We sometime update
5271 : * old records that did not yet have a cmax (e.g. pg_class' own
5272 : * entry while rewriting it) during rewrites, so allow that.
5273 : */
5274 : Assert(ent->cmin == InvalidCommandId || ent->cmin == new_ent->cmin);
5275 : Assert(ent->cmax == InvalidCommandId || ent->cmax == new_ent->cmax);
5276 : }
5277 : else
5278 : {
5279 : /* update mapping */
5280 406 : new_ent->cmin = ent->cmin;
5281 406 : new_ent->cmax = ent->cmax;
5282 406 : new_ent->combocid = ent->combocid;
5283 : }
5284 : }
5285 :
5286 54 : if (CloseTransientFile(fd) != 0)
5287 0 : ereport(ERROR,
5288 : (errcode_for_file_access(),
5289 : errmsg("could not close file \"%s\": %m", path)));
5290 54 : }
5291 :
5292 :
5293 : /*
5294 : * Check whether the TransactionId 'xid' is in the pre-sorted array 'xip'.
5295 : */
5296 : static bool
5297 696 : TransactionIdInArray(TransactionId xid, TransactionId *xip, Size num)
5298 : {
5299 696 : return bsearch(&xid, xip, num,
5300 696 : sizeof(TransactionId), xidComparator) != NULL;
5301 : }
5302 :
5303 : /*
5304 : * list_sort() comparator for sorting RewriteMappingFiles in LSN order.
5305 : */
5306 : static int
5307 84 : file_sort_by_lsn(const ListCell *a_p, const ListCell *b_p)
5308 : {
5309 84 : RewriteMappingFile *a = (RewriteMappingFile *) lfirst(a_p);
5310 84 : RewriteMappingFile *b = (RewriteMappingFile *) lfirst(b_p);
5311 :
5312 84 : return pg_cmp_u64(a->lsn, b->lsn);
5313 : }
5314 :
5315 : /*
5316 : * Apply any existing logical remapping files if there are any targeted at our
5317 : * transaction for relid.
5318 : */
5319 : static void
5320 12 : UpdateLogicalMappings(HTAB *tuplecid_data, Oid relid, Snapshot snapshot)
5321 : {
5322 : DIR *mapping_dir;
5323 : struct dirent *mapping_de;
5324 12 : List *files = NIL;
5325 : ListCell *file;
5326 12 : Oid dboid = IsSharedRelation(relid) ? InvalidOid : MyDatabaseId;
5327 :
5328 12 : mapping_dir = AllocateDir(PG_LOGICAL_MAPPINGS_DIR);
5329 1116 : while ((mapping_de = ReadDir(mapping_dir, PG_LOGICAL_MAPPINGS_DIR)) != NULL)
5330 : {
5331 : Oid f_dboid;
5332 : Oid f_relid;
5333 : TransactionId f_mapped_xid;
5334 : TransactionId f_create_xid;
5335 : XLogRecPtr f_lsn;
5336 : uint32 f_hi,
5337 : f_lo;
5338 : RewriteMappingFile *f;
5339 :
5340 1104 : if (strcmp(mapping_de->d_name, ".") == 0 ||
5341 1092 : strcmp(mapping_de->d_name, "..") == 0)
5342 1050 : continue;
5343 :
5344 : /* Ignore files that aren't ours */
5345 1080 : if (strncmp(mapping_de->d_name, "map-", 4) != 0)
5346 0 : continue;
5347 :
5348 1080 : if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
5349 : &f_dboid, &f_relid, &f_hi, &f_lo,
5350 : &f_mapped_xid, &f_create_xid) != 6)
5351 0 : elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
5352 :
5353 1080 : f_lsn = ((uint64) f_hi) << 32 | f_lo;
5354 :
5355 : /* mapping for another database */
5356 1080 : if (f_dboid != dboid)
5357 0 : continue;
5358 :
5359 : /* mapping for another relation */
5360 1080 : if (f_relid != relid)
5361 120 : continue;
5362 :
5363 : /* did the creating transaction abort? */
5364 960 : if (!TransactionIdDidCommit(f_create_xid))
5365 264 : continue;
5366 :
5367 : /* not for our transaction */
5368 696 : if (!TransactionIdInArray(f_mapped_xid, snapshot->subxip, snapshot->subxcnt))
5369 642 : continue;
5370 :
5371 : /* ok, relevant, queue for apply */
5372 54 : f = palloc(sizeof(RewriteMappingFile));
5373 54 : f->lsn = f_lsn;
5374 54 : strcpy(f->fname, mapping_de->d_name);
5375 54 : files = lappend(files, f);
5376 : }
5377 12 : FreeDir(mapping_dir);
5378 :
5379 : /* sort files so we apply them in LSN order */
5380 12 : list_sort(files, file_sort_by_lsn);
5381 :
5382 66 : foreach(file, files)
5383 : {
5384 54 : RewriteMappingFile *f = (RewriteMappingFile *) lfirst(file);
5385 :
5386 54 : elog(DEBUG1, "applying mapping: \"%s\" in %u", f->fname,
5387 : snapshot->subxip[0]);
5388 54 : ApplyLogicalMappingFile(tuplecid_data, relid, f->fname);
5389 54 : pfree(f);
5390 : }
5391 12 : }
5392 :
5393 : /*
5394 : * Lookup cmin/cmax of a tuple, during logical decoding where we can't rely on
5395 : * combo CIDs.
5396 : */
5397 : bool
5398 1438 : ResolveCminCmaxDuringDecoding(HTAB *tuplecid_data,
5399 : Snapshot snapshot,
5400 : HeapTuple htup, Buffer buffer,
5401 : CommandId *cmin, CommandId *cmax)
5402 : {
5403 : ReorderBufferTupleCidKey key;
5404 : ReorderBufferTupleCidEnt *ent;
5405 : ForkNumber forkno;
5406 : BlockNumber blockno;
5407 1438 : bool updated_mapping = false;
5408 :
5409 : /*
5410 : * Return unresolved if tuplecid_data is not valid. That's because when
5411 : * streaming in-progress transactions we may run into tuples with the CID
5412 : * before actually decoding them. Think e.g. about INSERT followed by
5413 : * TRUNCATE, where the TRUNCATE may not be decoded yet when applying the
5414 : * INSERT. So in such cases, we assume the CID is from the future
5415 : * command.
5416 : */
5417 1438 : if (tuplecid_data == NULL)
5418 22 : return false;
5419 :
5420 : /* be careful about padding */
5421 1416 : memset(&key, 0, sizeof(key));
5422 :
5423 : Assert(!BufferIsLocal(buffer));
5424 :
5425 : /*
5426 : * get relfilelocator from the buffer, no convenient way to access it
5427 : * other than that.
5428 : */
5429 1416 : BufferGetTag(buffer, &key.rlocator, &forkno, &blockno);
5430 :
5431 : /* tuples can only be in the main fork */
5432 : Assert(forkno == MAIN_FORKNUM);
5433 : Assert(blockno == ItemPointerGetBlockNumber(&htup->t_self));
5434 :
5435 1416 : ItemPointerCopy(&htup->t_self,
5436 : &key.tid);
5437 :
5438 1428 : restart:
5439 : ent = (ReorderBufferTupleCidEnt *)
5440 1428 : hash_search(tuplecid_data, &key, HASH_FIND, NULL);
5441 :
5442 : /*
5443 : * failed to find a mapping, check whether the table was rewritten and
5444 : * apply mapping if so, but only do that once - there can be no new
5445 : * mappings while we are in here since we have to hold a lock on the
5446 : * relation.
5447 : */
5448 1428 : if (ent == NULL && !updated_mapping)
5449 : {
5450 12 : UpdateLogicalMappings(tuplecid_data, htup->t_tableOid, snapshot);
5451 : /* now check but don't update for a mapping again */
5452 12 : updated_mapping = true;
5453 12 : goto restart;
5454 : }
5455 1416 : else if (ent == NULL)
5456 0 : return false;
5457 :
5458 1416 : if (cmin)
5459 1416 : *cmin = ent->cmin;
5460 1416 : if (cmax)
5461 1416 : *cmax = ent->cmax;
5462 1416 : return true;
5463 : }
|
