TLA Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * reorderbuffer.c
4 : * PostgreSQL logical replay/reorder buffer management
5 : *
6 : *
7 : * Copyright (c) 2012-2023, 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 a 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 still rely on max_changes_in_memory when loading serialized changes
71 : * back into memory. At that point we can't use the memory limit directly
72 : * as we load the subxacts independently. One option to deal with this
73 : * would be to count the subxacts, and allow each to allocate 1/N of the
74 : * memory limit. That however does not seem very appealing, because with
75 : * many subtransactions it may easily cause thrashing (short cycles of
76 : * deserializing and applying very few changes). We probably should give
77 : * a bit more memory to the oldest subtransactions, because it's likely
78 : * they are the source for the next sequence of changes.
79 : *
80 : * -------------------------------------------------------------------------
81 : */
82 : #include "postgres.h"
83 :
84 : #include <unistd.h>
85 : #include <sys/stat.h>
86 :
87 : #include "access/detoast.h"
88 : #include "access/heapam.h"
89 : #include "access/rewriteheap.h"
90 : #include "access/transam.h"
91 : #include "access/xact.h"
92 : #include "access/xlog_internal.h"
93 : #include "catalog/catalog.h"
94 : #include "lib/binaryheap.h"
95 : #include "miscadmin.h"
96 : #include "pgstat.h"
97 : #include "replication/logical.h"
98 : #include "replication/reorderbuffer.h"
99 : #include "replication/slot.h"
100 : #include "replication/snapbuild.h" /* just for SnapBuildSnapDecRefcount */
101 : #include "storage/bufmgr.h"
102 : #include "storage/fd.h"
103 : #include "storage/sinval.h"
104 : #include "utils/builtins.h"
105 : #include "utils/combocid.h"
106 : #include "utils/memdebug.h"
107 : #include "utils/memutils.h"
108 : #include "utils/rel.h"
109 : #include "utils/relfilenumbermap.h"
110 :
111 :
112 : /* entry for a hash table we use to map from xid to our transaction state */
113 : typedef struct ReorderBufferTXNByIdEnt
114 : {
115 : TransactionId xid;
116 : ReorderBufferTXN *txn;
117 : } ReorderBufferTXNByIdEnt;
118 :
119 : /* data structures for (relfilelocator, ctid) => (cmin, cmax) mapping */
120 : typedef struct ReorderBufferTupleCidKey
121 : {
122 : RelFileLocator rlocator;
123 : ItemPointerData tid;
124 : } ReorderBufferTupleCidKey;
125 :
126 : typedef struct ReorderBufferTupleCidEnt
127 : {
128 : ReorderBufferTupleCidKey key;
129 : CommandId cmin;
130 : CommandId cmax;
131 : CommandId combocid; /* just for debugging */
132 : } ReorderBufferTupleCidEnt;
133 :
134 : /* Virtual file descriptor with file offset tracking */
135 : typedef struct TXNEntryFile
136 : {
137 : File vfd; /* -1 when the file is closed */
138 : off_t curOffset; /* offset for next write or read. Reset to 0
139 : * when vfd is opened. */
140 : } TXNEntryFile;
141 :
142 : /* k-way in-order change iteration support structures */
143 : typedef struct ReorderBufferIterTXNEntry
144 : {
145 : XLogRecPtr lsn;
146 : ReorderBufferChange *change;
147 : ReorderBufferTXN *txn;
148 : TXNEntryFile file;
149 : XLogSegNo segno;
150 : } ReorderBufferIterTXNEntry;
151 :
152 : typedef struct ReorderBufferIterTXNState
153 : {
154 : binaryheap *heap;
155 : Size nr_txns;
156 : dlist_head old_change;
157 : ReorderBufferIterTXNEntry entries[FLEXIBLE_ARRAY_MEMBER];
158 : } ReorderBufferIterTXNState;
159 :
160 : /* toast datastructures */
161 : typedef struct ReorderBufferToastEnt
162 : {
163 : Oid chunk_id; /* toast_table.chunk_id */
164 : int32 last_chunk_seq; /* toast_table.chunk_seq of the last chunk we
165 : * have seen */
166 : Size num_chunks; /* number of chunks we've already seen */
167 : Size size; /* combined size of chunks seen */
168 : dlist_head chunks; /* linked list of chunks */
169 : struct varlena *reconstructed; /* reconstructed varlena now pointed to in
170 : * main tup */
171 : } ReorderBufferToastEnt;
172 :
173 : /* Disk serialization support datastructures */
174 : typedef struct ReorderBufferDiskChange
175 : {
176 : Size size;
177 : ReorderBufferChange change;
178 : /* data follows */
179 : } ReorderBufferDiskChange;
180 :
181 : #define IsSpecInsert(action) \
182 : ( \
183 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT) \
184 : )
185 : #define IsSpecConfirmOrAbort(action) \
186 : ( \
187 : (((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM) || \
188 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT)) \
189 : )
190 : #define IsInsertOrUpdate(action) \
191 : ( \
192 : (((action) == REORDER_BUFFER_CHANGE_INSERT) || \
193 : ((action) == REORDER_BUFFER_CHANGE_UPDATE) || \
194 : ((action) == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT)) \
195 : )
196 :
197 : /*
198 : * Maximum number of changes kept in memory, per transaction. After that,
199 : * changes are spooled to disk.
200 : *
201 : * The current value should be sufficient to decode the entire transaction
202 : * without hitting disk in OLTP workloads, while starting to spool to disk in
203 : * other workloads reasonably fast.
204 : *
205 : * At some point in the future it probably makes sense to have a more elaborate
206 : * resource management here, but it's not entirely clear what that would look
207 : * like.
208 : */
209 : int logical_decoding_work_mem;
210 : static const Size max_changes_in_memory = 4096; /* XXX for restore only */
211 :
212 : /* GUC variable */
213 : int logical_replication_mode = LOGICAL_REP_MODE_BUFFERED;
214 :
215 : /* ---------------------------------------
216 : * primary reorderbuffer support routines
217 : * ---------------------------------------
218 : */
219 : static ReorderBufferTXN *ReorderBufferGetTXN(ReorderBuffer *rb);
220 : static void ReorderBufferReturnTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
221 : static ReorderBufferTXN *ReorderBufferTXNByXid(ReorderBuffer *rb,
222 : TransactionId xid, bool create, bool *is_new,
223 : XLogRecPtr lsn, bool create_as_top);
224 : static void ReorderBufferTransferSnapToParent(ReorderBufferTXN *txn,
225 : ReorderBufferTXN *subtxn);
226 :
227 : static void AssertTXNLsnOrder(ReorderBuffer *rb);
228 :
229 : /* ---------------------------------------
230 : * support functions for lsn-order iterating over the ->changes of a
231 : * transaction and its subtransactions
232 : *
233 : * used for iteration over the k-way heap merge of a transaction and its
234 : * subtransactions
235 : * ---------------------------------------
236 : */
237 : static void ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn,
238 : ReorderBufferIterTXNState *volatile *iter_state);
239 : static ReorderBufferChange *ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state);
240 : static void ReorderBufferIterTXNFinish(ReorderBuffer *rb,
241 : ReorderBufferIterTXNState *state);
242 : static void ReorderBufferExecuteInvalidations(uint32 nmsgs, SharedInvalidationMessage *msgs);
243 :
244 : /*
245 : * ---------------------------------------
246 : * Disk serialization support functions
247 : * ---------------------------------------
248 : */
249 : static void ReorderBufferCheckMemoryLimit(ReorderBuffer *rb);
250 : static void ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
251 : static void ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
252 : int fd, ReorderBufferChange *change);
253 : static Size ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
254 : TXNEntryFile *file, XLogSegNo *segno);
255 : static void ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
256 : char *data);
257 : static void ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn);
258 : static void ReorderBufferTruncateTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
259 : bool txn_prepared);
260 : static void ReorderBufferCleanupSerializedTXNs(const char *slotname);
261 : static void ReorderBufferSerializedPath(char *path, ReplicationSlot *slot,
262 : TransactionId xid, XLogSegNo segno);
263 :
264 : static void ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap);
265 : static Snapshot ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
266 : ReorderBufferTXN *txn, CommandId cid);
267 :
268 : /*
269 : * ---------------------------------------
270 : * Streaming support functions
271 : * ---------------------------------------
272 : */
273 : static inline bool ReorderBufferCanStream(ReorderBuffer *rb);
274 : static inline bool ReorderBufferCanStartStreaming(ReorderBuffer *rb);
275 : static void ReorderBufferStreamTXN(ReorderBuffer *rb, ReorderBufferTXN *txn);
276 : static void ReorderBufferStreamCommit(ReorderBuffer *rb, ReorderBufferTXN *txn);
277 :
278 : /* ---------------------------------------
279 : * toast reassembly support
280 : * ---------------------------------------
281 : */
282 : static void ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn);
283 : static void ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn);
284 : static void ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
285 : Relation relation, ReorderBufferChange *change);
286 : static void ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
287 : Relation relation, ReorderBufferChange *change);
288 :
289 : /*
290 : * ---------------------------------------
291 : * memory accounting
292 : * ---------------------------------------
293 : */
294 : static Size ReorderBufferChangeSize(ReorderBufferChange *change);
295 : static void ReorderBufferChangeMemoryUpdate(ReorderBuffer *rb,
296 : ReorderBufferChange *change,
297 : bool addition, Size sz);
298 :
299 : /*
300 : * Allocate a new ReorderBuffer and clean out any old serialized state from
301 : * prior ReorderBuffer instances for the same slot.
302 : */
303 : ReorderBuffer *
304 GIC 821 : ReorderBufferAllocate(void)
305 : {
306 : ReorderBuffer *buffer;
307 ECB : HASHCTL hash_ctl;
308 : MemoryContext new_ctx;
309 :
310 GIC 821 : Assert(MyReplicationSlot != NULL);
311 :
312 : /* allocate memory in own context, to have better accountability */
313 CBC 821 : new_ctx = AllocSetContextCreate(CurrentMemoryContext,
314 : "ReorderBuffer",
315 : ALLOCSET_DEFAULT_SIZES);
316 ECB :
317 : buffer =
318 GIC 821 : (ReorderBuffer *) MemoryContextAlloc(new_ctx, sizeof(ReorderBuffer));
319 :
320 821 : memset(&hash_ctl, 0, sizeof(hash_ctl));
321 ECB :
322 GIC 821 : buffer->context = new_ctx;
323 ECB :
324 GIC 821 : buffer->change_context = SlabContextCreate(new_ctx,
325 ECB : "Change",
326 : SLAB_DEFAULT_BLOCK_SIZE,
327 : sizeof(ReorderBufferChange));
328 :
329 GIC 821 : buffer->txn_context = SlabContextCreate(new_ctx,
330 : "TXN",
331 : SLAB_DEFAULT_BLOCK_SIZE,
332 ECB : sizeof(ReorderBufferTXN));
333 :
334 : /*
335 : * XXX the allocation sizes used below pre-date generation context's block
336 : * growing code. These values should likely be benchmarked and set to
337 : * more suitable values.
338 : */
339 GIC 821 : buffer->tup_context = GenerationContextCreate(new_ctx,
340 : "Tuples",
341 : SLAB_LARGE_BLOCK_SIZE,
342 ECB : SLAB_LARGE_BLOCK_SIZE,
343 : SLAB_LARGE_BLOCK_SIZE);
344 :
345 GIC 821 : hash_ctl.keysize = sizeof(TransactionId);
346 821 : hash_ctl.entrysize = sizeof(ReorderBufferTXNByIdEnt);
347 821 : hash_ctl.hcxt = buffer->context;
348 ECB :
349 CBC 821 : buffer->by_txn = hash_create("ReorderBufferByXid", 1000, &hash_ctl,
350 ECB : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
351 :
352 CBC 821 : buffer->by_txn_last_xid = InvalidTransactionId;
353 GIC 821 : buffer->by_txn_last_txn = NULL;
354 :
355 CBC 821 : buffer->outbuf = NULL;
356 821 : buffer->outbufsize = 0;
357 GIC 821 : buffer->size = 0;
358 ECB :
359 CBC 821 : buffer->spillTxns = 0;
360 821 : buffer->spillCount = 0;
361 GIC 821 : buffer->spillBytes = 0;
362 CBC 821 : buffer->streamTxns = 0;
363 821 : buffer->streamCount = 0;
364 821 : buffer->streamBytes = 0;
365 821 : buffer->totalTxns = 0;
366 821 : buffer->totalBytes = 0;
367 ECB :
368 CBC 821 : buffer->current_restart_decoding_lsn = InvalidXLogRecPtr;
369 ECB :
370 GIC 821 : dlist_init(&buffer->toplevel_by_lsn);
371 CBC 821 : dlist_init(&buffer->txns_by_base_snapshot_lsn);
372 GNC 821 : dclist_init(&buffer->catchange_txns);
373 :
374 ECB : /*
375 : * Ensure there's no stale data from prior uses of this slot, in case some
376 : * prior exit avoided calling ReorderBufferFree. Failure to do this can
377 : * produce duplicated txns, and it's very cheap if there's nothing there.
378 : */
379 GIC 821 : ReorderBufferCleanupSerializedTXNs(NameStr(MyReplicationSlot->data.name));
380 :
381 821 : return buffer;
382 : }
383 ECB :
384 : /*
385 : * Free a ReorderBuffer
386 : */
387 : void
388 GIC 679 : ReorderBufferFree(ReorderBuffer *rb)
389 : {
390 679 : MemoryContext context = rb->context;
391 :
392 ECB : /*
393 : * We free separately allocated data by entirely scrapping reorderbuffer's
394 : * memory context.
395 : */
396 GIC 679 : MemoryContextDelete(context);
397 :
398 : /* Free disk space used by unconsumed reorder buffers */
399 679 : ReorderBufferCleanupSerializedTXNs(NameStr(MyReplicationSlot->data.name));
400 CBC 679 : }
401 :
402 : /*
403 ECB : * Get an unused, possibly preallocated, ReorderBufferTXN.
404 : */
405 : static ReorderBufferTXN *
406 GIC 3310 : ReorderBufferGetTXN(ReorderBuffer *rb)
407 : {
408 : ReorderBufferTXN *txn;
409 :
410 ECB : txn = (ReorderBufferTXN *)
411 GIC 3310 : MemoryContextAlloc(rb->txn_context, sizeof(ReorderBufferTXN));
412 :
413 3310 : memset(txn, 0, sizeof(ReorderBufferTXN));
414 :
415 CBC 3310 : dlist_init(&txn->changes);
416 GIC 3310 : dlist_init(&txn->tuplecids);
417 CBC 3310 : dlist_init(&txn->subtxns);
418 :
419 ECB : /* InvalidCommandId is not zero, so set it explicitly */
420 CBC 3310 : txn->command_id = InvalidCommandId;
421 3310 : txn->output_plugin_private = NULL;
422 :
423 GIC 3310 : return txn;
424 ECB : }
425 :
426 : /*
427 : * Free a ReorderBufferTXN.
428 : */
429 : static void
430 GIC 3262 : ReorderBufferReturnTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
431 : {
432 : /* clean the lookup cache if we were cached (quite likely) */
433 3262 : if (rb->by_txn_last_xid == txn->xid)
434 ECB : {
435 GIC 3073 : rb->by_txn_last_xid = InvalidTransactionId;
436 3073 : rb->by_txn_last_txn = NULL;
437 ECB : }
438 :
439 : /* free data that's contained */
440 :
441 GIC 3262 : if (txn->gid != NULL)
442 : {
443 38 : pfree(txn->gid);
444 38 : txn->gid = NULL;
445 ECB : }
446 :
447 CBC 3262 : if (txn->tuplecid_hash != NULL)
448 ECB : {
449 GIC 432 : hash_destroy(txn->tuplecid_hash);
450 432 : txn->tuplecid_hash = NULL;
451 ECB : }
452 :
453 CBC 3262 : if (txn->invalidations)
454 ECB : {
455 GIC 982 : pfree(txn->invalidations);
456 982 : txn->invalidations = NULL;
457 ECB : }
458 :
459 : /* Reset the toast hash */
460 CBC 3262 : ReorderBufferToastReset(rb, txn);
461 :
462 GIC 3262 : pfree(txn);
463 3262 : }
464 ECB :
465 : /*
466 : * Get a fresh ReorderBufferChange.
467 : */
468 : ReorderBufferChange *
469 GIC 1915007 : ReorderBufferGetChange(ReorderBuffer *rb)
470 : {
471 : ReorderBufferChange *change;
472 :
473 ECB : change = (ReorderBufferChange *)
474 GIC 1915007 : MemoryContextAlloc(rb->change_context, sizeof(ReorderBufferChange));
475 :
476 1915007 : memset(change, 0, sizeof(ReorderBufferChange));
477 1915007 : return change;
478 ECB : }
479 :
480 : /*
481 : * Free a ReorderBufferChange and update memory accounting, if requested.
482 : */
483 : void
484 GIC 1914746 : ReorderBufferReturnChange(ReorderBuffer *rb, ReorderBufferChange *change,
485 : bool upd_mem)
486 : {
487 : /* update memory accounting info */
488 CBC 1914746 : if (upd_mem)
489 GIC 1914678 : ReorderBufferChangeMemoryUpdate(rb, change, false,
490 : ReorderBufferChangeSize(change));
491 :
492 ECB : /* free contained data */
493 CBC 1914746 : switch (change->action)
494 : {
495 GIC 1843893 : case REORDER_BUFFER_CHANGE_INSERT:
496 : case REORDER_BUFFER_CHANGE_UPDATE:
497 ECB : case REORDER_BUFFER_CHANGE_DELETE:
498 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
499 CBC 1843893 : if (change->data.tp.newtuple)
500 : {
501 GIC 1564706 : ReorderBufferReturnTupleBuf(rb, change->data.tp.newtuple);
502 1564706 : change->data.tp.newtuple = NULL;
503 ECB : }
504 :
505 CBC 1843893 : if (change->data.tp.oldtuple)
506 ECB : {
507 GIC 210918 : ReorderBufferReturnTupleBuf(rb, change->data.tp.oldtuple);
508 210918 : change->data.tp.oldtuple = NULL;
509 ECB : }
510 GIC 1843893 : break;
511 CBC 39 : case REORDER_BUFFER_CHANGE_MESSAGE:
512 39 : if (change->data.msg.prefix != NULL)
513 GIC 39 : pfree(change->data.msg.prefix);
514 CBC 39 : change->data.msg.prefix = NULL;
515 39 : if (change->data.msg.message != NULL)
516 39 : pfree(change->data.msg.message);
517 39 : change->data.msg.message = NULL;
518 39 : break;
519 4567 : case REORDER_BUFFER_CHANGE_INVALIDATION:
520 4567 : if (change->data.inval.invalidations)
521 4567 : pfree(change->data.inval.invalidations);
522 4567 : change->data.inval.invalidations = NULL;
523 4567 : break;
524 998 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
525 998 : if (change->data.snapshot)
526 ECB : {
527 CBC 998 : ReorderBufferFreeSnap(rb, change->data.snapshot);
528 998 : change->data.snapshot = NULL;
529 ECB : }
530 GIC 998 : break;
531 ECB : /* no data in addition to the struct itself */
532 CBC 38 : case REORDER_BUFFER_CHANGE_TRUNCATE:
533 GIC 38 : if (change->data.truncate.relids != NULL)
534 ECB : {
535 GIC 38 : ReorderBufferReturnRelids(rb, change->data.truncate.relids);
536 CBC 38 : change->data.truncate.relids = NULL;
537 ECB : }
538 GIC 38 : break;
539 CBC 65211 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
540 ECB : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
541 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
542 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
543 CBC 65211 : break;
544 : }
545 :
546 GIC 1914746 : pfree(change);
547 CBC 1914746 : }
548 :
549 : /*
550 ECB : * Get a fresh ReorderBufferTupleBuf fitting at least a tuple of size
551 : * tuple_len (excluding header overhead).
552 : */
553 : ReorderBufferTupleBuf *
554 GIC 1775664 : ReorderBufferGetTupleBuf(ReorderBuffer *rb, Size tuple_len)
555 : {
556 : ReorderBufferTupleBuf *tuple;
557 : Size alloc_len;
558 ECB :
559 GIC 1775664 : alloc_len = tuple_len + SizeofHeapTupleHeader;
560 :
561 : tuple = (ReorderBufferTupleBuf *)
562 1775664 : MemoryContextAlloc(rb->tup_context,
563 ECB : sizeof(ReorderBufferTupleBuf) +
564 : MAXIMUM_ALIGNOF + alloc_len);
565 GIC 1775664 : tuple->alloc_tuple_size = alloc_len;
566 CBC 1775664 : tuple->tuple.t_data = ReorderBufferTupleBufData(tuple);
567 :
568 GIC 1775664 : return tuple;
569 ECB : }
570 :
571 : /*
572 : * Free a ReorderBufferTupleBuf.
573 : */
574 : void
575 GIC 1775624 : ReorderBufferReturnTupleBuf(ReorderBuffer *rb, ReorderBufferTupleBuf *tuple)
576 : {
577 1775624 : pfree(tuple);
578 1775624 : }
579 ECB :
580 : /*
581 : * Get an array for relids of truncated relations.
582 : *
583 : * We use the global memory context (for the whole reorder buffer), because
584 : * none of the existing ones seems like a good match (some are SLAB, so we
585 : * can't use those, and tup_context is meant for tuple data, not relids). We
586 : * could add yet another context, but it seems like an overkill - TRUNCATE is
587 : * not particularly common operation, so it does not seem worth it.
588 : */
589 : Oid *
590 GIC 42 : ReorderBufferGetRelids(ReorderBuffer *rb, int nrelids)
591 : {
592 : Oid *relids;
593 : Size alloc_len;
594 ECB :
595 GIC 42 : alloc_len = sizeof(Oid) * nrelids;
596 :
597 42 : relids = (Oid *) MemoryContextAlloc(rb->context, alloc_len);
598 :
599 CBC 42 : return relids;
600 : }
601 ECB :
602 : /*
603 : * Free an array of relids.
604 : */
605 : void
606 GIC 38 : ReorderBufferReturnRelids(ReorderBuffer *rb, Oid *relids)
607 : {
608 38 : pfree(relids);
609 38 : }
610 ECB :
611 : /*
612 : * Return the ReorderBufferTXN from the given buffer, specified by Xid.
613 : * If create is true, and a transaction doesn't already exist, create it
614 : * (with the given LSN, and as top transaction if that's specified);
615 : * when this happens, is_new is set to true.
616 : */
617 : static ReorderBufferTXN *
618 GIC 6433598 : ReorderBufferTXNByXid(ReorderBuffer *rb, TransactionId xid, bool create,
619 : bool *is_new, XLogRecPtr lsn, bool create_as_top)
620 : {
621 : ReorderBufferTXN *txn;
622 ECB : ReorderBufferTXNByIdEnt *ent;
623 : bool found;
624 :
625 GIC 6433598 : Assert(TransactionIdIsValid(xid));
626 :
627 : /*
628 : * Check the one-entry lookup cache first
629 ECB : */
630 GIC 6433598 : if (TransactionIdIsValid(rb->by_txn_last_xid) &&
631 6430494 : rb->by_txn_last_xid == xid)
632 : {
633 5440881 : txn = rb->by_txn_last_txn;
634 ECB :
635 CBC 5440881 : if (txn != NULL)
636 : {
637 ECB : /* found it, and it's valid */
638 GIC 5440869 : if (is_new)
639 CBC 2546 : *is_new = false;
640 GIC 5440869 : return txn;
641 : }
642 ECB :
643 : /*
644 : * cached as non-existent, and asked not to create? Then nothing else
645 : * to do.
646 : */
647 GIC 12 : if (!create)
648 9 : return NULL;
649 : /* otherwise fall through to create it */
650 : }
651 ECB :
652 : /*
653 : * If the cache wasn't hit or it yielded a "does-not-exist" and we want to
654 : * create an entry.
655 : */
656 :
657 : /* search the lookup table */
658 : ent = (ReorderBufferTXNByIdEnt *)
659 GIC 992720 : hash_search(rb->by_txn,
660 : &xid,
661 : create ? HASH_ENTER : HASH_FIND,
662 : &found);
663 CBC 992720 : if (found)
664 GIC 988127 : txn = ent->txn;
665 4593 : else if (create)
666 : {
667 ECB : /* initialize the new entry, if creation was requested */
668 CBC 3310 : Assert(ent != NULL);
669 3310 : Assert(lsn != InvalidXLogRecPtr);
670 :
671 GIC 3310 : ent->txn = ReorderBufferGetTXN(rb);
672 CBC 3310 : ent->txn->xid = xid;
673 3310 : txn = ent->txn;
674 GIC 3310 : txn->first_lsn = lsn;
675 CBC 3310 : txn->restart_decoding_lsn = rb->current_restart_decoding_lsn;
676 ECB :
677 CBC 3310 : if (create_as_top)
678 ECB : {
679 CBC 2633 : dlist_push_tail(&rb->toplevel_by_lsn, &txn->node);
680 GIC 2633 : AssertTXNLsnOrder(rb);
681 ECB : }
682 : }
683 : else
684 CBC 1283 : txn = NULL; /* not found and not asked to create */
685 :
686 : /* update cache */
687 GIC 992720 : rb->by_txn_last_xid = xid;
688 CBC 992720 : rb->by_txn_last_txn = txn;
689 :
690 GIC 992720 : if (is_new)
691 CBC 1787 : *is_new = !found;
692 ECB :
693 GIC 992720 : Assert(!create || txn != NULL);
694 CBC 992720 : return txn;
695 ECB : }
696 :
697 : /*
698 : * Record the partial change for the streaming of in-progress transactions. We
699 : * can stream only complete changes so if we have a partial change like toast
700 : * table insert or speculative insert then we mark such a 'txn' so that it
701 : * can't be streamed. We also ensure that if the changes in such a 'txn' can
702 : * be streamed and are above logical_decoding_work_mem threshold then we stream
703 : * them as soon as we have a complete change.
704 : */
705 : static void
706 GIC 1713655 : ReorderBufferProcessPartialChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
707 : ReorderBufferChange *change,
708 : bool toast_insert)
709 : {
710 ECB : ReorderBufferTXN *toptxn;
711 :
712 : /*
713 : * The partial changes need to be processed only while streaming
714 : * in-progress transactions.
715 : */
716 GIC 1713655 : if (!ReorderBufferCanStream(rb))
717 1226598 : return;
718 :
719 : /* Get the top transaction. */
720 GNC 487057 : toptxn = rbtxn_get_toptxn(txn);
721 ECB :
722 : /*
723 : * Indicate a partial change for toast inserts. The change will be
724 : * considered as complete once we get the insert or update on the main
725 : * table and we are sure that the pending toast chunks are not required
726 : * anymore.
727 : *
728 : * If we allow streaming when there are pending toast chunks then such
729 : * chunks won't be released till the insert (multi_insert) is complete and
730 : * we expect the txn to have streamed all changes after streaming. This
731 : * restriction is mainly to ensure the correctness of streamed
732 : * transactions and it doesn't seem worth uplifting such a restriction
733 : * just to allow this case because anyway we will stream the transaction
734 : * once such an insert is complete.
735 : */
736 GIC 487057 : if (toast_insert)
737 CBC 1451 : toptxn->txn_flags |= RBTXN_HAS_PARTIAL_CHANGE;
738 485606 : else if (rbtxn_has_partial_change(toptxn) &&
739 33 : IsInsertOrUpdate(change->action) &&
740 33 : change->data.tp.clear_toast_afterwards)
741 23 : toptxn->txn_flags &= ~RBTXN_HAS_PARTIAL_CHANGE;
742 ECB :
743 : /*
744 : * Indicate a partial change for speculative inserts. The change will be
745 : * considered as complete once we get the speculative confirm or abort
746 : * token.
747 : */
748 GIC 487057 : if (IsSpecInsert(change->action))
749 LBC 0 : toptxn->txn_flags |= RBTXN_HAS_PARTIAL_CHANGE;
750 GBC 487057 : else if (rbtxn_has_partial_change(toptxn) &&
751 CBC 1461 : IsSpecConfirmOrAbort(change->action))
752 LBC 0 : toptxn->txn_flags &= ~RBTXN_HAS_PARTIAL_CHANGE;
753 EUB :
754 : /*
755 : * Stream the transaction if it is serialized before and the changes are
756 : * now complete in the top-level transaction.
757 : *
758 : * The reason for doing the streaming of such a transaction as soon as we
759 : * get the complete change for it is that previously it would have reached
760 : * the memory threshold and wouldn't get streamed because of incomplete
761 : * changes. Delaying such transactions would increase apply lag for them.
762 : */
763 GIC 487057 : if (ReorderBufferCanStartStreaming(rb) &&
764 CBC 157832 : !(rbtxn_has_partial_change(toptxn)) &&
765 GNC 156411 : rbtxn_is_serialized(txn) &&
766 6 : rbtxn_has_streamable_change(toptxn))
767 CBC 6 : ReorderBufferStreamTXN(rb, toptxn);
768 ECB : }
769 :
770 : /*
771 : * Queue a change into a transaction so it can be replayed upon commit or will be
772 : * streamed when we reach logical_decoding_work_mem threshold.
773 : */
774 : void
775 GIC 1713723 : ReorderBufferQueueChange(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
776 : ReorderBufferChange *change, bool toast_insert)
777 ECB : {
778 : ReorderBufferTXN *txn;
779 :
780 GIC 1713723 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
781 :
782 ECB : /*
783 : * While streaming the previous changes we have detected that the
784 : * transaction is aborted. So there is no point in collecting further
785 : * changes for it.
786 : */
787 GIC 1713723 : if (txn->concurrent_abort)
788 : {
789 ECB : /*
790 : * We don't need to update memory accounting for this change as we
791 : * have not added it to the queue yet.
792 : */
793 GIC 68 : ReorderBufferReturnChange(rb, change, false);
794 68 : return;
795 ECB : }
796 :
797 : /*
798 : * The changes that are sent downstream are considered streamable. We
799 : * remember such transactions so that only those will later be considered
800 : * for streaming.
801 : */
802 GNC 1713655 : if (change->action == REORDER_BUFFER_CHANGE_INSERT ||
803 538571 : change->action == REORDER_BUFFER_CHANGE_UPDATE ||
804 331637 : change->action == REORDER_BUFFER_CHANGE_DELETE ||
805 64237 : change->action == REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT ||
806 46321 : change->action == REORDER_BUFFER_CHANGE_TRUNCATE ||
807 46282 : change->action == REORDER_BUFFER_CHANGE_MESSAGE)
808 : {
809 1667411 : ReorderBufferTXN *toptxn = rbtxn_get_toptxn(txn);
810 :
811 1667411 : toptxn->txn_flags |= RBTXN_HAS_STREAMABLE_CHANGE;
812 : }
813 :
814 GIC 1713655 : change->lsn = lsn;
815 1713655 : change->txn = txn;
816 :
817 1713655 : Assert(InvalidXLogRecPtr != lsn);
818 1713655 : dlist_push_tail(&txn->changes, &change->node);
819 1713655 : txn->nentries++;
820 1713655 : txn->nentries_mem++;
821 ECB :
822 : /* update memory accounting information */
823 CBC 1713655 : ReorderBufferChangeMemoryUpdate(rb, change, true,
824 ECB : ReorderBufferChangeSize(change));
825 :
826 : /* process partial change */
827 GIC 1713655 : ReorderBufferProcessPartialChange(rb, txn, change, toast_insert);
828 ECB :
829 : /* check the memory limits and evict something if needed */
830 CBC 1713655 : ReorderBufferCheckMemoryLimit(rb);
831 : }
832 :
833 ECB : /*
834 : * A transactional message is queued to be processed upon commit and a
835 : * non-transactional message gets processed immediately.
836 : */
837 : void
838 CBC 44 : ReorderBufferQueueMessage(ReorderBuffer *rb, TransactionId xid,
839 : Snapshot snap, XLogRecPtr lsn,
840 : bool transactional, const char *prefix,
841 : Size message_size, const char *message)
842 ECB : {
843 GIC 44 : if (transactional)
844 : {
845 : MemoryContext oldcontext;
846 ECB : ReorderBufferChange *change;
847 :
848 GIC 38 : Assert(xid != InvalidTransactionId);
849 ECB :
850 : /*
851 : * We don't expect snapshots for transactional changes - we'll use the
852 : * snapshot derived later during apply (unless the change gets
853 : * skipped).
854 : */
855 GNC 38 : Assert(!snap);
856 :
857 CBC 38 : oldcontext = MemoryContextSwitchTo(rb->context);
858 :
859 GIC 38 : change = ReorderBufferGetChange(rb);
860 38 : change->action = REORDER_BUFFER_CHANGE_MESSAGE;
861 38 : change->data.msg.prefix = pstrdup(prefix);
862 CBC 38 : change->data.msg.message_size = message_size;
863 GIC 38 : change->data.msg.message = palloc(message_size);
864 38 : memcpy(change->data.msg.message, message, message_size);
865 :
866 38 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
867 ECB :
868 GIC 38 : MemoryContextSwitchTo(oldcontext);
869 : }
870 : else
871 : {
872 6 : ReorderBufferTXN *txn = NULL;
873 GNC 6 : volatile Snapshot snapshot_now = snap;
874 ECB :
875 : /* Non-transactional changes require a valid snapshot. */
876 CBC 6 : Assert(snapshot_now);
877 :
878 6 : if (xid != InvalidTransactionId)
879 3 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
880 ECB :
881 : /* setup snapshot to allow catalog access */
882 CBC 6 : SetupHistoricSnapshot(snapshot_now, NULL);
883 6 : PG_TRY();
884 : {
885 6 : rb->message(rb, txn, lsn, false, prefix, message_size, message);
886 :
887 6 : TeardownHistoricSnapshot(false);
888 : }
889 UIC 0 : PG_CATCH();
890 : {
891 LBC 0 : TeardownHistoricSnapshot(true);
892 0 : PG_RE_THROW();
893 : }
894 GIC 6 : PG_END_TRY();
895 ECB : }
896 GIC 44 : }
897 ECB :
898 : /*
899 : * AssertTXNLsnOrder
900 : * Verify LSN ordering of transaction lists in the reorderbuffer
901 : *
902 : * Other LSN-related invariants are checked too.
903 : *
904 : * No-op if assertions are not in use.
905 : */
906 : static void
907 GIC 6446 : AssertTXNLsnOrder(ReorderBuffer *rb)
908 EUB : {
909 : #ifdef USE_ASSERT_CHECKING
910 GBC 6446 : LogicalDecodingContext *ctx = rb->private_data;
911 EUB : dlist_iter iter;
912 GIC 6446 : XLogRecPtr prev_first_lsn = InvalidXLogRecPtr;
913 CBC 6446 : XLogRecPtr prev_base_snap_lsn = InvalidXLogRecPtr;
914 :
915 ECB : /*
916 : * Skip the verification if we don't reach the LSN at which we start
917 : * decoding the contents of transactions yet because until we reach the
918 : * LSN, we could have transactions that don't have the association between
919 : * the top-level transaction and subtransaction yet and consequently have
920 : * the same LSN. We don't guarantee this association until we try to
921 : * decode the actual contents of transaction. The ordering of the records
922 : * prior to the start_decoding_at LSN should have been checked before the
923 : * restart.
924 : */
925 GIC 6446 : if (SnapBuildXactNeedsSkip(ctx->snapshot_builder, ctx->reader->EndRecPtr))
926 CBC 3428 : return;
927 :
928 GIC 5819 : dlist_foreach(iter, &rb->toplevel_by_lsn)
929 ECB : {
930 GIC 2801 : ReorderBufferTXN *cur_txn = dlist_container(ReorderBufferTXN, node,
931 ECB : iter.cur);
932 :
933 : /* start LSN must be set */
934 GIC 2801 : Assert(cur_txn->first_lsn != InvalidXLogRecPtr);
935 :
936 : /* If there is an end LSN, it must be higher than start LSN */
937 2801 : if (cur_txn->end_lsn != InvalidXLogRecPtr)
938 20 : Assert(cur_txn->first_lsn <= cur_txn->end_lsn);
939 :
940 : /* Current initial LSN must be strictly higher than previous */
941 2801 : if (prev_first_lsn != InvalidXLogRecPtr)
942 236 : Assert(prev_first_lsn < cur_txn->first_lsn);
943 :
944 ECB : /* known-as-subtxn txns must not be listed */
945 CBC 2801 : Assert(!rbtxn_is_known_subxact(cur_txn));
946 :
947 2801 : prev_first_lsn = cur_txn->first_lsn;
948 : }
949 ECB :
950 GIC 4601 : dlist_foreach(iter, &rb->txns_by_base_snapshot_lsn)
951 : {
952 1583 : ReorderBufferTXN *cur_txn = dlist_container(ReorderBufferTXN,
953 ECB : base_snapshot_node,
954 : iter.cur);
955 :
956 : /* base snapshot (and its LSN) must be set */
957 CBC 1583 : Assert(cur_txn->base_snapshot != NULL);
958 GIC 1583 : Assert(cur_txn->base_snapshot_lsn != InvalidXLogRecPtr);
959 :
960 ECB : /* current LSN must be strictly higher than previous */
961 CBC 1583 : if (prev_base_snap_lsn != InvalidXLogRecPtr)
962 GIC 170 : Assert(prev_base_snap_lsn < cur_txn->base_snapshot_lsn);
963 :
964 ECB : /* known-as-subtxn txns must not be listed */
965 GIC 1583 : Assert(!rbtxn_is_known_subxact(cur_txn));
966 ECB :
967 GIC 1583 : prev_base_snap_lsn = cur_txn->base_snapshot_lsn;
968 : }
969 ECB : #endif
970 : }
971 :
972 : /*
973 : * AssertChangeLsnOrder
974 : *
975 : * Check ordering of changes in the (sub)transaction.
976 : */
977 : static void
978 GIC 2162 : AssertChangeLsnOrder(ReorderBufferTXN *txn)
979 : {
980 ECB : #ifdef USE_ASSERT_CHECKING
981 : dlist_iter iter;
982 GIC 2162 : XLogRecPtr prev_lsn = txn->first_lsn;
983 :
984 CBC 181917 : dlist_foreach(iter, &txn->changes)
985 : {
986 ECB : ReorderBufferChange *cur_change;
987 :
988 GIC 179755 : cur_change = dlist_container(ReorderBufferChange, node, iter.cur);
989 :
990 179755 : Assert(txn->first_lsn != InvalidXLogRecPtr);
991 179755 : Assert(cur_change->lsn != InvalidXLogRecPtr);
992 179755 : Assert(txn->first_lsn <= cur_change->lsn);
993 :
994 179755 : if (txn->end_lsn != InvalidXLogRecPtr)
995 26019 : Assert(cur_change->lsn <= txn->end_lsn);
996 :
997 CBC 179755 : Assert(prev_lsn <= cur_change->lsn);
998 :
999 GIC 179755 : prev_lsn = cur_change->lsn;
1000 : }
1001 ECB : #endif
1002 GIC 2162 : }
1003 ECB :
1004 : /*
1005 : * ReorderBufferGetOldestTXN
1006 : * Return oldest transaction in reorderbuffer
1007 : */
1008 : ReorderBufferTXN *
1009 CBC 255 : ReorderBufferGetOldestTXN(ReorderBuffer *rb)
1010 ECB : {
1011 : ReorderBufferTXN *txn;
1012 :
1013 CBC 255 : AssertTXNLsnOrder(rb);
1014 ECB :
1015 GIC 255 : if (dlist_is_empty(&rb->toplevel_by_lsn))
1016 CBC 223 : return NULL;
1017 :
1018 32 : txn = dlist_head_element(ReorderBufferTXN, node, &rb->toplevel_by_lsn);
1019 :
1020 GIC 32 : Assert(!rbtxn_is_known_subxact(txn));
1021 CBC 32 : Assert(txn->first_lsn != InvalidXLogRecPtr);
1022 GIC 32 : return txn;
1023 : }
1024 :
1025 : /*
1026 : * ReorderBufferGetOldestXmin
1027 : * Return oldest Xmin in reorderbuffer
1028 ECB : *
1029 : * Returns oldest possibly running Xid from the point of view of snapshots
1030 : * used in the transactions kept by reorderbuffer, or InvalidTransactionId if
1031 : * there are none.
1032 : *
1033 : * Since snapshots are assigned monotonically, this equals the Xmin of the
1034 : * base snapshot with minimal base_snapshot_lsn.
1035 : */
1036 : TransactionId
1037 CBC 274 : ReorderBufferGetOldestXmin(ReorderBuffer *rb)
1038 : {
1039 ECB : ReorderBufferTXN *txn;
1040 :
1041 CBC 274 : AssertTXNLsnOrder(rb);
1042 :
1043 GIC 274 : if (dlist_is_empty(&rb->txns_by_base_snapshot_lsn))
1044 242 : return InvalidTransactionId;
1045 :
1046 32 : txn = dlist_head_element(ReorderBufferTXN, base_snapshot_node,
1047 : &rb->txns_by_base_snapshot_lsn);
1048 32 : return txn->base_snapshot->xmin;
1049 : }
1050 :
1051 : void
1052 279 : ReorderBufferSetRestartPoint(ReorderBuffer *rb, XLogRecPtr ptr)
1053 : {
1054 279 : rb->current_restart_decoding_lsn = ptr;
1055 279 : }
1056 ECB :
1057 : /*
1058 : * ReorderBufferAssignChild
1059 : *
1060 : * Make note that we know that subxid is a subtransaction of xid, seen as of
1061 : * the given lsn.
1062 : */
1063 : void
1064 GIC 863 : ReorderBufferAssignChild(ReorderBuffer *rb, TransactionId xid,
1065 ECB : TransactionId subxid, XLogRecPtr lsn)
1066 : {
1067 : ReorderBufferTXN *txn;
1068 : ReorderBufferTXN *subtxn;
1069 : bool new_top;
1070 : bool new_sub;
1071 :
1072 GIC 863 : txn = ReorderBufferTXNByXid(rb, xid, true, &new_top, lsn, true);
1073 CBC 863 : subtxn = ReorderBufferTXNByXid(rb, subxid, true, &new_sub, lsn, false);
1074 ECB :
1075 GIC 863 : if (!new_sub)
1076 : {
1077 186 : if (rbtxn_is_known_subxact(subtxn))
1078 : {
1079 : /* already associated, nothing to do */
1080 186 : return;
1081 : }
1082 : else
1083 ECB : {
1084 : /*
1085 : * We already saw this transaction, but initially added it to the
1086 : * list of top-level txns. Now that we know it's not top-level,
1087 : * remove it from there.
1088 : */
1089 UIC 0 : dlist_delete(&subtxn->node);
1090 : }
1091 ECB : }
1092 :
1093 GIC 677 : subtxn->txn_flags |= RBTXN_IS_SUBXACT;
1094 CBC 677 : subtxn->toplevel_xid = xid;
1095 GIC 677 : Assert(subtxn->nsubtxns == 0);
1096 ECB :
1097 : /* set the reference to top-level transaction */
1098 GIC 677 : subtxn->toptxn = txn;
1099 ECB :
1100 : /* add to subtransaction list */
1101 GIC 677 : dlist_push_tail(&txn->subtxns, &subtxn->node);
1102 677 : txn->nsubtxns++;
1103 :
1104 : /* Possibly transfer the subtxn's snapshot to its top-level txn. */
1105 677 : ReorderBufferTransferSnapToParent(txn, subtxn);
1106 :
1107 : /* Verify LSN-ordering invariant */
1108 GBC 677 : AssertTXNLsnOrder(rb);
1109 : }
1110 :
1111 : /*
1112 ECB : * ReorderBufferTransferSnapToParent
1113 : * Transfer base snapshot from subtxn to top-level txn, if needed
1114 : *
1115 : * This is done if the top-level txn doesn't have a base snapshot, or if the
1116 : * subtxn's base snapshot has an earlier LSN than the top-level txn's base
1117 : * snapshot's LSN. This can happen if there are no changes in the toplevel
1118 : * txn but there are some in the subtxn, or the first change in subtxn has
1119 : * earlier LSN than first change in the top-level txn and we learned about
1120 : * their kinship only now.
1121 : *
1122 : * The subtransaction's snapshot is cleared regardless of the transfer
1123 : * happening, since it's not needed anymore in either case.
1124 : *
1125 : * We do this as soon as we become aware of their kinship, to avoid queueing
1126 : * extra snapshots to txns known-as-subtxns -- only top-level txns will
1127 : * receive further snapshots.
1128 : */
1129 : static void
1130 GIC 681 : ReorderBufferTransferSnapToParent(ReorderBufferTXN *txn,
1131 : ReorderBufferTXN *subtxn)
1132 : {
1133 681 : Assert(subtxn->toplevel_xid == txn->xid);
1134 :
1135 681 : if (subtxn->base_snapshot != NULL)
1136 : {
1137 UIC 0 : if (txn->base_snapshot == NULL ||
1138 0 : subtxn->base_snapshot_lsn < txn->base_snapshot_lsn)
1139 : {
1140 : /*
1141 : * If the toplevel transaction already has a base snapshot but
1142 : * it's newer than the subxact's, purge it.
1143 : */
1144 0 : if (txn->base_snapshot != NULL)
1145 : {
1146 0 : SnapBuildSnapDecRefcount(txn->base_snapshot);
1147 0 : dlist_delete(&txn->base_snapshot_node);
1148 : }
1149 ECB :
1150 : /*
1151 : * The snapshot is now the top transaction's; transfer it, and
1152 : * adjust the list position of the top transaction in the list by
1153 : * moving it to where the subtransaction is.
1154 : */
1155 UIC 0 : txn->base_snapshot = subtxn->base_snapshot;
1156 UBC 0 : txn->base_snapshot_lsn = subtxn->base_snapshot_lsn;
1157 0 : dlist_insert_before(&subtxn->base_snapshot_node,
1158 : &txn->base_snapshot_node);
1159 :
1160 : /*
1161 : * The subtransaction doesn't have a snapshot anymore (so it
1162 : * mustn't be in the list.)
1163 EUB : */
1164 UIC 0 : subtxn->base_snapshot = NULL;
1165 UBC 0 : subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
1166 0 : dlist_delete(&subtxn->base_snapshot_node);
1167 : }
1168 : else
1169 : {
1170 : /* Base snap of toplevel is fine, so subxact's is not needed */
1171 UIC 0 : SnapBuildSnapDecRefcount(subtxn->base_snapshot);
1172 0 : dlist_delete(&subtxn->base_snapshot_node);
1173 0 : subtxn->base_snapshot = NULL;
1174 UBC 0 : subtxn->base_snapshot_lsn = InvalidXLogRecPtr;
1175 EUB : }
1176 : }
1177 GIC 681 : }
1178 :
1179 : /*
1180 : * Associate a subtransaction with its toplevel transaction at commit
1181 : * time. There may be no further changes added after this.
1182 : */
1183 EUB : void
1184 GBC 267 : ReorderBufferCommitChild(ReorderBuffer *rb, TransactionId xid,
1185 EUB : TransactionId subxid, XLogRecPtr commit_lsn,
1186 : XLogRecPtr end_lsn)
1187 : {
1188 : ReorderBufferTXN *subtxn;
1189 :
1190 GBC 267 : subtxn = ReorderBufferTXNByXid(rb, subxid, false, NULL,
1191 EUB : InvalidXLogRecPtr, false);
1192 :
1193 : /*
1194 : * No need to do anything if that subtxn didn't contain any changes
1195 : */
1196 CBC 267 : if (!subtxn)
1197 GIC 81 : return;
1198 :
1199 186 : subtxn->final_lsn = commit_lsn;
1200 186 : subtxn->end_lsn = end_lsn;
1201 :
1202 : /*
1203 ECB : * Assign this subxact as a child of the toplevel xact (no-op if already
1204 : * done.)
1205 : */
1206 GIC 186 : ReorderBufferAssignChild(rb, xid, subxid, InvalidXLogRecPtr);
1207 : }
1208 :
1209 ECB :
1210 : /*
1211 : * Support for efficiently iterating over a transaction's and its
1212 : * subtransactions' changes.
1213 : *
1214 : * We do by doing a k-way merge between transactions/subtransactions. For that
1215 : * we model the current heads of the different transactions as a binary heap
1216 : * so we easily know which (sub-)transaction has the change with the smallest
1217 : * lsn next.
1218 : *
1219 : * We assume the changes in individual transactions are already sorted by LSN.
1220 : */
1221 :
1222 : /*
1223 : * Binary heap comparison function.
1224 : */
1225 : static int
1226 GIC 52311 : ReorderBufferIterCompare(Datum a, Datum b, void *arg)
1227 : {
1228 52311 : ReorderBufferIterTXNState *state = (ReorderBufferIterTXNState *) arg;
1229 52311 : XLogRecPtr pos_a = state->entries[DatumGetInt32(a)].lsn;
1230 52311 : XLogRecPtr pos_b = state->entries[DatumGetInt32(b)].lsn;
1231 :
1232 52311 : if (pos_a < pos_b)
1233 50985 : return 1;
1234 1326 : else if (pos_a == pos_b)
1235 UIC 0 : return 0;
1236 GIC 1326 : return -1;
1237 : }
1238 :
1239 : /*
1240 : * Allocate & initialize an iterator which iterates in lsn order over a
1241 : * transaction and all its subtransactions.
1242 : *
1243 : * Note: The iterator state is returned through iter_state parameter rather
1244 : * than the function's return value. This is because the state gets cleaned up
1245 ECB : * in a PG_CATCH block in the caller, so we want to make sure the caller gets
1246 : * back the state even if this function throws an exception.
1247 : */
1248 : static void
1249 CBC 1700 : ReorderBufferIterTXNInit(ReorderBuffer *rb, ReorderBufferTXN *txn,
1250 : ReorderBufferIterTXNState *volatile *iter_state)
1251 ECB : {
1252 CBC 1700 : Size nr_txns = 0;
1253 ECB : ReorderBufferIterTXNState *state;
1254 EUB : dlist_iter cur_txn_i;
1255 ECB : int32 off;
1256 :
1257 GIC 1700 : *iter_state = NULL;
1258 :
1259 : /* Check ordering of changes in the toplevel transaction. */
1260 1700 : AssertChangeLsnOrder(txn);
1261 :
1262 : /*
1263 : * Calculate the size of our heap: one element for every transaction that
1264 : * contains changes. (Besides the transactions already in the reorder
1265 : * buffer, we count the one we were directly passed.)
1266 : */
1267 1700 : if (txn->nentries > 0)
1268 CBC 1524 : nr_txns++;
1269 :
1270 GIC 2162 : dlist_foreach(cur_txn_i, &txn->subtxns)
1271 ECB : {
1272 : ReorderBufferTXN *cur_txn;
1273 :
1274 GIC 462 : cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
1275 :
1276 ECB : /* Check ordering of changes in this subtransaction. */
1277 GIC 462 : AssertChangeLsnOrder(cur_txn);
1278 :
1279 CBC 462 : if (cur_txn->nentries > 0)
1280 GIC 300 : nr_txns++;
1281 : }
1282 :
1283 : /* allocate iteration state */
1284 : state = (ReorderBufferIterTXNState *)
1285 1700 : MemoryContextAllocZero(rb->context,
1286 ECB : sizeof(ReorderBufferIterTXNState) +
1287 CBC 1700 : sizeof(ReorderBufferIterTXNEntry) * nr_txns);
1288 :
1289 1700 : state->nr_txns = nr_txns;
1290 GIC 1700 : dlist_init(&state->old_change);
1291 :
1292 3524 : for (off = 0; off < state->nr_txns; off++)
1293 ECB : {
1294 GIC 1824 : state->entries[off].file.vfd = -1;
1295 1824 : state->entries[off].segno = 0;
1296 ECB : }
1297 :
1298 : /* allocate heap */
1299 CBC 1700 : state->heap = binaryheap_allocate(state->nr_txns,
1300 : ReorderBufferIterCompare,
1301 : state);
1302 :
1303 : /* Now that the state fields are initialized, it is safe to return it. */
1304 1700 : *iter_state = state;
1305 :
1306 ECB : /*
1307 : * Now insert items into the binary heap, in an unordered fashion. (We
1308 : * will run a heap assembly step at the end; this is more efficient.)
1309 : */
1310 :
1311 CBC 1700 : off = 0;
1312 :
1313 ECB : /* add toplevel transaction if it contains changes */
1314 CBC 1700 : if (txn->nentries > 0)
1315 : {
1316 : ReorderBufferChange *cur_change;
1317 :
1318 1524 : if (rbtxn_is_serialized(txn))
1319 : {
1320 : /* serialize remaining changes */
1321 GIC 21 : ReorderBufferSerializeTXN(rb, txn);
1322 21 : ReorderBufferRestoreChanges(rb, txn, &state->entries[off].file,
1323 ECB : &state->entries[off].segno);
1324 : }
1325 :
1326 GIC 1524 : cur_change = dlist_head_element(ReorderBufferChange, node,
1327 : &txn->changes);
1328 :
1329 1524 : state->entries[off].lsn = cur_change->lsn;
1330 CBC 1524 : state->entries[off].change = cur_change;
1331 GIC 1524 : state->entries[off].txn = txn;
1332 :
1333 CBC 1524 : binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
1334 : }
1335 :
1336 : /* add subtransactions if they contain changes */
1337 2162 : dlist_foreach(cur_txn_i, &txn->subtxns)
1338 : {
1339 : ReorderBufferTXN *cur_txn;
1340 ECB :
1341 CBC 462 : cur_txn = dlist_container(ReorderBufferTXN, node, cur_txn_i.cur);
1342 :
1343 GIC 462 : if (cur_txn->nentries > 0)
1344 : {
1345 ECB : ReorderBufferChange *cur_change;
1346 :
1347 GIC 300 : if (rbtxn_is_serialized(cur_txn))
1348 ECB : {
1349 : /* serialize remaining changes */
1350 CBC 16 : ReorderBufferSerializeTXN(rb, cur_txn);
1351 GIC 16 : ReorderBufferRestoreChanges(rb, cur_txn,
1352 ECB : &state->entries[off].file,
1353 : &state->entries[off].segno);
1354 : }
1355 GIC 300 : cur_change = dlist_head_element(ReorderBufferChange, node,
1356 ECB : &cur_txn->changes);
1357 :
1358 GIC 300 : state->entries[off].lsn = cur_change->lsn;
1359 300 : state->entries[off].change = cur_change;
1360 CBC 300 : state->entries[off].txn = cur_txn;
1361 :
1362 300 : binaryheap_add_unordered(state->heap, Int32GetDatum(off++));
1363 : }
1364 : }
1365 :
1366 ECB : /* assemble a valid binary heap */
1367 GIC 1700 : binaryheap_build(state->heap);
1368 1700 : }
1369 ECB :
1370 : /*
1371 : * Return the next change when iterating over a transaction and its
1372 : * subtransactions.
1373 : *
1374 : * Returns NULL when no further changes exist.
1375 : */
1376 : static ReorderBufferChange *
1377 CBC 354474 : ReorderBufferIterTXNNext(ReorderBuffer *rb, ReorderBufferIterTXNState *state)
1378 ECB : {
1379 : ReorderBufferChange *change;
1380 : ReorderBufferIterTXNEntry *entry;
1381 : int32 off;
1382 :
1383 : /* nothing there anymore */
1384 GIC 354474 : if (state->heap->bh_size == 0)
1385 1689 : return NULL;
1386 ECB :
1387 CBC 352785 : off = DatumGetInt32(binaryheap_first(state->heap));
1388 GIC 352785 : entry = &state->entries[off];
1389 :
1390 : /* free memory we might have "leaked" in the previous *Next call */
1391 352785 : if (!dlist_is_empty(&state->old_change))
1392 : {
1393 45 : change = dlist_container(ReorderBufferChange, node,
1394 : dlist_pop_head_node(&state->old_change));
1395 45 : ReorderBufferReturnChange(rb, change, true);
1396 CBC 45 : Assert(dlist_is_empty(&state->old_change));
1397 : }
1398 :
1399 GIC 352785 : change = entry->change;
1400 :
1401 : /*
1402 : * update heap with information about which transaction has the next
1403 ECB : * relevant change in LSN order
1404 : */
1405 :
1406 : /* there are in-memory changes */
1407 CBC 352785 : if (dlist_has_next(&entry->txn->changes, &entry->change->node))
1408 : {
1409 GIC 350928 : dlist_node *next = dlist_next_node(&entry->txn->changes, &change->node);
1410 CBC 350928 : ReorderBufferChange *next_change =
1411 GIC 350928 : dlist_container(ReorderBufferChange, node, next);
1412 ECB :
1413 : /* txn stays the same */
1414 CBC 350928 : state->entries[off].lsn = next_change->lsn;
1415 350928 : state->entries[off].change = next_change;
1416 :
1417 GIC 350928 : binaryheap_replace_first(state->heap, Int32GetDatum(off));
1418 CBC 350928 : return change;
1419 : }
1420 :
1421 : /* try to load changes from disk */
1422 GIC 1857 : if (entry->txn->nentries != entry->txn->nentries_mem)
1423 : {
1424 : /*
1425 : * Ugly: restoring changes will reuse *Change records, thus delete the
1426 ECB : * current one from the per-tx list and only free in the next call.
1427 : */
1428 CBC 65 : dlist_delete(&change->node);
1429 65 : dlist_push_tail(&state->old_change, &change->node);
1430 ECB :
1431 : /*
1432 : * Update the total bytes processed by the txn for which we are
1433 : * releasing the current set of changes and restoring the new set of
1434 : * changes.
1435 : */
1436 CBC 65 : rb->totalBytes += entry->txn->size;
1437 65 : if (ReorderBufferRestoreChanges(rb, entry->txn, &entry->file,
1438 : &state->entries[off].segno))
1439 : {
1440 : /* successfully restored changes from disk */
1441 ECB : ReorderBufferChange *next_change =
1442 GIC 36 : dlist_head_element(ReorderBufferChange, node,
1443 : &entry->txn->changes);
1444 :
1445 36 : elog(DEBUG2, "restored %u/%u changes from disk",
1446 : (uint32) entry->txn->nentries_mem,
1447 ECB : (uint32) entry->txn->nentries);
1448 :
1449 GIC 36 : Assert(entry->txn->nentries_mem);
1450 : /* txn stays the same */
1451 36 : state->entries[off].lsn = next_change->lsn;
1452 36 : state->entries[off].change = next_change;
1453 36 : binaryheap_replace_first(state->heap, Int32GetDatum(off));
1454 :
1455 CBC 36 : return change;
1456 ECB : }
1457 : }
1458 :
1459 : /* ok, no changes there anymore, remove */
1460 GIC 1821 : binaryheap_remove_first(state->heap);
1461 ECB :
1462 GIC 1821 : return change;
1463 : }
1464 ECB :
1465 : /*
1466 : * Deallocate the iterator
1467 : */
1468 : static void
1469 GIC 1699 : ReorderBufferIterTXNFinish(ReorderBuffer *rb,
1470 ECB : ReorderBufferIterTXNState *state)
1471 : {
1472 : int32 off;
1473 :
1474 CBC 3522 : for (off = 0; off < state->nr_txns; off++)
1475 : {
1476 GIC 1823 : if (state->entries[off].file.vfd != -1)
1477 UIC 0 : FileClose(state->entries[off].file.vfd);
1478 : }
1479 ECB :
1480 : /* free memory we might have "leaked" in the last *Next call */
1481 CBC 1699 : if (!dlist_is_empty(&state->old_change))
1482 : {
1483 : ReorderBufferChange *change;
1484 :
1485 GIC 19 : change = dlist_container(ReorderBufferChange, node,
1486 : dlist_pop_head_node(&state->old_change));
1487 19 : ReorderBufferReturnChange(rb, change, true);
1488 CBC 19 : Assert(dlist_is_empty(&state->old_change));
1489 : }
1490 :
1491 GIC 1699 : binaryheap_free(state->heap);
1492 1699 : pfree(state);
1493 CBC 1699 : }
1494 :
1495 ECB : /*
1496 EUB : * Cleanup the contents of a transaction, usually after the transaction
1497 : * committed or aborted.
1498 : */
1499 : static void
1500 CBC 3262 : ReorderBufferCleanupTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
1501 : {
1502 : bool found;
1503 : dlist_mutable_iter iter;
1504 ECB :
1505 : /* cleanup subtransactions & their changes */
1506 CBC 3447 : dlist_foreach_modify(iter, &txn->subtxns)
1507 ECB : {
1508 : ReorderBufferTXN *subtxn;
1509 :
1510 CBC 185 : subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1511 ECB :
1512 : /*
1513 : * Subtransactions are always associated to the toplevel TXN, even if
1514 : * they originally were happening inside another subtxn, so we won't
1515 : * ever recurse more than one level deep here.
1516 : */
1517 GIC 185 : Assert(rbtxn_is_known_subxact(subtxn));
1518 185 : Assert(subtxn->nsubtxns == 0);
1519 ECB :
1520 GIC 185 : ReorderBufferCleanupTXN(rb, subtxn);
1521 : }
1522 :
1523 : /* cleanup changes in the txn */
1524 75313 : dlist_foreach_modify(iter, &txn->changes)
1525 ECB : {
1526 : ReorderBufferChange *change;
1527 :
1528 GIC 72051 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1529 ECB :
1530 : /* Check we're not mixing changes from different transactions. */
1531 GIC 72051 : Assert(change->txn == txn);
1532 :
1533 72051 : ReorderBufferReturnChange(rb, change, true);
1534 : }
1535 :
1536 ECB : /*
1537 : * Cleanup the tuplecids we stored for decoding catalog snapshot access.
1538 : * They are always stored in the toplevel transaction.
1539 : */
1540 GIC 25831 : dlist_foreach_modify(iter, &txn->tuplecids)
1541 : {
1542 : ReorderBufferChange *change;
1543 ECB :
1544 GIC 22569 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1545 :
1546 : /* Check we're not mixing changes from different transactions. */
1547 CBC 22569 : Assert(change->txn == txn);
1548 GIC 22569 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1549 :
1550 CBC 22569 : ReorderBufferReturnChange(rb, change, true);
1551 : }
1552 ECB :
1553 : /*
1554 : * Cleanup the base snapshot, if set.
1555 : */
1556 GIC 3262 : if (txn->base_snapshot != NULL)
1557 : {
1558 2568 : SnapBuildSnapDecRefcount(txn->base_snapshot);
1559 CBC 2568 : dlist_delete(&txn->base_snapshot_node);
1560 : }
1561 :
1562 : /*
1563 ECB : * Cleanup the snapshot for the last streamed run.
1564 : */
1565 GIC 3262 : if (txn->snapshot_now != NULL)
1566 ECB : {
1567 CBC 62 : Assert(rbtxn_is_streamed(txn));
1568 GIC 62 : ReorderBufferFreeSnap(rb, txn->snapshot_now);
1569 ECB : }
1570 :
1571 : /*
1572 : * Remove TXN from its containing lists.
1573 : *
1574 : * Note: if txn is known as subxact, we are deleting the TXN from its
1575 : * parent's list of known subxacts; this leaves the parent's nsubxacts
1576 : * count too high, but we don't care. Otherwise, we are deleting the TXN
1577 : * from the LSN-ordered list of toplevel TXNs. We remove the TXN from the
1578 : * list of catalog modifying transactions as well.
1579 : */
1580 GIC 3262 : dlist_delete(&txn->node);
1581 GNC 3262 : if (rbtxn_has_catalog_changes(txn))
1582 1023 : dclist_delete_from(&rb->catchange_txns, &txn->catchange_node);
1583 :
1584 : /* now remove reference from buffer */
1585 3262 : hash_search(rb->by_txn, &txn->xid, HASH_REMOVE, &found);
1586 CBC 3262 : Assert(found);
1587 ECB :
1588 : /* remove entries spilled to disk */
1589 GIC 3262 : if (rbtxn_is_serialized(txn))
1590 292 : ReorderBufferRestoreCleanup(rb, txn);
1591 :
1592 : /* deallocate */
1593 3262 : ReorderBufferReturnTXN(rb, txn);
1594 3262 : }
1595 :
1596 : /*
1597 : * Discard changes from a transaction (and subtransactions), either after
1598 : * streaming or decoding them at PREPARE. Keep the remaining info -
1599 ECB : * transactions, tuplecids, invalidations and snapshots.
1600 : *
1601 : * We additionally remove tuplecids after decoding the transaction at prepare
1602 : * time as we only need to perform invalidation at rollback or commit prepared.
1603 : *
1604 : * 'txn_prepared' indicates that we have decoded the transaction at prepare
1605 : * time.
1606 : */
1607 : static void
1608 CBC 993 : ReorderBufferTruncateTXN(ReorderBuffer *rb, ReorderBufferTXN *txn, bool txn_prepared)
1609 ECB : {
1610 : dlist_mutable_iter iter;
1611 :
1612 : /* cleanup subtransactions & their changes */
1613 CBC 1289 : dlist_foreach_modify(iter, &txn->subtxns)
1614 : {
1615 : ReorderBufferTXN *subtxn;
1616 :
1617 GIC 296 : subtxn = dlist_container(ReorderBufferTXN, node, iter.cur);
1618 :
1619 : /*
1620 : * Subtransactions are always associated to the toplevel TXN, even if
1621 : * they originally were happening inside another subtxn, so we won't
1622 : * ever recurse more than one level deep here.
1623 : */
1624 296 : Assert(rbtxn_is_known_subxact(subtxn));
1625 296 : Assert(subtxn->nsubtxns == 0);
1626 :
1627 CBC 296 : ReorderBufferTruncateTXN(rb, subtxn, txn_prepared);
1628 : }
1629 :
1630 : /* cleanup changes in the txn */
1631 GIC 157382 : dlist_foreach_modify(iter, &txn->changes)
1632 ECB : {
1633 : ReorderBufferChange *change;
1634 :
1635 GIC 156389 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1636 ECB :
1637 : /* Check we're not mixing changes from different transactions. */
1638 GIC 156389 : Assert(change->txn == txn);
1639 :
1640 : /* remove the change from it's containing list */
1641 156389 : dlist_delete(&change->node);
1642 :
1643 CBC 156389 : ReorderBufferReturnChange(rb, change, true);
1644 ECB : }
1645 :
1646 : /*
1647 : * Mark the transaction as streamed.
1648 : *
1649 : * The top-level transaction, is marked as streamed always, even if it
1650 : * does not contain any changes (that is, when all the changes are in
1651 : * subtransactions).
1652 : *
1653 : * For subtransactions, we only mark them as streamed when there are
1654 : * changes in them.
1655 : *
1656 : * We do it this way because of aborts - we don't want to send aborts for
1657 : * XIDs the downstream is not aware of. And of course, it always knows
1658 : * about the toplevel xact (we send the XID in all messages), but we never
1659 : * stream XIDs of empty subxacts.
1660 : */
1661 GNC 993 : if ((!txn_prepared) && (rbtxn_is_toptxn(txn) || (txn->nentries_mem != 0)))
1662 CBC 781 : txn->txn_flags |= RBTXN_IS_STREAMED;
1663 :
1664 GIC 993 : if (txn_prepared)
1665 : {
1666 : /*
1667 : * If this is a prepared txn, cleanup the tuplecids we stored for
1668 : * decoding catalog snapshot access. They are always stored in the
1669 : * toplevel transaction.
1670 : */
1671 174 : dlist_foreach_modify(iter, &txn->tuplecids)
1672 : {
1673 : ReorderBufferChange *change;
1674 :
1675 123 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1676 :
1677 : /* Check we're not mixing changes from different transactions. */
1678 123 : Assert(change->txn == txn);
1679 123 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1680 ECB :
1681 : /* Remove the change from its containing list. */
1682 GIC 123 : dlist_delete(&change->node);
1683 ECB :
1684 GIC 123 : ReorderBufferReturnChange(rb, change, true);
1685 : }
1686 : }
1687 :
1688 : /*
1689 : * Destroy the (relfilelocator, ctid) hashtable, so that we don't leak any
1690 ECB : * memory. We could also keep the hash table and update it with new ctid
1691 : * values, but this seems simpler and good enough for now.
1692 : */
1693 GIC 993 : if (txn->tuplecid_hash != NULL)
1694 ECB : {
1695 GIC 22 : hash_destroy(txn->tuplecid_hash);
1696 22 : txn->tuplecid_hash = NULL;
1697 ECB : }
1698 :
1699 : /* If this txn is serialized then clean the disk space. */
1700 GIC 993 : if (rbtxn_is_serialized(txn))
1701 ECB : {
1702 GIC 6 : ReorderBufferRestoreCleanup(rb, txn);
1703 CBC 6 : txn->txn_flags &= ~RBTXN_IS_SERIALIZED;
1704 :
1705 : /*
1706 : * We set this flag to indicate if the transaction is ever serialized.
1707 : * We need this to accurately update the stats as otherwise the same
1708 : * transaction can be counted as serialized multiple times.
1709 : */
1710 GIC 6 : txn->txn_flags |= RBTXN_IS_SERIALIZED_CLEAR;
1711 : }
1712 ECB :
1713 : /* also reset the number of entries in the transaction */
1714 CBC 993 : txn->nentries_mem = 0;
1715 993 : txn->nentries = 0;
1716 GIC 993 : }
1717 :
1718 : /*
1719 : * Build a hash with a (relfilelocator, ctid) -> (cmin, cmax) mapping for use by
1720 : * HeapTupleSatisfiesHistoricMVCC.
1721 ECB : */
1722 : static void
1723 GIC 1700 : ReorderBufferBuildTupleCidHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
1724 : {
1725 : dlist_iter iter;
1726 : HASHCTL hash_ctl;
1727 :
1728 1700 : if (!rbtxn_has_catalog_changes(txn) || dlist_is_empty(&txn->tuplecids))
1729 CBC 1246 : return;
1730 :
1731 GIC 454 : hash_ctl.keysize = sizeof(ReorderBufferTupleCidKey);
1732 454 : hash_ctl.entrysize = sizeof(ReorderBufferTupleCidEnt);
1733 CBC 454 : hash_ctl.hcxt = rb->context;
1734 ECB :
1735 : /*
1736 : * create the hash with the exact number of to-be-stored tuplecids from
1737 : * the start
1738 : */
1739 GIC 454 : txn->tuplecid_hash =
1740 454 : hash_create("ReorderBufferTupleCid", txn->ntuplecids, &hash_ctl,
1741 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1742 ECB :
1743 GIC 10013 : dlist_foreach(iter, &txn->tuplecids)
1744 : {
1745 : ReorderBufferTupleCidKey key;
1746 : ReorderBufferTupleCidEnt *ent;
1747 ECB : bool found;
1748 : ReorderBufferChange *change;
1749 :
1750 CBC 9559 : change = dlist_container(ReorderBufferChange, node, iter.cur);
1751 ECB :
1752 CBC 9559 : Assert(change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID);
1753 :
1754 : /* be careful about padding */
1755 GIC 9559 : memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
1756 :
1757 GNC 9559 : key.rlocator = change->data.tuplecid.locator;
1758 ECB :
1759 CBC 9559 : ItemPointerCopy(&change->data.tuplecid.tid,
1760 : &key.tid);
1761 :
1762 ECB : ent = (ReorderBufferTupleCidEnt *)
1763 GNC 9559 : hash_search(txn->tuplecid_hash, &key, HASH_ENTER, &found);
1764 GIC 9559 : if (!found)
1765 : {
1766 CBC 8027 : ent->cmin = change->data.tuplecid.cmin;
1767 GIC 8027 : ent->cmax = change->data.tuplecid.cmax;
1768 CBC 8027 : ent->combocid = change->data.tuplecid.combocid;
1769 : }
1770 : else
1771 ECB : {
1772 : /*
1773 : * Maybe we already saw this tuple before in this transaction, but
1774 : * if so it must have the same cmin.
1775 : */
1776 GIC 1532 : Assert(ent->cmin == change->data.tuplecid.cmin);
1777 :
1778 : /*
1779 ECB : * cmax may be initially invalid, but once set it can only grow,
1780 : * and never become invalid again.
1781 : */
1782 CBC 1532 : Assert((ent->cmax == InvalidCommandId) ||
1783 ECB : ((change->data.tuplecid.cmax != InvalidCommandId) &&
1784 : (change->data.tuplecid.cmax > ent->cmax)));
1785 GIC 1532 : ent->cmax = change->data.tuplecid.cmax;
1786 : }
1787 : }
1788 : }
1789 :
1790 : /*
1791 : * Copy a provided snapshot so we can modify it privately. This is needed so
1792 ECB : * that catalog modifying transactions can look into intermediate catalog
1793 : * states.
1794 : */
1795 : static Snapshot
1796 GIC 1534 : ReorderBufferCopySnap(ReorderBuffer *rb, Snapshot orig_snap,
1797 : ReorderBufferTXN *txn, CommandId cid)
1798 ECB : {
1799 : Snapshot snap;
1800 : dlist_iter iter;
1801 CBC 1534 : int i = 0;
1802 : Size size;
1803 :
1804 GIC 1534 : size = sizeof(SnapshotData) +
1805 1534 : sizeof(TransactionId) * orig_snap->xcnt +
1806 1534 : sizeof(TransactionId) * (txn->nsubtxns + 1);
1807 :
1808 1534 : snap = MemoryContextAllocZero(rb->context, size);
1809 1534 : memcpy(snap, orig_snap, sizeof(SnapshotData));
1810 :
1811 1534 : snap->copied = true;
1812 CBC 1534 : snap->active_count = 1; /* mark as active so nobody frees it */
1813 GIC 1534 : snap->regd_count = 0;
1814 1534 : snap->xip = (TransactionId *) (snap + 1);
1815 :
1816 1534 : memcpy(snap->xip, orig_snap->xip, sizeof(TransactionId) * snap->xcnt);
1817 ECB :
1818 : /*
1819 : * snap->subxip contains all txids that belong to our transaction which we
1820 : * need to check via cmin/cmax. That's why we store the toplevel
1821 : * transaction in there as well.
1822 : */
1823 GIC 1534 : snap->subxip = snap->xip + snap->xcnt;
1824 CBC 1534 : snap->subxip[i++] = txn->xid;
1825 ECB :
1826 : /*
1827 : * subxcnt isn't decreased when subtransactions abort, so count manually.
1828 : * Since it's an upper boundary it is safe to use it for the allocation
1829 : * above.
1830 : */
1831 GIC 1534 : snap->subxcnt = 1;
1832 ECB :
1833 GIC 1842 : dlist_foreach(iter, &txn->subtxns)
1834 : {
1835 : ReorderBufferTXN *sub_txn;
1836 :
1837 308 : sub_txn = dlist_container(ReorderBufferTXN, node, iter.cur);
1838 308 : snap->subxip[i++] = sub_txn->xid;
1839 CBC 308 : snap->subxcnt++;
1840 ECB : }
1841 :
1842 : /* sort so we can bsearch() later */
1843 GIC 1534 : qsort(snap->subxip, snap->subxcnt, sizeof(TransactionId), xidComparator);
1844 :
1845 : /* store the specified current CommandId */
1846 1534 : snap->curcid = cid;
1847 ECB :
1848 GIC 1534 : return snap;
1849 ECB : }
1850 :
1851 : /*
1852 : * Free a previously ReorderBufferCopySnap'ed snapshot
1853 : */
1854 : static void
1855 CBC 2528 : ReorderBufferFreeSnap(ReorderBuffer *rb, Snapshot snap)
1856 : {
1857 GIC 2528 : if (snap->copied)
1858 1532 : pfree(snap);
1859 ECB : else
1860 GIC 996 : SnapBuildSnapDecRefcount(snap);
1861 2528 : }
1862 ECB :
1863 : /*
1864 : * If the transaction was (partially) streamed, we need to prepare or commit
1865 : * it in a 'streamed' way. That is, we first stream the remaining part of the
1866 : * transaction, and then invoke stream_prepare or stream_commit message as per
1867 : * the case.
1868 : */
1869 : static void
1870 GIC 61 : ReorderBufferStreamCommit(ReorderBuffer *rb, ReorderBufferTXN *txn)
1871 ECB : {
1872 : /* we should only call this for previously streamed transactions */
1873 CBC 61 : Assert(rbtxn_is_streamed(txn));
1874 ECB :
1875 GIC 61 : ReorderBufferStreamTXN(rb, txn);
1876 ECB :
1877 CBC 61 : if (rbtxn_prepared(txn))
1878 : {
1879 : /*
1880 : * Note, we send stream prepare even if a concurrent abort is
1881 : * detected. See DecodePrepare for more information.
1882 : */
1883 GIC 12 : rb->stream_prepare(rb, txn, txn->final_lsn);
1884 :
1885 : /*
1886 ECB : * This is a PREPARED transaction, part of a two-phase commit. The
1887 : * full cleanup will happen as part of the COMMIT PREPAREDs, so now
1888 : * just truncate txn by removing changes and tuple_cids.
1889 : */
1890 GIC 12 : ReorderBufferTruncateTXN(rb, txn, true);
1891 ECB : /* Reset the CheckXidAlive */
1892 GIC 12 : CheckXidAlive = InvalidTransactionId;
1893 ECB : }
1894 : else
1895 : {
1896 GIC 49 : rb->stream_commit(rb, txn, txn->final_lsn);
1897 49 : ReorderBufferCleanupTXN(rb, txn);
1898 : }
1899 CBC 61 : }
1900 :
1901 : /*
1902 : * Set xid to detect concurrent aborts.
1903 : *
1904 : * While streaming an in-progress transaction or decoding a prepared
1905 : * transaction there is a possibility that the (sub)transaction might get
1906 ECB : * aborted concurrently. In such case if the (sub)transaction has catalog
1907 : * update then we might decode the tuple using wrong catalog version. For
1908 : * example, suppose there is one catalog tuple with (xmin: 500, xmax: 0). Now,
1909 : * the transaction 501 updates the catalog tuple and after that we will have
1910 : * two tuples (xmin: 500, xmax: 501) and (xmin: 501, xmax: 0). Now, if 501 is
1911 : * aborted and some other transaction say 502 updates the same catalog tuple
1912 : * then the first tuple will be changed to (xmin: 500, xmax: 502). So, the
1913 : * problem is that when we try to decode the tuple inserted/updated in 501
1914 : * after the catalog update, we will see the catalog tuple with (xmin: 500,
1915 : * xmax: 502) as visible because it will consider that the tuple is deleted by
1916 : * xid 502 which is not visible to our snapshot. And when we will try to
1917 : * decode with that catalog tuple, it can lead to a wrong result or a crash.
1918 : * So, it is necessary to detect concurrent aborts to allow streaming of
1919 : * in-progress transactions or decoding of prepared transactions.
1920 : *
1921 : * For detecting the concurrent abort we set CheckXidAlive to the current
1922 : * (sub)transaction's xid for which this change belongs to. And, during
1923 : * catalog scan we can check the status of the xid and if it is aborted we will
1924 : * report a specific error so that we can stop streaming current transaction
1925 : * and discard the already streamed changes on such an error. We might have
1926 : * already streamed some of the changes for the aborted (sub)transaction, but
1927 : * that is fine because when we decode the abort we will stream abort message
1928 : * to truncate the changes in the subscriber. Similarly, for prepared
1929 : * transactions, we stop decoding if concurrent abort is detected and then
1930 : * rollback the changes when rollback prepared is encountered. See
1931 : * DecodePrepare.
1932 : */
1933 : static inline void
1934 GIC 177674 : SetupCheckXidLive(TransactionId xid)
1935 : {
1936 : /*
1937 : * If the input transaction id is already set as a CheckXidAlive then
1938 : * nothing to do.
1939 : */
1940 177674 : if (TransactionIdEquals(CheckXidAlive, xid))
1941 91730 : return;
1942 :
1943 : /*
1944 : * setup CheckXidAlive if it's not committed yet. We don't check if the
1945 : * xid is aborted. That will happen during catalog access.
1946 : */
1947 85944 : if (!TransactionIdDidCommit(xid))
1948 322 : CheckXidAlive = xid;
1949 : else
1950 CBC 85622 : CheckXidAlive = InvalidTransactionId;
1951 : }
1952 :
1953 : /*
1954 : * Helper function for ReorderBufferProcessTXN for applying change.
1955 : */
1956 ECB : static inline void
1957 CBC 333687 : ReorderBufferApplyChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
1958 : Relation relation, ReorderBufferChange *change,
1959 : bool streaming)
1960 : {
1961 GIC 333687 : if (streaming)
1962 175984 : rb->stream_change(rb, txn, relation, change);
1963 ECB : else
1964 CBC 157703 : rb->apply_change(rb, txn, relation, change);
1965 GIC 333683 : }
1966 ECB :
1967 : /*
1968 : * Helper function for ReorderBufferProcessTXN for applying the truncate.
1969 : */
1970 : static inline void
1971 GIC 17 : ReorderBufferApplyTruncate(ReorderBuffer *rb, ReorderBufferTXN *txn,
1972 : int nrelations, Relation *relations,
1973 ECB : ReorderBufferChange *change, bool streaming)
1974 : {
1975 GIC 17 : if (streaming)
1976 UIC 0 : rb->stream_truncate(rb, txn, nrelations, relations, change);
1977 ECB : else
1978 CBC 17 : rb->apply_truncate(rb, txn, nrelations, relations, change);
1979 GIC 17 : }
1980 ECB :
1981 : /*
1982 : * Helper function for ReorderBufferProcessTXN for applying the message.
1983 : */
1984 : static inline void
1985 GIC 11 : ReorderBufferApplyMessage(ReorderBuffer *rb, ReorderBufferTXN *txn,
1986 : ReorderBufferChange *change, bool streaming)
1987 ECB : {
1988 GIC 11 : if (streaming)
1989 3 : rb->stream_message(rb, txn, change->lsn, true,
1990 3 : change->data.msg.prefix,
1991 ECB : change->data.msg.message_size,
1992 GBC 3 : change->data.msg.message);
1993 : else
1994 CBC 8 : rb->message(rb, txn, change->lsn, true,
1995 8 : change->data.msg.prefix,
1996 : change->data.msg.message_size,
1997 GIC 8 : change->data.msg.message);
1998 11 : }
1999 :
2000 : /*
2001 ECB : * Function to store the command id and snapshot at the end of the current
2002 : * stream so that we can reuse the same while sending the next stream.
2003 : */
2004 : static inline void
2005 CBC 660 : ReorderBufferSaveTXNSnapshot(ReorderBuffer *rb, ReorderBufferTXN *txn,
2006 ECB : Snapshot snapshot_now, CommandId command_id)
2007 : {
2008 CBC 660 : txn->command_id = command_id;
2009 :
2010 ECB : /* Avoid copying if it's already copied. */
2011 CBC 660 : if (snapshot_now->copied)
2012 GIC 660 : txn->snapshot_now = snapshot_now;
2013 ECB : else
2014 LBC 0 : txn->snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
2015 : txn, command_id);
2016 GIC 660 : }
2017 :
2018 : /*
2019 : * Helper function for ReorderBufferProcessTXN to handle the concurrent
2020 : * abort of the streaming transaction. This resets the TXN such that it
2021 ECB : * can be used to stream the remaining data of transaction being processed.
2022 : * This can happen when the subtransaction is aborted and we still want to
2023 : * continue processing the main or other subtransactions data.
2024 : */
2025 : static void
2026 GIC 7 : ReorderBufferResetTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
2027 ECB : Snapshot snapshot_now,
2028 : CommandId command_id,
2029 : XLogRecPtr last_lsn,
2030 EUB : ReorderBufferChange *specinsert)
2031 : {
2032 ECB : /* Discard the changes that we just streamed */
2033 GIC 7 : ReorderBufferTruncateTXN(rb, txn, rbtxn_prepared(txn));
2034 :
2035 : /* Free all resources allocated for toast reconstruction */
2036 7 : ReorderBufferToastReset(rb, txn);
2037 :
2038 : /* Return the spec insert change if it is not NULL */
2039 7 : if (specinsert != NULL)
2040 : {
2041 UIC 0 : ReorderBufferReturnChange(rb, specinsert, true);
2042 LBC 0 : specinsert = NULL;
2043 : }
2044 :
2045 : /*
2046 : * For the streaming case, stop the stream and remember the command ID and
2047 : * snapshot for the streaming run.
2048 : */
2049 CBC 7 : if (rbtxn_is_streamed(txn))
2050 : {
2051 GIC 7 : rb->stream_stop(rb, txn, last_lsn);
2052 CBC 7 : ReorderBufferSaveTXNSnapshot(rb, txn, snapshot_now, command_id);
2053 : }
2054 GIC 7 : }
2055 ECB :
2056 : /*
2057 EUB : * Helper function for ReorderBufferReplay and ReorderBufferStreamTXN.
2058 : *
2059 : * Send data of a transaction (and its subtransactions) to the
2060 : * output plugin. We iterate over the top and subtransactions (using a k-way
2061 : * merge) and replay the changes in lsn order.
2062 : *
2063 : * If streaming is true then data will be sent using stream API.
2064 : *
2065 ECB : * Note: "volatile" markers on some parameters are to avoid trouble with
2066 : * PG_TRY inside the function.
2067 : */
2068 : static void
2069 GIC 1700 : ReorderBufferProcessTXN(ReorderBuffer *rb, ReorderBufferTXN *txn,
2070 ECB : XLogRecPtr commit_lsn,
2071 : volatile Snapshot snapshot_now,
2072 : volatile CommandId command_id,
2073 : bool streaming)
2074 : {
2075 : bool using_subtxn;
2076 GIC 1700 : MemoryContext ccxt = CurrentMemoryContext;
2077 1700 : ReorderBufferIterTXNState *volatile iterstate = NULL;
2078 1700 : volatile XLogRecPtr prev_lsn = InvalidXLogRecPtr;
2079 1700 : ReorderBufferChange *volatile specinsert = NULL;
2080 1700 : volatile bool stream_started = false;
2081 1700 : ReorderBufferTXN *volatile curtxn = NULL;
2082 :
2083 : /* build data to be able to lookup the CommandIds of catalog tuples */
2084 1700 : ReorderBufferBuildTupleCidHash(rb, txn);
2085 ECB :
2086 : /* setup the initial snapshot */
2087 GIC 1700 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2088 :
2089 : /*
2090 : * Decoding needs access to syscaches et al., which in turn use
2091 : * heavyweight locks and such. Thus we need to have enough state around to
2092 ECB : * keep track of those. The easiest way is to simply use a transaction
2093 : * internally. That also allows us to easily enforce that nothing writes
2094 : * to the database by checking for xid assignments.
2095 : *
2096 : * When we're called via the SQL SRF there's already a transaction
2097 : * started, so start an explicit subtransaction there.
2098 : */
2099 GIC 1700 : using_subtxn = IsTransactionOrTransactionBlock();
2100 ECB :
2101 GIC 1700 : PG_TRY();
2102 : {
2103 ECB : ReorderBufferChange *change;
2104 GNC 1700 : int changes_count = 0; /* used to accumulate the number of
2105 : * changes */
2106 :
2107 GIC 1700 : if (using_subtxn)
2108 427 : BeginInternalSubTransaction(streaming ? "stream" : "replay");
2109 : else
2110 1273 : StartTransactionCommand();
2111 :
2112 : /*
2113 : * We only need to send begin/begin-prepare for non-streamed
2114 : * transactions.
2115 : */
2116 1700 : if (!streaming)
2117 ECB : {
2118 GIC 1040 : if (rbtxn_prepared(txn))
2119 CBC 25 : rb->begin_prepare(rb, txn);
2120 : else
2121 GIC 1015 : rb->begin(rb, txn);
2122 ECB : }
2123 :
2124 GIC 1700 : ReorderBufferIterTXNInit(rb, txn, &iterstate);
2125 CBC 356174 : while ((change = ReorderBufferIterTXNNext(rb, iterstate)) != NULL)
2126 ECB : {
2127 GIC 352785 : Relation relation = NULL;
2128 ECB : Oid reloid;
2129 :
2130 GIC 352785 : CHECK_FOR_INTERRUPTS();
2131 :
2132 : /*
2133 : * We can't call start stream callback before processing first
2134 ECB : * change.
2135 : */
2136 CBC 352785 : if (prev_lsn == InvalidXLogRecPtr)
2137 ECB : {
2138 GIC 1666 : if (streaming)
2139 ECB : {
2140 GIC 626 : txn->origin_id = change->origin_id;
2141 626 : rb->stream_start(rb, txn, change->lsn);
2142 CBC 626 : stream_started = true;
2143 ECB : }
2144 : }
2145 :
2146 : /*
2147 : * Enforce correct ordering of changes, merged from multiple
2148 : * subtransactions. The changes may have the same LSN due to
2149 : * MULTI_INSERT xlog records.
2150 : */
2151 GIC 352785 : Assert(prev_lsn == InvalidXLogRecPtr || prev_lsn <= change->lsn);
2152 :
2153 352785 : prev_lsn = change->lsn;
2154 ECB :
2155 : /*
2156 : * Set the current xid to detect concurrent aborts. This is
2157 : * required for the cases when we decode the changes before the
2158 : * COMMIT record is processed.
2159 : */
2160 CBC 352785 : if (streaming || rbtxn_prepared(change->txn))
2161 : {
2162 GIC 177674 : curtxn = change->txn;
2163 177674 : SetupCheckXidLive(curtxn->xid);
2164 : }
2165 :
2166 352785 : switch (change->action)
2167 : {
2168 1782 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
2169 ECB :
2170 : /*
2171 : * Confirmation for speculative insertion arrived. Simply
2172 : * use as a normal record. It'll be cleaned up at the end
2173 : * of INSERT processing.
2174 : */
2175 GIC 1782 : if (specinsert == NULL)
2176 UIC 0 : elog(ERROR, "invalid ordering of speculative insertion changes");
2177 GIC 1782 : Assert(specinsert->data.tp.oldtuple == NULL);
2178 CBC 1782 : change = specinsert;
2179 GIC 1782 : change->action = REORDER_BUFFER_CHANGE_INSERT;
2180 ECB :
2181 : /* intentionally fall through */
2182 GIC 339295 : case REORDER_BUFFER_CHANGE_INSERT:
2183 : case REORDER_BUFFER_CHANGE_UPDATE:
2184 ECB : case REORDER_BUFFER_CHANGE_DELETE:
2185 GIC 339295 : Assert(snapshot_now);
2186 ECB :
2187 GNC 339295 : reloid = RelidByRelfilenumber(change->data.tp.rlocator.spcOid,
2188 : change->data.tp.rlocator.relNumber);
2189 :
2190 : /*
2191 : * Mapped catalog tuple without data, emitted while
2192 : * catalog table was in the process of being rewritten. We
2193 : * can fail to look up the relfilenumber, because the
2194 EUB : * relmapper has no "historic" view, in contrast to the
2195 ECB : * normal catalog during decoding. Thus repeated rewrites
2196 : * can cause a lookup failure. That's OK because we do not
2197 : * decode catalog changes anyway. Normally such tuples
2198 : * would be skipped over below, but we can't identify
2199 : * whether the table should be logically logged without
2200 : * mapping the relfilenumber to the oid.
2201 : */
2202 GIC 339288 : if (reloid == InvalidOid &&
2203 CBC 76 : change->data.tp.newtuple == NULL &&
2204 GIC 76 : change->data.tp.oldtuple == NULL)
2205 CBC 76 : goto change_done;
2206 GIC 339212 : else if (reloid == InvalidOid)
2207 UNC 0 : elog(ERROR, "could not map filenumber \"%s\" to relation OID",
2208 : relpathperm(change->data.tp.rlocator,
2209 : MAIN_FORKNUM));
2210 :
2211 GIC 339212 : relation = RelationIdGetRelation(reloid);
2212 :
2213 339212 : if (!RelationIsValid(relation))
2214 UNC 0 : elog(ERROR, "could not open relation with OID %u (for filenumber \"%s\")",
2215 : reloid,
2216 : relpathperm(change->data.tp.rlocator,
2217 : MAIN_FORKNUM));
2218 :
2219 GIC 339212 : if (!RelationIsLogicallyLogged(relation))
2220 CBC 3550 : goto change_done;
2221 ECB :
2222 : /*
2223 : * Ignore temporary heaps created during DDL unless the
2224 : * plugin has asked for them.
2225 EUB : */
2226 GIC 335662 : if (relation->rd_rel->relrewrite && !rb->output_rewrites)
2227 24 : goto change_done;
2228 :
2229 ECB : /*
2230 : * For now ignore sequence changes entirely. Most of the
2231 : * time they don't log changes using records we
2232 EUB : * understand, so it doesn't make sense to handle the few
2233 : * cases we do.
2234 : */
2235 GIC 335638 : if (relation->rd_rel->relkind == RELKIND_SEQUENCE)
2236 UIC 0 : goto change_done;
2237 ECB :
2238 : /* user-triggered change */
2239 GIC 335638 : if (!IsToastRelation(relation))
2240 : {
2241 333687 : ReorderBufferToastReplace(rb, txn, relation, change);
2242 333687 : ReorderBufferApplyChange(rb, txn, relation, change,
2243 : streaming);
2244 ECB :
2245 : /*
2246 : * Only clear reassembled toast chunks if we're sure
2247 : * they're not required anymore. The creator of the
2248 : * tuple tells us.
2249 : */
2250 GIC 333683 : if (change->data.tp.clear_toast_afterwards)
2251 333462 : ReorderBufferToastReset(rb, txn);
2252 : }
2253 ECB : /* we're not interested in toast deletions */
2254 GBC 1951 : else if (change->action == REORDER_BUFFER_CHANGE_INSERT)
2255 : {
2256 : /*
2257 ECB : * Need to reassemble the full toasted Datum in
2258 : * memory, to ensure the chunks don't get reused till
2259 : * we're done remove it from the list of this
2260 : * transaction's changes. Otherwise it will get
2261 : * freed/reused while restoring spooled data from
2262 : * disk.
2263 : */
2264 GIC 1720 : Assert(change->data.tp.newtuple != NULL);
2265 :
2266 1720 : dlist_delete(&change->node);
2267 1720 : ReorderBufferToastAppendChunk(rb, txn, relation,
2268 ECB : change);
2269 : }
2270 :
2271 GIC 231 : change_done:
2272 ECB :
2273 : /*
2274 : * If speculative insertion was confirmed, the record
2275 : * isn't needed anymore.
2276 : */
2277 GIC 339284 : if (specinsert != NULL)
2278 : {
2279 1782 : ReorderBufferReturnChange(rb, specinsert, true);
2280 1782 : specinsert = NULL;
2281 : }
2282 ECB :
2283 GIC 339284 : if (RelationIsValid(relation))
2284 ECB : {
2285 CBC 339208 : RelationClose(relation);
2286 GIC 339208 : relation = NULL;
2287 : }
2288 339284 : break;
2289 ECB :
2290 GIC 1782 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
2291 :
2292 : /*
2293 : * Speculative insertions are dealt with by delaying the
2294 : * processing of the insert until the confirmation record
2295 ECB : * arrives. For that we simply unlink the record from the
2296 : * chain, so it does not get freed/reused while restoring
2297 : * spooled data from disk.
2298 : *
2299 : * This is safe in the face of concurrent catalog changes
2300 : * because the relevant relation can't be changed between
2301 : * speculative insertion and confirmation due to
2302 : * CheckTableNotInUse() and locking.
2303 : */
2304 :
2305 : /* clear out a pending (and thus failed) speculation */
2306 CBC 1782 : if (specinsert != NULL)
2307 : {
2308 LBC 0 : ReorderBufferReturnChange(rb, specinsert, true);
2309 UIC 0 : specinsert = NULL;
2310 : }
2311 :
2312 : /* and memorize the pending insertion */
2313 GIC 1782 : dlist_delete(&change->node);
2314 1782 : specinsert = change;
2315 1782 : break;
2316 :
2317 UIC 0 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
2318 :
2319 : /*
2320 : * Abort for speculative insertion arrived. So cleanup the
2321 : * specinsert tuple and toast hash.
2322 : *
2323 : * Note that we get the spec abort change for each toast
2324 ECB : * entry but we need to perform the cleanup only the first
2325 : * time we get it for the main table.
2326 EUB : */
2327 UBC 0 : if (specinsert != NULL)
2328 : {
2329 : /*
2330 : * We must clean the toast hash before processing a
2331 ECB : * completely new tuple to avoid confusion about the
2332 : * previous tuple's toast chunks.
2333 : */
2334 UIC 0 : Assert(change->data.tp.clear_toast_afterwards);
2335 UBC 0 : ReorderBufferToastReset(rb, txn);
2336 :
2337 : /* We don't need this record anymore. */
2338 UIC 0 : ReorderBufferReturnChange(rb, specinsert, true);
2339 0 : specinsert = NULL;
2340 : }
2341 0 : break;
2342 :
2343 GIC 17 : case REORDER_BUFFER_CHANGE_TRUNCATE:
2344 : {
2345 EUB : int i;
2346 GIC 17 : int nrelids = change->data.truncate.nrelids;
2347 17 : int nrelations = 0;
2348 : Relation *relations;
2349 :
2350 17 : relations = palloc0(nrelids * sizeof(Relation));
2351 44 : for (i = 0; i < nrelids; i++)
2352 EUB : {
2353 GBC 27 : Oid relid = change->data.truncate.relids[i];
2354 : Relation rel;
2355 :
2356 GNC 27 : rel = RelationIdGetRelation(relid);
2357 EUB :
2358 GNC 27 : if (!RelationIsValid(rel))
2359 UBC 0 : elog(ERROR, "could not open relation with OID %u", relid);
2360 :
2361 GNC 27 : if (!RelationIsLogicallyLogged(rel))
2362 UIC 0 : continue;
2363 :
2364 GNC 27 : relations[nrelations++] = rel;
2365 ECB : }
2366 :
2367 : /* Apply the truncate. */
2368 CBC 17 : ReorderBufferApplyTruncate(rb, txn, nrelations,
2369 ECB : relations, change,
2370 : streaming);
2371 :
2372 GIC 44 : for (i = 0; i < nrelations; i++)
2373 27 : RelationClose(relations[i]);
2374 ECB :
2375 GIC 17 : break;
2376 ECB : }
2377 EUB :
2378 GIC 11 : case REORDER_BUFFER_CHANGE_MESSAGE:
2379 CBC 11 : ReorderBufferApplyMessage(rb, txn, change, streaming);
2380 GBC 11 : break;
2381 :
2382 CBC 1845 : case REORDER_BUFFER_CHANGE_INVALIDATION:
2383 : /* Execute the invalidation messages locally */
2384 GIC 1845 : ReorderBufferExecuteInvalidations(change->data.inval.ninvalidations,
2385 : change->data.inval.invalidations);
2386 CBC 1845 : break;
2387 :
2388 GIC 460 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
2389 : /* get rid of the old */
2390 CBC 460 : TeardownHistoricSnapshot(false);
2391 ECB :
2392 GIC 460 : if (snapshot_now->copied)
2393 ECB : {
2394 GIC 440 : ReorderBufferFreeSnap(rb, snapshot_now);
2395 440 : snapshot_now =
2396 CBC 440 : ReorderBufferCopySnap(rb, change->data.snapshot,
2397 ECB : txn, command_id);
2398 : }
2399 :
2400 : /*
2401 : * Restored from disk, need to be careful not to double
2402 : * free. We could introduce refcounting for that, but for
2403 : * now this seems infrequent enough not to care.
2404 : */
2405 GIC 20 : else if (change->data.snapshot->copied)
2406 ECB : {
2407 UIC 0 : snapshot_now =
2408 LBC 0 : ReorderBufferCopySnap(rb, change->data.snapshot,
2409 : txn, command_id);
2410 ECB : }
2411 : else
2412 : {
2413 CBC 20 : snapshot_now = change->data.snapshot;
2414 ECB : }
2415 :
2416 : /* and continue with the new one */
2417 GIC 460 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2418 460 : break;
2419 :
2420 9375 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
2421 9375 : Assert(change->data.command_id != InvalidCommandId);
2422 :
2423 CBC 9375 : if (command_id < change->data.command_id)
2424 : {
2425 GBC 1581 : command_id = change->data.command_id;
2426 EUB :
2427 GIC 1581 : if (!snapshot_now->copied)
2428 : {
2429 : /* we don't use the global one anymore */
2430 434 : snapshot_now = ReorderBufferCopySnap(rb, snapshot_now,
2431 ECB : txn, command_id);
2432 : }
2433 :
2434 GIC 1581 : snapshot_now->curcid = command_id;
2435 ECB :
2436 CBC 1581 : TeardownHistoricSnapshot(false);
2437 GIC 1581 : SetupHistoricSnapshot(snapshot_now, txn->tuplecid_hash);
2438 ECB : }
2439 :
2440 GIC 9375 : break;
2441 ECB :
2442 UIC 0 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
2443 LBC 0 : elog(ERROR, "tuplecid value in changequeue");
2444 : break;
2445 ECB : }
2446 :
2447 : /*
2448 : * It is possible that the data is not sent to downstream for a
2449 : * long time either because the output plugin filtered it or there
2450 : * is a DDL that generates a lot of data that is not processed by
2451 : * the plugin. So, in such cases, the downstream can timeout. To
2452 : * avoid that we try to send a keepalive message if required.
2453 : * Trying to send a keepalive message after every change has some
2454 : * overhead, but testing showed there is no noticeable overhead if
2455 : * we do it after every ~100 changes.
2456 : */
2457 : #define CHANGES_THRESHOLD 100
2458 :
2459 GNC 352774 : if (++changes_count >= CHANGES_THRESHOLD)
2460 : {
2461 3097 : rb->update_progress_txn(rb, txn, change->lsn);
2462 3097 : changes_count = 0;
2463 : }
2464 : }
2465 :
2466 ECB : /* speculative insertion record must be freed by now */
2467 GIC 1689 : Assert(!specinsert);
2468 :
2469 : /* clean up the iterator */
2470 CBC 1689 : ReorderBufferIterTXNFinish(rb, iterstate);
2471 GIC 1689 : iterstate = NULL;
2472 ECB :
2473 : /*
2474 : * Update total transaction count and total bytes processed by the
2475 : * transaction and its subtransactions. Ensure to not count the
2476 : * streamed transaction multiple times.
2477 : *
2478 EUB : * Note that the statistics computation has to be done after
2479 : * ReorderBufferIterTXNFinish as it releases the serialized change
2480 : * which we have already accounted in ReorderBufferIterTXNNext.
2481 : */
2482 GIC 1689 : if (!rbtxn_is_streamed(txn))
2483 1098 : rb->totalTxns++;
2484 :
2485 1689 : rb->totalBytes += txn->total_size;
2486 :
2487 : /*
2488 : * Done with current changes, send the last message for this set of
2489 : * changes depending upon streaming mode.
2490 : */
2491 1689 : if (streaming)
2492 : {
2493 653 : if (stream_started)
2494 : {
2495 CBC 619 : rb->stream_stop(rb, txn, prev_lsn);
2496 GIC 619 : stream_started = false;
2497 ECB : }
2498 : }
2499 : else
2500 : {
2501 : /*
2502 : * Call either PREPARE (for two-phase transactions) or COMMIT (for
2503 : * regular ones).
2504 : */
2505 GIC 1036 : if (rbtxn_prepared(txn))
2506 CBC 25 : rb->prepare(rb, txn, commit_lsn);
2507 ECB : else
2508 GIC 1011 : rb->commit(rb, txn, commit_lsn);
2509 : }
2510 :
2511 : /* this is just a sanity check against bad output plugin behaviour */
2512 1684 : if (GetCurrentTransactionIdIfAny() != InvalidTransactionId)
2513 UIC 0 : elog(ERROR, "output plugin used XID %u",
2514 : GetCurrentTransactionId());
2515 :
2516 : /*
2517 : * Remember the command ID and snapshot for the next set of changes in
2518 ECB : * streaming mode.
2519 : */
2520 GIC 1684 : if (streaming)
2521 CBC 653 : ReorderBufferSaveTXNSnapshot(rb, txn, snapshot_now, command_id);
2522 GIC 1031 : else if (snapshot_now->copied)
2523 434 : ReorderBufferFreeSnap(rb, snapshot_now);
2524 :
2525 : /* cleanup */
2526 1684 : TeardownHistoricSnapshot(false);
2527 ECB :
2528 : /*
2529 : * Aborting the current (sub-)transaction as a whole has the right
2530 : * semantics. We want all locks acquired in here to be released, not
2531 : * reassigned to the parent and we do not want any database access
2532 : * have persistent effects.
2533 : */
2534 GIC 1684 : AbortCurrentTransaction();
2535 :
2536 : /* make sure there's no cache pollution */
2537 1684 : ReorderBufferExecuteInvalidations(txn->ninvalidations, txn->invalidations);
2538 :
2539 1684 : if (using_subtxn)
2540 424 : RollbackAndReleaseCurrentSubTransaction();
2541 ECB :
2542 : /*
2543 : * We are here due to one of the four reasons: 1. Decoding an
2544 : * in-progress txn. 2. Decoding a prepared txn. 3. Decoding of a
2545 : * prepared txn that was (partially) streamed. 4. Decoding a committed
2546 : * txn.
2547 : *
2548 : * For 1, we allow truncation of txn data by removing the changes
2549 EUB : * already streamed but still keeping other things like invalidations,
2550 : * snapshot, and tuplecids. For 2 and 3, we indicate
2551 : * ReorderBufferTruncateTXN to do more elaborate truncation of txn
2552 : * data as the entire transaction has been decoded except for commit.
2553 : * For 4, as the entire txn has been decoded, we can fully clean up
2554 : * the TXN reorder buffer.
2555 : */
2556 CBC 1684 : if (streaming || rbtxn_prepared(txn))
2557 ECB : {
2558 CBC 678 : ReorderBufferTruncateTXN(rb, txn, rbtxn_prepared(txn));
2559 ECB : /* Reset the CheckXidAlive */
2560 GIC 678 : CheckXidAlive = InvalidTransactionId;
2561 : }
2562 ECB : else
2563 GIC 1006 : ReorderBufferCleanupTXN(rb, txn);
2564 : }
2565 10 : PG_CATCH();
2566 : {
2567 10 : MemoryContext ecxt = MemoryContextSwitchTo(ccxt);
2568 10 : ErrorData *errdata = CopyErrorData();
2569 :
2570 ECB : /* TODO: Encapsulate cleanup from the PG_TRY and PG_CATCH blocks */
2571 GIC 10 : if (iterstate)
2572 10 : ReorderBufferIterTXNFinish(rb, iterstate);
2573 ECB :
2574 GIC 10 : TeardownHistoricSnapshot(true);
2575 ECB :
2576 : /*
2577 : * Force cache invalidation to happen outside of a valid transaction
2578 : * to prevent catalog access as we just caught an error.
2579 : */
2580 GIC 10 : AbortCurrentTransaction();
2581 :
2582 : /* make sure there's no cache pollution */
2583 10 : ReorderBufferExecuteInvalidations(txn->ninvalidations,
2584 : txn->invalidations);
2585 :
2586 10 : if (using_subtxn)
2587 3 : RollbackAndReleaseCurrentSubTransaction();
2588 :
2589 : /*
2590 : * The error code ERRCODE_TRANSACTION_ROLLBACK indicates a concurrent
2591 : * abort of the (sub)transaction we are streaming or preparing. We
2592 ECB : * need to do the cleanup and return gracefully on this error, see
2593 : * SetupCheckXidLive.
2594 : *
2595 : * This error code can be thrown by one of the callbacks we call
2596 : * during decoding so we need to ensure that we return gracefully only
2597 : * when we are sending the data in streaming mode and the streaming is
2598 : * not finished yet or when we are sending the data out on a PREPARE
2599 : * during a two-phase commit.
2600 : */
2601 CBC 10 : if (errdata->sqlerrcode == ERRCODE_TRANSACTION_ROLLBACK &&
2602 GIC 7 : (stream_started || rbtxn_prepared(txn)))
2603 ECB : {
2604 : /* curtxn must be set for streaming or prepared transactions */
2605 GIC 7 : Assert(curtxn);
2606 :
2607 ECB : /* Cleanup the temporary error state. */
2608 CBC 7 : FlushErrorState();
2609 GIC 7 : FreeErrorData(errdata);
2610 CBC 7 : errdata = NULL;
2611 GIC 7 : curtxn->concurrent_abort = true;
2612 :
2613 : /* Reset the TXN so that it is allowed to stream remaining data. */
2614 7 : ReorderBufferResetTXN(rb, txn, snapshot_now,
2615 : command_id, prev_lsn,
2616 ECB : specinsert);
2617 : }
2618 : else
2619 : {
2620 GIC 3 : ReorderBufferCleanupTXN(rb, txn);
2621 3 : MemoryContextSwitchTo(ecxt);
2622 CBC 3 : PG_RE_THROW();
2623 ECB : }
2624 : }
2625 GIC 1691 : PG_END_TRY();
2626 1691 : }
2627 :
2628 : /*
2629 : * Perform the replay of a transaction and its non-aborted subtransactions.
2630 : *
2631 : * Subtransactions previously have to be processed by
2632 : * ReorderBufferCommitChild(), even if previously assigned to the toplevel
2633 : * transaction with ReorderBufferAssignChild.
2634 : *
2635 : * This interface is called once a prepare or toplevel commit is read for both
2636 : * streamed as well as non-streamed transactions.
2637 ECB : */
2638 : static void
2639 GIC 1102 : ReorderBufferReplay(ReorderBufferTXN *txn,
2640 : ReorderBuffer *rb, TransactionId xid,
2641 ECB : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2642 : TimestampTz commit_time,
2643 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2644 : {
2645 : Snapshot snapshot_now;
2646 CBC 1102 : CommandId command_id = FirstCommandId;
2647 ECB :
2648 GIC 1102 : txn->final_lsn = commit_lsn;
2649 1102 : txn->end_lsn = end_lsn;
2650 CBC 1102 : txn->xact_time.commit_time = commit_time;
2651 GIC 1102 : txn->origin_id = origin_id;
2652 1102 : txn->origin_lsn = origin_lsn;
2653 :
2654 : /*
2655 : * If the transaction was (partially) streamed, we need to commit it in a
2656 ECB : * 'streamed' way. That is, we first stream the remaining part of the
2657 : * transaction, and then invoke stream_commit message.
2658 : *
2659 : * Called after everything (origin ID, LSN, ...) is stored in the
2660 : * transaction to avoid passing that information directly.
2661 : */
2662 CBC 1102 : if (rbtxn_is_streamed(txn))
2663 : {
2664 GIC 61 : ReorderBufferStreamCommit(rb, txn);
2665 61 : return;
2666 : }
2667 :
2668 : /*
2669 : * If this transaction has no snapshot, it didn't make any changes to the
2670 : * database, so there's nothing to decode. Note that
2671 : * ReorderBufferCommitChild will have transferred any snapshots from
2672 : * subtransactions if there were any.
2673 : */
2674 1041 : if (txn->base_snapshot == NULL)
2675 ECB : {
2676 GIC 1 : Assert(txn->ninvalidations == 0);
2677 :
2678 : /*
2679 : * Removing this txn before a commit might result in the computation
2680 : * of an incorrect restart_lsn. See SnapBuildProcessRunningXacts.
2681 : */
2682 CBC 1 : if (!rbtxn_prepared(txn))
2683 GIC 1 : ReorderBufferCleanupTXN(rb, txn);
2684 CBC 1 : return;
2685 ECB : }
2686 :
2687 CBC 1040 : snapshot_now = txn->base_snapshot;
2688 ECB :
2689 : /* Process and send the changes to output plugin. */
2690 GIC 1040 : ReorderBufferProcessTXN(rb, txn, commit_lsn, snapshot_now,
2691 : command_id, false);
2692 : }
2693 :
2694 : /*
2695 : * Commit a transaction.
2696 : *
2697 : * See comments for ReorderBufferReplay().
2698 ECB : */
2699 : void
2700 CBC 1066 : ReorderBufferCommit(ReorderBuffer *rb, TransactionId xid,
2701 ECB : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2702 : TimestampTz commit_time,
2703 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2704 : {
2705 : ReorderBufferTXN *txn;
2706 :
2707 GIC 1066 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2708 : false);
2709 :
2710 ECB : /* unknown transaction, nothing to replay */
2711 GIC 1066 : if (txn == NULL)
2712 CBC 1 : return;
2713 :
2714 GIC 1065 : ReorderBufferReplay(txn, rb, xid, commit_lsn, end_lsn, commit_time,
2715 : origin_id, origin_lsn);
2716 : }
2717 :
2718 ECB : /*
2719 : * Record the prepare information for a transaction.
2720 : */
2721 : bool
2722 GIC 124 : ReorderBufferRememberPrepareInfo(ReorderBuffer *rb, TransactionId xid,
2723 ECB : XLogRecPtr prepare_lsn, XLogRecPtr end_lsn,
2724 : TimestampTz prepare_time,
2725 : RepOriginId origin_id, XLogRecPtr origin_lsn)
2726 : {
2727 : ReorderBufferTXN *txn;
2728 :
2729 GIC 124 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr, false);
2730 :
2731 : /* unknown transaction, nothing to do */
2732 124 : if (txn == NULL)
2733 UIC 0 : return false;
2734 :
2735 : /*
2736 ECB : * Remember the prepare information to be later used by commit prepared in
2737 : * case we skip doing prepare.
2738 : */
2739 GIC 124 : txn->final_lsn = prepare_lsn;
2740 124 : txn->end_lsn = end_lsn;
2741 124 : txn->xact_time.prepare_time = prepare_time;
2742 124 : txn->origin_id = origin_id;
2743 CBC 124 : txn->origin_lsn = origin_lsn;
2744 :
2745 GIC 124 : return true;
2746 : }
2747 ECB :
2748 : /* Remember that we have skipped prepare */
2749 : void
2750 CBC 88 : ReorderBufferSkipPrepare(ReorderBuffer *rb, TransactionId xid)
2751 : {
2752 : ReorderBufferTXN *txn;
2753 :
2754 GIC 88 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr, false);
2755 :
2756 : /* unknown transaction, nothing to do */
2757 88 : if (txn == NULL)
2758 LBC 0 : return;
2759 :
2760 GIC 88 : txn->txn_flags |= RBTXN_SKIPPED_PREPARE;
2761 : }
2762 :
2763 : /*
2764 : * Prepare a two-phase transaction.
2765 ECB : *
2766 : * See comments for ReorderBufferReplay().
2767 : */
2768 : void
2769 GBC 36 : ReorderBufferPrepare(ReorderBuffer *rb, TransactionId xid,
2770 : char *gid)
2771 : {
2772 : ReorderBufferTXN *txn;
2773 :
2774 GIC 36 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2775 ECB : false);
2776 :
2777 : /* unknown transaction, nothing to replay */
2778 CBC 36 : if (txn == NULL)
2779 LBC 0 : return;
2780 :
2781 CBC 36 : txn->txn_flags |= RBTXN_PREPARE;
2782 GIC 36 : txn->gid = pstrdup(gid);
2783 :
2784 : /* The prepare info must have been updated in txn by now. */
2785 36 : Assert(txn->final_lsn != InvalidXLogRecPtr);
2786 ECB :
2787 GIC 36 : ReorderBufferReplay(txn, rb, xid, txn->final_lsn, txn->end_lsn,
2788 36 : txn->xact_time.prepare_time, txn->origin_id, txn->origin_lsn);
2789 :
2790 ECB : /*
2791 : * We send the prepare for the concurrently aborted xacts so that later
2792 : * when rollback prepared is decoded and sent, the downstream should be
2793 : * able to rollback such a xact. See comments atop DecodePrepare.
2794 EUB : *
2795 : * Note, for the concurrent_abort + streaming case a stream_prepare was
2796 ECB : * already sent within the ReorderBufferReplay call above.
2797 : */
2798 GIC 36 : if (txn->concurrent_abort && !rbtxn_is_streamed(txn))
2799 UIC 0 : rb->prepare(rb, txn, txn->final_lsn);
2800 : }
2801 :
2802 : /*
2803 : * This is used to handle COMMIT/ROLLBACK PREPARED.
2804 : */
2805 ECB : void
2806 GIC 38 : ReorderBufferFinishPrepared(ReorderBuffer *rb, TransactionId xid,
2807 : XLogRecPtr commit_lsn, XLogRecPtr end_lsn,
2808 : XLogRecPtr two_phase_at,
2809 : TimestampTz commit_time, RepOriginId origin_id,
2810 ECB : XLogRecPtr origin_lsn, char *gid, bool is_commit)
2811 : {
2812 : ReorderBufferTXN *txn;
2813 : XLogRecPtr prepare_end_lsn;
2814 : TimestampTz prepare_time;
2815 EUB :
2816 GIC 38 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, commit_lsn, false);
2817 ECB :
2818 : /* unknown transaction, nothing to do */
2819 GIC 38 : if (txn == NULL)
2820 UIC 0 : return;
2821 ECB :
2822 : /*
2823 : * By this time the txn has the prepare record information, remember it to
2824 : * be later used for rollback.
2825 : */
2826 GIC 38 : prepare_end_lsn = txn->end_lsn;
2827 38 : prepare_time = txn->xact_time.prepare_time;
2828 :
2829 : /* add the gid in the txn */
2830 38 : txn->gid = pstrdup(gid);
2831 :
2832 : /*
2833 : * It is possible that this transaction is not decoded at prepare time
2834 ECB : * either because by that time we didn't have a consistent snapshot, or
2835 EUB : * two_phase was not enabled, or it was decoded earlier but we have
2836 : * restarted. We only need to send the prepare if it was not decoded
2837 : * earlier. We don't need to decode the xact for aborts if it is not done
2838 : * already.
2839 : */
2840 GIC 38 : if ((txn->final_lsn < two_phase_at) && is_commit)
2841 : {
2842 CBC 1 : txn->txn_flags |= RBTXN_PREPARE;
2843 :
2844 : /*
2845 : * The prepare info must have been updated in txn even if we skip
2846 : * prepare.
2847 : */
2848 GIC 1 : Assert(txn->final_lsn != InvalidXLogRecPtr);
2849 :
2850 : /*
2851 : * By this time the txn has the prepare record information and it is
2852 ECB : * important to use that so that downstream gets the accurate
2853 : * information. If instead, we have passed commit information here
2854 : * then downstream can behave as it has already replayed commit
2855 : * prepared after the restart.
2856 EUB : */
2857 GIC 1 : ReorderBufferReplay(txn, rb, xid, txn->final_lsn, txn->end_lsn,
2858 1 : txn->xact_time.prepare_time, txn->origin_id, txn->origin_lsn);
2859 : }
2860 :
2861 38 : txn->final_lsn = commit_lsn;
2862 CBC 38 : txn->end_lsn = end_lsn;
2863 38 : txn->xact_time.commit_time = commit_time;
2864 GIC 38 : txn->origin_id = origin_id;
2865 38 : txn->origin_lsn = origin_lsn;
2866 ECB :
2867 GIC 38 : if (is_commit)
2868 29 : rb->commit_prepared(rb, txn, commit_lsn);
2869 : else
2870 9 : rb->rollback_prepared(rb, txn, prepare_end_lsn, prepare_time);
2871 :
2872 : /* cleanup: make sure there's no cache pollution */
2873 38 : ReorderBufferExecuteInvalidations(txn->ninvalidations,
2874 : txn->invalidations);
2875 38 : ReorderBufferCleanupTXN(rb, txn);
2876 ECB : }
2877 :
2878 : /*
2879 : * Abort a transaction that possibly has previous changes. Needs to be first
2880 : * called for subtransactions and then for the toplevel xid.
2881 : *
2882 : * NB: Transactions handled here have to have actively aborted (i.e. have
2883 : * produced an abort record). Implicitly aborted transactions are handled via
2884 : * ReorderBufferAbortOld(); transactions we're just not interested in, but
2885 : * which have committed are handled in ReorderBufferForget().
2886 : *
2887 : * This function purges this transaction and its contents from memory and
2888 : * disk.
2889 : */
2890 : void
2891 GNC 105 : ReorderBufferAbort(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn,
2892 : TimestampTz abort_time)
2893 : {
2894 ECB : ReorderBufferTXN *txn;
2895 :
2896 GIC 105 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2897 : false);
2898 ECB :
2899 : /* unknown, nothing to remove */
2900 CBC 105 : if (txn == NULL)
2901 LBC 0 : return;
2902 ECB :
2903 GNC 105 : txn->xact_time.abort_time = abort_time;
2904 :
2905 : /* For streamed transactions notify the remote node about the abort. */
2906 CBC 105 : if (rbtxn_is_streamed(txn))
2907 ECB : {
2908 GIC 29 : rb->stream_abort(rb, txn, lsn);
2909 ECB :
2910 : /*
2911 : * We might have decoded changes for this transaction that could load
2912 : * the cache as per the current transaction's view (consider DDL's
2913 : * happened in this transaction). We don't want the decoding of future
2914 : * transactions to use those cache entries so execute invalidations.
2915 : */
2916 GIC 29 : if (txn->ninvalidations > 0)
2917 UIC 0 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
2918 : txn->invalidations);
2919 : }
2920 :
2921 : /* cosmetic... */
2922 GIC 105 : txn->final_lsn = lsn;
2923 :
2924 : /* remove potential on-disk data, and deallocate */
2925 105 : ReorderBufferCleanupTXN(rb, txn);
2926 : }
2927 :
2928 : /*
2929 : * Abort all transactions that aren't actually running anymore because the
2930 ECB : * server restarted.
2931 : *
2932 : * NB: These really have to be transactions that have aborted due to a server
2933 : * crash/immediate restart, as we don't deal with invalidations here.
2934 : */
2935 : void
2936 GIC 1024 : ReorderBufferAbortOld(ReorderBuffer *rb, TransactionId oldestRunningXid)
2937 : {
2938 : dlist_mutable_iter it;
2939 ECB :
2940 EUB : /*
2941 : * Iterate through all (potential) toplevel TXNs and abort all that are
2942 ECB : * older than what possibly can be running. Once we've found the first
2943 : * that is alive we stop, there might be some that acquired an xid earlier
2944 : * but started writing later, but it's unlikely and they will be cleaned
2945 : * up in a later call to this function.
2946 : */
2947 CBC 1031 : dlist_foreach_modify(it, &rb->toplevel_by_lsn)
2948 : {
2949 : ReorderBufferTXN *txn;
2950 :
2951 GIC 41 : txn = dlist_container(ReorderBufferTXN, node, it.cur);
2952 :
2953 41 : if (TransactionIdPrecedes(txn->xid, oldestRunningXid))
2954 : {
2955 CBC 7 : elog(DEBUG2, "aborting old transaction %u", txn->xid);
2956 EUB :
2957 : /* Notify the remote node about the crash/immediate restart. */
2958 GIC 7 : if (rbtxn_is_streamed(txn))
2959 UIC 0 : rb->stream_abort(rb, txn, InvalidXLogRecPtr);
2960 :
2961 ECB : /* remove potential on-disk data, and deallocate this tx */
2962 GIC 7 : ReorderBufferCleanupTXN(rb, txn);
2963 : }
2964 ECB : else
2965 GIC 34 : return;
2966 : }
2967 : }
2968 :
2969 : /*
2970 : * Forget the contents of a transaction if we aren't interested in its
2971 : * contents. Needs to be first called for subtransactions and then for the
2972 : * toplevel xid.
2973 : *
2974 : * This is significantly different to ReorderBufferAbort() because
2975 ECB : * transactions that have committed need to be treated differently from aborted
2976 : * ones since they may have modified the catalog.
2977 : *
2978 : * Note that this is only allowed to be called in the moment a transaction
2979 : * commit has just been read, not earlier; otherwise later records referring
2980 : * to this xid might re-create the transaction incompletely.
2981 : */
2982 : void
2983 GIC 2429 : ReorderBufferForget(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
2984 : {
2985 : ReorderBufferTXN *txn;
2986 ECB :
2987 GIC 2429 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
2988 : false);
2989 :
2990 ECB : /* unknown, nothing to forget */
2991 GIC 2429 : if (txn == NULL)
2992 CBC 561 : return;
2993 :
2994 : /* this transaction mustn't be streamed */
2995 GNC 1868 : Assert(!rbtxn_is_streamed(txn));
2996 ECB :
2997 EUB : /* cosmetic... */
2998 GIC 1868 : txn->final_lsn = lsn;
2999 :
3000 ECB : /*
3001 : * Process cache invalidation messages if there are any. Even if we're not
3002 : * interested in the transaction's contents, it could have manipulated the
3003 : * catalog and we need to update the caches according to that.
3004 : */
3005 GIC 1868 : if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
3006 541 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3007 : txn->invalidations);
3008 : else
3009 1327 : Assert(txn->ninvalidations == 0);
3010 :
3011 : /* remove potential on-disk data, and deallocate */
3012 1868 : ReorderBufferCleanupTXN(rb, txn);
3013 : }
3014 :
3015 : /*
3016 : * Invalidate cache for those transactions that need to be skipped just in case
3017 : * catalogs were manipulated as part of the transaction.
3018 : *
3019 : * Note that this is a special-purpose function for prepared transactions where
3020 : * we don't want to clean up the TXN even when we decide to skip it. See
3021 ECB : * DecodePrepare.
3022 : */
3023 : void
3024 GIC 85 : ReorderBufferInvalidate(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3025 ECB : {
3026 : ReorderBufferTXN *txn;
3027 :
3028 GIC 85 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3029 ECB : false);
3030 :
3031 : /* unknown, nothing to do */
3032 GIC 85 : if (txn == NULL)
3033 LBC 0 : return;
3034 :
3035 : /*
3036 ECB : * Process cache invalidation messages if there are any. Even if we're not
3037 : * interested in the transaction's contents, it could have manipulated the
3038 : * catalog and we need to update the caches according to that.
3039 : */
3040 GIC 85 : if (txn->base_snapshot != NULL && txn->ninvalidations > 0)
3041 23 : ReorderBufferImmediateInvalidation(rb, txn->ninvalidations,
3042 : txn->invalidations);
3043 ECB : else
3044 CBC 62 : Assert(txn->ninvalidations == 0);
3045 : }
3046 :
3047 ECB :
3048 : /*
3049 : * Execute invalidations happening outside the context of a decoded
3050 : * transaction. That currently happens either for xid-less commits
3051 : * (cf. RecordTransactionCommit()) or for invalidations in uninteresting
3052 : * transactions (via ReorderBufferForget()).
3053 : */
3054 : void
3055 GIC 566 : ReorderBufferImmediateInvalidation(ReorderBuffer *rb, uint32 ninvalidations,
3056 : SharedInvalidationMessage *invalidations)
3057 : {
3058 566 : bool use_subtxn = IsTransactionOrTransactionBlock();
3059 : int i;
3060 :
3061 566 : if (use_subtxn)
3062 CBC 397 : BeginInternalSubTransaction("replay");
3063 :
3064 : /*
3065 : * Force invalidations to happen outside of a valid transaction - that way
3066 ECB : * entries will just be marked as invalid without accessing the catalog.
3067 : * That's advantageous because we don't need to setup the full state
3068 : * necessary for catalog access.
3069 : */
3070 CBC 566 : if (use_subtxn)
3071 GBC 397 : AbortCurrentTransaction();
3072 :
3073 GIC 24954 : for (i = 0; i < ninvalidations; i++)
3074 24388 : LocalExecuteInvalidationMessage(&invalidations[i]);
3075 :
3076 566 : if (use_subtxn)
3077 397 : RollbackAndReleaseCurrentSubTransaction();
3078 CBC 566 : }
3079 ECB :
3080 : /*
3081 : * Tell reorderbuffer about an xid seen in the WAL stream. Has to be called at
3082 : * least once for every xid in XLogRecord->xl_xid (other places in records
3083 : * may, but do not have to be passed through here).
3084 : *
3085 : * Reorderbuffer keeps some datastructures about transactions in LSN order,
3086 : * for efficiency. To do that it has to know about when transactions are seen
3087 : * first in the WAL. As many types of records are not actually interesting for
3088 : * logical decoding, they do not necessarily pass though here.
3089 : */
3090 : void
3091 GIC 2465084 : ReorderBufferProcessXid(ReorderBuffer *rb, TransactionId xid, XLogRecPtr lsn)
3092 : {
3093 ECB : /* many records won't have an xid assigned, centralize check here */
3094 GIC 2465084 : if (xid != InvalidTransactionId)
3095 2463336 : ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3096 CBC 2465084 : }
3097 :
3098 : /*
3099 ECB : * Add a new snapshot to this transaction that may only used after lsn 'lsn'
3100 : * because the previous snapshot doesn't describe the catalog correctly for
3101 : * following rows.
3102 : */
3103 : void
3104 GIC 1002 : ReorderBufferAddSnapshot(ReorderBuffer *rb, TransactionId xid,
3105 : XLogRecPtr lsn, Snapshot snap)
3106 : {
3107 1002 : ReorderBufferChange *change = ReorderBufferGetChange(rb);
3108 ECB :
3109 CBC 1002 : change->data.snapshot = snap;
3110 GIC 1002 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT;
3111 ECB :
3112 CBC 1002 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3113 GIC 1002 : }
3114 ECB :
3115 : /*
3116 : * Set up the transaction's base snapshot.
3117 : *
3118 : * If we know that xid is a subtransaction, set the base snapshot on the
3119 : * top-level transaction instead.
3120 : */
3121 : void
3122 GIC 2607 : ReorderBufferSetBaseSnapshot(ReorderBuffer *rb, TransactionId xid,
3123 : XLogRecPtr lsn, Snapshot snap)
3124 : {
3125 : ReorderBufferTXN *txn;
3126 : bool is_new;
3127 :
3128 GNC 2607 : Assert(snap != NULL);
3129 ECB :
3130 : /*
3131 : * Fetch the transaction to operate on. If we know it's a subtransaction,
3132 : * operate on its top-level transaction instead.
3133 : */
3134 CBC 2607 : txn = ReorderBufferTXNByXid(rb, xid, true, &is_new, lsn, true);
3135 GIC 2607 : if (rbtxn_is_known_subxact(txn))
3136 120 : txn = ReorderBufferTXNByXid(rb, txn->toplevel_xid, false,
3137 : NULL, InvalidXLogRecPtr, false);
3138 2607 : Assert(txn->base_snapshot == NULL);
3139 :
3140 2607 : txn->base_snapshot = snap;
3141 2607 : txn->base_snapshot_lsn = lsn;
3142 CBC 2607 : dlist_push_tail(&rb->txns_by_base_snapshot_lsn, &txn->base_snapshot_node);
3143 :
3144 GIC 2607 : AssertTXNLsnOrder(rb);
3145 CBC 2607 : }
3146 :
3147 ECB : /*
3148 : * Access the catalog with this CommandId at this point in the changestream.
3149 : *
3150 : * May only be called for command ids > 1
3151 : */
3152 : void
3153 GIC 22808 : ReorderBufferAddNewCommandId(ReorderBuffer *rb, TransactionId xid,
3154 : XLogRecPtr lsn, CommandId cid)
3155 : {
3156 22808 : ReorderBufferChange *change = ReorderBufferGetChange(rb);
3157 :
3158 22808 : change->data.command_id = cid;
3159 22808 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID;
3160 ECB :
3161 GIC 22808 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3162 22808 : }
3163 :
3164 : /*
3165 : * Update memory counters to account for the new or removed change.
3166 ECB : *
3167 : * We update two counters - in the reorder buffer, and in the transaction
3168 : * containing the change. The reorder buffer counter allows us to quickly
3169 : * decide if we reached the memory limit, the transaction counter allows
3170 : * us to quickly pick the largest transaction for eviction.
3171 : *
3172 : * When streaming is enabled, we need to update the toplevel transaction
3173 : * counters instead - we don't really care about subtransactions as we
3174 : * can't stream them individually anyway, and we only pick toplevel
3175 : * transactions for eviction. So only toplevel transactions matter.
3176 : */
3177 : static void
3178 CBC 3807299 : ReorderBufferChangeMemoryUpdate(ReorderBuffer *rb,
3179 ECB : ReorderBufferChange *change,
3180 : bool addition, Size sz)
3181 : {
3182 : ReorderBufferTXN *txn;
3183 : ReorderBufferTXN *toptxn;
3184 :
3185 GIC 3807299 : Assert(change->txn);
3186 :
3187 : /*
3188 : * Ignore tuple CID changes, because those are not evicted when reaching
3189 : * memory limit. So we just don't count them, because it might easily
3190 : * trigger a pointless attempt to spill.
3191 ECB : */
3192 GIC 3807299 : if (change->action == REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID)
3193 22692 : return;
3194 ECB :
3195 GIC 3784607 : txn = change->txn;
3196 ECB :
3197 : /*
3198 : * Update the total size in top level as well. This is later used to
3199 : * compute the decoding stats.
3200 : */
3201 GNC 3784607 : toptxn = rbtxn_get_toptxn(txn);
3202 :
3203 GIC 3784607 : if (addition)
3204 : {
3205 1892376 : txn->size += sz;
3206 1892376 : rb->size += sz;
3207 :
3208 : /* Update the total size in the top transaction. */
3209 1892376 : toptxn->total_size += sz;
3210 : }
3211 : else
3212 : {
3213 CBC 1892231 : Assert((rb->size >= sz) && (txn->size >= sz));
3214 GIC 1892231 : txn->size -= sz;
3215 1892231 : rb->size -= sz;
3216 :
3217 : /* Update the total size in the top transaction. */
3218 1892231 : toptxn->total_size -= sz;
3219 : }
3220 ECB :
3221 GIC 3784607 : Assert(txn->size <= rb->size);
3222 : }
3223 :
3224 : /*
3225 : * Add new (relfilelocator, tid) -> (cmin, cmax) mappings.
3226 : *
3227 ECB : * We do not include this change type in memory accounting, because we
3228 : * keep CIDs in a separate list and do not evict them when reaching
3229 : * the memory limit.
3230 : */
3231 : void
3232 GIC 22808 : ReorderBufferAddNewTupleCids(ReorderBuffer *rb, TransactionId xid,
3233 : XLogRecPtr lsn, RelFileLocator locator,
3234 : ItemPointerData tid, CommandId cmin,
3235 : CommandId cmax, CommandId combocid)
3236 ECB : {
3237 GIC 22808 : ReorderBufferChange *change = ReorderBufferGetChange(rb);
3238 ECB : ReorderBufferTXN *txn;
3239 :
3240 CBC 22808 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3241 ECB :
3242 GNC 22808 : change->data.tuplecid.locator = locator;
3243 GIC 22808 : change->data.tuplecid.tid = tid;
3244 CBC 22808 : change->data.tuplecid.cmin = cmin;
3245 GIC 22808 : change->data.tuplecid.cmax = cmax;
3246 22808 : change->data.tuplecid.combocid = combocid;
3247 22808 : change->lsn = lsn;
3248 CBC 22808 : change->txn = txn;
3249 22808 : change->action = REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID;
3250 ECB :
3251 GIC 22808 : dlist_push_tail(&txn->tuplecids, &change->node);
3252 22808 : txn->ntuplecids++;
3253 CBC 22808 : }
3254 :
3255 : /*
3256 ECB : * Accumulate the invalidations for executing them later.
3257 : *
3258 : * This needs to be called for each XLOG_XACT_INVALIDATIONS message and
3259 : * accumulates all the invalidation messages in the toplevel transaction, if
3260 : * available, otherwise in the current transaction, as well as in the form of
3261 : * change in reorder buffer. We require to record it in form of the change
3262 : * so that we can execute only the required invalidations instead of executing
3263 : * all the invalidations on each CommandId increment. We also need to
3264 : * accumulate these in the txn buffer because in some cases where we skip
3265 : * processing the transaction (see ReorderBufferForget), we need to execute
3266 : * all the invalidations together.
3267 : */
3268 : void
3269 GIC 4559 : ReorderBufferAddInvalidations(ReorderBuffer *rb, TransactionId xid,
3270 : XLogRecPtr lsn, Size nmsgs,
3271 : SharedInvalidationMessage *msgs)
3272 ECB : {
3273 : ReorderBufferTXN *txn;
3274 : MemoryContext oldcontext;
3275 : ReorderBufferChange *change;
3276 :
3277 CBC 4559 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3278 ECB :
3279 CBC 4559 : oldcontext = MemoryContextSwitchTo(rb->context);
3280 ECB :
3281 : /*
3282 : * Collect all the invalidations under the top transaction, if available,
3283 : * so that we can execute them all together. See comments atop this
3284 : * function.
3285 : */
3286 GNC 4559 : txn = rbtxn_get_toptxn(txn);
3287 ECB :
3288 GIC 4559 : Assert(nmsgs > 0);
3289 :
3290 : /* Accumulate invalidations. */
3291 4559 : if (txn->ninvalidations == 0)
3292 : {
3293 995 : txn->ninvalidations = nmsgs;
3294 995 : txn->invalidations = (SharedInvalidationMessage *)
3295 995 : palloc(sizeof(SharedInvalidationMessage) * nmsgs);
3296 995 : memcpy(txn->invalidations, msgs,
3297 : sizeof(SharedInvalidationMessage) * nmsgs);
3298 : }
3299 : else
3300 : {
3301 3564 : txn->invalidations = (SharedInvalidationMessage *)
3302 3564 : repalloc(txn->invalidations, sizeof(SharedInvalidationMessage) *
3303 CBC 3564 : (txn->ninvalidations + nmsgs));
3304 :
3305 GIC 3564 : memcpy(txn->invalidations + txn->ninvalidations, msgs,
3306 : nmsgs * sizeof(SharedInvalidationMessage));
3307 3564 : txn->ninvalidations += nmsgs;
3308 : }
3309 :
3310 4559 : change = ReorderBufferGetChange(rb);
3311 CBC 4559 : change->action = REORDER_BUFFER_CHANGE_INVALIDATION;
3312 GIC 4559 : change->data.inval.ninvalidations = nmsgs;
3313 CBC 4559 : change->data.inval.invalidations = (SharedInvalidationMessage *)
3314 GIC 4559 : palloc(sizeof(SharedInvalidationMessage) * nmsgs);
3315 4559 : memcpy(change->data.inval.invalidations, msgs,
3316 : sizeof(SharedInvalidationMessage) * nmsgs);
3317 :
3318 4559 : ReorderBufferQueueChange(rb, xid, lsn, change, false);
3319 :
3320 CBC 4559 : MemoryContextSwitchTo(oldcontext);
3321 GIC 4559 : }
3322 ECB :
3323 : /*
3324 : * Apply all invalidations we know. Possibly we only need parts at this point
3325 : * in the changestream but we don't know which those are.
3326 : */
3327 : static void
3328 CBC 3577 : ReorderBufferExecuteInvalidations(uint32 nmsgs, SharedInvalidationMessage *msgs)
3329 ECB : {
3330 : int i;
3331 :
3332 GIC 37885 : for (i = 0; i < nmsgs; i++)
3333 34308 : LocalExecuteInvalidationMessage(&msgs[i]);
3334 3577 : }
3335 ECB :
3336 : /*
3337 : * Mark a transaction as containing catalog changes
3338 : */
3339 : void
3340 GIC 28337 : ReorderBufferXidSetCatalogChanges(ReorderBuffer *rb, TransactionId xid,
3341 ECB : XLogRecPtr lsn)
3342 : {
3343 : ReorderBufferTXN *txn;
3344 :
3345 CBC 28337 : txn = ReorderBufferTXNByXid(rb, xid, true, NULL, lsn, true);
3346 ECB :
3347 GNC 28337 : if (!rbtxn_has_catalog_changes(txn))
3348 : {
3349 1022 : txn->txn_flags |= RBTXN_HAS_CATALOG_CHANGES;
3350 1022 : dclist_push_tail(&rb->catchange_txns, &txn->catchange_node);
3351 : }
3352 ECB :
3353 : /*
3354 : * Mark top-level transaction as having catalog changes too if one of its
3355 : * children has so that the ReorderBufferBuildTupleCidHash can
3356 : * conveniently check just top-level transaction and decide whether to
3357 : * build the hash table or not.
3358 : */
3359 GNC 28337 : if (rbtxn_is_subtxn(txn))
3360 : {
3361 903 : ReorderBufferTXN *toptxn = rbtxn_get_toptxn(txn);
3362 :
3363 903 : if (!rbtxn_has_catalog_changes(toptxn))
3364 : {
3365 18 : toptxn->txn_flags |= RBTXN_HAS_CATALOG_CHANGES;
3366 18 : dclist_push_tail(&rb->catchange_txns, &toptxn->catchange_node);
3367 : }
3368 : }
3369 28337 : }
3370 :
3371 : /*
3372 : * Return palloc'ed array of the transactions that have changed catalogs.
3373 : * The returned array is sorted in xidComparator order.
3374 : *
3375 : * The caller must free the returned array when done with it.
3376 : */
3377 : TransactionId *
3378 208 : ReorderBufferGetCatalogChangesXacts(ReorderBuffer *rb)
3379 : {
3380 : dlist_iter iter;
3381 208 : TransactionId *xids = NULL;
3382 208 : size_t xcnt = 0;
3383 :
3384 : /* Quick return if the list is empty */
3385 208 : if (dclist_count(&rb->catchange_txns) == 0)
3386 202 : return NULL;
3387 :
3388 : /* Initialize XID array */
3389 6 : xids = (TransactionId *) palloc(sizeof(TransactionId) *
3390 6 : dclist_count(&rb->catchange_txns));
3391 14 : dclist_foreach(iter, &rb->catchange_txns)
3392 : {
3393 8 : ReorderBufferTXN *txn = dclist_container(ReorderBufferTXN,
3394 : catchange_node,
3395 : iter.cur);
3396 :
3397 8 : Assert(rbtxn_has_catalog_changes(txn));
3398 :
3399 8 : xids[xcnt++] = txn->xid;
3400 : }
3401 :
3402 6 : qsort(xids, xcnt, sizeof(TransactionId), xidComparator);
3403 :
3404 6 : Assert(xcnt == dclist_count(&rb->catchange_txns));
3405 6 : return xids;
3406 : }
3407 :
3408 : /*
3409 : * Query whether a transaction is already *known* to contain catalog
3410 : * changes. This can be wrong until directly before the commit!
3411 ECB : */
3412 : bool
3413 GIC 3790 : ReorderBufferXidHasCatalogChanges(ReorderBuffer *rb, TransactionId xid)
3414 : {
3415 ECB : ReorderBufferTXN *txn;
3416 :
3417 CBC 3790 : txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
3418 : false);
3419 GIC 3790 : if (txn == NULL)
3420 646 : return false;
3421 :
3422 3144 : return rbtxn_has_catalog_changes(txn);
3423 ECB : }
3424 :
3425 : /*
3426 : * ReorderBufferXidHasBaseSnapshot
3427 : * Have we already set the base snapshot for the given txn/subtxn?
3428 : */
3429 : bool
3430 CBC 1696415 : ReorderBufferXidHasBaseSnapshot(ReorderBuffer *rb, TransactionId xid)
3431 : {
3432 ECB : ReorderBufferTXN *txn;
3433 :
3434 GIC 1696415 : txn = ReorderBufferTXNByXid(rb, xid, false,
3435 : NULL, InvalidXLogRecPtr, false);
3436 :
3437 : /* transaction isn't known yet, ergo no snapshot */
3438 1696415 : if (txn == NULL)
3439 3 : return false;
3440 :
3441 : /* a known subtxn? operate on top-level txn instead */
3442 CBC 1696412 : if (rbtxn_is_known_subxact(txn))
3443 GIC 491936 : txn = ReorderBufferTXNByXid(rb, txn->toplevel_xid, false,
3444 ECB : NULL, InvalidXLogRecPtr, false);
3445 :
3446 CBC 1696412 : return txn->base_snapshot != NULL;
3447 : }
3448 ECB :
3449 :
3450 : /*
3451 : * ---------------------------------------
3452 : * Disk serialization support
3453 : * ---------------------------------------
3454 : */
3455 :
3456 : /*
3457 : * Ensure the IO buffer is >= sz.
3458 : */
3459 : static void
3460 GIC 3309901 : ReorderBufferSerializeReserve(ReorderBuffer *rb, Size sz)
3461 ECB : {
3462 GIC 3309901 : if (!rb->outbufsize)
3463 : {
3464 CBC 49 : rb->outbuf = MemoryContextAlloc(rb->context, sz);
3465 49 : rb->outbufsize = sz;
3466 : }
3467 GIC 3309852 : else if (rb->outbufsize < sz)
3468 ECB : {
3469 CBC 300 : rb->outbuf = repalloc(rb->outbuf, sz);
3470 GIC 300 : rb->outbufsize = sz;
3471 : }
3472 CBC 3309901 : }
3473 ECB :
3474 : /*
3475 : * Find the largest transaction (toplevel or subxact) to evict (spill to disk).
3476 : *
3477 : * XXX With many subtransactions this might be quite slow, because we'll have
3478 : * to walk through all of them. There are some options how we could improve
3479 : * that: (a) maintain some secondary structure with transactions sorted by
3480 : * amount of changes, (b) not looking for the entirely largest transaction,
3481 : * but e.g. for transaction using at least some fraction of the memory limit,
3482 : * and (c) evicting multiple transactions at once, e.g. to free a given portion
3483 : * of the memory limit (e.g. 50%).
3484 : */
3485 : static ReorderBufferTXN *
3486 GIC 3698 : ReorderBufferLargestTXN(ReorderBuffer *rb)
3487 ECB : {
3488 : HASH_SEQ_STATUS hash_seq;
3489 : ReorderBufferTXNByIdEnt *ent;
3490 GIC 3698 : ReorderBufferTXN *largest = NULL;
3491 :
3492 3698 : hash_seq_init(&hash_seq, rb->by_txn);
3493 9509 : while ((ent = hash_seq_search(&hash_seq)) != NULL)
3494 : {
3495 5811 : ReorderBufferTXN *txn = ent->txn;
3496 ECB :
3497 : /* if the current transaction is larger, remember it */
3498 GIC 5811 : if ((!largest) || (txn->size > largest->size))
3499 4756 : largest = txn;
3500 ECB : }
3501 :
3502 CBC 3698 : Assert(largest);
3503 3698 : Assert(largest->size > 0);
3504 GIC 3698 : Assert(largest->size <= rb->size);
3505 ECB :
3506 GIC 3698 : return largest;
3507 : }
3508 :
3509 : /*
3510 : * Find the largest streamable toplevel transaction to evict (by streaming).
3511 : *
3512 : * This can be seen as an optimized version of ReorderBufferLargestTXN, which
3513 ECB : * should give us the same transaction (because we don't update memory account
3514 : * for subtransaction with streaming, so it's always 0). But we can simply
3515 : * iterate over the limited number of toplevel transactions that have a base
3516 : * snapshot. There is no use of selecting a transaction that doesn't have base
3517 : * snapshot because we don't decode such transactions. Also, we do not select
3518 : * the transaction which doesn't have any streamable change.
3519 : *
3520 : * Note that, we skip transactions that contains incomplete changes. There
3521 : * is a scope of optimization here such that we can select the largest
3522 : * transaction which has incomplete changes. But that will make the code and
3523 : * design quite complex and that might not be worth the benefit. If we plan to
3524 : * stream the transactions that contains incomplete changes then we need to
3525 : * find a way to partially stream/truncate the transaction changes in-memory
3526 : * and build a mechanism to partially truncate the spilled files.
3527 : * Additionally, whenever we partially stream the transaction we need to
3528 : * maintain the last streamed lsn and next time we need to restore from that
3529 : * segment and the offset in WAL. As we stream the changes from the top
3530 : * transaction and restore them subtransaction wise, we need to even remember
3531 : * the subxact from where we streamed the last change.
3532 : */
3533 : static ReorderBufferTXN *
3534 GNC 637 : ReorderBufferLargestStreamableTopTXN(ReorderBuffer *rb)
3535 : {
3536 : dlist_iter iter;
3537 GIC 637 : Size largest_size = 0;
3538 637 : ReorderBufferTXN *largest = NULL;
3539 :
3540 : /* Find the largest top-level transaction having a base snapshot. */
3541 1387 : dlist_foreach(iter, &rb->txns_by_base_snapshot_lsn)
3542 : {
3543 : ReorderBufferTXN *txn;
3544 ECB :
3545 GIC 750 : txn = dlist_container(ReorderBufferTXN, base_snapshot_node, iter.cur);
3546 ECB :
3547 : /* must not be a subtxn */
3548 CBC 750 : Assert(!rbtxn_is_known_subxact(txn));
3549 ECB : /* base_snapshot must be set */
3550 GIC 750 : Assert(txn->base_snapshot != NULL);
3551 ECB :
3552 GIC 750 : if ((largest == NULL || txn->total_size > largest_size) &&
3553 GNC 750 : (txn->total_size > 0) && !(rbtxn_has_partial_change(txn)) &&
3554 659 : rbtxn_has_streamable_change(txn))
3555 ECB : {
3556 GIC 659 : largest = txn;
3557 CBC 659 : largest_size = txn->total_size;
3558 : }
3559 : }
3560 :
3561 GIC 637 : return largest;
3562 : }
3563 :
3564 : /*
3565 : * Check whether the logical_decoding_work_mem limit was reached, and if yes
3566 : * pick the largest (sub)transaction at-a-time to evict and spill its changes to
3567 : * disk or send to the output plugin until we reach under the memory limit.
3568 : *
3569 : * If logical_replication_mode is set to "immediate", stream or serialize the
3570 : * changes immediately.
3571 : *
3572 : * XXX At this point we select the transactions until we reach under the memory
3573 : * limit, but we might also adapt a more elaborate eviction strategy - for example
3574 ECB : * evicting enough transactions to free certain fraction (e.g. 50%) of the memory
3575 : * limit.
3576 : */
3577 : static void
3578 CBC 1713655 : ReorderBufferCheckMemoryLimit(ReorderBuffer *rb)
3579 : {
3580 ECB : ReorderBufferTXN *txn;
3581 :
3582 : /*
3583 : * Bail out if logical_replication_mode is buffered and we haven't exceeded
3584 : * the memory limit.
3585 : */
3586 GNC 1713655 : if (logical_replication_mode == LOGICAL_REP_MODE_BUFFERED &&
3587 1713241 : rb->size < logical_decoding_work_mem * 1024L)
3588 GIC 1709364 : return;
3589 :
3590 ECB : /*
3591 : * If logical_replication_mode is immediate, loop until there's no change.
3592 : * Otherwise, loop until we reach under the memory limit. One might think
3593 : * that just by evicting the largest (sub)transaction we will come under
3594 : * the memory limit based on assumption that the selected transaction is
3595 : * at least as large as the most recent change (which caused us to go over
3596 : * the memory limit). However, that is not true because a user can reduce
3597 : * the logical_decoding_work_mem to a smaller value before the most recent
3598 : * change.
3599 : */
3600 GNC 8582 : while (rb->size >= logical_decoding_work_mem * 1024L ||
3601 4705 : (logical_replication_mode == LOGICAL_REP_MODE_IMMEDIATE &&
3602 828 : rb->size > 0))
3603 : {
3604 : /*
3605 : * Pick the largest transaction (or subtransaction) and evict it from
3606 : * memory by streaming, if possible. Otherwise, spill to disk.
3607 : */
3608 GIC 4928 : if (ReorderBufferCanStartStreaming(rb) &&
3609 GNC 637 : (txn = ReorderBufferLargestStreamableTopTXN(rb)) != NULL)
3610 : {
3611 : /* we know there has to be one, because the size is not zero */
3612 593 : Assert(txn && rbtxn_is_toptxn(txn));
3613 GIC 593 : Assert(txn->total_size > 0);
3614 593 : Assert(rb->size >= txn->total_size);
3615 :
3616 593 : ReorderBufferStreamTXN(rb, txn);
3617 : }
3618 : else
3619 : {
3620 : /*
3621 : * Pick the largest transaction (or subtransaction) and evict it
3622 : * from memory by serializing it to disk.
3623 : */
3624 3698 : txn = ReorderBufferLargestTXN(rb);
3625 :
3626 : /* we know there has to be one, because the size is not zero */
3627 3698 : Assert(txn);
3628 3698 : Assert(txn->size > 0);
3629 CBC 3698 : Assert(rb->size >= txn->size);
3630 :
3631 GIC 3698 : ReorderBufferSerializeTXN(rb, txn);
3632 ECB : }
3633 :
3634 : /*
3635 : * After eviction, the transaction should have no entries in memory,
3636 : * and should use 0 bytes for changes.
3637 : */
3638 GIC 4291 : Assert(txn->size == 0);
3639 4291 : Assert(txn->nentries_mem == 0);
3640 ECB : }
3641 :
3642 : /* We must be under the memory limit now. */
3643 CBC 4291 : Assert(rb->size < logical_decoding_work_mem * 1024L);
3644 : }
3645 ECB :
3646 : /*
3647 : * Spill data of a large transaction (and its subtransactions) to disk.
3648 : */
3649 : static void
3650 GIC 4003 : ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
3651 ECB : {
3652 : dlist_iter subtxn_i;
3653 : dlist_mutable_iter change_i;
3654 GIC 4003 : int fd = -1;
3655 4003 : XLogSegNo curOpenSegNo = 0;
3656 CBC 4003 : Size spilled = 0;
3657 GIC 4003 : Size size = txn->size;
3658 :
3659 4003 : elog(DEBUG2, "spill %u changes in XID %u to disk",
3660 : (uint32) txn->nentries_mem, txn->xid);
3661 :
3662 : /* do the same to all child TXs */
3663 4271 : dlist_foreach(subtxn_i, &txn->subtxns)
3664 : {
3665 : ReorderBufferTXN *subtxn;
3666 :
3667 268 : subtxn = dlist_container(ReorderBufferTXN, node, subtxn_i.cur);
3668 268 : ReorderBufferSerializeTXN(rb, subtxn);
3669 : }
3670 :
3671 : /* serialize changestream */
3672 1489026 : dlist_foreach_modify(change_i, &txn->changes)
3673 ECB : {
3674 : ReorderBufferChange *change;
3675 :
3676 GIC 1485023 : change = dlist_container(ReorderBufferChange, node, change_i.cur);
3677 :
3678 : /*
3679 : * store in segment in which it belongs by start lsn, don't split over
3680 : * multiple segments tho
3681 ECB : */
3682 CBC 1485023 : if (fd == -1 ||
3683 1481271 : !XLByteInSeg(change->lsn, curOpenSegNo, wal_segment_size))
3684 : {
3685 : char path[MAXPGPATH];
3686 :
3687 GIC 3756 : if (fd != -1)
3688 4 : CloseTransientFile(fd);
3689 :
3690 3756 : XLByteToSeg(change->lsn, curOpenSegNo, wal_segment_size);
3691 :
3692 : /*
3693 : * No need to care about TLIs here, only used during a single run,
3694 : * so each LSN only maps to a specific WAL record.
3695 ECB : */
3696 CBC 3756 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid,
3697 ECB : curOpenSegNo);
3698 :
3699 : /* open segment, create it if necessary */
3700 GIC 3756 : fd = OpenTransientFile(path,
3701 : O_CREAT | O_WRONLY | O_APPEND | PG_BINARY);
3702 :
3703 CBC 3756 : if (fd < 0)
3704 LBC 0 : ereport(ERROR,
3705 : (errcode_for_file_access(),
3706 : errmsg("could not open file \"%s\": %m", path)));
3707 ECB : }
3708 :
3709 CBC 1485023 : ReorderBufferSerializeChange(rb, txn, fd, change);
3710 GIC 1485023 : dlist_delete(&change->node);
3711 CBC 1485023 : ReorderBufferReturnChange(rb, change, true);
3712 :
3713 GIC 1485023 : spilled++;
3714 : }
3715 :
3716 : /* update the statistics iff we have spilled anything */
3717 4003 : if (spilled)
3718 : {
3719 CBC 3752 : rb->spillCount += 1;
3720 GIC 3752 : rb->spillBytes += size;
3721 :
3722 ECB : /* don't consider already serialized transactions */
3723 CBC 3752 : rb->spillTxns += (rbtxn_is_serialized(txn) || rbtxn_is_serialized_clear(txn)) ? 0 : 1;
3724 ECB :
3725 : /* update the decoding stats */
3726 CBC 3752 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
3727 : }
3728 :
3729 GIC 4003 : Assert(spilled == txn->nentries_mem);
3730 4003 : Assert(dlist_is_empty(&txn->changes));
3731 4003 : txn->nentries_mem = 0;
3732 4003 : txn->txn_flags |= RBTXN_IS_SERIALIZED;
3733 ECB :
3734 CBC 4003 : if (fd != -1)
3735 GIC 3752 : CloseTransientFile(fd);
3736 4003 : }
3737 :
3738 ECB : /*
3739 : * Serialize individual change to disk.
3740 : */
3741 : static void
3742 GIC 1485023 : ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
3743 : int fd, ReorderBufferChange *change)
3744 : {
3745 ECB : ReorderBufferDiskChange *ondisk;
3746 GIC 1485023 : Size sz = sizeof(ReorderBufferDiskChange);
3747 :
3748 1485023 : ReorderBufferSerializeReserve(rb, sz);
3749 ECB :
3750 CBC 1485023 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
3751 1485023 : memcpy(&ondisk->change, change, sizeof(ReorderBufferChange));
3752 ECB :
3753 GIC 1485023 : switch (change->action)
3754 ECB : {
3755 : /* fall through these, they're all similar enough */
3756 GIC 1467750 : case REORDER_BUFFER_CHANGE_INSERT:
3757 : case REORDER_BUFFER_CHANGE_UPDATE:
3758 ECB : case REORDER_BUFFER_CHANGE_DELETE:
3759 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
3760 : {
3761 : char *data;
3762 : ReorderBufferTupleBuf *oldtup,
3763 : *newtup;
3764 GIC 1467750 : Size oldlen = 0;
3765 1467750 : Size newlen = 0;
3766 :
3767 CBC 1467750 : oldtup = change->data.tp.oldtuple;
3768 GIC 1467750 : newtup = change->data.tp.newtuple;
3769 :
3770 1467750 : if (oldtup)
3771 ECB : {
3772 GIC 160127 : sz += sizeof(HeapTupleData);
3773 160127 : oldlen = oldtup->tuple.t_len;
3774 160127 : sz += oldlen;
3775 : }
3776 :
3777 CBC 1467750 : if (newtup)
3778 ECB : {
3779 GIC 1253921 : sz += sizeof(HeapTupleData);
3780 1253921 : newlen = newtup->tuple.t_len;
3781 1253921 : sz += newlen;
3782 ECB : }
3783 :
3784 : /* make sure we have enough space */
3785 CBC 1467750 : ReorderBufferSerializeReserve(rb, sz);
3786 :
3787 GIC 1467750 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
3788 : /* might have been reallocated above */
3789 1467750 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
3790 :
3791 CBC 1467750 : if (oldlen)
3792 : {
3793 GIC 160127 : memcpy(data, &oldtup->tuple, sizeof(HeapTupleData));
3794 160127 : data += sizeof(HeapTupleData);
3795 ECB :
3796 GIC 160127 : memcpy(data, oldtup->tuple.t_data, oldlen);
3797 160127 : data += oldlen;
3798 ECB : }
3799 EUB :
3800 GIC 1467750 : if (newlen)
3801 : {
3802 1253921 : memcpy(data, &newtup->tuple, sizeof(HeapTupleData));
3803 1253921 : data += sizeof(HeapTupleData);
3804 ECB :
3805 CBC 1253921 : memcpy(data, newtup->tuple.t_data, newlen);
3806 1253921 : data += newlen;
3807 : }
3808 1467750 : break;
3809 : }
3810 GIC 19 : case REORDER_BUFFER_CHANGE_MESSAGE:
3811 : {
3812 ECB : char *data;
3813 GIC 19 : Size prefix_size = strlen(change->data.msg.prefix) + 1;
3814 ECB :
3815 CBC 19 : sz += prefix_size + change->data.msg.message_size +
3816 : sizeof(Size) + sizeof(Size);
3817 GIC 19 : ReorderBufferSerializeReserve(rb, sz);
3818 ECB :
3819 GIC 19 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
3820 :
3821 ECB : /* might have been reallocated above */
3822 GIC 19 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
3823 :
3824 ECB : /* write the prefix including the size */
3825 CBC 19 : memcpy(data, &prefix_size, sizeof(Size));
3826 19 : data += sizeof(Size);
3827 19 : memcpy(data, change->data.msg.prefix,
3828 : prefix_size);
3829 19 : data += prefix_size;
3830 ECB :
3831 : /* write the message including the size */
3832 GIC 19 : memcpy(data, &change->data.msg.message_size, sizeof(Size));
3833 19 : data += sizeof(Size);
3834 19 : memcpy(data, change->data.msg.message,
3835 : change->data.msg.message_size);
3836 19 : data += change->data.msg.message_size;
3837 ECB :
3838 GIC 19 : break;
3839 : }
3840 117 : case REORDER_BUFFER_CHANGE_INVALIDATION:
3841 ECB : {
3842 : char *data;
3843 CBC 117 : Size inval_size = sizeof(SharedInvalidationMessage) *
3844 GIC 117 : change->data.inval.ninvalidations;
3845 ECB :
3846 CBC 117 : sz += inval_size;
3847 :
3848 117 : ReorderBufferSerializeReserve(rb, sz);
3849 GIC 117 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
3850 :
3851 ECB : /* might have been reallocated above */
3852 GIC 117 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
3853 117 : memcpy(data, change->data.inval.invalidations, inval_size);
3854 117 : data += inval_size;
3855 :
3856 117 : break;
3857 : }
3858 2 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
3859 ECB : {
3860 : Snapshot snap;
3861 : char *data;
3862 :
3863 CBC 2 : snap = change->data.snapshot;
3864 :
3865 2 : sz += sizeof(SnapshotData) +
3866 GIC 2 : sizeof(TransactionId) * snap->xcnt +
3867 CBC 2 : sizeof(TransactionId) * snap->subxcnt;
3868 ECB :
3869 : /* make sure we have enough space */
3870 GIC 2 : ReorderBufferSerializeReserve(rb, sz);
3871 2 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
3872 ECB : /* might have been reallocated above */
3873 GIC 2 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
3874 ECB :
3875 CBC 2 : memcpy(data, snap, sizeof(SnapshotData));
3876 2 : data += sizeof(SnapshotData);
3877 :
3878 GIC 2 : if (snap->xcnt)
3879 : {
3880 CBC 2 : memcpy(data, snap->xip,
3881 GIC 2 : sizeof(TransactionId) * snap->xcnt);
3882 CBC 2 : data += sizeof(TransactionId) * snap->xcnt;
3883 : }
3884 ECB :
3885 GIC 2 : if (snap->subxcnt)
3886 ECB : {
3887 UIC 0 : memcpy(data, snap->subxip,
3888 LBC 0 : sizeof(TransactionId) * snap->subxcnt);
3889 0 : data += sizeof(TransactionId) * snap->subxcnt;
3890 : }
3891 CBC 2 : break;
3892 ECB : }
3893 UIC 0 : case REORDER_BUFFER_CHANGE_TRUNCATE:
3894 : {
3895 ECB : Size size;
3896 : char *data;
3897 :
3898 : /* account for the OIDs of truncated relations */
3899 UIC 0 : size = sizeof(Oid) * change->data.truncate.nrelids;
3900 LBC 0 : sz += size;
3901 ECB :
3902 : /* make sure we have enough space */
3903 LBC 0 : ReorderBufferSerializeReserve(rb, sz);
3904 :
3905 0 : data = ((char *) rb->outbuf) + sizeof(ReorderBufferDiskChange);
3906 : /* might have been reallocated above */
3907 UIC 0 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
3908 ECB :
3909 UIC 0 : memcpy(data, change->data.truncate.relids, size);
3910 LBC 0 : data += size;
3911 :
3912 0 : break;
3913 : }
3914 CBC 17135 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
3915 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
3916 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
3917 ECB : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
3918 : /* ReorderBufferChange contains everything important */
3919 GIC 17135 : break;
3920 ECB : }
3921 :
3922 CBC 1485023 : ondisk->size = sz;
3923 :
3924 1485023 : errno = 0;
3925 GIC 1485023 : pgstat_report_wait_start(WAIT_EVENT_REORDER_BUFFER_WRITE);
3926 1485023 : if (write(fd, rb->outbuf, ondisk->size) != ondisk->size)
3927 ECB : {
3928 LBC 0 : int save_errno = errno;
3929 ECB :
3930 UIC 0 : CloseTransientFile(fd);
3931 ECB :
3932 : /* if write didn't set errno, assume problem is no disk space */
3933 LBC 0 : errno = save_errno ? save_errno : ENOSPC;
3934 UIC 0 : ereport(ERROR,
3935 ECB : (errcode_for_file_access(),
3936 : errmsg("could not write to data file for XID %u: %m",
3937 : txn->xid)));
3938 : }
3939 CBC 1485023 : pgstat_report_wait_end();
3940 :
3941 ECB : /*
3942 : * Keep the transaction's final_lsn up to date with each change we send to
3943 : * disk, so that ReorderBufferRestoreCleanup works correctly. (We used to
3944 : * only do this on commit and abort records, but that doesn't work if a
3945 : * system crash leaves a transaction without its abort record).
3946 : *
3947 : * Make sure not to move it backwards.
3948 : */
3949 CBC 1485023 : if (txn->final_lsn < change->lsn)
3950 GIC 1480548 : txn->final_lsn = change->lsn;
3951 ECB :
3952 GIC 1485023 : Assert(ondisk->change.action == change->action);
3953 CBC 1485023 : }
3954 :
3955 : /* Returns true, if the output plugin supports streaming, false, otherwise. */
3956 : static inline bool
3957 GIC 2205003 : ReorderBufferCanStream(ReorderBuffer *rb)
3958 ECB : {
3959 GIC 2205003 : LogicalDecodingContext *ctx = rb->private_data;
3960 ECB :
3961 CBC 2205003 : return ctx->streaming;
3962 ECB : }
3963 :
3964 : /* Returns true, if the streaming can be started now, false, otherwise. */
3965 : static inline bool
3966 CBC 491348 : ReorderBufferCanStartStreaming(ReorderBuffer *rb)
3967 : {
3968 491348 : LogicalDecodingContext *ctx = rb->private_data;
3969 GIC 491348 : SnapBuild *builder = ctx->snapshot_builder;
3970 ECB :
3971 : /* We can't start streaming unless a consistent state is reached. */
3972 GIC 491348 : if (SnapBuildCurrentState(builder) < SNAPBUILD_CONSISTENT)
3973 LBC 0 : return false;
3974 :
3975 ECB : /*
3976 : * We can't start streaming immediately even if the streaming is enabled
3977 : * because we previously decoded this transaction and now just are
3978 : * restarting.
3979 : */
3980 CBC 491348 : if (ReorderBufferCanStream(rb) &&
3981 GNC 488688 : !SnapBuildXactNeedsSkip(builder, ctx->reader->ReadRecPtr))
3982 GBC 158469 : return true;
3983 EUB :
3984 GBC 332879 : return false;
3985 : }
3986 ECB :
3987 : /*
3988 EUB : * Send data of a large transaction (and its subtransactions) to the
3989 : * output plugin, but using the stream API.
3990 : */
3991 : static void
3992 GIC 660 : ReorderBufferStreamTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
3993 : {
3994 EUB : Snapshot snapshot_now;
3995 : CommandId command_id;
3996 : Size stream_bytes;
3997 : bool txn_is_streamed;
3998 :
3999 : /* We can never reach here for a subtransaction. */
4000 GNC 660 : Assert(rbtxn_is_toptxn(txn));
4001 :
4002 EUB : /*
4003 : * We can't make any assumptions about base snapshot here, similar to what
4004 : * ReorderBufferCommit() does. That relies on base_snapshot getting
4005 : * transferred from subxact in ReorderBufferCommitChild(), but that was
4006 : * not yet called as the transaction is in-progress.
4007 : *
4008 : * So just walk the subxacts and use the same logic here. But we only need
4009 ECB : * to do that once, when the transaction is streamed for the first time.
4010 : * After that we need to reuse the snapshot from the previous run.
4011 : *
4012 : * Unlike DecodeCommit which adds xids of all the subtransactions in
4013 : * snapshot's xip array via SnapBuildCommittedTxn, we can't do that here
4014 : * but we do add them to subxip array instead via ReorderBufferCopySnap.
4015 : * This allows the catalog changes made in subtransactions decoded till
4016 : * now to be visible.
4017 : */
4018 GIC 660 : if (txn->snapshot_now == NULL)
4019 ECB : {
4020 : dlist_iter subxact_i;
4021 :
4022 : /* make sure this transaction is streamed for the first time */
4023 GBC 66 : Assert(!rbtxn_is_streamed(txn));
4024 :
4025 EUB : /* at the beginning we should have invalid command ID */
4026 GIC 66 : Assert(txn->command_id == InvalidCommandId);
4027 :
4028 GBC 70 : dlist_foreach(subxact_i, &txn->subtxns)
4029 EUB : {
4030 : ReorderBufferTXN *subtxn;
4031 :
4032 GIC 4 : subtxn = dlist_container(ReorderBufferTXN, node, subxact_i.cur);
4033 4 : ReorderBufferTransferSnapToParent(txn, subtxn);
4034 ECB : }
4035 :
4036 : /*
4037 : * If this transaction has no snapshot, it didn't make any changes to
4038 : * the database till now, so there's nothing to decode.
4039 : */
4040 GIC 66 : if (txn->base_snapshot == NULL)
4041 : {
4042 UIC 0 : Assert(txn->ninvalidations == 0);
4043 0 : return;
4044 ECB : }
4045 :
4046 GIC 66 : command_id = FirstCommandId;
4047 CBC 66 : snapshot_now = ReorderBufferCopySnap(rb, txn->base_snapshot,
4048 ECB : txn, command_id);
4049 : }
4050 : else
4051 : {
4052 : /* the transaction must have been already streamed */
4053 GIC 594 : Assert(rbtxn_is_streamed(txn));
4054 ECB :
4055 : /*
4056 : * Nah, we already have snapshot from the previous streaming run. We
4057 : * assume new subxacts can't move the LSN backwards, and so can't beat
4058 : * the LSN condition in the previous branch (so no need to walk
4059 : * through subxacts again). In fact, we must not do that as we may be
4060 : * using snapshot half-way through the subxact.
4061 : */
4062 GIC 594 : command_id = txn->command_id;
4063 ECB :
4064 : /*
4065 : * We can't use txn->snapshot_now directly because after the last
4066 : * streaming run, we might have got some new sub-transactions. So we
4067 : * need to add them to the snapshot.
4068 EUB : */
4069 GIC 594 : snapshot_now = ReorderBufferCopySnap(rb, txn->snapshot_now,
4070 : txn, command_id);
4071 :
4072 : /* Free the previously copied snapshot. */
4073 594 : Assert(txn->snapshot_now->copied);
4074 594 : ReorderBufferFreeSnap(rb, txn->snapshot_now);
4075 CBC 594 : txn->snapshot_now = NULL;
4076 ECB : }
4077 :
4078 : /*
4079 : * Remember this information to be used later to update stats. We can't
4080 : * update the stats here as an error while processing the changes would
4081 : * lead to the accumulation of stats even though we haven't streamed all
4082 : * the changes.
4083 : */
4084 GIC 660 : txn_is_streamed = rbtxn_is_streamed(txn);
4085 660 : stream_bytes = txn->total_size;
4086 :
4087 ECB : /* Process and send the changes to output plugin. */
4088 GIC 660 : ReorderBufferProcessTXN(rb, txn, InvalidXLogRecPtr, snapshot_now,
4089 : command_id, true);
4090 :
4091 660 : rb->streamCount += 1;
4092 660 : rb->streamBytes += stream_bytes;
4093 :
4094 : /* Don't consider already streamed transaction. */
4095 CBC 660 : rb->streamTxns += (txn_is_streamed) ? 0 : 1;
4096 :
4097 : /* update the decoding stats */
4098 GIC 660 : UpdateDecodingStats((LogicalDecodingContext *) rb->private_data);
4099 :
4100 660 : Assert(dlist_is_empty(&txn->changes));
4101 660 : Assert(txn->nentries == 0);
4102 660 : Assert(txn->nentries_mem == 0);
4103 : }
4104 :
4105 : /*
4106 : * Size of a change in memory.
4107 : */
4108 : static Size
4109 3807299 : ReorderBufferChangeSize(ReorderBufferChange *change)
4110 : {
4111 3807299 : Size sz = sizeof(ReorderBufferChange);
4112 :
4113 CBC 3807299 : switch (change->action)
4114 : {
4115 : /* fall through these, they're all similar enough */
4116 GIC 3688274 : case REORDER_BUFFER_CHANGE_INSERT:
4117 : case REORDER_BUFFER_CHANGE_UPDATE:
4118 ECB : case REORDER_BUFFER_CHANGE_DELETE:
4119 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4120 : {
4121 : ReorderBufferTupleBuf *oldtup,
4122 : *newtup;
4123 CBC 3688274 : Size oldlen = 0;
4124 GIC 3688274 : Size newlen = 0;
4125 :
4126 3688274 : oldtup = change->data.tp.oldtuple;
4127 CBC 3688274 : newtup = change->data.tp.newtuple;
4128 ECB :
4129 GIC 3688274 : if (oldtup)
4130 : {
4131 421844 : sz += sizeof(HeapTupleData);
4132 421844 : oldlen = oldtup->tuple.t_len;
4133 421844 : sz += oldlen;
4134 : }
4135 ECB :
4136 GIC 3688274 : if (newtup)
4137 EUB : {
4138 GBC 3129918 : sz += sizeof(HeapTupleData);
4139 GIC 3129918 : newlen = newtup->tuple.t_len;
4140 3129918 : sz += newlen;
4141 ECB : }
4142 :
4143 GIC 3688274 : break;
4144 : }
4145 78 : case REORDER_BUFFER_CHANGE_MESSAGE:
4146 : {
4147 78 : Size prefix_size = strlen(change->data.msg.prefix) + 1;
4148 ECB :
4149 GIC 78 : sz += prefix_size + change->data.msg.message_size +
4150 : sizeof(Size) + sizeof(Size);
4151 :
4152 78 : break;
4153 : }
4154 9115 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4155 : {
4156 9115 : sz += sizeof(SharedInvalidationMessage) *
4157 CBC 9115 : change->data.inval.ninvalidations;
4158 GIC 9115 : break;
4159 : }
4160 2002 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4161 : {
4162 : Snapshot snap;
4163 :
4164 CBC 2002 : snap = change->data.snapshot;
4165 :
4166 GIC 2002 : sz += sizeof(SnapshotData) +
4167 2002 : sizeof(TransactionId) * snap->xcnt +
4168 CBC 2002 : sizeof(TransactionId) * snap->subxcnt;
4169 ECB :
4170 CBC 2002 : break;
4171 : }
4172 GIC 74 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4173 : {
4174 74 : sz += sizeof(Oid) * change->data.truncate.nrelids;
4175 :
4176 74 : break;
4177 : }
4178 107756 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4179 ECB : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4180 : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4181 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4182 : /* ReorderBufferChange contains everything important */
4183 CBC 107756 : break;
4184 : }
4185 :
4186 3807299 : return sz;
4187 ECB : }
4188 :
4189 :
4190 : /*
4191 : * Restore a number of changes spilled to disk back into memory.
4192 : */
4193 : static Size
4194 GIC 102 : ReorderBufferRestoreChanges(ReorderBuffer *rb, ReorderBufferTXN *txn,
4195 ECB : TXNEntryFile *file, XLogSegNo *segno)
4196 : {
4197 CBC 102 : Size restored = 0;
4198 : XLogSegNo last_segno;
4199 : dlist_mutable_iter cleanup_iter;
4200 GIC 102 : File *fd = &file->vfd;
4201 :
4202 102 : Assert(txn->first_lsn != InvalidXLogRecPtr);
4203 102 : Assert(txn->final_lsn != InvalidXLogRecPtr);
4204 ECB :
4205 : /* free current entries, so we have memory for more */
4206 CBC 175059 : dlist_foreach_modify(cleanup_iter, &txn->changes)
4207 : {
4208 174957 : ReorderBufferChange *cleanup =
4209 GIC 174957 : dlist_container(ReorderBufferChange, node, cleanup_iter.cur);
4210 :
4211 CBC 174957 : dlist_delete(&cleanup->node);
4212 GIC 174957 : ReorderBufferReturnChange(rb, cleanup, true);
4213 : }
4214 102 : txn->nentries_mem = 0;
4215 102 : Assert(dlist_is_empty(&txn->changes));
4216 :
4217 102 : XLByteToSeg(txn->final_lsn, last_segno, wal_segment_size);
4218 ECB :
4219 CBC 178616 : while (restored < max_changes_in_memory && *segno <= last_segno)
4220 : {
4221 ECB : int readBytes;
4222 : ReorderBufferDiskChange *ondisk;
4223 :
4224 CBC 178514 : CHECK_FOR_INTERRUPTS();
4225 :
4226 178514 : if (*fd == -1)
4227 ECB : {
4228 : char path[MAXPGPATH];
4229 :
4230 : /* first time in */
4231 CBC 38 : if (*segno == 0)
4232 GIC 37 : XLByteToSeg(txn->first_lsn, *segno, wal_segment_size);
4233 ECB :
4234 CBC 38 : Assert(*segno != 0 || dlist_is_empty(&txn->changes));
4235 ECB :
4236 : /*
4237 : * No need to care about TLIs here, only used during a single run,
4238 : * so each LSN only maps to a specific WAL record.
4239 : */
4240 CBC 38 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid,
4241 : *segno);
4242 ECB :
4243 GIC 38 : *fd = PathNameOpenFile(path, O_RDONLY | PG_BINARY);
4244 ECB :
4245 : /* No harm in resetting the offset even in case of failure */
4246 GIC 38 : file->curOffset = 0;
4247 ECB :
4248 GIC 38 : if (*fd < 0 && errno == ENOENT)
4249 ECB : {
4250 UIC 0 : *fd = -1;
4251 LBC 0 : (*segno)++;
4252 0 : continue;
4253 ECB : }
4254 GIC 38 : else if (*fd < 0)
4255 LBC 0 : ereport(ERROR,
4256 : (errcode_for_file_access(),
4257 : errmsg("could not open file \"%s\": %m",
4258 : path)));
4259 ECB : }
4260 :
4261 : /*
4262 : * Read the statically sized part of a change which has information
4263 : * about the total size. If we couldn't read a record, we're at the
4264 : * end of this file.
4265 : */
4266 GIC 178514 : ReorderBufferSerializeReserve(rb, sizeof(ReorderBufferDiskChange));
4267 CBC 178514 : readBytes = FileRead(file->vfd, rb->outbuf,
4268 : sizeof(ReorderBufferDiskChange),
4269 ECB : file->curOffset, WAIT_EVENT_REORDER_BUFFER_READ);
4270 :
4271 : /* eof */
4272 GIC 178514 : if (readBytes == 0)
4273 ECB : {
4274 GIC 38 : FileClose(*fd);
4275 38 : *fd = -1;
4276 38 : (*segno)++;
4277 38 : continue;
4278 ECB : }
4279 GIC 178476 : else if (readBytes < 0)
4280 UIC 0 : ereport(ERROR,
4281 ECB : (errcode_for_file_access(),
4282 : errmsg("could not read from reorderbuffer spill file: %m")));
4283 GIC 178476 : else if (readBytes != sizeof(ReorderBufferDiskChange))
4284 UIC 0 : ereport(ERROR,
4285 : (errcode_for_file_access(),
4286 : errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
4287 : readBytes,
4288 : (uint32) sizeof(ReorderBufferDiskChange))));
4289 ECB :
4290 GIC 178476 : file->curOffset += readBytes;
4291 :
4292 CBC 178476 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4293 :
4294 GIC 178476 : ReorderBufferSerializeReserve(rb,
4295 CBC 178476 : sizeof(ReorderBufferDiskChange) + ondisk->size);
4296 GIC 178476 : ondisk = (ReorderBufferDiskChange *) rb->outbuf;
4297 ECB :
4298 CBC 178476 : readBytes = FileRead(file->vfd,
4299 GIC 178476 : rb->outbuf + sizeof(ReorderBufferDiskChange),
4300 178476 : ondisk->size - sizeof(ReorderBufferDiskChange),
4301 ECB : file->curOffset,
4302 : WAIT_EVENT_REORDER_BUFFER_READ);
4303 :
4304 CBC 178476 : if (readBytes < 0)
4305 UIC 0 : ereport(ERROR,
4306 ECB : (errcode_for_file_access(),
4307 : errmsg("could not read from reorderbuffer spill file: %m")));
4308 GIC 178476 : else if (readBytes != ondisk->size - sizeof(ReorderBufferDiskChange))
4309 LBC 0 : ereport(ERROR,
4310 ECB : (errcode_for_file_access(),
4311 : errmsg("could not read from reorderbuffer spill file: read %d instead of %u bytes",
4312 : readBytes,
4313 : (uint32) (ondisk->size - sizeof(ReorderBufferDiskChange)))));
4314 :
4315 GIC 178476 : file->curOffset += readBytes;
4316 :
4317 : /*
4318 : * ok, read a full change from disk, now restore it into proper
4319 ECB : * in-memory format
4320 : */
4321 CBC 178476 : ReorderBufferRestoreChange(rb, txn, rb->outbuf);
4322 GIC 178476 : restored++;
4323 : }
4324 :
4325 102 : return restored;
4326 ECB : }
4327 :
4328 : /*
4329 : * Convert change from its on-disk format to in-memory format and queue it onto
4330 : * the TXN's ->changes list.
4331 : *
4332 : * Note: although "data" is declared char*, at entry it points to a
4333 : * maxalign'd buffer, making it safe in most of this function to assume
4334 : * that the pointed-to data is suitably aligned for direct access.
4335 : */
4336 : static void
4337 GIC 178476 : ReorderBufferRestoreChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
4338 ECB : char *data)
4339 : {
4340 : ReorderBufferDiskChange *ondisk;
4341 : ReorderBufferChange *change;
4342 :
4343 CBC 178476 : ondisk = (ReorderBufferDiskChange *) data;
4344 :
4345 GBC 178476 : change = ReorderBufferGetChange(rb);
4346 EUB :
4347 : /* copy static part */
4348 GIC 178476 : memcpy(change, &ondisk->change, sizeof(ReorderBufferChange));
4349 ECB :
4350 GBC 178476 : data += sizeof(ReorderBufferDiskChange);
4351 :
4352 : /* restore individual stuff */
4353 GIC 178476 : switch (change->action)
4354 : {
4355 : /* fall through these, they're all similar enough */
4356 176581 : case REORDER_BUFFER_CHANGE_INSERT:
4357 : case REORDER_BUFFER_CHANGE_UPDATE:
4358 : case REORDER_BUFFER_CHANGE_DELETE:
4359 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT:
4360 176581 : if (change->data.tp.oldtuple)
4361 ECB : {
4362 CBC 5006 : uint32 tuplelen = ((HeapTuple) data)->t_len;
4363 :
4364 GIC 5006 : change->data.tp.oldtuple =
4365 5006 : ReorderBufferGetTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
4366 :
4367 ECB : /* restore ->tuple */
4368 GIC 5006 : memcpy(&change->data.tp.oldtuple->tuple, data,
4369 ECB : sizeof(HeapTupleData));
4370 CBC 5006 : data += sizeof(HeapTupleData);
4371 ECB :
4372 : /* reset t_data pointer into the new tuplebuf */
4373 GIC 5006 : change->data.tp.oldtuple->tuple.t_data =
4374 CBC 5006 : ReorderBufferTupleBufData(change->data.tp.oldtuple);
4375 EUB :
4376 : /* restore tuple data itself */
4377 GIC 5006 : memcpy(change->data.tp.oldtuple->tuple.t_data, data, tuplelen);
4378 CBC 5006 : data += tuplelen;
4379 EUB : }
4380 :
4381 GIC 176581 : if (change->data.tp.newtuple)
4382 : {
4383 : /* here, data might not be suitably aligned! */
4384 : uint32 tuplelen;
4385 ECB :
4386 GIC 166361 : memcpy(&tuplelen, data + offsetof(HeapTupleData, t_len),
4387 ECB : sizeof(uint32));
4388 :
4389 CBC 166361 : change->data.tp.newtuple =
4390 166361 : ReorderBufferGetTupleBuf(rb, tuplelen - SizeofHeapTupleHeader);
4391 ECB :
4392 : /* restore ->tuple */
4393 CBC 166361 : memcpy(&change->data.tp.newtuple->tuple, data,
4394 ECB : sizeof(HeapTupleData));
4395 CBC 166361 : data += sizeof(HeapTupleData);
4396 :
4397 : /* reset t_data pointer into the new tuplebuf */
4398 GIC 166361 : change->data.tp.newtuple->tuple.t_data =
4399 CBC 166361 : ReorderBufferTupleBufData(change->data.tp.newtuple);
4400 EUB :
4401 : /* restore tuple data itself */
4402 GIC 166361 : memcpy(change->data.tp.newtuple->tuple.t_data, data, tuplelen);
4403 CBC 166361 : data += tuplelen;
4404 EUB : }
4405 :
4406 GIC 176581 : break;
4407 1 : case REORDER_BUFFER_CHANGE_MESSAGE:
4408 : {
4409 : Size prefix_size;
4410 ECB :
4411 : /* read prefix */
4412 GIC 1 : memcpy(&prefix_size, data, sizeof(Size));
4413 1 : data += sizeof(Size);
4414 1 : change->data.msg.prefix = MemoryContextAlloc(rb->context,
4415 : prefix_size);
4416 CBC 1 : memcpy(change->data.msg.prefix, data, prefix_size);
4417 1 : Assert(change->data.msg.prefix[prefix_size - 1] == '\0');
4418 GIC 1 : data += prefix_size;
4419 :
4420 ECB : /* read the message */
4421 GIC 1 : memcpy(&change->data.msg.message_size, data, sizeof(Size));
4422 1 : data += sizeof(Size);
4423 1 : change->data.msg.message = MemoryContextAlloc(rb->context,
4424 : change->data.msg.message_size);
4425 1 : memcpy(change->data.msg.message, data,
4426 : change->data.msg.message_size);
4427 1 : data += change->data.msg.message_size;
4428 :
4429 1 : break;
4430 : }
4431 19 : case REORDER_BUFFER_CHANGE_INVALIDATION:
4432 ECB : {
4433 GIC 19 : Size inval_size = sizeof(SharedInvalidationMessage) *
4434 19 : change->data.inval.ninvalidations;
4435 :
4436 19 : change->data.inval.invalidations =
4437 19 : MemoryContextAlloc(rb->context, inval_size);
4438 ECB :
4439 : /* read the message */
4440 CBC 19 : memcpy(change->data.inval.invalidations, data, inval_size);
4441 :
4442 GIC 19 : break;
4443 ECB : }
4444 GIC 2 : case REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT:
4445 ECB : {
4446 : Snapshot oldsnap;
4447 : Snapshot newsnap;
4448 : Size size;
4449 :
4450 GIC 2 : oldsnap = (Snapshot) data;
4451 ECB :
4452 GIC 2 : size = sizeof(SnapshotData) +
4453 2 : sizeof(TransactionId) * oldsnap->xcnt +
4454 2 : sizeof(TransactionId) * (oldsnap->subxcnt + 0);
4455 ECB :
4456 GIC 2 : change->data.snapshot = MemoryContextAllocZero(rb->context, size);
4457 ECB :
4458 GIC 2 : newsnap = change->data.snapshot;
4459 ECB :
4460 CBC 2 : memcpy(newsnap, data, size);
4461 GIC 2 : newsnap->xip = (TransactionId *)
4462 : (((char *) newsnap) + sizeof(SnapshotData));
4463 CBC 2 : newsnap->subxip = newsnap->xip + newsnap->xcnt;
4464 GIC 2 : newsnap->copied = true;
4465 CBC 2 : break;
4466 : }
4467 : /* the base struct contains all the data, easy peasy */
4468 LBC 0 : case REORDER_BUFFER_CHANGE_TRUNCATE:
4469 ECB : {
4470 : Oid *relids;
4471 :
4472 LBC 0 : relids = ReorderBufferGetRelids(rb,
4473 0 : change->data.truncate.nrelids);
4474 UIC 0 : memcpy(relids, data, change->data.truncate.nrelids * sizeof(Oid));
4475 0 : change->data.truncate.relids = relids;
4476 ECB :
4477 UIC 0 : break;
4478 : }
4479 GIC 1873 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_CONFIRM:
4480 : case REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT:
4481 ECB : case REORDER_BUFFER_CHANGE_INTERNAL_COMMAND_ID:
4482 : case REORDER_BUFFER_CHANGE_INTERNAL_TUPLECID:
4483 GIC 1873 : break;
4484 ECB : }
4485 :
4486 GIC 178476 : dlist_push_tail(&txn->changes, &change->node);
4487 178476 : txn->nentries_mem++;
4488 ECB :
4489 : /*
4490 : * Update memory accounting for the restored change. We need to do this
4491 : * although we don't check the memory limit when restoring the changes in
4492 : * this branch (we only do that when initially queueing the changes after
4493 : * decoding), because we will release the changes later, and that will
4494 : * update the accounting too (subtracting the size from the counters). And
4495 : * we don't want to underflow there.
4496 : */
4497 CBC 178476 : ReorderBufferChangeMemoryUpdate(rb, change, true,
4498 ECB : ReorderBufferChangeSize(change));
4499 GIC 178476 : }
4500 :
4501 ECB : /*
4502 : * Remove all on-disk stored for the passed in transaction.
4503 : */
4504 : static void
4505 GIC 298 : ReorderBufferRestoreCleanup(ReorderBuffer *rb, ReorderBufferTXN *txn)
4506 : {
4507 ECB : XLogSegNo first;
4508 : XLogSegNo cur;
4509 : XLogSegNo last;
4510 :
4511 CBC 298 : Assert(txn->first_lsn != InvalidXLogRecPtr);
4512 298 : Assert(txn->final_lsn != InvalidXLogRecPtr);
4513 ECB :
4514 GIC 298 : XLByteToSeg(txn->first_lsn, first, wal_segment_size);
4515 298 : XLByteToSeg(txn->final_lsn, last, wal_segment_size);
4516 ECB :
4517 : /* iterate over all possible filenames, and delete them */
4518 CBC 600 : for (cur = first; cur <= last; cur++)
4519 : {
4520 ECB : char path[MAXPGPATH];
4521 :
4522 CBC 302 : ReorderBufferSerializedPath(path, MyReplicationSlot, txn->xid, cur);
4523 GIC 302 : if (unlink(path) != 0 && errno != ENOENT)
4524 LBC 0 : ereport(ERROR,
4525 : (errcode_for_file_access(),
4526 ECB : errmsg("could not remove file \"%s\": %m", path)));
4527 : }
4528 CBC 298 : }
4529 ECB :
4530 : /*
4531 : * Remove any leftover serialized reorder buffers from a slot directory after a
4532 : * prior crash or decoding session exit.
4533 : */
4534 : static void
4535 CBC 1534 : ReorderBufferCleanupSerializedTXNs(const char *slotname)
4536 : {
4537 ECB : DIR *spill_dir;
4538 : struct dirent *spill_de;
4539 : struct stat statbuf;
4540 : char path[MAXPGPATH * 2 + 12];
4541 :
4542 GIC 1534 : sprintf(path, "pg_replslot/%s", slotname);
4543 :
4544 : /* we're only handling directories here, skip if it's not ours */
4545 CBC 1534 : if (lstat(path, &statbuf) == 0 && !S_ISDIR(statbuf.st_mode))
4546 UIC 0 : return;
4547 ECB :
4548 CBC 1534 : spill_dir = AllocateDir(path);
4549 7670 : while ((spill_de = ReadDirExtended(spill_dir, path, INFO)) != NULL)
4550 : {
4551 ECB : /* only look at names that can be ours */
4552 GIC 4602 : if (strncmp(spill_de->d_name, "xid", 3) == 0)
4553 ECB : {
4554 UIC 0 : snprintf(path, sizeof(path),
4555 ECB : "pg_replslot/%s/%s", slotname,
4556 LBC 0 : spill_de->d_name);
4557 :
4558 0 : if (unlink(path) != 0)
4559 0 : ereport(ERROR,
4560 ECB : (errcode_for_file_access(),
4561 : errmsg("could not remove file \"%s\" during removal of pg_replslot/%s/xid*: %m",
4562 : path, slotname)));
4563 EUB : }
4564 : }
4565 GIC 1534 : FreeDir(spill_dir);
4566 : }
4567 EUB :
4568 : /*
4569 : * Given a replication slot, transaction ID and segment number, fill in the
4570 : * corresponding spill file into 'path', which is a caller-owned buffer of size
4571 : * at least MAXPGPATH.
4572 : */
4573 : static void
4574 CBC 4096 : ReorderBufferSerializedPath(char *path, ReplicationSlot *slot, TransactionId xid,
4575 : XLogSegNo segno)
4576 : {
4577 : XLogRecPtr recptr;
4578 ECB :
4579 GIC 4096 : XLogSegNoOffsetToRecPtr(segno, 0, wal_segment_size, recptr);
4580 :
4581 CBC 4096 : snprintf(path, MAXPGPATH, "pg_replslot/%s/xid-%u-lsn-%X-%X.spill",
4582 4096 : NameStr(MyReplicationSlot->data.name),
4583 GIC 4096 : xid, LSN_FORMAT_ARGS(recptr));
4584 4096 : }
4585 :
4586 : /*
4587 : * Delete all data spilled to disk after we've restarted/crashed. It will be
4588 : * recreated when the respective slots are reused.
4589 : */
4590 : void
4591 1176 : StartupReorderBuffer(void)
4592 ECB : {
4593 : DIR *logical_dir;
4594 : struct dirent *logical_de;
4595 :
4596 GIC 1176 : logical_dir = AllocateDir("pg_replslot");
4597 3562 : while ((logical_de = ReadDir(logical_dir, "pg_replslot")) != NULL)
4598 : {
4599 2386 : if (strcmp(logical_de->d_name, ".") == 0 ||
4600 CBC 1210 : strcmp(logical_de->d_name, "..") == 0)
4601 GIC 2352 : continue;
4602 :
4603 : /* if it cannot be a slot, skip the directory */
4604 34 : if (!ReplicationSlotValidateName(logical_de->d_name, DEBUG2))
4605 UIC 0 : continue;
4606 ECB :
4607 : /*
4608 : * ok, has to be a surviving logical slot, iterate and delete
4609 : * everything starting with xid-*
4610 : */
4611 GIC 34 : ReorderBufferCleanupSerializedTXNs(logical_de->d_name);
4612 : }
4613 CBC 1176 : FreeDir(logical_dir);
4614 GIC 1176 : }
4615 :
4616 : /* ---------------------------------------
4617 ECB : * toast reassembly support
4618 : * ---------------------------------------
4619 EUB : */
4620 :
4621 : /*
4622 : * Initialize per tuple toast reconstruction support.
4623 ECB : */
4624 : static void
4625 GIC 33 : ReorderBufferToastInitHash(ReorderBuffer *rb, ReorderBufferTXN *txn)
4626 : {
4627 : HASHCTL hash_ctl;
4628 :
4629 33 : Assert(txn->toast_hash == NULL);
4630 ECB :
4631 GIC 33 : hash_ctl.keysize = sizeof(Oid);
4632 33 : hash_ctl.entrysize = sizeof(ReorderBufferToastEnt);
4633 33 : hash_ctl.hcxt = rb->context;
4634 33 : txn->toast_hash = hash_create("ReorderBufferToastHash", 5, &hash_ctl,
4635 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
4636 33 : }
4637 ECB :
4638 : /*
4639 : * Per toast-chunk handling for toast reconstruction
4640 : *
4641 EUB : * Appends a toast chunk so we can reconstruct it when the tuple "owning" the
4642 : * toasted Datum comes along.
4643 ECB : */
4644 : static void
4645 GIC 1720 : ReorderBufferToastAppendChunk(ReorderBuffer *rb, ReorderBufferTXN *txn,
4646 : Relation relation, ReorderBufferChange *change)
4647 ECB : {
4648 : ReorderBufferToastEnt *ent;
4649 EUB : ReorderBufferTupleBuf *newtup;
4650 : bool found;
4651 : int32 chunksize;
4652 : bool isnull;
4653 : Pointer chunk;
4654 GBC 1720 : TupleDesc desc = RelationGetDescr(relation);
4655 : Oid chunk_id;
4656 : int32 chunk_seq;
4657 :
4658 GIC 1720 : if (txn->toast_hash == NULL)
4659 33 : ReorderBufferToastInitHash(rb, txn);
4660 ECB :
4661 GIC 1720 : Assert(IsToastRelation(relation));
4662 :
4663 1720 : newtup = change->data.tp.newtuple;
4664 1720 : chunk_id = DatumGetObjectId(fastgetattr(&newtup->tuple, 1, desc, &isnull));
4665 1720 : Assert(!isnull);
4666 1720 : chunk_seq = DatumGetInt32(fastgetattr(&newtup->tuple, 2, desc, &isnull));
4667 1720 : Assert(!isnull);
4668 :
4669 ECB : ent = (ReorderBufferToastEnt *)
4670 GNC 1720 : hash_search(txn->toast_hash, &chunk_id, HASH_ENTER, &found);
4671 ECB :
4672 GIC 1720 : if (!found)
4673 ECB : {
4674 CBC 47 : Assert(ent->chunk_id == chunk_id);
4675 47 : ent->num_chunks = 0;
4676 47 : ent->last_chunk_seq = 0;
4677 GIC 47 : ent->size = 0;
4678 47 : ent->reconstructed = NULL;
4679 47 : dlist_init(&ent->chunks);
4680 :
4681 47 : if (chunk_seq != 0)
4682 UIC 0 : elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq 0",
4683 ECB : chunk_seq, chunk_id);
4684 : }
4685 GIC 1673 : else if (found && chunk_seq != ent->last_chunk_seq + 1)
4686 UIC 0 : elog(ERROR, "got sequence entry %d for toast chunk %u instead of seq %d",
4687 : chunk_seq, chunk_id, ent->last_chunk_seq + 1);
4688 ECB :
4689 CBC 1720 : chunk = DatumGetPointer(fastgetattr(&newtup->tuple, 3, desc, &isnull));
4690 GIC 1720 : Assert(!isnull);
4691 ECB :
4692 : /* calculate size so we can allocate the right size at once later */
4693 CBC 1720 : if (!VARATT_IS_EXTENDED(chunk))
4694 GIC 1720 : chunksize = VARSIZE(chunk) - VARHDRSZ;
4695 UIC 0 : else if (VARATT_IS_SHORT(chunk))
4696 ECB : /* could happen due to heap_form_tuple doing its thing */
4697 UBC 0 : chunksize = VARSIZE_SHORT(chunk) - VARHDRSZ_SHORT;
4698 : else
4699 UIC 0 : elog(ERROR, "unexpected type of toast chunk");
4700 :
4701 GIC 1720 : ent->size += chunksize;
4702 1720 : ent->last_chunk_seq = chunk_seq;
4703 CBC 1720 : ent->num_chunks++;
4704 GIC 1720 : dlist_push_tail(&ent->chunks, &change->node);
4705 CBC 1720 : }
4706 ECB :
4707 : /*
4708 : * Rejigger change->newtuple to point to in-memory toast tuples instead to
4709 : * on-disk toast tuples that may not longer exist (think DROP TABLE or VACUUM).
4710 : *
4711 : * We cannot replace unchanged toast tuples though, so those will still point
4712 : * to on-disk toast data.
4713 : *
4714 : * While updating the existing change with detoasted tuple data, we need to
4715 : * update the memory accounting info, because the change size will differ.
4716 : * Otherwise the accounting may get out of sync, triggering serialization
4717 : * at unexpected times.
4718 : *
4719 : * We simply subtract size of the change before rejiggering the tuple, and
4720 : * then adding the new size. This makes it look like the change was removed
4721 : * and then added back, except it only tweaks the accounting info.
4722 : *
4723 : * In particular it can't trigger serialization, which would be pointless
4724 : * anyway as it happens during commit processing right before handing
4725 : * the change to the output plugin.
4726 : */
4727 : static void
4728 CBC 333687 : ReorderBufferToastReplace(ReorderBuffer *rb, ReorderBufferTXN *txn,
4729 : Relation relation, ReorderBufferChange *change)
4730 : {
4731 : TupleDesc desc;
4732 : int natt;
4733 : Datum *attrs;
4734 : bool *isnull;
4735 : bool *free;
4736 : HeapTuple tmphtup;
4737 ECB : Relation toast_rel;
4738 : TupleDesc toast_desc;
4739 : MemoryContext oldcontext;
4740 : ReorderBufferTupleBuf *newtup;
4741 : Size old_size;
4742 :
4743 : /* no toast tuples changed */
4744 GIC 333687 : if (txn->toast_hash == NULL)
4745 333442 : return;
4746 ECB :
4747 : /*
4748 : * We're going to modify the size of the change. So, to make sure the
4749 : * accounting is correct we record the current change size and then after
4750 : * re-computing the change we'll subtract the recorded size and then
4751 : * re-add the new change size at the end. We don't immediately subtract
4752 : * the old size because if there is any error before we add the new size,
4753 : * we will release the changes and that will update the accounting info
4754 : * (subtracting the size from the counters). And we don't want to
4755 : * underflow there.
4756 : */
4757 CBC 245 : old_size = ReorderBufferChangeSize(change);
4758 ECB :
4759 CBC 245 : oldcontext = MemoryContextSwitchTo(rb->context);
4760 :
4761 : /* we should only have toast tuples in an INSERT or UPDATE */
4762 245 : Assert(change->data.tp.newtuple);
4763 :
4764 245 : desc = RelationGetDescr(relation);
4765 :
4766 245 : toast_rel = RelationIdGetRelation(relation->rd_rel->reltoastrelid);
4767 245 : if (!RelationIsValid(toast_rel))
4768 LBC 0 : elog(ERROR, "could not open toast relation with OID %u (base relation \"%s\")",
4769 ECB : relation->rd_rel->reltoastrelid, RelationGetRelationName(relation));
4770 :
4771 CBC 245 : toast_desc = RelationGetDescr(toast_rel);
4772 :
4773 ECB : /* should we allocate from stack instead? */
4774 GBC 245 : attrs = palloc0(sizeof(Datum) * desc->natts);
4775 GIC 245 : isnull = palloc0(sizeof(bool) * desc->natts);
4776 245 : free = palloc0(sizeof(bool) * desc->natts);
4777 ECB :
4778 GBC 245 : newtup = change->data.tp.newtuple;
4779 :
4780 GIC 245 : heap_deform_tuple(&newtup->tuple, desc, attrs, isnull);
4781 ECB :
4782 CBC 755 : for (natt = 0; natt < desc->natts; natt++)
4783 : {
4784 GIC 510 : Form_pg_attribute attr = TupleDescAttr(desc, natt);
4785 ECB : ReorderBufferToastEnt *ent;
4786 : struct varlena *varlena;
4787 EUB :
4788 : /* va_rawsize is the size of the original datum -- including header */
4789 : struct varatt_external toast_pointer;
4790 : struct varatt_indirect redirect_pointer;
4791 GBC 510 : struct varlena *new_datum = NULL;
4792 : struct varlena *reconstructed;
4793 ECB : dlist_iter it;
4794 CBC 510 : Size data_done = 0;
4795 ECB :
4796 : /* system columns aren't toasted */
4797 CBC 510 : if (attr->attnum < 0)
4798 GIC 463 : continue;
4799 :
4800 510 : if (attr->attisdropped)
4801 UIC 0 : continue;
4802 :
4803 : /* not a varlena datatype */
4804 GIC 510 : if (attr->attlen != -1)
4805 241 : continue;
4806 :
4807 : /* no data */
4808 269 : if (isnull[natt])
4809 12 : continue;
4810 :
4811 : /* ok, we know we have a toast datum */
4812 257 : varlena = (struct varlena *) DatumGetPointer(attrs[natt]);
4813 :
4814 : /* no need to do anything if the tuple isn't external */
4815 257 : if (!VARATT_IS_EXTERNAL(varlena))
4816 202 : continue;
4817 :
4818 55 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, varlena);
4819 :
4820 ECB : /*
4821 : * Check whether the toast tuple changed, replace if so.
4822 : */
4823 : ent = (ReorderBufferToastEnt *)
4824 GIC 55 : hash_search(txn->toast_hash,
4825 : &toast_pointer.va_valueid,
4826 : HASH_FIND,
4827 : NULL);
4828 55 : if (ent == NULL)
4829 8 : continue;
4830 :
4831 : new_datum =
4832 47 : (struct varlena *) palloc0(INDIRECT_POINTER_SIZE);
4833 :
4834 47 : free[natt] = true;
4835 :
4836 CBC 47 : reconstructed = palloc0(toast_pointer.va_rawsize);
4837 ECB :
4838 GIC 47 : ent->reconstructed = reconstructed;
4839 :
4840 : /* stitch toast tuple back together from its parts */
4841 1767 : dlist_foreach(it, &ent->chunks)
4842 : {
4843 : bool isnull;
4844 : ReorderBufferChange *cchange;
4845 : ReorderBufferTupleBuf *ctup;
4846 : Pointer chunk;
4847 :
4848 1720 : cchange = dlist_container(ReorderBufferChange, node, it.cur);
4849 CBC 1720 : ctup = cchange->data.tp.newtuple;
4850 GIC 1720 : chunk = DatumGetPointer(fastgetattr(&ctup->tuple, 3, toast_desc, &isnull));
4851 ECB :
4852 GIC 1720 : Assert(!isnull);
4853 1720 : Assert(!VARATT_IS_EXTERNAL(chunk));
4854 CBC 1720 : Assert(!VARATT_IS_SHORT(chunk));
4855 :
4856 1720 : memcpy(VARDATA(reconstructed) + data_done,
4857 GIC 1720 : VARDATA(chunk),
4858 CBC 1720 : VARSIZE(chunk) - VARHDRSZ);
4859 1720 : data_done += VARSIZE(chunk) - VARHDRSZ;
4860 EUB : }
4861 GIC 47 : Assert(data_done == VARATT_EXTERNAL_GET_EXTSIZE(toast_pointer));
4862 :
4863 ECB : /* make sure its marked as compressed or not */
4864 GIC 47 : if (VARATT_EXTERNAL_IS_COMPRESSED(toast_pointer))
4865 5 : SET_VARSIZE_COMPRESSED(reconstructed, data_done + VARHDRSZ);
4866 ECB : else
4867 CBC 42 : SET_VARSIZE(reconstructed, data_done + VARHDRSZ);
4868 ECB :
4869 GIC 47 : memset(&redirect_pointer, 0, sizeof(redirect_pointer));
4870 CBC 47 : redirect_pointer.pointer = reconstructed;
4871 :
4872 47 : SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
4873 GIC 47 : memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
4874 ECB : sizeof(redirect_pointer));
4875 :
4876 CBC 47 : attrs[natt] = PointerGetDatum(new_datum);
4877 : }
4878 :
4879 : /*
4880 : * Build tuple in separate memory & copy tuple back into the tuplebuf
4881 : * passed to the output plugin. We can't directly heap_fill_tuple() into
4882 : * the tuplebuf because attrs[] will point back into the current content.
4883 ECB : */
4884 GIC 245 : tmphtup = heap_form_tuple(desc, attrs, isnull);
4885 245 : Assert(newtup->tuple.t_len <= MaxHeapTupleSize);
4886 CBC 245 : Assert(ReorderBufferTupleBufData(newtup) == newtup->tuple.t_data);
4887 :
4888 GIC 245 : memcpy(newtup->tuple.t_data, tmphtup->t_data, tmphtup->t_len);
4889 CBC 245 : newtup->tuple.t_len = tmphtup->t_len;
4890 ECB :
4891 : /*
4892 : * free resources we won't further need, more persistent stuff will be
4893 EUB : * free'd in ReorderBufferToastReset().
4894 : */
4895 GIC 245 : RelationClose(toast_rel);
4896 CBC 245 : pfree(tmphtup);
4897 755 : for (natt = 0; natt < desc->natts; natt++)
4898 : {
4899 GIC 510 : if (free[natt])
4900 CBC 47 : pfree(DatumGetPointer(attrs[natt]));
4901 ECB : }
4902 GIC 245 : pfree(attrs);
4903 245 : pfree(free);
4904 CBC 245 : pfree(isnull);
4905 :
4906 GIC 245 : MemoryContextSwitchTo(oldcontext);
4907 ECB :
4908 : /* subtract the old change size */
4909 GIC 245 : ReorderBufferChangeMemoryUpdate(rb, change, false, old_size);
4910 ECB : /* now add the change back, with the correct size */
4911 GIC 245 : ReorderBufferChangeMemoryUpdate(rb, change, true,
4912 : ReorderBufferChangeSize(change));
4913 : }
4914 :
4915 : /*
4916 ECB : * Free all resources allocated for toast reconstruction.
4917 : */
4918 : static void
4919 GIC 336731 : ReorderBufferToastReset(ReorderBuffer *rb, ReorderBufferTXN *txn)
4920 ECB : {
4921 : HASH_SEQ_STATUS hstat;
4922 : ReorderBufferToastEnt *ent;
4923 :
4924 CBC 336731 : if (txn->toast_hash == NULL)
4925 GIC 336698 : return;
4926 ECB :
4927 : /* sequentially walk over the hash and free everything */
4928 CBC 33 : hash_seq_init(&hstat, txn->toast_hash);
4929 GIC 80 : while ((ent = (ReorderBufferToastEnt *) hash_seq_search(&hstat)) != NULL)
4930 ECB : {
4931 : dlist_mutable_iter it;
4932 :
4933 CBC 47 : if (ent->reconstructed != NULL)
4934 GIC 47 : pfree(ent->reconstructed);
4935 :
4936 1767 : dlist_foreach_modify(it, &ent->chunks)
4937 : {
4938 1720 : ReorderBufferChange *change =
4939 1720 : dlist_container(ReorderBufferChange, node, it.cur);
4940 ECB :
4941 CBC 1720 : dlist_delete(&change->node);
4942 1720 : ReorderBufferReturnChange(rb, change, true);
4943 : }
4944 ECB : }
4945 :
4946 CBC 33 : hash_destroy(txn->toast_hash);
4947 GIC 33 : txn->toast_hash = NULL;
4948 ECB : }
4949 :
4950 :
4951 : /* ---------------------------------------
4952 : * Visibility support for logical decoding
4953 : *
4954 : *
4955 : * Lookup actual cmin/cmax values when using decoding snapshot. We can't
4956 : * always rely on stored cmin/cmax values because of two scenarios:
4957 : *
4958 : * * A tuple got changed multiple times during a single transaction and thus
4959 : * has got a combo CID. Combo CIDs are only valid for the duration of a
4960 : * single transaction.
4961 : * * A tuple with a cmin but no cmax (and thus no combo CID) got
4962 : * deleted/updated in another transaction than the one which created it
4963 : * which we are looking at right now. As only one of cmin, cmax or combo CID
4964 : * is actually stored in the heap we don't have access to the value we
4965 : * need anymore.
4966 : *
4967 : * To resolve those problems we have a per-transaction hash of (cmin,
4968 : * cmax) tuples keyed by (relfilelocator, ctid) which contains the actual
4969 : * (cmin, cmax) values. That also takes care of combo CIDs by simply
4970 : * not caring about them at all. As we have the real cmin/cmax values
4971 : * combo CIDs aren't interesting.
4972 : *
4973 : * As we only care about catalog tuples here the overhead of this
4974 : * hashtable should be acceptable.
4975 : *
4976 : * Heap rewrites complicate this a bit, check rewriteheap.c for
4977 : * details.
4978 : * -------------------------------------------------------------------------
4979 : */
4980 :
4981 : /* struct for sorting mapping files by LSN efficiently */
4982 : typedef struct RewriteMappingFile
4983 : {
4984 : XLogRecPtr lsn;
4985 : char fname[MAXPGPATH];
4986 : } RewriteMappingFile;
4987 :
4988 : #ifdef NOT_USED
4989 : static void
4990 : DisplayMapping(HTAB *tuplecid_data)
4991 : {
4992 : HASH_SEQ_STATUS hstat;
4993 : ReorderBufferTupleCidEnt *ent;
4994 :
4995 : hash_seq_init(&hstat, tuplecid_data);
4996 : while ((ent = (ReorderBufferTupleCidEnt *) hash_seq_search(&hstat)) != NULL)
4997 : {
4998 : elog(DEBUG3, "mapping: node: %u/%u/%u tid: %u/%u cmin: %u, cmax: %u",
4999 : ent->key.rlocator.dbOid,
5000 : ent->key.rlocator.spcOid,
5001 : ent->key.rlocator.relNumber,
5002 : ItemPointerGetBlockNumber(&ent->key.tid),
5003 : ItemPointerGetOffsetNumber(&ent->key.tid),
5004 : ent->cmin,
5005 : ent->cmax
5006 : );
5007 : }
5008 : }
5009 : #endif
5010 :
5011 : /*
5012 : * Apply a single mapping file to tuplecid_data.
5013 : *
5014 : * The mapping file has to have been verified to be a) committed b) for our
5015 : * transaction c) applied in LSN order.
5016 : */
5017 : static void
5018 GIC 22 : ApplyLogicalMappingFile(HTAB *tuplecid_data, Oid relid, const char *fname)
5019 : {
5020 ECB : char path[MAXPGPATH];
5021 : int fd;
5022 : int readBytes;
5023 : LogicalRewriteMappingData map;
5024 :
5025 CBC 22 : sprintf(path, "pg_logical/mappings/%s", fname);
5026 22 : fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
5027 GIC 22 : if (fd < 0)
5028 LBC 0 : ereport(ERROR,
5029 : (errcode_for_file_access(),
5030 ECB : errmsg("could not open file \"%s\": %m", path)));
5031 :
5032 : while (true)
5033 CBC 119 : {
5034 ECB : ReorderBufferTupleCidKey key;
5035 : ReorderBufferTupleCidEnt *ent;
5036 : ReorderBufferTupleCidEnt *new_ent;
5037 : bool found;
5038 :
5039 : /* be careful about padding */
5040 GIC 141 : memset(&key, 0, sizeof(ReorderBufferTupleCidKey));
5041 :
5042 : /* read all mappings till the end of the file */
5043 141 : pgstat_report_wait_start(WAIT_EVENT_REORDER_LOGICAL_MAPPING_READ);
5044 141 : readBytes = read(fd, &map, sizeof(LogicalRewriteMappingData));
5045 141 : pgstat_report_wait_end();
5046 :
5047 141 : if (readBytes < 0)
5048 UIC 0 : ereport(ERROR,
5049 : (errcode_for_file_access(),
5050 : errmsg("could not read file \"%s\": %m",
5051 : path)));
5052 GIC 141 : else if (readBytes == 0) /* EOF */
5053 22 : break;
5054 119 : else if (readBytes != sizeof(LogicalRewriteMappingData))
5055 UIC 0 : ereport(ERROR,
5056 : (errcode_for_file_access(),
5057 : errmsg("could not read from file \"%s\": read %d instead of %d bytes",
5058 : path, readBytes,
5059 : (int32) sizeof(LogicalRewriteMappingData))));
5060 :
5061 GNC 119 : key.rlocator = map.old_locator;
5062 GIC 119 : ItemPointerCopy(&map.old_tid,
5063 : &key.tid);
5064 :
5065 :
5066 : ent = (ReorderBufferTupleCidEnt *)
5067 GNC 119 : hash_search(tuplecid_data, &key, HASH_FIND, NULL);
5068 :
5069 : /* no existing mapping, no need to update */
5070 GIC 119 : if (!ent)
5071 UIC 0 : continue;
5072 :
5073 GNC 119 : key.rlocator = map.new_locator;
5074 GIC 119 : ItemPointerCopy(&map.new_tid,
5075 : &key.tid);
5076 :
5077 : new_ent = (ReorderBufferTupleCidEnt *)
5078 GNC 119 : hash_search(tuplecid_data, &key, HASH_ENTER, &found);
5079 :
5080 GIC 119 : if (found)
5081 : {
5082 : /*
5083 : * Make sure the existing mapping makes sense. We sometime update
5084 : * old records that did not yet have a cmax (e.g. pg_class' own
5085 : * entry while rewriting it) during rewrites, so allow that.
5086 : */
5087 6 : Assert(ent->cmin == InvalidCommandId || ent->cmin == new_ent->cmin);
5088 6 : Assert(ent->cmax == InvalidCommandId || ent->cmax == new_ent->cmax);
5089 : }
5090 : else
5091 : {
5092 : /* update mapping */
5093 113 : new_ent->cmin = ent->cmin;
5094 113 : new_ent->cmax = ent->cmax;
5095 113 : new_ent->combocid = ent->combocid;
5096 : }
5097 : }
5098 :
5099 22 : if (CloseTransientFile(fd) != 0)
5100 UIC 0 : ereport(ERROR,
5101 : (errcode_for_file_access(),
5102 : errmsg("could not close file \"%s\": %m", path)));
5103 GIC 22 : }
5104 ECB :
5105 :
5106 : /*
5107 : * Check whether the TransactionId 'xid' is in the pre-sorted array 'xip'.
5108 : */
5109 : static bool
5110 GIC 290 : TransactionIdInArray(TransactionId xid, TransactionId *xip, Size num)
5111 ECB : {
5112 CBC 290 : return bsearch(&xid, xip, num,
5113 290 : sizeof(TransactionId), xidComparator) != NULL;
5114 EUB : }
5115 :
5116 : /*
5117 : * list_sort() comparator for sorting RewriteMappingFiles in LSN order.
5118 : */
5119 ECB : static int
5120 GIC 32 : file_sort_by_lsn(const ListCell *a_p, const ListCell *b_p)
5121 : {
5122 32 : RewriteMappingFile *a = (RewriteMappingFile *) lfirst(a_p);
5123 32 : RewriteMappingFile *b = (RewriteMappingFile *) lfirst(b_p);
5124 :
5125 32 : if (a->lsn < b->lsn)
5126 CBC 9 : return -1;
5127 GIC 23 : else if (a->lsn > b->lsn)
5128 23 : return 1;
5129 LBC 0 : return 0;
5130 ECB : }
5131 :
5132 : /*
5133 : * Apply any existing logical remapping files if there are any targeted at our
5134 EUB : * transaction for relid.
5135 : */
5136 : static void
5137 GIC 5 : UpdateLogicalMappings(HTAB *tuplecid_data, Oid relid, Snapshot snapshot)
5138 ECB : {
5139 : DIR *mapping_dir;
5140 : struct dirent *mapping_de;
5141 GBC 5 : List *files = NIL;
5142 : ListCell *file;
5143 GIC 5 : Oid dboid = IsSharedRelation(relid) ? InvalidOid : MyDatabaseId;
5144 :
5145 5 : mapping_dir = AllocateDir("pg_logical/mappings");
5146 460 : while ((mapping_de = ReadDir(mapping_dir, "pg_logical/mappings")) != NULL)
5147 ECB : {
5148 : Oid f_dboid;
5149 : Oid f_relid;
5150 : TransactionId f_mapped_xid;
5151 : TransactionId f_create_xid;
5152 : XLogRecPtr f_lsn;
5153 : uint32 f_hi,
5154 : f_lo;
5155 : RewriteMappingFile *f;
5156 :
5157 GBC 455 : if (strcmp(mapping_de->d_name, ".") == 0 ||
5158 GIC 450 : strcmp(mapping_de->d_name, "..") == 0)
5159 CBC 433 : continue;
5160 ECB :
5161 : /* Ignore files that aren't ours */
5162 GIC 445 : if (strncmp(mapping_de->d_name, "map-", 4) != 0)
5163 UIC 0 : continue;
5164 ECB :
5165 GIC 445 : if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
5166 ECB : &f_dboid, &f_relid, &f_hi, &f_lo,
5167 : &f_mapped_xid, &f_create_xid) != 6)
5168 UIC 0 : elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
5169 :
5170 GIC 445 : f_lsn = ((uint64) f_hi) << 32 | f_lo;
5171 :
5172 : /* mapping for another database */
5173 CBC 445 : if (f_dboid != dboid)
5174 LBC 0 : continue;
5175 :
5176 : /* mapping for another relation */
5177 GIC 445 : if (f_relid != relid)
5178 45 : continue;
5179 ECB :
5180 : /* did the creating transaction abort? */
5181 CBC 400 : if (!TransactionIdDidCommit(f_create_xid))
5182 GIC 110 : continue;
5183 :
5184 : /* not for our transaction */
5185 CBC 290 : if (!TransactionIdInArray(f_mapped_xid, snapshot->subxip, snapshot->subxcnt))
5186 GBC 268 : continue;
5187 :
5188 : /* ok, relevant, queue for apply */
5189 CBC 22 : f = palloc(sizeof(RewriteMappingFile));
5190 GIC 22 : f->lsn = f_lsn;
5191 22 : strcpy(f->fname, mapping_de->d_name);
5192 22 : files = lappend(files, f);
5193 : }
5194 5 : FreeDir(mapping_dir);
5195 :
5196 ECB : /* sort files so we apply them in LSN order */
5197 GIC 5 : list_sort(files, file_sort_by_lsn);
5198 ECB :
5199 CBC 27 : foreach(file, files)
5200 : {
5201 GIC 22 : RewriteMappingFile *f = (RewriteMappingFile *) lfirst(file);
5202 :
5203 22 : elog(DEBUG1, "applying mapping: \"%s\" in %u", f->fname,
5204 : snapshot->subxip[0]);
5205 22 : ApplyLogicalMappingFile(tuplecid_data, relid, f->fname);
5206 CBC 22 : pfree(f);
5207 : }
5208 5 : }
5209 ECB :
5210 : /*
5211 : * Lookup cmin/cmax of a tuple, during logical decoding where we can't rely on
5212 : * combo CIDs.
5213 : */
5214 : bool
5215 GBC 601 : ResolveCminCmaxDuringDecoding(HTAB *tuplecid_data,
5216 : Snapshot snapshot,
5217 : HeapTuple htup, Buffer buffer,
5218 : CommandId *cmin, CommandId *cmax)
5219 : {
5220 : ReorderBufferTupleCidKey key;
5221 : ReorderBufferTupleCidEnt *ent;
5222 : ForkNumber forkno;
5223 ECB : BlockNumber blockno;
5224 GIC 601 : bool updated_mapping = false;
5225 :
5226 : /*
5227 ECB : * Return unresolved if tuplecid_data is not valid. That's because when
5228 : * streaming in-progress transactions we may run into tuples with the CID
5229 : * before actually decoding them. Think e.g. about INSERT followed by
5230 : * TRUNCATE, where the TRUNCATE may not be decoded yet when applying the
5231 : * INSERT. So in such cases, we assume the CID is from the future
5232 : * command.
5233 : */
5234 GIC 601 : if (tuplecid_data == NULL)
5235 9 : return false;
5236 :
5237 : /* be careful about padding */
5238 592 : memset(&key, 0, sizeof(key));
5239 :
5240 592 : Assert(!BufferIsLocal(buffer));
5241 :
5242 : /*
5243 : * get relfilelocator from the buffer, no convenient way to access it
5244 : * other than that.
5245 ECB : */
5246 GNC 592 : BufferGetTag(buffer, &key.rlocator, &forkno, &blockno);
5247 :
5248 ECB : /* tuples can only be in the main fork */
5249 GBC 592 : Assert(forkno == MAIN_FORKNUM);
5250 GIC 592 : Assert(blockno == ItemPointerGetBlockNumber(&htup->t_self));
5251 ECB :
5252 GIC 592 : ItemPointerCopy(&htup->t_self,
5253 : &key.tid);
5254 EUB :
5255 GIC 597 : restart:
5256 ECB : ent = (ReorderBufferTupleCidEnt *)
5257 GNC 597 : hash_search(tuplecid_data, &key, HASH_FIND, NULL);
5258 :
5259 : /*
5260 ECB : * failed to find a mapping, check whether the table was rewritten and
5261 : * apply mapping if so, but only do that once - there can be no new
5262 : * mappings while we are in here since we have to hold a lock on the
5263 : * relation.
5264 : */
5265 CBC 597 : if (ent == NULL && !updated_mapping)
5266 : {
5267 GIC 5 : UpdateLogicalMappings(tuplecid_data, htup->t_tableOid, snapshot);
5268 ECB : /* now check but don't update for a mapping again */
5269 CBC 5 : updated_mapping = true;
5270 GIC 5 : goto restart;
5271 : }
5272 CBC 592 : else if (ent == NULL)
5273 LBC 0 : return false;
5274 ECB :
5275 CBC 592 : if (cmin)
5276 GIC 592 : *cmin = ent->cmin;
5277 CBC 592 : if (cmax)
5278 GIC 592 : *cmax = ent->cmax;
5279 592 : return true;
5280 ECB : }
|
