PostgreSQL Source Code  git master
rewriteheap.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * rewriteheap.c
4  * Support functions to rewrite tables.
5  *
6  * These functions provide a facility to completely rewrite a heap, while
7  * preserving visibility information and update chains.
8  *
9  * INTERFACE
10  *
11  * The caller is responsible for creating the new heap, all catalog
12  * changes, supplying the tuples to be written to the new heap, and
13  * rebuilding indexes. The caller must hold AccessExclusiveLock on the
14  * target table, because we assume no one else is writing into it.
15  *
16  * To use the facility:
17  *
18  * begin_heap_rewrite
19  * while (fetch next tuple)
20  * {
21  * if (tuple is dead)
22  * rewrite_heap_dead_tuple
23  * else
24  * {
25  * // do any transformations here if required
26  * rewrite_heap_tuple
27  * }
28  * }
29  * end_heap_rewrite
30  *
31  * The contents of the new relation shouldn't be relied on until after
32  * end_heap_rewrite is called.
33  *
34  *
35  * IMPLEMENTATION
36  *
37  * This would be a fairly trivial affair, except that we need to maintain
38  * the ctid chains that link versions of an updated tuple together.
39  * Since the newly stored tuples will have tids different from the original
40  * ones, if we just copied t_ctid fields to the new table the links would
41  * be wrong. When we are required to copy a (presumably recently-dead or
42  * delete-in-progress) tuple whose ctid doesn't point to itself, we have
43  * to substitute the correct ctid instead.
44  *
45  * For each ctid reference from A -> B, we might encounter either A first
46  * or B first. (Note that a tuple in the middle of a chain is both A and B
47  * of different pairs.)
48  *
49  * If we encounter A first, we'll store the tuple in the unresolved_tups
50  * hash table. When we later encounter B, we remove A from the hash table,
51  * fix the ctid to point to the new location of B, and insert both A and B
52  * to the new heap.
53  *
54  * If we encounter B first, we can insert B to the new heap right away.
55  * We then add an entry to the old_new_tid_map hash table showing B's
56  * original tid (in the old heap) and new tid (in the new heap).
57  * When we later encounter A, we get the new location of B from the table,
58  * and can write A immediately with the correct ctid.
59  *
60  * Entries in the hash tables can be removed as soon as the later tuple
61  * is encountered. That helps to keep the memory usage down. At the end,
62  * both tables are usually empty; we should have encountered both A and B
63  * of each pair. However, it's possible for A to be RECENTLY_DEAD and B
64  * entirely DEAD according to HeapTupleSatisfiesVacuum, because the test
65  * for deadness using OldestXmin is not exact. In such a case we might
66  * encounter B first, and skip it, and find A later. Then A would be added
67  * to unresolved_tups, and stay there until end of the rewrite. Since
68  * this case is very unusual, we don't worry about the memory usage.
69  *
70  * Using in-memory hash tables means that we use some memory for each live
71  * update chain in the table, from the time we find one end of the
72  * reference until we find the other end. That shouldn't be a problem in
73  * practice, but if you do something like an UPDATE without a where-clause
74  * on a large table, and then run CLUSTER in the same transaction, you
75  * could run out of memory. It doesn't seem worthwhile to add support for
76  * spill-to-disk, as there shouldn't be that many RECENTLY_DEAD tuples in a
77  * table under normal circumstances. Furthermore, in the typical scenario
78  * of CLUSTERing on an unchanging key column, we'll see all the versions
79  * of a given tuple together anyway, and so the peak memory usage is only
80  * proportional to the number of RECENTLY_DEAD versions of a single row, not
81  * in the whole table. Note that if we do fail halfway through a CLUSTER,
82  * the old table is still valid, so failure is not catastrophic.
83  *
84  * We can't use the normal heap_insert function to insert into the new
85  * heap, because heap_insert overwrites the visibility information.
86  * We use a special-purpose raw_heap_insert function instead, which
87  * is optimized for bulk inserting a lot of tuples, knowing that we have
88  * exclusive access to the heap. raw_heap_insert builds new pages in
89  * local storage. When a page is full, or at the end of the process,
90  * we insert it to WAL as a single record and then write it to disk
91  * directly through smgr. Note, however, that any data sent to the new
92  * heap's TOAST table will go through the normal bufmgr.
93  *
94  *
95  * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
96  * Portions Copyright (c) 1994-5, Regents of the University of California
97  *
98  * IDENTIFICATION
99  * src/backend/access/heap/rewriteheap.c
100  *
101  *-------------------------------------------------------------------------
102  */
103 #include "postgres.h"
104 
105 #include <sys/stat.h>
106 #include <unistd.h>
107 
108 #include "access/heapam.h"
109 #include "access/heapam_xlog.h"
110 #include "access/heaptoast.h"
111 #include "access/rewriteheap.h"
112 #include "access/transam.h"
113 #include "access/xact.h"
114 #include "access/xloginsert.h"
115 #include "catalog/catalog.h"
116 #include "lib/ilist.h"
117 #include "miscadmin.h"
118 #include "pgstat.h"
119 #include "replication/logical.h"
120 #include "replication/slot.h"
121 #include "storage/bufmgr.h"
122 #include "storage/fd.h"
123 #include "storage/procarray.h"
124 #include "storage/smgr.h"
125 #include "utils/memutils.h"
126 #include "utils/rel.h"
127 
128 /*
129  * State associated with a rewrite operation. This is opaque to the user
130  * of the rewrite facility.
131  */
132 typedef struct RewriteStateData
133 {
134  Relation rs_old_rel; /* source heap */
135  Relation rs_new_rel; /* destination heap */
136  Page rs_buffer; /* page currently being built */
137  BlockNumber rs_blockno; /* block where page will go */
138  bool rs_buffer_valid; /* T if any tuples in buffer */
139  bool rs_use_wal; /* must we WAL-log inserts? */
140  bool rs_logical_rewrite; /* do we need to do logical rewriting */
141  TransactionId rs_oldest_xmin; /* oldest xmin used by caller to determine
142  * tuple visibility */
143  TransactionId rs_freeze_xid; /* Xid that will be used as freeze cutoff
144  * point */
145  TransactionId rs_logical_xmin; /* Xid that will be used as cutoff point
146  * for logical rewrites */
147  MultiXactId rs_cutoff_multi; /* MultiXactId that will be used as cutoff
148  * point for multixacts */
149  MemoryContext rs_cxt; /* for hash tables and entries and tuples in
150  * them */
151  XLogRecPtr rs_begin_lsn; /* XLogInsertLsn when starting the rewrite */
152  HTAB *rs_unresolved_tups; /* unmatched A tuples */
153  HTAB *rs_old_new_tid_map; /* unmatched B tuples */
154  HTAB *rs_logical_mappings; /* logical remapping files */
155  uint32 rs_num_rewrite_mappings; /* # in memory mappings */
157 
158 /*
159  * The lookup keys for the hash tables are tuple TID and xmin (we must check
160  * both to avoid false matches from dead tuples). Beware that there is
161  * probably some padding space in this struct; it must be zeroed out for
162  * correct hashtable operation.
163  */
164 typedef struct
165 {
166  TransactionId xmin; /* tuple xmin */
167  ItemPointerData tid; /* tuple location in old heap */
168 } TidHashKey;
169 
170 /*
171  * Entry structures for the hash tables
172  */
173 typedef struct
174 {
175  TidHashKey key; /* expected xmin/old location of B tuple */
176  ItemPointerData old_tid; /* A's location in the old heap */
177  HeapTuple tuple; /* A's tuple contents */
179 
181 
182 typedef struct
183 {
184  TidHashKey key; /* actual xmin/old location of B tuple */
185  ItemPointerData new_tid; /* where we put it in the new heap */
187 
189 
190 /*
191  * In-Memory data for an xid that might need logical remapping entries
192  * to be logged.
193  */
194 typedef struct RewriteMappingFile
195 {
196  TransactionId xid; /* xid that might need to see the row */
197  int vfd; /* fd of mappings file */
198  off_t off; /* how far have we written yet */
199  uint32 num_mappings; /* number of in-memory mappings */
200  dlist_head mappings; /* list of in-memory mappings */
201  char path[MAXPGPATH]; /* path, for error messages */
203 
204 /*
205  * A single In-Memory logical rewrite mapping, hanging off
206  * RewriteMappingFile->mappings.
207  */
209 {
210  LogicalRewriteMappingData map; /* map between old and new location of the
211  * tuple */
214 
215 
216 /* prototypes for internal functions */
217 static void raw_heap_insert(RewriteState state, HeapTuple tup);
218 
219 /* internal logical remapping prototypes */
223 
224 
225 /*
226  * Begin a rewrite of a table
227  *
228  * old_heap old, locked heap relation tuples will be read from
229  * new_heap new, locked heap relation to insert tuples to
230  * oldest_xmin xid used by the caller to determine which tuples are dead
231  * freeze_xid xid before which tuples will be frozen
232  * cutoff_multi multixact before which multis will be removed
233  * use_wal should the inserts to the new heap be WAL-logged?
234  *
235  * Returns an opaque RewriteState, allocated in current memory context,
236  * to be used in subsequent calls to the other functions.
237  */
239 begin_heap_rewrite(Relation old_heap, Relation new_heap, TransactionId oldest_xmin,
240  TransactionId freeze_xid, MultiXactId cutoff_multi,
241  bool use_wal)
242 {
244  MemoryContext rw_cxt;
245  MemoryContext old_cxt;
246  HASHCTL hash_ctl;
247 
248  /*
249  * To ease cleanup, make a separate context that will contain the
250  * RewriteState struct itself plus all subsidiary data.
251  */
253  "Table rewrite",
255  old_cxt = MemoryContextSwitchTo(rw_cxt);
256 
257  /* Create and fill in the state struct */
258  state = palloc0(sizeof(RewriteStateData));
259 
260  state->rs_old_rel = old_heap;
261  state->rs_new_rel = new_heap;
262  state->rs_buffer = (Page) palloc(BLCKSZ);
263  /* new_heap needn't be empty, just locked */
264  state->rs_blockno = RelationGetNumberOfBlocks(new_heap);
265  state->rs_buffer_valid = false;
266  state->rs_use_wal = use_wal;
267  state->rs_oldest_xmin = oldest_xmin;
268  state->rs_freeze_xid = freeze_xid;
269  state->rs_cutoff_multi = cutoff_multi;
270  state->rs_cxt = rw_cxt;
271 
272  /* Initialize hash tables used to track update chains */
273  memset(&hash_ctl, 0, sizeof(hash_ctl));
274  hash_ctl.keysize = sizeof(TidHashKey);
275  hash_ctl.entrysize = sizeof(UnresolvedTupData);
276  hash_ctl.hcxt = state->rs_cxt;
277 
278  state->rs_unresolved_tups =
279  hash_create("Rewrite / Unresolved ctids",
280  128, /* arbitrary initial size */
281  &hash_ctl,
283 
284  hash_ctl.entrysize = sizeof(OldToNewMappingData);
285 
286  state->rs_old_new_tid_map =
287  hash_create("Rewrite / Old to new tid map",
288  128, /* arbitrary initial size */
289  &hash_ctl,
291 
292  MemoryContextSwitchTo(old_cxt);
293 
295 
296  return state;
297 }
298 
299 /*
300  * End a rewrite.
301  *
302  * state and any other resources are freed.
303  */
304 void
306 {
307  HASH_SEQ_STATUS seq_status;
308  UnresolvedTup unresolved;
309 
310  /*
311  * Write any remaining tuples in the UnresolvedTups table. If we have any
312  * left, they should in fact be dead, but let's err on the safe side.
313  */
314  hash_seq_init(&seq_status, state->rs_unresolved_tups);
315 
316  while ((unresolved = hash_seq_search(&seq_status)) != NULL)
317  {
318  ItemPointerSetInvalid(&unresolved->tuple->t_data->t_ctid);
319  raw_heap_insert(state, unresolved->tuple);
320  }
321 
322  /* Write the last page, if any */
323  if (state->rs_buffer_valid)
324  {
325  if (state->rs_use_wal)
326  log_newpage(&state->rs_new_rel->rd_node,
327  MAIN_FORKNUM,
328  state->rs_blockno,
329  state->rs_buffer,
330  true);
332 
334 
336  (char *) state->rs_buffer, true);
337  }
338 
339  /*
340  * If the rel is WAL-logged, must fsync before commit. We use heap_sync
341  * to ensure that the toast table gets fsync'd too.
342  *
343  * It's obvious that we must do this when not WAL-logging. It's less
344  * obvious that we have to do it even if we did WAL-log the pages. The
345  * reason is the same as in storage.c's RelationCopyStorage(): we're
346  * writing data that's not in shared buffers, and so a CHECKPOINT
347  * occurring during the rewriteheap operation won't have fsync'd data we
348  * wrote before the checkpoint.
349  */
350  if (RelationNeedsWAL(state->rs_new_rel))
351  heap_sync(state->rs_new_rel);
352 
354 
355  /* Deleting the context frees everything */
356  MemoryContextDelete(state->rs_cxt);
357 }
358 
359 /*
360  * Add a tuple to the new heap.
361  *
362  * Visibility information is copied from the original tuple, except that
363  * we "freeze" very-old tuples. Note that since we scribble on new_tuple,
364  * it had better be temp storage not a pointer to the original tuple.
365  *
366  * state opaque state as returned by begin_heap_rewrite
367  * old_tuple original tuple in the old heap
368  * new_tuple new, rewritten tuple to be inserted to new heap
369  */
370 void
372  HeapTuple old_tuple, HeapTuple new_tuple)
373 {
374  MemoryContext old_cxt;
375  ItemPointerData old_tid;
376  TidHashKey hashkey;
377  bool found;
378  bool free_new;
379 
380  old_cxt = MemoryContextSwitchTo(state->rs_cxt);
381 
382  /*
383  * Copy the original tuple's visibility information into new_tuple.
384  *
385  * XXX we might later need to copy some t_infomask2 bits, too? Right now,
386  * we intentionally clear the HOT status bits.
387  */
388  memcpy(&new_tuple->t_data->t_choice.t_heap,
389  &old_tuple->t_data->t_choice.t_heap,
390  sizeof(HeapTupleFields));
391 
392  new_tuple->t_data->t_infomask &= ~HEAP_XACT_MASK;
393  new_tuple->t_data->t_infomask2 &= ~HEAP2_XACT_MASK;
394  new_tuple->t_data->t_infomask |=
395  old_tuple->t_data->t_infomask & HEAP_XACT_MASK;
396 
397  /*
398  * While we have our hands on the tuple, we may as well freeze any
399  * eligible xmin or xmax, so that future VACUUM effort can be saved.
400  */
401  heap_freeze_tuple(new_tuple->t_data,
402  state->rs_old_rel->rd_rel->relfrozenxid,
403  state->rs_old_rel->rd_rel->relminmxid,
404  state->rs_freeze_xid,
405  state->rs_cutoff_multi);
406 
407  /*
408  * Invalid ctid means that ctid should point to the tuple itself. We'll
409  * override it later if the tuple is part of an update chain.
410  */
411  ItemPointerSetInvalid(&new_tuple->t_data->t_ctid);
412 
413  /*
414  * If the tuple has been updated, check the old-to-new mapping hash table.
415  */
416  if (!((old_tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
417  HeapTupleHeaderIsOnlyLocked(old_tuple->t_data)) &&
419  !(ItemPointerEquals(&(old_tuple->t_self),
420  &(old_tuple->t_data->t_ctid))))
421  {
422  OldToNewMapping mapping;
423 
424  memset(&hashkey, 0, sizeof(hashkey));
425  hashkey.xmin = HeapTupleHeaderGetUpdateXid(old_tuple->t_data);
426  hashkey.tid = old_tuple->t_data->t_ctid;
427 
428  mapping = (OldToNewMapping)
429  hash_search(state->rs_old_new_tid_map, &hashkey,
430  HASH_FIND, NULL);
431 
432  if (mapping != NULL)
433  {
434  /*
435  * We've already copied the tuple that t_ctid points to, so we can
436  * set the ctid of this tuple to point to the new location, and
437  * insert it right away.
438  */
439  new_tuple->t_data->t_ctid = mapping->new_tid;
440 
441  /* We don't need the mapping entry anymore */
442  hash_search(state->rs_old_new_tid_map, &hashkey,
443  HASH_REMOVE, &found);
444  Assert(found);
445  }
446  else
447  {
448  /*
449  * We haven't seen the tuple t_ctid points to yet. Stash this
450  * tuple into unresolved_tups to be written later.
451  */
452  UnresolvedTup unresolved;
453 
454  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
455  HASH_ENTER, &found);
456  Assert(!found);
457 
458  unresolved->old_tid = old_tuple->t_self;
459  unresolved->tuple = heap_copytuple(new_tuple);
460 
461  /*
462  * We can't do anything more now, since we don't know where the
463  * tuple will be written.
464  */
465  MemoryContextSwitchTo(old_cxt);
466  return;
467  }
468  }
469 
470  /*
471  * Now we will write the tuple, and then check to see if it is the B tuple
472  * in any new or known pair. When we resolve a known pair, we will be
473  * able to write that pair's A tuple, and then we have to check if it
474  * resolves some other pair. Hence, we need a loop here.
475  */
476  old_tid = old_tuple->t_self;
477  free_new = false;
478 
479  for (;;)
480  {
481  ItemPointerData new_tid;
482 
483  /* Insert the tuple and find out where it's put in new_heap */
484  raw_heap_insert(state, new_tuple);
485  new_tid = new_tuple->t_self;
486 
487  logical_rewrite_heap_tuple(state, old_tid, new_tuple);
488 
489  /*
490  * If the tuple is the updated version of a row, and the prior version
491  * wouldn't be DEAD yet, then we need to either resolve the prior
492  * version (if it's waiting in rs_unresolved_tups), or make an entry
493  * in rs_old_new_tid_map (so we can resolve it when we do see it). The
494  * previous tuple's xmax would equal this one's xmin, so it's
495  * RECENTLY_DEAD if and only if the xmin is not before OldestXmin.
496  */
497  if ((new_tuple->t_data->t_infomask & HEAP_UPDATED) &&
499  state->rs_oldest_xmin))
500  {
501  /*
502  * Okay, this is B in an update pair. See if we've seen A.
503  */
504  UnresolvedTup unresolved;
505 
506  memset(&hashkey, 0, sizeof(hashkey));
507  hashkey.xmin = HeapTupleHeaderGetXmin(new_tuple->t_data);
508  hashkey.tid = old_tid;
509 
510  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
511  HASH_FIND, NULL);
512 
513  if (unresolved != NULL)
514  {
515  /*
516  * We have seen and memorized the previous tuple already. Now
517  * that we know where we inserted the tuple its t_ctid points
518  * to, fix its t_ctid and insert it to the new heap.
519  */
520  if (free_new)
521  heap_freetuple(new_tuple);
522  new_tuple = unresolved->tuple;
523  free_new = true;
524  old_tid = unresolved->old_tid;
525  new_tuple->t_data->t_ctid = new_tid;
526 
527  /*
528  * We don't need the hash entry anymore, but don't free its
529  * tuple just yet.
530  */
531  hash_search(state->rs_unresolved_tups, &hashkey,
532  HASH_REMOVE, &found);
533  Assert(found);
534 
535  /* loop back to insert the previous tuple in the chain */
536  continue;
537  }
538  else
539  {
540  /*
541  * Remember the new tid of this tuple. We'll use it to set the
542  * ctid when we find the previous tuple in the chain.
543  */
544  OldToNewMapping mapping;
545 
546  mapping = hash_search(state->rs_old_new_tid_map, &hashkey,
547  HASH_ENTER, &found);
548  Assert(!found);
549 
550  mapping->new_tid = new_tid;
551  }
552  }
553 
554  /* Done with this (chain of) tuples, for now */
555  if (free_new)
556  heap_freetuple(new_tuple);
557  break;
558  }
559 
560  MemoryContextSwitchTo(old_cxt);
561 }
562 
563 /*
564  * Register a dead tuple with an ongoing rewrite. Dead tuples are not
565  * copied to the new table, but we still make note of them so that we
566  * can release some resources earlier.
567  *
568  * Returns true if a tuple was removed from the unresolved_tups table.
569  * This indicates that that tuple, previously thought to be "recently dead",
570  * is now known really dead and won't be written to the output.
571  */
572 bool
574 {
575  /*
576  * If we have already seen an earlier tuple in the update chain that
577  * points to this tuple, let's forget about that earlier tuple. It's in
578  * fact dead as well, our simple xmax < OldestXmin test in
579  * HeapTupleSatisfiesVacuum just wasn't enough to detect it. It happens
580  * when xmin of a tuple is greater than xmax, which sounds
581  * counter-intuitive but is perfectly valid.
582  *
583  * We don't bother to try to detect the situation the other way round,
584  * when we encounter the dead tuple first and then the recently dead one
585  * that points to it. If that happens, we'll have some unmatched entries
586  * in the UnresolvedTups hash table at the end. That can happen anyway,
587  * because a vacuum might have removed the dead tuple in the chain before
588  * us.
589  */
590  UnresolvedTup unresolved;
591  TidHashKey hashkey;
592  bool found;
593 
594  memset(&hashkey, 0, sizeof(hashkey));
595  hashkey.xmin = HeapTupleHeaderGetXmin(old_tuple->t_data);
596  hashkey.tid = old_tuple->t_self;
597 
598  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
599  HASH_FIND, NULL);
600 
601  if (unresolved != NULL)
602  {
603  /* Need to free the contained tuple as well as the hashtable entry */
604  heap_freetuple(unresolved->tuple);
605  hash_search(state->rs_unresolved_tups, &hashkey,
606  HASH_REMOVE, &found);
607  Assert(found);
608  return true;
609  }
610 
611  return false;
612 }
613 
614 /*
615  * Insert a tuple to the new relation. This has to track heap_insert
616  * and its subsidiary functions!
617  *
618  * t_self of the tuple is set to the new TID of the tuple. If t_ctid of the
619  * tuple is invalid on entry, it's replaced with the new TID as well (in
620  * the inserted data only, not in the caller's copy).
621  */
622 static void
624 {
625  Page page = state->rs_buffer;
626  Size pageFreeSpace,
627  saveFreeSpace;
628  Size len;
629  OffsetNumber newoff;
630  HeapTuple heaptup;
631 
632  /*
633  * If the new tuple is too big for storage or contains already toasted
634  * out-of-line attributes from some other relation, invoke the toaster.
635  *
636  * Note: below this point, heaptup is the data we actually intend to store
637  * into the relation; tup is the caller's original untoasted data.
638  */
639  if (state->rs_new_rel->rd_rel->relkind == RELKIND_TOASTVALUE)
640  {
641  /* toast table entries should never be recursively toasted */
643  heaptup = tup;
644  }
645  else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
646  {
648 
649  if (!state->rs_use_wal)
650  options |= HEAP_INSERT_SKIP_WAL;
651 
652  /*
653  * While rewriting the heap for VACUUM FULL / CLUSTER, make sure data
654  * for the TOAST table are not logically decoded. The main heap is
655  * WAL-logged as XLOG FPI records, which are not logically decoded.
656  */
657  options |= HEAP_INSERT_NO_LOGICAL;
658 
659  heaptup = heap_toast_insert_or_update(state->rs_new_rel, tup, NULL,
660  options);
661  }
662  else
663  heaptup = tup;
664 
665  len = MAXALIGN(heaptup->t_len); /* be conservative */
666 
667  /*
668  * If we're gonna fail for oversize tuple, do it right away
669  */
670  if (len > MaxHeapTupleSize)
671  ereport(ERROR,
672  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
673  errmsg("row is too big: size %zu, maximum size %zu",
674  len, MaxHeapTupleSize)));
675 
676  /* Compute desired extra freespace due to fillfactor option */
677  saveFreeSpace = RelationGetTargetPageFreeSpace(state->rs_new_rel,
679 
680  /* Now we can check to see if there's enough free space already. */
681  if (state->rs_buffer_valid)
682  {
683  pageFreeSpace = PageGetHeapFreeSpace(page);
684 
685  if (len + saveFreeSpace > pageFreeSpace)
686  {
687  /* Doesn't fit, so write out the existing page */
688 
689  /* XLOG stuff */
690  if (state->rs_use_wal)
691  log_newpage(&state->rs_new_rel->rd_node,
692  MAIN_FORKNUM,
693  state->rs_blockno,
694  page,
695  true);
696 
697  /*
698  * Now write the page. We say skipFsync = true because there's no
699  * need for smgr to schedule an fsync for this write; we'll do it
700  * ourselves in end_heap_rewrite.
701  */
703 
704  PageSetChecksumInplace(page, state->rs_blockno);
705 
707  state->rs_blockno, (char *) page, true);
708 
709  state->rs_blockno++;
710  state->rs_buffer_valid = false;
711  }
712  }
713 
714  if (!state->rs_buffer_valid)
715  {
716  /* Initialize a new empty page */
717  PageInit(page, BLCKSZ, 0);
718  state->rs_buffer_valid = true;
719  }
720 
721  /* And now we can insert the tuple into the page */
722  newoff = PageAddItem(page, (Item) heaptup->t_data, heaptup->t_len,
723  InvalidOffsetNumber, false, true);
724  if (newoff == InvalidOffsetNumber)
725  elog(ERROR, "failed to add tuple");
726 
727  /* Update caller's t_self to the actual position where it was stored */
728  ItemPointerSet(&(tup->t_self), state->rs_blockno, newoff);
729 
730  /*
731  * Insert the correct position into CTID of the stored tuple, too, if the
732  * caller didn't supply a valid CTID.
733  */
734  if (!ItemPointerIsValid(&tup->t_data->t_ctid))
735  {
736  ItemId newitemid;
737  HeapTupleHeader onpage_tup;
738 
739  newitemid = PageGetItemId(page, newoff);
740  onpage_tup = (HeapTupleHeader) PageGetItem(page, newitemid);
741 
742  onpage_tup->t_ctid = tup->t_self;
743  }
744 
745  /* If heaptup is a private copy, release it. */
746  if (heaptup != tup)
747  heap_freetuple(heaptup);
748 }
749 
750 /* ------------------------------------------------------------------------
751  * Logical rewrite support
752  *
753  * When doing logical decoding - which relies on using cmin/cmax of catalog
754  * tuples, via xl_heap_new_cid records - heap rewrites have to log enough
755  * information to allow the decoding backend to updates its internal mapping
756  * of (relfilenode,ctid) => (cmin, cmax) to be correct for the rewritten heap.
757  *
758  * For that, every time we find a tuple that's been modified in a catalog
759  * relation within the xmin horizon of any decoding slot, we log a mapping
760  * from the old to the new location.
761  *
762  * To deal with rewrites that abort the filename of a mapping file contains
763  * the xid of the transaction performing the rewrite, which then can be
764  * checked before being read in.
765  *
766  * For efficiency we don't immediately spill every single map mapping for a
767  * row to disk but only do so in batches when we've collected several of them
768  * in memory or when end_heap_rewrite() has been called.
769  *
770  * Crash-Safety: This module diverts from the usual patterns of doing WAL
771  * since it cannot rely on checkpoint flushing out all buffers and thus
772  * waiting for exclusive locks on buffers. Usually the XLogInsert() covering
773  * buffer modifications is performed while the buffer(s) that are being
774  * modified are exclusively locked guaranteeing that both the WAL record and
775  * the modified heap are on either side of the checkpoint. But since the
776  * mapping files we log aren't in shared_buffers that interlock doesn't work.
777  *
778  * Instead we simply write the mapping files out to disk, *before* the
779  * XLogInsert() is performed. That guarantees that either the XLogInsert() is
780  * inserted after the checkpoint's redo pointer or that the checkpoint (via
781  * CheckPointLogicalRewriteHeap()) has flushed the (partial) mapping file to
782  * disk. That leaves the tail end that has not yet been flushed open to
783  * corruption, which is solved by including the current offset in the
784  * xl_heap_rewrite_mapping records and truncating the mapping file to it
785  * during replay. Every time a rewrite is finished all generated mapping files
786  * are synced to disk.
787  *
788  * Note that if we were only concerned about crash safety we wouldn't have to
789  * deal with WAL logging at all - an fsync() at the end of a rewrite would be
790  * sufficient for crash safety. Any mapping that hasn't been safely flushed to
791  * disk has to be by an aborted (explicitly or via a crash) transaction and is
792  * ignored by virtue of the xid in its name being subject to a
793  * TransactionDidCommit() check. But we want to support having standbys via
794  * physical replication, both for availability and to do logical decoding
795  * there.
796  * ------------------------------------------------------------------------
797  */
798 
799 /*
800  * Do preparations for logging logical mappings during a rewrite if
801  * necessary. If we detect that we don't need to log anything we'll prevent
802  * any further action by the various logical rewrite functions.
803  */
804 static void
806 {
807  HASHCTL hash_ctl;
808  TransactionId logical_xmin;
809 
810  /*
811  * We only need to persist these mappings if the rewritten table can be
812  * accessed during logical decoding, if not, we can skip doing any
813  * additional work.
814  */
815  state->rs_logical_rewrite =
817 
818  if (!state->rs_logical_rewrite)
819  return;
820 
821  ProcArrayGetReplicationSlotXmin(NULL, &logical_xmin);
822 
823  /*
824  * If there are no logical slots in progress we don't need to do anything,
825  * there cannot be any remappings for relevant rows yet. The relation's
826  * lock protects us against races.
827  */
828  if (logical_xmin == InvalidTransactionId)
829  {
830  state->rs_logical_rewrite = false;
831  return;
832  }
833 
834  state->rs_logical_xmin = logical_xmin;
836  state->rs_num_rewrite_mappings = 0;
837 
838  memset(&hash_ctl, 0, sizeof(hash_ctl));
839  hash_ctl.keysize = sizeof(TransactionId);
840  hash_ctl.entrysize = sizeof(RewriteMappingFile);
841  hash_ctl.hcxt = state->rs_cxt;
842 
843  state->rs_logical_mappings =
844  hash_create("Logical rewrite mapping",
845  128, /* arbitrary initial size */
846  &hash_ctl,
848 }
849 
850 /*
851  * Flush all logical in-memory mappings to disk, but don't fsync them yet.
852  */
853 static void
855 {
856  HASH_SEQ_STATUS seq_status;
857  RewriteMappingFile *src;
858  dlist_mutable_iter iter;
859 
860  Assert(state->rs_logical_rewrite);
861 
862  /* no logical rewrite in progress, no need to iterate over mappings */
863  if (state->rs_num_rewrite_mappings == 0)
864  return;
865 
866  elog(DEBUG1, "flushing %u logical rewrite mapping entries",
867  state->rs_num_rewrite_mappings);
868 
869  hash_seq_init(&seq_status, state->rs_logical_mappings);
870  while ((src = (RewriteMappingFile *) hash_seq_search(&seq_status)) != NULL)
871  {
872  char *waldata;
873  char *waldata_start;
875  Oid dboid;
876  uint32 len;
877  int written;
878 
879  /* this file hasn't got any new mappings */
880  if (src->num_mappings == 0)
881  continue;
882 
883  if (state->rs_old_rel->rd_rel->relisshared)
884  dboid = InvalidOid;
885  else
886  dboid = MyDatabaseId;
887 
888  xlrec.num_mappings = src->num_mappings;
889  xlrec.mapped_rel = RelationGetRelid(state->rs_old_rel);
890  xlrec.mapped_xid = src->xid;
891  xlrec.mapped_db = dboid;
892  xlrec.offset = src->off;
893  xlrec.start_lsn = state->rs_begin_lsn;
894 
895  /* write all mappings consecutively */
896  len = src->num_mappings * sizeof(LogicalRewriteMappingData);
897  waldata_start = waldata = palloc(len);
898 
899  /*
900  * collect data we need to write out, but don't modify ondisk data yet
901  */
902  dlist_foreach_modify(iter, &src->mappings)
903  {
905 
906  pmap = dlist_container(RewriteMappingDataEntry, node, iter.cur);
907 
908  memcpy(waldata, &pmap->map, sizeof(pmap->map));
909  waldata += sizeof(pmap->map);
910 
911  /* remove from the list and free */
912  dlist_delete(&pmap->node);
913  pfree(pmap);
914 
915  /* update bookkeeping */
916  state->rs_num_rewrite_mappings--;
917  src->num_mappings--;
918  }
919 
920  Assert(src->num_mappings == 0);
921  Assert(waldata == waldata_start + len);
922 
923  /*
924  * Note that we deviate from the usual WAL coding practices here,
925  * check the above "Logical rewrite support" comment for reasoning.
926  */
927  written = FileWrite(src->vfd, waldata_start, len, src->off,
929  if (written != len)
930  ereport(ERROR,
932  errmsg("could not write to file \"%s\", wrote %d of %d: %m", src->path,
933  written, len)));
934  src->off += len;
935 
936  XLogBeginInsert();
937  XLogRegisterData((char *) (&xlrec), sizeof(xlrec));
938  XLogRegisterData(waldata_start, len);
939 
940  /* write xlog record */
941  XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_REWRITE);
942 
943  pfree(waldata_start);
944  }
945  Assert(state->rs_num_rewrite_mappings == 0);
946 }
947 
948 /*
949  * Logical remapping part of end_heap_rewrite().
950  */
951 static void
953 {
954  HASH_SEQ_STATUS seq_status;
955  RewriteMappingFile *src;
956 
957  /* done, no logical rewrite in progress */
958  if (!state->rs_logical_rewrite)
959  return;
960 
961  /* writeout remaining in-memory entries */
962  if (state->rs_num_rewrite_mappings > 0)
964 
965  /* Iterate over all mappings we have written and fsync the files. */
966  hash_seq_init(&seq_status, state->rs_logical_mappings);
967  while ((src = (RewriteMappingFile *) hash_seq_search(&seq_status)) != NULL)
968  {
972  errmsg("could not fsync file \"%s\": %m", src->path)));
973  FileClose(src->vfd);
974  }
975  /* memory context cleanup will deal with the rest */
976 }
977 
978 /*
979  * Log a single (old->new) mapping for 'xid'.
980  */
981 static void
984 {
985  RewriteMappingFile *src;
987  Oid relid;
988  bool found;
989 
990  relid = RelationGetRelid(state->rs_old_rel);
991 
992  /* look for existing mappings for this 'mapped' xid */
993  src = hash_search(state->rs_logical_mappings, &xid,
994  HASH_ENTER, &found);
995 
996  /*
997  * We haven't yet had the need to map anything for this xid, create
998  * per-xid data structures.
999  */
1000  if (!found)
1001  {
1002  char path[MAXPGPATH];
1003  Oid dboid;
1004 
1005  if (state->rs_old_rel->rd_rel->relisshared)
1006  dboid = InvalidOid;
1007  else
1008  dboid = MyDatabaseId;
1009 
1010  snprintf(path, MAXPGPATH,
1011  "pg_logical/mappings/" LOGICAL_REWRITE_FORMAT,
1012  dboid, relid,
1013  (uint32) (state->rs_begin_lsn >> 32),
1014  (uint32) state->rs_begin_lsn,
1015  xid, GetCurrentTransactionId());
1016 
1017  dlist_init(&src->mappings);
1018  src->num_mappings = 0;
1019  src->off = 0;
1020  memcpy(src->path, path, sizeof(path));
1021  src->vfd = PathNameOpenFile(path,
1022  O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
1023  if (src->vfd < 0)
1024  ereport(ERROR,
1026  errmsg("could not create file \"%s\": %m", path)));
1027  }
1028 
1029  pmap = MemoryContextAlloc(state->rs_cxt,
1030  sizeof(RewriteMappingDataEntry));
1031  memcpy(&pmap->map, map, sizeof(LogicalRewriteMappingData));
1032  dlist_push_tail(&src->mappings, &pmap->node);
1033  src->num_mappings++;
1034  state->rs_num_rewrite_mappings++;
1035 
1036  /*
1037  * Write out buffer every time we've too many in-memory entries across all
1038  * mapping files.
1039  */
1040  if (state->rs_num_rewrite_mappings >= 1000 /* arbitrary number */ )
1042 }
1043 
1044 /*
1045  * Perform logical remapping for a tuple that's mapped from old_tid to
1046  * new_tuple->t_self by rewrite_heap_tuple() if necessary for the tuple.
1047  */
1048 static void
1050  HeapTuple new_tuple)
1051 {
1052  ItemPointerData new_tid = new_tuple->t_self;
1053  TransactionId cutoff = state->rs_logical_xmin;
1054  TransactionId xmin;
1055  TransactionId xmax;
1056  bool do_log_xmin = false;
1057  bool do_log_xmax = false;
1059 
1060  /* no logical rewrite in progress, we don't need to log anything */
1061  if (!state->rs_logical_rewrite)
1062  return;
1063 
1064  xmin = HeapTupleHeaderGetXmin(new_tuple->t_data);
1065  /* use *GetUpdateXid to correctly deal with multixacts */
1066  xmax = HeapTupleHeaderGetUpdateXid(new_tuple->t_data);
1067 
1068  /*
1069  * Log the mapping iff the tuple has been created recently.
1070  */
1071  if (TransactionIdIsNormal(xmin) && !TransactionIdPrecedes(xmin, cutoff))
1072  do_log_xmin = true;
1073 
1074  if (!TransactionIdIsNormal(xmax))
1075  {
1076  /*
1077  * no xmax is set, can't have any permanent ones, so this check is
1078  * sufficient
1079  */
1080  }
1081  else if (HEAP_XMAX_IS_LOCKED_ONLY(new_tuple->t_data->t_infomask))
1082  {
1083  /* only locked, we don't care */
1084  }
1085  else if (!TransactionIdPrecedes(xmax, cutoff))
1086  {
1087  /* tuple has been deleted recently, log */
1088  do_log_xmax = true;
1089  }
1090 
1091  /* if neither needs to be logged, we're done */
1092  if (!do_log_xmin && !do_log_xmax)
1093  return;
1094 
1095  /* fill out mapping information */
1096  map.old_node = state->rs_old_rel->rd_node;
1097  map.old_tid = old_tid;
1098  map.new_node = state->rs_new_rel->rd_node;
1099  map.new_tid = new_tid;
1100 
1101  /* ---
1102  * Now persist the mapping for the individual xids that are affected. We
1103  * need to log for both xmin and xmax if they aren't the same transaction
1104  * since the mapping files are per "affected" xid.
1105  * We don't muster all that much effort detecting whether xmin and xmax
1106  * are actually the same transaction, we just check whether the xid is the
1107  * same disregarding subtransactions. Logging too much is relatively
1108  * harmless and we could never do the check fully since subtransaction
1109  * data is thrown away during restarts.
1110  * ---
1111  */
1112  if (do_log_xmin)
1113  logical_rewrite_log_mapping(state, xmin, &map);
1114  /* separately log mapping for xmax unless it'd be redundant */
1115  if (do_log_xmax && !TransactionIdEquals(xmin, xmax))
1116  logical_rewrite_log_mapping(state, xmax, &map);
1117 }
1118 
1119 /*
1120  * Replay XLOG_HEAP2_REWRITE records
1121  */
1122 void
1124 {
1125  char path[MAXPGPATH];
1126  int fd;
1127  xl_heap_rewrite_mapping *xlrec;
1128  uint32 len;
1129  char *data;
1130 
1131  xlrec = (xl_heap_rewrite_mapping *) XLogRecGetData(r);
1132 
1133  snprintf(path, MAXPGPATH,
1134  "pg_logical/mappings/" LOGICAL_REWRITE_FORMAT,
1135  xlrec->mapped_db, xlrec->mapped_rel,
1136  (uint32) (xlrec->start_lsn >> 32),
1137  (uint32) xlrec->start_lsn,
1138  xlrec->mapped_xid, XLogRecGetXid(r));
1139 
1140  fd = OpenTransientFile(path,
1141  O_CREAT | O_WRONLY | PG_BINARY);
1142  if (fd < 0)
1143  ereport(ERROR,
1145  errmsg("could not create file \"%s\": %m", path)));
1146 
1147  /*
1148  * Truncate all data that's not guaranteed to have been safely fsynced (by
1149  * previous record or by the last checkpoint).
1150  */
1152  if (ftruncate(fd, xlrec->offset) != 0)
1153  ereport(ERROR,
1155  errmsg("could not truncate file \"%s\" to %u: %m",
1156  path, (uint32) xlrec->offset)));
1158 
1159  /* now seek to the position we want to write our data to */
1160  if (lseek(fd, xlrec->offset, SEEK_SET) != xlrec->offset)
1161  ereport(ERROR,
1163  errmsg("could not seek to end of file \"%s\": %m",
1164  path)));
1165 
1166  data = XLogRecGetData(r) + sizeof(*xlrec);
1167 
1168  len = xlrec->num_mappings * sizeof(LogicalRewriteMappingData);
1169 
1170  /* write out tail end of mapping file (again) */
1171  errno = 0;
1173  if (write(fd, data, len) != len)
1174  {
1175  /* if write didn't set errno, assume problem is no disk space */
1176  if (errno == 0)
1177  errno = ENOSPC;
1178  ereport(ERROR,
1180  errmsg("could not write to file \"%s\": %m", path)));
1181  }
1183 
1184  /*
1185  * Now fsync all previously written data. We could improve things and only
1186  * do this for the last write to a file, but the required bookkeeping
1187  * doesn't seem worth the trouble.
1188  */
1190  if (pg_fsync(fd) != 0)
1193  errmsg("could not fsync file \"%s\": %m", path)));
1195 
1196  if (CloseTransientFile(fd) != 0)
1197  ereport(ERROR,
1199  errmsg("could not close file \"%s\": %m", path)));
1200 }
1201 
1202 /* ---
1203  * Perform a checkpoint for logical rewrite mappings
1204  *
1205  * This serves two tasks:
1206  * 1) Remove all mappings not needed anymore based on the logical restart LSN
1207  * 2) Flush all remaining mappings to disk, so that replay after a checkpoint
1208  * only has to deal with the parts of a mapping that have been written out
1209  * after the checkpoint started.
1210  * ---
1211  */
1212 void
1214 {
1215  XLogRecPtr cutoff;
1216  XLogRecPtr redo;
1217  DIR *mappings_dir;
1218  struct dirent *mapping_de;
1219  char path[MAXPGPATH + 20];
1220 
1221  /*
1222  * We start of with a minimum of the last redo pointer. No new decoding
1223  * slot will start before that, so that's a safe upper bound for removal.
1224  */
1225  redo = GetRedoRecPtr();
1226 
1227  /* now check for the restart ptrs from existing slots */
1229 
1230  /* don't start earlier than the restart lsn */
1231  if (cutoff != InvalidXLogRecPtr && redo < cutoff)
1232  cutoff = redo;
1233 
1234  mappings_dir = AllocateDir("pg_logical/mappings");
1235  while ((mapping_de = ReadDir(mappings_dir, "pg_logical/mappings")) != NULL)
1236  {
1237  struct stat statbuf;
1238  Oid dboid;
1239  Oid relid;
1240  XLogRecPtr lsn;
1241  TransactionId rewrite_xid;
1242  TransactionId create_xid;
1243  uint32 hi,
1244  lo;
1245 
1246  if (strcmp(mapping_de->d_name, ".") == 0 ||
1247  strcmp(mapping_de->d_name, "..") == 0)
1248  continue;
1249 
1250  snprintf(path, sizeof(path), "pg_logical/mappings/%s", mapping_de->d_name);
1251  if (lstat(path, &statbuf) == 0 && !S_ISREG(statbuf.st_mode))
1252  continue;
1253 
1254  /* Skip over files that cannot be ours. */
1255  if (strncmp(mapping_de->d_name, "map-", 4) != 0)
1256  continue;
1257 
1258  if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
1259  &dboid, &relid, &hi, &lo, &rewrite_xid, &create_xid) != 6)
1260  elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
1261 
1262  lsn = ((uint64) hi) << 32 | lo;
1263 
1264  if (lsn < cutoff || cutoff == InvalidXLogRecPtr)
1265  {
1266  elog(DEBUG1, "removing logical rewrite file \"%s\"", path);
1267  if (unlink(path) < 0)
1268  ereport(ERROR,
1270  errmsg("could not remove file \"%s\": %m", path)));
1271  }
1272  else
1273  {
1274  /* on some operating systems fsyncing a file requires O_RDWR */
1275  int fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
1276 
1277  /*
1278  * The file cannot vanish due to concurrency since this function
1279  * is the only one removing logical mappings and it's run while
1280  * CheckpointLock is held exclusively.
1281  */
1282  if (fd < 0)
1283  ereport(ERROR,
1285  errmsg("could not open file \"%s\": %m", path)));
1286 
1287  /*
1288  * We could try to avoid fsyncing files that either haven't
1289  * changed or have only been created since the checkpoint's start,
1290  * but it's currently not deemed worth the effort.
1291  */
1293  if (pg_fsync(fd) != 0)
1296  errmsg("could not fsync file \"%s\": %m", path)));
1298 
1299  if (CloseTransientFile(fd) != 0)
1300  ereport(ERROR,
1302  errmsg("could not close file \"%s\": %m", path)));
1303  }
1304  }
1305  FreeDir(mappings_dir);
1306 }
#define HeapTupleHeaderGetUpdateXid(tup)
Definition: htup_details.h:365
#define ItemPointerIsValid(pointer)
Definition: itemptr.h:82
HeapTuple heap_copytuple(HeapTuple tuple)
Definition: heaptuple.c:680
union HeapTupleHeaderData::@45 t_choice
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
void MemoryContextDelete(MemoryContext context)
Definition: mcxt.c:211
#define AllocSetContextCreate
Definition: memutils.h:170
void end_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:305
#define DEBUG1
Definition: elog.h:25
dlist_node * cur
Definition: ilist.h:180
TidHashKey key
Definition: rewriteheap.c:175
void heap_xlog_logical_rewrite(XLogReaderState *r)
Definition: rewriteheap.c:1123
HeapTupleFields t_heap
Definition: htup_details.h:156
TransactionId rs_logical_xmin
Definition: rewriteheap.c:145
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1358
#define TransactionIdEquals(id1, id2)
Definition: transam.h:43
#define HASH_CONTEXT
Definition: hsearch.h:93
#define HASH_ELEM
Definition: hsearch.h:87
#define dlist_foreach_modify(iter, lhead)
Definition: ilist.h:524
uint32 TransactionId
Definition: c.h:514
MemoryContext hcxt
Definition: hsearch.h:78
RewriteState begin_heap_rewrite(Relation old_heap, Relation new_heap, TransactionId oldest_xmin, TransactionId freeze_xid, MultiXactId cutoff_multi, bool use_wal)
Definition: rewriteheap.c:239
static void logical_rewrite_heap_tuple(RewriteState state, ItemPointerData old_tid, HeapTuple new_tuple)
Definition: rewriteheap.c:1049
static void logical_end_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:952
#define write(a, b, c)
Definition: win32.h:14
HeapTupleHeaderData * HeapTupleHeader
Definition: htup.h:23
struct SMgrRelationData * rd_smgr
Definition: rel.h:56
TransactionId rs_freeze_xid
Definition: rewriteheap.c:143
#define XLOG_HEAP2_REWRITE
Definition: heapam_xlog.h:53
static void dlist_push_tail(dlist_head *head, dlist_node *node)
Definition: ilist.h:317
bool HeapTupleHeaderIsOnlyLocked(HeapTupleHeader tuple)
MultiXactId rs_cutoff_multi
Definition: rewriteheap.c:147
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
Pointer Item
Definition: item.h:17
Size entrysize
Definition: hsearch.h:73
Relation rs_new_rel
Definition: rewriteheap.c:135
int errcode(int sqlerrcode)
Definition: elog.c:608
#define HEAP2_XACT_MASK
Definition: htup_details.h:283
#define PageAddItem(page, item, size, offsetNumber, overwrite, is_heap)
Definition: bufpage.h:416
UnresolvedTupData * UnresolvedTup
Definition: rewriteheap.c:180
void heap_sync(Relation rel)
Definition: heapam.c:8938
#define HEAP_INSERT_SKIP_WAL
Definition: heapam.h:32
#define HeapTupleHeaderIndicatesMovedPartitions(tup)
Definition: htup_details.h:445
void CheckPointLogicalRewriteHeap(void)
Definition: rewriteheap.c:1213
uint32 BlockNumber
Definition: block.h:31
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:906
#define HEAP_UPDATED
Definition: htup_details.h:209
ItemPointerData old_tid
Definition: rewriteheap.c:176
Form_pg_class rd_rel
Definition: rel.h:83
void heap_freetuple(HeapTuple htup)
Definition: heaptuple.c:1338
unsigned int Oid
Definition: postgres_ext.h:31
Definition: dirent.h:9
void ProcArrayGetReplicationSlotXmin(TransactionId *xmin, TransactionId *catalog_xmin)
Definition: procarray.c:3117
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1222
uint16 OffsetNumber
Definition: off.h:24
HeapTupleHeader t_data
Definition: htup.h:68
#define RelationOpenSmgr(relation)
Definition: rel.h:479
Definition: dynahash.c:208
#define dlist_container(type, membername, ptr)
Definition: ilist.h:477
void pfree(void *pointer)
Definition: mcxt.c:1056
uint32 rs_num_rewrite_mappings
Definition: rewriteheap.c:155
#define XLogRecGetData(decoder)
Definition: xlogreader.h:283
Definition: dirent.c:25
#define ERROR
Definition: elog.h:43
ItemPointerData new_tid
Definition: rewriteheap.c:185
Size PageGetHeapFreeSpace(Page page)
Definition: bufpage.c:665
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2292
#define HEAP_XMAX_INVALID
Definition: htup_details.h:207
char path[MAXPGPATH]
Definition: rewriteheap.c:201
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:11191
ItemPointerData t_ctid
Definition: htup_details.h:160
HTAB * rs_unresolved_tups
Definition: rewriteheap.c:152
#define MAXPGPATH
ItemPointerData t_self
Definition: htup.h:65
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:192
int FileSync(File file, uint32 wait_event_info)
Definition: fd.c:2049
TransactionId xid
Definition: rewriteheap.c:196
TransactionId GetCurrentTransactionId(void)
Definition: xact.c:422
uint32 t_len
Definition: htup.h:64
static void logical_rewrite_log_mapping(RewriteState state, TransactionId xid, LogicalRewriteMappingData *map)
Definition: rewriteheap.c:982
#define MaxHeapTupleSize
Definition: htup_details.h:560
int errcode_for_file_access(void)
Definition: elog.c:631
XLogRecPtr ReplicationSlotsComputeLogicalRestartLSN(void)
Definition: slot.c:791
#define InvalidTransactionId
Definition: transam.h:31
unsigned int uint32
Definition: c.h:359
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2503
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1342
MemoryContext CurrentMemoryContext
Definition: mcxt.c:38
bool heap_freeze_tuple(HeapTupleHeader tuple, TransactionId relfrozenxid, TransactionId relminmxid, TransactionId cutoff_xid, TransactionId cutoff_multi)
Definition: heapam.c:6368
static void dlist_delete(dlist_node *node)
Definition: ilist.h:358
#define ereport(elevel, rest)
Definition: elog.h:141
XLogRecPtr rs_begin_lsn
Definition: rewriteheap.c:151
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
#define RelationGetTargetPageFreeSpace(relation, defaultff)
Definition: rel.h:306
#define S_ISREG(m)
Definition: win32_port.h:299
HTAB * rs_logical_mappings
Definition: rewriteheap.c:154
int CloseTransientFile(int fd)
Definition: fd.c:2469
TransactionId rs_oldest_xmin
Definition: rewriteheap.c:141
#define stat(a, b)
Definition: win32_port.h:255
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:235
#define TOAST_TUPLE_THRESHOLD
Definition: heaptoast.h:48
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
#define HASH_BLOBS
Definition: hsearch.h:88
#define XLogRecGetXid(decoder)
Definition: xlogreader.h:281
int FileWrite(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:1951
MemoryContext rs_cxt
Definition: rewriteheap.c:149
void * palloc0(Size size)
Definition: mcxt.c:980
#define RelationIsAccessibleInLogicalDecoding(relation)
Definition: rel.h:578
int data_sync_elevel(int elevel)
Definition: fd.c:3519
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:316
#define HEAP_XMAX_IS_LOCKED_ONLY(infomask)
Definition: htup_details.h:230
Oid MyDatabaseId
Definition: globals.c:85
Size keysize
Definition: hsearch.h:72
#define RelationGetNumberOfBlocks(reln)
Definition: bufmgr.h:198
ItemPointerData tid
Definition: rewriteheap.c:167
#define InvalidOffsetNumber
Definition: off.h:26
#define InvalidOid
Definition: postgres_ext.h:36
static void dlist_init(dlist_head *head)
Definition: ilist.h:278
struct RewriteStateData RewriteStateData
TransactionId mapped_xid
Definition: heapam_xlog.h:378
TransactionId MultiXactId
Definition: c.h:524
RelFileNode rd_node
Definition: rel.h:54
void FileClose(File file)
Definition: fd.c:1748
static void logical_begin_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:805
ItemPointerData new_tid
Definition: rewriteheap.h:40
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:739
HTAB * rs_old_new_tid_map
Definition: rewriteheap.c:153
Definition: regguts.h:298
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2569
#define HeapTupleHeaderGetXmin(tup)
Definition: htup_details.h:313
size_t Size
Definition: c.h:467
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1318
dlist_head mappings
Definition: rewriteheap.c:200
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1198
#define MAXALIGN(LEN)
Definition: c.h:692
XLogRecPtr GetRedoRecPtr(void)
Definition: xlog.c:8208
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1389
#define RelationNeedsWAL(relation)
Definition: rel.h:524
bool ItemPointerEquals(ItemPointer pointer1, ItemPointer pointer2)
Definition: itemptr.c:29
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1379
static void logical_heap_rewrite_flush_mappings(RewriteState state)
Definition: rewriteheap.c:854
struct RewriteMappingFile RewriteMappingFile
#define LOGICAL_REWRITE_FORMAT
Definition: rewriteheap.h:54
void smgrextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:483
#define lstat(path, sb)
Definition: win32_port.h:244
#define HEAP_INSERT_SKIP_FSM
Definition: heapam.h:33
bool rewrite_heap_dead_tuple(RewriteState state, HeapTuple old_tuple)
Definition: rewriteheap.c:573
#define ItemPointerSetInvalid(pointer)
Definition: itemptr.h:172
#define HeapTupleHasExternal(tuple)
Definition: htup_details.h:673
void * palloc(Size size)
Definition: mcxt.c:949
int errmsg(const char *fmt,...)
Definition: elog.c:822
#define HEAP_INSERT_NO_LOGICAL
Definition: heapam.h:35
TransactionId xmin
Definition: rewriteheap.c:166
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:796
LogicalRewriteMappingData map
Definition: rewriteheap.c:210
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:972
#define elog(elevel,...)
Definition: elog.h:228
static void raw_heap_insert(RewriteState state, HeapTuple tup)
Definition: rewriteheap.c:623
struct LogicalRewriteMappingData LogicalRewriteMappingData
#define HEAP_XACT_MASK
Definition: htup_details.h:218
int pg_fsync(int fd)
Definition: fd.c:330
Relation rs_old_rel
Definition: rewriteheap.c:134
char d_name[MAX_PATH]
Definition: dirent.h:14
#define HEAP_DEFAULT_FILLFACTOR
Definition: rel.h:277
struct RewriteMappingDataEntry RewriteMappingDataEntry
HeapTuple heap_toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup, int options)
Definition: heaptoast.c:94
#define TransactionIdIsNormal(xid)
Definition: transam.h:42
BlockNumber rs_blockno
Definition: rewriteheap.c:137
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define snprintf
Definition: port.h:192
void rewrite_heap_tuple(RewriteState state, HeapTuple old_tuple, HeapTuple new_tuple)
Definition: rewriteheap.c:371
#define RelationGetRelid(relation)
Definition: rel.h:422
OldToNewMappingData * OldToNewMapping
Definition: rewriteheap.c:188
int FreeDir(DIR *dir)
Definition: fd.c:2621
#define PageGetItem(page, itemId)
Definition: bufpage.h:340
Pointer Page
Definition: bufpage.h:78
#define ItemPointerSet(pointer, blockNumber, offNum)
Definition: itemptr.h:127
void PageInit(Page page, Size pageSize, Size specialSize)
Definition: bufpage.c:42
ItemPointerData old_tid
Definition: rewriteheap.h:39
#define ftruncate(a, b)
Definition: win32_port.h:60