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) 1996-2018, 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 "miscadmin.h"
109 
110 #include "access/heapam.h"
111 #include "access/heapam_xlog.h"
112 #include "access/rewriteheap.h"
113 #include "access/transam.h"
114 #include "access/tuptoaster.h"
115 #include "access/xact.h"
116 #include "access/xloginsert.h"
117 
118 #include "catalog/catalog.h"
119 
120 #include "lib/ilist.h"
121 
122 #include "pgstat.h"
123 
124 #include "replication/logical.h"
125 #include "replication/slot.h"
126 
127 #include "storage/bufmgr.h"
128 #include "storage/fd.h"
129 #include "storage/smgr.h"
130 
131 #include "utils/memutils.h"
132 #include "utils/rel.h"
133 #include "utils/tqual.h"
134 
135 #include "storage/procarray.h"
136 
137 /*
138  * State associated with a rewrite operation. This is opaque to the user
139  * of the rewrite facility.
140  */
141 typedef struct RewriteStateData
142 {
143  Relation rs_old_rel; /* source heap */
144  Relation rs_new_rel; /* destination heap */
145  Page rs_buffer; /* page currently being built */
146  BlockNumber rs_blockno; /* block where page will go */
147  bool rs_buffer_valid; /* T if any tuples in buffer */
148  bool rs_use_wal; /* must we WAL-log inserts? */
149  bool rs_logical_rewrite; /* do we need to do logical rewriting */
150  TransactionId rs_oldest_xmin; /* oldest xmin used by caller to determine
151  * tuple visibility */
152  TransactionId rs_freeze_xid; /* Xid that will be used as freeze cutoff
153  * point */
154  TransactionId rs_logical_xmin; /* Xid that will be used as cutoff point
155  * for logical rewrites */
156  MultiXactId rs_cutoff_multi; /* MultiXactId that will be used as cutoff
157  * point for multixacts */
158  MemoryContext rs_cxt; /* for hash tables and entries and tuples in
159  * them */
160  XLogRecPtr rs_begin_lsn; /* XLogInsertLsn when starting the rewrite */
161  HTAB *rs_unresolved_tups; /* unmatched A tuples */
162  HTAB *rs_old_new_tid_map; /* unmatched B tuples */
163  HTAB *rs_logical_mappings; /* logical remapping files */
164  uint32 rs_num_rewrite_mappings; /* # in memory mappings */
166 
167 /*
168  * The lookup keys for the hash tables are tuple TID and xmin (we must check
169  * both to avoid false matches from dead tuples). Beware that there is
170  * probably some padding space in this struct; it must be zeroed out for
171  * correct hashtable operation.
172  */
173 typedef struct
174 {
175  TransactionId xmin; /* tuple xmin */
176  ItemPointerData tid; /* tuple location in old heap */
177 } TidHashKey;
178 
179 /*
180  * Entry structures for the hash tables
181  */
182 typedef struct
183 {
184  TidHashKey key; /* expected xmin/old location of B tuple */
185  ItemPointerData old_tid; /* A's location in the old heap */
186  HeapTuple tuple; /* A's tuple contents */
188 
190 
191 typedef struct
192 {
193  TidHashKey key; /* actual xmin/old location of B tuple */
194  ItemPointerData new_tid; /* where we put it in the new heap */
196 
198 
199 /*
200  * In-Memory data for an xid that might need logical remapping entries
201  * to be logged.
202  */
203 typedef struct RewriteMappingFile
204 {
205  TransactionId xid; /* xid that might need to see the row */
206  int vfd; /* fd of mappings file */
207  off_t off; /* how far have we written yet */
208  uint32 num_mappings; /* number of in-memory mappings */
209  dlist_head mappings; /* list of in-memory mappings */
210  char path[MAXPGPATH]; /* path, for error messages */
212 
213 /*
214  * A single In-Memory logical rewrite mapping, hanging off
215  * RewriteMappingFile->mappings.
216  */
218 {
219  LogicalRewriteMappingData map; /* map between old and new location of the
220  * tuple */
223 
224 
225 /* prototypes for internal functions */
226 static void raw_heap_insert(RewriteState state, HeapTuple tup);
227 
228 /* internal logical remapping prototypes */
232 
233 
234 /*
235  * Begin a rewrite of a table
236  *
237  * old_heap old, locked heap relation tuples will be read from
238  * new_heap new, locked heap relation to insert tuples to
239  * oldest_xmin xid used by the caller to determine which tuples are dead
240  * freeze_xid xid before which tuples will be frozen
241  * min_multi multixact before which multis will be removed
242  * use_wal should the inserts to the new heap be WAL-logged?
243  *
244  * Returns an opaque RewriteState, allocated in current memory context,
245  * to be used in subsequent calls to the other functions.
246  */
248 begin_heap_rewrite(Relation old_heap, Relation new_heap, TransactionId oldest_xmin,
249  TransactionId freeze_xid, MultiXactId cutoff_multi,
250  bool use_wal)
251 {
253  MemoryContext rw_cxt;
254  MemoryContext old_cxt;
255  HASHCTL hash_ctl;
256 
257  /*
258  * To ease cleanup, make a separate context that will contain the
259  * RewriteState struct itself plus all subsidiary data.
260  */
262  "Table rewrite",
264  old_cxt = MemoryContextSwitchTo(rw_cxt);
265 
266  /* Create and fill in the state struct */
267  state = palloc0(sizeof(RewriteStateData));
268 
269  state->rs_old_rel = old_heap;
270  state->rs_new_rel = new_heap;
271  state->rs_buffer = (Page) palloc(BLCKSZ);
272  /* new_heap needn't be empty, just locked */
273  state->rs_blockno = RelationGetNumberOfBlocks(new_heap);
274  state->rs_buffer_valid = false;
275  state->rs_use_wal = use_wal;
276  state->rs_oldest_xmin = oldest_xmin;
277  state->rs_freeze_xid = freeze_xid;
278  state->rs_cutoff_multi = cutoff_multi;
279  state->rs_cxt = rw_cxt;
280 
281  /* Initialize hash tables used to track update chains */
282  memset(&hash_ctl, 0, sizeof(hash_ctl));
283  hash_ctl.keysize = sizeof(TidHashKey);
284  hash_ctl.entrysize = sizeof(UnresolvedTupData);
285  hash_ctl.hcxt = state->rs_cxt;
286 
287  state->rs_unresolved_tups =
288  hash_create("Rewrite / Unresolved ctids",
289  128, /* arbitrary initial size */
290  &hash_ctl,
292 
293  hash_ctl.entrysize = sizeof(OldToNewMappingData);
294 
295  state->rs_old_new_tid_map =
296  hash_create("Rewrite / Old to new tid map",
297  128, /* arbitrary initial size */
298  &hash_ctl,
300 
301  MemoryContextSwitchTo(old_cxt);
302 
304 
305  return state;
306 }
307 
308 /*
309  * End a rewrite.
310  *
311  * state and any other resources are freed.
312  */
313 void
315 {
316  HASH_SEQ_STATUS seq_status;
317  UnresolvedTup unresolved;
318 
319  /*
320  * Write any remaining tuples in the UnresolvedTups table. If we have any
321  * left, they should in fact be dead, but let's err on the safe side.
322  */
323  hash_seq_init(&seq_status, state->rs_unresolved_tups);
324 
325  while ((unresolved = hash_seq_search(&seq_status)) != NULL)
326  {
327  ItemPointerSetInvalid(&unresolved->tuple->t_data->t_ctid);
328  raw_heap_insert(state, unresolved->tuple);
329  }
330 
331  /* Write the last page, if any */
332  if (state->rs_buffer_valid)
333  {
334  if (state->rs_use_wal)
335  log_newpage(&state->rs_new_rel->rd_node,
336  MAIN_FORKNUM,
337  state->rs_blockno,
338  state->rs_buffer,
339  true);
341 
343 
345  (char *) state->rs_buffer, true);
346  }
347 
348  /*
349  * If the rel is WAL-logged, must fsync before commit. We use heap_sync
350  * to ensure that the toast table gets fsync'd too.
351  *
352  * It's obvious that we must do this when not WAL-logging. It's less
353  * obvious that we have to do it even if we did WAL-log the pages. The
354  * reason is the same as in tablecmds.c's copy_relation_data(): we're
355  * writing data that's not in shared buffers, and so a CHECKPOINT
356  * occurring during the rewriteheap operation won't have fsync'd data we
357  * wrote before the checkpoint.
358  */
359  if (RelationNeedsWAL(state->rs_new_rel))
360  heap_sync(state->rs_new_rel);
361 
363 
364  /* Deleting the context frees everything */
365  MemoryContextDelete(state->rs_cxt);
366 }
367 
368 /*
369  * Add a tuple to the new heap.
370  *
371  * Visibility information is copied from the original tuple, except that
372  * we "freeze" very-old tuples. Note that since we scribble on new_tuple,
373  * it had better be temp storage not a pointer to the original tuple.
374  *
375  * state opaque state as returned by begin_heap_rewrite
376  * old_tuple original tuple in the old heap
377  * new_tuple new, rewritten tuple to be inserted to new heap
378  */
379 void
381  HeapTuple old_tuple, HeapTuple new_tuple)
382 {
383  MemoryContext old_cxt;
384  ItemPointerData old_tid;
385  TidHashKey hashkey;
386  bool found;
387  bool free_new;
388 
389  old_cxt = MemoryContextSwitchTo(state->rs_cxt);
390 
391  /*
392  * Copy the original tuple's visibility information into new_tuple.
393  *
394  * XXX we might later need to copy some t_infomask2 bits, too? Right now,
395  * we intentionally clear the HOT status bits.
396  */
397  memcpy(&new_tuple->t_data->t_choice.t_heap,
398  &old_tuple->t_data->t_choice.t_heap,
399  sizeof(HeapTupleFields));
400 
401  new_tuple->t_data->t_infomask &= ~HEAP_XACT_MASK;
402  new_tuple->t_data->t_infomask2 &= ~HEAP2_XACT_MASK;
403  new_tuple->t_data->t_infomask |=
404  old_tuple->t_data->t_infomask & HEAP_XACT_MASK;
405 
406  /*
407  * While we have our hands on the tuple, we may as well freeze any
408  * eligible xmin or xmax, so that future VACUUM effort can be saved.
409  */
410  heap_freeze_tuple(new_tuple->t_data,
411  state->rs_old_rel->rd_rel->relfrozenxid,
412  state->rs_old_rel->rd_rel->relminmxid,
413  state->rs_freeze_xid,
414  state->rs_cutoff_multi);
415 
416  /*
417  * Invalid ctid means that ctid should point to the tuple itself. We'll
418  * override it later if the tuple is part of an update chain.
419  */
420  ItemPointerSetInvalid(&new_tuple->t_data->t_ctid);
421 
422  /*
423  * If the tuple has been updated, check the old-to-new mapping hash table.
424  */
425  if (!((old_tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
426  HeapTupleHeaderIsOnlyLocked(old_tuple->t_data)) &&
428  !(ItemPointerEquals(&(old_tuple->t_self),
429  &(old_tuple->t_data->t_ctid))))
430  {
431  OldToNewMapping mapping;
432 
433  memset(&hashkey, 0, sizeof(hashkey));
434  hashkey.xmin = HeapTupleHeaderGetUpdateXid(old_tuple->t_data);
435  hashkey.tid = old_tuple->t_data->t_ctid;
436 
437  mapping = (OldToNewMapping)
438  hash_search(state->rs_old_new_tid_map, &hashkey,
439  HASH_FIND, NULL);
440 
441  if (mapping != NULL)
442  {
443  /*
444  * We've already copied the tuple that t_ctid points to, so we can
445  * set the ctid of this tuple to point to the new location, and
446  * insert it right away.
447  */
448  new_tuple->t_data->t_ctid = mapping->new_tid;
449 
450  /* We don't need the mapping entry anymore */
451  hash_search(state->rs_old_new_tid_map, &hashkey,
452  HASH_REMOVE, &found);
453  Assert(found);
454  }
455  else
456  {
457  /*
458  * We haven't seen the tuple t_ctid points to yet. Stash this
459  * tuple into unresolved_tups to be written later.
460  */
461  UnresolvedTup unresolved;
462 
463  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
464  HASH_ENTER, &found);
465  Assert(!found);
466 
467  unresolved->old_tid = old_tuple->t_self;
468  unresolved->tuple = heap_copytuple(new_tuple);
469 
470  /*
471  * We can't do anything more now, since we don't know where the
472  * tuple will be written.
473  */
474  MemoryContextSwitchTo(old_cxt);
475  return;
476  }
477  }
478 
479  /*
480  * Now we will write the tuple, and then check to see if it is the B tuple
481  * in any new or known pair. When we resolve a known pair, we will be
482  * able to write that pair's A tuple, and then we have to check if it
483  * resolves some other pair. Hence, we need a loop here.
484  */
485  old_tid = old_tuple->t_self;
486  free_new = false;
487 
488  for (;;)
489  {
490  ItemPointerData new_tid;
491 
492  /* Insert the tuple and find out where it's put in new_heap */
493  raw_heap_insert(state, new_tuple);
494  new_tid = new_tuple->t_self;
495 
496  logical_rewrite_heap_tuple(state, old_tid, new_tuple);
497 
498  /*
499  * If the tuple is the updated version of a row, and the prior version
500  * wouldn't be DEAD yet, then we need to either resolve the prior
501  * version (if it's waiting in rs_unresolved_tups), or make an entry
502  * in rs_old_new_tid_map (so we can resolve it when we do see it). The
503  * previous tuple's xmax would equal this one's xmin, so it's
504  * RECENTLY_DEAD if and only if the xmin is not before OldestXmin.
505  */
506  if ((new_tuple->t_data->t_infomask & HEAP_UPDATED) &&
508  state->rs_oldest_xmin))
509  {
510  /*
511  * Okay, this is B in an update pair. See if we've seen A.
512  */
513  UnresolvedTup unresolved;
514 
515  memset(&hashkey, 0, sizeof(hashkey));
516  hashkey.xmin = HeapTupleHeaderGetXmin(new_tuple->t_data);
517  hashkey.tid = old_tid;
518 
519  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
520  HASH_FIND, NULL);
521 
522  if (unresolved != NULL)
523  {
524  /*
525  * We have seen and memorized the previous tuple already. Now
526  * that we know where we inserted the tuple its t_ctid points
527  * to, fix its t_ctid and insert it to the new heap.
528  */
529  if (free_new)
530  heap_freetuple(new_tuple);
531  new_tuple = unresolved->tuple;
532  free_new = true;
533  old_tid = unresolved->old_tid;
534  new_tuple->t_data->t_ctid = new_tid;
535 
536  /*
537  * We don't need the hash entry anymore, but don't free its
538  * tuple just yet.
539  */
540  hash_search(state->rs_unresolved_tups, &hashkey,
541  HASH_REMOVE, &found);
542  Assert(found);
543 
544  /* loop back to insert the previous tuple in the chain */
545  continue;
546  }
547  else
548  {
549  /*
550  * Remember the new tid of this tuple. We'll use it to set the
551  * ctid when we find the previous tuple in the chain.
552  */
553  OldToNewMapping mapping;
554 
555  mapping = hash_search(state->rs_old_new_tid_map, &hashkey,
556  HASH_ENTER, &found);
557  Assert(!found);
558 
559  mapping->new_tid = new_tid;
560  }
561  }
562 
563  /* Done with this (chain of) tuples, for now */
564  if (free_new)
565  heap_freetuple(new_tuple);
566  break;
567  }
568 
569  MemoryContextSwitchTo(old_cxt);
570 }
571 
572 /*
573  * Register a dead tuple with an ongoing rewrite. Dead tuples are not
574  * copied to the new table, but we still make note of them so that we
575  * can release some resources earlier.
576  *
577  * Returns true if a tuple was removed from the unresolved_tups table.
578  * This indicates that that tuple, previously thought to be "recently dead",
579  * is now known really dead and won't be written to the output.
580  */
581 bool
583 {
584  /*
585  * If we have already seen an earlier tuple in the update chain that
586  * points to this tuple, let's forget about that earlier tuple. It's in
587  * fact dead as well, our simple xmax < OldestXmin test in
588  * HeapTupleSatisfiesVacuum just wasn't enough to detect it. It happens
589  * when xmin of a tuple is greater than xmax, which sounds
590  * counter-intuitive but is perfectly valid.
591  *
592  * We don't bother to try to detect the situation the other way round,
593  * when we encounter the dead tuple first and then the recently dead one
594  * that points to it. If that happens, we'll have some unmatched entries
595  * in the UnresolvedTups hash table at the end. That can happen anyway,
596  * because a vacuum might have removed the dead tuple in the chain before
597  * us.
598  */
599  UnresolvedTup unresolved;
600  TidHashKey hashkey;
601  bool found;
602 
603  memset(&hashkey, 0, sizeof(hashkey));
604  hashkey.xmin = HeapTupleHeaderGetXmin(old_tuple->t_data);
605  hashkey.tid = old_tuple->t_self;
606 
607  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
608  HASH_FIND, NULL);
609 
610  if (unresolved != NULL)
611  {
612  /* Need to free the contained tuple as well as the hashtable entry */
613  heap_freetuple(unresolved->tuple);
614  hash_search(state->rs_unresolved_tups, &hashkey,
615  HASH_REMOVE, &found);
616  Assert(found);
617  return true;
618  }
619 
620  return false;
621 }
622 
623 /*
624  * Insert a tuple to the new relation. This has to track heap_insert
625  * and its subsidiary functions!
626  *
627  * t_self of the tuple is set to the new TID of the tuple. If t_ctid of the
628  * tuple is invalid on entry, it's replaced with the new TID as well (in
629  * the inserted data only, not in the caller's copy).
630  */
631 static void
633 {
634  Page page = state->rs_buffer;
635  Size pageFreeSpace,
636  saveFreeSpace;
637  Size len;
638  OffsetNumber newoff;
639  HeapTuple heaptup;
640 
641  /*
642  * If the new tuple is too big for storage or contains already toasted
643  * out-of-line attributes from some other relation, invoke the toaster.
644  *
645  * Note: below this point, heaptup is the data we actually intend to store
646  * into the relation; tup is the caller's original untoasted data.
647  */
648  if (state->rs_new_rel->rd_rel->relkind == RELKIND_TOASTVALUE)
649  {
650  /* toast table entries should never be recursively toasted */
652  heaptup = tup;
653  }
654  else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
655  heaptup = toast_insert_or_update(state->rs_new_rel, tup, NULL,
657  (state->rs_use_wal ?
658  0 : HEAP_INSERT_SKIP_WAL));
659  else
660  heaptup = tup;
661 
662  len = MAXALIGN(heaptup->t_len); /* be conservative */
663 
664  /*
665  * If we're gonna fail for oversize tuple, do it right away
666  */
667  if (len > MaxHeapTupleSize)
668  ereport(ERROR,
669  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
670  errmsg("row is too big: size %zu, maximum size %zu",
671  len, MaxHeapTupleSize)));
672 
673  /* Compute desired extra freespace due to fillfactor option */
674  saveFreeSpace = RelationGetTargetPageFreeSpace(state->rs_new_rel,
676 
677  /* Now we can check to see if there's enough free space already. */
678  if (state->rs_buffer_valid)
679  {
680  pageFreeSpace = PageGetHeapFreeSpace(page);
681 
682  if (len + saveFreeSpace > pageFreeSpace)
683  {
684  /* Doesn't fit, so write out the existing page */
685 
686  /* XLOG stuff */
687  if (state->rs_use_wal)
688  log_newpage(&state->rs_new_rel->rd_node,
689  MAIN_FORKNUM,
690  state->rs_blockno,
691  page,
692  true);
693 
694  /*
695  * Now write the page. We say isTemp = true even if it's not a
696  * temp table, because there's no need for smgr to schedule an
697  * fsync for this write; we'll do it ourselves in
698  * end_heap_rewrite.
699  */
701 
702  PageSetChecksumInplace(page, state->rs_blockno);
703 
705  state->rs_blockno, (char *) page, true);
706 
707  state->rs_blockno++;
708  state->rs_buffer_valid = false;
709  }
710  }
711 
712  if (!state->rs_buffer_valid)
713  {
714  /* Initialize a new empty page */
715  PageInit(page, BLCKSZ, 0);
716  state->rs_buffer_valid = true;
717  }
718 
719  /* And now we can insert the tuple into the page */
720  newoff = PageAddItem(page, (Item) heaptup->t_data, heaptup->t_len,
721  InvalidOffsetNumber, false, true);
722  if (newoff == InvalidOffsetNumber)
723  elog(ERROR, "failed to add tuple");
724 
725  /* Update caller's t_self to the actual position where it was stored */
726  ItemPointerSet(&(tup->t_self), state->rs_blockno, newoff);
727 
728  /*
729  * Insert the correct position into CTID of the stored tuple, too, if the
730  * caller didn't supply a valid CTID.
731  */
732  if (!ItemPointerIsValid(&tup->t_data->t_ctid))
733  {
734  ItemId newitemid;
735  HeapTupleHeader onpage_tup;
736 
737  newitemid = PageGetItemId(page, newoff);
738  onpage_tup = (HeapTupleHeader) PageGetItem(page, newitemid);
739 
740  onpage_tup->t_ctid = tup->t_self;
741  }
742 
743  /* If heaptup is a private copy, release it. */
744  if (heaptup != tup)
745  heap_freetuple(heaptup);
746 }
747 
748 /* ------------------------------------------------------------------------
749  * Logical rewrite support
750  *
751  * When doing logical decoding - which relies on using cmin/cmax of catalog
752  * tuples, via xl_heap_new_cid records - heap rewrites have to log enough
753  * information to allow the decoding backend to updates its internal mapping
754  * of (relfilenode,ctid) => (cmin, cmax) to be correct for the rewritten heap.
755  *
756  * For that, every time we find a tuple that's been modified in a catalog
757  * relation within the xmin horizon of any decoding slot, we log a mapping
758  * from the old to the new location.
759  *
760  * To deal with rewrites that abort the filename of a mapping file contains
761  * the xid of the transaction performing the rewrite, which then can be
762  * checked before being read in.
763  *
764  * For efficiency we don't immediately spill every single map mapping for a
765  * row to disk but only do so in batches when we've collected several of them
766  * in memory or when end_heap_rewrite() has been called.
767  *
768  * Crash-Safety: This module diverts from the usual patterns of doing WAL
769  * since it cannot rely on checkpoint flushing out all buffers and thus
770  * waiting for exclusive locks on buffers. Usually the XLogInsert() covering
771  * buffer modifications is performed while the buffer(s) that are being
772  * modified are exclusively locked guaranteeing that both the WAL record and
773  * the modified heap are on either side of the checkpoint. But since the
774  * mapping files we log aren't in shared_buffers that interlock doesn't work.
775  *
776  * Instead we simply write the mapping files out to disk, *before* the
777  * XLogInsert() is performed. That guarantees that either the XLogInsert() is
778  * inserted after the checkpoint's redo pointer or that the checkpoint (via
779  * LogicalRewriteHeapCheckpoint()) has flushed the (partial) mapping file to
780  * disk. That leaves the tail end that has not yet been flushed open to
781  * corruption, which is solved by including the current offset in the
782  * xl_heap_rewrite_mapping records and truncating the mapping file to it
783  * during replay. Every time a rewrite is finished all generated mapping files
784  * are synced to disk.
785  *
786  * Note that if we were only concerned about crash safety we wouldn't have to
787  * deal with WAL logging at all - an fsync() at the end of a rewrite would be
788  * sufficient for crash safety. Any mapping that hasn't been safely flushed to
789  * disk has to be by an aborted (explicitly or via a crash) transaction and is
790  * ignored by virtue of the xid in its name being subject to a
791  * TransactionDidCommit() check. But we want to support having standbys via
792  * physical replication, both for availability and to do logical decoding
793  * there.
794  * ------------------------------------------------------------------------
795  */
796 
797 /*
798  * Do preparations for logging logical mappings during a rewrite if
799  * necessary. If we detect that we don't need to log anything we'll prevent
800  * any further action by the various logical rewrite functions.
801  */
802 static void
804 {
805  HASHCTL hash_ctl;
806  TransactionId logical_xmin;
807 
808  /*
809  * We only need to persist these mappings if the rewritten table can be
810  * accessed during logical decoding, if not, we can skip doing any
811  * additional work.
812  */
813  state->rs_logical_rewrite =
815 
816  if (!state->rs_logical_rewrite)
817  return;
818 
819  ProcArrayGetReplicationSlotXmin(NULL, &logical_xmin);
820 
821  /*
822  * If there are no logical slots in progress we don't need to do anything,
823  * there cannot be any remappings for relevant rows yet. The relation's
824  * lock protects us against races.
825  */
826  if (logical_xmin == InvalidTransactionId)
827  {
828  state->rs_logical_rewrite = false;
829  return;
830  }
831 
832  state->rs_logical_xmin = logical_xmin;
834  state->rs_num_rewrite_mappings = 0;
835 
836  memset(&hash_ctl, 0, sizeof(hash_ctl));
837  hash_ctl.keysize = sizeof(TransactionId);
838  hash_ctl.entrysize = sizeof(RewriteMappingFile);
839  hash_ctl.hcxt = state->rs_cxt;
840 
841  state->rs_logical_mappings =
842  hash_create("Logical rewrite mapping",
843  128, /* arbitrary initial size */
844  &hash_ctl,
846 }
847 
848 /*
849  * Flush all logical in-memory mappings to disk, but don't fsync them yet.
850  */
851 static void
853 {
854  HASH_SEQ_STATUS seq_status;
855  RewriteMappingFile *src;
856  dlist_mutable_iter iter;
857 
858  Assert(state->rs_logical_rewrite);
859 
860  /* no logical rewrite in progress, no need to iterate over mappings */
861  if (state->rs_num_rewrite_mappings == 0)
862  return;
863 
864  elog(DEBUG1, "flushing %u logical rewrite mapping entries",
865  state->rs_num_rewrite_mappings);
866 
867  hash_seq_init(&seq_status, state->rs_logical_mappings);
868  while ((src = (RewriteMappingFile *) hash_seq_search(&seq_status)) != NULL)
869  {
870  char *waldata;
871  char *waldata_start;
873  Oid dboid;
874  uint32 len;
875  int written;
876 
877  /* this file hasn't got any new mappings */
878  if (src->num_mappings == 0)
879  continue;
880 
881  if (state->rs_old_rel->rd_rel->relisshared)
882  dboid = InvalidOid;
883  else
884  dboid = MyDatabaseId;
885 
886  xlrec.num_mappings = src->num_mappings;
887  xlrec.mapped_rel = RelationGetRelid(state->rs_old_rel);
888  xlrec.mapped_xid = src->xid;
889  xlrec.mapped_db = dboid;
890  xlrec.offset = src->off;
891  xlrec.start_lsn = state->rs_begin_lsn;
892 
893  /* write all mappings consecutively */
894  len = src->num_mappings * sizeof(LogicalRewriteMappingData);
895  waldata_start = waldata = palloc(len);
896 
897  /*
898  * collect data we need to write out, but don't modify ondisk data yet
899  */
900  dlist_foreach_modify(iter, &src->mappings)
901  {
903 
904  pmap = dlist_container(RewriteMappingDataEntry, node, iter.cur);
905 
906  memcpy(waldata, &pmap->map, sizeof(pmap->map));
907  waldata += sizeof(pmap->map);
908 
909  /* remove from the list and free */
910  dlist_delete(&pmap->node);
911  pfree(pmap);
912 
913  /* update bookkeeping */
914  state->rs_num_rewrite_mappings--;
915  src->num_mappings--;
916  }
917 
918  Assert(src->num_mappings == 0);
919  Assert(waldata == waldata_start + len);
920 
921  /*
922  * Note that we deviate from the usual WAL coding practices here,
923  * check the above "Logical rewrite support" comment for reasoning.
924  */
925  written = FileWrite(src->vfd, waldata_start, len,
927  if (written != len)
928  ereport(ERROR,
930  errmsg("could not write to file \"%s\", wrote %d of %d: %m", src->path,
931  written, len)));
932  src->off += len;
933 
934  XLogBeginInsert();
935  XLogRegisterData((char *) (&xlrec), sizeof(xlrec));
936  XLogRegisterData(waldata_start, len);
937 
938  /* write xlog record */
939  XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_REWRITE);
940 
941  pfree(waldata_start);
942  }
943  Assert(state->rs_num_rewrite_mappings == 0);
944 }
945 
946 /*
947  * Logical remapping part of end_heap_rewrite().
948  */
949 static void
951 {
952  HASH_SEQ_STATUS seq_status;
953  RewriteMappingFile *src;
954 
955  /* done, no logical rewrite in progress */
956  if (!state->rs_logical_rewrite)
957  return;
958 
959  /* writeout remaining in-memory entries */
960  if (state->rs_num_rewrite_mappings > 0)
962 
963  /* Iterate over all mappings we have written and fsync the files. */
964  hash_seq_init(&seq_status, state->rs_logical_mappings);
965  while ((src = (RewriteMappingFile *) hash_seq_search(&seq_status)) != NULL)
966  {
968  ereport(ERROR,
970  errmsg("could not fsync file \"%s\": %m", src->path)));
971  FileClose(src->vfd);
972  }
973  /* memory context cleanup will deal with the rest */
974 }
975 
976 /*
977  * Log a single (old->new) mapping for 'xid'.
978  */
979 static void
982 {
983  RewriteMappingFile *src;
985  Oid relid;
986  bool found;
987 
988  relid = RelationGetRelid(state->rs_old_rel);
989 
990  /* look for existing mappings for this 'mapped' xid */
991  src = hash_search(state->rs_logical_mappings, &xid,
992  HASH_ENTER, &found);
993 
994  /*
995  * We haven't yet had the need to map anything for this xid, create
996  * per-xid data structures.
997  */
998  if (!found)
999  {
1000  char path[MAXPGPATH];
1001  Oid dboid;
1002 
1003  if (state->rs_old_rel->rd_rel->relisshared)
1004  dboid = InvalidOid;
1005  else
1006  dboid = MyDatabaseId;
1007 
1008  snprintf(path, MAXPGPATH,
1009  "pg_logical/mappings/" LOGICAL_REWRITE_FORMAT,
1010  dboid, relid,
1011  (uint32) (state->rs_begin_lsn >> 32),
1012  (uint32) state->rs_begin_lsn,
1013  xid, GetCurrentTransactionId());
1014 
1015  dlist_init(&src->mappings);
1016  src->num_mappings = 0;
1017  src->off = 0;
1018  memcpy(src->path, path, sizeof(path));
1019  src->vfd = PathNameOpenFile(path,
1020  O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
1021  if (src->vfd < 0)
1022  ereport(ERROR,
1024  errmsg("could not create file \"%s\": %m", path)));
1025  }
1026 
1027  pmap = MemoryContextAlloc(state->rs_cxt,
1028  sizeof(RewriteMappingDataEntry));
1029  memcpy(&pmap->map, map, sizeof(LogicalRewriteMappingData));
1030  dlist_push_tail(&src->mappings, &pmap->node);
1031  src->num_mappings++;
1032  state->rs_num_rewrite_mappings++;
1033 
1034  /*
1035  * Write out buffer every time we've too many in-memory entries across all
1036  * mapping files.
1037  */
1038  if (state->rs_num_rewrite_mappings >= 1000 /* arbitrary number */ )
1040 }
1041 
1042 /*
1043  * Perform logical remapping for a tuple that's mapped from old_tid to
1044  * new_tuple->t_self by rewrite_heap_tuple() if necessary for the tuple.
1045  */
1046 static void
1048  HeapTuple new_tuple)
1049 {
1050  ItemPointerData new_tid = new_tuple->t_self;
1051  TransactionId cutoff = state->rs_logical_xmin;
1052  TransactionId xmin;
1053  TransactionId xmax;
1054  bool do_log_xmin = false;
1055  bool do_log_xmax = false;
1057 
1058  /* no logical rewrite in progress, we don't need to log anything */
1059  if (!state->rs_logical_rewrite)
1060  return;
1061 
1062  xmin = HeapTupleHeaderGetXmin(new_tuple->t_data);
1063  /* use *GetUpdateXid to correctly deal with multixacts */
1064  xmax = HeapTupleHeaderGetUpdateXid(new_tuple->t_data);
1065 
1066  /*
1067  * Log the mapping iff the tuple has been created recently.
1068  */
1069  if (TransactionIdIsNormal(xmin) && !TransactionIdPrecedes(xmin, cutoff))
1070  do_log_xmin = true;
1071 
1072  if (!TransactionIdIsNormal(xmax))
1073  {
1074  /*
1075  * no xmax is set, can't have any permanent ones, so this check is
1076  * sufficient
1077  */
1078  }
1079  else if (HEAP_XMAX_IS_LOCKED_ONLY(new_tuple->t_data->t_infomask))
1080  {
1081  /* only locked, we don't care */
1082  }
1083  else if (!TransactionIdPrecedes(xmax, cutoff))
1084  {
1085  /* tuple has been deleted recently, log */
1086  do_log_xmax = true;
1087  }
1088 
1089  /* if neither needs to be logged, we're done */
1090  if (!do_log_xmin && !do_log_xmax)
1091  return;
1092 
1093  /* fill out mapping information */
1094  map.old_node = state->rs_old_rel->rd_node;
1095  map.old_tid = old_tid;
1096  map.new_node = state->rs_new_rel->rd_node;
1097  map.new_tid = new_tid;
1098 
1099  /* ---
1100  * Now persist the mapping for the individual xids that are affected. We
1101  * need to log for both xmin and xmax if they aren't the same transaction
1102  * since the mapping files are per "affected" xid.
1103  * We don't muster all that much effort detecting whether xmin and xmax
1104  * are actually the same transaction, we just check whether the xid is the
1105  * same disregarding subtransactions. Logging too much is relatively
1106  * harmless and we could never do the check fully since subtransaction
1107  * data is thrown away during restarts.
1108  * ---
1109  */
1110  if (do_log_xmin)
1111  logical_rewrite_log_mapping(state, xmin, &map);
1112  /* separately log mapping for xmax unless it'd be redundant */
1113  if (do_log_xmax && !TransactionIdEquals(xmin, xmax))
1114  logical_rewrite_log_mapping(state, xmax, &map);
1115 }
1116 
1117 /*
1118  * Replay XLOG_HEAP2_REWRITE records
1119  */
1120 void
1122 {
1123  char path[MAXPGPATH];
1124  int fd;
1125  xl_heap_rewrite_mapping *xlrec;
1126  uint32 len;
1127  char *data;
1128 
1129  xlrec = (xl_heap_rewrite_mapping *) XLogRecGetData(r);
1130 
1131  snprintf(path, MAXPGPATH,
1132  "pg_logical/mappings/" LOGICAL_REWRITE_FORMAT,
1133  xlrec->mapped_db, xlrec->mapped_rel,
1134  (uint32) (xlrec->start_lsn >> 32),
1135  (uint32) xlrec->start_lsn,
1136  xlrec->mapped_xid, XLogRecGetXid(r));
1137 
1138  fd = OpenTransientFile(path,
1139  O_CREAT | O_WRONLY | PG_BINARY);
1140  if (fd < 0)
1141  ereport(ERROR,
1143  errmsg("could not create file \"%s\": %m", path)));
1144 
1145  /*
1146  * Truncate all data that's not guaranteed to have been safely fsynced (by
1147  * previous record or by the last checkpoint).
1148  */
1150  if (ftruncate(fd, xlrec->offset) != 0)
1151  ereport(ERROR,
1153  errmsg("could not truncate file \"%s\" to %u: %m",
1154  path, (uint32) xlrec->offset)));
1156 
1157  /* now seek to the position we want to write our data to */
1158  if (lseek(fd, xlrec->offset, SEEK_SET) != xlrec->offset)
1159  ereport(ERROR,
1161  errmsg("could not seek to end of file \"%s\": %m",
1162  path)));
1163 
1164  data = XLogRecGetData(r) + sizeof(*xlrec);
1165 
1166  len = xlrec->num_mappings * sizeof(LogicalRewriteMappingData);
1167 
1168  /* write out tail end of mapping file (again) */
1170  if (write(fd, data, len) != len)
1171  ereport(ERROR,
1173  errmsg("could not write to file \"%s\": %m", path)));
1175 
1176  /*
1177  * Now fsync all previously written data. We could improve things and only
1178  * do this for the last write to a file, but the required bookkeeping
1179  * doesn't seem worth the trouble.
1180  */
1182  if (pg_fsync(fd) != 0)
1183  ereport(ERROR,
1185  errmsg("could not fsync file \"%s\": %m", path)));
1187 
1188  CloseTransientFile(fd);
1189 }
1190 
1191 /* ---
1192  * Perform a checkpoint for logical rewrite mappings
1193  *
1194  * This serves two tasks:
1195  * 1) Remove all mappings not needed anymore based on the logical restart LSN
1196  * 2) Flush all remaining mappings to disk, so that replay after a checkpoint
1197  * only has to deal with the parts of a mapping that have been written out
1198  * after the checkpoint started.
1199  * ---
1200  */
1201 void
1203 {
1204  XLogRecPtr cutoff;
1205  XLogRecPtr redo;
1206  DIR *mappings_dir;
1207  struct dirent *mapping_de;
1208  char path[MAXPGPATH + 20];
1209 
1210  /*
1211  * We start of with a minimum of the last redo pointer. No new decoding
1212  * slot will start before that, so that's a safe upper bound for removal.
1213  */
1214  redo = GetRedoRecPtr();
1215 
1216  /* now check for the restart ptrs from existing slots */
1218 
1219  /* don't start earlier than the restart lsn */
1220  if (cutoff != InvalidXLogRecPtr && redo < cutoff)
1221  cutoff = redo;
1222 
1223  mappings_dir = AllocateDir("pg_logical/mappings");
1224  while ((mapping_de = ReadDir(mappings_dir, "pg_logical/mappings")) != NULL)
1225  {
1226  struct stat statbuf;
1227  Oid dboid;
1228  Oid relid;
1229  XLogRecPtr lsn;
1230  TransactionId rewrite_xid;
1231  TransactionId create_xid;
1232  uint32 hi,
1233  lo;
1234 
1235  if (strcmp(mapping_de->d_name, ".") == 0 ||
1236  strcmp(mapping_de->d_name, "..") == 0)
1237  continue;
1238 
1239  snprintf(path, sizeof(path), "pg_logical/mappings/%s", mapping_de->d_name);
1240  if (lstat(path, &statbuf) == 0 && !S_ISREG(statbuf.st_mode))
1241  continue;
1242 
1243  /* Skip over files that cannot be ours. */
1244  if (strncmp(mapping_de->d_name, "map-", 4) != 0)
1245  continue;
1246 
1247  if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
1248  &dboid, &relid, &hi, &lo, &rewrite_xid, &create_xid) != 6)
1249  elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
1250 
1251  lsn = ((uint64) hi) << 32 | lo;
1252 
1253  if (lsn < cutoff || cutoff == InvalidXLogRecPtr)
1254  {
1255  elog(DEBUG1, "removing logical rewrite file \"%s\"", path);
1256  if (unlink(path) < 0)
1257  ereport(ERROR,
1259  errmsg("could not remove file \"%s\": %m", path)));
1260  }
1261  else
1262  {
1263  int fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
1264 
1265  /*
1266  * The file cannot vanish due to concurrency since this function
1267  * is the only one removing logical mappings and it's run while
1268  * CheckpointLock is held exclusively.
1269  */
1270  if (fd < 0)
1271  ereport(ERROR,
1273  errmsg("could not open file \"%s\": %m", path)));
1274 
1275  /*
1276  * We could try to avoid fsyncing files that either haven't
1277  * changed or have only been created since the checkpoint's start,
1278  * but it's currently not deemed worth the effort.
1279  */
1281  if (pg_fsync(fd) != 0)
1282  ereport(ERROR,
1284  errmsg("could not fsync file \"%s\": %m", path)));
1286  CloseTransientFile(fd);
1287  }
1288  }
1289  FreeDir(mappings_dir);
1290 }
#define HeapTupleHeaderGetUpdateXid(tup)
Definition: htup_details.h:370
bool HeapTupleHeaderIsOnlyLocked(HeapTupleHeader tuple)
Definition: tqual.c:1596
#define ItemPointerIsValid(pointer)
Definition: itemptr.h:60
HeapTuple heap_copytuple(HeapTuple tuple)
Definition: heaptuple.c:722
#define TOAST_TUPLE_THRESHOLD
Definition: tuptoaster.h:55
union HeapTupleHeaderData::@45 t_choice
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
void MemoryContextDelete(MemoryContext context)
Definition: mcxt.c:211
void end_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:314
#define DEBUG1
Definition: elog.h:25
dlist_node * cur
Definition: ilist.h:180
TidHashKey key
Definition: rewriteheap.c:184
void heap_xlog_logical_rewrite(XLogReaderState *r)
Definition: rewriteheap.c:1121
HeapTupleFields t_heap
Definition: htup_details.h:153
TransactionId rs_logical_xmin
Definition: rewriteheap.c:154
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1351
#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:474
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:248
static void logical_rewrite_heap_tuple(RewriteState state, ItemPointerData old_tid, HeapTuple new_tuple)
Definition: rewriteheap.c:1047
static void logical_end_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:950
HeapTuple toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup, int options)
Definition: tuptoaster.c:534
#define write(a, b, c)
Definition: win32.h:14
HeapTupleHeaderData * HeapTupleHeader
Definition: htup.h:23
struct SMgrRelationData * rd_smgr
Definition: rel.h:57
TransactionId rs_freeze_xid
Definition: rewriteheap.c:152
#define XLOG_HEAP2_REWRITE
Definition: heapam_xlog.h:53
static void dlist_push_tail(dlist_head *head, dlist_node *node)
Definition: ilist.h:317
MultiXactId rs_cutoff_multi
Definition: rewriteheap.c:156
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:144
int errcode(int sqlerrcode)
Definition: elog.c:575
#define HEAP2_XACT_MASK
Definition: htup_details.h:280
#define PageAddItem(page, item, size, offsetNumber, overwrite, is_heap)
Definition: bufpage.h:412
UnresolvedTupData * UnresolvedTup
Definition: rewriteheap.c:189
void heap_sync(Relation rel)
Definition: heapam.c:9356
int snprintf(char *str, size_t count, const char *fmt,...) pg_attribute_printf(3
#define HEAP_INSERT_SKIP_WAL
Definition: heapam.h:28
#define HeapTupleHeaderIndicatesMovedPartitions(tup)
Definition: htup_details.h:453
void CheckPointLogicalRewriteHeap(void)
Definition: rewriteheap.c:1202
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:206
ItemPointerData old_tid
Definition: rewriteheap.c:185
Form_pg_class rd_rel
Definition: rel.h:84
void heap_freetuple(HeapTuple htup)
Definition: heaptuple.c:1773
unsigned int Oid
Definition: postgres_ext.h:31
Definition: dirent.h:9
void ProcArrayGetReplicationSlotXmin(TransactionId *xmin, TransactionId *catalog_xmin)
Definition: procarray.c:2986
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1080
uint16 OffsetNumber
Definition: off.h:24
HeapTupleHeader t_data
Definition: htup.h:68
#define RelationOpenSmgr(relation)
Definition: rel.h:465
Definition: dynahash.c:208
#define dlist_container(type, membername, ptr)
Definition: ilist.h:477
void pfree(void *pointer)
Definition: mcxt.c:1031
uint32 rs_num_rewrite_mappings
Definition: rewriteheap.c:164
#define XLogRecGetData(decoder)
Definition: xlogreader.h:226
Definition: dirent.c:25
#define ERROR
Definition: elog.h:43
ItemPointerData new_tid
Definition: rewriteheap.c:194
Size PageGetHeapFreeSpace(Page page)
Definition: bufpage.c:662
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2396
#define HEAP_XMAX_INVALID
Definition: htup_details.h:204
char path[MAXPGPATH]
Definition: rewriteheap.c:210
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:11203
ItemPointerData t_ctid
Definition: htup_details.h:157
HTAB * rs_unresolved_tups
Definition: rewriteheap.c:161
#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:2072
TransactionId xid
Definition: rewriteheap.c:205
TransactionId GetCurrentTransactionId(void)
Definition: xact.c:417
uint32 t_len
Definition: htup.h:64
static void logical_rewrite_log_mapping(RewriteState state, TransactionId xid, LogicalRewriteMappingData *map)
Definition: rewriteheap.c:980
#define MaxHeapTupleSize
Definition: htup_details.h:578
int errcode_for_file_access(void)
Definition: elog.c:598
XLogRecPtr ReplicationSlotsComputeLogicalRestartLSN(void)
Definition: slot.c:784
#define InvalidTransactionId
Definition: transam.h:31
unsigned int uint32
Definition: c.h:325
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2600
int FileWrite(File file, char *buffer, int amount, uint32 wait_event_info)
Definition: fd.c:1951
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1260
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:7030
static void dlist_delete(dlist_node *node)
Definition: ilist.h:358
#define ereport(elevel, rest)
Definition: elog.h:122
XLogRecPtr rs_begin_lsn
Definition: rewriteheap.c:160
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
#define RelationGetTargetPageFreeSpace(relation, defaultff)
Definition: rel.h:298
#define S_ISREG(m)
Definition: win32_port.h:310
HTAB * rs_logical_mappings
Definition: rewriteheap.c:163
int CloseTransientFile(int fd)
Definition: fd.c:2566
#define AllocSetContextCreate(parent, name, allocparams)
Definition: memutils.h:170
TransactionId rs_oldest_xmin
Definition: rewriteheap.c:150
#define stat(a, b)
Definition: win32_port.h:266
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:231
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:224
MemoryContext rs_cxt
Definition: rewriteheap.c:158
void * palloc0(Size size)
Definition: mcxt.c:955
#define RelationIsAccessibleInLogicalDecoding(relation)
Definition: rel.h:564
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:227
Oid MyDatabaseId
Definition: globals.c:86
Size keysize
Definition: hsearch.h:72
#define RelationGetNumberOfBlocks(reln)
Definition: bufmgr.h:199
ItemPointerData tid
Definition: rewriteheap.c:176
#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:484
RelFileNode rd_node
Definition: rel.h:55
void FileClose(File file)
Definition: fd.c:1742
static void logical_begin_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:803
ItemPointerData new_tid
Definition: rewriteheap.h:40
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:699
HTAB * rs_old_new_tid_map
Definition: rewriteheap.c:162
Definition: regguts.h:298
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2666
#define HeapTupleHeaderGetXmin(tup)
Definition: htup_details.h:318
size_t Size
Definition: c.h:433
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1236
dlist_head mappings
Definition: rewriteheap.c:209
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1195
#define MAXALIGN(LEN)
Definition: c.h:652
XLogRecPtr GetRedoRecPtr(void)
Definition: xlog.c:8212
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1389
#define RelationNeedsWAL(relation)
Definition: rel.h:510
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:852
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:600
#define lstat(path, sb)
Definition: win32_port.h:255
#define HEAP_INSERT_SKIP_FSM
Definition: heapam.h:29
bool rewrite_heap_dead_tuple(RewriteState state, HeapTuple old_tuple)
Definition: rewriteheap.c:582
#define ItemPointerSetInvalid(pointer)
Definition: itemptr.h:150
#define HeapTupleHasExternal(tuple)
Definition: htup_details.h:691
void * palloc(Size size)
Definition: mcxt.c:924
int errmsg(const char *fmt,...)
Definition: elog.c:797
TransactionId xmin
Definition: rewriteheap.c:175
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:771
LogicalRewriteMappingData map
Definition: rewriteheap.c:219
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:972
static void raw_heap_insert(RewriteState state, HeapTuple tup)
Definition: rewriteheap.c:632
struct LogicalRewriteMappingData LogicalRewriteMappingData
#define HEAP_XACT_MASK
Definition: htup_details.h:215
int pg_fsync(int fd)
Definition: fd.c:341
Relation rs_old_rel
Definition: rewriteheap.c:143
char d_name[MAX_PATH]
Definition: dirent.h:14
#define elog
Definition: elog.h:219
#define HEAP_DEFAULT_FILLFACTOR
Definition: rel.h:269
struct RewriteMappingDataEntry RewriteMappingDataEntry
#define TransactionIdIsNormal(xid)
Definition: transam.h:42
BlockNumber rs_blockno
Definition: rewriteheap.c:146
void XLogBeginInsert(void)
Definition: xloginsert.c:120
void rewrite_heap_tuple(RewriteState state, HeapTuple old_tuple, HeapTuple new_tuple)
Definition: rewriteheap.c:380
#define RelationGetRelid(relation)
Definition: rel.h:407
OldToNewMappingData * OldToNewMapping
Definition: rewriteheap.c:197
int FreeDir(DIR *dir)
Definition: fd.c:2718
#define PageGetItem(page, itemId)
Definition: bufpage.h:336
Pointer Page
Definition: bufpage.h:74
#define ItemPointerSet(pointer, blockNumber, offNum)
Definition: itemptr.h:105
void PageInit(Page page, Size pageSize, Size specialSize)
Definition: bufpage.c:41
ItemPointerData old_tid
Definition: rewriteheap.h:39
#define ftruncate(a, b)
Definition: win32_port.h:60