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) 1996-2020, 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_logical_rewrite; /* do we need to do logical rewriting */
140  TransactionId rs_oldest_xmin; /* oldest xmin used by caller to determine
141  * tuple visibility */
142  TransactionId rs_freeze_xid; /* Xid that will be used as freeze cutoff
143  * point */
144  TransactionId rs_logical_xmin; /* Xid that will be used as cutoff point
145  * for logical rewrites */
146  MultiXactId rs_cutoff_multi; /* MultiXactId that will be used as cutoff
147  * point for multixacts */
148  MemoryContext rs_cxt; /* for hash tables and entries and tuples in
149  * them */
150  XLogRecPtr rs_begin_lsn; /* XLogInsertLsn when starting the rewrite */
151  HTAB *rs_unresolved_tups; /* unmatched A tuples */
152  HTAB *rs_old_new_tid_map; /* unmatched B tuples */
153  HTAB *rs_logical_mappings; /* logical remapping files */
154  uint32 rs_num_rewrite_mappings; /* # in memory mappings */
156 
157 /*
158  * The lookup keys for the hash tables are tuple TID and xmin (we must check
159  * both to avoid false matches from dead tuples). Beware that there is
160  * probably some padding space in this struct; it must be zeroed out for
161  * correct hashtable operation.
162  */
163 typedef struct
164 {
165  TransactionId xmin; /* tuple xmin */
166  ItemPointerData tid; /* tuple location in old heap */
167 } TidHashKey;
168 
169 /*
170  * Entry structures for the hash tables
171  */
172 typedef struct
173 {
174  TidHashKey key; /* expected xmin/old location of B tuple */
175  ItemPointerData old_tid; /* A's location in the old heap */
176  HeapTuple tuple; /* A's tuple contents */
178 
180 
181 typedef struct
182 {
183  TidHashKey key; /* actual xmin/old location of B tuple */
184  ItemPointerData new_tid; /* where we put it in the new heap */
186 
188 
189 /*
190  * In-Memory data for an xid that might need logical remapping entries
191  * to be logged.
192  */
193 typedef struct RewriteMappingFile
194 {
195  TransactionId xid; /* xid that might need to see the row */
196  int vfd; /* fd of mappings file */
197  off_t off; /* how far have we written yet */
198  uint32 num_mappings; /* number of in-memory mappings */
199  dlist_head mappings; /* list of in-memory mappings */
200  char path[MAXPGPATH]; /* path, for error messages */
202 
203 /*
204  * A single In-Memory logical rewrite mapping, hanging off
205  * RewriteMappingFile->mappings.
206  */
208 {
209  LogicalRewriteMappingData map; /* map between old and new location of the
210  * tuple */
213 
214 
215 /* prototypes for internal functions */
216 static void raw_heap_insert(RewriteState state, HeapTuple tup);
217 
218 /* internal logical remapping prototypes */
222 
223 
224 /*
225  * Begin a rewrite of a table
226  *
227  * old_heap old, locked heap relation tuples will be read from
228  * new_heap new, locked heap relation to insert tuples to
229  * oldest_xmin xid used by the caller to determine which tuples are dead
230  * freeze_xid xid before which tuples will be frozen
231  * cutoff_multi multixact before which multis will be removed
232  *
233  * Returns an opaque RewriteState, allocated in current memory context,
234  * to be used in subsequent calls to the other functions.
235  */
237 begin_heap_rewrite(Relation old_heap, Relation new_heap, TransactionId oldest_xmin,
238  TransactionId freeze_xid, MultiXactId cutoff_multi)
239 {
241  MemoryContext rw_cxt;
242  MemoryContext old_cxt;
243  HASHCTL hash_ctl;
244 
245  /*
246  * To ease cleanup, make a separate context that will contain the
247  * RewriteState struct itself plus all subsidiary data.
248  */
250  "Table rewrite",
252  old_cxt = MemoryContextSwitchTo(rw_cxt);
253 
254  /* Create and fill in the state struct */
255  state = palloc0(sizeof(RewriteStateData));
256 
257  state->rs_old_rel = old_heap;
258  state->rs_new_rel = new_heap;
259  state->rs_buffer = (Page) palloc(BLCKSZ);
260  /* new_heap needn't be empty, just locked */
261  state->rs_blockno = RelationGetNumberOfBlocks(new_heap);
262  state->rs_buffer_valid = false;
263  state->rs_oldest_xmin = oldest_xmin;
264  state->rs_freeze_xid = freeze_xid;
265  state->rs_cutoff_multi = cutoff_multi;
266  state->rs_cxt = rw_cxt;
267 
268  /* Initialize hash tables used to track update chains */
269  memset(&hash_ctl, 0, sizeof(hash_ctl));
270  hash_ctl.keysize = sizeof(TidHashKey);
271  hash_ctl.entrysize = sizeof(UnresolvedTupData);
272  hash_ctl.hcxt = state->rs_cxt;
273 
274  state->rs_unresolved_tups =
275  hash_create("Rewrite / Unresolved ctids",
276  128, /* arbitrary initial size */
277  &hash_ctl,
279 
280  hash_ctl.entrysize = sizeof(OldToNewMappingData);
281 
282  state->rs_old_new_tid_map =
283  hash_create("Rewrite / Old to new tid map",
284  128, /* arbitrary initial size */
285  &hash_ctl,
287 
288  MemoryContextSwitchTo(old_cxt);
289 
291 
292  return state;
293 }
294 
295 /*
296  * End a rewrite.
297  *
298  * state and any other resources are freed.
299  */
300 void
302 {
303  HASH_SEQ_STATUS seq_status;
304  UnresolvedTup unresolved;
305 
306  /*
307  * Write any remaining tuples in the UnresolvedTups table. If we have any
308  * left, they should in fact be dead, but let's err on the safe side.
309  */
310  hash_seq_init(&seq_status, state->rs_unresolved_tups);
311 
312  while ((unresolved = hash_seq_search(&seq_status)) != NULL)
313  {
314  ItemPointerSetInvalid(&unresolved->tuple->t_data->t_ctid);
315  raw_heap_insert(state, unresolved->tuple);
316  }
317 
318  /* Write the last page, if any */
319  if (state->rs_buffer_valid)
320  {
321  if (RelationNeedsWAL(state->rs_new_rel))
322  log_newpage(&state->rs_new_rel->rd_node,
323  MAIN_FORKNUM,
324  state->rs_blockno,
325  state->rs_buffer,
326  true);
328 
330 
332  (char *) state->rs_buffer, true);
333  }
334 
335  /*
336  * When we WAL-logged rel pages, we must nonetheless fsync them. The
337  * reason is the same as in storage.c's RelationCopyStorage(): we're
338  * writing data that's not in shared buffers, and so a CHECKPOINT
339  * occurring during the rewriteheap operation won't have fsync'd data we
340  * wrote before the checkpoint.
341  */
342  if (RelationNeedsWAL(state->rs_new_rel))
344 
346 
347  /* Deleting the context frees everything */
348  MemoryContextDelete(state->rs_cxt);
349 }
350 
351 /*
352  * Add a tuple to the new heap.
353  *
354  * Visibility information is copied from the original tuple, except that
355  * we "freeze" very-old tuples. Note that since we scribble on new_tuple,
356  * it had better be temp storage not a pointer to the original tuple.
357  *
358  * state opaque state as returned by begin_heap_rewrite
359  * old_tuple original tuple in the old heap
360  * new_tuple new, rewritten tuple to be inserted to new heap
361  */
362 void
364  HeapTuple old_tuple, HeapTuple new_tuple)
365 {
366  MemoryContext old_cxt;
367  ItemPointerData old_tid;
368  TidHashKey hashkey;
369  bool found;
370  bool free_new;
371 
372  old_cxt = MemoryContextSwitchTo(state->rs_cxt);
373 
374  /*
375  * Copy the original tuple's visibility information into new_tuple.
376  *
377  * XXX we might later need to copy some t_infomask2 bits, too? Right now,
378  * we intentionally clear the HOT status bits.
379  */
380  memcpy(&new_tuple->t_data->t_choice.t_heap,
381  &old_tuple->t_data->t_choice.t_heap,
382  sizeof(HeapTupleFields));
383 
384  new_tuple->t_data->t_infomask &= ~HEAP_XACT_MASK;
385  new_tuple->t_data->t_infomask2 &= ~HEAP2_XACT_MASK;
386  new_tuple->t_data->t_infomask |=
387  old_tuple->t_data->t_infomask & HEAP_XACT_MASK;
388 
389  /*
390  * While we have our hands on the tuple, we may as well freeze any
391  * eligible xmin or xmax, so that future VACUUM effort can be saved.
392  */
393  heap_freeze_tuple(new_tuple->t_data,
394  state->rs_old_rel->rd_rel->relfrozenxid,
395  state->rs_old_rel->rd_rel->relminmxid,
396  state->rs_freeze_xid,
397  state->rs_cutoff_multi);
398 
399  /*
400  * Invalid ctid means that ctid should point to the tuple itself. We'll
401  * override it later if the tuple is part of an update chain.
402  */
403  ItemPointerSetInvalid(&new_tuple->t_data->t_ctid);
404 
405  /*
406  * If the tuple has been updated, check the old-to-new mapping hash table.
407  */
408  if (!((old_tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
409  HeapTupleHeaderIsOnlyLocked(old_tuple->t_data)) &&
411  !(ItemPointerEquals(&(old_tuple->t_self),
412  &(old_tuple->t_data->t_ctid))))
413  {
414  OldToNewMapping mapping;
415 
416  memset(&hashkey, 0, sizeof(hashkey));
417  hashkey.xmin = HeapTupleHeaderGetUpdateXid(old_tuple->t_data);
418  hashkey.tid = old_tuple->t_data->t_ctid;
419 
420  mapping = (OldToNewMapping)
421  hash_search(state->rs_old_new_tid_map, &hashkey,
422  HASH_FIND, NULL);
423 
424  if (mapping != NULL)
425  {
426  /*
427  * We've already copied the tuple that t_ctid points to, so we can
428  * set the ctid of this tuple to point to the new location, and
429  * insert it right away.
430  */
431  new_tuple->t_data->t_ctid = mapping->new_tid;
432 
433  /* We don't need the mapping entry anymore */
434  hash_search(state->rs_old_new_tid_map, &hashkey,
435  HASH_REMOVE, &found);
436  Assert(found);
437  }
438  else
439  {
440  /*
441  * We haven't seen the tuple t_ctid points to yet. Stash this
442  * tuple into unresolved_tups to be written later.
443  */
444  UnresolvedTup unresolved;
445 
446  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
447  HASH_ENTER, &found);
448  Assert(!found);
449 
450  unresolved->old_tid = old_tuple->t_self;
451  unresolved->tuple = heap_copytuple(new_tuple);
452 
453  /*
454  * We can't do anything more now, since we don't know where the
455  * tuple will be written.
456  */
457  MemoryContextSwitchTo(old_cxt);
458  return;
459  }
460  }
461 
462  /*
463  * Now we will write the tuple, and then check to see if it is the B tuple
464  * in any new or known pair. When we resolve a known pair, we will be
465  * able to write that pair's A tuple, and then we have to check if it
466  * resolves some other pair. Hence, we need a loop here.
467  */
468  old_tid = old_tuple->t_self;
469  free_new = false;
470 
471  for (;;)
472  {
473  ItemPointerData new_tid;
474 
475  /* Insert the tuple and find out where it's put in new_heap */
476  raw_heap_insert(state, new_tuple);
477  new_tid = new_tuple->t_self;
478 
479  logical_rewrite_heap_tuple(state, old_tid, new_tuple);
480 
481  /*
482  * If the tuple is the updated version of a row, and the prior version
483  * wouldn't be DEAD yet, then we need to either resolve the prior
484  * version (if it's waiting in rs_unresolved_tups), or make an entry
485  * in rs_old_new_tid_map (so we can resolve it when we do see it). The
486  * previous tuple's xmax would equal this one's xmin, so it's
487  * RECENTLY_DEAD if and only if the xmin is not before OldestXmin.
488  */
489  if ((new_tuple->t_data->t_infomask & HEAP_UPDATED) &&
491  state->rs_oldest_xmin))
492  {
493  /*
494  * Okay, this is B in an update pair. See if we've seen A.
495  */
496  UnresolvedTup unresolved;
497 
498  memset(&hashkey, 0, sizeof(hashkey));
499  hashkey.xmin = HeapTupleHeaderGetXmin(new_tuple->t_data);
500  hashkey.tid = old_tid;
501 
502  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
503  HASH_FIND, NULL);
504 
505  if (unresolved != NULL)
506  {
507  /*
508  * We have seen and memorized the previous tuple already. Now
509  * that we know where we inserted the tuple its t_ctid points
510  * to, fix its t_ctid and insert it to the new heap.
511  */
512  if (free_new)
513  heap_freetuple(new_tuple);
514  new_tuple = unresolved->tuple;
515  free_new = true;
516  old_tid = unresolved->old_tid;
517  new_tuple->t_data->t_ctid = new_tid;
518 
519  /*
520  * We don't need the hash entry anymore, but don't free its
521  * tuple just yet.
522  */
523  hash_search(state->rs_unresolved_tups, &hashkey,
524  HASH_REMOVE, &found);
525  Assert(found);
526 
527  /* loop back to insert the previous tuple in the chain */
528  continue;
529  }
530  else
531  {
532  /*
533  * Remember the new tid of this tuple. We'll use it to set the
534  * ctid when we find the previous tuple in the chain.
535  */
536  OldToNewMapping mapping;
537 
538  mapping = hash_search(state->rs_old_new_tid_map, &hashkey,
539  HASH_ENTER, &found);
540  Assert(!found);
541 
542  mapping->new_tid = new_tid;
543  }
544  }
545 
546  /* Done with this (chain of) tuples, for now */
547  if (free_new)
548  heap_freetuple(new_tuple);
549  break;
550  }
551 
552  MemoryContextSwitchTo(old_cxt);
553 }
554 
555 /*
556  * Register a dead tuple with an ongoing rewrite. Dead tuples are not
557  * copied to the new table, but we still make note of them so that we
558  * can release some resources earlier.
559  *
560  * Returns true if a tuple was removed from the unresolved_tups table.
561  * This indicates that that tuple, previously thought to be "recently dead",
562  * is now known really dead and won't be written to the output.
563  */
564 bool
566 {
567  /*
568  * If we have already seen an earlier tuple in the update chain that
569  * points to this tuple, let's forget about that earlier tuple. It's in
570  * fact dead as well, our simple xmax < OldestXmin test in
571  * HeapTupleSatisfiesVacuum just wasn't enough to detect it. It happens
572  * when xmin of a tuple is greater than xmax, which sounds
573  * counter-intuitive but is perfectly valid.
574  *
575  * We don't bother to try to detect the situation the other way round,
576  * when we encounter the dead tuple first and then the recently dead one
577  * that points to it. If that happens, we'll have some unmatched entries
578  * in the UnresolvedTups hash table at the end. That can happen anyway,
579  * because a vacuum might have removed the dead tuple in the chain before
580  * us.
581  */
582  UnresolvedTup unresolved;
583  TidHashKey hashkey;
584  bool found;
585 
586  memset(&hashkey, 0, sizeof(hashkey));
587  hashkey.xmin = HeapTupleHeaderGetXmin(old_tuple->t_data);
588  hashkey.tid = old_tuple->t_self;
589 
590  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
591  HASH_FIND, NULL);
592 
593  if (unresolved != NULL)
594  {
595  /* Need to free the contained tuple as well as the hashtable entry */
596  heap_freetuple(unresolved->tuple);
597  hash_search(state->rs_unresolved_tups, &hashkey,
598  HASH_REMOVE, &found);
599  Assert(found);
600  return true;
601  }
602 
603  return false;
604 }
605 
606 /*
607  * Insert a tuple to the new relation. This has to track heap_insert
608  * and its subsidiary functions!
609  *
610  * t_self of the tuple is set to the new TID of the tuple. If t_ctid of the
611  * tuple is invalid on entry, it's replaced with the new TID as well (in
612  * the inserted data only, not in the caller's copy).
613  */
614 static void
616 {
617  Page page = state->rs_buffer;
618  Size pageFreeSpace,
619  saveFreeSpace;
620  Size len;
621  OffsetNumber newoff;
622  HeapTuple heaptup;
623 
624  /*
625  * If the new tuple is too big for storage or contains already toasted
626  * out-of-line attributes from some other relation, invoke the toaster.
627  *
628  * Note: below this point, heaptup is the data we actually intend to store
629  * into the relation; tup is the caller's original untoasted data.
630  */
631  if (state->rs_new_rel->rd_rel->relkind == RELKIND_TOASTVALUE)
632  {
633  /* toast table entries should never be recursively toasted */
635  heaptup = tup;
636  }
637  else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
638  {
640 
641  /*
642  * While rewriting the heap for VACUUM FULL / CLUSTER, make sure data
643  * for the TOAST table are not logically decoded. The main heap is
644  * WAL-logged as XLOG FPI records, which are not logically decoded.
645  */
646  options |= HEAP_INSERT_NO_LOGICAL;
647 
648  heaptup = heap_toast_insert_or_update(state->rs_new_rel, tup, NULL,
649  options);
650  }
651  else
652  heaptup = tup;
653 
654  len = MAXALIGN(heaptup->t_len); /* be conservative */
655 
656  /*
657  * If we're gonna fail for oversize tuple, do it right away
658  */
659  if (len > MaxHeapTupleSize)
660  ereport(ERROR,
661  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
662  errmsg("row is too big: size %zu, maximum size %zu",
663  len, MaxHeapTupleSize)));
664 
665  /* Compute desired extra freespace due to fillfactor option */
666  saveFreeSpace = RelationGetTargetPageFreeSpace(state->rs_new_rel,
668 
669  /* Now we can check to see if there's enough free space already. */
670  if (state->rs_buffer_valid)
671  {
672  pageFreeSpace = PageGetHeapFreeSpace(page);
673 
674  if (len + saveFreeSpace > pageFreeSpace)
675  {
676  /* Doesn't fit, so write out the existing page */
677 
678  /* XLOG stuff */
679  if (RelationNeedsWAL(state->rs_new_rel))
680  log_newpage(&state->rs_new_rel->rd_node,
681  MAIN_FORKNUM,
682  state->rs_blockno,
683  page,
684  true);
685 
686  /*
687  * Now write the page. We say skipFsync = true because there's no
688  * need for smgr to schedule an fsync for this write; we'll do it
689  * ourselves in end_heap_rewrite.
690  */
692 
693  PageSetChecksumInplace(page, state->rs_blockno);
694 
696  state->rs_blockno, (char *) page, true);
697 
698  state->rs_blockno++;
699  state->rs_buffer_valid = false;
700  }
701  }
702 
703  if (!state->rs_buffer_valid)
704  {
705  /* Initialize a new empty page */
706  PageInit(page, BLCKSZ, 0);
707  state->rs_buffer_valid = true;
708  }
709 
710  /* And now we can insert the tuple into the page */
711  newoff = PageAddItem(page, (Item) heaptup->t_data, heaptup->t_len,
712  InvalidOffsetNumber, false, true);
713  if (newoff == InvalidOffsetNumber)
714  elog(ERROR, "failed to add tuple");
715 
716  /* Update caller's t_self to the actual position where it was stored */
717  ItemPointerSet(&(tup->t_self), state->rs_blockno, newoff);
718 
719  /*
720  * Insert the correct position into CTID of the stored tuple, too, if the
721  * caller didn't supply a valid CTID.
722  */
723  if (!ItemPointerIsValid(&tup->t_data->t_ctid))
724  {
725  ItemId newitemid;
726  HeapTupleHeader onpage_tup;
727 
728  newitemid = PageGetItemId(page, newoff);
729  onpage_tup = (HeapTupleHeader) PageGetItem(page, newitemid);
730 
731  onpage_tup->t_ctid = tup->t_self;
732  }
733 
734  /* If heaptup is a private copy, release it. */
735  if (heaptup != tup)
736  heap_freetuple(heaptup);
737 }
738 
739 /* ------------------------------------------------------------------------
740  * Logical rewrite support
741  *
742  * When doing logical decoding - which relies on using cmin/cmax of catalog
743  * tuples, via xl_heap_new_cid records - heap rewrites have to log enough
744  * information to allow the decoding backend to updates its internal mapping
745  * of (relfilenode,ctid) => (cmin, cmax) to be correct for the rewritten heap.
746  *
747  * For that, every time we find a tuple that's been modified in a catalog
748  * relation within the xmin horizon of any decoding slot, we log a mapping
749  * from the old to the new location.
750  *
751  * To deal with rewrites that abort the filename of a mapping file contains
752  * the xid of the transaction performing the rewrite, which then can be
753  * checked before being read in.
754  *
755  * For efficiency we don't immediately spill every single map mapping for a
756  * row to disk but only do so in batches when we've collected several of them
757  * in memory or when end_heap_rewrite() has been called.
758  *
759  * Crash-Safety: This module diverts from the usual patterns of doing WAL
760  * since it cannot rely on checkpoint flushing out all buffers and thus
761  * waiting for exclusive locks on buffers. Usually the XLogInsert() covering
762  * buffer modifications is performed while the buffer(s) that are being
763  * modified are exclusively locked guaranteeing that both the WAL record and
764  * the modified heap are on either side of the checkpoint. But since the
765  * mapping files we log aren't in shared_buffers that interlock doesn't work.
766  *
767  * Instead we simply write the mapping files out to disk, *before* the
768  * XLogInsert() is performed. That guarantees that either the XLogInsert() is
769  * inserted after the checkpoint's redo pointer or that the checkpoint (via
770  * CheckPointLogicalRewriteHeap()) has flushed the (partial) mapping file to
771  * disk. That leaves the tail end that has not yet been flushed open to
772  * corruption, which is solved by including the current offset in the
773  * xl_heap_rewrite_mapping records and truncating the mapping file to it
774  * during replay. Every time a rewrite is finished all generated mapping files
775  * are synced to disk.
776  *
777  * Note that if we were only concerned about crash safety we wouldn't have to
778  * deal with WAL logging at all - an fsync() at the end of a rewrite would be
779  * sufficient for crash safety. Any mapping that hasn't been safely flushed to
780  * disk has to be by an aborted (explicitly or via a crash) transaction and is
781  * ignored by virtue of the xid in its name being subject to a
782  * TransactionDidCommit() check. But we want to support having standbys via
783  * physical replication, both for availability and to do logical decoding
784  * there.
785  * ------------------------------------------------------------------------
786  */
787 
788 /*
789  * Do preparations for logging logical mappings during a rewrite if
790  * necessary. If we detect that we don't need to log anything we'll prevent
791  * any further action by the various logical rewrite functions.
792  */
793 static void
795 {
796  HASHCTL hash_ctl;
797  TransactionId logical_xmin;
798 
799  /*
800  * We only need to persist these mappings if the rewritten table can be
801  * accessed during logical decoding, if not, we can skip doing any
802  * additional work.
803  */
804  state->rs_logical_rewrite =
806 
807  if (!state->rs_logical_rewrite)
808  return;
809 
810  ProcArrayGetReplicationSlotXmin(NULL, &logical_xmin);
811 
812  /*
813  * If there are no logical slots in progress we don't need to do anything,
814  * there cannot be any remappings for relevant rows yet. The relation's
815  * lock protects us against races.
816  */
817  if (logical_xmin == InvalidTransactionId)
818  {
819  state->rs_logical_rewrite = false;
820  return;
821  }
822 
823  state->rs_logical_xmin = logical_xmin;
825  state->rs_num_rewrite_mappings = 0;
826 
827  memset(&hash_ctl, 0, sizeof(hash_ctl));
828  hash_ctl.keysize = sizeof(TransactionId);
829  hash_ctl.entrysize = sizeof(RewriteMappingFile);
830  hash_ctl.hcxt = state->rs_cxt;
831 
832  state->rs_logical_mappings =
833  hash_create("Logical rewrite mapping",
834  128, /* arbitrary initial size */
835  &hash_ctl,
837 }
838 
839 /*
840  * Flush all logical in-memory mappings to disk, but don't fsync them yet.
841  */
842 static void
844 {
845  HASH_SEQ_STATUS seq_status;
846  RewriteMappingFile *src;
847  dlist_mutable_iter iter;
848 
849  Assert(state->rs_logical_rewrite);
850 
851  /* no logical rewrite in progress, no need to iterate over mappings */
852  if (state->rs_num_rewrite_mappings == 0)
853  return;
854 
855  elog(DEBUG1, "flushing %u logical rewrite mapping entries",
856  state->rs_num_rewrite_mappings);
857 
858  hash_seq_init(&seq_status, state->rs_logical_mappings);
859  while ((src = (RewriteMappingFile *) hash_seq_search(&seq_status)) != NULL)
860  {
861  char *waldata;
862  char *waldata_start;
864  Oid dboid;
865  uint32 len;
866  int written;
867 
868  /* this file hasn't got any new mappings */
869  if (src->num_mappings == 0)
870  continue;
871 
872  if (state->rs_old_rel->rd_rel->relisshared)
873  dboid = InvalidOid;
874  else
875  dboid = MyDatabaseId;
876 
877  xlrec.num_mappings = src->num_mappings;
878  xlrec.mapped_rel = RelationGetRelid(state->rs_old_rel);
879  xlrec.mapped_xid = src->xid;
880  xlrec.mapped_db = dboid;
881  xlrec.offset = src->off;
882  xlrec.start_lsn = state->rs_begin_lsn;
883 
884  /* write all mappings consecutively */
885  len = src->num_mappings * sizeof(LogicalRewriteMappingData);
886  waldata_start = waldata = palloc(len);
887 
888  /*
889  * collect data we need to write out, but don't modify ondisk data yet
890  */
891  dlist_foreach_modify(iter, &src->mappings)
892  {
894 
895  pmap = dlist_container(RewriteMappingDataEntry, node, iter.cur);
896 
897  memcpy(waldata, &pmap->map, sizeof(pmap->map));
898  waldata += sizeof(pmap->map);
899 
900  /* remove from the list and free */
901  dlist_delete(&pmap->node);
902  pfree(pmap);
903 
904  /* update bookkeeping */
905  state->rs_num_rewrite_mappings--;
906  src->num_mappings--;
907  }
908 
909  Assert(src->num_mappings == 0);
910  Assert(waldata == waldata_start + len);
911 
912  /*
913  * Note that we deviate from the usual WAL coding practices here,
914  * check the above "Logical rewrite support" comment for reasoning.
915  */
916  written = FileWrite(src->vfd, waldata_start, len, src->off,
918  if (written != len)
919  ereport(ERROR,
921  errmsg("could not write to file \"%s\", wrote %d of %d: %m", src->path,
922  written, len)));
923  src->off += len;
924 
925  XLogBeginInsert();
926  XLogRegisterData((char *) (&xlrec), sizeof(xlrec));
927  XLogRegisterData(waldata_start, len);
928 
929  /* write xlog record */
930  XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_REWRITE);
931 
932  pfree(waldata_start);
933  }
934  Assert(state->rs_num_rewrite_mappings == 0);
935 }
936 
937 /*
938  * Logical remapping part of end_heap_rewrite().
939  */
940 static void
942 {
943  HASH_SEQ_STATUS seq_status;
944  RewriteMappingFile *src;
945 
946  /* done, no logical rewrite in progress */
947  if (!state->rs_logical_rewrite)
948  return;
949 
950  /* writeout remaining in-memory entries */
951  if (state->rs_num_rewrite_mappings > 0)
953 
954  /* Iterate over all mappings we have written and fsync the files. */
955  hash_seq_init(&seq_status, state->rs_logical_mappings);
956  while ((src = (RewriteMappingFile *) hash_seq_search(&seq_status)) != NULL)
957  {
961  errmsg("could not fsync file \"%s\": %m", src->path)));
962  FileClose(src->vfd);
963  }
964  /* memory context cleanup will deal with the rest */
965 }
966 
967 /*
968  * Log a single (old->new) mapping for 'xid'.
969  */
970 static void
973 {
974  RewriteMappingFile *src;
976  Oid relid;
977  bool found;
978 
979  relid = RelationGetRelid(state->rs_old_rel);
980 
981  /* look for existing mappings for this 'mapped' xid */
982  src = hash_search(state->rs_logical_mappings, &xid,
983  HASH_ENTER, &found);
984 
985  /*
986  * We haven't yet had the need to map anything for this xid, create
987  * per-xid data structures.
988  */
989  if (!found)
990  {
991  char path[MAXPGPATH];
992  Oid dboid;
993 
994  if (state->rs_old_rel->rd_rel->relisshared)
995  dboid = InvalidOid;
996  else
997  dboid = MyDatabaseId;
998 
999  snprintf(path, MAXPGPATH,
1000  "pg_logical/mappings/" LOGICAL_REWRITE_FORMAT,
1001  dboid, relid,
1002  (uint32) (state->rs_begin_lsn >> 32),
1003  (uint32) state->rs_begin_lsn,
1004  xid, GetCurrentTransactionId());
1005 
1006  dlist_init(&src->mappings);
1007  src->num_mappings = 0;
1008  src->off = 0;
1009  memcpy(src->path, path, sizeof(path));
1010  src->vfd = PathNameOpenFile(path,
1011  O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
1012  if (src->vfd < 0)
1013  ereport(ERROR,
1015  errmsg("could not create file \"%s\": %m", path)));
1016  }
1017 
1018  pmap = MemoryContextAlloc(state->rs_cxt,
1019  sizeof(RewriteMappingDataEntry));
1020  memcpy(&pmap->map, map, sizeof(LogicalRewriteMappingData));
1021  dlist_push_tail(&src->mappings, &pmap->node);
1022  src->num_mappings++;
1023  state->rs_num_rewrite_mappings++;
1024 
1025  /*
1026  * Write out buffer every time we've too many in-memory entries across all
1027  * mapping files.
1028  */
1029  if (state->rs_num_rewrite_mappings >= 1000 /* arbitrary number */ )
1031 }
1032 
1033 /*
1034  * Perform logical remapping for a tuple that's mapped from old_tid to
1035  * new_tuple->t_self by rewrite_heap_tuple() if necessary for the tuple.
1036  */
1037 static void
1039  HeapTuple new_tuple)
1040 {
1041  ItemPointerData new_tid = new_tuple->t_self;
1042  TransactionId cutoff = state->rs_logical_xmin;
1043  TransactionId xmin;
1044  TransactionId xmax;
1045  bool do_log_xmin = false;
1046  bool do_log_xmax = false;
1048 
1049  /* no logical rewrite in progress, we don't need to log anything */
1050  if (!state->rs_logical_rewrite)
1051  return;
1052 
1053  xmin = HeapTupleHeaderGetXmin(new_tuple->t_data);
1054  /* use *GetUpdateXid to correctly deal with multixacts */
1055  xmax = HeapTupleHeaderGetUpdateXid(new_tuple->t_data);
1056 
1057  /*
1058  * Log the mapping iff the tuple has been created recently.
1059  */
1060  if (TransactionIdIsNormal(xmin) && !TransactionIdPrecedes(xmin, cutoff))
1061  do_log_xmin = true;
1062 
1063  if (!TransactionIdIsNormal(xmax))
1064  {
1065  /*
1066  * no xmax is set, can't have any permanent ones, so this check is
1067  * sufficient
1068  */
1069  }
1070  else if (HEAP_XMAX_IS_LOCKED_ONLY(new_tuple->t_data->t_infomask))
1071  {
1072  /* only locked, we don't care */
1073  }
1074  else if (!TransactionIdPrecedes(xmax, cutoff))
1075  {
1076  /* tuple has been deleted recently, log */
1077  do_log_xmax = true;
1078  }
1079 
1080  /* if neither needs to be logged, we're done */
1081  if (!do_log_xmin && !do_log_xmax)
1082  return;
1083 
1084  /* fill out mapping information */
1085  map.old_node = state->rs_old_rel->rd_node;
1086  map.old_tid = old_tid;
1087  map.new_node = state->rs_new_rel->rd_node;
1088  map.new_tid = new_tid;
1089 
1090  /* ---
1091  * Now persist the mapping for the individual xids that are affected. We
1092  * need to log for both xmin and xmax if they aren't the same transaction
1093  * since the mapping files are per "affected" xid.
1094  * We don't muster all that much effort detecting whether xmin and xmax
1095  * are actually the same transaction, we just check whether the xid is the
1096  * same disregarding subtransactions. Logging too much is relatively
1097  * harmless and we could never do the check fully since subtransaction
1098  * data is thrown away during restarts.
1099  * ---
1100  */
1101  if (do_log_xmin)
1102  logical_rewrite_log_mapping(state, xmin, &map);
1103  /* separately log mapping for xmax unless it'd be redundant */
1104  if (do_log_xmax && !TransactionIdEquals(xmin, xmax))
1105  logical_rewrite_log_mapping(state, xmax, &map);
1106 }
1107 
1108 /*
1109  * Replay XLOG_HEAP2_REWRITE records
1110  */
1111 void
1113 {
1114  char path[MAXPGPATH];
1115  int fd;
1116  xl_heap_rewrite_mapping *xlrec;
1117  uint32 len;
1118  char *data;
1119 
1120  xlrec = (xl_heap_rewrite_mapping *) XLogRecGetData(r);
1121 
1122  snprintf(path, MAXPGPATH,
1123  "pg_logical/mappings/" LOGICAL_REWRITE_FORMAT,
1124  xlrec->mapped_db, xlrec->mapped_rel,
1125  (uint32) (xlrec->start_lsn >> 32),
1126  (uint32) xlrec->start_lsn,
1127  xlrec->mapped_xid, XLogRecGetXid(r));
1128 
1129  fd = OpenTransientFile(path,
1130  O_CREAT | O_WRONLY | PG_BINARY);
1131  if (fd < 0)
1132  ereport(ERROR,
1134  errmsg("could not create file \"%s\": %m", path)));
1135 
1136  /*
1137  * Truncate all data that's not guaranteed to have been safely fsynced (by
1138  * previous record or by the last checkpoint).
1139  */
1141  if (ftruncate(fd, xlrec->offset) != 0)
1142  ereport(ERROR,
1144  errmsg("could not truncate file \"%s\" to %u: %m",
1145  path, (uint32) xlrec->offset)));
1147 
1148  data = XLogRecGetData(r) + sizeof(*xlrec);
1149 
1150  len = xlrec->num_mappings * sizeof(LogicalRewriteMappingData);
1151 
1152  /* write out tail end of mapping file (again) */
1153  errno = 0;
1155  if (pg_pwrite(fd, data, len, xlrec->offset) != len)
1156  {
1157  /* if write didn't set errno, assume problem is no disk space */
1158  if (errno == 0)
1159  errno = ENOSPC;
1160  ereport(ERROR,
1162  errmsg("could not write to file \"%s\": %m", path)));
1163  }
1165 
1166  /*
1167  * Now fsync all previously written data. We could improve things and only
1168  * do this for the last write to a file, but the required bookkeeping
1169  * doesn't seem worth the trouble.
1170  */
1172  if (pg_fsync(fd) != 0)
1175  errmsg("could not fsync file \"%s\": %m", path)));
1177 
1178  if (CloseTransientFile(fd) != 0)
1179  ereport(ERROR,
1181  errmsg("could not close file \"%s\": %m", path)));
1182 }
1183 
1184 /* ---
1185  * Perform a checkpoint for logical rewrite mappings
1186  *
1187  * This serves two tasks:
1188  * 1) Remove all mappings not needed anymore based on the logical restart LSN
1189  * 2) Flush all remaining mappings to disk, so that replay after a checkpoint
1190  * only has to deal with the parts of a mapping that have been written out
1191  * after the checkpoint started.
1192  * ---
1193  */
1194 void
1196 {
1197  XLogRecPtr cutoff;
1198  XLogRecPtr redo;
1199  DIR *mappings_dir;
1200  struct dirent *mapping_de;
1201  char path[MAXPGPATH + 20];
1202 
1203  /*
1204  * We start of with a minimum of the last redo pointer. No new decoding
1205  * slot will start before that, so that's a safe upper bound for removal.
1206  */
1207  redo = GetRedoRecPtr();
1208 
1209  /* now check for the restart ptrs from existing slots */
1211 
1212  /* don't start earlier than the restart lsn */
1213  if (cutoff != InvalidXLogRecPtr && redo < cutoff)
1214  cutoff = redo;
1215 
1216  mappings_dir = AllocateDir("pg_logical/mappings");
1217  while ((mapping_de = ReadDir(mappings_dir, "pg_logical/mappings")) != NULL)
1218  {
1219  struct stat statbuf;
1220  Oid dboid;
1221  Oid relid;
1222  XLogRecPtr lsn;
1223  TransactionId rewrite_xid;
1224  TransactionId create_xid;
1225  uint32 hi,
1226  lo;
1227 
1228  if (strcmp(mapping_de->d_name, ".") == 0 ||
1229  strcmp(mapping_de->d_name, "..") == 0)
1230  continue;
1231 
1232  snprintf(path, sizeof(path), "pg_logical/mappings/%s", mapping_de->d_name);
1233  if (lstat(path, &statbuf) == 0 && !S_ISREG(statbuf.st_mode))
1234  continue;
1235 
1236  /* Skip over files that cannot be ours. */
1237  if (strncmp(mapping_de->d_name, "map-", 4) != 0)
1238  continue;
1239 
1240  if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
1241  &dboid, &relid, &hi, &lo, &rewrite_xid, &create_xid) != 6)
1242  elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
1243 
1244  lsn = ((uint64) hi) << 32 | lo;
1245 
1246  if (lsn < cutoff || cutoff == InvalidXLogRecPtr)
1247  {
1248  elog(DEBUG1, "removing logical rewrite file \"%s\"", path);
1249  if (unlink(path) < 0)
1250  ereport(ERROR,
1252  errmsg("could not remove file \"%s\": %m", path)));
1253  }
1254  else
1255  {
1256  /* on some operating systems fsyncing a file requires O_RDWR */
1257  int fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
1258 
1259  /*
1260  * The file cannot vanish due to concurrency since this function
1261  * is the only one removing logical mappings and it's run while
1262  * CheckpointLock is held exclusively.
1263  */
1264  if (fd < 0)
1265  ereport(ERROR,
1267  errmsg("could not open file \"%s\": %m", path)));
1268 
1269  /*
1270  * We could try to avoid fsyncing files that either haven't
1271  * changed or have only been created since the checkpoint's start,
1272  * but it's currently not deemed worth the effort.
1273  */
1275  if (pg_fsync(fd) != 0)
1278  errmsg("could not fsync file \"%s\": %m", path)));
1280 
1281  if (CloseTransientFile(fd) != 0)
1282  ereport(ERROR,
1284  errmsg("could not close file \"%s\": %m", path)));
1285  }
1286  }
1287  FreeDir(mappings_dir);
1288 }
#define HeapTupleHeaderGetUpdateXid(tup)
Definition: htup_details.h:365
#define ItemPointerIsValid(pointer)
Definition: itemptr.h:82
HeapTuple heap_copytuple(HeapTuple tuple)
Definition: heaptuple.c:680
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
void MemoryContextDelete(MemoryContext context)
Definition: mcxt.c:212
#define AllocSetContextCreate
Definition: memutils.h:170
void end_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:301
#define DEBUG1
Definition: elog.h:25
dlist_node * cur
Definition: ilist.h:180
TidHashKey key
Definition: rewriteheap.c:174
void heap_xlog_logical_rewrite(XLogReaderState *r)
Definition: rewriteheap.c:1112
HeapTupleFields t_heap
Definition: htup_details.h:156
TransactionId rs_logical_xmin
Definition: rewriteheap.c:144
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1437
#define TransactionIdEquals(id1, id2)
Definition: transam.h:43
#define HASH_CONTEXT
Definition: hsearch.h:91
#define HASH_ELEM
Definition: hsearch.h:85
#define dlist_foreach_modify(iter, lhead)
Definition: ilist.h:524
uint32 TransactionId
Definition: c.h:520
MemoryContext hcxt
Definition: hsearch.h:77
static void logical_rewrite_heap_tuple(RewriteState state, ItemPointerData old_tid, HeapTuple new_tuple)
Definition: rewriteheap.c:1038
static void logical_end_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:941
HeapTupleHeaderData * HeapTupleHeader
Definition: htup.h:23
struct SMgrRelationData * rd_smgr
Definition: rel.h:57
TransactionId rs_freeze_xid
Definition: rewriteheap.c:142
#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:146
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
Pointer Item
Definition: item.h:17
Size entrysize
Definition: hsearch.h:72
Relation rs_new_rel
Definition: rewriteheap.c:135
int errcode(int sqlerrcode)
Definition: elog.c:610
#define HEAP2_XACT_MASK
Definition: htup_details.h:283
#define PageAddItem(page, item, size, offsetNumber, overwrite, is_heap)
Definition: bufpage.h:410
UnresolvedTupData * UnresolvedTup
Definition: rewriteheap.c:179
#define HeapTupleHeaderIndicatesMovedPartitions(tup)
Definition: htup_details.h:445
void CheckPointLogicalRewriteHeap(void)
Definition: rewriteheap.c:1195
uint32 BlockNumber
Definition: block.h:31
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:919
#define HEAP_UPDATED
Definition: htup_details.h:209
ItemPointerData old_tid
Definition: rewriteheap.c:175
Form_pg_class rd_rel
Definition: rel.h:109
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:3730
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1211
ssize_t pg_pwrite(int fd, const void *buf, size_t nbyte, off_t offset)
Definition: pwrite.c:27
uint16 OffsetNumber
Definition: off.h:24
HeapTupleHeader t_data
Definition: htup.h:68
#define RelationOpenSmgr(relation)
Definition: rel.h:513
Definition: dynahash.c:218
#define dlist_container(type, membername, ptr)
Definition: ilist.h:477
void pfree(void *pointer)
Definition: mcxt.c:1057
uint32 rs_num_rewrite_mappings
Definition: rewriteheap.c:154
#define XLogRecGetData(decoder)
Definition: xlogreader.h:310
Definition: dirent.c:25
#define ERROR
Definition: elog.h:43
ItemPointerData new_tid
Definition: rewriteheap.c:184
Size PageGetHeapFreeSpace(Page page)
Definition: bufpage.c:866
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2372
#define HEAP_XMAX_INVALID
Definition: htup_details.h:207
char path[MAXPGPATH]
Definition: rewriteheap.c:200
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:11507
ItemPointerData t_ctid
Definition: htup_details.h:160
HTAB * rs_unresolved_tups
Definition: rewriteheap.c:151
#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:2127
TransactionId xid
Definition: rewriteheap.c:195
TransactionId GetCurrentTransactionId(void)
Definition: xact.c:438
uint32 t_len
Definition: htup.h:64
static void logical_rewrite_log_mapping(RewriteState state, TransactionId xid, LogicalRewriteMappingData *map)
Definition: rewriteheap.c:971
#define MaxHeapTupleSize
Definition: htup_details.h:560
int errcode_for_file_access(void)
Definition: elog.c:633
XLogRecPtr ReplicationSlotsComputeLogicalRestartLSN(void)
Definition: slot.c:848
#define InvalidTransactionId
Definition: transam.h:31
unsigned int uint32
Definition: c.h:374
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2583
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1386
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:6376
static void dlist_delete(dlist_node *node)
Definition: ilist.h:358
XLogRecPtr rs_begin_lsn
Definition: rewriteheap.c:150
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
#define RelationGetTargetPageFreeSpace(relation, defaultff)
Definition: rel.h:340
#define S_ISREG(m)
Definition: win32_port.h:299
HTAB * rs_logical_mappings
Definition: rewriteheap.c:153
int CloseTransientFile(int fd)
Definition: fd.c:2549
TransactionId rs_oldest_xmin
Definition: rewriteheap.c:140
#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:330
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:422
#define HASH_BLOBS
Definition: hsearch.h:86
#define XLogRecGetXid(decoder)
Definition: xlogreader.h:307
int FileWrite(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2029
MemoryContext rs_cxt
Definition: rewriteheap.c:148
void * palloc0(Size size)
Definition: mcxt.c:981
#define RelationIsAccessibleInLogicalDecoding(relation)
Definition: rel.h:619
int data_sync_elevel(int elevel)
Definition: fd.c:3603
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:326
#define HEAP_XMAX_IS_LOCKED_ONLY(infomask)
Definition: htup_details.h:230
Oid MyDatabaseId
Definition: globals.c:85
Size keysize
Definition: hsearch.h:71
#define RelationGetNumberOfBlocks(reln)
Definition: bufmgr.h:211
ItemPointerData tid
Definition: rewriteheap.c:166
#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
#define ereport(elevel,...)
Definition: elog.h:144
TransactionId mapped_xid
Definition: heapam_xlog.h:377
TransactionId MultiXactId
Definition: c.h:530
RelFileNode rd_node
Definition: rel.h:55
void FileClose(File file)
Definition: fd.c:1826
static void logical_begin_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:794
ItemPointerData new_tid
Definition: rewriteheap.h:40
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:745
HTAB * rs_old_new_tid_map
Definition: rewriteheap.c:152
Definition: regguts.h:298
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2649
#define HeapTupleHeaderGetXmin(tup)
Definition: htup_details.h:313
size_t Size
Definition: c.h:473
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1362
dlist_head mappings
Definition: rewriteheap.c:199
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1414
#define MAXALIGN(LEN)
Definition: c.h:698
XLogRecPtr GetRedoRecPtr(void)
Definition: xlog.c:8366
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1401
union HeapTupleHeaderData::@44 t_choice
#define RelationNeedsWAL(relation)
Definition: rel.h:562
bool ItemPointerEquals(ItemPointer pointer1, ItemPointer pointer2)
Definition: itemptr.c:29
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1391
static void logical_heap_rewrite_flush_mappings(RewriteState state)
Definition: rewriteheap.c:843
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:463
#define lstat(path, sb)
Definition: win32_port.h:244
#define HEAP_INSERT_SKIP_FSM
Definition: heapam.h:34
bool rewrite_heap_dead_tuple(RewriteState state, HeapTuple old_tuple)
Definition: rewriteheap.c:565
#define ItemPointerSetInvalid(pointer)
Definition: itemptr.h:172
#define HeapTupleHasExternal(tuple)
Definition: htup_details.h:673
void * palloc(Size size)
Definition: mcxt.c:950
int errmsg(const char *fmt,...)
Definition: elog.c:824
#define HEAP_INSERT_NO_LOGICAL
Definition: heapam.h:36
TransactionId xmin
Definition: rewriteheap.c:165
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:797
LogicalRewriteMappingData map
Definition: rewriteheap.c:209
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:996
#define elog(elevel,...)
Definition: elog.h:214
static void raw_heap_insert(RewriteState state, HeapTuple tup)
Definition: rewriteheap.c:615
struct LogicalRewriteMappingData LogicalRewriteMappingData
#define HEAP_XACT_MASK
Definition: htup_details.h:218
int pg_fsync(int fd)
Definition: fd.c:346
Relation rs_old_rel
Definition: rewriteheap.c:134
char d_name[MAX_PATH]
Definition: dirent.h:15
#define HEAP_DEFAULT_FILLFACTOR
Definition: rel.h:311
struct RewriteMappingDataEntry RewriteMappingDataEntry
HeapTuple heap_toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup, int options)
Definition: heaptoast.c:96
RewriteState begin_heap_rewrite(Relation old_heap, Relation new_heap, TransactionId oldest_xmin, TransactionId freeze_xid, MultiXactId cutoff_multi)
Definition: rewriteheap.c:237
#define TransactionIdIsNormal(xid)
Definition: transam.h:42
BlockNumber rs_blockno
Definition: rewriteheap.c:137
void XLogBeginInsert(void)
Definition: xloginsert.c:123
#define snprintf
Definition: port.h:193
void smgrimmedsync(SMgrRelation reln, ForkNumber forknum)
Definition: smgr.c:643
void rewrite_heap_tuple(RewriteState state, HeapTuple old_tuple, HeapTuple new_tuple)
Definition: rewriteheap.c:363
#define RelationGetRelid(relation)
Definition: rel.h:456
OldToNewMappingData * OldToNewMapping
Definition: rewriteheap.c:187
int FreeDir(DIR *dir)
Definition: fd.c:2701
#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:59