PostgreSQL Source Code  git master
rewriteheap.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * rewriteheap.c
4  * Support functions to rewrite tables.
5  *
6  * These functions provide a facility to completely rewrite a heap, while
7  * preserving visibility information and update chains.
8  *
9  * INTERFACE
10  *
11  * The caller is responsible for creating the new heap, all catalog
12  * changes, supplying the tuples to be written to the new heap, and
13  * rebuilding indexes. The caller must hold AccessExclusiveLock on the
14  * target table, because we assume no one else is writing into it.
15  *
16  * To use the facility:
17  *
18  * begin_heap_rewrite
19  * while (fetch next tuple)
20  * {
21  * if (tuple is dead)
22  * rewrite_heap_dead_tuple
23  * else
24  * {
25  * // do any transformations here if required
26  * rewrite_heap_tuple
27  * }
28  * }
29  * end_heap_rewrite
30  *
31  * The contents of the new relation shouldn't be relied on until after
32  * end_heap_rewrite is called.
33  *
34  *
35  * IMPLEMENTATION
36  *
37  * This would be a fairly trivial affair, except that we need to maintain
38  * the ctid chains that link versions of an updated tuple together.
39  * Since the newly stored tuples will have tids different from the original
40  * ones, if we just copied t_ctid fields to the new table the links would
41  * be wrong. When we are required to copy a (presumably recently-dead or
42  * delete-in-progress) tuple whose ctid doesn't point to itself, we have
43  * to substitute the correct ctid instead.
44  *
45  * For each ctid reference from A -> B, we might encounter either A first
46  * or B first. (Note that a tuple in the middle of a chain is both A and B
47  * of different pairs.)
48  *
49  * If we encounter A first, we'll store the tuple in the unresolved_tups
50  * hash table. When we later encounter B, we remove A from the hash table,
51  * fix the ctid to point to the new location of B, and insert both A and B
52  * to the new heap.
53  *
54  * If we encounter B first, we can insert B to the new heap right away.
55  * We then add an entry to the old_new_tid_map hash table showing B's
56  * original tid (in the old heap) and new tid (in the new heap).
57  * When we later encounter A, we get the new location of B from the table,
58  * and can write A immediately with the correct ctid.
59  *
60  * Entries in the hash tables can be removed as soon as the later tuple
61  * is encountered. That helps to keep the memory usage down. At the end,
62  * both tables are usually empty; we should have encountered both A and B
63  * of each pair. However, it's possible for A to be RECENTLY_DEAD and B
64  * entirely DEAD according to HeapTupleSatisfiesVacuum, because the test
65  * for deadness using OldestXmin is not exact. In such a case we might
66  * encounter B first, and skip it, and find A later. Then A would be added
67  * to unresolved_tups, and stay there until end of the rewrite. Since
68  * this case is very unusual, we don't worry about the memory usage.
69  *
70  * Using in-memory hash tables means that we use some memory for each live
71  * update chain in the table, from the time we find one end of the
72  * reference until we find the other end. That shouldn't be a problem in
73  * practice, but if you do something like an UPDATE without a where-clause
74  * on a large table, and then run CLUSTER in the same transaction, you
75  * could run out of memory. It doesn't seem worthwhile to add support for
76  * spill-to-disk, as there shouldn't be that many RECENTLY_DEAD tuples in a
77  * table under normal circumstances. Furthermore, in the typical scenario
78  * of CLUSTERing on an unchanging key column, we'll see all the versions
79  * of a given tuple together anyway, and so the peak memory usage is only
80  * proportional to the number of RECENTLY_DEAD versions of a single row, not
81  * in the whole table. Note that if we do fail halfway through a CLUSTER,
82  * the old table is still valid, so failure is not catastrophic.
83  *
84  * We can't use the normal heap_insert function to insert into the new
85  * heap, because heap_insert overwrites the visibility information.
86  * We use a special-purpose raw_heap_insert function instead, which
87  * is optimized for bulk inserting a lot of tuples, knowing that we have
88  * exclusive access to the heap. raw_heap_insert builds new pages in
89  * local storage. When a page is full, or at the end of the process,
90  * we insert it to WAL as a single record and then write it to disk
91  * directly through smgr. Note, however, that any data sent to the new
92  * heap's TOAST table will go through the normal bufmgr.
93  *
94  *
95  * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
96  * Portions Copyright (c) 1994-5, Regents of the University of California
97  *
98  * IDENTIFICATION
99  * src/backend/access/heap/rewriteheap.c
100  *
101  *-------------------------------------------------------------------------
102  */
103 #include "postgres.h"
104 
105 #include <sys/stat.h>
106 #include <unistd.h>
107 
108 #include "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 
134 #include "storage/procarray.h"
135 
136 /*
137  * State associated with a rewrite operation. This is opaque to the user
138  * of the rewrite facility.
139  */
140 typedef struct RewriteStateData
141 {
142  Relation rs_old_rel; /* source heap */
143  Relation rs_new_rel; /* destination heap */
144  Page rs_buffer; /* page currently being built */
145  BlockNumber rs_blockno; /* block where page will go */
146  bool rs_buffer_valid; /* T if any tuples in buffer */
147  bool rs_use_wal; /* must we WAL-log inserts? */
148  bool rs_logical_rewrite; /* do we need to do logical rewriting */
149  TransactionId rs_oldest_xmin; /* oldest xmin used by caller to determine
150  * tuple visibility */
151  TransactionId rs_freeze_xid; /* Xid that will be used as freeze cutoff
152  * point */
153  TransactionId rs_logical_xmin; /* Xid that will be used as cutoff point
154  * for logical rewrites */
155  MultiXactId rs_cutoff_multi; /* MultiXactId that will be used as cutoff
156  * point for multixacts */
157  MemoryContext rs_cxt; /* for hash tables and entries and tuples in
158  * them */
159  XLogRecPtr rs_begin_lsn; /* XLogInsertLsn when starting the rewrite */
160  HTAB *rs_unresolved_tups; /* unmatched A tuples */
161  HTAB *rs_old_new_tid_map; /* unmatched B tuples */
162  HTAB *rs_logical_mappings; /* logical remapping files */
163  uint32 rs_num_rewrite_mappings; /* # in memory mappings */
165 
166 /*
167  * The lookup keys for the hash tables are tuple TID and xmin (we must check
168  * both to avoid false matches from dead tuples). Beware that there is
169  * probably some padding space in this struct; it must be zeroed out for
170  * correct hashtable operation.
171  */
172 typedef struct
173 {
174  TransactionId xmin; /* tuple xmin */
175  ItemPointerData tid; /* tuple location in old heap */
176 } TidHashKey;
177 
178 /*
179  * Entry structures for the hash tables
180  */
181 typedef struct
182 {
183  TidHashKey key; /* expected xmin/old location of B tuple */
184  ItemPointerData old_tid; /* A's location in the old heap */
185  HeapTuple tuple; /* A's tuple contents */
187 
189 
190 typedef struct
191 {
192  TidHashKey key; /* actual xmin/old location of B tuple */
193  ItemPointerData new_tid; /* where we put it in the new heap */
195 
197 
198 /*
199  * In-Memory data for an xid that might need logical remapping entries
200  * to be logged.
201  */
202 typedef struct RewriteMappingFile
203 {
204  TransactionId xid; /* xid that might need to see the row */
205  int vfd; /* fd of mappings file */
206  off_t off; /* how far have we written yet */
207  uint32 num_mappings; /* number of in-memory mappings */
208  dlist_head mappings; /* list of in-memory mappings */
209  char path[MAXPGPATH]; /* path, for error messages */
211 
212 /*
213  * A single In-Memory logical rewrite mapping, hanging off
214  * RewriteMappingFile->mappings.
215  */
217 {
218  LogicalRewriteMappingData map; /* map between old and new location of the
219  * tuple */
222 
223 
224 /* prototypes for internal functions */
225 static void raw_heap_insert(RewriteState state, HeapTuple tup);
226 
227 /* internal logical remapping prototypes */
231 
232 
233 /*
234  * Begin a rewrite of a table
235  *
236  * old_heap old, locked heap relation tuples will be read from
237  * new_heap new, locked heap relation to insert tuples to
238  * oldest_xmin xid used by the caller to determine which tuples are dead
239  * freeze_xid xid before which tuples will be frozen
240  * cutoff_multi multixact before which multis will be removed
241  * use_wal should the inserts to the new heap be WAL-logged?
242  *
243  * Returns an opaque RewriteState, allocated in current memory context,
244  * to be used in subsequent calls to the other functions.
245  */
247 begin_heap_rewrite(Relation old_heap, Relation new_heap, TransactionId oldest_xmin,
248  TransactionId freeze_xid, MultiXactId cutoff_multi,
249  bool use_wal)
250 {
252  MemoryContext rw_cxt;
253  MemoryContext old_cxt;
254  HASHCTL hash_ctl;
255 
256  /*
257  * To ease cleanup, make a separate context that will contain the
258  * RewriteState struct itself plus all subsidiary data.
259  */
261  "Table rewrite",
263  old_cxt = MemoryContextSwitchTo(rw_cxt);
264 
265  /* Create and fill in the state struct */
266  state = palloc0(sizeof(RewriteStateData));
267 
268  state->rs_old_rel = old_heap;
269  state->rs_new_rel = new_heap;
270  state->rs_buffer = (Page) palloc(BLCKSZ);
271  /* new_heap needn't be empty, just locked */
272  state->rs_blockno = RelationGetNumberOfBlocks(new_heap);
273  state->rs_buffer_valid = false;
274  state->rs_use_wal = use_wal;
275  state->rs_oldest_xmin = oldest_xmin;
276  state->rs_freeze_xid = freeze_xid;
277  state->rs_cutoff_multi = cutoff_multi;
278  state->rs_cxt = rw_cxt;
279 
280  /* Initialize hash tables used to track update chains */
281  memset(&hash_ctl, 0, sizeof(hash_ctl));
282  hash_ctl.keysize = sizeof(TidHashKey);
283  hash_ctl.entrysize = sizeof(UnresolvedTupData);
284  hash_ctl.hcxt = state->rs_cxt;
285 
286  state->rs_unresolved_tups =
287  hash_create("Rewrite / Unresolved ctids",
288  128, /* arbitrary initial size */
289  &hash_ctl,
291 
292  hash_ctl.entrysize = sizeof(OldToNewMappingData);
293 
294  state->rs_old_new_tid_map =
295  hash_create("Rewrite / Old to new tid map",
296  128, /* arbitrary initial size */
297  &hash_ctl,
299 
300  MemoryContextSwitchTo(old_cxt);
301 
303 
304  return state;
305 }
306 
307 /*
308  * End a rewrite.
309  *
310  * state and any other resources are freed.
311  */
312 void
314 {
315  HASH_SEQ_STATUS seq_status;
316  UnresolvedTup unresolved;
317 
318  /*
319  * Write any remaining tuples in the UnresolvedTups table. If we have any
320  * left, they should in fact be dead, but let's err on the safe side.
321  */
322  hash_seq_init(&seq_status, state->rs_unresolved_tups);
323 
324  while ((unresolved = hash_seq_search(&seq_status)) != NULL)
325  {
326  ItemPointerSetInvalid(&unresolved->tuple->t_data->t_ctid);
327  raw_heap_insert(state, unresolved->tuple);
328  }
329 
330  /* Write the last page, if any */
331  if (state->rs_buffer_valid)
332  {
333  if (state->rs_use_wal)
334  log_newpage(&state->rs_new_rel->rd_node,
335  MAIN_FORKNUM,
336  state->rs_blockno,
337  state->rs_buffer,
338  true);
340 
342 
344  (char *) state->rs_buffer, true);
345  }
346 
347  /*
348  * If the rel is WAL-logged, must fsync before commit. We use heap_sync
349  * to ensure that the toast table gets fsync'd too.
350  *
351  * It's obvious that we must do this when not WAL-logging. It's less
352  * obvious that we have to do it even if we did WAL-log the pages. The
353  * reason is the same as in storage.c's RelationCopyStorage(): we're
354  * writing data that's not in shared buffers, and so a CHECKPOINT
355  * occurring during the rewriteheap operation won't have fsync'd data we
356  * wrote before the checkpoint.
357  */
358  if (RelationNeedsWAL(state->rs_new_rel))
359  heap_sync(state->rs_new_rel);
360 
362 
363  /* Deleting the context frees everything */
364  MemoryContextDelete(state->rs_cxt);
365 }
366 
367 /*
368  * Add a tuple to the new heap.
369  *
370  * Visibility information is copied from the original tuple, except that
371  * we "freeze" very-old tuples. Note that since we scribble on new_tuple,
372  * it had better be temp storage not a pointer to the original tuple.
373  *
374  * state opaque state as returned by begin_heap_rewrite
375  * old_tuple original tuple in the old heap
376  * new_tuple new, rewritten tuple to be inserted to new heap
377  */
378 void
380  HeapTuple old_tuple, HeapTuple new_tuple)
381 {
382  MemoryContext old_cxt;
383  ItemPointerData old_tid;
384  TidHashKey hashkey;
385  bool found;
386  bool free_new;
387 
388  old_cxt = MemoryContextSwitchTo(state->rs_cxt);
389 
390  /*
391  * Copy the original tuple's visibility information into new_tuple.
392  *
393  * XXX we might later need to copy some t_infomask2 bits, too? Right now,
394  * we intentionally clear the HOT status bits.
395  */
396  memcpy(&new_tuple->t_data->t_choice.t_heap,
397  &old_tuple->t_data->t_choice.t_heap,
398  sizeof(HeapTupleFields));
399 
400  new_tuple->t_data->t_infomask &= ~HEAP_XACT_MASK;
401  new_tuple->t_data->t_infomask2 &= ~HEAP2_XACT_MASK;
402  new_tuple->t_data->t_infomask |=
403  old_tuple->t_data->t_infomask & HEAP_XACT_MASK;
404 
405  /*
406  * While we have our hands on the tuple, we may as well freeze any
407  * eligible xmin or xmax, so that future VACUUM effort can be saved.
408  */
409  heap_freeze_tuple(new_tuple->t_data,
410  state->rs_old_rel->rd_rel->relfrozenxid,
411  state->rs_old_rel->rd_rel->relminmxid,
412  state->rs_freeze_xid,
413  state->rs_cutoff_multi);
414 
415  /*
416  * Invalid ctid means that ctid should point to the tuple itself. We'll
417  * override it later if the tuple is part of an update chain.
418  */
419  ItemPointerSetInvalid(&new_tuple->t_data->t_ctid);
420 
421  /*
422  * If the tuple has been updated, check the old-to-new mapping hash table.
423  */
424  if (!((old_tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
425  HeapTupleHeaderIsOnlyLocked(old_tuple->t_data)) &&
427  !(ItemPointerEquals(&(old_tuple->t_self),
428  &(old_tuple->t_data->t_ctid))))
429  {
430  OldToNewMapping mapping;
431 
432  memset(&hashkey, 0, sizeof(hashkey));
433  hashkey.xmin = HeapTupleHeaderGetUpdateXid(old_tuple->t_data);
434  hashkey.tid = old_tuple->t_data->t_ctid;
435 
436  mapping = (OldToNewMapping)
437  hash_search(state->rs_old_new_tid_map, &hashkey,
438  HASH_FIND, NULL);
439 
440  if (mapping != NULL)
441  {
442  /*
443  * We've already copied the tuple that t_ctid points to, so we can
444  * set the ctid of this tuple to point to the new location, and
445  * insert it right away.
446  */
447  new_tuple->t_data->t_ctid = mapping->new_tid;
448 
449  /* We don't need the mapping entry anymore */
450  hash_search(state->rs_old_new_tid_map, &hashkey,
451  HASH_REMOVE, &found);
452  Assert(found);
453  }
454  else
455  {
456  /*
457  * We haven't seen the tuple t_ctid points to yet. Stash this
458  * tuple into unresolved_tups to be written later.
459  */
460  UnresolvedTup unresolved;
461 
462  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
463  HASH_ENTER, &found);
464  Assert(!found);
465 
466  unresolved->old_tid = old_tuple->t_self;
467  unresolved->tuple = heap_copytuple(new_tuple);
468 
469  /*
470  * We can't do anything more now, since we don't know where the
471  * tuple will be written.
472  */
473  MemoryContextSwitchTo(old_cxt);
474  return;
475  }
476  }
477 
478  /*
479  * Now we will write the tuple, and then check to see if it is the B tuple
480  * in any new or known pair. When we resolve a known pair, we will be
481  * able to write that pair's A tuple, and then we have to check if it
482  * resolves some other pair. Hence, we need a loop here.
483  */
484  old_tid = old_tuple->t_self;
485  free_new = false;
486 
487  for (;;)
488  {
489  ItemPointerData new_tid;
490 
491  /* Insert the tuple and find out where it's put in new_heap */
492  raw_heap_insert(state, new_tuple);
493  new_tid = new_tuple->t_self;
494 
495  logical_rewrite_heap_tuple(state, old_tid, new_tuple);
496 
497  /*
498  * If the tuple is the updated version of a row, and the prior version
499  * wouldn't be DEAD yet, then we need to either resolve the prior
500  * version (if it's waiting in rs_unresolved_tups), or make an entry
501  * in rs_old_new_tid_map (so we can resolve it when we do see it). The
502  * previous tuple's xmax would equal this one's xmin, so it's
503  * RECENTLY_DEAD if and only if the xmin is not before OldestXmin.
504  */
505  if ((new_tuple->t_data->t_infomask & HEAP_UPDATED) &&
507  state->rs_oldest_xmin))
508  {
509  /*
510  * Okay, this is B in an update pair. See if we've seen A.
511  */
512  UnresolvedTup unresolved;
513 
514  memset(&hashkey, 0, sizeof(hashkey));
515  hashkey.xmin = HeapTupleHeaderGetXmin(new_tuple->t_data);
516  hashkey.tid = old_tid;
517 
518  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
519  HASH_FIND, NULL);
520 
521  if (unresolved != NULL)
522  {
523  /*
524  * We have seen and memorized the previous tuple already. Now
525  * that we know where we inserted the tuple its t_ctid points
526  * to, fix its t_ctid and insert it to the new heap.
527  */
528  if (free_new)
529  heap_freetuple(new_tuple);
530  new_tuple = unresolved->tuple;
531  free_new = true;
532  old_tid = unresolved->old_tid;
533  new_tuple->t_data->t_ctid = new_tid;
534 
535  /*
536  * We don't need the hash entry anymore, but don't free its
537  * tuple just yet.
538  */
539  hash_search(state->rs_unresolved_tups, &hashkey,
540  HASH_REMOVE, &found);
541  Assert(found);
542 
543  /* loop back to insert the previous tuple in the chain */
544  continue;
545  }
546  else
547  {
548  /*
549  * Remember the new tid of this tuple. We'll use it to set the
550  * ctid when we find the previous tuple in the chain.
551  */
552  OldToNewMapping mapping;
553 
554  mapping = hash_search(state->rs_old_new_tid_map, &hashkey,
555  HASH_ENTER, &found);
556  Assert(!found);
557 
558  mapping->new_tid = new_tid;
559  }
560  }
561 
562  /* Done with this (chain of) tuples, for now */
563  if (free_new)
564  heap_freetuple(new_tuple);
565  break;
566  }
567 
568  MemoryContextSwitchTo(old_cxt);
569 }
570 
571 /*
572  * Register a dead tuple with an ongoing rewrite. Dead tuples are not
573  * copied to the new table, but we still make note of them so that we
574  * can release some resources earlier.
575  *
576  * Returns true if a tuple was removed from the unresolved_tups table.
577  * This indicates that that tuple, previously thought to be "recently dead",
578  * is now known really dead and won't be written to the output.
579  */
580 bool
582 {
583  /*
584  * If we have already seen an earlier tuple in the update chain that
585  * points to this tuple, let's forget about that earlier tuple. It's in
586  * fact dead as well, our simple xmax < OldestXmin test in
587  * HeapTupleSatisfiesVacuum just wasn't enough to detect it. It happens
588  * when xmin of a tuple is greater than xmax, which sounds
589  * counter-intuitive but is perfectly valid.
590  *
591  * We don't bother to try to detect the situation the other way round,
592  * when we encounter the dead tuple first and then the recently dead one
593  * that points to it. If that happens, we'll have some unmatched entries
594  * in the UnresolvedTups hash table at the end. That can happen anyway,
595  * because a vacuum might have removed the dead tuple in the chain before
596  * us.
597  */
598  UnresolvedTup unresolved;
599  TidHashKey hashkey;
600  bool found;
601 
602  memset(&hashkey, 0, sizeof(hashkey));
603  hashkey.xmin = HeapTupleHeaderGetXmin(old_tuple->t_data);
604  hashkey.tid = old_tuple->t_self;
605 
606  unresolved = hash_search(state->rs_unresolved_tups, &hashkey,
607  HASH_FIND, NULL);
608 
609  if (unresolved != NULL)
610  {
611  /* Need to free the contained tuple as well as the hashtable entry */
612  heap_freetuple(unresolved->tuple);
613  hash_search(state->rs_unresolved_tups, &hashkey,
614  HASH_REMOVE, &found);
615  Assert(found);
616  return true;
617  }
618 
619  return false;
620 }
621 
622 /*
623  * Insert a tuple to the new relation. This has to track heap_insert
624  * and its subsidiary functions!
625  *
626  * t_self of the tuple is set to the new TID of the tuple. If t_ctid of the
627  * tuple is invalid on entry, it's replaced with the new TID as well (in
628  * the inserted data only, not in the caller's copy).
629  */
630 static void
632 {
633  Page page = state->rs_buffer;
634  Size pageFreeSpace,
635  saveFreeSpace;
636  Size len;
637  OffsetNumber newoff;
638  HeapTuple heaptup;
639 
640  /*
641  * If the new tuple is too big for storage or contains already toasted
642  * out-of-line attributes from some other relation, invoke the toaster.
643  *
644  * Note: below this point, heaptup is the data we actually intend to store
645  * into the relation; tup is the caller's original untoasted data.
646  */
647  if (state->rs_new_rel->rd_rel->relkind == RELKIND_TOASTVALUE)
648  {
649  /* toast table entries should never be recursively toasted */
651  heaptup = tup;
652  }
653  else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
654  {
656 
657  if (!state->rs_use_wal)
658  options |= HEAP_INSERT_SKIP_WAL;
659 
660  /*
661  * While rewriting the heap for VACUUM FULL / CLUSTER, make sure data
662  * for the TOAST table are not logically decoded. The main heap is
663  * WAL-logged as XLOG FPI records, which are not logically decoded.
664  */
665  options |= HEAP_INSERT_NO_LOGICAL;
666 
667  heaptup = toast_insert_or_update(state->rs_new_rel, tup, NULL,
668  options);
669  }
670  else
671  heaptup = tup;
672 
673  len = MAXALIGN(heaptup->t_len); /* be conservative */
674 
675  /*
676  * If we're gonna fail for oversize tuple, do it right away
677  */
678  if (len > MaxHeapTupleSize)
679  ereport(ERROR,
680  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
681  errmsg("row is too big: size %zu, maximum size %zu",
682  len, MaxHeapTupleSize)));
683 
684  /* Compute desired extra freespace due to fillfactor option */
685  saveFreeSpace = RelationGetTargetPageFreeSpace(state->rs_new_rel,
687 
688  /* Now we can check to see if there's enough free space already. */
689  if (state->rs_buffer_valid)
690  {
691  pageFreeSpace = PageGetHeapFreeSpace(page);
692 
693  if (len + saveFreeSpace > pageFreeSpace)
694  {
695  /* Doesn't fit, so write out the existing page */
696 
697  /* XLOG stuff */
698  if (state->rs_use_wal)
699  log_newpage(&state->rs_new_rel->rd_node,
700  MAIN_FORKNUM,
701  state->rs_blockno,
702  page,
703  true);
704 
705  /*
706  * Now write the page. We say isTemp = true even if it's not a
707  * temp table, because there's no need for smgr to schedule an
708  * fsync for this write; we'll do it ourselves in
709  * end_heap_rewrite.
710  */
712 
713  PageSetChecksumInplace(page, state->rs_blockno);
714 
716  state->rs_blockno, (char *) page, true);
717 
718  state->rs_blockno++;
719  state->rs_buffer_valid = false;
720  }
721  }
722 
723  if (!state->rs_buffer_valid)
724  {
725  /* Initialize a new empty page */
726  PageInit(page, BLCKSZ, 0);
727  state->rs_buffer_valid = true;
728  }
729 
730  /* And now we can insert the tuple into the page */
731  newoff = PageAddItem(page, (Item) heaptup->t_data, heaptup->t_len,
732  InvalidOffsetNumber, false, true);
733  if (newoff == InvalidOffsetNumber)
734  elog(ERROR, "failed to add tuple");
735 
736  /* Update caller's t_self to the actual position where it was stored */
737  ItemPointerSet(&(tup->t_self), state->rs_blockno, newoff);
738 
739  /*
740  * Insert the correct position into CTID of the stored tuple, too, if the
741  * caller didn't supply a valid CTID.
742  */
743  if (!ItemPointerIsValid(&tup->t_data->t_ctid))
744  {
745  ItemId newitemid;
746  HeapTupleHeader onpage_tup;
747 
748  newitemid = PageGetItemId(page, newoff);
749  onpage_tup = (HeapTupleHeader) PageGetItem(page, newitemid);
750 
751  onpage_tup->t_ctid = tup->t_self;
752  }
753 
754  /* If heaptup is a private copy, release it. */
755  if (heaptup != tup)
756  heap_freetuple(heaptup);
757 }
758 
759 /* ------------------------------------------------------------------------
760  * Logical rewrite support
761  *
762  * When doing logical decoding - which relies on using cmin/cmax of catalog
763  * tuples, via xl_heap_new_cid records - heap rewrites have to log enough
764  * information to allow the decoding backend to updates its internal mapping
765  * of (relfilenode,ctid) => (cmin, cmax) to be correct for the rewritten heap.
766  *
767  * For that, every time we find a tuple that's been modified in a catalog
768  * relation within the xmin horizon of any decoding slot, we log a mapping
769  * from the old to the new location.
770  *
771  * To deal with rewrites that abort the filename of a mapping file contains
772  * the xid of the transaction performing the rewrite, which then can be
773  * checked before being read in.
774  *
775  * For efficiency we don't immediately spill every single map mapping for a
776  * row to disk but only do so in batches when we've collected several of them
777  * in memory or when end_heap_rewrite() has been called.
778  *
779  * Crash-Safety: This module diverts from the usual patterns of doing WAL
780  * since it cannot rely on checkpoint flushing out all buffers and thus
781  * waiting for exclusive locks on buffers. Usually the XLogInsert() covering
782  * buffer modifications is performed while the buffer(s) that are being
783  * modified are exclusively locked guaranteeing that both the WAL record and
784  * the modified heap are on either side of the checkpoint. But since the
785  * mapping files we log aren't in shared_buffers that interlock doesn't work.
786  *
787  * Instead we simply write the mapping files out to disk, *before* the
788  * XLogInsert() is performed. That guarantees that either the XLogInsert() is
789  * inserted after the checkpoint's redo pointer or that the checkpoint (via
790  * CheckPointLogicalRewriteHeap()) has flushed the (partial) mapping file to
791  * disk. That leaves the tail end that has not yet been flushed open to
792  * corruption, which is solved by including the current offset in the
793  * xl_heap_rewrite_mapping records and truncating the mapping file to it
794  * during replay. Every time a rewrite is finished all generated mapping files
795  * are synced to disk.
796  *
797  * Note that if we were only concerned about crash safety we wouldn't have to
798  * deal with WAL logging at all - an fsync() at the end of a rewrite would be
799  * sufficient for crash safety. Any mapping that hasn't been safely flushed to
800  * disk has to be by an aborted (explicitly or via a crash) transaction and is
801  * ignored by virtue of the xid in its name being subject to a
802  * TransactionDidCommit() check. But we want to support having standbys via
803  * physical replication, both for availability and to do logical decoding
804  * there.
805  * ------------------------------------------------------------------------
806  */
807 
808 /*
809  * Do preparations for logging logical mappings during a rewrite if
810  * necessary. If we detect that we don't need to log anything we'll prevent
811  * any further action by the various logical rewrite functions.
812  */
813 static void
815 {
816  HASHCTL hash_ctl;
817  TransactionId logical_xmin;
818 
819  /*
820  * We only need to persist these mappings if the rewritten table can be
821  * accessed during logical decoding, if not, we can skip doing any
822  * additional work.
823  */
824  state->rs_logical_rewrite =
826 
827  if (!state->rs_logical_rewrite)
828  return;
829 
830  ProcArrayGetReplicationSlotXmin(NULL, &logical_xmin);
831 
832  /*
833  * If there are no logical slots in progress we don't need to do anything,
834  * there cannot be any remappings for relevant rows yet. The relation's
835  * lock protects us against races.
836  */
837  if (logical_xmin == InvalidTransactionId)
838  {
839  state->rs_logical_rewrite = false;
840  return;
841  }
842 
843  state->rs_logical_xmin = logical_xmin;
845  state->rs_num_rewrite_mappings = 0;
846 
847  memset(&hash_ctl, 0, sizeof(hash_ctl));
848  hash_ctl.keysize = sizeof(TransactionId);
849  hash_ctl.entrysize = sizeof(RewriteMappingFile);
850  hash_ctl.hcxt = state->rs_cxt;
851 
852  state->rs_logical_mappings =
853  hash_create("Logical rewrite mapping",
854  128, /* arbitrary initial size */
855  &hash_ctl,
857 }
858 
859 /*
860  * Flush all logical in-memory mappings to disk, but don't fsync them yet.
861  */
862 static void
864 {
865  HASH_SEQ_STATUS seq_status;
866  RewriteMappingFile *src;
867  dlist_mutable_iter iter;
868 
869  Assert(state->rs_logical_rewrite);
870 
871  /* no logical rewrite in progress, no need to iterate over mappings */
872  if (state->rs_num_rewrite_mappings == 0)
873  return;
874 
875  elog(DEBUG1, "flushing %u logical rewrite mapping entries",
876  state->rs_num_rewrite_mappings);
877 
878  hash_seq_init(&seq_status, state->rs_logical_mappings);
879  while ((src = (RewriteMappingFile *) hash_seq_search(&seq_status)) != NULL)
880  {
881  char *waldata;
882  char *waldata_start;
884  Oid dboid;
885  uint32 len;
886  int written;
887 
888  /* this file hasn't got any new mappings */
889  if (src->num_mappings == 0)
890  continue;
891 
892  if (state->rs_old_rel->rd_rel->relisshared)
893  dboid = InvalidOid;
894  else
895  dboid = MyDatabaseId;
896 
897  xlrec.num_mappings = src->num_mappings;
898  xlrec.mapped_rel = RelationGetRelid(state->rs_old_rel);
899  xlrec.mapped_xid = src->xid;
900  xlrec.mapped_db = dboid;
901  xlrec.offset = src->off;
902  xlrec.start_lsn = state->rs_begin_lsn;
903 
904  /* write all mappings consecutively */
905  len = src->num_mappings * sizeof(LogicalRewriteMappingData);
906  waldata_start = waldata = palloc(len);
907 
908  /*
909  * collect data we need to write out, but don't modify ondisk data yet
910  */
911  dlist_foreach_modify(iter, &src->mappings)
912  {
914 
915  pmap = dlist_container(RewriteMappingDataEntry, node, iter.cur);
916 
917  memcpy(waldata, &pmap->map, sizeof(pmap->map));
918  waldata += sizeof(pmap->map);
919 
920  /* remove from the list and free */
921  dlist_delete(&pmap->node);
922  pfree(pmap);
923 
924  /* update bookkeeping */
925  state->rs_num_rewrite_mappings--;
926  src->num_mappings--;
927  }
928 
929  Assert(src->num_mappings == 0);
930  Assert(waldata == waldata_start + len);
931 
932  /*
933  * Note that we deviate from the usual WAL coding practices here,
934  * check the above "Logical rewrite support" comment for reasoning.
935  */
936  written = FileWrite(src->vfd, waldata_start, len, src->off,
938  if (written != len)
939  ereport(ERROR,
941  errmsg("could not write to file \"%s\", wrote %d of %d: %m", src->path,
942  written, len)));
943  src->off += len;
944 
945  XLogBeginInsert();
946  XLogRegisterData((char *) (&xlrec), sizeof(xlrec));
947  XLogRegisterData(waldata_start, len);
948 
949  /* write xlog record */
950  XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_REWRITE);
951 
952  pfree(waldata_start);
953  }
954  Assert(state->rs_num_rewrite_mappings == 0);
955 }
956 
957 /*
958  * Logical remapping part of end_heap_rewrite().
959  */
960 static void
962 {
963  HASH_SEQ_STATUS seq_status;
964  RewriteMappingFile *src;
965 
966  /* done, no logical rewrite in progress */
967  if (!state->rs_logical_rewrite)
968  return;
969 
970  /* writeout remaining in-memory entries */
971  if (state->rs_num_rewrite_mappings > 0)
973 
974  /* Iterate over all mappings we have written and fsync the files. */
975  hash_seq_init(&seq_status, state->rs_logical_mappings);
976  while ((src = (RewriteMappingFile *) hash_seq_search(&seq_status)) != NULL)
977  {
981  errmsg("could not fsync file \"%s\": %m", src->path)));
982  FileClose(src->vfd);
983  }
984  /* memory context cleanup will deal with the rest */
985 }
986 
987 /*
988  * Log a single (old->new) mapping for 'xid'.
989  */
990 static void
993 {
994  RewriteMappingFile *src;
996  Oid relid;
997  bool found;
998 
999  relid = RelationGetRelid(state->rs_old_rel);
1000 
1001  /* look for existing mappings for this 'mapped' xid */
1002  src = hash_search(state->rs_logical_mappings, &xid,
1003  HASH_ENTER, &found);
1004 
1005  /*
1006  * We haven't yet had the need to map anything for this xid, create
1007  * per-xid data structures.
1008  */
1009  if (!found)
1010  {
1011  char path[MAXPGPATH];
1012  Oid dboid;
1013 
1014  if (state->rs_old_rel->rd_rel->relisshared)
1015  dboid = InvalidOid;
1016  else
1017  dboid = MyDatabaseId;
1018 
1019  snprintf(path, MAXPGPATH,
1020  "pg_logical/mappings/" LOGICAL_REWRITE_FORMAT,
1021  dboid, relid,
1022  (uint32) (state->rs_begin_lsn >> 32),
1023  (uint32) state->rs_begin_lsn,
1024  xid, GetCurrentTransactionId());
1025 
1026  dlist_init(&src->mappings);
1027  src->num_mappings = 0;
1028  src->off = 0;
1029  memcpy(src->path, path, sizeof(path));
1030  src->vfd = PathNameOpenFile(path,
1031  O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
1032  if (src->vfd < 0)
1033  ereport(ERROR,
1035  errmsg("could not create file \"%s\": %m", path)));
1036  }
1037 
1038  pmap = MemoryContextAlloc(state->rs_cxt,
1039  sizeof(RewriteMappingDataEntry));
1040  memcpy(&pmap->map, map, sizeof(LogicalRewriteMappingData));
1041  dlist_push_tail(&src->mappings, &pmap->node);
1042  src->num_mappings++;
1043  state->rs_num_rewrite_mappings++;
1044 
1045  /*
1046  * Write out buffer every time we've too many in-memory entries across all
1047  * mapping files.
1048  */
1049  if (state->rs_num_rewrite_mappings >= 1000 /* arbitrary number */ )
1051 }
1052 
1053 /*
1054  * Perform logical remapping for a tuple that's mapped from old_tid to
1055  * new_tuple->t_self by rewrite_heap_tuple() if necessary for the tuple.
1056  */
1057 static void
1059  HeapTuple new_tuple)
1060 {
1061  ItemPointerData new_tid = new_tuple->t_self;
1062  TransactionId cutoff = state->rs_logical_xmin;
1063  TransactionId xmin;
1064  TransactionId xmax;
1065  bool do_log_xmin = false;
1066  bool do_log_xmax = false;
1068 
1069  /* no logical rewrite in progress, we don't need to log anything */
1070  if (!state->rs_logical_rewrite)
1071  return;
1072 
1073  xmin = HeapTupleHeaderGetXmin(new_tuple->t_data);
1074  /* use *GetUpdateXid to correctly deal with multixacts */
1075  xmax = HeapTupleHeaderGetUpdateXid(new_tuple->t_data);
1076 
1077  /*
1078  * Log the mapping iff the tuple has been created recently.
1079  */
1080  if (TransactionIdIsNormal(xmin) && !TransactionIdPrecedes(xmin, cutoff))
1081  do_log_xmin = true;
1082 
1083  if (!TransactionIdIsNormal(xmax))
1084  {
1085  /*
1086  * no xmax is set, can't have any permanent ones, so this check is
1087  * sufficient
1088  */
1089  }
1090  else if (HEAP_XMAX_IS_LOCKED_ONLY(new_tuple->t_data->t_infomask))
1091  {
1092  /* only locked, we don't care */
1093  }
1094  else if (!TransactionIdPrecedes(xmax, cutoff))
1095  {
1096  /* tuple has been deleted recently, log */
1097  do_log_xmax = true;
1098  }
1099 
1100  /* if neither needs to be logged, we're done */
1101  if (!do_log_xmin && !do_log_xmax)
1102  return;
1103 
1104  /* fill out mapping information */
1105  map.old_node = state->rs_old_rel->rd_node;
1106  map.old_tid = old_tid;
1107  map.new_node = state->rs_new_rel->rd_node;
1108  map.new_tid = new_tid;
1109 
1110  /* ---
1111  * Now persist the mapping for the individual xids that are affected. We
1112  * need to log for both xmin and xmax if they aren't the same transaction
1113  * since the mapping files are per "affected" xid.
1114  * We don't muster all that much effort detecting whether xmin and xmax
1115  * are actually the same transaction, we just check whether the xid is the
1116  * same disregarding subtransactions. Logging too much is relatively
1117  * harmless and we could never do the check fully since subtransaction
1118  * data is thrown away during restarts.
1119  * ---
1120  */
1121  if (do_log_xmin)
1122  logical_rewrite_log_mapping(state, xmin, &map);
1123  /* separately log mapping for xmax unless it'd be redundant */
1124  if (do_log_xmax && !TransactionIdEquals(xmin, xmax))
1125  logical_rewrite_log_mapping(state, xmax, &map);
1126 }
1127 
1128 /*
1129  * Replay XLOG_HEAP2_REWRITE records
1130  */
1131 void
1133 {
1134  char path[MAXPGPATH];
1135  int fd;
1136  xl_heap_rewrite_mapping *xlrec;
1137  uint32 len;
1138  char *data;
1139 
1140  xlrec = (xl_heap_rewrite_mapping *) XLogRecGetData(r);
1141 
1142  snprintf(path, MAXPGPATH,
1143  "pg_logical/mappings/" LOGICAL_REWRITE_FORMAT,
1144  xlrec->mapped_db, xlrec->mapped_rel,
1145  (uint32) (xlrec->start_lsn >> 32),
1146  (uint32) xlrec->start_lsn,
1147  xlrec->mapped_xid, XLogRecGetXid(r));
1148 
1149  fd = OpenTransientFile(path,
1150  O_CREAT | O_WRONLY | PG_BINARY);
1151  if (fd < 0)
1152  ereport(ERROR,
1154  errmsg("could not create file \"%s\": %m", path)));
1155 
1156  /*
1157  * Truncate all data that's not guaranteed to have been safely fsynced (by
1158  * previous record or by the last checkpoint).
1159  */
1161  if (ftruncate(fd, xlrec->offset) != 0)
1162  ereport(ERROR,
1164  errmsg("could not truncate file \"%s\" to %u: %m",
1165  path, (uint32) xlrec->offset)));
1167 
1168  /* now seek to the position we want to write our data to */
1169  if (lseek(fd, xlrec->offset, SEEK_SET) != xlrec->offset)
1170  ereport(ERROR,
1172  errmsg("could not seek to end of file \"%s\": %m",
1173  path)));
1174 
1175  data = XLogRecGetData(r) + sizeof(*xlrec);
1176 
1177  len = xlrec->num_mappings * sizeof(LogicalRewriteMappingData);
1178 
1179  /* write out tail end of mapping file (again) */
1180  errno = 0;
1182  if (write(fd, data, len) != len)
1183  {
1184  /* if write didn't set errno, assume problem is no disk space */
1185  if (errno == 0)
1186  errno = ENOSPC;
1187  ereport(ERROR,
1189  errmsg("could not write to file \"%s\": %m", path)));
1190  }
1192 
1193  /*
1194  * Now fsync all previously written data. We could improve things and only
1195  * do this for the last write to a file, but the required bookkeeping
1196  * doesn't seem worth the trouble.
1197  */
1199  if (pg_fsync(fd) != 0)
1202  errmsg("could not fsync file \"%s\": %m", path)));
1204 
1205  if (CloseTransientFile(fd) != 0)
1206  ereport(ERROR,
1208  errmsg("could not close file \"%s\": %m", path)));
1209 }
1210 
1211 /* ---
1212  * Perform a checkpoint for logical rewrite mappings
1213  *
1214  * This serves two tasks:
1215  * 1) Remove all mappings not needed anymore based on the logical restart LSN
1216  * 2) Flush all remaining mappings to disk, so that replay after a checkpoint
1217  * only has to deal with the parts of a mapping that have been written out
1218  * after the checkpoint started.
1219  * ---
1220  */
1221 void
1223 {
1224  XLogRecPtr cutoff;
1225  XLogRecPtr redo;
1226  DIR *mappings_dir;
1227  struct dirent *mapping_de;
1228  char path[MAXPGPATH + 20];
1229 
1230  /*
1231  * We start of with a minimum of the last redo pointer. No new decoding
1232  * slot will start before that, so that's a safe upper bound for removal.
1233  */
1234  redo = GetRedoRecPtr();
1235 
1236  /* now check for the restart ptrs from existing slots */
1238 
1239  /* don't start earlier than the restart lsn */
1240  if (cutoff != InvalidXLogRecPtr && redo < cutoff)
1241  cutoff = redo;
1242 
1243  mappings_dir = AllocateDir("pg_logical/mappings");
1244  while ((mapping_de = ReadDir(mappings_dir, "pg_logical/mappings")) != NULL)
1245  {
1246  struct stat statbuf;
1247  Oid dboid;
1248  Oid relid;
1249  XLogRecPtr lsn;
1250  TransactionId rewrite_xid;
1251  TransactionId create_xid;
1252  uint32 hi,
1253  lo;
1254 
1255  if (strcmp(mapping_de->d_name, ".") == 0 ||
1256  strcmp(mapping_de->d_name, "..") == 0)
1257  continue;
1258 
1259  snprintf(path, sizeof(path), "pg_logical/mappings/%s", mapping_de->d_name);
1260  if (lstat(path, &statbuf) == 0 && !S_ISREG(statbuf.st_mode))
1261  continue;
1262 
1263  /* Skip over files that cannot be ours. */
1264  if (strncmp(mapping_de->d_name, "map-", 4) != 0)
1265  continue;
1266 
1267  if (sscanf(mapping_de->d_name, LOGICAL_REWRITE_FORMAT,
1268  &dboid, &relid, &hi, &lo, &rewrite_xid, &create_xid) != 6)
1269  elog(ERROR, "could not parse filename \"%s\"", mapping_de->d_name);
1270 
1271  lsn = ((uint64) hi) << 32 | lo;
1272 
1273  if (lsn < cutoff || cutoff == InvalidXLogRecPtr)
1274  {
1275  elog(DEBUG1, "removing logical rewrite file \"%s\"", path);
1276  if (unlink(path) < 0)
1277  ereport(ERROR,
1279  errmsg("could not remove file \"%s\": %m", path)));
1280  }
1281  else
1282  {
1283  int fd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
1284 
1285  /*
1286  * The file cannot vanish due to concurrency since this function
1287  * is the only one removing logical mappings and it's run while
1288  * CheckpointLock is held exclusively.
1289  */
1290  if (fd < 0)
1291  ereport(ERROR,
1293  errmsg("could not open file \"%s\": %m", path)));
1294 
1295  /*
1296  * We could try to avoid fsyncing files that either haven't
1297  * changed or have only been created since the checkpoint's start,
1298  * but it's currently not deemed worth the effort.
1299  */
1301  if (pg_fsync(fd) != 0)
1304  errmsg("could not fsync file \"%s\": %m", path)));
1306 
1307  if (CloseTransientFile(fd) != 0)
1308  ereport(ERROR,
1310  errmsg("could not close file \"%s\": %m", path)));
1311  }
1312  }
1313  FreeDir(mappings_dir);
1314 }
#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 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
#define AllocSetContextCreate
Definition: memutils.h:169
void end_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:313
#define DEBUG1
Definition: elog.h:25
dlist_node * cur
Definition: ilist.h:180
TidHashKey key
Definition: rewriteheap.c:183
void heap_xlog_logical_rewrite(XLogReaderState *r)
Definition: rewriteheap.c:1132
HeapTupleFields t_heap
Definition: htup_details.h:156
TransactionId rs_logical_xmin
Definition: rewriteheap.c:153
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1323
#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:507
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:247
static void logical_rewrite_heap_tuple(RewriteState state, ItemPointerData old_tid, HeapTuple new_tuple)
Definition: rewriteheap.c:1058
static void logical_end_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:961
HeapTuple toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup, int options)
Definition: tuptoaster.c:538
#define write(a, b, c)
Definition: win32.h:14
HeapTupleHeaderData * HeapTupleHeader
Definition: htup.h:23
struct SMgrRelationData * rd_smgr
Definition: rel.h:56
TransactionId rs_freeze_xid
Definition: rewriteheap.c:151
#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:155
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:143
int errcode(int sqlerrcode)
Definition: elog.c:570
#define HEAP2_XACT_MASK
Definition: htup_details.h:283
#define PageAddItem(page, item, size, offsetNumber, overwrite, is_heap)
Definition: bufpage.h:416
UnresolvedTupData * UnresolvedTup
Definition: rewriteheap.c:188
void heap_sync(Relation rel)
Definition: heapam.c:8934
#define HEAP_INSERT_SKIP_WAL
Definition: heapam.h:32
#define HeapTupleHeaderIndicatesMovedPartitions(tup)
Definition: htup_details.h:445
void CheckPointLogicalRewriteHeap(void)
Definition: rewriteheap.c:1222
uint32 BlockNumber
Definition: block.h:31
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:906
#define HEAP_UPDATED
Definition: htup_details.h:209
ItemPointerData old_tid
Definition: rewriteheap.c:184
Form_pg_class rd_rel
Definition: rel.h:83
void heap_freetuple(HeapTuple htup)
Definition: heaptuple.c:1338
unsigned int Oid
Definition: postgres_ext.h:31
Definition: dirent.h:9
void ProcArrayGetReplicationSlotXmin(TransactionId *xmin, TransactionId *catalog_xmin)
Definition: procarray.c:3003
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1191
uint16 OffsetNumber
Definition: off.h:24
HeapTupleHeader t_data
Definition: htup.h:68
#define RelationOpenSmgr(relation)
Definition: rel.h:473
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:163
#define XLogRecGetData(decoder)
Definition: xlogreader.h:237
Definition: dirent.c:25
#define ERROR
Definition: elog.h:43
ItemPointerData new_tid
Definition: rewriteheap.c:193
Size PageGetHeapFreeSpace(Page page)
Definition: bufpage.c:665
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2257
#define HEAP_XMAX_INVALID
Definition: htup_details.h:207
char path[MAXPGPATH]
Definition: rewriteheap.c:209
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:11149
ItemPointerData t_ctid
Definition: htup_details.h:160
HTAB * rs_unresolved_tups
Definition: rewriteheap.c:160
#define MAXPGPATH
ItemPointerData t_self
Definition: htup.h:65
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:191
int FileSync(File file, uint32 wait_event_info)
Definition: fd.c:2014
TransactionId xid
Definition: rewriteheap.c:204
TransactionId GetCurrentTransactionId(void)
Definition: xact.c:423
uint32 t_len
Definition: htup.h:64
static void logical_rewrite_log_mapping(RewriteState state, TransactionId xid, LogicalRewriteMappingData *map)
Definition: rewriteheap.c:991
#define MaxHeapTupleSize
Definition: htup_details.h:560
int errcode_for_file_access(void)
Definition: elog.c:593
XLogRecPtr ReplicationSlotsComputeLogicalRestartLSN(void)
Definition: slot.c:791
#define InvalidTransactionId
Definition: transam.h:31
unsigned int uint32
Definition: c.h:358
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2468
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1342
MemoryContext CurrentMemoryContext
Definition: mcxt.c:38
bool heap_freeze_tuple(HeapTupleHeader tuple, TransactionId relfrozenxid, TransactionId relminmxid, TransactionId cutoff_xid, TransactionId cutoff_multi)
Definition: heapam.c:6367
static void dlist_delete(dlist_node *node)
Definition: ilist.h:358
#define ereport(elevel, rest)
Definition: elog.h:141
XLogRecPtr rs_begin_lsn
Definition: rewriteheap.c:159
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:300
#define RelationGetTargetPageFreeSpace(relation, defaultff)
Definition: rel.h:307
#define S_ISREG(m)
Definition: win32_port.h:308
HTAB * rs_logical_mappings
Definition: rewriteheap.c:162
int CloseTransientFile(int fd)
Definition: fd.c:2434
TransactionId rs_oldest_xmin
Definition: rewriteheap.c:149
#define stat(a, b)
Definition: win32_port.h:264
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:235
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:235
int FileWrite(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:1916
MemoryContext rs_cxt
Definition: rewriteheap.c:157
void * palloc0(Size size)
Definition: mcxt.c:955
#define RelationIsAccessibleInLogicalDecoding(relation)
Definition: rel.h:572
int data_sync_elevel(int elevel)
Definition: fd.c:3485
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:316
#define HEAP_XMAX_IS_LOCKED_ONLY(infomask)
Definition: htup_details.h:230
Oid MyDatabaseId
Definition: globals.c:85
Size keysize
Definition: hsearch.h:72
#define RelationGetNumberOfBlocks(reln)
Definition: bufmgr.h:198
ItemPointerData tid
Definition: rewriteheap.c:175
#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:517
RelFileNode rd_node
Definition: rel.h:54
void FileClose(File file)
Definition: fd.c:1713
static void logical_begin_heap_rewrite(RewriteState state)
Definition: rewriteheap.c:814
ItemPointerData new_tid
Definition: rewriteheap.h:40
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:732
HTAB * rs_old_new_tid_map
Definition: rewriteheap.c:161
Definition: regguts.h:298
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2534
#define HeapTupleHeaderGetXmin(tup)
Definition: htup_details.h:313
size_t Size
Definition: c.h:466
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1318
dlist_head mappings
Definition: rewriteheap.c:208
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1198
#define MAXALIGN(LEN)
Definition: c.h:685
XLogRecPtr GetRedoRecPtr(void)
Definition: xlog.c:8171
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1389
#define RelationNeedsWAL(relation)
Definition: rel.h:518
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:863
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:537
#define lstat(path, sb)
Definition: win32_port.h:253
#define HEAP_INSERT_SKIP_FSM
Definition: heapam.h:33
bool rewrite_heap_dead_tuple(RewriteState state, HeapTuple old_tuple)
Definition: rewriteheap.c:581
#define ItemPointerSetInvalid(pointer)
Definition: itemptr.h:172
#define HeapTupleHasExternal(tuple)
Definition: htup_details.h:673
void * palloc(Size size)
Definition: mcxt.c:924
int errmsg(const char *fmt,...)
Definition: elog.c:784
#define HEAP_INSERT_NO_LOGICAL
Definition: heapam.h:35
TransactionId xmin
Definition: rewriteheap.c:174
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:771
LogicalRewriteMappingData map
Definition: rewriteheap.c:218
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:972
#define elog(elevel,...)
Definition: elog.h:226
static void raw_heap_insert(RewriteState state, HeapTuple tup)
Definition: rewriteheap.c:631
struct LogicalRewriteMappingData LogicalRewriteMappingData
#define HEAP_XACT_MASK
Definition: htup_details.h:218
int pg_fsync(int fd)
Definition: fd.c:333
Relation rs_old_rel
Definition: rewriteheap.c:142
char d_name[MAX_PATH]
Definition: dirent.h:14
#define HEAP_DEFAULT_FILLFACTOR
Definition: rel.h:278
struct RewriteMappingDataEntry RewriteMappingDataEntry
#define TransactionIdIsNormal(xid)
Definition: transam.h:42
BlockNumber rs_blockno
Definition: rewriteheap.c:145
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define snprintf
Definition: port.h:192
void rewrite_heap_tuple(RewriteState state, HeapTuple old_tuple, HeapTuple new_tuple)
Definition: rewriteheap.c:379
#define RelationGetRelid(relation)
Definition: rel.h:416
OldToNewMappingData * OldToNewMapping
Definition: rewriteheap.c:196
int FreeDir(DIR *dir)
Definition: fd.c:2586
#define PageGetItem(page, itemId)
Definition: bufpage.h:340
Pointer Page
Definition: bufpage.h:78
#define ItemPointerSet(pointer, blockNumber, offNum)
Definition: itemptr.h:127
void PageInit(Page page, Size pageSize, Size specialSize)
Definition: bufpage.c:42
ItemPointerData old_tid
Definition: rewriteheap.h:39
#define ftruncate(a, b)
Definition: win32_port.h:60