PostgreSQL Source Code  git master
clog.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * clog.c
4  * PostgreSQL transaction-commit-log manager
5  *
6  * This module replaces the old "pg_log" access code, which treated pg_log
7  * essentially like a relation, in that it went through the regular buffer
8  * manager. The problem with that was that there wasn't any good way to
9  * recycle storage space for transactions so old that they'll never be
10  * looked up again. Now we use specialized access code so that the commit
11  * log can be broken into relatively small, independent segments.
12  *
13  * XLOG interactions: this module generates an XLOG record whenever a new
14  * CLOG page is initialized to zeroes. Other writes of CLOG come from
15  * recording of transaction commit or abort in xact.c, which generates its
16  * own XLOG records for these events and will re-perform the status update
17  * on redo; so we need make no additional XLOG entry here. For synchronous
18  * transaction commits, the XLOG is guaranteed flushed through the XLOG commit
19  * record before we are called to log a commit, so the WAL rule "write xlog
20  * before data" is satisfied automatically. However, for async commits we
21  * must track the latest LSN affecting each CLOG page, so that we can flush
22  * XLOG that far and satisfy the WAL rule. We don't have to worry about this
23  * for aborts (whether sync or async), since the post-crash assumption would
24  * be that such transactions failed anyway.
25  *
26  * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
27  * Portions Copyright (c) 1994, Regents of the University of California
28  *
29  * src/backend/access/transam/clog.c
30  *
31  *-------------------------------------------------------------------------
32  */
33 #include "postgres.h"
34 
35 #include "access/clog.h"
36 #include "access/slru.h"
37 #include "access/transam.h"
38 #include "access/xlog.h"
39 #include "access/xloginsert.h"
40 #include "access/xlogutils.h"
41 #include "miscadmin.h"
42 #include "pg_trace.h"
43 #include "pgstat.h"
44 #include "storage/proc.h"
45 #include "storage/sync.h"
46 
47 /*
48  * Defines for CLOG page sizes. A page is the same BLCKSZ as is used
49  * everywhere else in Postgres.
50  *
51  * Note: because TransactionIds are 32 bits and wrap around at 0xFFFFFFFF,
52  * CLOG page numbering also wraps around at 0xFFFFFFFF/CLOG_XACTS_PER_PAGE,
53  * and CLOG segment numbering at
54  * 0xFFFFFFFF/CLOG_XACTS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. We need take no
55  * explicit notice of that fact in this module, except when comparing segment
56  * and page numbers in TruncateCLOG (see CLOGPagePrecedes).
57  */
58 
59 /* We need two bits per xact, so four xacts fit in a byte */
60 #define CLOG_BITS_PER_XACT 2
61 #define CLOG_XACTS_PER_BYTE 4
62 #define CLOG_XACTS_PER_PAGE (BLCKSZ * CLOG_XACTS_PER_BYTE)
63 #define CLOG_XACT_BITMASK ((1 << CLOG_BITS_PER_XACT) - 1)
64 
65 #define TransactionIdToPage(xid) ((xid) / (TransactionId) CLOG_XACTS_PER_PAGE)
66 #define TransactionIdToPgIndex(xid) ((xid) % (TransactionId) CLOG_XACTS_PER_PAGE)
67 #define TransactionIdToByte(xid) (TransactionIdToPgIndex(xid) / CLOG_XACTS_PER_BYTE)
68 #define TransactionIdToBIndex(xid) ((xid) % (TransactionId) CLOG_XACTS_PER_BYTE)
69 
70 /* We store the latest async LSN for each group of transactions */
71 #define CLOG_XACTS_PER_LSN_GROUP 32 /* keep this a power of 2 */
72 #define CLOG_LSNS_PER_PAGE (CLOG_XACTS_PER_PAGE / CLOG_XACTS_PER_LSN_GROUP)
73 
74 #define GetLSNIndex(slotno, xid) ((slotno) * CLOG_LSNS_PER_PAGE + \
75  ((xid) % (TransactionId) CLOG_XACTS_PER_PAGE) / CLOG_XACTS_PER_LSN_GROUP)
76 
77 /*
78  * The number of subtransactions below which we consider to apply clog group
79  * update optimization. Testing reveals that the number higher than this can
80  * hurt performance.
81  */
82 #define THRESHOLD_SUBTRANS_CLOG_OPT 5
83 
84 /*
85  * Link to shared-memory data structures for CLOG control
86  */
88 
89 #define XactCtl (&XactCtlData)
90 
91 
92 static int ZeroCLOGPage(int pageno, bool writeXlog);
93 static bool CLOGPagePrecedes(int page1, int page2);
94 static void WriteZeroPageXlogRec(int pageno);
95 static void WriteTruncateXlogRec(int pageno, TransactionId oldestXact,
96  Oid oldestXactDb);
97 static void TransactionIdSetPageStatus(TransactionId xid, int nsubxids,
98  TransactionId *subxids, XidStatus status,
99  XLogRecPtr lsn, int pageno,
100  bool all_xact_same_page);
102  XLogRecPtr lsn, int slotno);
103 static void set_status_by_pages(int nsubxids, TransactionId *subxids,
106  XidStatus status, XLogRecPtr lsn, int pageno);
107 static void TransactionIdSetPageStatusInternal(TransactionId xid, int nsubxids,
108  TransactionId *subxids, XidStatus status,
109  XLogRecPtr lsn, int pageno);
110 
111 
112 /*
113  * TransactionIdSetTreeStatus
114  *
115  * Record the final state of transaction entries in the commit log for
116  * a transaction and its subtransaction tree. Take care to ensure this is
117  * efficient, and as atomic as possible.
118  *
119  * xid is a single xid to set status for. This will typically be
120  * the top level transactionid for a top level commit or abort. It can
121  * also be a subtransaction when we record transaction aborts.
122  *
123  * subxids is an array of xids of length nsubxids, representing subtransactions
124  * in the tree of xid. In various cases nsubxids may be zero.
125  *
126  * lsn must be the WAL location of the commit record when recording an async
127  * commit. For a synchronous commit it can be InvalidXLogRecPtr, since the
128  * caller guarantees the commit record is already flushed in that case. It
129  * should be InvalidXLogRecPtr for abort cases, too.
130  *
131  * In the commit case, atomicity is limited by whether all the subxids are in
132  * the same CLOG page as xid. If they all are, then the lock will be grabbed
133  * only once, and the status will be set to committed directly. Otherwise
134  * we must
135  * 1. set sub-committed all subxids that are not on the same page as the
136  * main xid
137  * 2. atomically set committed the main xid and the subxids on the same page
138  * 3. go over the first bunch again and set them committed
139  * Note that as far as concurrent checkers are concerned, main transaction
140  * commit as a whole is still atomic.
141  *
142  * Example:
143  * TransactionId t commits and has subxids t1, t2, t3, t4
144  * t is on page p1, t1 is also on p1, t2 and t3 are on p2, t4 is on p3
145  * 1. update pages2-3:
146  * page2: set t2,t3 as sub-committed
147  * page3: set t4 as sub-committed
148  * 2. update page1:
149  * set t1 as sub-committed,
150  * then set t as committed,
151  then set t1 as committed
152  * 3. update pages2-3:
153  * page2: set t2,t3 as committed
154  * page3: set t4 as committed
155  *
156  * NB: this is a low-level routine and is NOT the preferred entry point
157  * for most uses; functions in transam.c are the intended callers.
158  *
159  * XXX Think about issuing POSIX_FADV_WILLNEED on pages that we will need,
160  * but aren't yet in cache, as well as hinting pages not to fall out of
161  * cache yet.
162  */
163 void
165  TransactionId *subxids, XidStatus status, XLogRecPtr lsn)
166 {
167  int pageno = TransactionIdToPage(xid); /* get page of parent */
168  int i;
169 
172 
173  /*
174  * See how many subxids, if any, are on the same page as the parent, if
175  * any.
176  */
177  for (i = 0; i < nsubxids; i++)
178  {
179  if (TransactionIdToPage(subxids[i]) != pageno)
180  break;
181  }
182 
183  /*
184  * Do all items fit on a single page?
185  */
186  if (i == nsubxids)
187  {
188  /*
189  * Set the parent and all subtransactions in a single call
190  */
191  TransactionIdSetPageStatus(xid, nsubxids, subxids, status, lsn,
192  pageno, true);
193  }
194  else
195  {
196  int nsubxids_on_first_page = i;
197 
198  /*
199  * If this is a commit then we care about doing this correctly (i.e.
200  * using the subcommitted intermediate status). By here, we know
201  * we're updating more than one page of clog, so we must mark entries
202  * that are *not* on the first page so that they show as subcommitted
203  * before we then return to update the status to fully committed.
204  *
205  * To avoid touching the first page twice, skip marking subcommitted
206  * for the subxids on that first page.
207  */
209  set_status_by_pages(nsubxids - nsubxids_on_first_page,
210  subxids + nsubxids_on_first_page,
212 
213  /*
214  * Now set the parent and subtransactions on same page as the parent,
215  * if any
216  */
217  pageno = TransactionIdToPage(xid);
218  TransactionIdSetPageStatus(xid, nsubxids_on_first_page, subxids, status,
219  lsn, pageno, false);
220 
221  /*
222  * Now work through the rest of the subxids one clog page at a time,
223  * starting from the second page onwards, like we did above.
224  */
225  set_status_by_pages(nsubxids - nsubxids_on_first_page,
226  subxids + nsubxids_on_first_page,
227  status, lsn);
228  }
229 }
230 
231 /*
232  * Helper for TransactionIdSetTreeStatus: set the status for a bunch of
233  * transactions, chunking in the separate CLOG pages involved. We never
234  * pass the whole transaction tree to this function, only subtransactions
235  * that are on different pages to the top level transaction id.
236  */
237 static void
238 set_status_by_pages(int nsubxids, TransactionId *subxids,
240 {
241  int pageno = TransactionIdToPage(subxids[0]);
242  int offset = 0;
243  int i = 0;
244 
245  Assert(nsubxids > 0); /* else the pageno fetch above is unsafe */
246 
247  while (i < nsubxids)
248  {
249  int num_on_page = 0;
250  int nextpageno;
251 
252  do
253  {
254  nextpageno = TransactionIdToPage(subxids[i]);
255  if (nextpageno != pageno)
256  break;
257  num_on_page++;
258  i++;
259  } while (i < nsubxids);
260 
262  num_on_page, subxids + offset,
263  status, lsn, pageno, false);
264  offset = i;
265  pageno = nextpageno;
266  }
267 }
268 
269 /*
270  * Record the final state of transaction entries in the commit log for all
271  * entries on a single page. Atomic only on this page.
272  */
273 static void
275  TransactionId *subxids, XidStatus status,
276  XLogRecPtr lsn, int pageno,
277  bool all_xact_same_page)
278 {
279  /* Can't use group update when PGPROC overflows. */
281  "group clog threshold less than PGPROC cached subxids");
282 
283  /*
284  * When there is contention on XactSLRULock, we try to group multiple
285  * updates; a single leader process will perform transaction status
286  * updates for multiple backends so that the number of times XactSLRULock
287  * needs to be acquired is reduced.
288  *
289  * For this optimization to be safe, the XID and subxids in MyProc must be
290  * the same as the ones for which we're setting the status. Check that
291  * this is the case.
292  *
293  * For this optimization to be efficient, we shouldn't have too many
294  * sub-XIDs and all of the XIDs for which we're adjusting clog should be
295  * on the same page. Check those conditions, too.
296  */
297  if (all_xact_same_page && xid == MyProc->xid &&
298  nsubxids <= THRESHOLD_SUBTRANS_CLOG_OPT &&
299  nsubxids == MyProc->subxidStatus.count &&
300  (nsubxids == 0 ||
301  memcmp(subxids, MyProc->subxids.xids,
302  nsubxids * sizeof(TransactionId)) == 0))
303  {
304  /*
305  * If we can immediately acquire XactSLRULock, we update the status of
306  * our own XID and release the lock. If not, try use group XID
307  * update. If that doesn't work out, fall back to waiting for the
308  * lock to perform an update for this transaction only.
309  */
310  if (LWLockConditionalAcquire(XactSLRULock, LW_EXCLUSIVE))
311  {
312  /* Got the lock without waiting! Do the update. */
313  TransactionIdSetPageStatusInternal(xid, nsubxids, subxids, status,
314  lsn, pageno);
315  LWLockRelease(XactSLRULock);
316  return;
317  }
318  else if (TransactionGroupUpdateXidStatus(xid, status, lsn, pageno))
319  {
320  /* Group update mechanism has done the work. */
321  return;
322  }
323 
324  /* Fall through only if update isn't done yet. */
325  }
326 
327  /* Group update not applicable, or couldn't accept this page number. */
328  LWLockAcquire(XactSLRULock, LW_EXCLUSIVE);
329  TransactionIdSetPageStatusInternal(xid, nsubxids, subxids, status,
330  lsn, pageno);
331  LWLockRelease(XactSLRULock);
332 }
333 
334 /*
335  * Record the final state of transaction entry in the commit log
336  *
337  * We don't do any locking here; caller must handle that.
338  */
339 static void
341  TransactionId *subxids, XidStatus status,
342  XLogRecPtr lsn, int pageno)
343 {
344  int slotno;
345  int i;
346 
350  Assert(LWLockHeldByMeInMode(XactSLRULock, LW_EXCLUSIVE));
351 
352  /*
353  * If we're doing an async commit (ie, lsn is valid), then we must wait
354  * for any active write on the page slot to complete. Otherwise our
355  * update could reach disk in that write, which will not do since we
356  * mustn't let it reach disk until we've done the appropriate WAL flush.
357  * But when lsn is invalid, it's OK to scribble on a page while it is
358  * write-busy, since we don't care if the update reaches disk sooner than
359  * we think.
360  */
361  slotno = SimpleLruReadPage(XactCtl, pageno, XLogRecPtrIsInvalid(lsn), xid);
362 
363  /*
364  * Set the main transaction id, if any.
365  *
366  * If we update more than one xid on this page while it is being written
367  * out, we might find that some of the bits go to disk and others don't.
368  * If we are updating commits on the page with the top-level xid that
369  * could break atomicity, so we subcommit the subxids first before we mark
370  * the top-level commit.
371  */
372  if (TransactionIdIsValid(xid))
373  {
374  /* Subtransactions first, if needed ... */
376  {
377  for (i = 0; i < nsubxids; i++)
378  {
379  Assert(XactCtl->shared->page_number[slotno] == TransactionIdToPage(subxids[i]));
380  TransactionIdSetStatusBit(subxids[i],
382  lsn, slotno);
383  }
384  }
385 
386  /* ... then the main transaction */
387  TransactionIdSetStatusBit(xid, status, lsn, slotno);
388  }
389 
390  /* Set the subtransactions */
391  for (i = 0; i < nsubxids; i++)
392  {
393  Assert(XactCtl->shared->page_number[slotno] == TransactionIdToPage(subxids[i]));
394  TransactionIdSetStatusBit(subxids[i], status, lsn, slotno);
395  }
396 
397  XactCtl->shared->page_dirty[slotno] = true;
398 }
399 
400 /*
401  * When we cannot immediately acquire XactSLRULock in exclusive mode at
402  * commit time, add ourselves to a list of processes that need their XIDs
403  * status update. The first process to add itself to the list will acquire
404  * XactSLRULock in exclusive mode and set transaction status as required
405  * on behalf of all group members. This avoids a great deal of contention
406  * around XactSLRULock when many processes are trying to commit at once,
407  * since the lock need not be repeatedly handed off from one committing
408  * process to the next.
409  *
410  * Returns true when transaction status has been updated in clog; returns
411  * false if we decided against applying the optimization because the page
412  * number we need to update differs from those processes already waiting.
413  */
414 static bool
416  XLogRecPtr lsn, int pageno)
417 {
418  volatile PROC_HDR *procglobal = ProcGlobal;
419  PGPROC *proc = MyProc;
420  uint32 nextidx;
421  uint32 wakeidx;
422 
423  /* We should definitely have an XID whose status needs to be updated. */
425 
426  /*
427  * Add ourselves to the list of processes needing a group XID status
428  * update.
429  */
430  proc->clogGroupMember = true;
431  proc->clogGroupMemberXid = xid;
433  proc->clogGroupMemberPage = pageno;
434  proc->clogGroupMemberLsn = lsn;
435 
436  nextidx = pg_atomic_read_u32(&procglobal->clogGroupFirst);
437 
438  while (true)
439  {
440  /*
441  * Add the proc to list, if the clog page where we need to update the
442  * current transaction status is same as group leader's clog page.
443  *
444  * There is a race condition here, which is that after doing the below
445  * check and before adding this proc's clog update to a group, the
446  * group leader might have already finished the group update for this
447  * page and becomes group leader of another group. This will lead to a
448  * situation where a single group can have different clog page
449  * updates. This isn't likely and will still work, just maybe a bit
450  * less efficiently.
451  */
452  if (nextidx != INVALID_PGPROCNO &&
454  {
455  /*
456  * Ensure that this proc is not a member of any clog group that
457  * needs an XID status update.
458  */
459  proc->clogGroupMember = false;
461  return false;
462  }
463 
464  pg_atomic_write_u32(&proc->clogGroupNext, nextidx);
465 
467  &nextidx,
468  (uint32) proc->pgprocno))
469  break;
470  }
471 
472  /*
473  * If the list was not empty, the leader will update the status of our
474  * XID. It is impossible to have followers without a leader because the
475  * first process that has added itself to the list will always have
476  * nextidx as INVALID_PGPROCNO.
477  */
478  if (nextidx != INVALID_PGPROCNO)
479  {
480  int extraWaits = 0;
481 
482  /* Sleep until the leader updates our XID status. */
484  for (;;)
485  {
486  /* acts as a read barrier */
487  PGSemaphoreLock(proc->sem);
488  if (!proc->clogGroupMember)
489  break;
490  extraWaits++;
491  }
493 
495 
496  /* Fix semaphore count for any absorbed wakeups */
497  while (extraWaits-- > 0)
498  PGSemaphoreUnlock(proc->sem);
499  return true;
500  }
501 
502  /* We are the leader. Acquire the lock on behalf of everyone. */
503  LWLockAcquire(XactSLRULock, LW_EXCLUSIVE);
504 
505  /*
506  * Now that we've got the lock, clear the list of processes waiting for
507  * group XID status update, saving a pointer to the head of the list.
508  * Trying to pop elements one at a time could lead to an ABA problem.
509  */
510  nextidx = pg_atomic_exchange_u32(&procglobal->clogGroupFirst,
512 
513  /* Remember head of list so we can perform wakeups after dropping lock. */
514  wakeidx = nextidx;
515 
516  /* Walk the list and update the status of all XIDs. */
517  while (nextidx != INVALID_PGPROCNO)
518  {
519  PGPROC *proc = &ProcGlobal->allProcs[nextidx];
520 
521  /*
522  * Transactions with more than THRESHOLD_SUBTRANS_CLOG_OPT sub-XIDs
523  * should not use group XID status update mechanism.
524  */
526 
528  proc->subxidStatus.count,
529  proc->subxids.xids,
531  proc->clogGroupMemberLsn,
532  proc->clogGroupMemberPage);
533 
534  /* Move to next proc in list. */
535  nextidx = pg_atomic_read_u32(&proc->clogGroupNext);
536  }
537 
538  /* We're done with the lock now. */
539  LWLockRelease(XactSLRULock);
540 
541  /*
542  * Now that we've released the lock, go back and wake everybody up. We
543  * don't do this under the lock so as to keep lock hold times to a
544  * minimum.
545  */
546  while (wakeidx != INVALID_PGPROCNO)
547  {
548  PGPROC *proc = &ProcGlobal->allProcs[wakeidx];
549 
550  wakeidx = pg_atomic_read_u32(&proc->clogGroupNext);
552 
553  /* ensure all previous writes are visible before follower continues. */
555 
556  proc->clogGroupMember = false;
557 
558  if (proc != MyProc)
559  PGSemaphoreUnlock(proc->sem);
560  }
561 
562  return true;
563 }
564 
565 /*
566  * Sets the commit status of a single transaction.
567  *
568  * Must be called with XactSLRULock held
569  */
570 static void
572 {
573  int byteno = TransactionIdToByte(xid);
574  int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
575  char *byteptr;
576  char byteval;
577  char curval;
578 
579  byteptr = XactCtl->shared->page_buffer[slotno] + byteno;
580  curval = (*byteptr >> bshift) & CLOG_XACT_BITMASK;
581 
582  /*
583  * When replaying transactions during recovery we still need to perform
584  * the two phases of subcommit and then commit. However, some transactions
585  * are already correctly marked, so we just treat those as a no-op which
586  * allows us to keep the following Assert as restrictive as possible.
587  */
590  return;
591 
592  /*
593  * Current state change should be from 0 or subcommitted to target state
594  * or we should already be there when replaying changes during recovery.
595  */
596  Assert(curval == 0 ||
599  curval == status);
600 
601  /* note this assumes exclusive access to the clog page */
602  byteval = *byteptr;
603  byteval &= ~(((1 << CLOG_BITS_PER_XACT) - 1) << bshift);
604  byteval |= (status << bshift);
605  *byteptr = byteval;
606 
607  /*
608  * Update the group LSN if the transaction completion LSN is higher.
609  *
610  * Note: lsn will be invalid when supplied during InRecovery processing,
611  * so we don't need to do anything special to avoid LSN updates during
612  * recovery. After recovery completes the next clog change will set the
613  * LSN correctly.
614  */
615  if (!XLogRecPtrIsInvalid(lsn))
616  {
617  int lsnindex = GetLSNIndex(slotno, xid);
618 
619  if (XactCtl->shared->group_lsn[lsnindex] < lsn)
620  XactCtl->shared->group_lsn[lsnindex] = lsn;
621  }
622 }
623 
624 /*
625  * Interrogate the state of a transaction in the commit log.
626  *
627  * Aside from the actual commit status, this function returns (into *lsn)
628  * an LSN that is late enough to be able to guarantee that if we flush up to
629  * that LSN then we will have flushed the transaction's commit record to disk.
630  * The result is not necessarily the exact LSN of the transaction's commit
631  * record! For example, for long-past transactions (those whose clog pages
632  * already migrated to disk), we'll return InvalidXLogRecPtr. Also, because
633  * we group transactions on the same clog page to conserve storage, we might
634  * return the LSN of a later transaction that falls into the same group.
635  *
636  * NB: this is a low-level routine and is NOT the preferred entry point
637  * for most uses; TransactionLogFetch() in transam.c is the intended caller.
638  */
639 XidStatus
641 {
642  int pageno = TransactionIdToPage(xid);
643  int byteno = TransactionIdToByte(xid);
644  int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
645  int slotno;
646  int lsnindex;
647  char *byteptr;
649 
650  /* lock is acquired by SimpleLruReadPage_ReadOnly */
651 
652  slotno = SimpleLruReadPage_ReadOnly(XactCtl, pageno, xid);
653  byteptr = XactCtl->shared->page_buffer[slotno] + byteno;
654 
655  status = (*byteptr >> bshift) & CLOG_XACT_BITMASK;
656 
657  lsnindex = GetLSNIndex(slotno, xid);
658  *lsn = XactCtl->shared->group_lsn[lsnindex];
659 
660  LWLockRelease(XactSLRULock);
661 
662  return status;
663 }
664 
665 /*
666  * Number of shared CLOG buffers.
667  *
668  * On larger multi-processor systems, it is possible to have many CLOG page
669  * requests in flight at one time which could lead to disk access for CLOG
670  * page if the required page is not found in memory. Testing revealed that we
671  * can get the best performance by having 128 CLOG buffers, more than that it
672  * doesn't improve performance.
673  *
674  * Unconditionally keeping the number of CLOG buffers to 128 did not seem like
675  * a good idea, because it would increase the minimum amount of shared memory
676  * required to start, which could be a problem for people running very small
677  * configurations. The following formula seems to represent a reasonable
678  * compromise: people with very low values for shared_buffers will get fewer
679  * CLOG buffers as well, and everyone else will get 128.
680  */
681 Size
683 {
684  return Min(128, Max(4, NBuffers / 512));
685 }
686 
687 /*
688  * Initialization of shared memory for CLOG
689  */
690 Size
692 {
694 }
695 
696 void
698 {
699  XactCtl->PagePrecedes = CLOGPagePrecedes;
701  XactSLRULock, "pg_xact", LWTRANCHE_XACT_BUFFER,
704 }
705 
706 /*
707  * This func must be called ONCE on system install. It creates
708  * the initial CLOG segment. (The CLOG directory is assumed to
709  * have been created by initdb, and CLOGShmemInit must have been
710  * called already.)
711  */
712 void
714 {
715  int slotno;
716 
717  LWLockAcquire(XactSLRULock, LW_EXCLUSIVE);
718 
719  /* Create and zero the first page of the commit log */
720  slotno = ZeroCLOGPage(0, false);
721 
722  /* Make sure it's written out */
723  SimpleLruWritePage(XactCtl, slotno);
724  Assert(!XactCtl->shared->page_dirty[slotno]);
725 
726  LWLockRelease(XactSLRULock);
727 }
728 
729 /*
730  * Initialize (or reinitialize) a page of CLOG to zeroes.
731  * If writeXlog is true, also emit an XLOG record saying we did this.
732  *
733  * The page is not actually written, just set up in shared memory.
734  * The slot number of the new page is returned.
735  *
736  * Control lock must be held at entry, and will be held at exit.
737  */
738 static int
739 ZeroCLOGPage(int pageno, bool writeXlog)
740 {
741  int slotno;
742 
743  slotno = SimpleLruZeroPage(XactCtl, pageno);
744 
745  if (writeXlog)
746  WriteZeroPageXlogRec(pageno);
747 
748  return slotno;
749 }
750 
751 /*
752  * This must be called ONCE during postmaster or standalone-backend startup,
753  * after StartupXLOG has initialized ShmemVariableCache->nextXid.
754  */
755 void
757 {
759  int pageno = TransactionIdToPage(xid);
760 
761  LWLockAcquire(XactSLRULock, LW_EXCLUSIVE);
762 
763  /*
764  * Initialize our idea of the latest page number.
765  */
766  XactCtl->shared->latest_page_number = pageno;
767 
768  LWLockRelease(XactSLRULock);
769 }
770 
771 /*
772  * This must be called ONCE at the end of startup/recovery.
773  */
774 void
775 TrimCLOG(void)
776 {
778  int pageno = TransactionIdToPage(xid);
779 
780  LWLockAcquire(XactSLRULock, LW_EXCLUSIVE);
781 
782  /*
783  * Zero out the remainder of the current clog page. Under normal
784  * circumstances it should be zeroes already, but it seems at least
785  * theoretically possible that XLOG replay will have settled on a nextXID
786  * value that is less than the last XID actually used and marked by the
787  * previous database lifecycle (since subtransaction commit writes clog
788  * but makes no WAL entry). Let's just be safe. (We need not worry about
789  * pages beyond the current one, since those will be zeroed when first
790  * used. For the same reason, there is no need to do anything when
791  * nextXid is exactly at a page boundary; and it's likely that the
792  * "current" page doesn't exist yet in that case.)
793  */
794  if (TransactionIdToPgIndex(xid) != 0)
795  {
796  int byteno = TransactionIdToByte(xid);
797  int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
798  int slotno;
799  char *byteptr;
800 
801  slotno = SimpleLruReadPage(XactCtl, pageno, false, xid);
802  byteptr = XactCtl->shared->page_buffer[slotno] + byteno;
803 
804  /* Zero so-far-unused positions in the current byte */
805  *byteptr &= (1 << bshift) - 1;
806  /* Zero the rest of the page */
807  MemSet(byteptr + 1, 0, BLCKSZ - byteno - 1);
808 
809  XactCtl->shared->page_dirty[slotno] = true;
810  }
811 
812  LWLockRelease(XactSLRULock);
813 }
814 
815 /*
816  * Perform a checkpoint --- either during shutdown, or on-the-fly
817  */
818 void
820 {
821  /*
822  * Write dirty CLOG pages to disk. This may result in sync requests
823  * queued for later handling by ProcessSyncRequests(), as part of the
824  * checkpoint.
825  */
826  TRACE_POSTGRESQL_CLOG_CHECKPOINT_START(true);
827  SimpleLruWriteAll(XactCtl, true);
828  TRACE_POSTGRESQL_CLOG_CHECKPOINT_DONE(true);
829 }
830 
831 
832 /*
833  * Make sure that CLOG has room for a newly-allocated XID.
834  *
835  * NB: this is called while holding XidGenLock. We want it to be very fast
836  * most of the time; even when it's not so fast, no actual I/O need happen
837  * unless we're forced to write out a dirty clog or xlog page to make room
838  * in shared memory.
839  */
840 void
842 {
843  int pageno;
844 
845  /*
846  * No work except at first XID of a page. But beware: just after
847  * wraparound, the first XID of page zero is FirstNormalTransactionId.
848  */
849  if (TransactionIdToPgIndex(newestXact) != 0 &&
851  return;
852 
853  pageno = TransactionIdToPage(newestXact);
854 
855  LWLockAcquire(XactSLRULock, LW_EXCLUSIVE);
856 
857  /* Zero the page and make an XLOG entry about it */
858  ZeroCLOGPage(pageno, true);
859 
860  LWLockRelease(XactSLRULock);
861 }
862 
863 
864 /*
865  * Remove all CLOG segments before the one holding the passed transaction ID
866  *
867  * Before removing any CLOG data, we must flush XLOG to disk, to ensure
868  * that any recently-emitted FREEZE_PAGE records have reached disk; otherwise
869  * a crash and restart might leave us with some unfrozen tuples referencing
870  * removed CLOG data. We choose to emit a special TRUNCATE XLOG record too.
871  * Replaying the deletion from XLOG is not critical, since the files could
872  * just as well be removed later, but doing so prevents a long-running hot
873  * standby server from acquiring an unreasonably bloated CLOG directory.
874  *
875  * Since CLOG segments hold a large number of transactions, the opportunity to
876  * actually remove a segment is fairly rare, and so it seems best not to do
877  * the XLOG flush unless we have confirmed that there is a removable segment.
878  */
879 void
880 TruncateCLOG(TransactionId oldestXact, Oid oldestxid_datoid)
881 {
882  int cutoffPage;
883 
884  /*
885  * The cutoff point is the start of the segment containing oldestXact. We
886  * pass the *page* containing oldestXact to SimpleLruTruncate.
887  */
888  cutoffPage = TransactionIdToPage(oldestXact);
889 
890  /* Check to see if there's any files that could be removed */
892  return; /* nothing to remove */
893 
894  /*
895  * Advance oldestClogXid before truncating clog, so concurrent xact status
896  * lookups can ensure they don't attempt to access truncated-away clog.
897  *
898  * It's only necessary to do this if we will actually truncate away clog
899  * pages.
900  */
901  AdvanceOldestClogXid(oldestXact);
902 
903  /*
904  * Write XLOG record and flush XLOG to disk. We record the oldest xid
905  * we're keeping information about here so we can ensure that it's always
906  * ahead of clog truncation in case we crash, and so a standby finds out
907  * the new valid xid before the next checkpoint.
908  */
909  WriteTruncateXlogRec(cutoffPage, oldestXact, oldestxid_datoid);
910 
911  /* Now we can remove the old CLOG segment(s) */
912  SimpleLruTruncate(XactCtl, cutoffPage);
913 }
914 
915 
916 /*
917  * Decide whether a CLOG page number is "older" for truncation purposes.
918  *
919  * We need to use comparison of TransactionIds here in order to do the right
920  * thing with wraparound XID arithmetic. However, TransactionIdPrecedes()
921  * would get weird about permanent xact IDs. So, offset both such that xid1,
922  * xid2, and xid2 + CLOG_XACTS_PER_PAGE - 1 are all normal XIDs; this offset
923  * is relevant to page 0 and to the page preceding page 0.
924  *
925  * The page containing oldestXact-2^31 is the important edge case. The
926  * portion of that page equaling or following oldestXact-2^31 is expendable,
927  * but the portion preceding oldestXact-2^31 is not. When oldestXact-2^31 is
928  * the first XID of a page and segment, the entire page and segment is
929  * expendable, and we could truncate the segment. Recognizing that case would
930  * require making oldestXact, not just the page containing oldestXact,
931  * available to this callback. The benefit would be rare and small, so we
932  * don't optimize that edge case.
933  */
934 static bool
935 CLOGPagePrecedes(int page1, int page2)
936 {
937  TransactionId xid1;
938  TransactionId xid2;
939 
940  xid1 = ((TransactionId) page1) * CLOG_XACTS_PER_PAGE;
941  xid1 += FirstNormalTransactionId + 1;
942  xid2 = ((TransactionId) page2) * CLOG_XACTS_PER_PAGE;
943  xid2 += FirstNormalTransactionId + 1;
944 
945  return (TransactionIdPrecedes(xid1, xid2) &&
946  TransactionIdPrecedes(xid1, xid2 + CLOG_XACTS_PER_PAGE - 1));
947 }
948 
949 
950 /*
951  * Write a ZEROPAGE xlog record
952  */
953 static void
955 {
956  XLogBeginInsert();
957  XLogRegisterData((char *) (&pageno), sizeof(int));
958  (void) XLogInsert(RM_CLOG_ID, CLOG_ZEROPAGE);
959 }
960 
961 /*
962  * Write a TRUNCATE xlog record
963  *
964  * We must flush the xlog record to disk before returning --- see notes
965  * in TruncateCLOG().
966  */
967 static void
968 WriteTruncateXlogRec(int pageno, TransactionId oldestXact, Oid oldestXactDb)
969 {
970  XLogRecPtr recptr;
971  xl_clog_truncate xlrec;
972 
973  xlrec.pageno = pageno;
974  xlrec.oldestXact = oldestXact;
975  xlrec.oldestXactDb = oldestXactDb;
976 
977  XLogBeginInsert();
978  XLogRegisterData((char *) (&xlrec), sizeof(xl_clog_truncate));
979  recptr = XLogInsert(RM_CLOG_ID, CLOG_TRUNCATE);
980  XLogFlush(recptr);
981 }
982 
983 /*
984  * CLOG resource manager's routines
985  */
986 void
988 {
989  uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
990 
991  /* Backup blocks are not used in clog records */
992  Assert(!XLogRecHasAnyBlockRefs(record));
993 
994  if (info == CLOG_ZEROPAGE)
995  {
996  int pageno;
997  int slotno;
998 
999  memcpy(&pageno, XLogRecGetData(record), sizeof(int));
1000 
1001  LWLockAcquire(XactSLRULock, LW_EXCLUSIVE);
1002 
1003  slotno = ZeroCLOGPage(pageno, false);
1004  SimpleLruWritePage(XactCtl, slotno);
1005  Assert(!XactCtl->shared->page_dirty[slotno]);
1006 
1007  LWLockRelease(XactSLRULock);
1008  }
1009  else if (info == CLOG_TRUNCATE)
1010  {
1011  xl_clog_truncate xlrec;
1012 
1013  memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_clog_truncate));
1014 
1016 
1018  }
1019  else
1020  elog(PANIC, "clog_redo: unknown op code %u", info);
1021 }
1022 
1023 /*
1024  * Entrypoint for sync.c to sync clog files.
1025  */
1026 int
1027 clogsyncfiletag(const FileTag *ftag, char *path)
1028 {
1029  return SlruSyncFileTag(XactCtl, ftag, path);
1030 }
static bool pg_atomic_compare_exchange_u32(volatile pg_atomic_uint32 *ptr, uint32 *expected, uint32 newval)
Definition: atomics.h:311
#define pg_write_barrier()
Definition: atomics.h:159
static void pg_atomic_write_u32(volatile pg_atomic_uint32 *ptr, uint32 val)
Definition: atomics.h:258
static uint32 pg_atomic_read_u32(volatile pg_atomic_uint32 *ptr)
Definition: atomics.h:241
static uint32 pg_atomic_exchange_u32(volatile pg_atomic_uint32 *ptr, uint32 newval)
Definition: atomics.h:292
unsigned int uint32
Definition: c.h:452
#define Min(x, y)
Definition: c.h:997
#define Max(x, y)
Definition: c.h:991
unsigned char uint8
Definition: c.h:450
#define MemSet(start, val, len)
Definition: c.h:1019
#define StaticAssertStmt(condition, errmessage)
Definition: c.h:929
uint32 TransactionId
Definition: c.h:598
size_t Size
Definition: c.h:551
#define CLOG_XACT_BITMASK
Definition: clog.c:63
#define CLOG_XACTS_PER_PAGE
Definition: clog.c:62
static void WriteZeroPageXlogRec(int pageno)
Definition: clog.c:954
static bool TransactionGroupUpdateXidStatus(TransactionId xid, XidStatus status, XLogRecPtr lsn, int pageno)
Definition: clog.c:415
static bool CLOGPagePrecedes(int page1, int page2)
Definition: clog.c:935
#define THRESHOLD_SUBTRANS_CLOG_OPT
Definition: clog.c:82
static void TransactionIdSetStatusBit(TransactionId xid, XidStatus status, XLogRecPtr lsn, int slotno)
Definition: clog.c:571
XidStatus TransactionIdGetStatus(TransactionId xid, XLogRecPtr *lsn)
Definition: clog.c:640
void TransactionIdSetTreeStatus(TransactionId xid, int nsubxids, TransactionId *subxids, XidStatus status, XLogRecPtr lsn)
Definition: clog.c:164
void CLOGShmemInit(void)
Definition: clog.c:697
#define TransactionIdToBIndex(xid)
Definition: clog.c:68
Size CLOGShmemBuffers(void)
Definition: clog.c:682
void ExtendCLOG(TransactionId newestXact)
Definition: clog.c:841
void clog_redo(XLogReaderState *record)
Definition: clog.c:987
void TruncateCLOG(TransactionId oldestXact, Oid oldestxid_datoid)
Definition: clog.c:880
Size CLOGShmemSize(void)
Definition: clog.c:691
static void TransactionIdSetPageStatus(TransactionId xid, int nsubxids, TransactionId *subxids, XidStatus status, XLogRecPtr lsn, int pageno, bool all_xact_same_page)
Definition: clog.c:274
int clogsyncfiletag(const FileTag *ftag, char *path)
Definition: clog.c:1027
void BootStrapCLOG(void)
Definition: clog.c:713
#define CLOG_BITS_PER_XACT
Definition: clog.c:60
#define CLOG_LSNS_PER_PAGE
Definition: clog.c:72
#define TransactionIdToByte(xid)
Definition: clog.c:67
#define TransactionIdToPgIndex(xid)
Definition: clog.c:66
void StartupCLOG(void)
Definition: clog.c:756
static void set_status_by_pages(int nsubxids, TransactionId *subxids, XidStatus status, XLogRecPtr lsn)
Definition: clog.c:238
#define GetLSNIndex(slotno, xid)
Definition: clog.c:74
void CheckPointCLOG(void)
Definition: clog.c:819
static SlruCtlData XactCtlData
Definition: clog.c:87
#define XactCtl
Definition: clog.c:89
void TrimCLOG(void)
Definition: clog.c:775
static void TransactionIdSetPageStatusInternal(TransactionId xid, int nsubxids, TransactionId *subxids, XidStatus status, XLogRecPtr lsn, int pageno)
Definition: clog.c:340
static void WriteTruncateXlogRec(int pageno, TransactionId oldestXact, Oid oldestXactDb)
Definition: clog.c:968
#define TransactionIdToPage(xid)
Definition: clog.c:65
static int ZeroCLOGPage(int pageno, bool writeXlog)
Definition: clog.c:739
#define TRANSACTION_STATUS_IN_PROGRESS
Definition: clog.h:27
int XidStatus
Definition: clog.h:25
#define CLOG_ZEROPAGE
Definition: clog.h:56
#define TRANSACTION_STATUS_ABORTED
Definition: clog.h:29
#define TRANSACTION_STATUS_SUB_COMMITTED
Definition: clog.h:30
#define CLOG_TRUNCATE
Definition: clog.h:57
#define TRANSACTION_STATUS_COMMITTED
Definition: clog.h:28
#define PANIC
Definition: elog.h:36
#define elog(elevel,...)
Definition: elog.h:218
int NBuffers
Definition: globals.c:136
int i
Definition: isn.c:73
Assert(fmt[strlen(fmt) - 1] !='\n')
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1196
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1800
bool LWLockHeldByMeInMode(LWLock *l, LWLockMode mode)
Definition: lwlock.c:1934
bool LWLockConditionalAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1367
@ LWTRANCHE_XACT_BUFFER
Definition: lwlock.h:169
@ LW_EXCLUSIVE
Definition: lwlock.h:104
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:229
void PGSemaphoreUnlock(PGSemaphore sema)
Definition: posix_sema.c:340
void PGSemaphoreLock(PGSemaphore sema)
Definition: posix_sema.c:320
unsigned int Oid
Definition: postgres_ext.h:31
#define INVALID_PGPROCNO
Definition: proc.h:83
#define PGPROC_MAX_CACHED_SUBXIDS
Definition: proc.h:36
void SimpleLruWritePage(SlruCtl ctl, int slotno)
Definition: slru.c:614
void SimpleLruWriteAll(SlruCtl ctl, bool allow_redirtied)
Definition: slru.c:1156
bool SlruScanDirCbReportPresence(SlruCtl ctl, char *filename, int segpage, void *data)
Definition: slru.c:1500
void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns, LWLock *ctllock, const char *subdir, int tranche_id, SyncRequestHandler sync_handler)
Definition: slru.c:187
void SimpleLruTruncate(SlruCtl ctl, int cutoffPage)
Definition: slru.c:1226
bool SlruScanDirectory(SlruCtl ctl, SlruScanCallback callback, void *data)
Definition: slru.c:1553
int SlruSyncFileTag(SlruCtl ctl, const FileTag *ftag, char *path)
Definition: slru.c:1593
int SimpleLruZeroPage(SlruCtl ctl, int pageno)
Definition: slru.c:280
int SimpleLruReadPage_ReadOnly(SlruCtl ctl, int pageno, TransactionId xid)
Definition: slru.c:495
int SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok, TransactionId xid)
Definition: slru.c:395
Size SimpleLruShmemSize(int nslots, int nlsns)
Definition: slru.c:156
#define SlruPagePrecedesUnitTests(ctl, per_page)
Definition: slru.h:156
PGPROC * MyProc
Definition: proc.c:68
PROC_HDR * ProcGlobal
Definition: proc.c:80
Definition: sync.h:51
Definition: proc.h:160
XLogRecPtr clogGroupMemberLsn
Definition: proc.h:274
int clogGroupMemberPage
Definition: proc.h:272
TransactionId clogGroupMemberXid
Definition: proc.h:269
bool clogGroupMember
Definition: proc.h:267
pg_atomic_uint32 clogGroupNext
Definition: proc.h:268
XidStatus clogGroupMemberXidStatus
Definition: proc.h:270
XidCacheStatus subxidStatus
Definition: proc.h:248
TransactionId xid
Definition: proc.h:171
int pgprocno
Definition: proc.h:188
struct XidCache subxids
Definition: proc.h:250
PGSemaphore sem
Definition: proc.h:165
Definition: proc.h:354
PGPROC * allProcs
Definition: proc.h:356
pg_atomic_uint32 clogGroupFirst
Definition: proc.h:386
FullTransactionId nextXid
Definition: transam.h:220
uint8 count
Definition: proc.h:41
TransactionId xids[PGPROC_MAX_CACHED_SUBXIDS]
Definition: proc.h:48
Oid oldestXactDb
Definition: clog.h:36
int pageno
Definition: clog.h:34
TransactionId oldestXact
Definition: clog.h:35
@ SYNC_HANDLER_CLOG
Definition: sync.h:38
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:273
#define InvalidTransactionId
Definition: transam.h:31
#define TransactionIdEquals(id1, id2)
Definition: transam.h:43
#define XidFromFullTransactionId(x)
Definition: transam.h:48
#define FirstNormalTransactionId
Definition: transam.h:34
#define TransactionIdIsValid(xid)
Definition: transam.h:41
void AdvanceOldestClogXid(TransactionId oldest_datfrozenxid)
Definition: varsup.c:328
VariableCache ShmemVariableCache
Definition: varsup.c:34
@ WAIT_EVENT_XACT_GROUP_UPDATE
Definition: wait_event.h:131
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: wait_event.h:266
static void pgstat_report_wait_end(void)
Definition: wait_event.h:282
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2509
#define XLogRecPtrIsInvalid(r)
Definition: xlogdefs.h:29
uint64 XLogRecPtr
Definition: xlogdefs.h:21
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:443
void XLogBeginInsert(void)
Definition: xloginsert.c:150
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:351
#define XLogRecGetInfo(decoder)
Definition: xlogreader.h:408
#define XLogRecGetData(decoder)
Definition: xlogreader.h:413
#define XLogRecHasAnyBlockRefs(decoder)
Definition: xlogreader.h:415
#define XLR_INFO_MASK
Definition: xlogrecord.h:62
bool InRecovery
Definition: xlogutils.c:53