1 /*-------------------------------------------------------------------------
4 * PostgreSQL transaction-commit-log manager
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.
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.
26 * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
27 * Portions Copyright (c) 1994, Regents of the University of California
29 * $PostgreSQL: pgsql/src/backend/access/transam/clog.c,v 1.50 2008/11/03 19:26:07 alvherre Exp $
31 *-------------------------------------------------------------------------
35 #include "access/clog.h"
36 #include "access/slru.h"
37 #include "access/transam.h"
39 #include "postmaster/bgwriter.h"
42 * Defines for CLOG page sizes. A page is the same BLCKSZ as is used
43 * everywhere else in Postgres.
45 * Note: because TransactionIds are 32 bits and wrap around at 0xFFFFFFFF,
46 * CLOG page numbering also wraps around at 0xFFFFFFFF/CLOG_XACTS_PER_PAGE,
47 * and CLOG segment numbering at 0xFFFFFFFF/CLOG_XACTS_PER_SEGMENT. We need
48 * take no explicit notice of that fact in this module, except when comparing
49 * segment and page numbers in TruncateCLOG (see CLOGPagePrecedes).
52 /* We need two bits per xact, so four xacts fit in a byte */
53 #define CLOG_BITS_PER_XACT 2
54 #define CLOG_XACTS_PER_BYTE 4
55 #define CLOG_XACTS_PER_PAGE (BLCKSZ * CLOG_XACTS_PER_BYTE)
56 #define CLOG_XACT_BITMASK ((1 << CLOG_BITS_PER_XACT) - 1)
58 #define TransactionIdToPage(xid) ((xid) / (TransactionId) CLOG_XACTS_PER_PAGE)
59 #define TransactionIdToPgIndex(xid) ((xid) % (TransactionId) CLOG_XACTS_PER_PAGE)
60 #define TransactionIdToByte(xid) (TransactionIdToPgIndex(xid) / CLOG_XACTS_PER_BYTE)
61 #define TransactionIdToBIndex(xid) ((xid) % (TransactionId) CLOG_XACTS_PER_BYTE)
63 /* We store the latest async LSN for each group of transactions */
64 #define CLOG_XACTS_PER_LSN_GROUP 32 /* keep this a power of 2 */
65 #define CLOG_LSNS_PER_PAGE (CLOG_XACTS_PER_PAGE / CLOG_XACTS_PER_LSN_GROUP)
67 #define GetLSNIndex(slotno, xid) ((slotno) * CLOG_LSNS_PER_PAGE + \
68 ((xid) % (TransactionId) CLOG_XACTS_PER_PAGE) / CLOG_XACTS_PER_LSN_GROUP)
72 * Link to shared-memory data structures for CLOG control
74 static SlruCtlData ClogCtlData;
76 #define ClogCtl (&ClogCtlData)
79 static int ZeroCLOGPage(int pageno, bool writeXlog);
80 static bool CLOGPagePrecedes(int page1, int page2);
81 static void WriteZeroPageXlogRec(int pageno);
82 static void WriteTruncateXlogRec(int pageno);
83 static void TransactionIdSetPageStatus(TransactionId xid, int nsubxids,
84 TransactionId *subxids, XidStatus status,
85 XLogRecPtr lsn, int pageno);
86 static void TransactionIdSetStatusBit(TransactionId xid, XidStatus status,
87 XLogRecPtr lsn, int slotno);
88 static void set_status_by_pages(int nsubxids, TransactionId *subxids,
89 XidStatus status, XLogRecPtr lsn);
93 * TransactionIdSetTreeStatus
95 * Record the final state of transaction entries in the commit log for
96 * a transaction and its subtransaction tree. Take care to ensure this is
97 * efficient, and as atomic as possible.
99 * xid is a single xid to set status for. This will typically be
100 * the top level transactionid for a top level commit or abort. It can
101 * also be a subtransaction when we record transaction aborts.
103 * subxids is an array of xids of length nsubxids, representing subtransactions
104 * in the tree of xid. In various cases nsubxids may be zero.
106 * lsn must be the WAL location of the commit record when recording an async
107 * commit. For a synchronous commit it can be InvalidXLogRecPtr, since the
108 * caller guarantees the commit record is already flushed in that case. It
109 * should be InvalidXLogRecPtr for abort cases, too.
111 * In the commit case, atomicity is limited by whether all the subxids are in
112 * the same CLOG page as xid. If they all are, then the lock will be grabbed
113 * only once, and the status will be set to committed directly. Otherwise
115 * 1. set sub-committed all subxids that are not on the same page as the
117 * 2. atomically set committed the main xid and the subxids on the same page
118 * 3. go over the first bunch again and set them committed
119 * Note that as far as concurrent checkers are concerned, main transaction
120 * commit as a whole is still atomic.
123 * TransactionId t commits and has subxids t1, t2, t3, t4
124 * t is on page p1, t1 is also on p1, t2 and t3 are on p2, t4 is on p3
125 * 1. update pages2-3:
126 * page2: set t2,t3 as sub-committed
127 * page3: set t4 as sub-committed
129 * set t1 as sub-committed,
130 * then set t as committed,
131 then set t1 as committed
132 * 3. update pages2-3:
133 * page2: set t2,t3 as committed
134 * page3: set t4 as committed
136 * NB: this is a low-level routine and is NOT the preferred entry point
137 * for most uses; functions in transam.c are the intended callers.
139 * XXX Think about issuing FADVISE_WILLNEED on pages that we will need,
140 * but aren't yet in cache, as well as hinting pages not to fall out of
144 TransactionIdSetTreeStatus(TransactionId xid, int nsubxids,
145 TransactionId *subxids, XidStatus status, XLogRecPtr lsn)
147 int pageno = TransactionIdToPage(xid); /* get page of parent */
150 Assert(status == TRANSACTION_STATUS_COMMITTED ||
151 status == TRANSACTION_STATUS_ABORTED);
154 * See how many subxids, if any, are on the same page as the parent, if any.
156 for (i = 0; i < nsubxids; i++)
158 if (TransactionIdToPage(subxids[i]) != pageno)
163 * Do all items fit on a single page?
168 * Set the parent and all subtransactions in a single call
170 TransactionIdSetPageStatus(xid, nsubxids, subxids, status, lsn,
175 int nsubxids_on_first_page = i;
178 * If this is a commit then we care about doing this correctly (i.e.
179 * using the subcommitted intermediate status). By here, we know we're
180 * updating more than one page of clog, so we must mark entries that
181 * are *not* on the first page so that they show as subcommitted before
182 * we then return to update the status to fully committed.
184 * To avoid touching the first page twice, skip marking subcommitted
185 * for the subxids on that first page.
187 if (status == TRANSACTION_STATUS_COMMITTED)
188 set_status_by_pages(nsubxids - nsubxids_on_first_page,
189 subxids + nsubxids_on_first_page,
190 TRANSACTION_STATUS_SUB_COMMITTED, lsn);
193 * Now set the parent and subtransactions on same page as the parent,
196 pageno = TransactionIdToPage(xid);
197 TransactionIdSetPageStatus(xid, nsubxids_on_first_page, subxids, status,
201 * Now work through the rest of the subxids one clog page at a time,
202 * starting from the second page onwards, like we did above.
204 set_status_by_pages(nsubxids - nsubxids_on_first_page,
205 subxids + nsubxids_on_first_page,
211 * Helper for TransactionIdSetTreeStatus: set the status for a bunch of
212 * transactions, chunking in the separate CLOG pages involved. We never
213 * pass the whole transaction tree to this function, only subtransactions
214 * that are on different pages to the top level transaction id.
217 set_status_by_pages(int nsubxids, TransactionId *subxids,
218 XidStatus status, XLogRecPtr lsn)
220 int pageno = TransactionIdToPage(subxids[0]);
228 while (TransactionIdToPage(subxids[i]) == pageno && i < nsubxids)
234 TransactionIdSetPageStatus(InvalidTransactionId,
235 num_on_page, subxids + offset,
236 status, lsn, pageno);
238 pageno = TransactionIdToPage(subxids[offset]);
243 * Record the final state of transaction entries in the commit log for
244 * all entries on a single page. Atomic only on this page.
246 * Otherwise API is same as TransactionIdSetTreeStatus()
249 TransactionIdSetPageStatus(TransactionId xid, int nsubxids,
250 TransactionId *subxids, XidStatus status,
251 XLogRecPtr lsn, int pageno)
256 Assert(status == TRANSACTION_STATUS_COMMITTED ||
257 status == TRANSACTION_STATUS_ABORTED ||
258 (status == TRANSACTION_STATUS_SUB_COMMITTED && !TransactionIdIsValid(xid)));
260 LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
263 * If we're doing an async commit (ie, lsn is valid), then we must wait
264 * for any active write on the page slot to complete. Otherwise our
265 * update could reach disk in that write, which will not do since we
266 * mustn't let it reach disk until we've done the appropriate WAL flush.
267 * But when lsn is invalid, it's OK to scribble on a page while it is
268 * write-busy, since we don't care if the update reaches disk sooner than
271 slotno = SimpleLruReadPage(ClogCtl, pageno, XLogRecPtrIsInvalid(lsn), xid);
274 * Set the main transaction id, if any.
276 * If we update more than one xid on this page while it is being written
277 * out, we might find that some of the bits go to disk and others don't.
278 * If we are updating commits on the page with the top-level xid that could
279 * break atomicity, so we subcommit the subxids first before we mark the
282 if (TransactionIdIsValid(xid))
284 /* Subtransactions first, if needed ... */
285 if (status == TRANSACTION_STATUS_COMMITTED)
287 for (i = 0; i < nsubxids; i++)
289 Assert(ClogCtl->shared->page_number[slotno] == TransactionIdToPage(subxids[i]));
290 TransactionIdSetStatusBit(subxids[i],
291 TRANSACTION_STATUS_SUB_COMMITTED,
296 /* ... then the main transaction */
297 TransactionIdSetStatusBit(xid, status, lsn, slotno);
300 /* Set the subtransactions */
301 for (i = 0; i < nsubxids; i++)
303 Assert(ClogCtl->shared->page_number[slotno] == TransactionIdToPage(subxids[i]));
304 TransactionIdSetStatusBit(subxids[i], status, lsn, slotno);
307 ClogCtl->shared->page_dirty[slotno] = true;
309 LWLockRelease(CLogControlLock);
313 * Sets the commit status of a single transaction.
315 * Must be called with CLogControlLock held
318 TransactionIdSetStatusBit(TransactionId xid, XidStatus status, XLogRecPtr lsn, int slotno)
320 int byteno = TransactionIdToByte(xid);
321 int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
326 byteptr = ClogCtl->shared->page_buffer[slotno] + byteno;
327 curval = (*byteptr >> bshift) & CLOG_XACT_BITMASK;
330 * When replaying transactions during recovery we still need to perform
331 * the two phases of subcommit and then commit. However, some transactions
332 * are already correctly marked, so we just treat those as a no-op which
333 * allows us to keep the following Assert as restrictive as possible.
335 if (InRecovery && status == TRANSACTION_STATUS_SUB_COMMITTED &&
336 curval == TRANSACTION_STATUS_COMMITTED)
340 * Current state change should be from 0 or subcommitted to target state
341 * or we should already be there when replaying changes during recovery.
343 Assert(curval == 0 ||
344 (curval == TRANSACTION_STATUS_SUB_COMMITTED &&
345 status != TRANSACTION_STATUS_IN_PROGRESS) ||
348 /* note this assumes exclusive access to the clog page */
350 byteval &= ~(((1 << CLOG_BITS_PER_XACT) - 1) << bshift);
351 byteval |= (status << bshift);
355 * Update the group LSN if the transaction completion LSN is higher.
357 * Note: lsn will be invalid when supplied during InRecovery processing,
358 * so we don't need to do anything special to avoid LSN updates during
359 * recovery. After recovery completes the next clog change will set the
362 if (!XLogRecPtrIsInvalid(lsn))
364 int lsnindex = GetLSNIndex(slotno, xid);
366 if (XLByteLT(ClogCtl->shared->group_lsn[lsnindex], lsn))
367 ClogCtl->shared->group_lsn[lsnindex] = lsn;
372 * Interrogate the state of a transaction in the commit log.
374 * Aside from the actual commit status, this function returns (into *lsn)
375 * an LSN that is late enough to be able to guarantee that if we flush up to
376 * that LSN then we will have flushed the transaction's commit record to disk.
377 * The result is not necessarily the exact LSN of the transaction's commit
378 * record! For example, for long-past transactions (those whose clog pages
379 * already migrated to disk), we'll return InvalidXLogRecPtr. Also, because
380 * we group transactions on the same clog page to conserve storage, we might
381 * return the LSN of a later transaction that falls into the same group.
383 * NB: this is a low-level routine and is NOT the preferred entry point
384 * for most uses; TransactionLogFetch() in transam.c is the intended caller.
387 TransactionIdGetStatus(TransactionId xid, XLogRecPtr *lsn)
389 int pageno = TransactionIdToPage(xid);
390 int byteno = TransactionIdToByte(xid);
391 int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
397 /* lock is acquired by SimpleLruReadPage_ReadOnly */
399 slotno = SimpleLruReadPage_ReadOnly(ClogCtl, pageno, xid);
400 byteptr = ClogCtl->shared->page_buffer[slotno] + byteno;
402 status = (*byteptr >> bshift) & CLOG_XACT_BITMASK;
404 lsnindex = GetLSNIndex(slotno, xid);
405 *lsn = ClogCtl->shared->group_lsn[lsnindex];
407 LWLockRelease(CLogControlLock);
414 * Initialization of shared memory for CLOG
419 return SimpleLruShmemSize(NUM_CLOG_BUFFERS, CLOG_LSNS_PER_PAGE);
425 ClogCtl->PagePrecedes = CLOGPagePrecedes;
426 SimpleLruInit(ClogCtl, "CLOG Ctl", NUM_CLOG_BUFFERS, CLOG_LSNS_PER_PAGE,
427 CLogControlLock, "pg_clog");
431 * This func must be called ONCE on system install. It creates
432 * the initial CLOG segment. (The CLOG directory is assumed to
433 * have been created by the initdb shell script, and CLOGShmemInit
434 * must have been called already.)
441 LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
443 /* Create and zero the first page of the commit log */
444 slotno = ZeroCLOGPage(0, false);
446 /* Make sure it's written out */
447 SimpleLruWritePage(ClogCtl, slotno, NULL);
448 Assert(!ClogCtl->shared->page_dirty[slotno]);
450 LWLockRelease(CLogControlLock);
454 * Initialize (or reinitialize) a page of CLOG to zeroes.
455 * If writeXlog is TRUE, also emit an XLOG record saying we did this.
457 * The page is not actually written, just set up in shared memory.
458 * The slot number of the new page is returned.
460 * Control lock must be held at entry, and will be held at exit.
463 ZeroCLOGPage(int pageno, bool writeXlog)
467 slotno = SimpleLruZeroPage(ClogCtl, pageno);
470 WriteZeroPageXlogRec(pageno);
476 * This must be called ONCE during postmaster or standalone-backend startup,
477 * after StartupXLOG has initialized ShmemVariableCache->nextXid.
482 TransactionId xid = ShmemVariableCache->nextXid;
483 int pageno = TransactionIdToPage(xid);
485 LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
488 * Initialize our idea of the latest page number.
490 ClogCtl->shared->latest_page_number = pageno;
493 * Zero out the remainder of the current clog page. Under normal
494 * circumstances it should be zeroes already, but it seems at least
495 * theoretically possible that XLOG replay will have settled on a nextXID
496 * value that is less than the last XID actually used and marked by the
497 * previous database lifecycle (since subtransaction commit writes clog
498 * but makes no WAL entry). Let's just be safe. (We need not worry about
499 * pages beyond the current one, since those will be zeroed when first
500 * used. For the same reason, there is no need to do anything when
501 * nextXid is exactly at a page boundary; and it's likely that the
502 * "current" page doesn't exist yet in that case.)
504 if (TransactionIdToPgIndex(xid) != 0)
506 int byteno = TransactionIdToByte(xid);
507 int bshift = TransactionIdToBIndex(xid) * CLOG_BITS_PER_XACT;
511 slotno = SimpleLruReadPage(ClogCtl, pageno, false, xid);
512 byteptr = ClogCtl->shared->page_buffer[slotno] + byteno;
514 /* Zero so-far-unused positions in the current byte */
515 *byteptr &= (1 << bshift) - 1;
516 /* Zero the rest of the page */
517 MemSet(byteptr + 1, 0, BLCKSZ - byteno - 1);
519 ClogCtl->shared->page_dirty[slotno] = true;
522 LWLockRelease(CLogControlLock);
526 * This must be called ONCE during postmaster or standalone-backend shutdown
531 /* Flush dirty CLOG pages to disk */
532 TRACE_POSTGRESQL_CLOG_CHECKPOINT_START(false);
533 SimpleLruFlush(ClogCtl, false);
534 TRACE_POSTGRESQL_CLOG_CHECKPOINT_DONE(false);
538 * Perform a checkpoint --- either during shutdown, or on-the-fly
543 /* Flush dirty CLOG pages to disk */
544 TRACE_POSTGRESQL_CLOG_CHECKPOINT_START(true);
545 SimpleLruFlush(ClogCtl, true);
546 TRACE_POSTGRESQL_CLOG_CHECKPOINT_DONE(true);
551 * Make sure that CLOG has room for a newly-allocated XID.
553 * NB: this is called while holding XidGenLock. We want it to be very fast
554 * most of the time; even when it's not so fast, no actual I/O need happen
555 * unless we're forced to write out a dirty clog or xlog page to make room
559 ExtendCLOG(TransactionId newestXact)
564 * No work except at first XID of a page. But beware: just after
565 * wraparound, the first XID of page zero is FirstNormalTransactionId.
567 if (TransactionIdToPgIndex(newestXact) != 0 &&
568 !TransactionIdEquals(newestXact, FirstNormalTransactionId))
571 pageno = TransactionIdToPage(newestXact);
573 LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
575 /* Zero the page and make an XLOG entry about it */
576 ZeroCLOGPage(pageno, true);
578 LWLockRelease(CLogControlLock);
583 * Remove all CLOG segments before the one holding the passed transaction ID
585 * Before removing any CLOG data, we must flush XLOG to disk, to ensure
586 * that any recently-emitted HEAP_FREEZE records have reached disk; otherwise
587 * a crash and restart might leave us with some unfrozen tuples referencing
588 * removed CLOG data. We choose to emit a special TRUNCATE XLOG record too.
589 * Replaying the deletion from XLOG is not critical, since the files could
590 * just as well be removed later, but doing so prevents a long-running hot
591 * standby server from acquiring an unreasonably bloated CLOG directory.
593 * Since CLOG segments hold a large number of transactions, the opportunity to
594 * actually remove a segment is fairly rare, and so it seems best not to do
595 * the XLOG flush unless we have confirmed that there is a removable segment.
598 TruncateCLOG(TransactionId oldestXact)
603 * The cutoff point is the start of the segment containing oldestXact. We
604 * pass the *page* containing oldestXact to SimpleLruTruncate.
606 cutoffPage = TransactionIdToPage(oldestXact);
608 /* Check to see if there's any files that could be removed */
609 if (!SlruScanDirectory(ClogCtl, cutoffPage, false))
610 return; /* nothing to remove */
612 /* Write XLOG record and flush XLOG to disk */
613 WriteTruncateXlogRec(cutoffPage);
615 /* Now we can remove the old CLOG segment(s) */
616 SimpleLruTruncate(ClogCtl, cutoffPage);
621 * Decide which of two CLOG page numbers is "older" for truncation purposes.
623 * We need to use comparison of TransactionIds here in order to do the right
624 * thing with wraparound XID arithmetic. However, if we are asked about
625 * page number zero, we don't want to hand InvalidTransactionId to
626 * TransactionIdPrecedes: it'll get weird about permanent xact IDs. So,
627 * offset both xids by FirstNormalTransactionId to avoid that.
630 CLOGPagePrecedes(int page1, int page2)
635 xid1 = ((TransactionId) page1) * CLOG_XACTS_PER_PAGE;
636 xid1 += FirstNormalTransactionId;
637 xid2 = ((TransactionId) page2) * CLOG_XACTS_PER_PAGE;
638 xid2 += FirstNormalTransactionId;
640 return TransactionIdPrecedes(xid1, xid2);
645 * Write a ZEROPAGE xlog record
648 WriteZeroPageXlogRec(int pageno)
652 rdata.data = (char *) (&pageno);
653 rdata.len = sizeof(int);
654 rdata.buffer = InvalidBuffer;
656 (void) XLogInsert(RM_CLOG_ID, CLOG_ZEROPAGE, &rdata);
660 * Write a TRUNCATE xlog record
662 * We must flush the xlog record to disk before returning --- see notes
666 WriteTruncateXlogRec(int pageno)
671 rdata.data = (char *) (&pageno);
672 rdata.len = sizeof(int);
673 rdata.buffer = InvalidBuffer;
675 recptr = XLogInsert(RM_CLOG_ID, CLOG_TRUNCATE, &rdata);
680 * CLOG resource manager's routines
683 clog_redo(XLogRecPtr lsn, XLogRecord *record)
685 uint8 info = record->xl_info & ~XLR_INFO_MASK;
687 if (info == CLOG_ZEROPAGE)
692 memcpy(&pageno, XLogRecGetData(record), sizeof(int));
694 LWLockAcquire(CLogControlLock, LW_EXCLUSIVE);
696 slotno = ZeroCLOGPage(pageno, false);
697 SimpleLruWritePage(ClogCtl, slotno, NULL);
698 Assert(!ClogCtl->shared->page_dirty[slotno]);
700 LWLockRelease(CLogControlLock);
702 else if (info == CLOG_TRUNCATE)
706 memcpy(&pageno, XLogRecGetData(record), sizeof(int));
709 * During XLOG replay, latest_page_number isn't set up yet; insert a
710 * suitable value to bypass the sanity test in SimpleLruTruncate.
712 ClogCtl->shared->latest_page_number = pageno;
714 SimpleLruTruncate(ClogCtl, pageno);
717 elog(PANIC, "clog_redo: unknown op code %u", info);
721 clog_desc(StringInfo buf, uint8 xl_info, char *rec)
723 uint8 info = xl_info & ~XLR_INFO_MASK;
725 if (info == CLOG_ZEROPAGE)
729 memcpy(&pageno, rec, sizeof(int));
730 appendStringInfo(buf, "zeropage: %d", pageno);
732 else if (info == CLOG_TRUNCATE)
736 memcpy(&pageno, rec, sizeof(int));
737 appendStringInfo(buf, "truncate before: %d", pageno);
740 appendStringInfo(buf, "UNKNOWN");