Re: [BUGS] Bug #613: Sequence values fall back to previously chec - Mailing list pgsql-hackers

From Tom Lane
Subject Re: [BUGS] Bug #613: Sequence values fall back to previously chec
Date
Msg-id 20915.1016203144@sss.pgh.pa.us
Whole thread Raw
Responses Re: [BUGS] Bug #613: Sequence values fall back to previously  (Bruce Momjian <pgman@candle.pha.pa.us>)
Re: [BUGS] Bug #613: Sequence values fall back to previously  (Bruce Momjian <pgman@candle.pha.pa.us>)
List pgsql-hackers
Attached is a patch against current CVS that fixes both of the known
problems with sequences: failure to flush XLOG after a transaction
that only does "SELECT nextval()", and failure to force a new WAL
record to be written on the first nextval after a checkpoint.
(The latter uses Vadim's idea of looking at the sequence page LSN.)
I haven't tested it really extensively, but it seems to cure the
reported problems.

Some notes:

1. I found what I believe is another bug in the sequence logic:
        fetch = log = fetch - log + SEQ_LOG_VALS;
should be
        fetch = log = fetch + SEQ_LOG_VALS;
I can't see any reason to reduce the number of values prefetched
by the number formerly prefetched.  Also, if the sequence's "cache"
setting is large (more than SEQ_LOG_VALS), the original code could
easily fail to fetch as many values as it was supposed to cache,
let alone additional ones to be prefetched and logged.

2. I renamed XLogCtl->RedoRecPtr to SavedRedoRecPtr, and renamed
the associated routines to SetSavedRedoRecPtr/GetSavedRedoRecPtr,
in hopes of reducing confusion.

3. I believe it'd now be possible to remove SavedRedoRecPtr and
SetSavedRedoRecPtr/GetSavedRedoRecPtr entirely, in favor of letting
the postmaster fetch the updated pointer with GetRedoRecPtr just
like a backend would.  This would be cleaner and less code ... but
someone might object that it introduces a risk of postmaster hangup,
if some backend crashes whilst holding info_lck.  I consider that
risk minuscule given the short intervals in which info_lck is held,
but it can't be denied that the risk is not zero.  Thoughts?

Comments?  Unless I hear objections I will patch this in current
and the 7.2 branch.  (If we agree to remove SavedRedoRecPtr,
though, I don't think we should back-patch that change.)

            regards, tom lane

*** src/backend/access/transam/xact.c.orig    Tue Mar 12 07:56:31 2002
--- src/backend/access/transam/xact.c    Thu Mar 14 20:00:50 2002
***************
*** 546,577 ****
      xid = GetCurrentTransactionId();

      /*
!      * We needn't write anything in xlog or clog if the transaction was
!      * read-only, which we check by testing if it made any xlog entries.
       */
!     if (MyLastRecPtr.xrecoff != 0)
      {
-         XLogRecData rdata;
-         xl_xact_commit xlrec;
          XLogRecPtr    recptr;

          BufmgrCommit();

-         xlrec.xtime = time(NULL);
-         rdata.buffer = InvalidBuffer;
-         rdata.data = (char *) (&xlrec);
-         rdata.len = SizeOfXactCommit;
-         rdata.next = NULL;
-
          START_CRIT_SECTION();

!         /*
!          * SHOULD SAVE ARRAY OF RELFILENODE-s TO DROP
!          */
!         recptr = XLogInsert(RM_XACT_ID, XLOG_XACT_COMMIT, &rdata);

          /*
!          * Sleep before commit! So we can flush more than one commit
           * records per single fsync.  (The idea is some other backend may
           * do the XLogFlush while we're sleeping.  This needs work still,
           * because on most Unixen, the minimum select() delay is 10msec or
--- 546,593 ----
      xid = GetCurrentTransactionId();

      /*
!      * We only need to log the commit in xlog and clog if the transaction made
!      * any transaction-controlled XLOG entries.  (Otherwise, its XID appears
!      * nowhere in permanent storage, so no one will ever care if it
!      * committed.)  However, we must flush XLOG to disk if we made any XLOG
!      * entries, whether in or out of transaction control.  For example, if we
!      * reported a nextval() result to the client, this ensures that any XLOG
!      * record generated by nextval will hit the disk before we report the
!      * transaction committed.
       */
!     if (MyXactMadeXLogEntry)
      {
          XLogRecPtr    recptr;

          BufmgrCommit();

          START_CRIT_SECTION();

!         if (MyLastRecPtr.xrecoff != 0)
!         {
!             /* Need to emit a commit record */
!             XLogRecData rdata;
!             xl_xact_commit xlrec;
!
!             xlrec.xtime = time(NULL);
!             rdata.buffer = InvalidBuffer;
!             rdata.data = (char *) (&xlrec);
!             rdata.len = SizeOfXactCommit;
!             rdata.next = NULL;
!
!             /*
!              * XXX SHOULD SAVE ARRAY OF RELFILENODE-s TO DROP
!              */
!             recptr = XLogInsert(RM_XACT_ID, XLOG_XACT_COMMIT, &rdata);
!         }
!         else
!         {
!             /* Just flush through last record written by me */
!             recptr = ProcLastRecEnd;
!         }

          /*
!          * Sleep before flush! So we can flush more than one commit
           * records per single fsync.  (The idea is some other backend may
           * do the XLogFlush while we're sleeping.  This needs work still,
           * because on most Unixen, the minimum select() delay is 10msec or
***************
*** 593,607 ****

          XLogFlush(recptr);

!         /* Break the chain of back-links in the XLOG records I output */
!         MyLastRecPtr.xrecoff = 0;
!
!         /* Mark the transaction committed in clog */
!         TransactionIdCommit(xid);

          END_CRIT_SECTION();
      }

      /* Show myself as out of the transaction in PROC array */
      MyProc->logRec.xrecoff = 0;

--- 609,625 ----

          XLogFlush(recptr);

!         /* Mark the transaction committed in clog, if needed */
!         if (MyLastRecPtr.xrecoff != 0)
!             TransactionIdCommit(xid);

          END_CRIT_SECTION();
      }

+     /* Break the chain of back-links in the XLOG records I output */
+     MyLastRecPtr.xrecoff = 0;
+     MyXactMadeXLogEntry = false;
+
      /* Show myself as out of the transaction in PROC array */
      MyProc->logRec.xrecoff = 0;

***************
*** 689,696 ****
      TransactionId xid = GetCurrentTransactionId();

      /*
!      * We needn't write anything in xlog or clog if the transaction was
!      * read-only, which we check by testing if it made any xlog entries.
       *
       * Extra check here is to catch case that we aborted partway through
       * RecordTransactionCommit ...
--- 707,717 ----
      TransactionId xid = GetCurrentTransactionId();

      /*
!      * We only need to log the abort in xlog and clog if the transaction made
!      * any transaction-controlled XLOG entries.  (Otherwise, its XID appears
!      * nowhere in permanent storage, so no one will ever care if it
!      * committed.)  We do not flush XLOG to disk in any case, since the
!      * default assumption after a crash would be that we aborted, anyway.
       *
       * Extra check here is to catch case that we aborted partway through
       * RecordTransactionCommit ...
***************
*** 714,724 ****
           */
          recptr = XLogInsert(RM_XACT_ID, XLOG_XACT_ABORT, &rdata);

-         /*
-          * There's no need for XLogFlush here, since the default
-          * assumption would be that we aborted, anyway.
-          */
-
          /* Mark the transaction aborted in clog */
          TransactionIdAbort(xid);

--- 735,740 ----
***************
*** 727,732 ****
--- 743,750 ----

      /* Break the chain of back-links in the XLOG records I output */
      MyLastRecPtr.xrecoff = 0;
+     MyXactMadeXLogEntry = false;
+
      /* Show myself as out of the transaction in PROC array */
      MyProc->logRec.xrecoff = 0;

*** src/backend/access/transam/xlog.c.orig    Tue Mar 12 07:56:31 2002
--- src/backend/access/transam/xlog.c    Thu Mar 14 20:29:51 2002
***************
*** 131,157 ****

  /*
   * MyLastRecPtr points to the start of the last XLOG record inserted by the
!  * current transaction.  If MyLastRecPtr.xrecoff == 0, then we are not in
!  * a transaction or the transaction has not yet made any loggable changes.
   *
   * Note that XLOG records inserted outside transaction control are not
!  * reflected into MyLastRecPtr.
   */
  XLogRecPtr    MyLastRecPtr = {0, 0};

  /*
   * ProcLastRecPtr points to the start of the last XLOG record inserted by the
   * current backend.  It is updated for all inserts, transaction-controlled
!  * or not.
   */
  static XLogRecPtr ProcLastRecPtr = {0, 0};

  /*
   * RedoRecPtr is this backend's local copy of the REDO record pointer
   * (which is almost but not quite the same as a pointer to the most recent
   * CHECKPOINT record).    We update this from the shared-memory copy,
   * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
!  * hold the Insert lock).  See XLogInsert for details.
   */
  static XLogRecPtr RedoRecPtr;

--- 131,166 ----

  /*
   * MyLastRecPtr points to the start of the last XLOG record inserted by the
!  * current transaction.  If MyLastRecPtr.xrecoff == 0, then the current
!  * xact hasn't yet inserted any transaction-controlled XLOG records.
   *
   * Note that XLOG records inserted outside transaction control are not
!  * reflected into MyLastRecPtr.  They do, however, cause MyXactMadeXLogEntry
!  * to be set true.  The latter can be used to test whether the current xact
!  * made any loggable changes (including out-of-xact changes, such as
!  * sequence updates).
   */
  XLogRecPtr    MyLastRecPtr = {0, 0};

+ bool        MyXactMadeXLogEntry = false;
+
  /*
   * ProcLastRecPtr points to the start of the last XLOG record inserted by the
   * current backend.  It is updated for all inserts, transaction-controlled
!  * or not.  ProcLastRecEnd is similar but points to end+1 of last record.
   */
  static XLogRecPtr ProcLastRecPtr = {0, 0};

+ XLogRecPtr    ProcLastRecEnd = {0, 0};
+
  /*
   * RedoRecPtr is this backend's local copy of the REDO record pointer
   * (which is almost but not quite the same as a pointer to the most recent
   * CHECKPOINT record).    We update this from the shared-memory copy,
   * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
!  * hold the Insert lock).  See XLogInsert for details.  We are also allowed
!  * to update from XLogCtl->Insert.RedoRecPtr if we hold the info_lck;
!  * see GetRedoRecPtr.
   */
  static XLogRecPtr RedoRecPtr;

***************
*** 272,278 ****
      StartUpID    ThisStartUpID;

      /* This value is not protected by *any* lock... */
!     XLogRecPtr    RedoRecPtr;        /* see SetRedoRecPtr/GetRedoRecPtr */

      slock_t        info_lck;        /* locks shared LogwrtRqst/LogwrtResult */
  } XLogCtlData;
--- 281,288 ----
      StartUpID    ThisStartUpID;

      /* This value is not protected by *any* lock... */
!     /* see SetSavedRedoRecPtr/GetSavedRedoRecPtr */
!     XLogRecPtr    SavedRedoRecPtr;

      slock_t        info_lck;        /* locks shared LogwrtRqst/LogwrtResult */
  } XLogCtlData;
***************
*** 777,782 ****
--- 787,793 ----
          MyLastRecPtr = RecPtr;
      ProcLastRecPtr = RecPtr;
      Insert->PrevRecord = RecPtr;
+     MyXactMadeXLogEntry = true;

      Insert->currpos += SizeOfXLogRecord;
      freespace -= SizeOfXLogRecord;
***************
*** 855,860 ****
--- 866,873 ----
          SpinLockRelease_NoHoldoff(&xlogctl->info_lck);
      }

+     ProcLastRecEnd = RecPtr;
+
      END_CRIT_SECTION();

      return (RecPtr);
***************
*** 2538,2544 ****

      ThisStartUpID = checkPoint.ThisStartUpID;
      RedoRecPtr = XLogCtl->Insert.RedoRecPtr =
!         XLogCtl->RedoRecPtr = checkPoint.redo;

      if (XLByteLT(RecPtr, checkPoint.redo))
          elog(PANIC, "invalid redo in checkpoint record");
--- 2551,2557 ----

      ThisStartUpID = checkPoint.ThisStartUpID;
      RedoRecPtr = XLogCtl->Insert.RedoRecPtr =
!         XLogCtl->SavedRedoRecPtr = checkPoint.redo;

      if (XLByteLT(RecPtr, checkPoint.redo))
          elog(PANIC, "invalid redo in checkpoint record");
***************
*** 2824,2855 ****
  SetThisStartUpID(void)
  {
      ThisStartUpID = XLogCtl->ThisStartUpID;
!     RedoRecPtr = XLogCtl->RedoRecPtr;
  }

  /*
   * CheckPoint process called by postmaster saves copy of new RedoRecPtr
!  * in shmem (using SetRedoRecPtr).    When checkpointer completes, postmaster
!  * calls GetRedoRecPtr to update its own copy of RedoRecPtr, so that
!  * subsequently-spawned backends will start out with a reasonably up-to-date
!  * local RedoRecPtr.  Since these operations are not protected by any lock
!  * and copying an XLogRecPtr isn't atomic, it's unsafe to use either of these
!  * routines at other times!
!  *
!  * Note: once spawned, a backend must update its local RedoRecPtr from
!  * XLogCtl->Insert.RedoRecPtr while holding the insert lock.  This is
!  * done in XLogInsert().
   */
  void
! SetRedoRecPtr(void)
  {
!     XLogCtl->RedoRecPtr = RedoRecPtr;
  }

  void
  GetRedoRecPtr(void)
  {
!     RedoRecPtr = XLogCtl->RedoRecPtr;
  }

  /*
--- 2837,2883 ----
  SetThisStartUpID(void)
  {
      ThisStartUpID = XLogCtl->ThisStartUpID;
!     RedoRecPtr = XLogCtl->SavedRedoRecPtr;
  }

  /*
   * CheckPoint process called by postmaster saves copy of new RedoRecPtr
!  * in shmem (using SetSavedRedoRecPtr).  When checkpointer completes,
!  * postmaster calls GetSavedRedoRecPtr to update its own copy of RedoRecPtr,
!  * so that subsequently-spawned backends will start out with a reasonably
!  * up-to-date local RedoRecPtr.  Since these operations are not protected by
!  * any lock and copying an XLogRecPtr isn't atomic, it's unsafe to use either
!  * of these routines at other times!
   */
  void
! SetSavedRedoRecPtr(void)
  {
!     XLogCtl->SavedRedoRecPtr = RedoRecPtr;
  }

  void
+ GetSavedRedoRecPtr(void)
+ {
+     RedoRecPtr = XLogCtl->SavedRedoRecPtr;
+ }
+
+ /*
+  * Once spawned, a backend may update its local RedoRecPtr from
+  * XLogCtl->Insert.RedoRecPtr; it must hold the insert lock or info_lck
+  * to do so.  This is done in XLogInsert() or GetRedoRecPtr().
+  */
+ XLogRecPtr
  GetRedoRecPtr(void)
  {
!     /* use volatile pointer to prevent code rearrangement */
!     volatile XLogCtlData *xlogctl = XLogCtl;
!
!     SpinLockAcquire_NoHoldoff(&xlogctl->info_lck);
!     Assert(XLByteLE(RedoRecPtr, xlogctl->Insert.RedoRecPtr));
!     RedoRecPtr = xlogctl->Insert.RedoRecPtr;
!     SpinLockRelease_NoHoldoff(&xlogctl->info_lck);
!
!     return RedoRecPtr;
  }

  /*
***************
*** 2862,2867 ****
--- 2890,2896 ----

      /* suppress in-transaction check in CreateCheckPoint */
      MyLastRecPtr.xrecoff = 0;
+     MyXactMadeXLogEntry = false;

      CritSectionCount++;
      CreateDummyCaches();
***************
*** 2886,2892 ****
      uint32        _logId;
      uint32        _logSeg;

!     if (MyLastRecPtr.xrecoff != 0)
          elog(ERROR, "CreateCheckPoint: cannot be called inside transaction block");

      /*
--- 2915,2921 ----
      uint32        _logId;
      uint32        _logSeg;

!     if (MyXactMadeXLogEntry)
          elog(ERROR, "CreateCheckPoint: cannot be called inside transaction block");

      /*
***************
*** 2972,2980 ****

      /*
       * Here we update the shared RedoRecPtr for future XLogInsert calls;
!      * this must be done while holding the insert lock.
       */
!     RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;

      /*
       * Get UNDO record ptr - this is oldest of PROC->logRec values. We do
--- 3001,3016 ----

      /*
       * Here we update the shared RedoRecPtr for future XLogInsert calls;
!      * this must be done while holding the insert lock AND the info_lck.
       */
!     {
!         /* use volatile pointer to prevent code rearrangement */
!         volatile XLogCtlData *xlogctl = XLogCtl;
!
!         SpinLockAcquire_NoHoldoff(&xlogctl->info_lck);
!         RedoRecPtr = xlogctl->Insert.RedoRecPtr = checkPoint.redo;
!         SpinLockRelease_NoHoldoff(&xlogctl->info_lck);
!     }

      /*
       * Get UNDO record ptr - this is oldest of PROC->logRec values. We do
*** src/backend/bootstrap/bootstrap.c.orig    Tue Mar 12 07:56:32 2002
--- src/backend/bootstrap/bootstrap.c    Thu Mar 14 20:19:51 2002
***************
*** 386,392 ****
                  InitDummyProcess();        /* needed to get LWLocks */
              CreateDummyCaches();
              CreateCheckPoint(false);
!             SetRedoRecPtr();
              proc_exit(0);        /* done */

          case BS_XLOG_STARTUP:
--- 386,392 ----
                  InitDummyProcess();        /* needed to get LWLocks */
              CreateDummyCaches();
              CreateCheckPoint(false);
!             SetSavedRedoRecPtr(); /* pass redo ptr back to postmaster */
              proc_exit(0);        /* done */

          case BS_XLOG_STARTUP:
*** src/backend/commands/sequence.c.orig    Tue Mar 12 07:56:35 2002
--- src/backend/commands/sequence.c    Thu Mar 14 22:06:22 2002
***************
*** 286,291 ****
--- 286,292 ----
      char       *seqname = get_seq_name(seqin);
      SeqTable    elm;
      Buffer        buf;
+     Page        page;
      Form_pg_sequence seq;
      int64        incby,
                  maxv,
***************
*** 316,321 ****
--- 317,323 ----

      seq = read_info("nextval", elm, &buf);        /* lock page' buffer and
                                                   * read tuple */
+     page = BufferGetPage(buf);

      last = next = result = seq->last_value;
      incby = seq->increment_by;
***************
*** 331,341 ****
          log--;
      }

      if (log < fetch)
      {
!         fetch = log = fetch - log + SEQ_LOG_VALS;
          logit = true;
      }

      while (fetch)                /* try to fetch cache [+ log ] numbers */
      {
--- 333,365 ----
          log--;
      }

+     /*
+      * Decide whether we should emit a WAL log record.  If so, force up
+      * the fetch count to grab SEQ_LOG_VALS more values than we actually
+      * need to cache.  (These will then be usable without logging.)
+      *
+      * If this is the first nextval after a checkpoint, we must force
+      * a new WAL record to be written anyway, else replay starting from the
+      * checkpoint would fail to advance the sequence past the logged
+      * values.  In this case we may as well fetch extra values.
+      */
      if (log < fetch)
      {
!         /* forced log to satisfy local demand for values */
!         fetch = log = fetch + SEQ_LOG_VALS;
          logit = true;
      }
+     else
+     {
+         XLogRecPtr    redoptr = GetRedoRecPtr();
+
+         if (XLByteLE(PageGetLSN(page), redoptr))
+         {
+             /* last update of seq was before checkpoint */
+             fetch = log = fetch + SEQ_LOG_VALS;
+             logit = true;
+         }
+     }

      while (fetch)                /* try to fetch cache [+ log ] numbers */
      {
***************
*** 386,391 ****
--- 410,418 ----
          }
      }

+     log -= fetch;                /* adjust for any unfetched numbers */
+     Assert(log >= 0);
+
      /* save info in local cache */
      elm->last = result;            /* last returned number */
      elm->cached = last;            /* last fetched number */
***************
*** 396,402 ****
          xl_seq_rec    xlrec;
          XLogRecPtr    recptr;
          XLogRecData rdata[2];
-         Page        page = BufferGetPage(buf);

          xlrec.node = elm->rel->rd_node;
          rdata[0].buffer = InvalidBuffer;
--- 423,428 ----
***************
*** 417,431 ****

          PageSetLSN(page, recptr);
          PageSetSUI(page, ThisStartUpID);
-
-         if (fetch)                /* not all numbers were fetched */
-             log -= fetch;
      }

      /* update on-disk data */
      seq->last_value = last;        /* last fetched number */
      seq->is_called = true;
-     Assert(log >= 0);
      seq->log_cnt = log;            /* how much is logged */
      END_CRIT_SECTION();

--- 443,453 ----
*** src/backend/postmaster/postmaster.c.orig    Tue Mar 12 07:57:01 2002
--- src/backend/postmaster/postmaster.c    Thu Mar 14 20:20:01 2002
***************
*** 1683,1689 ****
              {
                  checkpointed = time(NULL);
                  /* Update RedoRecPtr for future child backends */
!                 GetRedoRecPtr();
              }
          }
          else
--- 1683,1689 ----
              {
                  checkpointed = time(NULL);
                  /* Update RedoRecPtr for future child backends */
!                 GetSavedRedoRecPtr();
              }
          }
          else
*** src/include/access/xlog.h.orig    Fri Nov 16 12:17:19 2001
--- src/include/access/xlog.h    Thu Mar 14 20:20:51 2002
***************
*** 178,183 ****
--- 178,185 ----
  extern StartUpID ThisStartUpID; /* current SUI */
  extern bool InRecovery;
  extern XLogRecPtr MyLastRecPtr;
+ extern bool MyXactMadeXLogEntry;
+ extern XLogRecPtr ProcLastRecEnd;

  /* these variables are GUC parameters related to XLOG */
  extern int    CheckPointSegments;
***************
*** 205,212 ****
  extern void CreateCheckPoint(bool shutdown);
  extern void SetThisStartUpID(void);
  extern void XLogPutNextOid(Oid nextOid);
! extern void SetRedoRecPtr(void);
! extern void GetRedoRecPtr(void);

  /* in storage/ipc/sinval.c, but don't want to declare in sinval.h because
   * we'd have to include xlog.h into that ...
--- 207,215 ----
  extern void CreateCheckPoint(bool shutdown);
  extern void SetThisStartUpID(void);
  extern void XLogPutNextOid(Oid nextOid);
! extern void SetSavedRedoRecPtr(void);
! extern void GetSavedRedoRecPtr(void);
! extern XLogRecPtr GetRedoRecPtr(void);

  /* in storage/ipc/sinval.c, but don't want to declare in sinval.h because
   * we'd have to include xlog.h into that ...

pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: [BUGS] Bug #613: Sequence values fall back to previously chec
Next
From: Tom Lane
Date:
Subject: Re: User Level Lock question