Re: [HACKERS] Infrastructure changes for recovery

[Simon Riggs]

On Thu, 2008-08-07 at 12:44 +0100, Simon Riggs wrote:

> I would like to propose some changes to the infrastructure for recovery.
> These changes are beneficial in themselves, but also form the basis for
> other work we might later contemplate.
>
> Currently
> * the startup process performs restartpoints during recovery
> * the death of the startup process is tied directly to the change of
> state in the postmaster following recovery
>
> I propose to
> * have startup process signal postmaster when it starts Redo phase (if
> it starts it)

> Decoupling things in this way allows us to
> 1. arrange for the bgwriter to start during Redo, so it can:
> i) clean dirty blocks for the startup process
> ii) perform restartpoints in background
>   Both of these aspects will increase performance of recovery

Taking into account comments from Tom and Alvaro

Included patch with the following changes:

* new postmaster mode known as consistent recovery, entered only when
recovery passes safe/consistent point. InRedo is now set in all
processes when started/connected in consistent recovery mode.

* bgwriter and stats process starts in consistent recovery mode.
bgwriter changes mode when startup process completes.

* bgwriter now performs restartpoints and also cleans shared_buffers
while the startup process performs redo apply

* recovery.conf parameter log_restartpoints is now deprecated, since
function overlaps with log_checkpoints too much. I've kept the
distinction between restartpoints and checkpoints in code, to avoid
convoluted code. Minor change, not critical.

[Replying to one of Alvaro's other comments: Startup process still uses
XLogReadBuffer. I'm not planning on changing that either, at least not
in this patch.]

Patch doesn't conflict with rmgr plugin patch.

Passes make check, but that's easy.
Various other tests all seem to be working.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support

[recovery_infrastruc.v2.patch]

---

Index: src/backend/access/transam/xlog.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/access/transam/xlog.c,v
retrieving revision 1.317
diff -c -r1.317 xlog.c
*** src/backend/access/transam/xlog.c 11 Aug 2008 11:05:10 -0000 1.317
--- src/backend/access/transam/xlog.c 31 Aug 2008 19:07:40 -0000
***************
*** 131,137 ****
  static bool recoveryTarget = false;
  static bool recoveryTargetExact = false;
  static bool recoveryTargetInclusive = true;
- static bool recoveryLogRestartpoints = false;
  static TransactionId recoveryTargetXid;
  static TimestampTz recoveryTargetTime;
  static TimestampTz recoveryLastXTime = 0;
--- 131,136 ----
***************
*** 386,392 ****
  static XLogRecord *nextRecord = NULL;
  static TimeLineID lastPageTLI = 0;
 
! static bool InRedo = false;
 
 
  static void XLogArchiveNotify(const char *xlog);
--- 385,391 ----
  static XLogRecord *nextRecord = NULL;
  static TimeLineID lastPageTLI = 0;
 
! bool InRedo = false;
 
 
  static void XLogArchiveNotify(const char *xlog);
***************
*** 480,485 ****
--- 479,488 ----
  bool doPageWrites;
  bool isLogSwitch = (rmid == RM_XLOG_ID && info == XLOG_SWITCH);
 
+ /* cross-check on whether we should be here or not */
+ if (InRedo)
+ elog(FATAL, "cannot write new WAL data during recovery mode");
+
  /* info's high bits are reserved for use by me */
  if (info & XLR_INFO_MASK)
  elog(PANIC, "invalid xlog info mask %02X", info);
***************
*** 2051,2057 ****
  unlink(tmppath);
  }
 
! elog(DEBUG2, "done creating and filling new WAL file");
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
--- 2054,2061 ----
  unlink(tmppath);
  }
 
! XLogFileName(tmppath, ThisTimeLineID, log, seg);
! elog(DEBUG2, "done creating and filling new WAL file %s", tmppath);
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
***************
*** 4532,4546 ****
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
- /*
- * does nothing if a recovery_target is not also set
- */
- if (!parse_bool(tok2, &recoveryLogRestartpoints))
-  ereport(ERROR,
- (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-  errmsg("parameter \"log_restartpoints\" requires a Boolean value")));
  ereport(LOG,
! (errmsg("log_restartpoints = %s", tok2)));
  }
  else
  ereport(FATAL,
--- 4536,4544 ----
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
  ereport(LOG,
! (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
!  errmsg("parameter \"log_restartpoints\" has been deprecated")));
  }
  else
  ereport(FATAL,
***************
*** 4811,4816 ****
--- 4809,4815 ----
  CheckPoint checkPoint;
  bool wasShutdown;
  bool reachedStopPoint = false;
+ bool reachedSafeStopPoint = false;
  bool haveBackupLabel = false;
  XLogRecPtr RecPtr,
  LastRec,
***************
*** 5039,5044 ****
--- 5038,5048 ----
  UpdateControlFile();
 
  /*
+ * Reset pgstat data, because it may be invalid after recovery.
+ */
+ pgstat_reset_all();
+
+ /*
  * If there was a backup label file, it's done its job and the info
  * has now been propagated into pg_control.  We must get rid of the
  * label file so that if we crash during recovery, we'll pick up at
***************
*** 5148,5153 ****
--- 5152,5172 ----
 
  LastRec = ReadRecPtr;
 
+ /*
+ * Have we reached our safe stopping point? If so, we can
+ * signal Postmaster to enter consistent recovery mode
+ */
+ if (!reachedSafeStopPoint &&
+ XLByteLE(ControlFile->minRecoveryPoint, EndRecPtr))
+ {
+ reachedSafeStopPoint = true;
+ ereport(LOG,
+ (errmsg("consistent recovery state reached at %X/%X",
+ EndRecPtr.xlogid, EndRecPtr.xrecoff)));
+ if (IsUnderPostmaster)
+ SendPostmasterSignal(PMSIGNAL_RECOVERY_START);
+ }
+
  record = ReadRecord(NULL, LOG);
  } while (record != NULL && recoveryContinue);
 
***************
*** 5169,5174 ****
--- 5188,5194 ----
  /* there are no WAL records following the checkpoint */
  ereport(LOG,
  (errmsg("redo is not required")));
+ reachedSafeStopPoint = true;
  }
  }
 
***************
*** 5184,5190 ****
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (XLByteLT(EndOfLog, ControlFile->minRecoveryPoint))
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
--- 5204,5210 ----
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (InRecovery && !reachedSafeStopPoint)
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
***************
*** 5305,5314 ****
  */
  XLogCheckInvalidPages();
 
! /*
! * Reset pgstat data, because it may be invalid after recovery.
! */
! pgstat_reset_all();
 
  /*
  * Perform a checkpoint to update all our recovery activity to disk.
--- 5325,5332 ----
  */
  XLogCheckInvalidPages();
 
! if (IsUnderPostmaster)
! BgWriterCompleteRestartPointImmediately();
 
  /*
  * Perform a checkpoint to update all our recovery activity to disk.
***************
*** 5318,5323 ****
--- 5336,5344 ----
  * assigning a new TLI, using a shutdown checkpoint allows us to have
  * the rule that TLI only changes in shutdown checkpoints, which
  * allows some extra error checking in xlog_redo.
+ *
+ * Note that this will wait behind any restartpoint that the bgwriter
+ * is currently performing, though will be much faster as a result.
  */
  CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
  }
***************
*** 5372,5377 ****
--- 5393,5401 ----
  readRecordBuf = NULL;
  readRecordBufSize = 0;
  }
+
+ if (IsUnderPostmaster)
+ BgWriterRecoveryComplete();
  }
 
  /*
***************
*** 5642,5648 ****
   * Log end of a checkpoint.
   */
  static void
! LogCheckpointEnd(void)
  {
  long write_secs,
  sync_secs,
--- 5666,5672 ----
   * Log end of a checkpoint.
   */
  static void
! LogCheckpointEnd(bool checkpoint)
  {
  long write_secs,
  sync_secs,
***************
*** 5665,5673 ****
  CheckpointStats.ckpt_sync_end_t,
  &sync_secs, &sync_usecs);
 
! elog(LOG, "checkpoint complete: wrote %d buffers (%.1f%%); "
  "%d transaction log file(s) added, %d removed, %d recycled; "
  "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
  CheckpointStats.ckpt_bufs_written,
  (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
  CheckpointStats.ckpt_segs_added,
--- 5689,5698 ----
  CheckpointStats.ckpt_sync_end_t,
  &sync_secs, &sync_usecs);
 
! elog(LOG, "%s complete: wrote %d buffers (%.1f%%); "
  "%d transaction log file(s) added, %d removed, %d recycled; "
  "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
+ (checkpoint ? "  checkpoint" : "restartpoint"),
  CheckpointStats.ckpt_bufs_written,
  (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
  CheckpointStats.ckpt_segs_added,
***************
*** 6002,6008 ****
 
  /* All real work is done, but log before releasing lock. */
  if (log_checkpoints)
! LogCheckpointEnd();
 
  LWLockRelease(CheckpointLock);
  }
--- 6027,6033 ----
 
  /* All real work is done, but log before releasing lock. */
  if (log_checkpoints)
! LogCheckpointEnd(true);
 
  LWLockRelease(CheckpointLock);
  }
***************
*** 6071,6099 ****
  }
  }
 
  /*
! * OK, force data out to disk
! */
! CheckPointGuts(checkPoint->redo, CHECKPOINT_IMMEDIATE);
!
! /*
! * Update pg_control so that any subsequent crash will restart from this
! * checkpoint. Note: ReadRecPtr gives the XLOG address of the checkpoint
! * record itself.
  */
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
  ControlFile->checkPoint = ReadRecPtr;
  ControlFile->checkPointCopy = *checkPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
 
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
  checkPoint->redo.xlogid, checkPoint->redo.xrecoff)));
! if (recoveryLastXTime)
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
! (errmsg("last completed transaction was at log time %s",
! timestamptz_to_str(recoveryLastXTime))));
  }
 
  /*
--- 6096,6164 ----
  }
  }
 
+ if (recoveryLastXTime)
+ ereport((log_checkpoints ? LOG : DEBUG2),
+ (errmsg("last completed transaction was at log time %s",
+ timestamptz_to_str(recoveryLastXTime))));
  /*
! * Update ControlFile data in shared memory.
! * Note: ReadRecPtr gives the XLOG address of the checkpoint record itself.
  */
+ LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
  ControlFile->checkPoint = ReadRecPtr;
  ControlFile->checkPointCopy = *checkPoint;
+ RequestRestartPoint();
+ LWLockRelease(ControlFileLock);
+ }
+
+ /*
+  * As of 8.4, RestartPoints are always created by the bgwriter
+  */
+ void
+ CreateRestartPoint(void)
+ {
+ CheckPoint *checkPoint;
+
+ if (log_checkpoints)
+ {
+ /*
+ * Prepare to accumulate statistics.
+ */
+
+ MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
+
+ elog(LOG, "restartpoint starting:");
+ }
+
+ LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
+
+ checkPoint = &ControlFile->checkPointCopy;
+
+ /*
+ * OK, write out dirty blocks smoothly
+ */
+ CheckPointGuts(checkPoint->redo, 0);
+
+ /*
+ * Update pg_control, using current time
+ */
+ LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
+ LWLockRelease(ControlFileLock);
+
+ /* All real work is done, but log before releasing lock. */
+ if (log_checkpoints)
+ LogCheckpointEnd(true);
 
! ereport((log_checkpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
  checkPoint->redo.xlogid, checkPoint->redo.xrecoff)));
!
! LWLockRelease(CheckpointLock);
!
  }
 
  /*
Index: src/backend/postmaster/bgwriter.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/bgwriter.c,v
retrieving revision 1.51
diff -c -r1.51 bgwriter.c
*** src/backend/postmaster/bgwriter.c 11 Aug 2008 11:05:11 -0000 1.51
--- src/backend/postmaster/bgwriter.c 31 Aug 2008 19:45:39 -0000
***************
*** 122,127 ****
--- 122,129 ----
  {
  pid_t bgwriter_pid; /* PID of bgwriter (0 if not started) */
 
+ bool InRedo;
+
  slock_t ckpt_lck; /* protects all the ckpt_* fields */
 
  int ckpt_started; /* advances when checkpoint starts */
***************
*** 166,172 ****
 
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr;
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
--- 168,174 ----
 
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr; /* not used if InRedo */
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
***************
*** 186,191 ****
--- 188,208 ----
  static void ReqCheckpointHandler(SIGNAL_ARGS);
  static void ReqShutdownHandler(SIGNAL_ARGS);
 
+ /* notify bgwriter of change in mode */
+ void
+ BgWriterRecoveryComplete(void)
+ {
+ BgWriterShmem->InRedo = false;
+ elog(DEBUG1, "recovery complete");
+ }
+
+ /* ask bgwriter to complete any restartpoint, if any, with zero delay */
+ void
+ BgWriterCompleteRestartPointImmediately(void)
+ {
+ BgWriterShmem->ckpt_flags = CHECKPOINT_IMMEDIATE;
+ elog(DEBUG2, "asking bgwriter to complete any restartpoint with zero delay");
+ }
 
  /*
   * Main entry point for bgwriter process
***************
*** 202,207 ****
--- 219,230 ----
  BgWriterShmem->bgwriter_pid = MyProcPid;
  am_bg_writer = true;
 
+ /*
+ * Follow the postmaster's current mode at startup. If we are InRedo then
+ * the startup process will later tell us when it is complete.
+ */
+ BgWriterShmem->InRedo = InRedo;
+
  /*
  * If possible, make this process a group leader, so that the postmaster
  * can signal any child processes too. (bgwriter probably never has any
***************
*** 356,371 ****
  */
  PG_SETMASK(&UnBlockSig);
 
  /*
  * Loop forever
  */
  for (;;)
  {
- bool do_checkpoint = false;
- int flags = 0;
- pg_time_t now;
- int elapsed_secs;
-
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
--- 379,393 ----
  */
  PG_SETMASK(&UnBlockSig);
 
+ if (InRedo)
+ elog(DEBUG1, "bgwriter starting in recovery mode, pid = %u",
+ BgWriterShmem->bgwriter_pid);
+
  /*
  * Loop forever
  */
  for (;;)
  {
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
***************
*** 383,501 ****
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
- if (checkpoint_requested)
- {
- checkpoint_requested = false;
- do_checkpoint = true;
- BgWriterStats.m_requested_checkpoints++;
- }
- if (shutdown_requested)
- {
- /*
- * From here on, elog(ERROR) should end with exit(1), not send
- * control back to the sigsetjmp block above
- */
- ExitOnAnyError = true;
- /* Close down the database */
- ShutdownXLOG(0, 0);
- DumpFreeSpaceMap(0, 0);
- /* Normal exit from the bgwriter is here */
- proc_exit(0); /* done */
- }
-
- /*
- * Force a checkpoint if too much time has elapsed since the last one.
- * Note that we count a timed checkpoint in stats only when this
- * occurs without an external request, but we set the CAUSE_TIME flag
- * bit even if there is also an external request.
- */
- now = (pg_time_t) time(NULL);
- elapsed_secs = now - last_checkpoint_time;
- if (elapsed_secs >= CheckPointTimeout)
- {
- if (!do_checkpoint)
- BgWriterStats.m_timed_checkpoints++;
- do_checkpoint = true;
- flags |= CHECKPOINT_CAUSE_TIME;
- }
 
! /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
! */
! if (do_checkpoint)
  {
- /* use volatile pointer to prevent code rearrangement */
- volatile BgWriterShmemStruct *bgs = BgWriterShmem;
-
  /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
  */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
 
! /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
! */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
 
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
 
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
 
  /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
  */
! smgrcloseall();
 
  /*
! * Indicate checkpoint completion to any waiting backends.
  */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
 
  /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
  */
! last_checkpoint_time = now;
! }
! else
! BgBufferSync();
 
! /* Check for archive_timeout and switch xlog files if necessary. */
! CheckArchiveTimeout();
 
! /* Nap for the configured time. */
! BgWriterNap();
  }
  }
 
--- 405,610 ----
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
 
! if (InRedo)
  {
  /*
! * Check to see whether startup process has completed redo.
! * If so, we can permanently change out of recovery mode.
  */
! if (BgWriterShmem->InRedo == false)
! {
 
! elog(LOG, "changing to InRedo = false");
 
! InitXLOGAccess();
! InRedo = false;
 
! /*
! * Start time-driven events from now
! */
! last_checkpoint_time = last_xlog_switch_time = (pg_time_t) time(NULL);
! }
!
! if (checkpoint_requested)
! {
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_time = (pg_time_t) time(NULL);
! ckpt_cached_elapsed = 0;
!
! CreateRestartPoint();
!
! ckpt_active = false;
! checkpoint_requested = false;
! /*
! * Reset any flags if we requested immediate completion part
! * way through the restart point
! */
! BgWriterShmem->ckpt_flags = 0;
! }
! else
! {
! /* Clean buffers dirtied by recovery */
! BgBufferSync();
!
! /* Nap for the configured time. */
! BgWriterNap();
! }
!
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
 
  /*
! * Check to see whether startup process has completed redo.
! * If so, we can permanently change out of recovery mode.
  */
! if (BgWriterShmem->InRedo == false)
! {
! elog(DEBUG2, "changing to InRedo = false");
!
! InitXLOGAccess();
! InRedo = false;
!
! /*
! * Start time-driven events from now
! */
! last_checkpoint_time = last_xlog_switch_time = (pg_time_t) time(NULL);
! }
! }
! else /* Normal processing */
! {
! bool do_checkpoint = false;
! int flags = 0;
! pg_time_t now;
! int elapsed_secs;
!
! Assert(!InRedo);
!
! if (checkpoint_requested)
! {
! checkpoint_requested = false;
! do_checkpoint = true;
! BgWriterStats.m_requested_checkpoints++;
! }
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Close down the database */
! ShutdownXLOG(0, 0);
! DumpFreeSpaceMap(0, 0);
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
 
  /*
! * Force a checkpoint if too much time has elapsed since the last one.
! * Note that we count a timed checkpoint in stats only when this
! * occurs without an external request, but we set the CAUSE_TIME flag
! * bit even if there is also an external request.
  */
! now = (pg_time_t) time(NULL);
! elapsed_secs = now - last_checkpoint_time;
! if (elapsed_secs >= CheckPointTimeout)
! {
! if (!do_checkpoint)
! BgWriterStats.m_timed_checkpoints++;
! do_checkpoint = true;
! flags |= CHECKPOINT_CAUSE_TIME;
! }
 
  /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
  */
! if (do_checkpoint)
! {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
!
! /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
!
! /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
! */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
!
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
!
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
!
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
!
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
! }
! else
! BgBufferSync();
 
! /* Check for archive_timeout and switch xlog files if necessary. */
! CheckArchiveTimeout();
 
! /* Nap for the configured time. */
! BgWriterNap();
! }
  }
  }
 
***************
*** 588,594 ****
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
--- 697,704 ----
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! if (!InRedo)
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
***************
*** 642,647 ****
--- 752,770 ----
  if (!am_bg_writer)
  return;
 
+ /* Perform minimal duties during recovery and skip wait if requested */
+ if (InRedo)
+ {
+ BgBufferSync();
+
+ if (!ImmediateCheckpointRequested() &&
+ !shutdown_requested &&
+ IsCheckpointOnSchedule(progress))
+ BgWriterNap();
+
+ return;
+ }
+
  /*
  * Perform the usual bgwriter duties and take a nap, unless we're behind
  * schedule, in which case we just try to catch up as quickly as possible.
***************
*** 716,731 ****
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
  {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
  }
 
  /*
--- 839,857 ----
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! if (!InRedo)
  {
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
! {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
! }
  }
 
  /*
***************
*** 966,971 ****
--- 1092,1118 ----
  }
  }
 
+ void
+ RequestRestartPoint(void)
+ {
+ /*
+ * If in a standalone backend, just do it ourselves.
+ */
+ if (!IsPostmasterEnvironment)
+ {
+ CreateRestartPoint();
+ return;
+ }
+
+ /*
+ * Send signal to request restartpoint.
+ */
+ if (BgWriterShmem->bgwriter_pid == 0)
+ elog(LOG, "could not request restartpoint because bgwriter not running");
+ if (kill(BgWriterShmem->bgwriter_pid, SIGINT) != 0)
+ elog(LOG, "could not signal for restartpoint: %m");
+ }
+
  /*
   * ForwardFsyncRequest
   * Forward a file-fsync request from a backend to the bgwriter
Index: src/backend/postmaster/postmaster.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/postmaster.c,v
retrieving revision 1.561
diff -c -r1.561 postmaster.c
*** src/backend/postmaster/postmaster.c 26 Jun 2008 02:47:19 -0000 1.561
--- src/backend/postmaster/postmaster.c 31 Aug 2008 17:10:31 -0000
***************
*** 254,259 ****
--- 254,264 ----
  {
  PM_INIT, /* postmaster starting */
  PM_STARTUP, /* waiting for startup subprocess */
+ PM_RECOVERY, /* consistent recovery mode; state only
+ * entered for archive and streaming recovery,
+ * and only after the point where the
+ * all data is in consistent state.
+ */
  PM_RUN, /* normal "database is alive" state */
  PM_WAIT_BACKUP, /* waiting for online backup mode to end */
  PM_WAIT_BACKENDS, /* waiting for live backends to exit */
***************
*** 2104,2110 ****
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
--- 2109,2115 ----
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP || pmState == PM_RECOVERY);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
***************
*** 2136,2141 ****
--- 2141,2147 ----
  * Otherwise, commence normal operations.
  */
  pmState = PM_RUN;
+ InRedo = false;
 
  /*
  * Load the flat authorization file into postmaster's cache. The
***************
*** 2148,2155 ****
  * Crank up the background writer. It doesn't matter if this
  * fails, we'll just try again later.
  */
! Assert(BgWriterPID == 0);
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
--- 2154,2161 ----
  * Crank up the background writer. It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0)
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
***************
*** 2812,2817 ****
--- 2818,2825 ----
  */
  MyCancelKey = PostmasterRandom();
 
+ InRedo = (pmState != PM_RUN);
+
  /*
  * Make room for backend data structure.  Better before the fork() so we
  * can handle failure cleanly.
***************
*** 3821,3826 ****
--- 3829,3880 ----
 
  PG_SETMASK(&BlockSig);
 
+ if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_START))
+ {
+ Assert(pmState == PM_STARTUP);
+
+ /*
+ * Go to shutdown mode if a shutdown request was pending.
+ */
+ if (Shutdown > NoShutdown)
+ {
+ pmState = PM_WAIT_BACKENDS;
+ /* PostmasterStateMachine logic does the rest */
+ }
+ else
+ {
+ /*
+ * Startup process has entered recovery
+ */
+ pmState = PM_RECOVERY;
+ InRedo = true;
+
+ /*
+ * Load the flat authorization file into postmaster's cache. The
+ * startup process won't have recomputed this from the database yet,
+ * so we it may change following recovery.
+ */
+ load_role();
+
+ /*
+ * Crank up the background writer. It doesn't matter if this
+ * fails, we'll just try again later.
+ */
+ Assert(BgWriterPID == 0);
+ BgWriterPID = StartBackgroundWriter();
+
+ /*
+ * Likewise, start other special children as needed.
+ */
+ Assert(PgStatPID == 0);
+ PgStatPID = pgstat_start();
+
+ /* XXX at this point we could accept read-only connections */
+ ereport(DEBUG1,
+ (errmsg("database system is in consistent recovery mode")));
+ }
+ }
+
  if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
  {
  /*
Index: src/include/miscadmin.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/miscadmin.h,v
retrieving revision 1.202
diff -c -r1.202 miscadmin.h
*** src/include/miscadmin.h 23 Apr 2008 13:44:59 -0000 1.202
--- src/include/miscadmin.h 31 Aug 2008 11:23:45 -0000
***************
*** 64,69 ****
--- 64,71 ----
   *
   *****************************************************************************/
 
+ extern bool InRedo;
+
  /* in globals.c */
  /* these are marked volatile because they are set by signal handlers: */
  extern PGDLLIMPORT volatile bool InterruptPending;
Index: src/include/access/xlog.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog.h,v
retrieving revision 1.88
diff -c -r1.88 xlog.h
*** src/include/access/xlog.h 12 May 2008 08:35:05 -0000 1.88
--- src/include/access/xlog.h 31 Aug 2008 13:33:43 -0000
***************
*** 205,210 ****
--- 205,211 ----
  extern void ShutdownXLOG(int code, Datum arg);
  extern void InitXLOGAccess(void);
  extern void CreateCheckPoint(int flags);
+ extern void CreateRestartPoint(void);
  extern void XLogPutNextOid(Oid nextOid);
  extern XLogRecPtr GetRedoRecPtr(void);
  extern XLogRecPtr GetInsertRecPtr(void);
Index: src/include/postmaster/bgwriter.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/postmaster/bgwriter.h,v
retrieving revision 1.12
diff -c -r1.12 bgwriter.h
*** src/include/postmaster/bgwriter.h 11 Aug 2008 11:05:11 -0000 1.12
--- src/include/postmaster/bgwriter.h 31 Aug 2008 15:35:30 -0000
***************
*** 25,36 ****
--- 25,40 ----
  extern void BackgroundWriterMain(void);
 
  extern void RequestCheckpoint(int flags);
+ extern void RequestRestartPoint(void);
  extern void CheckpointWriteDelay(int flags, double progress);
 
  extern bool ForwardFsyncRequest(RelFileNode rnode, ForkNumber forknum,
  BlockNumber segno);
  extern void AbsorbFsyncRequests(void);
 
+ extern void BgWriterRecoveryComplete(void);
+ extern void BgWriterCompleteRestartPointImmediately(void);
+
  extern Size BgWriterShmemSize(void);
  extern void BgWriterShmemInit(void);
 
Index: src/include/storage/pmsignal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/storage/pmsignal.h,v
retrieving revision 1.20
diff -c -r1.20 pmsignal.h
*** src/include/storage/pmsignal.h 19 Jun 2008 21:32:56 -0000 1.20
--- src/include/storage/pmsignal.h 31 Aug 2008 11:26:40 -0000
***************
*** 22,27 ****
--- 22,28 ----
   */
  typedef enum
  {
+ PMSIGNAL_RECOVERY_START, /* move to PM_RECOVERY state */
  PMSIGNAL_PASSWORD_CHANGE, /* pg_auth file has changed */
  PMSIGNAL_WAKEN_ARCHIVER, /* send a NOTIFY signal to xlog archiver */
  PMSIGNAL_ROTATE_LOGFILE, /* send SIGUSR1 to syslogger to rotate logfile */

---

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Tom Lane-2]

Simon Riggs <simon@...> writes:
> Included patch with the following changes:

> * new postmaster mode known as consistent recovery, entered only when
> recovery passes safe/consistent point. InRedo is now set in all
> processes when started/connected in consistent recovery mode.

I looked at this and didn't like the InRedo signaling at all.  In the
first place, it looks like you're expecting the flag to be passed down
via fork(), but that won't work in EXEC_BACKEND mode.  (Yes, easily
fixed, but it's wrong as-is.)  In the second place, the method of
signaling it to the bgwriter looks to have race conditions: in
principle, at least, I think the startup process could try to clear
the shared-memory InRedo flag before the bgwriter has gotten as far as
setting it.  (This might work if the startup process can't exit redo
mode without the bgwriter having finished a checkpoint, but such a
dependency is too rube goldbergian for me.)

ISTM that it would probably be better if there were exactly one InRedo
flag in shared memory, probably in xlog.c's shared state, with the
postmaster not being responsible for setting or clearing it; rather
the startup process should do those things.

> * bgwriter and stats process starts in consistent recovery mode.
> bgwriter changes mode when startup process completes.

I'm not sure about the interaction of this.  In particular, what about
recovery restart points before we have reached the safe stop point?
I don't think we want to give up the capability of having those.

Also, it seems pretty bogus to update the in-memory ControlFile
checkpoint values before the restart point is actually done.  It looks
to me like what you have done is to try to use those fields as signaling
for the restart request in addition to their existing purposes, which
I think is confusing and probably dangerous.  I'd rather there were a
different signaling path and ControlFile maintains its currrent
definition.

I found the changes in bgwriter.c unreadable.  You've evidently
moved blocks of code around, but exactly what did you mean to
change?  Why is there so much duplicated code now?

> I've kept the distinction between restartpoints and checkpoints in
> code, to avoid convoluted code.

Hmm, I dunno, it seems like that might be a bad choice.  Are you sure
it's not cleaner to just use the regular checkpoint code?

                        regards, tom lane

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Mon, 2008-09-08 at 13:34 -0400, Tom Lane wrote:

> Simon Riggs <simon@...> writes:
> > Included patch with the following changes:
>
> > * new postmaster mode known as consistent recovery, entered only when
> > recovery passes safe/consistent point. InRedo is now set in all
> > processes when started/connected in consistent recovery mode.
>
> I looked at this and didn't like the InRedo signaling at all.  In the
> first place, it looks like you're expecting the flag to be passed down
> via fork(), but that won't work in EXEC_BACKEND mode.  (Yes, easily
> fixed, but it's wrong as-is.)

OK, thanks. I didn't check that.

>  In the second place, the method of
> signaling it to the bgwriter looks to have race conditions: in
> principle, at least, I think the startup process could try to clear
> the shared-memory InRedo flag before the bgwriter has gotten as far as
> setting it.  (This might work if the startup process can't exit redo
> mode without the bgwriter having finished a checkpoint, but such a
> dependency is too rube goldbergian for me.)

We never interrupt what the bgwriter does. If it is in the middle of
doing a checkpoint when recovery finishes, then Startup process just
waits behind it and then issues a shutdown checkpoint. If there's loads
of blocks to be written it doesn't matter who writes them. There's only
one soup spoon, and it doesn't matter who stirs the soup.

> ISTM that it would probably be better if there were exactly one InRedo
> flag in shared memory, probably in xlog.c's shared state, with the
> postmaster not being responsible for setting or clearing it; rather
> the startup process should do those things.

So replace InRedo with an IsInRedo() function that accesses XLogCtl?

Sounds good

> > * bgwriter and stats process starts in consistent recovery mode.
> > bgwriter changes mode when startup process completes.
>
> I'm not sure about the interaction of this.  In particular, what about
> recovery restart points before we have reached the safe stop point?
> I don't think we want to give up the capability of having those.

Maybe. I felt it made the code cleaner to give that up. Realistically
its a fairly short window of time. But shouldn't be too hard to put
back.

> Also, it seems pretty bogus to update the in-memory ControlFile
> checkpoint values before the restart point is actually done.  It looks
> to me like what you have done is to try to use those fields as signaling
> for the restart request in addition to their existing purposes, which
> I think is confusing and probably dangerous.  I'd rather there were a
> different signaling path and ControlFile maintains its currrent
> definition.

OK

> I found the changes in bgwriter.c unreadable.  You've evidently
> moved blocks of code around, but exactly what did you mean to
> change?  Why is there so much duplicated code now?

The patch utility did that. Some code reuse confused it. It's real easy
to read though if you apply the patch and then read the main loop.
You'll see what I mean.

> > I've kept the distinction between restartpoints and checkpoints in
> > code, to avoid convoluted code.
>
> Hmm, I dunno, it seems like that might be a bad choice.  Are you sure
> it's not cleaner to just use the regular checkpoint code?

When I tried to write it, it just looked to my eyes like every single
line had a caveat which looked ugly and multiplied the testing. You're
the code dude, always happy to structure things as you suggest. If
you're sure, that is.

Thanks for the review.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support


--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Tom Lane-2]

Simon Riggs <simon@...> writes:
> On Mon, 2008-09-08 at 13:34 -0400, Tom Lane wrote:
>> Hmm, I dunno, it seems like that might be a bad choice.  Are you sure
>> it's not cleaner to just use the regular checkpoint code?

> When I tried to write it, it just looked to my eyes like every single
> line had a caveat which looked ugly and multiplied the testing. You're
> the code dude, always happy to structure things as you suggest. If
> you're sure, that is.

No, was just wondering if the other way would be better.  If you're
sure not, that's fine.

                        regards, tom lane

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Mon, 2008-09-08 at 13:34 -0400, Tom Lane wrote:

> ISTM that it would probably be better if there were exactly one InRedo
> flag in shared memory, probably in xlog.c's shared state, with the
> postmaster not being responsible for setting or clearing it; rather
> the startup process should do those things.

Done

> > * bgwriter and stats process starts in consistent recovery mode.
> > bgwriter changes mode when startup process completes.
>
> I'm not sure about the interaction of this.  In particular, what about
> recovery restart points before we have reached the safe stop point?
> I don't think we want to give up the capability of having those.
>
> Also, it seems pretty bogus to update the in-memory ControlFile
> checkpoint values before the restart point is actually done.  It looks
> to me like what you have done is to try to use those fields as signaling
> for the restart request in addition to their existing purposes, which
> I think is confusing and probably dangerous.  I'd rather there were a
> different signaling path and ControlFile maintains its currrent
> definition.

Done


Testing takes a while on this, I probably won't complete it until
Friday. So enclosed patch is for eyeballs only at this stage.

I added in the XLogCtl padding we've discussed before, while I'm there.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support

[recovery_infrastruc.v3.patch]

---

Index: src/backend/access/transam/xlog.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/access/transam/xlog.c,v
retrieving revision 1.317
diff -c -r1.317 xlog.c
*** src/backend/access/transam/xlog.c 11 Aug 2008 11:05:10 -0000 1.317
--- src/backend/access/transam/xlog.c 10 Sep 2008 20:48:52 -0000
***************
*** 119,124 ****
--- 119,125 ----
 
  /* Are we doing recovery from XLOG? */
  bool InRecovery = false;
+ bool reachedSafeStopPoint = false;
 
  /* Are we recovering using offline XLOG archives? */
  static bool InArchiveRecovery = false;
***************
*** 131,137 ****
  static bool recoveryTarget = false;
  static bool recoveryTargetExact = false;
  static bool recoveryTargetInclusive = true;
- static bool recoveryLogRestartpoints = false;
  static TransactionId recoveryTargetXid;
  static TimestampTz recoveryTargetTime;
  static TimestampTz recoveryLastXTime = 0;
--- 132,137 ----
***************
*** 286,295 ****
--- 286,297 ----
  /*
   * Total shared-memory state for XLOG.
   */
+ #define XLOGCTL_BUFFER_SPACING 128
  typedef struct XLogCtlData
  {
  /* Protected by WALInsertLock: */
  XLogCtlInsert Insert;
+ char InsertPadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogCtlInsert)];
 
  /* Protected by info_lck: */
  XLogwrtRqst LogwrtRqst;
***************
*** 297,305 ****
--- 299,314 ----
  uint32 ckptXidEpoch; /* nextXID & epoch of latest checkpoint */
  TransactionId ckptXid;
  XLogRecPtr asyncCommitLSN; /* LSN of newest async commit */
+ /* add data structure padding for above info_lck declarations */
+ char InfoPadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogwrtRqst)
+ - sizeof(XLogwrtResult)
+ - sizeof(uint32)
+ - sizeof(TransactionId)
+ - sizeof(XLogRecPtr)];
 
  /* Protected by WALWriteLock: */
  XLogCtlWrite Write;
+ char WritePadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogCtlWrite)];
 
  /*
  * These values do not change after startup, although the pointed-to pages
***************
*** 311,316 ****
--- 320,336 ----
  int XLogCacheBlck; /* highest allocated xlog buffer index */
  TimeLineID ThisTimeLineID;
 
+ /*
+ * InStartupXLog changes at the end of recovery only and is only ever set
+ * by the Startup process. We assume that changes to it are atomic,
+ * so accesses to it is never locked. When it does change bgwriter
+ * must immediately begin using it, since this helps it decide whether
+ * to flush WAL or not when it writes dirty blocks. If bgwriter does
+ * it too soon, we will write invalid WAL records and if it reflects the
+ * change too late it could skip flushing WAL for a data block change.
+ */
+ bool InStartupXLog;
+
  slock_t info_lck; /* locks shared variables shown above */
  } XLogCtlData;
 
***************
*** 480,485 ****
--- 500,509 ----
  bool doPageWrites;
  bool isLogSwitch = (rmid == RM_XLOG_ID && info == XLOG_SWITCH);
 
+ /* cross-check on whether we should be here or not */
+ if (InStartupXLog())
+ elog(FATAL, "cannot write new WAL data during recovery mode");
+
  /* info's high bits are reserved for use by me */
  if (info & XLR_INFO_MASK)
  elog(PANIC, "invalid xlog info mask %02X", info);
***************
*** 1677,1684 ****
  XLogRecPtr WriteRqstPtr;
  XLogwrtRqst WriteRqst;
 
! /* Disabled during REDO */
! if (InRedo)
  return;
 
  /* Quick exit if already known flushed */
--- 1701,1708 ----
  XLogRecPtr WriteRqstPtr;
  XLogwrtRqst WriteRqst;
 
! /* Disabled during StartupXLog */
! if (InStartupXLog())
  return;
 
  /* Quick exit if already known flushed */
***************
*** 1766,1774 ****
  * the bad page is encountered again during recovery then we would be
  * unable to restart the database at all!  (This scenario has actually
  * happened in the field several times with 7.1 releases. Note that we
! * cannot get here while InRedo is true, but if the bad page is brought in
! * and marked dirty during recovery then CreateCheckPoint will try to
! * flush it at the end of recovery.)
  *
  * The current approach is to ERROR under normal conditions, but only
  * WARNING during recovery, so that the system can be brought up even if
--- 1790,1798 ----
  * the bad page is encountered again during recovery then we would be
  * unable to restart the database at all!  (This scenario has actually
  * happened in the field several times with 7.1 releases. Note that we
! * cannot get here while InStartupXLog() is true, but if the bad page is
! * brought in and marked dirty during recovery then CreateCheckPoint will
! * try to flush it at the end of recovery.)
  *
  * The current approach is to ERROR under normal conditions, but only
  * WARNING during recovery, so that the system can be brought up even if
***************
*** 2051,2057 ****
  unlink(tmppath);
  }
 
! elog(DEBUG2, "done creating and filling new WAL file");
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
--- 2075,2082 ----
  unlink(tmppath);
  }
 
! XLogFileName(tmppath, ThisTimeLineID, log, seg);
! elog(DEBUG2, "done creating and filling new WAL file %s", tmppath);
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
***************
*** 4532,4546 ****
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
- /*
- * does nothing if a recovery_target is not also set
- */
- if (!parse_bool(tok2, &recoveryLogRestartpoints))
-  ereport(ERROR,
- (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-  errmsg("parameter \"log_restartpoints\" requires a Boolean value")));
  ereport(LOG,
! (errmsg("log_restartpoints = %s", tok2)));
  }
  else
  ereport(FATAL,
--- 4557,4565 ----
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
  ereport(LOG,
! (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
!  errmsg("parameter \"log_restartpoints\" has been deprecated")));
  }
  else
  ereport(FATAL,
***************
*** 4823,4828 ****
--- 4842,4849 ----
  uint32 freespace;
  TransactionId oldestActiveXID;
 
+ XLogCtl->InStartupXLog = true;
+
  /*
  * Read control file and check XLOG status looks valid.
  *
***************
*** 5039,5044 ****
--- 5060,5070 ----
  UpdateControlFile();
 
  /*
+ * Reset pgstat data, because it may be invalid after recovery.
+ */
+ pgstat_reset_all();
+
+ /*
  * If there was a backup label file, it's done its job and the info
  * has now been propagated into pg_control.  We must get rid of the
  * label file so that if we crash during recovery, we'll pick up at
***************
*** 5148,5153 ****
--- 5174,5194 ----
 
  LastRec = ReadRecPtr;
 
+ /*
+ * Have we reached our safe stopping point? If so, we can
+ * signal Postmaster to enter consistent recovery mode
+ */
+ if (!reachedSafeStopPoint &&
+ XLByteLE(ControlFile->minRecoveryPoint, EndRecPtr))
+ {
+ reachedSafeStopPoint = true;
+ ereport(LOG,
+ (errmsg("consistent recovery state reached at %X/%X",
+ EndRecPtr.xlogid, EndRecPtr.xrecoff)));
+ if (IsUnderPostmaster)
+ SendPostmasterSignal(PMSIGNAL_RECOVERY_START);
+ }
+
  record = ReadRecord(NULL, LOG);
  } while (record != NULL && recoveryContinue);
 
***************
*** 5169,5174 ****
--- 5210,5216 ----
  /* there are no WAL records following the checkpoint */
  ereport(LOG,
  (errmsg("redo is not required")));
+ reachedSafeStopPoint = true;
  }
  }
 
***************
*** 5184,5190 ****
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (XLByteLT(EndOfLog, ControlFile->minRecoveryPoint))
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
--- 5226,5232 ----
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (InRecovery && !reachedSafeStopPoint)
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
***************
*** 5305,5314 ****
  */
  XLogCheckInvalidPages();
 
! /*
! * Reset pgstat data, because it may be invalid after recovery.
! */
! pgstat_reset_all();
 
  /*
  * Perform a checkpoint to update all our recovery activity to disk.
--- 5347,5354 ----
  */
  XLogCheckInvalidPages();
 
! if (IsUnderPostmaster)
! BgWriterCompleteRestartPointImmediately();
 
  /*
  * Perform a checkpoint to update all our recovery activity to disk.
***************
*** 5318,5323 ****
--- 5358,5366 ----
  * assigning a new TLI, using a shutdown checkpoint allows us to have
  * the rule that TLI only changes in shutdown checkpoints, which
  * allows some extra error checking in xlog_redo.
+ *
+ * Note that this will wait behind any restartpoint that the bgwriter
+ * is currently performing, though will be much faster as a result.
  */
  CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
  }
***************
*** 5372,5377 ****
--- 5415,5428 ----
  readRecordBuf = NULL;
  readRecordBufSize = 0;
  }
+
+ XLogCtl->InStartupXLog = false;
+ }
+
+ bool
+ InStartupXLog(void)
+ {
+ return XLogCtl->InStartupXLog;
  }
 
  /*
***************
*** 5642,5648 ****
   * Log end of a checkpoint.
   */
  static void
! LogCheckpointEnd(void)
  {
  long write_secs,
  sync_secs,
--- 5693,5699 ----
   * Log end of a checkpoint.
   */
  static void
! LogCheckpointEnd(bool checkpoint)
  {
  long write_secs,
  sync_secs,
***************
*** 5665,5673 ****
  CheckpointStats.ckpt_sync_end_t,
  &sync_secs, &sync_usecs);
 
! elog(LOG, "checkpoint complete: wrote %d buffers (%.1f%%); "
  "%d transaction log file(s) added, %d removed, %d recycled; "
  "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
  CheckpointStats.ckpt_bufs_written,
  (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
  CheckpointStats.ckpt_segs_added,
--- 5716,5725 ----
  CheckpointStats.ckpt_sync_end_t,
  &sync_secs, &sync_usecs);
 
! elog(LOG, "%s complete: wrote %d buffers (%.1f%%); "
  "%d transaction log file(s) added, %d removed, %d recycled; "
  "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
+ (checkpoint ? "  checkpoint" : "restartpoint"),
  CheckpointStats.ckpt_bufs_written,
  (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
  CheckpointStats.ckpt_segs_added,
***************
*** 6002,6008 ****
 
  /* All real work is done, but log before releasing lock. */
  if (log_checkpoints)
! LogCheckpointEnd();
 
  LWLockRelease(CheckpointLock);
  }
--- 6054,6060 ----
 
  /* All real work is done, but log before releasing lock. */
  if (log_checkpoints)
! LogCheckpointEnd(true);
 
  LWLockRelease(CheckpointLock);
  }
***************
*** 6071,6099 ****
  }
  }
 
  /*
! * OK, force data out to disk
  */
! CheckPointGuts(checkPoint->redo, CHECKPOINT_IMMEDIATE);
 
  /*
! * Update pg_control so that any subsequent crash will restart from this
! * checkpoint. Note: ReadRecPtr gives the XLOG address of the checkpoint
! * record itself.
  */
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
! ControlFile->checkPoint = ReadRecPtr;
! ControlFile->checkPointCopy = *checkPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
 
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
! checkPoint->redo.xlogid, checkPoint->redo.xrecoff)));
! if (recoveryLastXTime)
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
! (errmsg("last completed transaction was at log time %s",
! timestamptz_to_str(recoveryLastXTime))));
  }
 
  /*
--- 6123,6184 ----
  }
  }
 
+ if (recoveryLastXTime)
+ ereport((log_checkpoints ? LOG : DEBUG2),
+ (errmsg("last completed transaction was at log time %s",
+ timestamptz_to_str(recoveryLastXTime))));
+
+ RequestRestartPoint(ReadRecPtr, checkPoint, reachedSafeStopPoint);
+ }
+
+ /*
+  * As of 8.4, RestartPoints are always created by the bgwriter
+  * once we have reachedSafeStopPoint. We use bgwriter's shared memory
+  * area wherever we call it from, to keep better code structure.
+  */
+ void
+ CreateRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint)
+ {
+ if (log_checkpoints)
+ {
+ /*
+ * Prepare to accumulate statistics.
+ */
+
+ MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
+
+ elog(LOG, "restartpoint starting:");
+ }
+
+ LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
+
  /*
! * OK, write out dirty blocks smoothly
  */
! CheckPointGuts(restartPoint->redo, 0);
 
  /*
! * Update pg_control, using current time
  */
+ LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
! ControlFile->checkPoint = ReadPtr;
! ControlFile->checkPointCopy = *restartPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
+ LWLockRelease(ControlFileLock);
 
! /* All real work is done, but log before releasing lock. */
! if (log_checkpoints)
! LogCheckpointEnd(true);
!
! ereport((log_checkpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
! restartPoint->redo.xlogid, restartPoint->redo.xrecoff)));
!
! LWLockRelease(CheckpointLock);
!
  }
 
  /*
Index: src/backend/postmaster/bgwriter.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/bgwriter.c,v
retrieving revision 1.51
diff -c -r1.51 bgwriter.c
*** src/backend/postmaster/bgwriter.c 11 Aug 2008 11:05:11 -0000 1.51
--- src/backend/postmaster/bgwriter.c 10 Sep 2008 20:53:07 -0000
***************
*** 49,54 ****
--- 49,55 ----
  #include <unistd.h>
 
  #include "access/xlog_internal.h"
+ #include "catalog/pg_control.h"
  #include "libpq/pqsignal.h"
  #include "miscadmin.h"
  #include "pgstat.h"
***************
*** 130,135 ****
--- 131,144 ----
 
  int ckpt_flags; /* checkpoint flags, as defined in xlog.h */
 
+ /*
+ * When the Startup process wants a restartpoint, it sets these fields
+ * so that whoever performs the restartpoint can update the control file,
+ * allowing the caller to continue, if it is running in another process.
+ */
+ XLogRecPtr ReadPtr; /* ReadRecPtr for RestartPoint request */
+ CheckPoint *restartPoint; /* restartPoint data for ControlFile */
+
  uint32 num_backend_writes; /* counts non-bgwriter buffer writes */
 
  int num_requests; /* current # of requests */
***************
*** 164,172 ****
 
  static bool ckpt_active = false;
 
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr;
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
--- 173,184 ----
 
  static bool ckpt_active = false;
 
+ static bool BgWriterInStartup;
+
+
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr; /* not used if InStartupXLog */
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
***************
*** 186,191 ****
--- 198,212 ----
  static void ReqCheckpointHandler(SIGNAL_ARGS);
  static void ReqShutdownHandler(SIGNAL_ARGS);
 
+ /* ask bgwriter to complete any restartpoint, if any, with zero delay */
+ void
+ BgWriterCompleteRestartPointImmediately(void)
+ {
+ SpinLockAcquire(&BgWriterShmem->ckpt_lck);
+ BgWriterShmem->ckpt_flags = CHECKPOINT_IMMEDIATE;
+ SpinLockRelease(&BgWriterShmem->ckpt_lck);
+ elog(DEBUG2, "asking bgwriter to complete any restartpoint with zero delay");
+ }
 
  /*
   * Main entry point for bgwriter process
***************
*** 202,207 ****
--- 223,230 ----
  BgWriterShmem->bgwriter_pid = MyProcPid;
  am_bg_writer = true;
 
+ BgWriterInStartup = InStartupXLog();
+
  /*
  * If possible, make this process a group leader, so that the postmaster
  * can signal any child processes too. (bgwriter probably never has any
***************
*** 356,371 ****
  */
  PG_SETMASK(&UnBlockSig);
 
  /*
  * Loop forever
  */
  for (;;)
  {
- bool do_checkpoint = false;
- int flags = 0;
- pg_time_t now;
- int elapsed_secs;
-
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
--- 379,393 ----
  */
  PG_SETMASK(&UnBlockSig);
 
+ if (BgWriterInStartup)
+ elog(DEBUG1, "bgwriter starting in recovery mode, pid = %u",
+ BgWriterShmem->bgwriter_pid);
+
  /*
  * Loop forever
  */
  for (;;)
  {
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
***************
*** 383,501 ****
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
- if (checkpoint_requested)
- {
- checkpoint_requested = false;
- do_checkpoint = true;
- BgWriterStats.m_requested_checkpoints++;
- }
- if (shutdown_requested)
- {
- /*
- * From here on, elog(ERROR) should end with exit(1), not send
- * control back to the sigsetjmp block above
- */
- ExitOnAnyError = true;
- /* Close down the database */
- ShutdownXLOG(0, 0);
- DumpFreeSpaceMap(0, 0);
- /* Normal exit from the bgwriter is here */
- proc_exit(0); /* done */
- }
-
- /*
- * Force a checkpoint if too much time has elapsed since the last one.
- * Note that we count a timed checkpoint in stats only when this
- * occurs without an external request, but we set the CAUSE_TIME flag
- * bit even if there is also an external request.
- */
- now = (pg_time_t) time(NULL);
- elapsed_secs = now - last_checkpoint_time;
- if (elapsed_secs >= CheckPointTimeout)
- {
- if (!do_checkpoint)
- BgWriterStats.m_timed_checkpoints++;
- do_checkpoint = true;
- flags |= CHECKPOINT_CAUSE_TIME;
- }
 
! /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
! */
! if (do_checkpoint)
  {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
!
! /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
 
  /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
  */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
 
  /*
! * Do the checkpoint.
  */
! CreateCheckPoint(flags);
 
  /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
  */
! smgrcloseall();
 
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
 
! ckpt_active = false;
!
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
  }
- else
- BgBufferSync();
-
- /* Check for archive_timeout and switch xlog files if necessary. */
- CheckArchiveTimeout();
-
- /* Nap for the configured time. */
- BgWriterNap();
  }
  }
 
--- 405,607 ----
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
 
! if (BgWriterInStartup)
  {
! if (checkpoint_requested)
! {
! XLogRecPtr ReadPtr;
! CheckPoint restartPoint;
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_time = (pg_time_t) time(NULL);
! ckpt_cached_elapsed = 0;
!
! /*
! * Get the requested values from shared memory that the
! * Startup process has put there for us
! */
! SpinLockAcquire(&BgWriterShmem->ckpt_lck);
! ReadPtr = BgWriterShmem->ReadPtr;
! memcpy(&restartPoint, &BgWriterShmem->restartPoint, sizeof(CheckPoint));
! SpinLockRelease(&BgWriterShmem->ckpt_lck);
!
! CreateRestartPoint(ReadPtr, &restartPoint);
!
! /*
! * Reset any flags if we requested immediate completion part
! * way through the restart point
! */
! SpinLockAcquire(&BgWriterShmem->ckpt_lck);
! BgWriterShmem->ckpt_flags = 0;
! SpinLockRelease(&BgWriterShmem->ckpt_lck);
!
! ckpt_active = false;
! checkpoint_requested = false;
! }
! else
! {
! /* Clean buffers dirtied by recovery */
! BgBufferSync(true);
!
! /* Nap for the configured time. */
! BgWriterNap();
! }
!
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
 
  /*
! * Check to see whether startup process has completed redo.
! * If so, we can permanently change out of recovery mode.
  */
! if (InStartupXLog())
! {
! elog(DEBUG2, "changing from recovery to normal mode");
!
! InitXLOGAccess();
! BgWriterInStartup = false;
!
! /*
! * Start time-driven events from now
! */
! last_checkpoint_time = last_xlog_switch_time = (pg_time_t) time(NULL);
! }
! }
! else /* Normal processing */
! {
! bool do_checkpoint = false;
! int flags = 0;
! pg_time_t now;
! int elapsed_secs;
!
! Assert(!InStartupXLog());
!
! if (checkpoint_requested)
! {
! checkpoint_requested = false;
! do_checkpoint = true;
! BgWriterStats.m_requested_checkpoints++;
! }
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Close down the database */
! ShutdownXLOG(0, 0);
! DumpFreeSpaceMap(0, 0);
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
 
  /*
! * Force a checkpoint if too much time has elapsed since the last one.
! * Note that we count a timed checkpoint in stats only when this
! * occurs without an external request, but we set the CAUSE_TIME flag
! * bit even if there is also an external request.
  */
! now = (pg_time_t) time(NULL);
! elapsed_secs = now - last_checkpoint_time;
! if (elapsed_secs >= CheckPointTimeout)
! {
! if (!do_checkpoint)
! BgWriterStats.m_timed_checkpoints++;
! do_checkpoint = true;
! flags |= CHECKPOINT_CAUSE_TIME;
! }
 
  /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
  */
! if (do_checkpoint)
! {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
!
! /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
!
! /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
! */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
!
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
!
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
!
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
!
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
! }
! else
! BgBufferSync(false);
 
! /* Check for archive_timeout and switch xlog files if necessary. */
! CheckArchiveTimeout();
 
! /* Nap for the configured time. */
! BgWriterNap();
  }
  }
  }
 
***************
*** 588,594 ****
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
--- 694,701 ----
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! if (!InStartupXLog())
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
***************
*** 642,647 ****
--- 749,767 ----
  if (!am_bg_writer)
  return;
 
+ /* Perform minimal duties during recovery and skip wait if requested */
+ if (InStartupXLog())
+ {
+ BgBufferSync(true);
+
+ if (!ImmediateCheckpointRequested() &&
+ !shutdown_requested &&
+ IsCheckpointOnSchedule(progress))
+ BgWriterNap();
+
+ return;
+ }
+
  /*
  * Perform the usual bgwriter duties and take a nap, unless we're behind
  * schedule, in which case we just try to catch up as quickly as possible.
***************
*** 660,666 ****
  AbsorbFsyncRequests();
  absorb_counter = WRITES_PER_ABSORB;
 
! BgBufferSync();
  CheckArchiveTimeout();
  BgWriterNap();
  }
--- 780,786 ----
  AbsorbFsyncRequests();
  absorb_counter = WRITES_PER_ABSORB;
 
! BgBufferSync(false);
  CheckArchiveTimeout();
  BgWriterNap();
  }
***************
*** 716,731 ****
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
  {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
  }
 
  /*
--- 836,854 ----
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! if (!InStartupXLog())
  {
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
! {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
! }
  }
 
  /*
***************
*** 967,972 ****
--- 1090,1126 ----
  }
 
  /*
+  * Always runs in Startup process (see xlog.c)
+  */
+ void
+ RequestRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, bool sendToBGWriter)
+ {
+ /*
+ * Should we just do it ourselves?
+ */
+ if (!IsPostmasterEnvironment || !sendToBGWriter)
+ {
+ CreateRestartPoint(ReadPtr, restartPoint);
+ return;
+ }
+
+ /*
+ * Push requested values into shared memory, then signal to request restartpoint.
+ */
+ if (BgWriterShmem->bgwriter_pid == 0)
+ elog(LOG, "could not request restartpoint because bgwriter not running");
+
+ SpinLockAcquire(&BgWriterShmem->ckpt_lck);
+ BgWriterShmem->ReadPtr = ReadPtr;
+ memcpy(&BgWriterShmem->restartPoint, restartPoint, sizeof(CheckPoint));
+ SpinLockRelease(&BgWriterShmem->ckpt_lck);
+
+
+ if (kill(BgWriterShmem->bgwriter_pid, SIGINT) != 0)
+ elog(LOG, "could not signal for restartpoint: %m");
+ }
+
+ /*
   * ForwardFsyncRequest
   * Forward a file-fsync request from a backend to the bgwriter
   *
Index: src/backend/postmaster/postmaster.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/postmaster.c,v
retrieving revision 1.561
diff -c -r1.561 postmaster.c
*** src/backend/postmaster/postmaster.c 26 Jun 2008 02:47:19 -0000 1.561
--- src/backend/postmaster/postmaster.c 10 Sep 2008 20:46:59 -0000
***************
*** 254,259 ****
--- 254,264 ----
  {
  PM_INIT, /* postmaster starting */
  PM_STARTUP, /* waiting for startup subprocess */
+ PM_RECOVERY, /* consistent recovery mode; state only
+ * entered for archive and streaming recovery,
+ * and only after the point where the
+ * all data is in consistent state.
+ */
  PM_RUN, /* normal "database is alive" state */
  PM_WAIT_BACKUP, /* waiting for online backup mode to end */
  PM_WAIT_BACKENDS, /* waiting for live backends to exit */
***************
*** 2104,2110 ****
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
--- 2109,2115 ----
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP || pmState == PM_RECOVERY);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
***************
*** 2148,2155 ****
  * Crank up the background writer. It doesn't matter if this
  * fails, we'll just try again later.
  */
! Assert(BgWriterPID == 0);
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
--- 2153,2160 ----
  * Crank up the background writer. It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0)
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
***************
*** 3821,3826 ****
--- 3826,3876 ----
 
  PG_SETMASK(&BlockSig);
 
+ if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_START))
+ {
+ Assert(pmState == PM_STARTUP);
+
+ /*
+ * Go to shutdown mode if a shutdown request was pending.
+ */
+ if (Shutdown > NoShutdown)
+ {
+ pmState = PM_WAIT_BACKENDS;
+ /* PostmasterStateMachine logic does the rest */
+ }
+ else
+ {
+ /*
+ * Startup process has entered recovery
+ */
+ pmState = PM_RECOVERY;
+
+ /*
+ * Load the flat authorization file into postmaster's cache. The
+ * startup process won't have recomputed this from the database yet,
+ * so we it may change following recovery.
+ */
+ load_role();
+
+ /*
+ * Crank up the background writer. It doesn't matter if this
+ * fails, we'll just try again later.
+ */
+ Assert(BgWriterPID == 0);
+ BgWriterPID = StartBackgroundWriter();
+
+ /*
+ * Likewise, start other special children as needed.
+ */
+ Assert(PgStatPID == 0);
+ PgStatPID = pgstat_start();
+
+ /* XXX at this point we could accept read-only connections */
+ ereport(DEBUG1,
+ (errmsg("database system is in consistent recovery mode")));
+ }
+ }
+
  if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
  {
  /*
Index: src/backend/storage/buffer/bufmgr.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/storage/buffer/bufmgr.c,v
retrieving revision 1.237
diff -c -r1.237 bufmgr.c
*** src/backend/storage/buffer/bufmgr.c 11 Aug 2008 11:05:11 -0000 1.237
--- src/backend/storage/buffer/bufmgr.c 10 Sep 2008 20:42:10 -0000
***************
*** 1211,1217 ****
   * This is called periodically by the background writer process.
   */
  void
! BgBufferSync(void)
  {
  /* info obtained from freelist.c */
  int strategy_buf_id;
--- 1211,1217 ----
   * This is called periodically by the background writer process.
   */
  void
! BgBufferSync(bool InStartup)
  {
  /* info obtained from freelist.c */
  int strategy_buf_id;
***************
*** 1423,1428 ****
--- 1423,1434 ----
  {
  int buffer_state = SyncOneBuffer(next_to_clean, true);
 
+ /*
+ * Quit scanning if bgwriter state changes while we're here
+ */
+ if (InStartup && !InStartupXLog())
+ break;
+
  if (++next_to_clean >= NBuffers)
  {
  next_to_clean = 0;
Index: src/include/access/xlog.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog.h,v
retrieving revision 1.88
diff -c -r1.88 xlog.h
*** src/include/access/xlog.h 12 May 2008 08:35:05 -0000 1.88
--- src/include/access/xlog.h 10 Sep 2008 20:42:55 -0000
***************
*** 197,202 ****
--- 197,203 ----
  extern void xlog_redo(XLogRecPtr lsn, XLogRecord *record);
  extern void xlog_desc(StringInfo buf, uint8 xl_info, char *rec);
 
+ extern bool InStartupXLog(void);
  extern void UpdateControlFile(void);
  extern Size XLOGShmemSize(void);
  extern void XLOGShmemInit(void);
Index: src/include/access/xlog_internal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog_internal.h,v
retrieving revision 1.24
diff -c -r1.24 xlog_internal.h
*** src/include/access/xlog_internal.h 11 Aug 2008 11:05:11 -0000 1.24
--- src/include/access/xlog_internal.h 10 Sep 2008 20:42:43 -0000
***************
*** 17,22 ****
--- 17,23 ----
  #define XLOG_INTERNAL_H
 
  #include "access/xlog.h"
+ #include "catalog/pg_control.h"
  #include "fmgr.h"
  #include "pgtime.h"
  #include "storage/block.h"
***************
*** 245,250 ****
--- 246,254 ----
  extern pg_time_t GetLastSegSwitchTime(void);
  extern XLogRecPtr RequestXLogSwitch(void);
 
+
+ extern void CreateRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint);
+
  /*
   * These aren't in xlog.h because I'd rather not include fmgr.h there.
   */
Index: src/include/postmaster/bgwriter.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/postmaster/bgwriter.h,v
retrieving revision 1.12
diff -c -r1.12 bgwriter.h
*** src/include/postmaster/bgwriter.h 11 Aug 2008 11:05:11 -0000 1.12
--- src/include/postmaster/bgwriter.h 10 Sep 2008 20:51:50 -0000
***************
*** 12,17 ****
--- 12,18 ----
  #ifndef _BGWRITER_H
  #define _BGWRITER_H
 
+ #include "catalog/pg_control.h"
  #include "storage/block.h"
  #include "storage/relfilenode.h"
 
***************
*** 25,36 ****
--- 26,40 ----
  extern void BackgroundWriterMain(void);
 
  extern void RequestCheckpoint(int flags);
+ extern void RequestRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, bool sendToBGWriter);
  extern void CheckpointWriteDelay(int flags, double progress);
 
  extern bool ForwardFsyncRequest(RelFileNode rnode, ForkNumber forknum,
  BlockNumber segno);
  extern void AbsorbFsyncRequests(void);
 
+ extern void BgWriterCompleteRestartPointImmediately(void);
+
  extern Size BgWriterShmemSize(void);
  extern void BgWriterShmemInit(void);
 
Index: src/include/storage/bufmgr.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/storage/bufmgr.h,v
retrieving revision 1.115
diff -c -r1.115 bufmgr.h
*** src/include/storage/bufmgr.h 11 Aug 2008 11:05:11 -0000 1.115
--- src/include/storage/bufmgr.h 10 Sep 2008 19:38:38 -0000
***************
*** 193,199 ****
  extern void AbortBufferIO(void);
 
  extern void BufmgrCommit(void);
! extern void BgBufferSync(void);
 
  extern void AtProcExit_LocalBuffers(void);
 
--- 193,199 ----
  extern void AbortBufferIO(void);
 
  extern void BufmgrCommit(void);
! extern void BgBufferSync(bool InStartup);
 
  extern void AtProcExit_LocalBuffers(void);
 
Index: src/include/storage/pmsignal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/storage/pmsignal.h,v
retrieving revision 1.20
diff -c -r1.20 pmsignal.h
*** src/include/storage/pmsignal.h 19 Jun 2008 21:32:56 -0000 1.20
--- src/include/storage/pmsignal.h 31 Aug 2008 19:54:58 -0000
***************
*** 22,27 ****
--- 22,28 ----
   */
  typedef enum
  {
+ PMSIGNAL_RECOVERY_START, /* move to PM_RECOVERY state */
  PMSIGNAL_PASSWORD_CHANGE, /* pg_auth file has changed */
  PMSIGNAL_WAKEN_ARCHIVER, /* send a NOTIFY signal to xlog archiver */
  PMSIGNAL_ROTATE_LOGFILE, /* send SIGUSR1 to syslogger to rotate logfile */

---

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Alvaro Herrera-7]

Simon Riggs wrote:

> --- 5716,5725 ----
>   CheckpointStats.ckpt_sync_end_t,
>   &sync_secs, &sync_usecs);
>  
> ! elog(LOG, "%s complete: wrote %d buffers (%.1f%%); "
>   "%d transaction log file(s) added, %d removed, %d recycled; "
>   "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
> + (checkpoint ? "  checkpoint" : "restartpoint"),
>   CheckpointStats.ckpt_bufs_written,
>   (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
>   CheckpointStats.ckpt_segs_added,

Very minor nit: this really needs a rework.  It is relatively OK in the
previous code, but it was already stuffing too much in a single message.
Maybe

ereport(LOG,
(errmsg(checkpoint ? "checkpoint complete" : "restartpoint complete"),
errdetail("Wrote %d buffers (%.1f%%); "
        "%d transaction log file(s) added, %d removed, %d recycled; "
        "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s.",
        ...
)))

--
Alvaro Herrera                                http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Fri, 2008-09-12 at 14:14 -0400, Alvaro Herrera wrote:

> Simon Riggs wrote:
>
> > --- 5716,5725 ----
> >   CheckpointStats.ckpt_sync_end_t,
> >   &sync_secs, &sync_usecs);
> >  
> > ! elog(LOG, "%s complete: wrote %d buffers (%.1f%%); "
> >   "%d transaction log file(s) added, %d removed, %d recycled; "
> >   "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
> > + (checkpoint ? "  checkpoint" : "restartpoint"),
> >   CheckpointStats.ckpt_bufs_written,
> >   (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
> >   CheckpointStats.ckpt_segs_added,
>
> Very minor nit: this really needs a rework.  

All I changed was the word "restartpoint"... its otherwise identical to
existing message. I'd rather not change that.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support


--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Alvaro Herrera-7]

Simon Riggs wrote:

>
> On Fri, 2008-09-12 at 14:14 -0400, Alvaro Herrera wrote:
> > Simon Riggs wrote:
> >
> > > --- 5716,5725 ----
> > >   CheckpointStats.ckpt_sync_end_t,
> > >   &sync_secs, &sync_usecs);
> > >  
> > > ! elog(LOG, "%s complete: wrote %d buffers (%.1f%%); "
> > >   "%d transaction log file(s) added, %d removed, %d recycled; "
> > >   "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
> > > + (checkpoint ? "  checkpoint" : "restartpoint"),
> > >   CheckpointStats.ckpt_bufs_written,
> > >   (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
> > >   CheckpointStats.ckpt_segs_added,
> >
> > Very minor nit: this really needs a rework.  
>
> All I changed was the word "restartpoint"... its otherwise identical to
> existing message. I'd rather not change that.

The new message is not translatable, the original was.

--
Alvaro Herrera                                http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Fri, 2008-09-12 at 15:31 -0400, Alvaro Herrera wrote:
> Simon Riggs wrote:
> >
> > All I changed was the word "restartpoint"... its otherwise identical to
> > existing message. I'd rather not change that.
>
> The new message is not translatable, the original was.

OK, perhaps the word "restartpoint" is redundant anyhow. Anybody looking
at the log can see we're in recovery. So I'll switch it back to say just
checkpoint whether we do restartpoint or checkpoints.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support


--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Tom Lane-2]

Simon Riggs <simon@...> wrote:
> Testing takes a while on this, I probably won't complete it until
> Friday. So enclosed patch is for eyeballs only at this stage.

What's the status on that patch?

                        regards, tom lane

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Tue, 2008-09-16 at 15:45 -0400, Tom Lane wrote:
> Simon Riggs <simon@...> wrote:
> > Testing takes a while on this, I probably won't complete it until
> > Friday. So enclosed patch is for eyeballs only at this stage.
>
> What's the status on that patch?

Checking EXEC_BACKEND case and that reporting back later today.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support


--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Tue, 2008-09-16 at 15:45 -0400, Tom Lane wrote:
> Simon Riggs <simon@...> wrote:
> > Testing takes a while on this, I probably won't complete it until
> > Friday. So enclosed patch is for eyeballs only at this stage.
>
> What's the status on that patch?

Having some trouble trying to get a clean state change from recovery to
normal mode.

Startup needs to be able to write WAL at the end of recovery so it can
write a ShutdownCheckpoint, yet must not be allowed to write WAL before
that. Other services are still not fully up yet. So every other process
apart from Startup musn't write WAL until Startup has fully completed
its actions, which is sometime later.

bgwriter needs to know about the impending state change so it can finish
off any checkpoint its currently working on. But then can't start doing
normal processing yet either. So it must have a third state that is
between recovery and normal processing. When normal processing is
reached, it *must* write WAL immediately from that point onwards, yet
using the new timeline that startup decides upon.

So the idea of a single database-wide boolean state seems impossible. We
need a local canInsertWAL flag that is set at different times in
different processes.

Global states changes are now

not started
started - postmaster process then startup process
safestoppingpoint - bgwriter starts
startup_does_shutdown_checkpoint - bgwriter finishes up, just waits,
   startup is now allowed to insert WAL for checkpoint record
startup completes StartupXLog - bgwriter begins normal processing
startup exits - postmaster in normal state

We *might* solve this by making the final checkpoint that Startup writes
into an online checkpoint. That would then allow a straight and safe
transition for bgwriter from just one state to the other. But that
leaves some other ugliness that I don't want to deal with, 'cos there's
other fish frying.

Feels like I should shutdown the bgwriter after recovery and then allow
it to be cranked up again after normal processing starts, and do all of
this through postmaster state changes. That way bgwriter doesn't need to
do a dynamic state change.

Comments anyone?

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support


--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Thu, 2008-09-18 at 09:05 +0100, Simon Riggs wrote:

> Feels like I should shutdown the bgwriter after recovery and then
> allow it to be cranked up again after normal processing starts, and do
> all of this through postmaster state changes. That way bgwriter
> doesn't need to do a dynamic state change.

This approach appears to be working nicely so far. Some ugly bits of
former patch removed.

Patch passes basic tests and changes state cleanly.

Restarting test cycle on this patch now, confirm tomorrow.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support

[recovery_infrastruc.v5.patch]

---

Index: src/backend/access/transam/xlog.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/access/transam/xlog.c,v
retrieving revision 1.317
diff -c -r1.317 xlog.c
*** src/backend/access/transam/xlog.c 11 Aug 2008 11:05:10 -0000 1.317
--- src/backend/access/transam/xlog.c 18 Sep 2008 11:39:07 -0000
***************
*** 119,124 ****
--- 119,126 ----
 
  /* Are we doing recovery from XLOG? */
  bool InRecovery = false;
+ bool reachedSafeStopPoint = false;
+ bool StartupCanWriteWAL = false;
 
  /* Are we recovering using offline XLOG archives? */
  static bool InArchiveRecovery = false;
***************
*** 131,137 ****
  static bool recoveryTarget = false;
  static bool recoveryTargetExact = false;
  static bool recoveryTargetInclusive = true;
- static bool recoveryLogRestartpoints = false;
  static TransactionId recoveryTargetXid;
  static TimestampTz recoveryTargetTime;
  static TimestampTz recoveryLastXTime = 0;
--- 133,138 ----
***************
*** 286,295 ****
--- 287,298 ----
  /*
   * Total shared-memory state for XLOG.
   */
+ #define XLOGCTL_BUFFER_SPACING 128
  typedef struct XLogCtlData
  {
  /* Protected by WALInsertLock: */
  XLogCtlInsert Insert;
+ char InsertPadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogCtlInsert)];
 
  /* Protected by info_lck: */
  XLogwrtRqst LogwrtRqst;
***************
*** 297,305 ****
--- 300,315 ----
  uint32 ckptXidEpoch; /* nextXID & epoch of latest checkpoint */
  TransactionId ckptXid;
  XLogRecPtr asyncCommitLSN; /* LSN of newest async commit */
+ /* add data structure padding for above info_lck declarations */
+ char InfoPadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogwrtRqst)
+ - sizeof(XLogwrtResult)
+ - sizeof(uint32)
+ - sizeof(TransactionId)
+ - sizeof(XLogRecPtr)];
 
  /* Protected by WALWriteLock: */
  XLogCtlWrite Write;
+ char WritePadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogCtlWrite)];
 
  /*
  * These values do not change after startup, although the pointed-to pages
***************
*** 311,316 ****
--- 321,337 ----
  int XLogCacheBlck; /* highest allocated xlog buffer index */
  TimeLineID ThisTimeLineID;
 
+ /*
+ * canWriteWAL changes at the end of recovery only and is only ever set
+ * by the Startup process. We assume that changes to it are atomic,
+ * so accesses to it is never locked. When it does change bgwriter
+ * must immediately begin using it, since this helps it decide whether
+ * to flush WAL or not when it writes dirty blocks. If bgwriter does
+ * it too soon, we will write invalid WAL records and if it reflects the
+ * change too late it could skip flushing WAL for a data block change.
+ */
+ bool canWriteWAL;
+
  slock_t info_lck; /* locks shared variables shown above */
  } XLogCtlData;
 
***************
*** 480,485 ****
--- 501,510 ----
  bool doPageWrites;
  bool isLogSwitch = (rmid == RM_XLOG_ID && info == XLOG_SWITCH);
 
+ /* cross-check on whether we should be here or not */
+ if (!(canWriteWAL() || (InRecovery && StartupCanWriteWAL)))
+ elog(FATAL, "cannot make new WAL entries during recovery");
+
  /* info's high bits are reserved for use by me */
  if (info & XLR_INFO_MASK)
  elog(PANIC, "invalid xlog info mask %02X", info);
***************
*** 1677,1684 ****
  XLogRecPtr WriteRqstPtr;
  XLogwrtRqst WriteRqst;
 
! /* Disabled during REDO */
! if (InRedo)
  return;
 
  /* Quick exit if already known flushed */
--- 1702,1709 ----
  XLogRecPtr WriteRqstPtr;
  XLogwrtRqst WriteRqst;
 
! /* Disabled during StartupXLog */
! if (!canWriteWAL())
  return;
 
  /* Quick exit if already known flushed */
***************
*** 1766,1774 ****
  * the bad page is encountered again during recovery then we would be
  * unable to restart the database at all!  (This scenario has actually
  * happened in the field several times with 7.1 releases. Note that we
! * cannot get here while InRedo is true, but if the bad page is brought in
! * and marked dirty during recovery then CreateCheckPoint will try to
! * flush it at the end of recovery.)
  *
  * The current approach is to ERROR under normal conditions, but only
  * WARNING during recovery, so that the system can be brought up even if
--- 1791,1799 ----
  * the bad page is encountered again during recovery then we would be
  * unable to restart the database at all!  (This scenario has actually
  * happened in the field several times with 7.1 releases. Note that we
! * cannot get here while canWriteWAL() is false, but if the bad page is
! * brought in and marked dirty during recovery then CreateCheckPoint will
! * try to flush it at the end of recovery.)
  *
  * The current approach is to ERROR under normal conditions, but only
  * WARNING during recovery, so that the system can be brought up even if
***************
*** 2051,2057 ****
  unlink(tmppath);
  }
 
! elog(DEBUG2, "done creating and filling new WAL file");
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
--- 2076,2083 ----
  unlink(tmppath);
  }
 
! XLogFileName(tmppath, ThisTimeLineID, log, seg);
! elog(DEBUG2, "done creating and filling new WAL file %s", tmppath);
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
***************
*** 4532,4546 ****
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
- /*
- * does nothing if a recovery_target is not also set
- */
- if (!parse_bool(tok2, &recoveryLogRestartpoints))
-  ereport(ERROR,
- (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-  errmsg("parameter \"log_restartpoints\" requires a Boolean value")));
  ereport(LOG,
! (errmsg("log_restartpoints = %s", tok2)));
  }
  else
  ereport(FATAL,
--- 4558,4566 ----
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
  ereport(LOG,
! (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
!  errmsg("parameter \"log_restartpoints\" has been deprecated")));
  }
  else
  ereport(FATAL,
***************
*** 4823,4828 ****
--- 4843,4850 ----
  uint32 freespace;
  TransactionId oldestActiveXID;
 
+ XLogCtl->canWriteWAL = false;
+
  /*
  * Read control file and check XLOG status looks valid.
  *
***************
*** 5039,5044 ****
--- 5061,5071 ----
  UpdateControlFile();
 
  /*
+ * Reset pgstat data, because it may be invalid after recovery.
+ */
+ pgstat_reset_all();
+
+ /*
  * If there was a backup label file, it's done its job and the info
  * has now been propagated into pg_control.  We must get rid of the
  * label file so that if we crash during recovery, we'll pick up at
***************
*** 5148,5153 ****
--- 5175,5195 ----
 
  LastRec = ReadRecPtr;
 
+ /*
+ * Have we reached our safe stopping point? If so, we can
+ * signal Postmaster to enter consistent recovery mode
+ */
+ if (!reachedSafeStopPoint &&
+ XLByteLE(ControlFile->minRecoveryPoint, EndRecPtr))
+ {
+ reachedSafeStopPoint = true;
+ ereport(LOG,
+ (errmsg("consistent recovery state reached at %X/%X",
+ EndRecPtr.xlogid, EndRecPtr.xrecoff)));
+ if (IsUnderPostmaster)
+ SendPostmasterSignal(PMSIGNAL_RECOVERY_START);
+ }
+
  record = ReadRecord(NULL, LOG);
  } while (record != NULL && recoveryContinue);
 
***************
*** 5169,5174 ****
--- 5211,5217 ----
  /* there are no WAL records following the checkpoint */
  ereport(LOG,
  (errmsg("redo is not required")));
+ reachedSafeStopPoint = true;
  }
  }
 
***************
*** 5184,5190 ****
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (XLByteLT(EndOfLog, ControlFile->minRecoveryPoint))
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
--- 5227,5233 ----
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (InRecovery && !reachedSafeStopPoint)
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
***************
*** 5306,5314 ****
  XLogCheckInvalidPages();
 
  /*
! * Reset pgstat data, because it may be invalid after recovery.
  */
! pgstat_reset_all();
 
  /*
  * Perform a checkpoint to update all our recovery activity to disk.
--- 5349,5365 ----
  XLogCheckInvalidPages();
 
  /*
! * Let Postmaster know we are about to write Shutdown checkpoint,
! * so that we can shutdown bgwriter cleanly and be left in peace
! * to get on with this.
  */
! if (IsUnderPostmaster)
! SendPostmasterSignal(PMSIGNAL_RECOVERY_CKPT);
!
! /*
! * Startup process is now allowed to write WAL records
! */
! StartupCanWriteWAL = true;
 
  /*
  * Perform a checkpoint to update all our recovery activity to disk.
***************
*** 5318,5323 ****
--- 5369,5377 ----
  * assigning a new TLI, using a shutdown checkpoint allows us to have
  * the rule that TLI only changes in shutdown checkpoints, which
  * allows some extra error checking in xlog_redo.
+ *
+ * Note that this will wait behind any restartpoint that the bgwriter
+ * is currently performing, though will be much faster as a result.
  */
  CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
  }
***************
*** 5372,5377 ****
--- 5426,5442 ----
  readRecordBuf = NULL;
  readRecordBufSize = 0;
  }
+
+ /*
+ * Lastly, allow everybody else to write WAL now
+ */
+ XLogCtl->canWriteWAL = true;
+ }
+
+ bool
+ canWriteWAL(void)
+ {
+ return XLogCtl->canWriteWAL;
  }
 
  /*
***************
*** 6071,6099 ****
  }
  }
 
  /*
! * OK, force data out to disk
  */
! CheckPointGuts(checkPoint->redo, CHECKPOINT_IMMEDIATE);
 
  /*
! * Update pg_control so that any subsequent crash will restart from this
! * checkpoint. Note: ReadRecPtr gives the XLOG address of the checkpoint
! * record itself.
  */
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
! ControlFile->checkPoint = ReadRecPtr;
! ControlFile->checkPointCopy = *checkPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
 
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
! checkPoint->redo.xlogid, checkPoint->redo.xrecoff)));
! if (recoveryLastXTime)
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
! (errmsg("last completed transaction was at log time %s",
! timestamptz_to_str(recoveryLastXTime))));
  }
 
  /*
--- 6136,6200 ----
  }
  }
 
+ if (recoveryLastXTime)
+ ereport((log_checkpoints ? LOG : DEBUG2),
+ (errmsg("last completed transaction was at log time %s",
+ timestamptz_to_str(recoveryLastXTime))));
+
+ RequestRestartPoint(ReadRecPtr, checkPoint, reachedSafeStopPoint);
+ }
+
+ /*
+  * As of 8.4, RestartPoints are always created by the bgwriter
+  * once we have reachedSafeStopPoint. We use bgwriter's shared memory
+  * area wherever we call it from, to keep better code structure.
+  */
+ void
+ CreateRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint)
+ {
+ if (log_checkpoints)
+ {
+ /*
+ * Prepare to accumulate statistics.
+ */
+
+ MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
+
+ /*
+ * Do the restartpoint equivalent of LogCheckpointStart()
+ */
+ elog(LOG, "restartpoint starting:");
+ }
+
+ LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
+
  /*
! * OK, write out dirty blocks smoothly
  */
! CheckPointGuts(restartPoint->redo, 0);
 
  /*
! * Update pg_control, using current time
  */
+ LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
! ControlFile->checkPoint = ReadPtr;
! ControlFile->checkPointCopy = *restartPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
+ LWLockRelease(ControlFileLock);
 
! /* All real work is done, but log before releasing lock. */
! if (log_checkpoints)
! LogCheckpointEnd();
!
! ereport((log_checkpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
! restartPoint->redo.xlogid, restartPoint->redo.xrecoff)));
!
! LWLockRelease(CheckpointLock);
!
  }
 
  /*
Index: src/backend/postmaster/bgwriter.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/bgwriter.c,v
retrieving revision 1.51
diff -c -r1.51 bgwriter.c
*** src/backend/postmaster/bgwriter.c 11 Aug 2008 11:05:11 -0000 1.51
--- src/backend/postmaster/bgwriter.c 18 Sep 2008 12:06:05 -0000
***************
*** 49,54 ****
--- 49,55 ----
  #include <unistd.h>
 
  #include "access/xlog_internal.h"
+ #include "catalog/pg_control.h"
  #include "libpq/pqsignal.h"
  #include "miscadmin.h"
  #include "pgstat.h"
***************
*** 130,135 ****
--- 131,144 ----
 
  int ckpt_flags; /* checkpoint flags, as defined in xlog.h */
 
+ /*
+ * When the Startup process wants a restartpoint, it sets these fields
+ * so that whoever performs the restartpoint can update the control file,
+ * allowing the caller to continue, if it is running in another process.
+ */
+ XLogRecPtr ReadPtr; /* ReadRecPtr for RestartPoint request */
+ CheckPoint *restartPoint; /* restartPoint data for ControlFile */
+
  uint32 num_backend_writes; /* counts non-bgwriter buffer writes */
 
  int num_requests; /* current # of requests */
***************
*** 164,172 ****
 
  static bool ckpt_active = false;
 
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr;
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
--- 173,184 ----
 
  static bool ckpt_active = false;
 
+ static bool BgWriterInRecovery = true;
+
+
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr; /* not used if canWriteWAL */
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
***************
*** 186,192 ****
  static void ReqCheckpointHandler(SIGNAL_ARGS);
  static void ReqShutdownHandler(SIGNAL_ARGS);
 
-
  /*
   * Main entry point for bgwriter process
   *
--- 198,203 ----
***************
*** 357,371 ****
  PG_SETMASK(&UnBlockSig);
 
  /*
  * Loop forever
  */
  for (;;)
  {
- bool do_checkpoint = false;
- int flags = 0;
- pg_time_t now;
- int elapsed_secs;
-
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
--- 368,389 ----
  PG_SETMASK(&UnBlockSig);
 
  /*
+ * We are either in recovery mode, or we're not. bgwriter never
+ * attepts to change state, we just shutdown and startup in new mode.
+ * That's fine don't really want to hang around while Startup process
+ * performs it's shutdown checkpoint.
+ */
+ BgWriterInRecovery = !canWriteWAL();
+
+ if (BgWriterInRecovery)
+ elog(DEBUG1, "bgwriter starting during recovery, pid = %u",
+ BgWriterShmem->bgwriter_pid);
+
+ /*
  * Loop forever
  */
  for (;;)
  {
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
***************
*** 383,501 ****
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
- if (checkpoint_requested)
- {
- checkpoint_requested = false;
- do_checkpoint = true;
- BgWriterStats.m_requested_checkpoints++;
- }
- if (shutdown_requested)
- {
- /*
- * From here on, elog(ERROR) should end with exit(1), not send
- * control back to the sigsetjmp block above
- */
- ExitOnAnyError = true;
- /* Close down the database */
- ShutdownXLOG(0, 0);
- DumpFreeSpaceMap(0, 0);
- /* Normal exit from the bgwriter is here */
- proc_exit(0); /* done */
- }
 
! /*
! * Force a checkpoint if too much time has elapsed since the last one.
! * Note that we count a timed checkpoint in stats only when this
! * occurs without an external request, but we set the CAUSE_TIME flag
! * bit even if there is also an external request.
! */
! now = (pg_time_t) time(NULL);
! elapsed_secs = now - last_checkpoint_time;
! if (elapsed_secs >= CheckPointTimeout)
  {
! if (!do_checkpoint)
! BgWriterStats.m_timed_checkpoints++;
! do_checkpoint = true;
! flags |= CHECKPOINT_CAUSE_TIME;
  }
!
! /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
! */
! if (do_checkpoint)
  {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
!
! /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
 
  /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
  */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
 
  /*
! * Initialize bgwriter-private variables used during checkpoint.
  */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
 
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
!
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
!
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
 
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
  }
- else
- BgBufferSync();
-
- /* Check for archive_timeout and switch xlog files if necessary. */
- CheckArchiveTimeout();
-
- /* Nap for the configured time. */
- BgWriterNap();
  }
  }
 
--- 401,586 ----
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
 
! if (BgWriterInRecovery)
  {
! if (checkpoint_requested)
! {
! XLogRecPtr ReadPtr;
! CheckPoint restartPoint;
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_time = (pg_time_t) time(NULL);
! ckpt_cached_elapsed = 0;
!
! /*
! * Get the requested values from shared memory that the
! * Startup process has put there for us
! */
! SpinLockAcquire(&BgWriterShmem->ckpt_lck);
! ReadPtr = BgWriterShmem->ReadPtr;
! memcpy(&restartPoint, &BgWriterShmem->restartPoint, sizeof(CheckPoint));
! SpinLockRelease(&BgWriterShmem->ckpt_lck);
!
! CreateRestartPoint(ReadPtr, &restartPoint);
!
! /*
! * Reset any flags if we requested immediate completion part
! * way through the restart point
! */
! SpinLockAcquire(&BgWriterShmem->ckpt_lck);
! BgWriterShmem->ckpt_flags = 0;
! SpinLockRelease(&BgWriterShmem->ckpt_lck);
!
! ckpt_active = false;
! checkpoint_requested = false;
! }
! else
! {
! /* Clean buffers dirtied by recovery */
! BgBufferSync();
!
! /* Nap for the configured time. */
! BgWriterNap();
! }
!
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
  }
! else /* Normal processing */
  {
! bool do_checkpoint = false;
! int flags = 0;
! pg_time_t now;
! int elapsed_secs;
!
! Assert(canWriteWAL());
!
! if (checkpoint_requested)
! {
! checkpoint_requested = false;
! do_checkpoint = true;
! BgWriterStats.m_requested_checkpoints++;
! }
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Close down the database */
! ShutdownXLOG(0, 0);
! DumpFreeSpaceMap(0, 0);
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
 
  /*
! * Force a checkpoint if too much time has elapsed since the last one.
! * Note that we count a timed checkpoint in stats only when this
! * occurs without an external request, but we set the CAUSE_TIME flag
! * bit even if there is also an external request.
  */
! now = (pg_time_t) time(NULL);
! elapsed_secs = now - last_checkpoint_time;
! if (elapsed_secs >= CheckPointTimeout)
! {
! if (!do_checkpoint)
! BgWriterStats.m_timed_checkpoints++;
! do_checkpoint = true;
! flags |= CHECKPOINT_CAUSE_TIME;
! }
 
  /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
  */
! if (do_checkpoint)
! {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
!
! /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
!
! /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
! */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
!
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
!
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
!
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
!
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
! }
! else
! BgBufferSync();
 
! /* Check for archive_timeout and switch xlog files if necessary. */
! CheckArchiveTimeout();
 
! /* Nap for the configured time. */
! BgWriterNap();
  }
  }
  }
 
***************
*** 588,594 ****
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
--- 673,680 ----
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! if (canWriteWAL())
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
***************
*** 642,647 ****
--- 728,745 ----
  if (!am_bg_writer)
  return;
 
+ /* Perform minimal duties during recovery and skip wait if requested */
+ if (!canWriteWAL())
+ {
+ BgBufferSync();
+
+ if (!shutdown_requested &&
+ IsCheckpointOnSchedule(progress))
+ BgWriterNap();
+
+ return;
+ }
+
  /*
  * Perform the usual bgwriter duties and take a nap, unless we're behind
  * schedule, in which case we just try to catch up as quickly as possible.
***************
*** 716,731 ****
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
  {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
  }
 
  /*
--- 814,832 ----
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! if (canWriteWAL())
  {
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
! {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
! }
  }
 
  /*
***************
*** 967,972 ****
--- 1068,1104 ----
  }
 
  /*
+  * Always runs in Startup process (see xlog.c)
+  */
+ void
+ RequestRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, bool sendToBGWriter)
+ {
+ /*
+ * Should we just do it ourselves?
+ */
+ if (!IsPostmasterEnvironment || !sendToBGWriter)
+ {
+ CreateRestartPoint(ReadPtr, restartPoint);
+ return;
+ }
+
+ /*
+ * Push requested values into shared memory, then signal to request restartpoint.
+ */
+ if (BgWriterShmem->bgwriter_pid == 0)
+ elog(LOG, "could not request restartpoint because bgwriter not running");
+
+ SpinLockAcquire(&BgWriterShmem->ckpt_lck);
+ BgWriterShmem->ReadPtr = ReadPtr;
+ memcpy(&BgWriterShmem->restartPoint, restartPoint, sizeof(CheckPoint));
+ SpinLockRelease(&BgWriterShmem->ckpt_lck);
+
+
+ if (kill(BgWriterShmem->bgwriter_pid, SIGINT) != 0)
+ elog(LOG, "could not signal for restartpoint: %m");
+ }
+
+ /*
   * ForwardFsyncRequest
   * Forward a file-fsync request from a backend to the bgwriter
   *
Index: src/backend/postmaster/postmaster.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/postmaster.c,v
retrieving revision 1.561
diff -c -r1.561 postmaster.c
*** src/backend/postmaster/postmaster.c 26 Jun 2008 02:47:19 -0000 1.561
--- src/backend/postmaster/postmaster.c 18 Sep 2008 12:03:07 -0000
***************
*** 254,259 ****
--- 254,267 ----
  {
  PM_INIT, /* postmaster starting */
  PM_STARTUP, /* waiting for startup subprocess */
+ PM_RECOVERY, /* consistent recovery mode; state only
+ * entered for archive and streaming recovery,
+ * and only after the point where the
+ * all data is in consistent state.
+ */
+ PM_RECOVERY_CKPT, /* Startup process writes shutdown checkpoint
+ * so bgwriter must shutdown immediately.
+ */
  PM_RUN, /* normal "database is alive" state */
  PM_WAIT_BACKUP, /* waiting for online backup mode to end */
  PM_WAIT_BACKENDS, /* waiting for live backends to exit */
***************
*** 1294,1300 ****
  * state that prevents it, start one.  It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0 && pmState == PM_RUN)
  BgWriterPID = StartBackgroundWriter();
 
  /*
--- 1302,1308 ----
  * state that prevents it, start one.  It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0 && (pmState == PM_RUN || pmState == PM_RECOVERY))
  BgWriterPID = StartBackgroundWriter();
 
  /*
***************
*** 2104,2110 ****
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
--- 2112,2119 ----
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP || pmState == PM_RECOVERY
! || pmState == PM_RECOVERY_CKPT);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
***************
*** 2116,2121 ****
--- 2125,2133 ----
  ExitPostmaster(1);
  }
 
+ /* We should never exit normally during PM_RECOVERY state */
+ Assert(pmState != PM_RECOVERY);
+
  /*
  * Startup succeeded - we are done with system startup or
  * recovery.
***************
*** 2148,2155 ****
  * Crank up the background writer. It doesn't matter if this
  * fails, we'll just try again later.
  */
! Assert(BgWriterPID == 0);
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
--- 2160,2167 ----
  * Crank up the background writer. It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0)
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
***************
*** 2211,2216 ****
--- 2223,2238 ----
  if (PgStatPID != 0)
  signal_child(PgStatPID, SIGQUIT);
  }
+ else if (EXIT_STATUS_0(exitstatus) && pmState == PM_RECOVERY_CKPT)
+ {
+ /*
+ * bgwriter has shutdown as it was requested, so this is OK.
+ * There is no state change associated with this event,
+ * since this was caused by the state change.
+ * We now wait for Startup process to complete, before
+ * moving to PM_RUN state.
+ */
+ }
  else
  {
  /*
***************
*** 2570,2575 ****
--- 2592,2619 ----
  static void
  PostmasterStateMachine(void)
  {
+ if (pmState == PM_RECOVERY)
+ {
+ /* Start the bgwriter if not running */
+ if (BgWriterPID == 0)
+ BgWriterPID = StartBackgroundWriter();
+
+ /* If we have lost the stats collector, try to start a new one */
+ if (PgStatPID == 0)
+ PgStatPID = pgstat_start();
+ }
+
+ if (pmState == PM_RECOVERY_CKPT)
+ {
+ /* Tell bgwriter to shut down, if its still active */
+ if (BgWriterPID != 0)
+ signal_child(BgWriterPID, SIGUSR2);
+
+ /* If we have lost the stats collector, try to start a new one */
+ if (PgStatPID == 0)
+ PgStatPID = pgstat_start();
+ }
+
  if (pmState == PM_WAIT_BACKUP)
  {
  /*
***************
*** 3821,3826 ****
--- 3865,3940 ----
 
  PG_SETMASK(&BlockSig);
 
+ if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_START))
+ {
+ Assert(pmState == PM_STARTUP);
+
+ /*
+ * Go to shutdown mode if a shutdown request was pending.
+ */
+ if (Shutdown > NoShutdown)
+ {
+ pmState = PM_WAIT_BACKENDS;
+ /* PostmasterStateMachine logic does the rest */
+ }
+ else
+ {
+ /*
+ * Startup process has entered recovery
+ */
+ pmState = PM_RECOVERY;
+
+ /*
+ * Load the flat authorization file into postmaster's cache. The
+ * startup process won't have recomputed this from the database yet,
+ * so we it may change following recovery.
+ */
+ load_role();
+
+ /*
+ * Crank up the background writer. It doesn't matter if this
+ * fails, we'll just try again later.
+ */
+ Assert(BgWriterPID == 0);
+ BgWriterPID = StartBackgroundWriter();
+
+ /*
+ * Likewise, start other special children as needed.
+ */
+ Assert(PgStatPID == 0);
+ PgStatPID = pgstat_start();
+
+ /* XXX at this point we could accept read-only connections */
+ ereport(DEBUG1,
+ (errmsg("database system is in consistent recovery mode")));
+ }
+ }
+
+ if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_CKPT))
+ {
+ Assert(pmState == PM_RECOVERY);
+
+ /*
+ * Go to shutdown mode if a shutdown request was pending.
+ */
+ if (Shutdown > NoShutdown)
+ {
+ pmState = PM_WAIT_BACKENDS;
+ /* PostmasterStateMachine logic does the rest */
+ }
+ else
+ {
+ /*
+ * Startup process has entered recovery checkpoint phase
+ */
+ pmState = PM_RECOVERY_CKPT;
+
+ /* Tell bgwriter to shut down */
+ if (BgWriterPID != 0)
+ signal_child(BgWriterPID, SIGUSR2);
+ }
+ }
+
  if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
  {
  /*
Index: src/include/access/xlog.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog.h,v
retrieving revision 1.88
diff -c -r1.88 xlog.h
*** src/include/access/xlog.h 12 May 2008 08:35:05 -0000 1.88
--- src/include/access/xlog.h 18 Sep 2008 11:40:27 -0000
***************
*** 197,202 ****
--- 197,204 ----
  extern void xlog_redo(XLogRecPtr lsn, XLogRecord *record);
  extern void xlog_desc(StringInfo buf, uint8 xl_info, char *rec);
 
+ extern bool canWriteWAL(void);
+
  extern void UpdateControlFile(void);
  extern Size XLOGShmemSize(void);
  extern void XLOGShmemInit(void);
Index: src/include/access/xlog_internal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog_internal.h,v
retrieving revision 1.24
diff -c -r1.24 xlog_internal.h
*** src/include/access/xlog_internal.h 11 Aug 2008 11:05:11 -0000 1.24
--- src/include/access/xlog_internal.h 18 Sep 2008 05:54:53 -0000
***************
*** 17,22 ****
--- 17,23 ----
  #define XLOG_INTERNAL_H
 
  #include "access/xlog.h"
+ #include "catalog/pg_control.h"
  #include "fmgr.h"
  #include "pgtime.h"
  #include "storage/block.h"
***************
*** 245,250 ****
--- 246,254 ----
  extern pg_time_t GetLastSegSwitchTime(void);
  extern XLogRecPtr RequestXLogSwitch(void);
 
+
+ extern void CreateRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint);
+
  /*
   * These aren't in xlog.h because I'd rather not include fmgr.h there.
   */
Index: src/include/postmaster/bgwriter.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/postmaster/bgwriter.h,v
retrieving revision 1.12
diff -c -r1.12 bgwriter.h
*** src/include/postmaster/bgwriter.h 11 Aug 2008 11:05:11 -0000 1.12
--- src/include/postmaster/bgwriter.h 18 Sep 2008 10:44:27 -0000
***************
*** 12,17 ****
--- 12,18 ----
  #ifndef _BGWRITER_H
  #define _BGWRITER_H
 
+ #include "catalog/pg_control.h"
  #include "storage/block.h"
  #include "storage/relfilenode.h"
 
***************
*** 25,30 ****
--- 26,32 ----
  extern void BackgroundWriterMain(void);
 
  extern void RequestCheckpoint(int flags);
+ extern void RequestRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, bool sendToBGWriter);
  extern void CheckpointWriteDelay(int flags, double progress);
 
  extern bool ForwardFsyncRequest(RelFileNode rnode, ForkNumber forknum,
Index: src/include/storage/pmsignal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/storage/pmsignal.h,v
retrieving revision 1.20
diff -c -r1.20 pmsignal.h
*** src/include/storage/pmsignal.h 19 Jun 2008 21:32:56 -0000 1.20
--- src/include/storage/pmsignal.h 18 Sep 2008 11:03:03 -0000
***************
*** 22,27 ****
--- 22,29 ----
   */
  typedef enum
  {
+ PMSIGNAL_RECOVERY_START, /* move to PM_RECOVERY state */
+ PMSIGNAL_RECOVERY_CKPT, /* move to PM_RECOVERY_CKPT state */
  PMSIGNAL_PASSWORD_CHANGE, /* pg_auth file has changed */
  PMSIGNAL_WAKEN_ARCHIVER, /* send a NOTIFY signal to xlog archiver */
  PMSIGNAL_ROTATE_LOGFILE, /* send SIGUSR1 to syslogger to rotate logfile */

---

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Tom Lane-2]

Simon Riggs <simon@...> writes:
> Having some trouble trying to get a clean state change from recovery to
> normal mode.

> Startup needs to be able to write WAL at the end of recovery so it can
> write a ShutdownCheckpoint, yet must not be allowed to write WAL before
> that. Other services are still not fully up yet. So every other process
> apart from Startup musn't write WAL until Startup has fully completed
> its actions, which is sometime later.
> ...
> We *might* solve this by making the final checkpoint that Startup writes
> into an online checkpoint.

Do we really need a checkpoint there at all?  I can't recall right now
if there was a good reason why Vadim made it do that.  I have a feeling
that it might have had to do with an assumption that the recovery
environment was completely distinct from live environment; which is a
statement that's certainly not going to be true in a query-answering
slave.

                        regards, tom lane

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Thu, 2008-09-18 at 09:06 -0400, Tom Lane wrote:

> Simon Riggs <simon@...> writes:
> > Having some trouble trying to get a clean state change from recovery to
> > normal mode.
>
> > Startup needs to be able to write WAL at the end of recovery so it can
> > write a ShutdownCheckpoint, yet must not be allowed to write WAL before
> > that. Other services are still not fully up yet. So every other process
> > apart from Startup musn't write WAL until Startup has fully completed
> > its actions, which is sometime later.
> > ...
> > We *might* solve this by making the final checkpoint that Startup writes
> > into an online checkpoint.
>
> Do we really need a checkpoint there at all?  I can't recall right now
> if there was a good reason why Vadim made it do that.  I have a feeling
> that it might have had to do with an assumption that the recovery
> environment was completely distinct from live environment; which is a
> statement that's certainly not going to be true in a query-answering
> slave.

Hmm, a question I hadn't considered.

We definitely don't want to do PreallocXlogFiles().

"Timelines only change at shutdown checkpoints". But its just as easy to
write a short timeline change record rather than the checkpoint itself,
so we can sort that out.

It should be sufficient to request bgwriter to perform an immediate
checkpoint, rather than have startup process perform it. That way
startup can exit and we can change to normal processing faster.

If we crash before the next checkpoint then we would revert to archive
recovery or crash recovery. The last restartpoint might well be
somewhere else. In streaming mode we would presumably have that data
locally, so not really a problem. But we musn't mark control file in
production yet, but we can do so following the next checkpoint.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support


--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Tom Lane-2]

Simon Riggs <simon@...> writes:
> On Thu, 2008-09-18 at 09:06 -0400, Tom Lane wrote:
>> Do we really need a checkpoint there at all?

> "Timelines only change at shutdown checkpoints".

Hmm.  I *think* that that is just a debugging crosscheck rather than a
critical property.  But yeah, it would take some close investigation,
which maybe isn't warranted if you have a less-invasive solution.

                        regards, tom lane

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Thu, 2008-09-18 at 10:09 -0400, Tom Lane wrote:
> Simon Riggs <simon@...> writes:
> > On Thu, 2008-09-18 at 09:06 -0400, Tom Lane wrote:
> >> Do we really need a checkpoint there at all?
>
> > "Timelines only change at shutdown checkpoints".
>
> Hmm.  I *think* that that is just a debugging crosscheck rather than a
> critical property.  But yeah, it would take some close investigation,
> which maybe isn't warranted if you have a less-invasive solution.

The important thing is that everybody uses the new timelineid.

AFAICS we actually write new timelineids into the WAL file if the
previous timelineid, so it can't be that risky.

Ignore v5 then and look for v6 on Monday. Hopefully the final one :-)

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support


--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Thu, 2008-09-18 at 10:09 -0400, Tom Lane wrote:
> Simon Riggs <simon@...> writes:
> > On Thu, 2008-09-18 at 09:06 -0400, Tom Lane wrote:
> >> Do we really need a checkpoint there at all?
>
> > "Timelines only change at shutdown checkpoints".
>
> Hmm.  I *think* that that is just a debugging crosscheck rather than a
> critical property.  But yeah, it would take some close investigation,
> which maybe isn't warranted if you have a less-invasive solution.

OK, new patch, version 6. Some major differences to previous patch.

* new IsRecoveryProcessingMode() in shmem
* padding in XLogCtl to ensure above call is cheap
* specific part of bgwriter shmem for passing restartpoint data
* avoid Shutdown checkpoint at end of recovery, with carefully
considered positioning of statements (beware!)
* only one new postmaster mode, PM_RECOVERY
* bgwriter changes state without stopping/starting

Modes I have tested so far
* make check
* Start, Stop
* Crash Recovery
* Archive Recovery
* Archive Recovery, switch in middle of restartpoint

Modes not yet tested
* EXEC_BACKEND

Ready for serious review prior to commit. I will be performing further
testing also.

 backend/access/transam/multixact.c |    2
 backend/access/transam/xlog.c      |  328 ++++++++++++---!!!!!!!!!!!!
 backend/postmaster/bgwriter.c      |  371 +++++---!!!!!!!!!!!!!!!!!!!!!
 backend/postmaster/postmaster.c    |   62 ++++!!
 backend/storage/buffer/README      |    5
 backend/storage/buffer/bufmgr.c    |   34 +!!
 include/access/xlog.h              |   14 !
 include/access/xlog_internal.h     |    3
 include/catalog/pg_control.h       |    2
 include/postmaster/bgwriter.h      |    2
 include/storage/bufmgr.h           |    2
 include/storage/pmsignal.h         |    1
 12 files changed, 279 insertions(+), 56 deletions(-), 491 mods(!)

There's a few subtle points along the way. I've tried to explain them
all in code comments, but questions welcome. At v6, most things are now
done a particular way for a specific reason.

Look especially at InRecovery, which is used extensively in different
parts of the code. The meaning of this has been subdivided into two
meanings, so only *some* of the places that use it have been changed.
All have been checked.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support

[recovery_infrastruc.v6.patch]

---

Index: src/backend/access/transam/multixact.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/access/transam/multixact.c,v
retrieving revision 1.28
diff -c -r1.28 multixact.c
*** src/backend/access/transam/multixact.c 1 Aug 2008 13:16:08 -0000 1.28
--- src/backend/access/transam/multixact.c 22 Sep 2008 19:28:56 -0000
***************
*** 1543,1549 ****
  * SimpleLruTruncate would get confused.  It seems best not to risk
  * removing any data during recovery anyway, so don't truncate.
  */
! if (!InRecovery)
  TruncateMultiXact();
 
  TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
--- 1543,1549 ----
  * SimpleLruTruncate would get confused.  It seems best not to risk
  * removing any data during recovery anyway, so don't truncate.
  */
! if (!IsRecoveryProcessingMode())
  TruncateMultiXact();
 
  TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
Index: src/backend/access/transam/xlog.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/access/transam/xlog.c,v
retrieving revision 1.317
diff -c -r1.317 xlog.c
*** src/backend/access/transam/xlog.c 11 Aug 2008 11:05:10 -0000 1.317
--- src/backend/access/transam/xlog.c 22 Sep 2008 21:30:24 -0000
***************
*** 119,124 ****
--- 119,125 ----
 
  /* Are we doing recovery from XLOG? */
  bool InRecovery = false;
+ bool reachedSafeStopPoint = false;
 
  /* Are we recovering using offline XLOG archives? */
  static bool InArchiveRecovery = false;
***************
*** 131,137 ****
  static bool recoveryTarget = false;
  static bool recoveryTargetExact = false;
  static bool recoveryTargetInclusive = true;
- static bool recoveryLogRestartpoints = false;
  static TransactionId recoveryTargetXid;
  static TimestampTz recoveryTargetTime;
  static TimestampTz recoveryLastXTime = 0;
--- 132,137 ----
***************
*** 286,295 ****
--- 286,297 ----
  /*
   * Total shared-memory state for XLOG.
   */
+ #define XLOGCTL_BUFFER_SPACING 128
  typedef struct XLogCtlData
  {
  /* Protected by WALInsertLock: */
  XLogCtlInsert Insert;
+ char InsertPadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogCtlInsert)];
 
  /* Protected by info_lck: */
  XLogwrtRqst LogwrtRqst;
***************
*** 297,305 ****
--- 299,314 ----
  uint32 ckptXidEpoch; /* nextXID & epoch of latest checkpoint */
  TransactionId ckptXid;
  XLogRecPtr asyncCommitLSN; /* LSN of newest async commit */
+ /* add data structure padding for above info_lck declarations */
+ char InfoPadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogwrtRqst)
+ - sizeof(XLogwrtResult)
+ - sizeof(uint32)
+ - sizeof(TransactionId)
+ - sizeof(XLogRecPtr)];
 
  /* Protected by WALWriteLock: */
  XLogCtlWrite Write;
+ char WritePadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogCtlWrite)];
 
  /*
  * These values do not change after startup, although the pointed-to pages
***************
*** 311,316 ****
--- 320,344 ----
  int XLogCacheBlck; /* highest allocated xlog buffer index */
  TimeLineID ThisTimeLineID;
 
+ /*
+ * IsRecoveryProcessingMode shows whether the postmaster is in a
+ * postmaster state earlier than PM_RUN, or not. This is a globally
+ * accessible state to allow EXEC_BACKEND case.
+ *
+ * We also retain a local state variable InRecovery. InRecovery=true
+ * means the code is being executed by Startup process and therefore
+ * always during RecoveryProcessingMode. This allows us to retain the
+ * often important distinction between code executed *during*
+ * RecoveryProcessingMode and but not necessarily by Startup process.
+ *
+ * Reviewer's note: all call points InRecovery and InRedo have been checked
+ * for correctness and have been changed to IsRecoveryProcessingMode()
+ * if appropriate.
+ */
+ bool IsRecoveryProcessingMode;
+
+ char InfoLockPadding[XLOGCTL_BUFFER_SPACING];
+
  slock_t info_lck; /* locks shared variables shown above */
  } XLogCtlData;
 
***************
*** 396,401 ****
--- 424,430 ----
  static void readRecoveryCommandFile(void);
  static void exitArchiveRecovery(TimeLineID endTLI,
  uint32 endLogId, uint32 endLogSeg);
+ static XLogRecPtr exitRecovery(void);
  static bool recoveryStopsHere(XLogRecord *record, bool *includeThis);
  static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
 
***************
*** 479,484 ****
--- 508,518 ----
  bool updrqst;
  bool doPageWrites;
  bool isLogSwitch = (rmid == RM_XLOG_ID && info == XLOG_SWITCH);
+ bool isRecoveryEnd = (rmid == RM_XLOG_ID && info == XLOG_RECOVERY_END);
+
+ /* cross-check on whether we should be here or not */
+ if (IsRecoveryProcessingMode() && !isRecoveryEnd)
+ elog(FATAL, "cannot make new WAL entries during recovery");
 
  /* info's high bits are reserved for use by me */
  if (info & XLR_INFO_MASK)
***************
*** 1677,1684 ****
  XLogRecPtr WriteRqstPtr;
  XLogwrtRqst WriteRqst;
 
! /* Disabled during REDO */
! if (InRedo)
  return;
 
  /* Quick exit if already known flushed */
--- 1711,1717 ----
  XLogRecPtr WriteRqstPtr;
  XLogwrtRqst WriteRqst;
 
! if (IsRecoveryProcessingMode())
  return;
 
  /* Quick exit if already known flushed */
***************
*** 1766,1774 ****
  * the bad page is encountered again during recovery then we would be
  * unable to restart the database at all!  (This scenario has actually
  * happened in the field several times with 7.1 releases. Note that we
! * cannot get here while InRedo is true, but if the bad page is brought in
! * and marked dirty during recovery then CreateCheckPoint will try to
! * flush it at the end of recovery.)
  *
  * The current approach is to ERROR under normal conditions, but only
  * WARNING during recovery, so that the system can be brought up even if
--- 1799,1807 ----
  * the bad page is encountered again during recovery then we would be
  * unable to restart the database at all!  (This scenario has actually
  * happened in the field several times with 7.1 releases. Note that we
! * cannot get here while IsRecoveryProcessingMode(), but if the bad page is
! * brought in and marked dirty during recovery then CreateCheckPoint will
! * try to flush it at the end of recovery.)
  *
  * The current approach is to ERROR under normal conditions, but only
  * WARNING during recovery, so that the system can be brought up even if
***************
*** 2051,2057 ****
  unlink(tmppath);
  }
 
! elog(DEBUG2, "done creating and filling new WAL file");
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
--- 2084,2091 ----
  unlink(tmppath);
  }
 
! XLogFileName(tmppath, ThisTimeLineID, log, seg);
! elog(DEBUG2, "done creating and filling new WAL file %s", tmppath);
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
***************
*** 4532,4546 ****
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
- /*
- * does nothing if a recovery_target is not also set
- */
- if (!parse_bool(tok2, &recoveryLogRestartpoints))
-  ereport(ERROR,
- (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-  errmsg("parameter \"log_restartpoints\" requires a Boolean value")));
  ereport(LOG,
! (errmsg("log_restartpoints = %s", tok2)));
  }
  else
  ereport(FATAL,
--- 4566,4574 ----
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
  ereport(LOG,
! (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
!  errmsg("parameter \"log_restartpoints\" has been deprecated")));
  }
  else
  ereport(FATAL,
***************
*** 4811,4828 ****
  CheckPoint checkPoint;
  bool wasShutdown;
  bool reachedStopPoint = false;
  bool haveBackupLabel = false;
  XLogRecPtr RecPtr,
  LastRec,
  checkPointLoc,
  minRecoveryLoc,
! EndOfLog;
  uint32 endLogId;
  uint32 endLogSeg;
  XLogRecord *record;
  uint32 freespace;
  TransactionId oldestActiveXID;
 
  /*
  * Read control file and check XLOG status looks valid.
  *
--- 4839,4860 ----
  CheckPoint checkPoint;
  bool wasShutdown;
  bool reachedStopPoint = false;
+ bool performedArchiveRecovery = false;
  bool haveBackupLabel = false;
  XLogRecPtr RecPtr,
  LastRec,
  checkPointLoc,
  minRecoveryLoc,
! EndOfLog,
! RecoveryCompletionPtr;
  uint32 endLogId;
  uint32 endLogSeg;
  XLogRecord *record;
  uint32 freespace;
  TransactionId oldestActiveXID;
 
+ XLogCtl->IsRecoveryProcessingMode = true;
+
  /*
  * Read control file and check XLOG status looks valid.
  *
***************
*** 5039,5044 ****
--- 5071,5081 ----
  UpdateControlFile();
 
  /*
+ * Reset pgstat data, because it may be invalid after recovery.
+ */
+ pgstat_reset_all();
+
+ /*
  * If there was a backup label file, it's done its job and the info
  * has now been propagated into pg_control.  We must get rid of the
  * label file so that if we crash during recovery, we'll pick up at
***************
*** 5148,5153 ****
--- 5185,5205 ----
 
  LastRec = ReadRecPtr;
 
+ /*
+ * Have we reached our safe stopping point? If so, we can
+ * signal Postmaster to enter consistent recovery mode
+ */
+ if (!reachedSafeStopPoint &&
+ XLByteLE(ControlFile->minRecoveryPoint, EndRecPtr))
+ {
+ reachedSafeStopPoint = true;
+ ereport(LOG,
+ (errmsg("consistent recovery state reached at %X/%X",
+ EndRecPtr.xlogid, EndRecPtr.xrecoff)));
+ if (IsUnderPostmaster)
+ SendPostmasterSignal(PMSIGNAL_RECOVERY_START);
+ }
+
  record = ReadRecord(NULL, LOG);
  } while (record != NULL && recoveryContinue);
 
***************
*** 5169,5174 ****
--- 5221,5227 ----
  /* there are no WAL records following the checkpoint */
  ereport(LOG,
  (errmsg("redo is not required")));
+ reachedSafeStopPoint = true;
  }
  }
 
***************
*** 5184,5190 ****
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (XLByteLT(EndOfLog, ControlFile->minRecoveryPoint))
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
--- 5237,5243 ----
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (InRecovery && !reachedSafeStopPoint)
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
***************
*** 5227,5233 ****
--- 5280,5289 ----
  * we will use that below.)
  */
  if (InArchiveRecovery)
+ {
+ performedArchiveRecovery = true;
  exitArchiveRecovery(curFileTLI, endLogId, endLogSeg);
+ }
 
  /*
  * Prepare to write WAL starting at EndOfLog position, and init xlog
***************
*** 5286,5291 ****
--- 5342,5349 ----
  /* Pre-scan prepared transactions to find out the range of XIDs present */
  oldestActiveXID = PrescanPreparedTransactions();
 
+ RecoveryCompletionPtr = EndOfLog;
+
  if (InRecovery)
  {
  int rmid;
***************
*** 5306,5343 ****
  XLogCheckInvalidPages();
 
  /*
! * Reset pgstat data, because it may be invalid after recovery.
! */
! pgstat_reset_all();
!
! /*
! * Perform a checkpoint to update all our recovery activity to disk.
! *
! * Note that we write a shutdown checkpoint rather than an on-line
! * one. This is not particularly critical, but since we may be
! * assigning a new TLI, using a shutdown checkpoint allows us to have
! * the rule that TLI only changes in shutdown checkpoints, which
! * allows some extra error checking in xlog_redo.
  */
! CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
  }
 
- /*
- * Preallocate additional log files, if wanted.
- */
- PreallocXlogFiles(EndOfLog);
-
- /*
- * Okay, we're officially UP.
- */
- InRecovery = false;
-
- ControlFile->state = DB_IN_PRODUCTION;
- ControlFile->time = (pg_time_t) time(NULL);
- UpdateControlFile();
-
  /* start the archive_timeout timer running */
! XLogCtl->Write.lastSegSwitchTime = ControlFile->time;
 
  /* initialize shared-memory copy of latest checkpoint XID/epoch */
  XLogCtl->ckptXidEpoch = ControlFile->checkPointCopy.nextXidEpoch;
--- 5364,5377 ----
  XLogCheckInvalidPages();
 
  /*
! * Finally exit recovery and mark that in WAL. Pre-8.4 we wrote
! * a shutdown checkpoint here, but we ask bgwriter to do that now.
  */
! RecoveryCompletionPtr = exitRecovery();
  }
 
  /* start the archive_timeout timer running */
! XLogCtl->Write.lastSegSwitchTime = (pg_time_t) time(NULL);
 
  /* initialize shared-memory copy of latest checkpoint XID/epoch */
  XLogCtl->ckptXidEpoch = ControlFile->checkPointCopy.nextXidEpoch;
***************
*** 5372,5377 ****
--- 5406,5446 ----
  readRecordBuf = NULL;
  readRecordBufSize = 0;
  }
+
+ /*
+ * Okay, we're officially UP.
+ */
+ XLogCtl->IsRecoveryProcessingMode = false;
+
+ /*
+ * If we had to perform archive recovery we don't mark the control file,
+ * yet, since we haven't definitely got a safe point to recover from that
+ * doesn't rely on archived WAL files. So we switch quickly into normal
+ * processing and rely on the bgwriter's checkpoint (NOT a restartpoint)
+ * to define a safe recovery point and put us into full production state.
+ * We specifically do not want to wait for checkpoint completion here,
+ * so we can reduce startup time in a standby mode replication failover.
+ * The checkpoint creation will also flush WAL, so we wait for that
+ * otherwise we may need to prepare WAL files ourselves.
+ *
+ * If we are doing crash recovery, we know we have WAL files accessible
+ * so we just get started again as quickly as possible.
+ */
+ if (performedArchiveRecovery)
+ RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_IMMEDIATE);
+ else
+ {
+ XLogFlush(RecoveryCompletionPtr);
+ ControlFile->state = DB_IN_PRODUCTION;
+ ControlFile->time = (pg_time_t) time(NULL);
+ UpdateControlFile();
+ }
+ }
+
+ bool
+ IsRecoveryProcessingMode(void)
+ {
+ return XLogCtl->IsRecoveryProcessingMode;
  }
 
  /*
***************
*** 5642,5648 ****
   * Log end of a checkpoint.
   */
  static void
! LogCheckpointEnd(void)
  {
  long write_secs,
  sync_secs,
--- 5711,5717 ----
   * Log end of a checkpoint.
   */
  static void
! LogCheckpointEnd(int flags)
  {
  long write_secs,
  sync_secs,
***************
*** 5665,5681 ****
  CheckpointStats.ckpt_sync_end_t,
  &sync_secs, &sync_usecs);
 
! elog(LOG, "checkpoint complete: wrote %d buffers (%.1f%%); "
! "%d transaction log file(s) added, %d removed, %d recycled; "
! "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
! CheckpointStats.ckpt_bufs_written,
! (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
! CheckpointStats.ckpt_segs_added,
! CheckpointStats.ckpt_segs_removed,
! CheckpointStats.ckpt_segs_recycled,
! write_secs, write_usecs / 1000,
! sync_secs, sync_usecs / 1000,
! total_secs, total_usecs / 1000);
  }
 
  /*
--- 5734,5759 ----
  CheckpointStats.ckpt_sync_end_t,
  &sync_secs, &sync_usecs);
 
! if (flags & CHECKPOINT_RESTARTPOINT)
! elog(LOG, "restartpoint complete: wrote %d buffers (%.1f%%); "
! "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
! CheckpointStats.ckpt_bufs_written,
! (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
! write_secs, write_usecs / 1000,
! sync_secs, sync_usecs / 1000,
! total_secs, total_usecs / 1000);
! else
! elog(LOG, "checkpoint complete: wrote %d buffers (%.1f%%); "
! "%d transaction log file(s) added, %d removed, %d recycled; "
! "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
! CheckpointStats.ckpt_bufs_written,
! (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
! CheckpointStats.ckpt_segs_added,
! CheckpointStats.ckpt_segs_removed,
! CheckpointStats.ckpt_segs_recycled,
! write_secs, write_usecs / 1000,
! sync_secs, sync_usecs / 1000,
! total_secs, total_usecs / 1000);
  }
 
  /*
***************
*** 5944,5949 ****
--- 6022,6029 ----
  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
  if (shutdown)
  ControlFile->state = DB_SHUTDOWNED;
+ else
+ ControlFile->state = DB_IN_PRODUCTION;
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
  ControlFile->checkPoint = ProcLastRecPtr;
  ControlFile->checkPointCopy = checkPoint;
***************
*** 6002,6008 ****
 
  /* All real work is done, but log before releasing lock. */
  if (log_checkpoints)
! LogCheckpointEnd();
 
  LWLockRelease(CheckpointLock);
  }
--- 6082,6088 ----
 
  /* All real work is done, but log before releasing lock. */
  if (log_checkpoints)
! LogCheckpointEnd(flags);
 
  LWLockRelease(CheckpointLock);
  }
***************
*** 6071,6099 ****
  }
  }
 
  /*
! * OK, force data out to disk
  */
! CheckPointGuts(checkPoint->redo, CHECKPOINT_IMMEDIATE);
 
  /*
! * Update pg_control so that any subsequent crash will restart from this
! * checkpoint. Note: ReadRecPtr gives the XLOG address of the checkpoint
! * record itself.
  */
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
! ControlFile->checkPoint = ReadRecPtr;
! ControlFile->checkPointCopy = *checkPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
 
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
! checkPoint->redo.xlogid, checkPoint->redo.xrecoff)));
! if (recoveryLastXTime)
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
! (errmsg("last completed transaction was at log time %s",
! timestamptz_to_str(recoveryLastXTime))));
  }
 
  /*
--- 6151,6215 ----
  }
  }
 
+ if (recoveryLastXTime)
+ ereport((log_checkpoints ? LOG : DEBUG2),
+ (errmsg("last completed transaction was at log time %s",
+ timestamptz_to_str(recoveryLastXTime))));
+
+ RequestRestartPoint(ReadRecPtr, checkPoint, reachedSafeStopPoint);
+ }
+
+ /*
+  * As of 8.4, RestartPoints are always created by the bgwriter
+  * once we have reachedSafeStopPoint. We use bgwriter's shared memory
+  * area wherever we call it from, to keep better code structure.
+  */
+ void
+ CreateRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint)
+ {
+ if (log_checkpoints)
+ {
+ /*
+ * Prepare to accumulate statistics.
+ */
+
+ MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
+
+ /*
+ * Do the restartpoint equivalent of LogCheckpointStart()
+ */
+ elog(LOG, "restartpoint starting");
+ }
+
+ LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
+
  /*
! * OK, write out dirty blocks smoothly
  */
! CheckPointGuts(restartPoint->redo, CHECKPOINT_RESTARTPOINT);
 
  /*
! * Update pg_control, using current time
  */
+ LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
! ControlFile->checkPoint = ReadPtr;
! ControlFile->checkPointCopy = *restartPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
+ LWLockRelease(ControlFileLock);
+
+ /* All real work is done, but log before releasing lock. */
+ if (log_checkpoints)
+ LogCheckpointEnd(CHECKPOINT_RESTARTPOINT);
 
! ereport((log_checkpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
! restartPoint->redo.xlogid, restartPoint->redo.xrecoff)));
!
! LWLockRelease(CheckpointLock);
!
  }
 
  /*
***************
*** 6158,6164 ****
  }
 
  /*
!  * XLOG resource manager's routines
   */
  void
  xlog_redo(XLogRecPtr lsn, XLogRecord *record)
--- 6274,6313 ----
  }
 
  /*
!  * exitRecovery()
!  *
!  * Exit recovery state and write a XLOG_RECOVERY_END record. This is the
!  * only record type that can record a change of timelineID. We assume
!  * caller has already set ThisTimeLineID, if appropriate.
!  */
! static XLogRecPtr
! exitRecovery(void)
! {
! XLogRecPtr RecPtr;
! XLogRecData rdata;
!
! rdata.buffer = InvalidBuffer;
! rdata.data = (char *) (&ThisTimeLineID);
! rdata.len = sizeof(TimeLineID);
! rdata.next = NULL;
!
! /*
! * This is the only type of WAL message that can be inserted during
! * recovery. This ensures that we don't allow others to get access
! * until after we have changed state.
! */
! RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RECOVERY_END, &rdata);
!
! InRecovery = false;
!
! return RecPtr;
! }
!
! /*
!  * XLOG resource manager's routines.
!  *
!  * Definitions of message info are in include/catalog/pg_control.h,
!  * though not all messages relate to control file processing.
   */
  void
  xlog_redo(XLogRecPtr lsn, XLogRecord *record)
***************
*** 6193,6213 ****
  ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
 
  /*
! * TLI may change in a shutdown checkpoint, but it shouldn't decrease
  */
! if (checkPoint.ThisTimeLineID != ThisTimeLineID)
  {
! if (checkPoint.ThisTimeLineID < ThisTimeLineID ||
  !list_member_int(expectedTLIs,
! (int) checkPoint.ThisTimeLineID))
  ereport(PANIC,
! (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
! checkPoint.ThisTimeLineID, ThisTimeLineID)));
  /* Following WAL records should be run with new TLI */
! ThisTimeLineID = checkPoint.ThisTimeLineID;
  }
-
- RecoveryRestartPoint(&checkPoint);
  }
  else if (info == XLOG_CHECKPOINT_ONLINE)
  {
--- 6342,6379 ----
  ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
 
  /*
! * TLI no longer changes at shutdown checkpoint, since as of 8.4,
! * shutdown checkpoints only occur at shutdown. Much less confusing.
  */
!
! RecoveryRestartPoint(&checkPoint);
! }
! else if (info == XLOG_RECOVERY_END)
! {
! TimeLineID tli;
!
! memcpy(&tli, XLogRecGetData(record), sizeof(TimeLineID));
!
! /*
! * TLI may change when recovery ends, but it shouldn't decrease.
! *
! * This is the only WAL record that can tell us to change timelineID
! * while we process WAL records.
! *
! * We can *choose* to stop recovery at any point, generating a
! * new timelineID which is recorded using this record type.
! */
! if (tli != ThisTimeLineID)
  {
! if (tli < ThisTimeLineID ||
  !list_member_int(expectedTLIs,
! (int) tli))
  ereport(PANIC,
! (errmsg("unexpected timeline ID %u (after %u) at recovery end record",
! tli, ThisTimeLineID)));
  /* Following WAL records should be run with new TLI */
! ThisTimeLineID = tli;
  }
  }
  else if (info == XLOG_CHECKPOINT_ONLINE)
  {
Index: src/backend/postmaster/bgwriter.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/bgwriter.c,v
retrieving revision 1.51
diff -c -r1.51 bgwriter.c
*** src/backend/postmaster/bgwriter.c 11 Aug 2008 11:05:11 -0000 1.51
--- src/backend/postmaster/bgwriter.c 22 Sep 2008 20:58:59 -0000
***************
*** 49,54 ****
--- 49,55 ----
  #include <unistd.h>
 
  #include "access/xlog_internal.h"
+ #include "catalog/pg_control.h"
  #include "libpq/pqsignal.h"
  #include "miscadmin.h"
  #include "pgstat.h"
***************
*** 130,135 ****
--- 131,143 ----
 
  int ckpt_flags; /* checkpoint flags, as defined in xlog.h */
 
+ /*
+ * When the Startup process wants bgwriter to perform a restartpoint, it
+ * sets these fields so that we can update the control file afterwards.
+ */
+ XLogRecPtr ReadPtr; /* ReadRecPtr for RestartPoint request */
+ CheckPoint *restartPoint; /* restartPoint data for ControlFile */
+
  uint32 num_backend_writes; /* counts non-bgwriter buffer writes */
 
  int num_requests; /* current # of requests */
***************
*** 166,172 ****
 
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr;
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
--- 174,180 ----
 
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr; /* not used if IsRecoveryProcessingMode */
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
***************
*** 198,203 ****
--- 206,212 ----
  {
  sigjmp_buf local_sigjmp_buf;
  MemoryContext bgwriter_context;
+ bool BgWriterRecoveryMode;
 
  BgWriterShmem->bgwriter_pid = MyProcPid;
  am_bg_writer = true;
***************
*** 356,371 ****
  */
  PG_SETMASK(&UnBlockSig);
 
  /*
  * Loop forever
  */
  for (;;)
  {
- bool do_checkpoint = false;
- int flags = 0;
- pg_time_t now;
- int elapsed_secs;
-
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
--- 365,381 ----
  */
  PG_SETMASK(&UnBlockSig);
 
+ BgWriterRecoveryMode = IsRecoveryProcessingMode();
+
+ if (BgWriterRecoveryMode)
+ elog(DEBUG1, "bgwriter starting during recovery, pid = %u",
+ BgWriterShmem->bgwriter_pid);
+
  /*
  * Loop forever
  */
  for (;;)
  {
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
***************
*** 383,501 ****
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
- if (checkpoint_requested)
- {
- checkpoint_requested = false;
- do_checkpoint = true;
- BgWriterStats.m_requested_checkpoints++;
- }
- if (shutdown_requested)
- {
- /*
- * From here on, elog(ERROR) should end with exit(1), not send
- * control back to the sigsetjmp block above
- */
- ExitOnAnyError = true;
- /* Close down the database */
- ShutdownXLOG(0, 0);
- DumpFreeSpaceMap(0, 0);
- /* Normal exit from the bgwriter is here */
- proc_exit(0); /* done */
- }
 
! /*
! * Force a checkpoint if too much time has elapsed since the last one.
! * Note that we count a timed checkpoint in stats only when this
! * occurs without an external request, but we set the CAUSE_TIME flag
! * bit even if there is also an external request.
! */
! now = (pg_time_t) time(NULL);
! elapsed_secs = now - last_checkpoint_time;
! if (elapsed_secs >= CheckPointTimeout)
  {
! if (!do_checkpoint)
! BgWriterStats.m_timed_checkpoints++;
! do_checkpoint = true;
! flags |= CHECKPOINT_CAUSE_TIME;
  }
!
! /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
! */
! if (do_checkpoint)
  {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
 
  /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
  */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
!
! /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
! */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
!
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
 
  /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
  */
! smgrcloseall();
 
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
 
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
  }
- else
- BgBufferSync();
-
- /* Check for archive_timeout and switch xlog files if necessary. */
- CheckArchiveTimeout();
-
- /* Nap for the configured time. */
- BgWriterNap();
  }
  }
 
--- 393,592 ----
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
 
! if (BgWriterRecoveryMode)
  {
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
!
! if (!IsRecoveryProcessingMode())
! {
! elog(DEBUG2, "bgwriter changing from recovery to normal mode");
!
! InitXLOGAccess();
! BgWriterRecoveryMode = false;
!
! /*
! * Start time-driven events from now
! */
! last_checkpoint_time = last_xlog_switch_time = (pg_time_t) time(NULL);
!
! continue;
! }
!
! if (checkpoint_requested)
! {
! XLogRecPtr ReadPtr;
! CheckPoint restartPoint;
!
! checkpoint_requested = false;
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_time = (pg_time_t) time(NULL);
! ckpt_cached_elapsed = 0;
!
! /*
! * Get the requested values from shared memory that the
! * Startup process has put there for us.
! */
! SpinLockAcquire(&BgWriterShmem->ckpt_lck);
! ReadPtr = BgWriterShmem->ReadPtr;
! memcpy(&restartPoint, &BgWriterShmem->restartPoint, sizeof(CheckPoint));
! SpinLockRelease(&BgWriterShmem->ckpt_lck);
!
! CreateRestartPoint(ReadPtr, &restartPoint);
!
! ckpt_active = false;
!
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
! }
! else
! {
! /* Clean buffers dirtied by recovery */
! BgBufferSync(true);
!
! /* Nap for the configured time. */
! BgWriterNap();
! }
  }
! else /* Normal processing */
  {
! bool do_checkpoint = false;
! int flags = 0;
! pg_time_t now;
! int elapsed_secs;
!
! Assert(!IsRecoveryProcessingMode());
!
! if (checkpoint_requested)
! {
! checkpoint_requested = false;
! do_checkpoint = true;
! BgWriterStats.m_requested_checkpoints++;
! }
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Close down the database */
! ShutdownXLOG(0, 0);
! DumpFreeSpaceMap(0, 0);
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
 
  /*
! * Force a checkpoint if too much time has elapsed since the last one.
! * Note that we count a timed checkpoint in stats only when this
! * occurs without an external request, but we set the CAUSE_TIME flag
! * bit even if there is also an external request.
  */
! now = (pg_time_t) time(NULL);
! elapsed_secs = now - last_checkpoint_time;
! if (elapsed_secs >= CheckPointTimeout)
! {
! if (!do_checkpoint)
! BgWriterStats.m_timed_checkpoints++;
! do_checkpoint = true;
! flags |= CHECKPOINT_CAUSE_TIME;
! }
 
  /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
  */
! if (do_checkpoint)
! {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
!
! /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
!
! /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
! */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
!
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
!
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
!
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
!
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
! }
! else
! BgBufferSync(false);
 
! /* Check for archive_timeout and switch xlog files if necessary. */
! CheckArchiveTimeout();
 
! /* Nap for the configured time. */
! BgWriterNap();
  }
  }
  }
 
***************
*** 588,594 ****
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
--- 679,686 ----
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! if (!IsRecoveryProcessingMode())
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
***************
*** 642,647 ****
--- 734,751 ----
  if (!am_bg_writer)
  return;
 
+ /* Perform minimal duties during recovery and skip wait if requested */
+ if (IsRecoveryProcessingMode())
+ {
+ BgBufferSync(true);
+
+ if (!shutdown_requested &&
+ IsCheckpointOnSchedule(progress))
+ BgWriterNap();
+
+ return;
+ }
+
  /*
  * Perform the usual bgwriter duties and take a nap, unless we're behind
  * schedule, in which case we just try to catch up as quickly as possible.
***************
*** 660,666 ****
  AbsorbFsyncRequests();
  absorb_counter = WRITES_PER_ABSORB;
 
! BgBufferSync();
  CheckArchiveTimeout();
  BgWriterNap();
  }
--- 764,770 ----
  AbsorbFsyncRequests();
  absorb_counter = WRITES_PER_ABSORB;
 
! BgBufferSync(false);
  CheckArchiveTimeout();
  BgWriterNap();
  }
***************
*** 716,731 ****
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
  {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
  }
 
  /*
--- 820,838 ----
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! if (!IsRecoveryProcessingMode())
  {
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
! {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
! }
  }
 
  /*
***************
*** 967,972 ****
--- 1074,1109 ----
  }
 
  /*
+  * Always runs in Startup process (see xlog.c)
+  */
+ void
+ RequestRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, bool sendToBGWriter)
+ {
+ /*
+ * Should we just do it ourselves?
+ */
+ if (!IsPostmasterEnvironment || !sendToBGWriter)
+ {
+ CreateRestartPoint(ReadPtr, restartPoint);
+ return;
+ }
+
+ /*
+ * Push requested values into shared memory, then signal to request restartpoint.
+ */
+ if (BgWriterShmem->bgwriter_pid == 0)
+ elog(LOG, "could not request restartpoint because bgwriter not running");
+
+ SpinLockAcquire(&BgWriterShmem->ckpt_lck);
+ BgWriterShmem->ReadPtr = ReadPtr;
+ memcpy(&BgWriterShmem->restartPoint, restartPoint, sizeof(CheckPoint));
+ SpinLockRelease(&BgWriterShmem->ckpt_lck);
+
+ if (kill(BgWriterShmem->bgwriter_pid, SIGINT) != 0)
+ elog(LOG, "could not signal for restartpoint: %m");
+ }
+
+ /*
   * ForwardFsyncRequest
   * Forward a file-fsync request from a backend to the bgwriter
   *
Index: src/backend/postmaster/postmaster.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/postmaster.c,v
retrieving revision 1.561
diff -c -r1.561 postmaster.c
*** src/backend/postmaster/postmaster.c 26 Jun 2008 02:47:19 -0000 1.561
--- src/backend/postmaster/postmaster.c 22 Sep 2008 20:25:36 -0000
***************
*** 254,259 ****
--- 254,264 ----
  {
  PM_INIT, /* postmaster starting */
  PM_STARTUP, /* waiting for startup subprocess */
+ PM_RECOVERY, /* consistent recovery mode; state only
+ * entered for archive and streaming recovery,
+ * and only after the point where the
+ * all data is in consistent state.
+ */
  PM_RUN, /* normal "database is alive" state */
  PM_WAIT_BACKUP, /* waiting for online backup mode to end */
  PM_WAIT_BACKENDS, /* waiting for live backends to exit */
***************
*** 1294,1300 ****
  * state that prevents it, start one.  It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0 && pmState == PM_RUN)
  BgWriterPID = StartBackgroundWriter();
 
  /*
--- 1299,1305 ----
  * state that prevents it, start one.  It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0 && (pmState == PM_RUN || pmState == PM_RECOVERY))
  BgWriterPID = StartBackgroundWriter();
 
  /*
***************
*** 2104,2110 ****
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
--- 2109,2115 ----
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP || pmState == PM_RECOVERY);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
***************
*** 2145,2155 ****
  load_role();
 
  /*
! * Crank up the background writer. It doesn't matter if this
! * fails, we'll just try again later.
  */
! Assert(BgWriterPID == 0);
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
--- 2150,2160 ----
  load_role();
 
  /*
! * Check whether we need to start background writer, if not
! * already running.
  */
! if (BgWriterPID == 0)
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
***************
*** 3821,3826 ****
--- 3826,3876 ----
 
  PG_SETMASK(&BlockSig);
 
+ if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_START))
+ {
+ Assert(pmState == PM_STARTUP);
+
+ /*
+ * Go to shutdown mode if a shutdown request was pending.
+ */
+ if (Shutdown > NoShutdown)
+ {
+ pmState = PM_WAIT_BACKENDS;
+ /* PostmasterStateMachine logic does the rest */
+ }
+ else
+ {
+ /*
+ * Startup process has entered recovery
+ */
+ pmState = PM_RECOVERY;
+
+ /*
+ * Load the flat authorization file into postmaster's cache. The
+ * startup process won't have recomputed this from the database yet,
+ * so we it may change following recovery.
+ */
+ load_role();
+
+ /*
+ * Crank up the background writer. It doesn't matter if this
+ * fails, we'll just try again later.
+ */
+ Assert(BgWriterPID == 0);
+ BgWriterPID = StartBackgroundWriter();
+
+ /*
+ * Likewise, start other special children as needed.
+ */
+ Assert(PgStatPID == 0);
+ PgStatPID = pgstat_start();
+
+ /* XXX at this point we could accept read-only connections */
+ ereport(DEBUG1,
+ (errmsg("database system is in consistent recovery mode")));
+ }
+ }
+
  if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
  {
  /*
Index: src/backend/storage/buffer/README
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/storage/buffer/README,v
retrieving revision 1.14
diff -c -r1.14 README
*** src/backend/storage/buffer/README 21 Mar 2008 13:23:28 -0000 1.14
--- src/backend/storage/buffer/README 22 Sep 2008 21:54:20 -0000
***************
*** 264,266 ****
--- 264,271 ----
  This ensures that the page image transferred to disk is reasonably consistent.
  We might miss a hint-bit update or two but that isn't a problem, for the same
  reasons mentioned under buffer access rules.
+
+ As of 8.4, background writer starts during recovery mode when there is
+ some form of potentially extended recovery to perform. It performs an
+ identical service to normal processing, except that checkpoints it
+ writes are technically restartpoints nor needs to flush WAL for dirty buffers.
Index: src/backend/storage/buffer/bufmgr.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/storage/buffer/bufmgr.c,v
retrieving revision 1.237
diff -c -r1.237 bufmgr.c
*** src/backend/storage/buffer/bufmgr.c 11 Aug 2008 11:05:11 -0000 1.237
--- src/backend/storage/buffer/bufmgr.c 22 Sep 2008 21:01:02 -0000
***************
*** 1080,1086 ****
   *
   * This is called at checkpoint time to write out all dirty shared buffers.
   * The checkpoint request flags should be passed in; currently the only one
!  * examined is CHECKPOINT_IMMEDIATE, which disables delays between writes.
   */
  static void
  BufferSync(int flags)
--- 1080,1087 ----
   *
   * This is called at checkpoint time to write out all dirty shared buffers.
   * The checkpoint request flags should be passed in; currently the only one
!  * examined is CHECKPOINT_IMMEDIATE, which disables delays between writes,
!  * except for restartpoints, which use CHECKPOINT_RESTARTPOINT.
   */
  static void
  BufferSync(int flags)
***************
*** 1089,1094 ****
--- 1090,1096 ----
  int num_to_scan;
  int num_to_write;
  int num_written;
+ bool cleanup_and_exit = false;
 
  /* Make sure we can handle the pin inside SyncOneBuffer */
  ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
***************
*** 1163,1169 ****
  */
  if (bufHdr->flags & BM_CHECKPOINT_NEEDED)
  {
! if (SyncOneBuffer(buf_id, false) & BUF_WRITTEN)
  {
  TRACE_POSTGRESQL_BUFFER_SYNC_WRITTEN(buf_id);
  BgWriterStats.m_buf_written_checkpoints++;
--- 1165,1180 ----
  */
  if (bufHdr->flags & BM_CHECKPOINT_NEEDED)
  {
! if (cleanup_and_exit)
! {
! LockBufHdr(bufHdr);
!
! if (bufHdr->flags & BM_CHECKPOINT_NEEDED)
! bufHdr->flags &= ~BM_CHECKPOINT_NEEDED;
!
! UnlockBufHdr(bufHdr);
! }
! else if (SyncOneBuffer(buf_id, false) & BUF_WRITTEN)
  {
  TRACE_POSTGRESQL_BUFFER_SYNC_WRITTEN(buf_id);
  BgWriterStats.m_buf_written_checkpoints++;
***************
*** 1194,1199 ****
--- 1205,1218 ----
 
  if (++buf_id >= NBuffers)
  buf_id = 0;
+
+ /*
+ * Quit scanning if state changes while we're here. If it
+ * does we want to stop performing the current restartpoint
+ * as quickly as possible. We'll write a real checkpoint instead.
+ */
+ if (flags & CHECKPOINT_RESTARTPOINT && !IsRecoveryProcessingMode())
+ cleanup_and_exit = true;
  }
 
  TRACE_POSTGRESQL_BUFFER_SYNC_DONE(NBuffers, num_written, num_to_write);
***************
*** 1211,1217 ****
   * This is called periodically by the background writer process.
   */
  void
! BgBufferSync(void)
  {
  /* info obtained from freelist.c */
  int strategy_buf_id;
--- 1230,1236 ----
   * This is called periodically by the background writer process.
   */
  void
! BgBufferSync(bool calledInRecovery)
  {
  /* info obtained from freelist.c */
  int strategy_buf_id;
***************
*** 1423,1428 ****
--- 1442,1456 ----
  {
  int buffer_state = SyncOneBuffer(next_to_clean, true);
 
+ /*
+ * Quit scanning if state changes while we're here, otherwise
+ * we might skip writing WAL for a buffer that was modified
+ * by a transaction that committed very soon after recovery
+ * mode completes.
+ */
+ if (calledInRecovery && !IsRecoveryProcessingMode())
+ break;
+
  if (++next_to_clean >= NBuffers)
  {
  next_to_clean = 0;
Index: src/include/access/xlog.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog.h,v
retrieving revision 1.88
diff -c -r1.88 xlog.h
*** src/include/access/xlog.h 12 May 2008 08:35:05 -0000 1.88
--- src/include/access/xlog.h 22 Sep 2008 19:04:56 -0000
***************
*** 133,139 ****
  } XLogRecData;
 
  extern TimeLineID ThisTimeLineID; /* current TLI */
! extern bool InRecovery;
  extern XLogRecPtr XactLastRecEnd;
 
  /* these variables are GUC parameters related to XLOG */
--- 133,148 ----
  } XLogRecData;
 
  extern TimeLineID ThisTimeLineID; /* current TLI */
!
! /*
!  * Prior to 8.4, all activity during recovery were carried out by Startup
!  * process. This local variable continues to be used in many parts of the
!  * code to indicate actions taken by RecoveryManagers. Other processes who
!  * potentially perform work during recovery should check
!  * IsRecoveryProcessingMode(), see XLogCtl notes in xlog.c
!  */
! extern bool InRecovery;
!
  extern XLogRecPtr XactLastRecEnd;
 
  /* these variables are GUC parameters related to XLOG */
***************
*** 166,171 ****
--- 175,181 ----
  /* These indicate the cause of a checkpoint request */
  #define CHECKPOINT_CAUSE_XLOG 0x0010 /* XLOG consumption */
  #define CHECKPOINT_CAUSE_TIME 0x0020 /* Elapsed time */
+ #define CHECKPOINT_RESTARTPOINT 0x0040 /* Restartpoint during recovery */
 
  /* Checkpoint statistics */
  typedef struct CheckpointStatsData
***************
*** 197,202 ****
--- 207,214 ----
  extern void xlog_redo(XLogRecPtr lsn, XLogRecord *record);
  extern void xlog_desc(StringInfo buf, uint8 xl_info, char *rec);
 
+ extern bool IsRecoveryProcessingMode(void);
+
  extern void UpdateControlFile(void);
  extern Size XLOGShmemSize(void);
  extern void XLOGShmemInit(void);
Index: src/include/access/xlog_internal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog_internal.h,v
retrieving revision 1.24
diff -c -r1.24 xlog_internal.h
*** src/include/access/xlog_internal.h 11 Aug 2008 11:05:11 -0000 1.24
--- src/include/access/xlog_internal.h 22 Sep 2008 20:26:56 -0000
***************
*** 17,22 ****
--- 17,23 ----
  #define XLOG_INTERNAL_H
 
  #include "access/xlog.h"
+ #include "catalog/pg_control.h"
  #include "fmgr.h"
  #include "pgtime.h"
  #include "storage/block.h"
***************
*** 245,250 ****
--- 246,253 ----
  extern pg_time_t GetLastSegSwitchTime(void);
  extern XLogRecPtr RequestXLogSwitch(void);
 
+ extern void CreateRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint);
+
  /*
   * These aren't in xlog.h because I'd rather not include fmgr.h there.
   */
Index: src/include/catalog/pg_control.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/catalog/pg_control.h,v
retrieving revision 1.41
diff -c -r1.41 pg_control.h
*** src/include/catalog/pg_control.h 21 Apr 2008 00:26:47 -0000 1.41
--- src/include/catalog/pg_control.h 22 Sep 2008 19:21:37 -0000
***************
*** 46,52 ****
  #define XLOG_NOOP 0x20
  #define XLOG_NEXTOID 0x30
  #define XLOG_SWITCH 0x40
!
 
  /* System status indicator */
  typedef enum DBState
--- 46,52 ----
  #define XLOG_NOOP 0x20
  #define XLOG_NEXTOID 0x30
  #define XLOG_SWITCH 0x40
! #define XLOG_RECOVERY_END 0x50
 
  /* System status indicator */
  typedef enum DBState
Index: src/include/postmaster/bgwriter.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/postmaster/bgwriter.h,v
retrieving revision 1.12
diff -c -r1.12 bgwriter.h
*** src/include/postmaster/bgwriter.h 11 Aug 2008 11:05:11 -0000 1.12
--- src/include/postmaster/bgwriter.h 22 Sep 2008 15:53:22 -0000
***************
*** 12,17 ****
--- 12,18 ----
  #ifndef _BGWRITER_H
  #define _BGWRITER_H
 
+ #include "catalog/pg_control.h"
  #include "storage/block.h"
  #include "storage/relfilenode.h"
 
***************
*** 25,30 ****
--- 26,32 ----
  extern void BackgroundWriterMain(void);
 
  extern void RequestCheckpoint(int flags);
+ extern void RequestRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, bool sendToBGWriter);
  extern void CheckpointWriteDelay(int flags, double progress);
 
  extern bool ForwardFsyncRequest(RelFileNode rnode, ForkNumber forknum,
Index: src/include/storage/bufmgr.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/storage/bufmgr.h,v
retrieving revision 1.115
diff -c -r1.115 bufmgr.h
*** src/include/storage/bufmgr.h 11 Aug 2008 11:05:11 -0000 1.115
--- src/include/storage/bufmgr.h 22 Sep 2008 18:36:54 -0000
***************
*** 193,199 ****
  extern void AbortBufferIO(void);
 
  extern void BufmgrCommit(void);
! extern void BgBufferSync(void);
 
  extern void AtProcExit_LocalBuffers(void);
 
--- 193,199 ----
  extern void AbortBufferIO(void);
 
  extern void BufmgrCommit(void);
! extern void BgBufferSync(bool calledInRecovery);
 
  extern void AtProcExit_LocalBuffers(void);
 
Index: src/include/storage/pmsignal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/storage/pmsignal.h,v
retrieving revision 1.20
diff -c -r1.20 pmsignal.h
*** src/include/storage/pmsignal.h 19 Jun 2008 21:32:56 -0000 1.20
--- src/include/storage/pmsignal.h 22 Sep 2008 15:54:57 -0000
***************
*** 22,27 ****
--- 22,28 ----
   */
  typedef enum
  {
+ PMSIGNAL_RECOVERY_START, /* move to PM_RECOVERY state */
  PMSIGNAL_PASSWORD_CHANGE, /* pg_auth file has changed */
  PMSIGNAL_WAKEN_ARCHIVER, /* send a NOTIFY signal to xlog archiver */
  PMSIGNAL_ROTATE_LOGFILE, /* send SIGUSR1 to syslogger to rotate logfile */

---

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Simon Riggs]

On Mon, 2008-09-22 at 23:06 +0100, Simon Riggs wrote:

> On Thu, 2008-09-18 at 10:09 -0400, Tom Lane wrote:
> > Simon Riggs <simon@...> writes:
> > > On Thu, 2008-09-18 at 09:06 -0400, Tom Lane wrote:
> > >> Do we really need a checkpoint there at all?
> >
> > > "Timelines only change at shutdown checkpoints".
> >
> > Hmm.  I *think* that that is just a debugging crosscheck rather than a
> > critical property.  But yeah, it would take some close investigation,
> > which maybe isn't warranted if you have a less-invasive solution.
>
> OK, new patch, version 6. Some major differences to previous patch.

> Ready for serious review prior to commit. I will be performing further
> testing also.

Version 7

I've removed the concept of interrupting a restartpoint half way
through, I found a fault there. It was more ugly than the alternative
and less robust. The code now waits at the end of recovery if we are in
the middle of a restartpoint, but forces a do-it-more-quickly also. That
means we won't always get a fast start even though we skip the shutdown
checkpoint, but at least we're sure there's no chance of breakage
because of concurrent activiy, state changes etc..

I'm happy with this now.

--
 Simon Riggs           www.2ndQuadrant.com
 PostgreSQL Training, Services and Support

[recovery_infrastruc.v7.patch]

---

Index: src/backend/access/transam/multixact.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/access/transam/multixact.c,v
retrieving revision 1.28
diff -c -r1.28 multixact.c
*** src/backend/access/transam/multixact.c 1 Aug 2008 13:16:08 -0000 1.28
--- src/backend/access/transam/multixact.c 22 Sep 2008 19:28:56 -0000
***************
*** 1543,1549 ****
  * SimpleLruTruncate would get confused.  It seems best not to risk
  * removing any data during recovery anyway, so don't truncate.
  */
! if (!InRecovery)
  TruncateMultiXact();
 
  TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
--- 1543,1549 ----
  * SimpleLruTruncate would get confused.  It seems best not to risk
  * removing any data during recovery anyway, so don't truncate.
  */
! if (!IsRecoveryProcessingMode())
  TruncateMultiXact();
 
  TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
Index: src/backend/access/transam/xlog.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/access/transam/xlog.c,v
retrieving revision 1.317
diff -c -r1.317 xlog.c
*** src/backend/access/transam/xlog.c 11 Aug 2008 11:05:10 -0000 1.317
--- src/backend/access/transam/xlog.c 23 Sep 2008 14:56:37 -0000
***************
*** 66,76 ****
  bool fullPageWrites = true;
  bool log_checkpoints = false;
  int sync_method = DEFAULT_SYNC_METHOD;
-
  #ifdef WAL_DEBUG
  bool XLOG_DEBUG = false;
  #endif
-
  /*
   * XLOGfileslop is the maximum number of preallocated future XLOG segments.
   * When we are done with an old XLOG segment file, we will recycle it as a
--- 66,74 ----
***************
*** 119,124 ****
--- 117,123 ----
 
  /* Are we doing recovery from XLOG? */
  bool InRecovery = false;
+ bool reachedSafeStopPoint = false;
 
  /* Are we recovering using offline XLOG archives? */
  static bool InArchiveRecovery = false;
***************
*** 131,137 ****
  static bool recoveryTarget = false;
  static bool recoveryTargetExact = false;
  static bool recoveryTargetInclusive = true;
- static bool recoveryLogRestartpoints = false;
  static TransactionId recoveryTargetXid;
  static TimestampTz recoveryTargetTime;
  static TimestampTz recoveryLastXTime = 0;
--- 130,135 ----
***************
*** 286,295 ****
--- 284,295 ----
  /*
   * Total shared-memory state for XLOG.
   */
+ #define XLOGCTL_BUFFER_SPACING 128
  typedef struct XLogCtlData
  {
  /* Protected by WALInsertLock: */
  XLogCtlInsert Insert;
+ char InsertPadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogCtlInsert)];
 
  /* Protected by info_lck: */
  XLogwrtRqst LogwrtRqst;
***************
*** 297,305 ****
--- 297,312 ----
  uint32 ckptXidEpoch; /* nextXID & epoch of latest checkpoint */
  TransactionId ckptXid;
  XLogRecPtr asyncCommitLSN; /* LSN of newest async commit */
+ /* add data structure padding for above info_lck declarations */
+ char InfoPadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogwrtRqst)
+ - sizeof(XLogwrtResult)
+ - sizeof(uint32)
+ - sizeof(TransactionId)
+ - sizeof(XLogRecPtr)];
 
  /* Protected by WALWriteLock: */
  XLogCtlWrite Write;
+ char WritePadding[XLOGCTL_BUFFER_SPACING - sizeof(XLogCtlWrite)];
 
  /*
  * These values do not change after startup, although the pointed-to pages
***************
*** 311,316 ****
--- 318,342 ----
  int XLogCacheBlck; /* highest allocated xlog buffer index */
  TimeLineID ThisTimeLineID;
 
+ /*
+ * IsRecoveryProcessingMode shows whether the postmaster is in a
+ * postmaster state earlier than PM_RUN, or not. This is a globally
+ * accessible state to allow EXEC_BACKEND case.
+ *
+ * We also retain a local state variable InRecovery. InRecovery=true
+ * means the code is being executed by Startup process and therefore
+ * always during RecoveryProcessingMode. This allows us to retain the
+ * often important distinction between code executed *during*
+ * RecoveryProcessingMode and but not necessarily by Startup process.
+ *
+ * Reviewer's note: all call points InRecovery and InRedo have been checked
+ * for correctness and have been changed to IsRecoveryProcessingMode()
+ * if appropriate.
+ */
+ bool IsRecoveryProcessingMode;
+
+ char InfoLockPadding[XLOGCTL_BUFFER_SPACING];
+
  slock_t info_lck; /* locks shared variables shown above */
  } XLogCtlData;
 
***************
*** 396,401 ****
--- 422,428 ----
  static void readRecoveryCommandFile(void);
  static void exitArchiveRecovery(TimeLineID endTLI,
  uint32 endLogId, uint32 endLogSeg);
+ static XLogRecPtr exitRecovery(void);
  static bool recoveryStopsHere(XLogRecord *record, bool *includeThis);
  static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
 
***************
*** 479,484 ****
--- 506,516 ----
  bool updrqst;
  bool doPageWrites;
  bool isLogSwitch = (rmid == RM_XLOG_ID && info == XLOG_SWITCH);
+ bool isRecoveryEnd = (rmid == RM_XLOG_ID && info == XLOG_RECOVERY_END);
+
+ /* cross-check on whether we should be here or not */
+ if (IsRecoveryProcessingMode() && !isRecoveryEnd)
+ elog(FATAL, "cannot make new WAL entries during recovery");
 
  /* info's high bits are reserved for use by me */
  if (info & XLR_INFO_MASK)
***************
*** 1677,1684 ****
  XLogRecPtr WriteRqstPtr;
  XLogwrtRqst WriteRqst;
 
! /* Disabled during REDO */
! if (InRedo)
  return;
 
  /* Quick exit if already known flushed */
--- 1709,1715 ----
  XLogRecPtr WriteRqstPtr;
  XLogwrtRqst WriteRqst;
 
! if (IsRecoveryProcessingMode())
  return;
 
  /* Quick exit if already known flushed */
***************
*** 1766,1774 ****
  * the bad page is encountered again during recovery then we would be
  * unable to restart the database at all!  (This scenario has actually
  * happened in the field several times with 7.1 releases. Note that we
! * cannot get here while InRedo is true, but if the bad page is brought in
! * and marked dirty during recovery then CreateCheckPoint will try to
! * flush it at the end of recovery.)
  *
  * The current approach is to ERROR under normal conditions, but only
  * WARNING during recovery, so that the system can be brought up even if
--- 1797,1805 ----
  * the bad page is encountered again during recovery then we would be
  * unable to restart the database at all!  (This scenario has actually
  * happened in the field several times with 7.1 releases. Note that we
! * cannot get here while IsRecoveryProcessingMode(), but if the bad page is
! * brought in and marked dirty during recovery then CreateCheckPoint will
! * try to flush it at the end of recovery.)
  *
  * The current approach is to ERROR under normal conditions, but only
  * WARNING during recovery, so that the system can be brought up even if
***************
*** 2051,2057 ****
  unlink(tmppath);
  }
 
! elog(DEBUG2, "done creating and filling new WAL file");
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
--- 2082,2089 ----
  unlink(tmppath);
  }
 
! XLogFileName(tmppath, ThisTimeLineID, log, seg);
! elog(DEBUG2, "done creating and filling new WAL file %s", tmppath);
 
  /* Set flag to tell caller there was no existent file */
  *use_existent = false;
***************
*** 4532,4546 ****
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
- /*
- * does nothing if a recovery_target is not also set
- */
- if (!parse_bool(tok2, &recoveryLogRestartpoints))
-  ereport(ERROR,
- (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-  errmsg("parameter \"log_restartpoints\" requires a Boolean value")));
  ereport(LOG,
! (errmsg("log_restartpoints = %s", tok2)));
  }
  else
  ereport(FATAL,
--- 4564,4572 ----
  }
  else if (strcmp(tok1, "log_restartpoints") == 0)
  {
  ereport(LOG,
! (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
!  errmsg("parameter \"log_restartpoints\" has been deprecated")));
  }
  else
  ereport(FATAL,
***************
*** 4811,4828 ****
  CheckPoint checkPoint;
  bool wasShutdown;
  bool reachedStopPoint = false;
  bool haveBackupLabel = false;
  XLogRecPtr RecPtr,
  LastRec,
  checkPointLoc,
  minRecoveryLoc,
! EndOfLog;
  uint32 endLogId;
  uint32 endLogSeg;
  XLogRecord *record;
  uint32 freespace;
  TransactionId oldestActiveXID;
 
  /*
  * Read control file and check XLOG status looks valid.
  *
--- 4837,4858 ----
  CheckPoint checkPoint;
  bool wasShutdown;
  bool reachedStopPoint = false;
+ bool performedArchiveRecovery = false;
  bool haveBackupLabel = false;
  XLogRecPtr RecPtr,
  LastRec,
  checkPointLoc,
  minRecoveryLoc,
! EndOfLog,
! RecoveryCompletionPtr;
  uint32 endLogId;
  uint32 endLogSeg;
  XLogRecord *record;
  uint32 freespace;
  TransactionId oldestActiveXID;
 
+ XLogCtl->IsRecoveryProcessingMode = true;
+
  /*
  * Read control file and check XLOG status looks valid.
  *
***************
*** 5039,5044 ****
--- 5069,5079 ----
  UpdateControlFile();
 
  /*
+ * Reset pgstat data, because it may be invalid after recovery.
+ */
+ pgstat_reset_all();
+
+ /*
  * If there was a backup label file, it's done its job and the info
  * has now been propagated into pg_control.  We must get rid of the
  * label file so that if we crash during recovery, we'll pick up at
***************
*** 5148,5153 ****
--- 5183,5203 ----
 
  LastRec = ReadRecPtr;
 
+ /*
+ * Have we reached our safe stopping point? If so, we can
+ * signal Postmaster to enter consistent recovery mode
+ */
+ if (!reachedSafeStopPoint &&
+ XLByteLE(ControlFile->minRecoveryPoint, EndRecPtr))
+ {
+ reachedSafeStopPoint = true;
+ ereport(LOG,
+ (errmsg("consistent recovery state reached at %X/%X",
+ EndRecPtr.xlogid, EndRecPtr.xrecoff)));
+ if (IsUnderPostmaster)
+ SendPostmasterSignal(PMSIGNAL_RECOVERY_START);
+ }
+
  record = ReadRecord(NULL, LOG);
  } while (record != NULL && recoveryContinue);
 
***************
*** 5169,5174 ****
--- 5219,5225 ----
  /* there are no WAL records following the checkpoint */
  ereport(LOG,
  (errmsg("redo is not required")));
+ reachedSafeStopPoint = true;
  }
  }
 
***************
*** 5184,5190 ****
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (XLByteLT(EndOfLog, ControlFile->minRecoveryPoint))
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
--- 5235,5241 ----
  * Complain if we did not roll forward far enough to render the backup
  * dump consistent.
  */
! if (InRecovery && !reachedSafeStopPoint)
  {
  if (reachedStopPoint) /* stopped because of stop request */
  ereport(FATAL,
***************
*** 5227,5233 ****
--- 5278,5287 ----
  * we will use that below.)
  */
  if (InArchiveRecovery)
+ {
+ performedArchiveRecovery = true;
  exitArchiveRecovery(curFileTLI, endLogId, endLogSeg);
+ }
 
  /*
  * Prepare to write WAL starting at EndOfLog position, and init xlog
***************
*** 5286,5291 ****
--- 5340,5347 ----
  /* Pre-scan prepared transactions to find out the range of XIDs present */
  oldestActiveXID = PrescanPreparedTransactions();
 
+ RecoveryCompletionPtr = EndOfLog;
+
  if (InRecovery)
  {
  int rmid;
***************
*** 5306,5343 ****
  XLogCheckInvalidPages();
 
  /*
! * Reset pgstat data, because it may be invalid after recovery.
  */
! pgstat_reset_all();
!
! /*
! * Perform a checkpoint to update all our recovery activity to disk.
! *
! * Note that we write a shutdown checkpoint rather than an on-line
! * one. This is not particularly critical, but since we may be
! * assigning a new TLI, using a shutdown checkpoint allows us to have
! * the rule that TLI only changes in shutdown checkpoints, which
! * allows some extra error checking in xlog_redo.
! */
! CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
  }
 
- /*
- * Preallocate additional log files, if wanted.
- */
- PreallocXlogFiles(EndOfLog);
-
- /*
- * Okay, we're officially UP.
- */
- InRecovery = false;
-
- ControlFile->state = DB_IN_PRODUCTION;
- ControlFile->time = (pg_time_t) time(NULL);
- UpdateControlFile();
-
  /* start the archive_timeout timer running */
! XLogCtl->Write.lastSegSwitchTime = ControlFile->time;
 
  /* initialize shared-memory copy of latest checkpoint XID/epoch */
  XLogCtl->ckptXidEpoch = ControlFile->checkPointCopy.nextXidEpoch;
--- 5362,5375 ----
  XLogCheckInvalidPages();
 
  /*
! * Finally exit recovery and mark that in WAL. Pre-8.4 we wrote
! * a shutdown checkpoint here, but we ask bgwriter to do that now.
  */
! RecoveryCompletionPtr = exitRecovery();
  }
 
  /* start the archive_timeout timer running */
! XLogCtl->Write.lastSegSwitchTime = (pg_time_t) time(NULL);
 
  /* initialize shared-memory copy of latest checkpoint XID/epoch */
  XLogCtl->ckptXidEpoch = ControlFile->checkPointCopy.nextXidEpoch;
***************
*** 5372,5377 ****
--- 5404,5446 ----
  readRecordBuf = NULL;
  readRecordBufSize = 0;
  }
+
+ /*
+ * Okay, we're officially UP.
+ */
+ XLogCtl->IsRecoveryProcessingMode = false;
+
+ /*
+ * If we had to perform archive recovery we don't mark the control file,
+ * yet, since we haven't definitely got a safe point to recover from that
+ * doesn't rely on archived WAL files. So we switch quickly into normal
+ * processing and rely on the bgwriter's checkpoint (NOT a restartpoint)
+ * to define a safe recovery point and put us into full production state.
+ * We specifically do not want to wait for checkpoint completion here,
+ * so we can reduce startup time in a standby mode replication failover.
+ * The checkpoint creation will also flush WAL, so we wait for that
+ * otherwise we may need to prepare WAL files ourselves.
+ *
+ * If we are doing crash recovery, we know we have WAL files accessible
+ * so we just get started again as quickly as possible.
+ */
+ if (performedArchiveRecovery)
+ RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_IMMEDIATE);
+ else
+ {
+ XLogFlush(RecoveryCompletionPtr);
+ LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
+ ControlFile->state = DB_IN_PRODUCTION;
+ ControlFile->time = (pg_time_t) time(NULL);
+ UpdateControlFile();
+ LWLockRelease(ControlFileLock);
+ }
+ }
+
+ bool
+ IsRecoveryProcessingMode(void)
+ {
+ return XLogCtl->IsRecoveryProcessingMode;
  }
 
  /*
***************
*** 5629,5648 ****
  static void
  LogCheckpointStart(int flags)
  {
! elog(LOG, "checkpoint starting:%s%s%s%s%s%s",
! (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
! (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
! (flags & CHECKPOINT_FORCE) ? " force" : "",
! (flags & CHECKPOINT_WAIT) ? " wait" : "",
! (flags & CHECKPOINT_CAUSE_XLOG) ? " xlog" : "",
! (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "");
  }
 
  /*
   * Log end of a checkpoint.
   */
  static void
! LogCheckpointEnd(void)
  {
  long write_secs,
  sync_secs,
--- 5698,5721 ----
  static void
  LogCheckpointStart(int flags)
  {
! if (flags & CHECKPOINT_RESTARTPOINT)
! elog(LOG, "restartpoint starting:%s",
! (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "");
! else
! elog(LOG, "checkpoint starting:%s%s%s%s%s%s",
! (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
! (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
! (flags & CHECKPOINT_FORCE) ? " force" : "",
! (flags & CHECKPOINT_WAIT) ? " wait" : "",
! (flags & CHECKPOINT_CAUSE_XLOG) ? " xlog" : "",
! (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "");
  }
 
  /*
   * Log end of a checkpoint.
   */
  static void
! LogCheckpointEnd(int flags)
  {
  long write_secs,
  sync_secs,
***************
*** 5665,5681 ****
  CheckpointStats.ckpt_sync_end_t,
  &sync_secs, &sync_usecs);
 
! elog(LOG, "checkpoint complete: wrote %d buffers (%.1f%%); "
! "%d transaction log file(s) added, %d removed, %d recycled; "
! "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
! CheckpointStats.ckpt_bufs_written,
! (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
! CheckpointStats.ckpt_segs_added,
! CheckpointStats.ckpt_segs_removed,
! CheckpointStats.ckpt_segs_recycled,
! write_secs, write_usecs / 1000,
! sync_secs, sync_usecs / 1000,
! total_secs, total_usecs / 1000);
  }
 
  /*
--- 5738,5763 ----
  CheckpointStats.ckpt_sync_end_t,
  &sync_secs, &sync_usecs);
 
! if (flags & CHECKPOINT_RESTARTPOINT)
! elog(LOG, "restartpoint complete: wrote %d buffers (%.1f%%); "
! "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
! CheckpointStats.ckpt_bufs_written,
! (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
! write_secs, write_usecs / 1000,
! sync_secs, sync_usecs / 1000,
! total_secs, total_usecs / 1000);
! else
! elog(LOG, "checkpoint complete: wrote %d buffers (%.1f%%); "
! "%d transaction log file(s) added, %d removed, %d recycled; "
! "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s",
! CheckpointStats.ckpt_bufs_written,
! (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
! CheckpointStats.ckpt_segs_added,
! CheckpointStats.ckpt_segs_removed,
! CheckpointStats.ckpt_segs_recycled,
! write_secs, write_usecs / 1000,
! sync_secs, sync_usecs / 1000,
! total_secs, total_usecs / 1000);
  }
 
  /*
***************
*** 5944,5949 ****
--- 6026,6033 ----
  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
  if (shutdown)
  ControlFile->state = DB_SHUTDOWNED;
+ else
+ ControlFile->state = DB_IN_PRODUCTION;
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
  ControlFile->checkPoint = ProcLastRecPtr;
  ControlFile->checkPointCopy = checkPoint;
***************
*** 6002,6008 ****
 
  /* All real work is done, but log before releasing lock. */
  if (log_checkpoints)
! LogCheckpointEnd();
 
  LWLockRelease(CheckpointLock);
  }
--- 6086,6092 ----
 
  /* All real work is done, but log before releasing lock. */
  if (log_checkpoints)
! LogCheckpointEnd(flags);
 
  LWLockRelease(CheckpointLock);
  }
***************
*** 6071,6099 ****
  }
  }
 
! /*
! * OK, force data out to disk
! */
! CheckPointGuts(checkPoint->redo, CHECKPOINT_IMMEDIATE);
 
  /*
! * Update pg_control so that any subsequent crash will restart from this
! * checkpoint. Note: ReadRecPtr gives the XLOG address of the checkpoint
! * record itself.
  */
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
! ControlFile->checkPoint = ReadRecPtr;
! ControlFile->checkPointCopy = *checkPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
 
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
! checkPoint->redo.xlogid, checkPoint->redo.xrecoff)));
! if (recoveryLastXTime)
! ereport((recoveryLogRestartpoints ? LOG : DEBUG2),
! (errmsg("last completed transaction was at log time %s",
! timestamptz_to_str(recoveryLastXTime))));
  }
 
  /*
--- 6155,6213 ----
  }
  }
 
! if (recoveryLastXTime)
! ereport((log_checkpoints ? LOG : DEBUG2),
! (errmsg("last completed transaction was at log time %s",
! timestamptz_to_str(recoveryLastXTime))));
!
! RequestRestartPoint(ReadRecPtr, checkPoint, reachedSafeStopPoint);
! }
!
! /*
!  * As of 8.4, RestartPoints are always created by the bgwriter
!  * once we have reachedSafeStopPoint. We use bgwriter's shared memory
!  * area wherever we call it from, to keep better code structure.
!  */
! void
! CreateRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, int flags)
! {
! if (log_checkpoints)
! {
! /*
! * Prepare to accumulate statistics.
! */
!
! MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
! CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
!
! LogCheckpointStart(CHECKPOINT_RESTARTPOINT | flags);
! }
!
! LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
!
! CheckPointGuts(restartPoint->redo, CHECKPOINT_RESTARTPOINT | flags);
 
  /*
! * Update pg_control, using current time
  */
+ LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
  ControlFile->prevCheckPoint = ControlFile->checkPoint;
! ControlFile->checkPoint = ReadPtr;
! ControlFile->checkPointCopy = *restartPoint;
  ControlFile->time = (pg_time_t) time(NULL);
  UpdateControlFile();
+ LWLockRelease(ControlFileLock);
 
! /* All real work is done, but log before releasing lock. */
! if (log_checkpoints)
! LogCheckpointEnd(CHECKPOINT_RESTARTPOINT);
!
! ereport((log_checkpoints ? LOG : DEBUG2),
  (errmsg("recovery restart point at %X/%X",
! restartPoint->redo.xlogid, restartPoint->redo.xrecoff)));
!
! LWLockRelease(CheckpointLock);
!
  }
 
  /*
***************
*** 6158,6164 ****
  }
 
  /*
!  * XLOG resource manager's routines
   */
  void
  xlog_redo(XLogRecPtr lsn, XLogRecord *record)
--- 6272,6326 ----
  }
 
  /*
!  * exitRecovery()
!  *
!  * Exit recovery state and write a XLOG_RECOVERY_END record. This is the
!  * only record type that can record a change of timelineID. We assume
!  * caller has already set ThisTimeLineID, if appropriate.
!  */
! static XLogRecPtr
! exitRecovery(void)
! {
! XLogRecPtr RecPtr;
! XLogRecData rdata;
!
! rdata.buffer = InvalidBuffer;
! rdata.data = (char *) (&ThisTimeLineID);
! rdata.len = sizeof(TimeLineID);
! rdata.next = NULL;
!
! /*
! * If a restartpoint is in progress, send a second signal to
! * get it go faster so we don't need to wait as long. Then wait for
! * the lock, so we know the restartpoint has completed. It's
! * difficult to interrupt it without risking robustness, plus any work
! * it does now will be work  we can avoid later later when we checkpoint.
! */
! if (!LWLockConditionalAcquire(CheckpointLock, LW_EXCLUSIVE))
! {
! RequestRestartPointCompletion();
! elog(LOG, "startup process waiting for restartpoint to complete");
! LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
! LWLockRelease(CheckpointLock);
! }
!
! /*
! * This is the only type of WAL message that can be inserted during
! * recovery. This ensures that we don't allow others to get access
! * until after we have changed state.
! */
! RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RECOVERY_END, &rdata);
!
! InRecovery = false;
!
! return RecPtr;
! }
!
! /*
!  * XLOG resource manager's routines.
!  *
!  * Definitions of message info are in include/catalog/pg_control.h,
!  * though not all messages relate to control file processing.
   */
  void
  xlog_redo(XLogRecPtr lsn, XLogRecord *record)
***************
*** 6193,6213 ****
  ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
 
  /*
! * TLI may change in a shutdown checkpoint, but it shouldn't decrease
  */
! if (checkPoint.ThisTimeLineID != ThisTimeLineID)
  {
! if (checkPoint.ThisTimeLineID < ThisTimeLineID ||
  !list_member_int(expectedTLIs,
! (int) checkPoint.ThisTimeLineID))
  ereport(PANIC,
! (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
! checkPoint.ThisTimeLineID, ThisTimeLineID)));
  /* Following WAL records should be run with new TLI */
! ThisTimeLineID = checkPoint.ThisTimeLineID;
  }
-
- RecoveryRestartPoint(&checkPoint);
  }
  else if (info == XLOG_CHECKPOINT_ONLINE)
  {
--- 6355,6392 ----
  ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
 
  /*
! * TLI no longer changes at shutdown checkpoint, since as of 8.4,
! * shutdown checkpoints only occur at shutdown. Much less confusing.
  */
!
! RecoveryRestartPoint(&checkPoint);
! }
! else if (info == XLOG_RECOVERY_END)
! {
! TimeLineID tli;
!
! memcpy(&tli, XLogRecGetData(record), sizeof(TimeLineID));
!
! /*
! * TLI may change when recovery ends, but it shouldn't decrease.
! *
! * This is the only WAL record that can tell us to change timelineID
! * while we process WAL records.
! *
! * We can *choose* to stop recovery at any point, generating a
! * new timelineID which is recorded using this record type.
! */
! if (tli != ThisTimeLineID)
  {
! if (tli < ThisTimeLineID ||
  !list_member_int(expectedTLIs,
! (int) tli))
  ereport(PANIC,
! (errmsg("unexpected timeline ID %u (after %u) at recovery end record",
! tli, ThisTimeLineID)));
  /* Following WAL records should be run with new TLI */
! ThisTimeLineID = tli;
  }
  }
  else if (info == XLOG_CHECKPOINT_ONLINE)
  {
***************
*** 6288,6294 ****
  }
 
  #ifdef WAL_DEBUG
-
  static void
  xlog_outrec(StringInfo buf, XLogRecord *record)
  {
--- 6467,6472 ----
***************
*** 6308,6314 ****
  }
  #endif   /* WAL_DEBUG */
 
-
  /*
   * Return the (possible) sync flag used for opening a file, depending on the
   * value of the GUC wal_sync_method.
--- 6486,6491 ----
Index: src/backend/postmaster/bgwriter.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/bgwriter.c,v
retrieving revision 1.51
diff -c -r1.51 bgwriter.c
*** src/backend/postmaster/bgwriter.c 11 Aug 2008 11:05:11 -0000 1.51
--- src/backend/postmaster/bgwriter.c 23 Sep 2008 14:56:29 -0000
***************
*** 49,54 ****
--- 49,55 ----
  #include <unistd.h>
 
  #include "access/xlog_internal.h"
+ #include "catalog/pg_control.h"
  #include "libpq/pqsignal.h"
  #include "miscadmin.h"
  #include "pgstat.h"
***************
*** 130,135 ****
--- 131,143 ----
 
  int ckpt_flags; /* checkpoint flags, as defined in xlog.h */
 
+ /*
+ * When the Startup process wants bgwriter to perform a restartpoint, it
+ * sets these fields so that we can update the control file afterwards.
+ */
+ XLogRecPtr ReadPtr; /* ReadRecPtr for RestartPoint request */
+ CheckPoint *restartPoint; /* restartPoint data for ControlFile */
+
  uint32 num_backend_writes; /* counts non-bgwriter buffer writes */
 
  int num_requests; /* current # of requests */
***************
*** 166,172 ****
 
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr;
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
--- 174,180 ----
 
  /* these values are valid when ckpt_active is true: */
  static pg_time_t ckpt_start_time;
! static XLogRecPtr ckpt_start_recptr; /* not used if IsRecoveryProcessingMode */
  static double ckpt_cached_elapsed;
 
  static pg_time_t last_checkpoint_time;
***************
*** 198,203 ****
--- 206,212 ----
  {
  sigjmp_buf local_sigjmp_buf;
  MemoryContext bgwriter_context;
+ bool BgWriterRecoveryMode;
 
  BgWriterShmem->bgwriter_pid = MyProcPid;
  am_bg_writer = true;
***************
*** 356,371 ****
  */
  PG_SETMASK(&UnBlockSig);
 
  /*
  * Loop forever
  */
  for (;;)
  {
- bool do_checkpoint = false;
- int flags = 0;
- pg_time_t now;
- int elapsed_secs;
-
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
--- 365,381 ----
  */
  PG_SETMASK(&UnBlockSig);
 
+ BgWriterRecoveryMode = IsRecoveryProcessingMode();
+
+ if (BgWriterRecoveryMode)
+ elog(DEBUG1, "bgwriter starting during recovery, pid = %u",
+ BgWriterShmem->bgwriter_pid);
+
  /*
  * Loop forever
  */
  for (;;)
  {
  /*
  * Emergency bailout if postmaster has died.  This is to avoid the
  * necessity for manual cleanup of all postmaster children.
***************
*** 383,501 ****
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
- if (checkpoint_requested)
- {
- checkpoint_requested = false;
- do_checkpoint = true;
- BgWriterStats.m_requested_checkpoints++;
- }
- if (shutdown_requested)
- {
- /*
- * From here on, elog(ERROR) should end with exit(1), not send
- * control back to the sigsetjmp block above
- */
- ExitOnAnyError = true;
- /* Close down the database */
- ShutdownXLOG(0, 0);
- DumpFreeSpaceMap(0, 0);
- /* Normal exit from the bgwriter is here */
- proc_exit(0); /* done */
- }
 
! /*
! * Force a checkpoint if too much time has elapsed since the last one.
! * Note that we count a timed checkpoint in stats only when this
! * occurs without an external request, but we set the CAUSE_TIME flag
! * bit even if there is also an external request.
! */
! now = (pg_time_t) time(NULL);
! elapsed_secs = now - last_checkpoint_time;
! if (elapsed_secs >= CheckPointTimeout)
  {
! if (!do_checkpoint)
! BgWriterStats.m_timed_checkpoints++;
! do_checkpoint = true;
! flags |= CHECKPOINT_CAUSE_TIME;
  }
!
! /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
! */
! if (do_checkpoint)
  {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
 
  /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
  */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
 
  /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
  */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
 
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
!
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
 
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
!
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
!
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
  }
- else
- BgBufferSync();
-
- /* Check for archive_timeout and switch xlog files if necessary. */
- CheckArchiveTimeout();
-
- /* Nap for the configured time. */
- BgWriterNap();
  }
  }
 
--- 393,599 ----
  got_SIGHUP = false;
  ProcessConfigFile(PGC_SIGHUP);
  }
 
! if (BgWriterRecoveryMode)
  {
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
!
! if (!IsRecoveryProcessingMode())
! {
! elog(DEBUG2, "bgwriter changing from recovery to normal mode");
!
! InitXLOGAccess();
! BgWriterRecoveryMode = false;
!
! /*
! * Start time-driven events from now
! */
! last_checkpoint_time = last_xlog_switch_time = (pg_time_t) time(NULL);
!
! /*
! * Notice that we do *not* act on a checkpoint_requested
! * state at this point. We have changed mode, so we wish to
! * perform a checkpoint not a restartpoint.
! */
! continue;
! }
!
! if (checkpoint_requested)
! {
! XLogRecPtr ReadPtr;
! CheckPoint restartPoint;
!
! checkpoint_requested = false;
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_time = (pg_time_t) time(NULL);
! ckpt_cached_elapsed = 0;
!
! /*
! * Get the requested values from shared memory that the
! * Startup process has put there for us.
! */
! SpinLockAcquire(&BgWriterShmem->ckpt_lck);
! ReadPtr = BgWriterShmem->ReadPtr;
! memcpy(&restartPoint, &BgWriterShmem->restartPoint, sizeof(CheckPoint));
! SpinLockRelease(&BgWriterShmem->ckpt_lck);
!
! /* Use smoothed writes, until interrupted if ever */
! CreateRestartPoint(ReadPtr, &restartPoint, 0);
!
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
!
! ckpt_active = false;
! checkpoint_requested = false;
! }
! else
! {
! /* Clean buffers dirtied by recovery */
! BgBufferSync();
!
! /* Nap for the configured time. */
! BgWriterNap();
! }
  }
! else /* Normal processing */
  {
! bool do_checkpoint = false;
! int flags = 0;
! pg_time_t now;
! int elapsed_secs;
!
! Assert(!IsRecoveryProcessingMode());
!
! if (checkpoint_requested)
! {
! checkpoint_requested = false;
! do_checkpoint = true;
! BgWriterStats.m_requested_checkpoints++;
! }
! if (shutdown_requested)
! {
! /*
! * From here on, elog(ERROR) should end with exit(1), not send
! * control back to the sigsetjmp block above
! */
! ExitOnAnyError = true;
! /* Close down the database */
! ShutdownXLOG(0, 0);
! DumpFreeSpaceMap(0, 0);
! /* Normal exit from the bgwriter is here */
! proc_exit(0); /* done */
! }
 
  /*
! * Force a checkpoint if too much time has elapsed since the last one.
! * Note that we count a timed checkpoint in stats only when this
! * occurs without an external request, but we set the CAUSE_TIME flag
! * bit even if there is also an external request.
  */
! now = (pg_time_t) time(NULL);
! elapsed_secs = now - last_checkpoint_time;
! if (elapsed_secs >= CheckPointTimeout)
! {
! if (!do_checkpoint)
! BgWriterStats.m_timed_checkpoints++;
! do_checkpoint = true;
! flags |= CHECKPOINT_CAUSE_TIME;
! }
 
  /*
! * Do a checkpoint if requested, otherwise do one cycle of
! * dirty-buffer writing.
  */
! if (do_checkpoint)
! {
! /* use volatile pointer to prevent code rearrangement */
! volatile BgWriterShmemStruct *bgs = BgWriterShmem;
!
! /*
! * Atomically fetch the request flags to figure out what kind of a
! * checkpoint we should perform, and increase the started-counter
! * to acknowledge that we've started a new checkpoint.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! flags |= bgs->ckpt_flags;
! bgs->ckpt_flags = 0;
! bgs->ckpt_started++;
! SpinLockRelease(&bgs->ckpt_lck);
!
! /*
! * We will warn if (a) too soon since last checkpoint (whatever
! * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
! * since the last checkpoint start.  Note in particular that this
! * implementation will not generate warnings caused by
! * CheckPointTimeout < CheckPointWarning.
! */
! if ((flags & CHECKPOINT_CAUSE_XLOG) &&
! elapsed_secs < CheckPointWarning)
! ereport(LOG,
! (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
! elapsed_secs),
! errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
!
! /*
! * Initialize bgwriter-private variables used during checkpoint.
! */
! ckpt_active = true;
! ckpt_start_recptr = GetInsertRecPtr();
! ckpt_start_time = now;
! ckpt_cached_elapsed = 0;
!
! /*
! * Do the checkpoint.
! */
! CreateCheckPoint(flags);
!
! /*
! * After any checkpoint, close all smgr files. This is so we
! * won't hang onto smgr references to deleted files indefinitely.
! */
! smgrcloseall();
!
! /*
! * Indicate checkpoint completion to any waiting backends.
! */
! SpinLockAcquire(&bgs->ckpt_lck);
! bgs->ckpt_done = bgs->ckpt_started;
! SpinLockRelease(&bgs->ckpt_lck);
!
! ckpt_active = false;
!
! /*
! * Note we record the checkpoint start time not end time as
! * last_checkpoint_time.  This is so that time-driven checkpoints
! * happen at a predictable spacing.
! */
! last_checkpoint_time = now;
! }
! else
! BgBufferSync();
 
! /* Check for archive_timeout and switch xlog files if necessary. */
! CheckArchiveTimeout();
 
! /* Nap for the configured time. */
! BgWriterNap();
  }
  }
  }
 
***************
*** 588,594 ****
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
--- 686,693 ----
  (ckpt_active ? ImmediateCheckpointRequested() : checkpoint_requested))
  break;
  pg_usleep(1000000L);
! if (!IsRecoveryProcessingMode())
! AbsorbFsyncRequests();
  udelay -= 1000000L;
  }
 
***************
*** 642,647 ****
--- 741,759 ----
  if (!am_bg_writer)
  return;
 
+ /* Perform minimal duties during recovery and skip wait if requested */
+ if (IsRecoveryProcessingMode())
+ {
+ BgBufferSync();
+
+ if (!shutdown_requested &&
+ !checkpoint_requested &&
+ IsCheckpointOnSchedule(progress))
+ BgWriterNap();
+
+ return;
+ }
+
  /*
  * Perform the usual bgwriter duties and take a nap, unless we're behind
  * schedule, in which case we just try to catch up as quickly as possible.
***************
*** 716,731 ****
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
  {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
  }
 
  /*
--- 828,846 ----
  * However, it's good enough for our purposes, we're only calculating an
  * estimate anyway.
  */
! if (!IsRecoveryProcessingMode())
  {
! recptr = GetInsertRecPtr();
! elapsed_xlogs =
! (((double) (int32) (recptr.xlogid - ckpt_start_recptr.xlogid)) * XLogSegsPerFile +
! ((double) recptr.xrecoff - (double) ckpt_start_recptr.xrecoff) / XLogSegSize) /
! CheckPointSegments;
!
! if (progress < elapsed_xlogs)
! {
! ckpt_cached_elapsed = elapsed_xlogs;
! return false;
! }
  }
 
  /*
***************
*** 967,972 ****
--- 1082,1130 ----
  }
 
  /*
+  * Always runs in Startup process (see xlog.c)
+  */
+ void
+ RequestRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, bool sendToBGWriter)
+ {
+ /*
+ * Should we just do it ourselves?
+ */
+ if (!IsPostmasterEnvironment || !sendToBGWriter)
+ {
+ CreateRestartPoint(ReadPtr, restartPoint, CHECKPOINT_IMMEDIATE);
+ return;
+ }
+
+ /*
+ * Push requested values into shared memory, then signal to request restartpoint.
+ */
+ if (BgWriterShmem->bgwriter_pid == 0)
+ elog(LOG, "could not request restartpoint because bgwriter not running");
+
+ SpinLockAcquire(&BgWriterShmem->ckpt_lck);
+ BgWriterShmem->ReadPtr = ReadPtr;
+ memcpy(&BgWriterShmem->restartPoint, restartPoint, sizeof(CheckPoint));
+ SpinLockRelease(&BgWriterShmem->ckpt_lck);
+
+ if (kill(BgWriterShmem->bgwriter_pid, SIGINT) != 0)
+ elog(LOG, "could not signal for restartpoint: %m");
+ }
+
+ /*
+  * Sends another checkpoint request signal to bgwriter, which causes it
+  * to avoid smoothed writes and continue processing as if it had been
+  * called with CHECKPOINT_IMMEDIATE. This is used at the end of recovery.
+  */
+ void
+ RequestRestartPointCompletion(void)
+ {
+ if (BgWriterShmem->bgwriter_pid != 0 &&
+ kill(BgWriterShmem->bgwriter_pid, SIGINT) != 0)
+ elog(LOG, "could not signal for restartpoint immediate: %m");
+ }
+
+ /*
   * ForwardFsyncRequest
   * Forward a file-fsync request from a backend to the bgwriter
   *
Index: src/backend/postmaster/postmaster.c
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/postmaster/postmaster.c,v
retrieving revision 1.561
diff -c -r1.561 postmaster.c
*** src/backend/postmaster/postmaster.c 26 Jun 2008 02:47:19 -0000 1.561
--- src/backend/postmaster/postmaster.c 22 Sep 2008 20:25:36 -0000
***************
*** 254,259 ****
--- 254,264 ----
  {
  PM_INIT, /* postmaster starting */
  PM_STARTUP, /* waiting for startup subprocess */
+ PM_RECOVERY, /* consistent recovery mode; state only
+ * entered for archive and streaming recovery,
+ * and only after the point where the
+ * all data is in consistent state.
+ */
  PM_RUN, /* normal "database is alive" state */
  PM_WAIT_BACKUP, /* waiting for online backup mode to end */
  PM_WAIT_BACKENDS, /* waiting for live backends to exit */
***************
*** 1294,1300 ****
  * state that prevents it, start one.  It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0 && pmState == PM_RUN)
  BgWriterPID = StartBackgroundWriter();
 
  /*
--- 1299,1305 ----
  * state that prevents it, start one.  It doesn't matter if this
  * fails, we'll just try again later.
  */
! if (BgWriterPID == 0 && (pmState == PM_RUN || pmState == PM_RECOVERY))
  BgWriterPID = StartBackgroundWriter();
 
  /*
***************
*** 2104,2110 ****
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
--- 2109,2115 ----
  if (pid == StartupPID)
  {
  StartupPID = 0;
! Assert(pmState == PM_STARTUP || pmState == PM_RECOVERY);
 
  /* FATAL exit of startup is treated as catastrophic */
  if (!EXIT_STATUS_0(exitstatus))
***************
*** 2145,2155 ****
  load_role();
 
  /*
! * Crank up the background writer. It doesn't matter if this
! * fails, we'll just try again later.
  */
! Assert(BgWriterPID == 0);
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
--- 2150,2160 ----
  load_role();
 
  /*
! * Check whether we need to start background writer, if not
! * already running.
  */
! if (BgWriterPID == 0)
! BgWriterPID = StartBackgroundWriter();
 
  /*
  * Likewise, start other special children as needed.  In a restart
***************
*** 3821,3826 ****
--- 3826,3876 ----
 
  PG_SETMASK(&BlockSig);
 
+ if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_START))
+ {
+ Assert(pmState == PM_STARTUP);
+
+ /*
+ * Go to shutdown mode if a shutdown request was pending.
+ */
+ if (Shutdown > NoShutdown)
+ {
+ pmState = PM_WAIT_BACKENDS;
+ /* PostmasterStateMachine logic does the rest */
+ }
+ else
+ {
+ /*
+ * Startup process has entered recovery
+ */
+ pmState = PM_RECOVERY;
+
+ /*
+ * Load the flat authorization file into postmaster's cache. The
+ * startup process won't have recomputed this from the database yet,
+ * so we it may change following recovery.
+ */
+ load_role();
+
+ /*
+ * Crank up the background writer. It doesn't matter if this
+ * fails, we'll just try again later.
+ */
+ Assert(BgWriterPID == 0);
+ BgWriterPID = StartBackgroundWriter();
+
+ /*
+ * Likewise, start other special children as needed.
+ */
+ Assert(PgStatPID == 0);
+ PgStatPID = pgstat_start();
+
+ /* XXX at this point we could accept read-only connections */
+ ereport(DEBUG1,
+ (errmsg("database system is in consistent recovery mode")));
+ }
+ }
+
  if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
  {
  /*
Index: src/backend/storage/buffer/README
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/backend/storage/buffer/README,v
retrieving revision 1.14
diff -c -r1.14 README
*** src/backend/storage/buffer/README 21 Mar 2008 13:23:28 -0000 1.14
--- src/backend/storage/buffer/README 22 Sep 2008 21:54:20 -0000
***************
*** 264,266 ****
--- 264,271 ----
  This ensures that the page image transferred to disk is reasonably consistent.
  We might miss a hint-bit update or two but that isn't a problem, for the same
  reasons mentioned under buffer access rules.
+
+ As of 8.4, background writer starts during recovery mode when there is
+ some form of potentially extended recovery to perform. It performs an
+ identical service to normal processing, except that checkpoints it
+ writes are technically restartpoints nor needs to flush WAL for dirty buffers.
Index: src/include/access/xlog.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog.h,v
retrieving revision 1.88
diff -c -r1.88 xlog.h
*** src/include/access/xlog.h 12 May 2008 08:35:05 -0000 1.88
--- src/include/access/xlog.h 22 Sep 2008 19:04:56 -0000
***************
*** 133,139 ****
  } XLogRecData;
 
  extern TimeLineID ThisTimeLineID; /* current TLI */
! extern bool InRecovery;
  extern XLogRecPtr XactLastRecEnd;
 
  /* these variables are GUC parameters related to XLOG */
--- 133,148 ----
  } XLogRecData;
 
  extern TimeLineID ThisTimeLineID; /* current TLI */
!
! /*
!  * Prior to 8.4, all activity during recovery were carried out by Startup
!  * process. This local variable continues to be used in many parts of the
!  * code to indicate actions taken by RecoveryManagers. Other processes who
!  * potentially perform work during recovery should check
!  * IsRecoveryProcessingMode(), see XLogCtl notes in xlog.c
!  */
! extern bool InRecovery;
!
  extern XLogRecPtr XactLastRecEnd;
 
  /* these variables are GUC parameters related to XLOG */
***************
*** 166,171 ****
--- 175,181 ----
  /* These indicate the cause of a checkpoint request */
  #define CHECKPOINT_CAUSE_XLOG 0x0010 /* XLOG consumption */
  #define CHECKPOINT_CAUSE_TIME 0x0020 /* Elapsed time */
+ #define CHECKPOINT_RESTARTPOINT 0x0040 /* Restartpoint during recovery */
 
  /* Checkpoint statistics */
  typedef struct CheckpointStatsData
***************
*** 197,202 ****
--- 207,214 ----
  extern void xlog_redo(XLogRecPtr lsn, XLogRecord *record);
  extern void xlog_desc(StringInfo buf, uint8 xl_info, char *rec);
 
+ extern bool IsRecoveryProcessingMode(void);
+
  extern void UpdateControlFile(void);
  extern Size XLOGShmemSize(void);
  extern void XLOGShmemInit(void);
Index: src/include/access/xlog_internal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/access/xlog_internal.h,v
retrieving revision 1.24
diff -c -r1.24 xlog_internal.h
*** src/include/access/xlog_internal.h 11 Aug 2008 11:05:11 -0000 1.24
--- src/include/access/xlog_internal.h 23 Sep 2008 10:59:40 -0000
***************
*** 17,22 ****
--- 17,23 ----
  #define XLOG_INTERNAL_H
 
  #include "access/xlog.h"
+ #include "catalog/pg_control.h"
  #include "fmgr.h"
  #include "pgtime.h"
  #include "storage/block.h"
***************
*** 245,250 ****
--- 246,254 ----
  extern pg_time_t GetLastSegSwitchTime(void);
  extern XLogRecPtr RequestXLogSwitch(void);
 
+ extern void CreateRestartPoint(const XLogRecPtr ReadPtr,
+ const CheckPoint *restartPoint, int flags);
+
  /*
   * These aren't in xlog.h because I'd rather not include fmgr.h there.
   */
Index: src/include/catalog/pg_control.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/catalog/pg_control.h,v
retrieving revision 1.41
diff -c -r1.41 pg_control.h
*** src/include/catalog/pg_control.h 21 Apr 2008 00:26:47 -0000 1.41
--- src/include/catalog/pg_control.h 22 Sep 2008 19:21:37 -0000
***************
*** 46,52 ****
  #define XLOG_NOOP 0x20
  #define XLOG_NEXTOID 0x30
  #define XLOG_SWITCH 0x40
!
 
  /* System status indicator */
  typedef enum DBState
--- 46,52 ----
  #define XLOG_NOOP 0x20
  #define XLOG_NEXTOID 0x30
  #define XLOG_SWITCH 0x40
! #define XLOG_RECOVERY_END 0x50
 
  /* System status indicator */
  typedef enum DBState
Index: src/include/postmaster/bgwriter.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/postmaster/bgwriter.h,v
retrieving revision 1.12
diff -c -r1.12 bgwriter.h
*** src/include/postmaster/bgwriter.h 11 Aug 2008 11:05:11 -0000 1.12
--- src/include/postmaster/bgwriter.h 23 Sep 2008 11:12:18 -0000
***************
*** 12,17 ****
--- 12,18 ----
  #ifndef _BGWRITER_H
  #define _BGWRITER_H
 
+ #include "catalog/pg_control.h"
  #include "storage/block.h"
  #include "storage/relfilenode.h"
 
***************
*** 25,30 ****
--- 26,33 ----
  extern void BackgroundWriterMain(void);
 
  extern void RequestCheckpoint(int flags);
+ extern void RequestRestartPoint(const XLogRecPtr ReadPtr, const CheckPoint *restartPoint, bool sendToBGWriter);
+ extern void RequestRestartPointCompletion(void);
  extern void CheckpointWriteDelay(int flags, double progress);
 
  extern bool ForwardFsyncRequest(RelFileNode rnode, ForkNumber forknum,
Index: src/include/storage/pmsignal.h
===================================================================
RCS file: /home/sriggs/pg/REPOSITORY/pgsql/src/include/storage/pmsignal.h,v
retrieving revision 1.20
diff -c -r1.20 pmsignal.h
*** src/include/storage/pmsignal.h 19 Jun 2008 21:32:56 -0000 1.20
--- src/include/storage/pmsignal.h 22 Sep 2008 15:54:57 -0000
***************
*** 22,27 ****
--- 22,28 ----
   */
  typedef enum
  {
+ PMSIGNAL_RECOVERY_START, /* move to PM_RECOVERY state */
  PMSIGNAL_PASSWORD_CHANGE, /* pg_auth file has changed */
  PMSIGNAL_WAKEN_ARCHIVER, /* send a NOTIFY signal to xlog archiver */
  PMSIGNAL_ROTATE_LOGFILE, /* send SIGUSR1 to syslogger to rotate logfile */

---

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches

[Tom Lane-2]

Simon Riggs <simon@...> writes:
> Version 7

After reading this for awhile, I realized that there is a rather
fundamental problem with it: it switches into "consistent recovery"
mode as soon as it's read WAL beyond ControlFile->minRecoveryPoint.
In a crash recovery situation that typically is before the last
checkpoint (if indeed it's not still zero), and what that means is
that this patch will activate the bgwriter and start letting in
backends instantaneously after a crash, long before we can have any
certainty that the DB state really is consistent.

In a normal crash recovery situation this would be easily fixed by
simply not letting it go to "consistent recovery" state at all, but
what about recovery from a restartpoint?  We don't want a slave that's
crashed once to never let backends in again.  But I don't see how to
determine that we're far enough past the restartpoint to be consistent
again.  In crash recovery we assume (without proof ;-)) that we're
consistent once we reach the end of valid-looking WAL, but that rule
doesn't help for a slave that's following a continuing WAL sequence.

Perhaps something could be done based on noting when we have to pull in
a WAL segment from the recovery_command, but it sounds like a pretty
fragile assumption.

Anyway, that's sufficiently bad that I'm bouncing the patch for
reconsideration.  Some other issues I noted before giving up:

* I'm a bit uncomfortable with the fact that the
IsRecoveryProcessingMode flag is read and written with no lock.
There's no atomicity or concurrent-write problem, sure, but on
a multiprocessor with weak memory ordering guarantees (eg PPC)
readers could get a fairly stale value of the flag.  The false
to true transition happens before anyone except the startup process is
running, so that's no problem; the net effect is then that backends
might think that recovery mode was still active for awhile after it
wasn't.  This seems a bit scary, eg in the patch as it stands that'll
disable XLogFlush calls that should have happened.  You could fix that
by grabbing/releasing some spinlock (any old one) around the accesses,
but are any of the call sites performance-critical?  The one in
XLogInsert looks like it is, if nothing else.

* I kinda think you broke XLogFlush anyway.  It's certainly clear
that the WARNING case at the bottom is unreachable with the patch,
and I think that means that you've messed up an important error
robustness behavior.  Is it still possible to get out of recovery mode
if there are any bad LSNs in the shared buffer pool?

* The use of InRecovery in CreateCheckPoint seems pretty bogus, since
that function can be called from the bgwriter in which the flag will
never be true.  Either this needs to be IsRecoveryProcessingMode(),
or it's useless because we'll never create ordinary checkpoints until
after subtrans.c is up anyway.

* The patch moves the clearing of InRecovery from after to before
StartupCLOG, RecoverPreparedTransactions, etc.  Is that really a
good idea?  It makes it hard for those modules to know if they are
coming up after a normal or recovery startup.  I think they may not
care at the moment, but I'd leave that alone without a darn good
reason to change it.

* The comment about CheckpointLock being only pro forma is now wrong,
if you are going to use locking it to implement a delay in exitRecovery.
But I don't understand why the delay there.  If it's needed, seems like
the path where a restartpoint *isn't* in progress is wrong --- don't you
actually need to start one and wait for it?  Also I note that if the
LWLockConditionalAcquire succeeds, you neglect to release the lock,
which surely can't be right.

* The tail end of StartupXLOG() looks pretty unsafe to me.  Surely
we mustn't clear IsRecoveryProcessingMode until after we have
established the safe-recovery checkpoint.  The comments there seem to
be only vaguely related to the current state of the patch, too.

* Logging of recoveryLastXTime seems pretty bogus now.  It's supposed to
be associated with a restartpoint completion report, but now it's just
out in the ether somewhere and doesn't represent a guarantee that we're
synchronized that far.

* backup.sgml needs to be updated to say that log_restartpoints is
obsolete.

* There are a bunch of disturbing assumptions in the SLRU-related
modules about their StartUp calls being executed without any concurrent
access.  This isn't really a problem that needs to be dealt with in this
patch, I think, but that will all have to be cleaned up before we dare
allow any backends to run concurrently with recovery.  btree's
suppression of relcache invals for metapage updates will be a problem
too.

                        regards, tom lane

--
Sent via pgsql-patches mailing list (pgsql-patches@...)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches