some comments

kelvich · kelvich · commit abea3800a892 · 2017-01-10T14:19:09.000+03:00
diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
@@ -45,8 +45,8 @@
  *		  fsynced
  *		* If COMMIT happens after checkpoint then backend reads state data from
  *		  files
- *		* In case of crash replay will move data from xlog to files, if that
- *		  hasn't happened before. XXX TODO - move to shmem in replay also
+ *		* Simplified version of the same scenario happens during recovery and
+ *		  replication. See comments to KnownPreparedXact structure.
  *
  *-------------------------------------------------------------------------
  */
@@ -181,6 +181,35 @@ static GlobalTransaction MyLockedGxact = NULL;
 
 static bool twophaseExitRegistered = false;
 
+/*
+ * During replay and replication KnownPreparedList holds info about active prepared
+ * transactions that weren't moved to files yet. We will need that info by the end of
+ * recovery (including promote) to restore memory state of that transactions.
+ *
+ * Naive approach here is to move each PREPARE record to disk, fsync it and don't have
+ * that list at all, but that provokes a lot of unnecessary fsyncs on small files
+ * causing replica to be slower than master.
+ *
+ * Replay of twophase records happens by the following rules:
+ *		* On PREPARE redo KnownPreparedAdd() is called to add that transaction to
+ *		  KnownPreparedList and no more actions taken.
+ *		* On checkpoint we iterate through KnownPreparedList, move all prepare
+ *		  records that behind redo_horizon to file and deleting items from list.
+ *		* On COMMIT/ABORT we delete file or entry in KnownPreparedList.
+ *		* At the end of recovery we move all known prepared transactions to disk
+ *		  to allow RecoverPreparedTransactions/StandbyRecoverPreparedTransactions
+ *		  do their work.
+ */
+typedef struct KnownPreparedXact
+{
+	TransactionId	xid;
+	XLogRecPtr		prepare_start_lsn;
+	XLogRecPtr		prepare_end_lsn;
+	dlist_node		list_node;
+} KnownPreparedXact;
+
+static dlist_head KnownPreparedList = DLIST_STATIC_INIT(KnownPreparedList);
+
 static void RecordTransactionCommitPrepared(TransactionId xid,
 								int nchildren,
 								TransactionId *children,
@@ -200,82 +229,6 @@ static void RemoveGXact(GlobalTransaction gxact);
 
 static void XlogReadTwoPhaseData(XLogRecPtr lsn, char **buf, int *len);
 
-
-dlist_head StandbyTwoPhaseStateData = DLIST_STATIC_INIT(StandbyTwoPhaseStateData);
-
-typedef struct StandbyPreparedTransaction
-{
-	TransactionId	xid;
-	XLogRecPtr		prepare_start_lsn;
-	XLogRecPtr		prepare_end_lsn;
-	dlist_node		list_node;
-}	StandbyPreparedTransaction;
-
-void
-StandbyCheckPointTwoPhase(XLogRecPtr redo_horizon)
-{
-	dlist_mutable_iter miter;
-	int			serialized_xacts = 0;
-
-	// Assert(RecoveryInProgress());
-
-	TRACE_POSTGRESQL_TWOPHASE_CHECKPOINT_START();
-
-	dlist_foreach_modify(miter, &StandbyTwoPhaseStateData)
-	{
-		StandbyPreparedTransaction   *xact = dlist_container(StandbyPreparedTransaction,
-															list_node, miter.cur);
-
-		if (redo_horizon == InvalidXLogRecPtr || xact->prepare_end_lsn <= redo_horizon)
-		{
-			char	   *buf;
-			int			len;
-
-			XlogReadTwoPhaseData(xact->prepare_start_lsn, &buf, &len);
-			RecreateTwoPhaseFile(xact->xid, buf, len);
-			pfree(buf);
-			dlist_delete(miter.cur);
-			serialized_xacts++;
-		}
-	}
-
-	TRACE_POSTGRESQL_TWOPHASE_CHECKPOINT_DONE();
-
-	if (log_checkpoints && serialized_xacts > 0)
-		ereport(LOG,
-				(errmsg_plural("%u two-phase state file was written "
-							   "for long-running prepared transactions",
-							   "%u two-phase state files were written "
-							   "for long-running prepared transactions",
-							   serialized_xacts,
-							   serialized_xacts)));
-}
-
-// XXX: rename to remove_standby_state
-void
-StandbyAtCommit(TransactionId xid)
-{
-	dlist_mutable_iter miter;
-
-	Assert(RecoveryInProgress());
-
-	dlist_foreach_modify(miter, &StandbyTwoPhaseStateData)
-	{
-		StandbyPreparedTransaction   *xact = dlist_container(StandbyPreparedTransaction,
-															list_node, miter.cur);
-
-		if (xact->xid == xid)
-		{
-			dlist_delete(miter.cur);
-			return;
-		}
-	}
-
-	RemoveTwoPhaseFile(xid, false);
-}
-
-
-
 /*
  * Initialization of shared memory
  */
@@ -1729,18 +1682,25 @@ CheckPointTwoPhase(XLogRecPtr redo_horizon)
 							   serialized_xacts)));
 }
 
+/*
+ * KnownPreparedAdd.
+ *
+ * Store correspondence of start/end lsn and xid in KnownPreparedList.
+ * This is called during redo of prepare record to have list of prepared
+ * transactions that aren't yet moved to 2PC files by the end of recovery.
+ */
 void
-StandbyAtPrepare(XLogReaderState *record)
+KnownPreparedAdd(XLogReaderState *record)
 {
-	StandbyPreparedTransaction *xact;
+	KnownPreparedXact *xact;
 	TwoPhaseFileHeader *hdr = (TwoPhaseFileHeader *) XLogRecGetData(record);
 
-	xact = (StandbyPreparedTransaction *) palloc(sizeof(StandbyPreparedTransaction));
+	xact = (KnownPreparedXact *) palloc(sizeof(KnownPreparedXact));
 	xact->xid = hdr->xid;
 	xact->prepare_start_lsn = record->ReadRecPtr;
 	xact->prepare_end_lsn = record->EndRecPtr;
 
-	dlist_push_tail(&StandbyTwoPhaseStateData, &xact->list_node);
+	dlist_push_tail(&KnownPreparedList, &xact->list_node);
 }
 
 /*
@@ -1781,7 +1741,7 @@ PrescanPreparedTransactions(TransactionId **xids_p, int *nxids_p)
 	int			nxids = 0;
 	int			allocsize = 0;
 
-	StandbyCheckPointTwoPhase(0);
+	KnownPreparedRecreateFiles(InvalidXLogRecPtr);
 
 	cldir = AllocateDir(TWOPHASE_DIR);
 	while ((clde = ReadDir(cldir, TWOPHASE_DIR)) != NULL)
@@ -2254,3 +2214,88 @@ RecordTransactionAbortPrepared(TransactionId xid,
 	 */
 	SyncRepWaitForLSN(recptr, false);
 }
+
+/*
+ * KnownPreparedRemoveByXid
+ *
+ * Forget about prepared transaction. Called durind commit/abort.
+ */
+void
+KnownPreparedRemoveByXid(TransactionId xid)
+{
+	dlist_mutable_iter miter;
+
+	Assert(RecoveryInProgress());
+
+	dlist_foreach_modify(miter, &KnownPreparedList)
+	{
+		KnownPreparedXact   *xact = dlist_container(KnownPreparedXact,
+														list_node, miter.cur);
+
+		if (xact->xid == xid)
+		{
+			dlist_delete(miter.cur);
+			/*
+			 * Since we found entry in KnownPreparedList we know that file isn't
+			 * on disk yet and we can end up here.
+			 */
+			return;
+		}
+	}
+
+	/*
+	 * Here we know that file should be moved to disk. But aborting recovery because
+	 * of absence of unnecessary file doesn't seems to be a good idea, so call remove
+	 * with giveWarning=false.
+	 */
+	RemoveTwoPhaseFile(xid, false);
+}
+
+/*
+ * KnownPreparedRecreateFiles
+ *
+ * Moves prepare records from WAL to files. Callend during checkpoint replay
+ * or PrescanPreparedTransactions.
+ *
+ * redo_horizon = InvalidXLogRecPtr indicates that all transactions from
+ *		KnownPreparedList should be moved to disk.
+ */
+void
+KnownPreparedRecreateFiles(XLogRecPtr redo_horizon)
+{
+	dlist_mutable_iter miter;
+	int			serialized_xacts = 0;
+
+	Assert(RecoveryInProgress());
+
+	TRACE_POSTGRESQL_TWOPHASE_CHECKPOINT_START();
+
+	dlist_foreach_modify(miter, &KnownPreparedList)
+	{
+		KnownPreparedXact   *xact = dlist_container(KnownPreparedXact,
+														list_node, miter.cur);
+
+		if (xact->prepare_end_lsn <= redo_horizon || redo_horizon == InvalidXLogRecPtr)
+		{
+			char	   *buf;
+			int			len;
+
+			XlogReadTwoPhaseData(xact->prepare_start_lsn, &buf, &len);
+			RecreateTwoPhaseFile(xact->xid, buf, len);
+			pfree(buf);
+			dlist_delete(miter.cur);
+			serialized_xacts++;
+		}
+	}
+
+	TRACE_POSTGRESQL_TWOPHASE_CHECKPOINT_DONE();
+
+	if (log_checkpoints && serialized_xacts > 0)
+		ereport(LOG,
+				(errmsg_plural("%u two-phase state file was written "
+							   "for long-running prepared transactions",
+							   "%u two-phase state files were written "
+							   "for long-running prepared transactions",
+							   serialized_xacts,
+							   serialized_xacts)));
+}
diff --git a/src/backend/access/transam/xact.c b/src/backend/access/transam/xact.c
@@ -5604,7 +5604,9 @@ xact_redo(XLogReaderState *record)
 			Assert(TransactionIdIsValid(parsed.twophase_xid));
 			xact_redo_commit(&parsed, parsed.twophase_xid,
 							 record->EndRecPtr, XLogRecGetOrigin(record));
-			StandbyAtCommit(parsed.twophase_xid);
+
+			/* Delete KnownPrepared entry or 2PC file. */
+			KnownPreparedRemoveByXid(parsed.twophase_xid);
 		}
 	}
 	else if (info == XLOG_XACT_ABORT || info == XLOG_XACT_ABORT_PREPARED)
@@ -5624,13 +5626,20 @@ xact_redo(XLogReaderState *record)
 		{
 			Assert(TransactionIdIsValid(parsed.twophase_xid));
 			xact_redo_abort(&parsed, parsed.twophase_xid);
-			StandbyAtCommit(parsed.twophase_xid);
+
+			/* Delete KnownPrepared entry or 2PC file. */
+			KnownPreparedRemoveByXid(parsed.twophase_xid);
 		}
 	}
 	else if (info == XLOG_XACT_PREPARE)
 	{
-		/* the record contents are exactly the 2PC file */
-		StandbyAtPrepare(record);
+		/*
+		 * If that transaction will not be commited by the end of recovery then we
+		 * will need 2PC file (the record contents is exactly the 2PC file) to be able
+		 * to commit that later.
+		 * For now store xid and pointers to that record in KnownPreparedList.
+		 */
+		KnownPreparedAdd(record);
 	}
 	else if (info == XLOG_XACT_ASSIGNMENT)
 	{
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
@@ -9547,8 +9547,8 @@ xlog_redo(XLogReaderState *record)
 					(errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
 							checkPoint.ThisTimeLineID, ThisTimeLineID)));
 
+		KnownPreparedRecreateFiles(checkPoint.redo);
 		RecoveryRestartPoint(&checkPoint);
-		StandbyCheckPointTwoPhase(checkPoint.redo);
 	}
 	else if (info == XLOG_END_OF_RECOVERY)
 	{
diff --git a/src/include/access/twophase.h b/src/include/access/twophase.h
@@ -57,8 +57,8 @@ extern void CheckPointTwoPhase(XLogRecPtr redo_horizon);
 
 extern void FinishPreparedTransaction(const char *gid, bool isCommit);
 
-extern void StandbyAtCommit(TransactionId xid);
-extern void StandbyAtPrepare(XLogReaderState *record);
-extern void StandbyCheckPointTwoPhase(XLogRecPtr redo_horizon);
+extern void KnownPreparedAdd(XLogReaderState *record);
+extern void KnownPreparedRemoveByXid(TransactionId xid);
+extern void KnownPreparedRecreateFiles(XLogRecPtr redo_horizon);
 
 #endif   /* TWOPHASE_H */

Original file line number	Diff line number	Diff line change
`@@ -5604,7 +5604,9 @@ xact_redo(XLogReaderState *record)`
`5604`	`5604`	`Assert(TransactionIdIsValid(parsed.twophase_xid));`
`5605`	`5605`	`xact_redo_commit(&parsed, parsed.twophase_xid,`
`5606`	`5606`	`record->EndRecPtr, XLogRecGetOrigin(record));`
`5607`		`- StandbyAtCommit(parsed.twophase_xid);`
	`5607`	`+`
	`5608`	`+ /* Delete KnownPrepared entry or 2PC file. */`
	`5609`	`+ KnownPreparedRemoveByXid(parsed.twophase_xid);`
`5608`	`5610`	`}`
`5609`	`5611`	`}`
`5610`	`5612`	`else if (info == XLOG_XACT_ABORT \|\| info == XLOG_XACT_ABORT_PREPARED)`
`@@ -5624,13 +5626,20 @@ xact_redo(XLogReaderState *record)`
`5624`	`5626`	`{`
`5625`	`5627`	`Assert(TransactionIdIsValid(parsed.twophase_xid));`
`5626`	`5628`	`xact_redo_abort(&parsed, parsed.twophase_xid);`
`5627`		`- StandbyAtCommit(parsed.twophase_xid);`
	`5629`	`+`
	`5630`	`+ /* Delete KnownPrepared entry or 2PC file. */`
	`5631`	`+ KnownPreparedRemoveByXid(parsed.twophase_xid);`
`5628`	`5632`	`}`
`5629`	`5633`	`}`
`5630`	`5634`	`else if (info == XLOG_XACT_PREPARE)`
`5631`	`5635`	`{`
`5632`		`- /* the record contents are exactly the 2PC file */`
`5633`		`- StandbyAtPrepare(record);`
	`5636`	`+ /*`
	`5637`	`+ * If that transaction will not be commited by the end of recovery then we`
	`5638`	`+ * will need 2PC file (the record contents is exactly the 2PC file) to be able`
	`5639`	`+ * to commit that later.`
	`5640`	`+ * For now store xid and pointers to that record in KnownPreparedList.`
	`5641`	`+ */`
	`5642`	`+ KnownPreparedAdd(record);`
`5634`	`5643`	`}`
`5635`	`5644`	`else if (info == XLOG_XACT_ASSIGNMENT)`
`5636`	`5645`	`{`
Original file line number	Diff line number	Diff line change
`@@ -9547,8 +9547,8 @@ xlog_redo(XLogReaderState *record)`
`9547`	`9547`	`(errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",`
`9548`	`9548`	`checkPoint.ThisTimeLineID, ThisTimeLineID)));`
`9549`	`9549`
	`9550`	`+ KnownPreparedRecreateFiles(checkPoint.redo);`
`9550`	`9551`	`RecoveryRestartPoint(&checkPoint);`
`9551`		`- StandbyCheckPointTwoPhase(checkPoint.redo);`
`9552`	`9552`	`}`
`9553`	`9553`	`else if (info == XLOG_END_OF_RECOVERY)`
`9554`	`9554`	`{`