If backup-end record is not seen, and we reach end of recovery from a

streamed backup, throw an error and refuse to start up. The restore has not
finished correctly in that case and the data directory is possibly corrupt.
We already errored out in case of archive recovery, but could not during
crash recovery because we couldn't distinguish between the case that
pg_start_backup() was called and the database then crashed (must not error,
data is OK), and the case that we're restoring from a backup and not all
the needed WAL was replayed (data can be corrupt).

To distinguish those cases, add a line to backup_label to indicate
whether the backup was taken with pg_start/stop_backup(), or by streaming
(ie. pg_basebackup).

This is a different implementation than what I committed to 9.2 a week ago.
That implementation was not back-patchable because it required re-initdb.

Fujii Masao

src/backend/access/transam/xlog.c

@@ -659,7 +659,8 @@ static bool CheckForStandbyTrigger(void);
 static void xlog_outrec(StringInfo buf, XLogRecord *record);
 #endif
 static void pg_start_backup_callback(int code, Datum arg);
-static bool read_backup_label(XLogRecPtr *checkPointLoc);
+static bool read_backup_label(XLogRecPtr *checkPointLoc,
+				  bool *backupEndRequired);
 static void rm_redo_error_callback(void *arg);
 static int	get_sync_bit(int method);
 
@@ -5995,6 +5996,7 @@ StartupXLOG(void)
 	XLogRecord *record;
 	uint32		freespace;
 	TransactionId oldestActiveXID;
+	bool		backupEndRequired = false;
 
 	/*
 	 * Read control file and check XLOG status looks valid.
@@ -6128,7 +6130,7 @@ StartupXLOG(void)
 	if (StandbyMode)
 		OwnLatch(&XLogCtl->recoveryWakeupLatch);
 
-	if (read_backup_label(&checkPointLoc))
+	if (read_backup_label(&checkPointLoc, &backupEndRequired))
 	{
 		/*
 		 * When a backup_label file is present, we want to roll forward from
@@ -6304,10 +6306,17 @@ StartupXLOG(void)
 		}
 
 		/*
-		 * set backupStartPoint if we're starting recovery from a base backup
+		 * Set backupStartPoint if we're starting recovery from a base backup.
+		 * However, if there was no recovery.conf, and the backup was taken
+		 * with pg_start_backup(), we don't know if the server crashed before
+		 * the backup was finished and we're doing crash recovery on the
+		 * original server, or if we're restoring from the base backup. We
+		 * have to assume we're doing crash recovery in that case, or the
+		 * database would refuse to start up after a crash.
 		 */
-		if (haveBackupLabel)
+		if ((InArchiveRecovery && haveBackupLabel) || backupEndRequired)
 			ControlFile->backupStartPoint = checkPoint.redo;
+
 		ControlFile->time = (pg_time_t) time(NULL);
 		/* No need to hold ControlFileLock yet, we aren't up far enough */
 		UpdateControlFile();
@@ -6670,15 +6679,8 @@ StartupXLOG(void)
 
 		/*
 		 * Ran off end of WAL before reaching end-of-backup WAL record, or
-		 * minRecoveryPoint. That's usually a bad sign, indicating that you
-		 * tried to recover from an online backup but never called
-		 * pg_stop_backup(), or you didn't archive all the WAL up to that
-		 * point. However, this also happens in crash recovery, if the system
-		 * crashes while an online backup is in progress. We must not treat
-		 * that as an error, or the database will refuse to start up.
+		 * minRecoveryPoint.
 		 */
-		if (InArchiveRecovery)
-		{
 		if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
 			ereport(FATAL,
 					(errmsg("WAL ends before end of online backup"),
@@ -6687,7 +6689,6 @@ StartupXLOG(void)
 			ereport(FATAL,
 					(errmsg("WAL ends before consistent recovery point")));
 	}
-	}
 
 	/*
 	 * Consider whether we need to assign a new timeline ID.
@@ -8990,6 +8991,8 @@ do_pg_start_backup(const char *backupidstr, bool fast, char **labelfile)
 						 startpoint.xlogid, startpoint.xrecoff, xlogfilename);
 		appendStringInfo(&labelfbuf, "CHECKPOINT LOCATION: %X/%X\n",
 						 checkpointloc.xlogid, checkpointloc.xrecoff);
+		appendStringInfo(&labelfbuf, "BACKUP METHOD: %s\n",
+						 exclusive ? "pg_start_backup" : "streamed");
 		appendStringInfo(&labelfbuf, "START TIME: %s\n", strfbuf);
 		appendStringInfo(&labelfbuf, "LABEL: %s\n", backupidstr);
 
@@ -9719,15 +9722,19 @@ pg_xlogfile_name(PG_FUNCTION_ARGS)
  *
  * Returns TRUE if a backup_label was found (and fills the checkpoint
  * location and its REDO location into *checkPointLoc and RedoStartLSN,
- * respectively); returns FALSE if not.
+ * respectively); returns FALSE if not. If this backup_label came from a
+ * streamed backup, *backupEndRequired is set to TRUE.
  */
 static bool
-read_backup_label(XLogRecPtr *checkPointLoc)
+read_backup_label(XLogRecPtr *checkPointLoc, bool *backupEndRequired)
 {
 	char		startxlogfilename[MAXFNAMELEN];
 	TimeLineID	tli;
 	FILE	   *lfp;
 	char		ch;
+	char		backuptype[20];
+
+	*backupEndRequired = false;
 
 	/*
 	 * See if label file is present
@@ -9760,6 +9767,16 @@ read_backup_label(XLogRecPtr *checkPointLoc)
 		ereport(FATAL,
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
+	/*
+	 * BACKUP METHOD line didn't exist in 9.1beta3 and earlier, so don't
+	 * error out if it doesn't exist.
+	 */
+	if (fscanf(lfp, "BACKUP METHOD: %19s", backuptype) == 1)
+	{
+		if (strcmp(backuptype, "streamed") == 0)
+			*backupEndRequired = true;
+	}
+
 	if (ferror(lfp) || FreeFile(lfp))
 		ereport(FATAL,
 				(errcode_for_file_access(),