[Issue #192] Handle 0 offset STOP LSN on master

Author: gsmolk

src/backup.c

@@ -1740,8 +1740,10 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
 
 		if (!XRecOffIsValid(stop_backup_lsn_tmp))
 		{
-			/* It is ok for replica to return STOP LSN with NullXRecOff */
-			if (backup->from_replica && XRecOffIsNull(stop_backup_lsn_tmp))
+			/* It is ok for replica to return STOP LSN with NullXRecOff
+			 * UPD: Apparently it is ok even for master.
+			 */
+			if (XRecOffIsNull(stop_backup_lsn_tmp))
 			{
 				char	   *xlog_path,
 							stream_xlog_path[MAXPGPATH];
@@ -1798,13 +1800,20 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
 
 				/* Get the first record in segment with current stop_lsn */
 				lsn_tmp = get_first_record_lsn(xlog_path, segno, backup->tli,
-										    instance_config.xlog_seg_size);
+										       instance_config.xlog_seg_size,
+										       instance_config.archive_timeout);
 
 				/* Check that returned LSN is valid and greater than stop_lsn */
 				if (XLogRecPtrIsInvalid(lsn_tmp) ||
 					!XRecOffIsValid(lsn_tmp) ||
 					lsn_tmp < stop_backup_lsn_tmp)
 				{
+					/* Backup from master should error out here */
+					if (!backup->from_replica)
+						elog(ERROR, "Failed to get next WAL record after %X/%X",
+									(uint32) (stop_backup_lsn_tmp >> 32),
+									(uint32) (stop_backup_lsn_tmp));
+
 					/* No luck, falling back to looking up for previous record */
 					elog(WARNING, "Failed to get next WAL record after %X/%X, "
 								"looking for previous WAL record",

src/parsexlog.c

@@ -631,13 +631,14 @@ wal_contains_lsn(const char *archivedir, XLogRecPtr target_lsn,
  */
 XLogRecPtr
 get_first_record_lsn(const char *archivedir, XLogSegNo	segno,
-				 TimeLineID tli, uint32 wal_seg_size)
+					 TimeLineID tli, uint32 wal_seg_size, int timeout)
 {
 	XLogReaderState *xlogreader;
-	XLogReaderData reader_data;
-	XLogRecPtr	record = InvalidXLogRecPtr;
-	XLogRecPtr	startpoint;
-	char	wal_segment[MAXFNAMELEN];
+	XLogReaderData   reader_data;
+	XLogRecPtr       record = InvalidXLogRecPtr;
+	XLogRecPtr       startpoint;
+	char             wal_segment[MAXFNAMELEN];
+	int              attempts = 0;
 
 	if (segno <= 1)
 		elog(ERROR, "Invalid WAL segment number " UINT64_FORMAT, segno);
@@ -653,13 +654,22 @@ get_first_record_lsn(const char *archivedir, XLogSegNo	segno,
 	/* Set startpoint to 0 in segno */
 	GetXLogRecPtr(segno, 0, wal_seg_size, startpoint);
 
-	record = XLogFindNextRecord(xlogreader, startpoint);
+	while (attempts <= timeout)
+	{
+		record = XLogFindNextRecord(xlogreader, startpoint);
 
-	if (XLogRecPtrIsInvalid(record))
-		record = InvalidXLogRecPtr;
-	else
-		elog(LOG, "First record in WAL segment \"%s\": %X/%X", wal_segment,
-				(uint32) (record >> 32), (uint32) (record));
+		if (XLogRecPtrIsInvalid(record))
+			record = InvalidXLogRecPtr;
+		else
+		{
+			elog(LOG, "First record in WAL segment \"%s\": %X/%X", wal_segment,
+					(uint32) (record >> 32), (uint32) (record));
+			break;
+		}
+
+		attempts++;
+		sleep(1);
+	}
 
 	/* cleanup */
 	CleanupXLogPageRead(xlogreader);

src/pg_probackup.h

@@ -957,7 +957,7 @@ extern XLogRecPtr get_prior_record_lsn(const char *archivedir, XLogRecPtr start_
 								   bool seek_prev_segment, uint32 seg_size);
 
 extern XLogRecPtr get_first_record_lsn(const char *archivedir, XLogRecPtr start_lsn,
-									TimeLineID tli, uint32 wal_seg_size);
+									   TimeLineID tli, uint32 wal_seg_size, int timeout);
 
 /* in util.c */
 extern TimeLineID get_current_timeline(PGconn *conn);