postmaster startup failure

Hannu Krosing <hannu@tm.ee> writes:

WHen running PostgreSQL 7.3.3-1 (from rpm's) on Redhat 9.0 I got the
following in logs and the postmaster will not start up.
PANIC: XLogWrite: write request 0/30504000 is past end of log
0/30504000

This looks like you've stumbled across some sort of boundary-condition
bug in xlog.c. It must be a pretty low-probability situation since
we've not seen it before.

How large is the database in question --- would it be feasible to send
it to me (as a tar of the whole $PGDATA directory) for examination?

regards, tom lane

Hannu Krosing <hannu@tm.ee> writes:

WHen running PostgreSQL 7.3.3-1 (from rpm's) on Redhat 9.0 I got the
following in logs and the postmaster will not start up.
PANIC: XLogWrite: write request 0/30504000 is past end of log
0/30504000

Ugh. The reason we hadn't seen this happen in the field was that it is
a bug I introduced in a patch two months ago :-(

7.3.3 will in fact fail to start up, with the above error, any time the
last record of the WAL file ends exactly at a page boundary. I think
we're gonna need a quick 7.3.4 ...

If you want a source patch for 7.3.3, here it is.

regards, tom lane

*** src/backend/access/transam/xlog.c.orig	Thu May 22 10:39:49 2003
--- src/backend/access/transam/xlog.c	Thu Jul 17 12:36:20 2003
***************
*** 2483,2488 ****
--- 2483,2489 ----
  				EndOfLog;
  	XLogRecord *record;
  	char	   *buffer;
+ 	uint32		freespace;

/* Use malloc() to ensure record buffer is MAXALIGNED */
buffer = (char *) malloc(_INTL_MAXLOGRECSZ);
***************
*** 2678,2685 ****
memcpy((char *) Insert->currpage, readBuf, BLCKSZ);
Insert->currpos = (char *) Insert->currpage +
(EndOfLog.xrecoff + BLCKSZ - XLogCtl->xlblocks[0].xrecoff);
- /* Make sure rest of page is zero */
- MemSet(Insert->currpos, 0, INSERT_FREESPACE(Insert));

LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;

--- 2679,2684 ----
***************
*** 2689,2694 ****
--- 2688,2714 ----

  	XLogCtl->LogwrtRqst.Write = EndOfLog;
  	XLogCtl->LogwrtRqst.Flush = EndOfLog;
+ 
+ 	freespace = INSERT_FREESPACE(Insert);
+ 	if (freespace > 0)
+ 	{
+ 		/* Make sure rest of page is zero */
+ 		MemSet(Insert->currpos, 0, freespace);
+ 		XLogCtl->Write.curridx = 0;
+ 	}
+ 	else
+ 	{
+ 		/*
+ 		 * Whenever Write.LogwrtResult points to exactly the end of a page,
+ 		 * Write.curridx must point to the *next* page (see XLogWrite()).
+ 		 *
+ 		 * Note: it might seem we should do AdvanceXLInsertBuffer() here,
+ 		 * but we can't since we haven't yet determined the correct StartUpID
+ 		 * to put into the new page's header.  The first actual attempt to
+ 		 * insert a log record will advance the insert state.
+ 		 */
+ 		XLogCtl->Write.curridx = NextBufIdx(0);
+ 	}

#ifdef NOT_USED
/* UNDO */

Tom Lane kirjutas N, 17.07.2003 kell 19:49:

Ugh. The reason we hadn't seen this happen in the field was that it is
a bug I introduced in a patch two months ago :-(

7.3.3 will in fact fail to start up, with the above error, any time the
last record of the WAL file ends exactly at a page boundary. I think
we're gonna need a quick 7.3.4 ...

If you want a source patch for 7.3.3, here it is.

Thanks!

-----------
Hannu

I ran into this too. Patched the code with Tom's change and it works fine.

Thanks again Tom!

Richard Schilling