On Tue, 2006-11-07 at 17:20 -0500, Tom Lane wrote:
> Jeff Davis <pgsql@j-davis.com> writes:
> > LOG: restored log file "00000001000000170000002B" from archive
> > LOG: record with zero length at 17/2B6EACC8
> > LOG: redo done at 17/2B6EAC84
>
> It looks to me like you archived this log file before it was fully
> written. You should take a close look at your archiving procedures.
>
Hmm... the only way the file gets there is archive command. I didn't do
any manual moves of the segments, certainly not last Sunday at 1am
(which is when this segment was put there).
From postgresql.conf:
archive_command = '/usr/local/pgsql/bin/archive_command.bash "%p" "%f"'
and:
-----------
$ cat /usr/local/pgsql/bin/archive_command.bash
#!/usr/local/bin/bash
# $1 is the full path to the file to archive
# $2 is the filename
WAL=$1
WAL_NAME=$2
WAL_ARCHIVE=/path/to/archive
echo test ! -f $WAL_ARCHIVE/$WAL_NAME && gzip -c $WAL >
$WAL_ARCHIVE/$WAL_NAME.gz
------------
Any obvious holes in there? Is gzip doing something weird? It worked
with all my other WAL segments, and I've tested the backups multiple
times.
Also, the WAL segment is exactly the right size, although I suppose
there's no way for it not to be.
Regards,
Jeff Davis