On Wed, Apr 13, 2011 at 4:45 AM, Henry C. <henka@cityweb.co.za> wrote:
> Greets,
>
> Pg 9.0.3.
>
> I'm trying out Pg's built-in replication for the first time and noticed
> something odd.
>
> On the slave I see the following in the logs (after rsyncing all from master
> to slave and firing up Pg on the slave):
>
> ...
> restored log file "000000010000018E0000000E" from archive
> restored log file "000000010000018E0000000F" from archive
> consistent recovery state reached at 18E/10000000
> restored log file "000000010000018E00000010" from archive
> cp: cannot stat `/home/arc/000000010000018E00000011': No such file or directory
> unexpected pageaddr 18D/91000000 in log file 398, segment 17, offset 0
> cp: cannot stat `/home/arc/000000010000018E00000011': No such file or directory
> streaming replication successfully connected to primary
> ...
>
> /home/arc is an NFS mount from master and is where the WAL archive is kept
> (yes, I'll move it eventually; for now I'm just testing).
>
> Things seem to run fine up until (and after) log file
> 000000010000018E00000011. That particular file is definitely present. Why
> would cp(1) fail to stat the file when it worked fine for all the others?
I guess that file didn't exist in the archive at the moment when cp failed.
It was archived after that. So you observed that file in the archive.
> I notice from another mailing list post that 'unexpected pageaddr' is possibly
> not that serious and is probably unrelated to the cp/stat error above.
>
> However, since recovery seems to have skipped a log file, what would that mean
> in terms of the slave being a true copy of master and integrity of the data?
When the standby fails to read the WAL file from the archive, it tries to read
that from the master via replication connection. So the standby would not skip
that file.
Regards,
--
Fujii Masao
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center
