On 07/27/2012 09:32 AM, Marek Kielar wrote:
> Hi, again,
>
> I'm sorry about the lack of version information - I concentrated so much on describing the problem correctly, that I
forgotto provide basic information. The version at locations is 9.1.4 (though this is irrelevant now), the server is
9.0.4.
>
> We found what the problem was. Another problem stems from it, however. Please read on.
>
> To add to the information already provided - we have a two-way backup of the template database. One is a WAL
replicationand the other is londiste (skytools) replication with periodic complete copy. As it turned out, the "stable"
scriptuses not, as we remembered, the actual template database but the londiste-replicated database which was to make
nextcomplete copy a few days ago. The copy did not complete, however - the schema-table-column structure transfer
completed,but the constraints and triggers did not get through somehow, as there was a lack of hard drive space.
Diggingon it, we found out that the drive's space was not used up by files in the filesystem, it was filled with
deletedfiles that postgresql server was still clinging on to, probably for a good while. After restarting the server
many,many gigabytes were suddenly made available on disk. And this is the new problem - the server has quite a
throughputand this is probably what causes the "leakage". How can we force
the server to let go of the files? Or maybe it is an actual leak that needs to be studied upon?
>
> On a side note, obviously, the Windows dump came out alright because it was from the proper database, not the
replicatedcopy.
What where the deleted files?
WAL, Logs, other?
What type of WAL replication are you doing?
Streaming, log shipping, etc?
What are your settings for the WAL replication?
In particular wal_keep_segments ?
Is the WAL replication actually working?
>
> Best regards,
> Marek Kielar
>
>
--
Adrian Klaver
adrian.klaver@gmail.com