Laurent Raufaste wrote:
> I have a big PG server dedicated to serve only SELECT queries.
> The database is updated permanently using Slony.
>
> [...] load is huge.
>
> In order to locate the problem, I stopped Slony (no updates anymore),
> mounted the database and index partitions with the sync option (no FS
> write cache), and the problem happens faster, like 2 minutes after
> having plugged the client (and the queries) to it.
> I can reproduce the problem at will.
>
> I tried to see if some file size were increasing a lot, and found
> nothing more than the usual DB increase (DB is constantly updated by
> Slony).
>
> What does it writes so much in the base directory ? If it's some
> temporary table or anything, how can I locate it so I can fix the
> problem ?
It could be a temporary file, although your work_mem setting is already
quite large.
Can you attach to the rogue backend with "strace" (if you have Linux, else
something else maybe) to see what it does and use "lsof" to see what files
it has open?
Yours,
Laurenz Albe