Claudio Freire <klaussfreire(at)gmail(dot)com> wrote:
> Now, I'm thinking those writes are catching the DB at a bad moment
-
> we do have regular very write-intensive peaks.
>
> Maybe I should look into increasing shared buffers?
As already pointed out, SELECT FOR UPDATE will require a disk write
of the tuple(s) read. If these are glutting, increasing
shared_buffers would tend to make things worse. You might want to
make the background writer more aggressive or *reduce*
shared_buffers to better spread the output.
Hopefully you have a BBU RAID controller configured for write-back.
If not you should. If you do, another thing which might help is
increasing the cache on that controller. Or you could move WAL to a
separate file system on a separate controller with its own BBU
write-back cache.
-Kevin