Adrian von Bidder wrote:
> With our test dump, the db (after import) is ca. 300M on disk, ca. half in
> WAL files (pg_xlog.) If I could mostly get rid of the WAL (keep it to a
> bare minimum and run pg without fsync, something like that), the remaining
> 160 to 180M would be ok.
>
Drop checkpoint_segments=1 and checkpoint_completion_target=0 before
running your load. It will take longer and the system will run slower,
but it will reduce the typical WAL footprint. Normally the number of
16MB WAL files will never been more than (2 +
checkpoint_completion_target) * checkpoint_segments + 1, although the
burst size can be bigger than that under load. You should settle into
where there's only 48MB being used there with those parameter changes.
--
Greg Smith 2ndQuadrant Baltimore, MD
PostgreSQL Training, Services and Support
greg@2ndQuadrant.com www.2ndQuadrant.com
