Gavan Schneider wrote:
> Therefore I wonder if I should just go down to 4MB, no
> archive_timeout, and accept the (max) 4MB of updated pages as my
> (worst case) loss if everything should go wrong at once? And,
> maybe do the segment downsize when a new version needs
> installing :)
You might want to consider grabbing the source for pg_clearxlogtail
from pgfoundry and piping through that followed by gzip in your
archive script. An "empty" WAL file tends to be about 16KB that
way, rather than (for us) ranging between 4MB and 8MB. That way you
can keep the archive_timeout fairly short.
-Kevin