http://www.postgresql.org/docs/8.3/static/wal-configuration.html
Is it right that checkpoint_timeout means the amount of time up to which
you agree to lose data in the event of a power crash? What if I set it
to 1 hour (and bump checkpoint_segments accordingly), does it mean that
I'm willing to lose up to 1 hour of data? I'm thinking about increasing
checkpoint_timeout to mitigate the full_page_writes bloat.
BTW how are transactions WAL logged? Do the logs include data too? In
this case, am I right that the effects of full_page_writes=on serve as a
starting data page on top of which to replay transactions when doing
crash recovery?