Rob Adams <robfadams@cox.net> writes:
> archive_timeout only seems to work if it's >= checkpoint_timeout.
Hmm, no, they should be pretty independent. Define "seems to work"
please?
One possible connection is that an xlog file switch will not actually
happen unless some xlog output has been generated since the last switch.
If you were watching an otherwise-idle system then maybe the checkpoint
records are needed to make it look like a switch is needed. OTOH if
it's *that* idle then the checkpoints should be no-ops too. So we
need a bit more context to understand what's happening. How often
do real updates happen on your database?
regards, tom lane