Alvaro Herrera <alvherre@commandprompt.com> writes:
> Simon Riggs wrote:
>> Ouch! We did discuss that also. Flushing the buffercache is nasty with
>> very large caches, so this makes autovacuum much less friendly - and
>> could take a seriously long time if you enforce the vacuum delay
>> costings.
> Hmm, isn't the buffer cache aware of a vacuum operation?
Yeah. What would probably happen is that we'd dump off most of the
dirtied pages to the kernel, which would likely still have a lot of them
in kernel buffers pending write. But then we'd have to fsync the table
--- so a physical write storm would ensue, which we have no way to
throttle.
I think the don't-truncate-clog approach is a much better answer.
regards, tom lane
