On 2013-06-26 23:42:55 -0700, Peter Geoghegan wrote:
> On Wed, Jun 26, 2013 at 11:27 PM, Andres Freund <andres@2ndquadrant.com> wrote:
> > Why not do this from a function/background worker in the backend where
> > you can go via the buffer manager to avoid torn pages et al. If you use
> > a buffer strategy the cache poisoning et al should be controlleable.
>
> I had considered that, but thought it might be a little bit
> aggressive, even with a strategy of BAS_BULKREAD.
Well, you can influence the pacing yourself, you don't need to rely on
the strategy for that. I'd only use it because of the ringbuffer logic
it has to avoid trashing the cache.
> Maybe the kludge I
> have in mind might not end up being that bad in practice, and would
> certainly perform better than an approach that used the buffer
> manager.
What do you have in mind then?
> But then, going through shared_buffers could be worth the
> overhead, if only for the peace of mind of not relying on something
> that is subtly broken.
Spurious alarms quickly lead to people ignoring them, consciously or
not, so trying to take care not to go there sounds like a good idea.
Greetings,
Andres Freund
-- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services