On Mon, 12 Mar 2007, Tom Lane wrote:
> It might also be interesting to know exactly how many buffers were
> pinned at the time the scan passed over them. In theory it should be a
> small fraction, but maybe it isn't ...
It is; the theory holds for all the tests I tried today. The actual
pinned buffers were so few (typically a fraction of the clients) that I
reverted to just lumping them in with the recently used ones. To better
reflect the vast majority of what it's interacting with, in my patch I
renamed the SyncOneBuffer "skip_pinned" to "skip_recently_used". It seems
natural that something currently pinned would also be considered recently
used, the current naming I didn't find so obvious.
I'm also now collecting clean vs. dirty usage histogram counts as well
since you suggested it. Nothing exciting to report there so far, may note
something interesting after I collect more data.
--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD