Merlin Moncure <mmonc...@gmail.com> writes:
> On Wed, Aug 24, 2011 at 1:24 PM, Daniel Farina <dan...@heroku.com> wrote:
>> At Heroku we use CREATE INDEX CONCURRENTLY with great success, but
>> recently when frobbing around some indexes I realized that there is no
>> equivalent for DROP INDEX, and this is a similar but lesser problem
>> (as CREATE INDEX takes much longer), as DROP INDEX takes an ACCESS
>> EXCLUSIVE lock on the parent table while doing the work to unlink
>> files, which nominally one would think to be trivial, but I assure you
>> it is not at times for even indexes that are a handful of gigabytes
>> (let's say ~=< a dozen).
> Are you sure that you are really waiting on the time to unlink the
> file? there's other stuff going on in there like waiting for lock,
> plan invalidation, etc. Point being, maybe the time consuming stuff
> can't really be deferred which would make the proposal moot.
Assuming the issue really is the physical unlinks (which I agree I'd
like to see some evidence for), I wonder whether the problem could be
addressed by moving smgrDoPendingDeletes() to after locks are released,
instead of before, in CommitTransaction/AbortTransaction. There does
not seem to be any strong reason why we have to do that before lock
release, since incoming potential users of a table should not be trying
to access the old physical storage after that anyway.
regards, tom lane
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
