On Tue, 27 Oct 2009, Alvaro Herrera wrote:
> Now 40 mins walking those pages to figure out that they need to be
> truncated, I concede that it's too much. Maybe we shouldn't be doing a
> backwards scan; perhaps this breaks the OS readahead and make it even
> slower.
I've watched that take hours before on a large table after purging
hundreds of gigabytes of old historical data. Improvements to speed that
up like scanning more efficiently would be welcome. But given the
potential for a really bad worst-case here, I have to wonder if this
really needs to get broken up into bits with finer locking instead of
micromanaging the details.
(Yes, I should have been using date-range partitioning instead and just
dropped the old partitions, but sometimes these things grow only after
you've made design decisions the wrong way)
--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
