On Aug 7, 2008, at 9:49 AM, Robert Haas wrote:
This proposal sounds like it would target batch jobs, because those are the kinds of jobs that where you can predict in advance what tables will be needed. I don't know whether my personal set of problems with MVCC syncs up with anyone else's, but this is rarely how I get bitten. Usually, what happens is that a user session (psql or web server connection) gets left in a transaction for days or weeks. Now the batch jobs (which are doing lots of updates) start creating tons of bloat, but it's not their snapshot that is causing the problem.
We have some cron'd scripts that check for long-running queries, idle in transaction and just plain idle. The scripts will kill sessions when the sit in different states for too long. It would be nice if the database could handle this (no, statement_timeout won't work, because the user can just change it to whatever they want), but I don't know how many other people have this need. I guess I could at least put the scripts up on pgFoundry...
-- Decibel!, aka Jim C. Nasby, Database Architect [EMAIL PROTECTED] Give your computer some brain candy! www.distributed.net Team #1828
smime.p7s
Description: S/MIME cryptographic signature
