On Mon, May 22, 2006 at 03:49:01PM -0700, Shelby Cain wrote:
> My experience with job queues comes from clients that mostly use Oracle as the backend. However, even with Oracle a
queuetable should be storing information about a job and not have records unnecessarily locked simply because they are
being"worked on" by another hypothetical "job runner" process... by this I mean that the status of a job should be
updatedto a distinct state at any given moment in time (eg: unprocessed, processing, processed). In the case I present
above,if you are using Postgresql you wouldnt have any open long-running transactions on that table and vacuuming
shouldwork... or am I misunderstanding the issue?
The issue is that vacuum has to base it's decisions not on the oldest
running transaction that has locks on a table, but on the oldest running
transaction in the entire database, because that transaction could start
reading any table at any time. Until that changes, long-running
transactions of any kind and heavy-update tables simply won't mix well
at all in a single database.
I recently proposed a way around this [1], but it didn't get much
traction.
[1] http://archives.postgresql.org/pgsql-general/2006-05/msg00184.php
--
Jim C. Nasby, Sr. Engineering Consultant jnasby@pervasive.com
Pervasive Software http://pervasive.com work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf cell: 512-569-9461