* vacstats has collected stats on all of our databases except koji
* vacstats.py now has an analyze mode. You can use it to create a
reasonable default vacuum policy.
* I've used vacstats to construct vacuum scripts for all of our
databases except koji. I've pushed them out to db2 and am watching them
to make sure that they run fine today. I don't anticipate a problem
but I'll have to evaluate whether to add a weekly script or turn the
dailies into every other day vacuumings once we add the koji tables.
* We've added three tables from koji to the hourlies since we know that
they are high update tables that need to be vacuumed frequently.
* The vacstat.py check mode now checks if we're in danger of running out
of transaction ids and prints a warning that should be emailed to
firstname.lastname@example.org if so. Dealing with that is a matter of performing a vacuum
of that particular db (rather than table by table). If the database
happens to be koji we might want to schedule that for the weekend as the
weekend is the lowest use time for koji. (Note -- it will result in a
slow down, not in an outage.)
Thanks to mbonnet for the information on the tables and queries for the
latter two points.
At this point the things left to do are:
* Get the rest of koji vacuuming integrated into the scripts. I'll be
doing this in the next few days.
* Find a time when we can do a vacuum full of some tables. The current
vacstats algorithm suggests the following tables:
Vacuum full mirrormanager host_category_dir: Freespace Percent
96.8641231524%, 735480624.0 Bytes
* Decide whether to upgrade postgres from 8.1 to 8.3. This has been
recommended by several postgres people but there's extremely little
chance that 8.3 will be going into RHEL5 so we'll have to decide whether
to move to F9, backport a package to RHEL5 or just stay with 8.1 for now.
Fedora-infrastructure-list mailing list