I've been running pg_autovacuum for a week or so, now and while I like
it I'm having trouble trusting it.
I have several small (~0-50 record) working tables that have high
insert/update/delete activity. When I first set up pg_autovacuum they
were being vacuumed every ~1-4 hours during the day and less often at
night. Most of the time the tuple count was reasonable.
Now these tables haven't been vacuumed for a couple of days although
the activity on them hasn't changed.
What has changed is that pg_autovacuum reports that the tuple count is
(actually, was at last vacuum) over 12,000 where it is actually more
like 16 so the next vacuum won't happen for quite a while. 12,000+ is
an unreasonable tuple count - it would represent several week's
activity happening at once.
This has happened to at least two working tables. I'm currently
running it as:
pg_autovacuum -d 2 >> autovacuum.log 2>&1 &
I have also encountered another oddity. If I run pg_autovacuum with
all defaults then it appears to analyze every table in every database
every pass.
Any ideas? My searches have turned up nothing useful. Version is
7.4.1.
Cheers,
Steve
---------------------------(end of broadcast)---------------------------
TIP 5: Have you checked our extensive FAQ?
http://www.postgresql.org/docs/faqs/FAQ.html