>> ERROR: canceling autovacuum task
>> CONTEXT: automatic vacuum of table "metadb.test.transactions"
>
> Are these happening regularly? They indicate that something is
> happening on the table that collides with what autovacuum needs to do,
> and autovacuum defers its task. For this to happen you need to be doing
> ALTER TABLE or similar however; normal UPDATE/INSERT/DELETE should not
> cause autovacuum to cancel itself.
>
I am not using an ALTER table command but I am doing periodic ANALYZEs
to evaluate the table size. Could this be causing the problem? I notice
that stopping the ANALYZE calls appears to eliminate the canceled
autovacuum.
What concerns me is that once the size has grown, even a VACUUM FULL
doesn't recover the space. Regular external VACUUMs keep the table at
around 10MB but if I use autovacuum and it grows to 40MB, a VACUUM FULL
will only get it down to 35MB. Is it possible that a canceled autovacuum
could result in permanently lost space?
Out of interest, what kind of fragmentation overhead should I expect if
I have a table in which I maintain a fixed number of rows. eg. A 20000
row table which is 6MB before rows are wrapped out will obviously use a
larger disk footprint as rows are added and deleted. Anyone have a rule
of thumb which works for them?
Thanks for the response,
Stuart