Mindaugas Riauba wrote:

Might e aggressive enough, but might not.  I have seen some people run
-V 0.1.  Also you probably don't need -A that low.  This could an issue
where analyze results in an inaccurate reltuples value which is
preventing autovacuum from doing it's job.  Could you please run it with
-d 2 and show us the relevant log output.


 Relevant parts are below. And we had to set so aggressive analyze because
otherwise planer statistics were getting old too fast. As I said table has
very
high turnover most of the records live here only for a few seconds.

Looked like pg_autovacuum is operating as expected. One of the annoying limitations of pg_autovacuum in current releases is that you can't set thresholds on a per table basis. It looks like this table might require an even more aggressive vacuum threshold. Couple of thoughts, are you sure it's the table that is growing and not the indexes? (assuming this table has indexes on it).
 And one more question - anyway why table keeps growing? It is shown that
it occupies
<10000 pages and max_fsm_pages = 200000 so vacuum should keep up with the
changes?
Or is it too low according to pg_class system table? What should be the
reasonable value?

Does the table keep growing? Or does it grow to a point an then stop growing? It's normal for a table to operate at a steady state size that is bigger that it's fresly "vacuum full"'d size. And with -V set at 0.5 it should be at a minimum 50% larger than it's minimum size. Your email before said that this table went from 20M to 70M but does it keep going? Perhaps it would start leveling off at this point, or some point shortly there-after.

Anyway, I'm not sure if there is something else going on here, but from the log it looks as though pg_autovacuum is working as advertised.

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

Reply via email to