Circling back on this one, I had a look at our analyze code. I found one place where *maybe* we weren't freeing memory and freed it, but analyzing a 2M record table I barely see any bump up in memory usage (from 22M up to 24M at peak) during analyze. And the change I made didn't appear to alter that (though the objects were probably all small enough that they weren't being detoasted into copies in any event). Though maybe with a really big table? (with really big objects?) Though still, doesn't analyze just pull a limited sample (30K approx max) so why would table size make any difference after a certain point?
P. On Tue, Mar 3, 2015 at 3:17 PM, Tom Lane <t...@sss.pgh.pa.us> wrote: > wambacher <wnordm...@gmx.de> writes: >> My system has 24GB of real memory but after some hours one autovacuum worker >> is using 80-90% of memory, the OOM-Killer (out of memory killer) kills the >> process with kill -9 and the postgresql-server is restarting because of that >> problem. > >> i changed the base configuration to use very small buffers, restartetd the >> server twice but the problem still exists. > >> i think, it's allways the same table and that table is huge: 111GB data and >> 3 Indices with 4GB, 128 GB and 12 GB. It's the table planet_osm_ways from >> openstreetmap. maybe that helps. > > Maybe you could reduce the statistics targets for that table. > > I think we've heard that the analyze functions for PostGIS data types are > memory hogs, too --- maybe it's worth inquiring about that on the postgis > mailing lists. > > regards, tom lane > > > -- > Sent via pgsql-general mailing list (pgsql-general@postgresql.org) > To make changes to your subscription: > http://www.postgresql.org/mailpref/pgsql-general -- Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
