Hello,
Here is an update on my problem :
- the problem was caused by "VACUUM ANALYZE", but by a plain "VACUUM" ;
- it was exactly the same with manual and automatic "VACUUM ANALYZE" ;
- it was caused by a GIN index on a tsvector, using a very high (10000)
statistics target.
Setting back the statistics to 1000 reduced the amount of RAM used to a
very reasonable amount.
The value of 10000 is indeed not very realistic, but I think that would
deserve some mention on the documentation, if possible with an estimate
of the maximal memory usage for a given statistics target and table
size.
Do you think it's a good idea, and if so, if that estimate can be
reasonably made ?
Regards,
--
Gaël Le Mignot - gael@pilotsystems.net
Pilot Systems - 9, rue Desargues - 75011 Paris
Tel : +33 1 44 53 05 55 - www.pilotsystems.net
Gérez vos contacts et vos newsletters : www.cockpit-mailing.com