On Fri, Apr 25, 2014 at 10:29 AM, Дмитрий Шалашов <skaurus@xxxxxxxxx> wrote: > Previously during load disk was 100% busy; now we have around 100 active > state queries, 100% loaded proc, but disk is virtually idle... That was happen after changing checkpoit_segments setting? > > > Best regards, > Dmitriy Shalashov > > > 2014-04-25 12:22 GMT+04:00 Дмитрий Шалашов <skaurus@xxxxxxxxx>: > >> Hi Ilya! >> >> > Actually, thise two things are tightly bound and there is no chance to >> > avoid vacuum, you can only postpone it, this kind of work eventually >> > supposed to be done. >> >> I understand that autovacuum has to be done, but not right after previous >> autovacuum? And then again and again. >> And after cancelling that first autovacuum I started another one by hand; >> from there no autovacuum was cancelled. >> >> > ionice autovacuum instead of mission critical ckeckpointer or bgwriter >> Yeah, that was desperate. I restarted server when I had a chance - to drop >> my ionice settings back to defaults. >> >> > Which exact values have you in the following settings: >> >> autovacuum_analyze_scale_factor = 0.1 >> autovacuum_analyze_threshold = 50 >> autovacuum_freeze_max_age = 200000000 >> autovacuum_max_workers = 3 >> autovacuum_naptime = 60 >> autovacuum_vacuum_cost_delay = 20 >> autovacuum_vacuum_cost_limit = -1 >> autovacuum_vacuum_scale_factor = 0.2 >> autovacuum_vacuum_threshold = 50 >> log_autovacuum_min_duration = 0 >> >> All defaults except last one I believe. >> >> >> Minwhile I noticed in the night logs: >> checkpoints are occurring too frequently (138 seconds apart) >> Consider increasing the configuration parameter "checkpoint_segments". >> >> Increased checkpoint_segments to 256 and reloaded config. >> >> >> Best regards, >> Dmitriy Shalashov >> >> >> 2014-04-25 12:12 GMT+04:00 Ilya Kosmodemiansky >> <ilya.kosmodemiansky@xxxxxxxxxxxxxxxxxxxxxxxxx>: >> >>> Hi Dmitry, >>> >>> On Fri, Apr 25, 2014 at 9:47 AM, Дмитрий Шалашов <skaurus@xxxxxxxxx> >>> wrote: >>> > cancelled autovacuum and it seems to help. >>> >>> > In the morning autovacuum was back. And then it finished and I gone to >>> > work. >>> >>> Actually, thise two things are tightly bound and there is no chance to >>> avoid vacuum, you can only postpone it, this kind of work eventually >>> supposed to be done. >>> >>> What you really need to do as a first thing - configure your >>> autovacuum aggressively enough and then mayde ionice autovacuum >>> instead of mission critical ckeckpointer or bgwriter. >>> >>> Which exact values have you in the following settings: >>> >>> autovacuum_analyze_scale_factor >>> autovacuum_analyze_threshold >>> autovacuum_freeze_max_age >>> autovacuum_max_workers >>> autovacuum_naptime >>> autovacuum_vacuum_cost_delay >>> autovacuum_vacuum_cost_limit >>> autovacuum_vacuum_scale_factor >>> autovacuum_vacuum_threshold >>> log_autovacuum_min_duration >>> >>> ? >>> >>> Best regards, Ilya >>> > >>> > Best regards, >>> > Dmitriy Shalashov >>> >>> >>> >>> -- >>> Ilya Kosmodemiansky, >>> >>> PostgreSQL-Consulting.com >>> tel. +14084142500 >>> cell. +4915144336040 >>> ik@xxxxxxxxxxxxxxxxxxxxxxxxx >> >> > -- Ilya Kosmodemiansky, PostgreSQL-Consulting.com tel. +14084142500 cell. +4915144336040 ik@xxxxxxxxxxxxxxxxxxxxxxxxx -- Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance