I'm running a Postgres database with a total disk occupation of 100Gb, largest and most used table up to 40Gb (about 30.000.000 tuples).
Overall performance degrades sometimes due to some queries that are not run by the final user app. I guess they are run by Postgres itself. They use to take up to 100% of CPU and delay user queries substantially.
From 'pg_stat_activity', you can see this pattern in "current_query" column for these queries:
SELECT * FROM "public"."tablename" ORDER BY "column1", "column2"... LIMIT 1000 OFFSET 144000
Sometimes with 'SET DATESTYLE = "ISO"'; before the SELECT.
These are always SELECTs on random tables without conditions (WHERE) and with 'ORDER BY' clause, what makes them -I guess- very heavy.
I use to cancel these queries with "pg_cancel_backend" to recover database functionality.
So, what are these queries indeed? Is it advisable to cancel them? Is there any way to prevent these situation to happen?
I use PostgreSQL 8.4.4 on x86_64-redhat-linux-gnu, compiled by GCC gcc (GCC) 4.1.2 20080704 (Red Hat 4.1.2-46), 64-bit.
