First time poster here - probably a good sign since I've been running postgresql with zero issues for the last several years! At any rate, I've recently run into a strange issue. Client-based EOFs are nothing new to our application; people can sometimes close a connection for a number of reasons. However, on the DB side these have always been released with no issue.
Today, we had 4 connections that saw either a client-based EOF or a 'canceling statement due to user request', and in all of these 4 cases the query remained active and started chewing heavily into the cpu such that they ground the db server to a halt until each of the procpids were manually issued a 'pg_cancel_backed ( procpid );' from the console. None of the 4 queries were particularly heavy; they registered approximately a 2300ms completion time using a query EXPLAIN ANALYZE. A little long, but not too far out of the ordinary on our usual database usage.
Has anyone run into something similar, or have any ideas on how I can track down what might have kept these queries in the pg_stat_activity table even after the procpids were gone on the originating device?
TIA,
-Chris.
