hubert depesz lubaczewski <depesz@gmail.com> writes: > my pg_stat_activity is raporting process working on a query for 12 hours > straight! (query_start). > 2 problems: > 1. this particular query runs typically between 80 and 200 miliseconds! > 2. procpid of this backend does not exists anymore in system It's possible that the "backend exited" message got dropped by the stats mechanism --- the stats code is deliberately designed to lose messages rather than delay the rest of the system, so it's not 100% reliable.
is it possible to "by hand" remove bad entries from this stats?