Обсуждение: Autovacuum vacuums not setting start time in pg_stat_activity
This is on a 8.2.3 database; vacuums started by autovacuum don't get their query_start time set. select * from pg_stat_activity where current_query like 'VACUUM%'; -[ RECORD 1 ]-+--------------------------------------------- datid | 16384 datname | emma2 procpid | 8674 usesysid | 10 usename | postgres current_query | VACUUM public.userdata_2409_messages_history waiting | f query_start | backend_start | 2007-06-12 15:29:53.376641-05 client_addr | client_port | -- Jim Nasby jim.nasby@enterprisedb.com EnterpriseDB http://enterprisedb.com 512.569.9461 (cell)
"Jim Nasby" <jim.nasby@enterprisedb.com> writes: > This is on a 8.2.3 database; vacuums started by autovacuum don't get > their query_start time set. Hmm. Query_start is actually defined as "time of last client command receipt", so it's null just like client_addr/port are. Not sure if there's anything we should put there that's more meaningful than the backend start time. regards, tom lane
Tom Lane wrote: > "Jim Nasby" <jim.nasby@enterprisedb.com> writes: > > This is on a 8.2.3 database; vacuums started by autovacuum don't get > > their query_start time set. > > Hmm. Query_start is actually defined as "time of last client command > receipt", so it's null just like client_addr/port are. Not sure if > there's anything we should put there that's more meaningful than the > backend start time. Let's put the current command (vacuum/analyze) start time in there. Backend start time, I assume, is the time at which the current autovac iteration was started. Should this be backpatched if changed in 8.3? -- Alvaro Herrera Developer, http://www.PostgreSQL.org/ "El hombre nunca sabe de lo que es capaz hasta que lo intenta" (C. Dickens)
Alvaro Herrera <alvherre@commandprompt.com> writes: > Should this be backpatched if changed in 8.3? I'd vote no; it's pretty hard to argue that it's a bug fix as opposed to a cosmetic improvement. regards, tom lane