A very simple delete (by key) on a small table (700 rows) every now and then stays “idle in transaction” for minutes (takes milliseconds usually) even though all the locks are marked as “granted”.
What can I do to pinpoint what causes it? I’m using this select:
SELECT a.datname, c.relname, l.transactionid, l.mode, l.GRANTED, a.usename, a.waiting, a.query, a.query_start, age(now(), a.query_start) AS "age", a.pid FROM pg_stat_activity a JOIN pg_locks l ON l.pid = a.pid JOIN pg_class c ON c.oid = l.relation ORDER BY a.query_start;
which shows a lot of “RowExclusiveLock”s but all are granted… so I don’t see what is causing this spikes of delays.
Advertisement
Answer
This is a problem of the application server.
Sessions are in state “idle in transaction” when the application does not end the transaction with COMMIT
or ROLLBACK
. This is to be considered a bug in the application.
The locks remain (and are of course granted, otherwise the session could not be idle) until the transaction ends.
From PostgreSQL 9.6 on, you can set the parameter idle_in_transaction_session_timeout
to terminate such transactions automatically with a ROLLBACK
, but that is a band-aid to avoid problems on the database rather than a solution.