Cancelling query send does not cancel query #1607

Activity

1. devadathanmb commented on Jul 4, 2026

   

   devadathanmb

   on Jul 4, 2026

   Last edited by devadathanmb

   Contributor

   More actions

   This seems to be already handled by psycopg, the driver pgcli uses to talk to Postgres.

   When you hit Ctrl-C during a query, psycopg catches it and sends a real cancel request to the server, then drains the resulting "canceling statement due to user request" error off the socket before letting the interrupt reach pgcli (see the source). That's essentially the same way psql also handles Ctrl-C. pgcli doesn't do anything extra on top of this, it just lets that exception bubble up and prints "cancelled query", the actual cancellation happens inside psycopg.

   I tried to reproduce this issue on the latest release and couldn't. The query gets cancelled server-side correctly every time I've tested it.

   There's one case where it could theoretically break: pressing Ctrl-C a second time before the first cancel finishes draining. That second drain is only guarded against a QueryCanceled error, so a second KeyboardInterrupt landing there escapes uncaught, before the pending result is fully read off the socket, and the connection is left desynced (next query fails with another command is already in progress until you reconnect). That window is roughly the cancel request's round trip time, which would be wider against a remote or high-latency database than on localhost.

   Which pgcli version are you on, and are you connecting directly or through something like PgBouncer or an SSH tunnel? Did you press Ctrl-C more than once when this happened?