Horton

PQcancel does not use tcp_user_timeout, connect_timeout and keepalive settings

Jelte Fennema-Nio <jelte.fennema@microsoft.com>

View thread View on postgr.es
From: Jelte Fennema <Jelte.Fennema@microsoft.com>
To: "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2021-09-30T14:44:45Z
Lists: pgsql-hackers

Attachments

The new connection made by PQcancel does not use the tcp_user_timeout, connect_timeout or any of the keepalive settings that are provided in the connection string. This means that a call to PQcancel can block for a much longer time than intended if there are network issues. This can be especially impactful, because PQcancel is a blocking function an there is no non blocking version of it. 

I attached a proposed patch to use the tcp_user_timeout from the connection string when connecting to Postgres in PQcancel. This resolves the issue for me, since this will make connecting timeout after a configurable time. So the other options are not strictly needed. It might still be nice for completeness to support them too though. I didn't do this yet, because I first wanted some feedback and also because implementing connect_timeout would require using non blocking TCP to connect and then use select to have a timeout.

Commits

  1. Make PQcancel use the PGconn's tcp_user_timeout and keepalives settings.

  2. Avoid calling gettext() in signal handlers.

  3. Avoid calling strerror[_r] in PQcancel().