Adding TCP_USER_TIMEOUT support for libpq/psqlodbc

From: AYahorau(at)ibagroup(dot)eu
To: pgsql-hackers(at)postgresql(dot)org
Cc: MikalaiKeida(at)ibagroup(dot)eu
Subject: Adding TCP_USER_TIMEOUT support for libpq/psqlodbc
Date: 2018-07-20 08:43:31
Message-ID: OF4C8A68CE.A350F319-ON432582D0.0028A5FF-432582D0.002FEE28@iba.by
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hello PostgreSQL Community!

Not long ago I faced the situation concerning ODBC/libpq client hanging in
case of some network problems.
I had a discussion regarding this issue within pgsql-odbc(at)postgresql(dot)org
and got some suggestions.
Here is this discussion:
https://www.postgresql.org/message-id/OF33DF00A3.D6444835-ON432582C3.003EA7C5-432582C3.0045562B%40iba.by

In a few words the suggestion was to use pqopt keepalive options for ODBC
configuration for example as follows:
pqopt = keepalives=1 keepalives_idle=5 keepalives_count=1
keepalives_interval=1

But under some circumstances it can be not reliable and the connection can
loose its keepalives properties and it can remain hanging.

Here is a quote from our discussion:
Hmm it seems keepalive stops while waiting for ack.
Therefore it's a matter of retransmission control
You can use TCP_USER_TIMEOUT on linux but the option is not used in libpq.

In my opinion it make sense to add the support of TCP_USER_TIMEOUT socket
option to libpq/psqlodbc connection.
The attachment contains a patch with the source code changes regarding
this issue.
In my implementation it can be configured by new keepalives_user_timeout
option within pqopt parameter.

Best regards,
Andrei Yahorau

Attachment Content-Type Size
0001_TCP_USER_TIMEOUT_libpq-int.patch application/octet-stream 747 bytes
0001_TCP_USER_TIMEOUT_fe-connect.patch application/octet-stream 1.9 KB

Browse pgsql-hackers by date

  From Date Subject
Next Message Heikki Linnakangas 2018-07-20 08:53:32 Re: [HACKERS] possible self-deadlock window after bad ProcessStartupPacket
Previous Message Heikki Linnakangas 2018-07-20 08:24:23 Re: [HACKERS] possible self-deadlock window after bad ProcessStartupPacket