Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov> writes:
>> The application LISTENs on channel tcn and a trigger function is
>> attached to most permanent tables to NOTIFY for DML on that
>> channel.
>> ...
>> The report to us was that testers were unable to start the
>> application. I believe that the above error on COMMIT kept the
>> application from getting past initial tests that the connection
>> was good.
>
> OK, so it was issuing a LISTEN and the LISTEN was failing at
> commit? (It's a bit hard to believe that you'd be triggering one
> of the NOTIFYs during "initial connection tests".)
In continued digging through logs I found something to indicate the
transaction on which COMMIT failed for 2528 of the failures. In all
cases the transaction made a change which would have fired a NOTIFY
at commit (in a deferred trigger).
So that initial assumption about the cause doesn't hold up. I'm not
sure at this point why the tester perception was that they couldn't
get in.
-Kevin