Re: deadlock in single-row select-for-update + update scenario? How could it happen?

On 08/22/2014 10:15 AM, hubert depesz lubaczewski wrote:
> On Fri, Aug 22, 2014 at 6:45 PM, Adrian Klaver
> <adrian.klaver@aklaver.com <mailto:adrian.klaver@aklaver.com>> wrote:
>
>     So process 66017 and 66014 are blocking each because they are
>     running the exact same queries. The interesting part is the process
>     with the lower pid is starting later then the none with the higher pid.
>
>
> Locking is obvious. But why deadlock? There is just single row, and it
> shouldn't be able to deadlock on it?!

Well both queries are doing SELECT .. FOR UPDATE as well as UPDATE. From
what I see there are four queries contending for the same row.

>
>     So what exactly is 'importer' and what does it do?
>
>
> Some software written by some guy. Runs lots of queries, but the only
> problem we have is with these transactions.
>
>     Also what is this (59303)?
>
>
> log_line_prefix is  '%m %r %p %u %d ' so it's port number.

So why are different processes running the exact same queries coming in
on different ports?


>
> depesz


--
Adrian Klaver
adrian.klaver@aklaver.com