Re: Notice lock waits

On Wed, Sep 28, 2016 at 11:57 PM, Haribabu Kommi <kommi(dot)haribabu(at)gmail(dot)com>
wrote:

>
>
> On Sat, Aug 6, 2016 at 3:00 AM, Jeff Janes <jeff(dot)janes(at)gmail(dot)com> wrote:
>
>> One time too many, I ran some minor change using psql on a production
>> server and was wondering why it was taking so much longer than it did
>> on the test server. Only to discover, after messing around with
>> opening new windows and running queries against pg_stat_activity and
>> pg_locks and so on, that it was waiting for a lock.
>>
>> So I created a new guc, notice_lock_waits, which acts like
>> log_lock_waits but sends the message as NOTICE so it will show up on
>> interactive connections like psql.
>>
>> I turn it on in my .psqlrc, as it doesn't make much sense for me to
>> turn it on in non-interactive sessions.
>>
>> A general facility for promoting selected LOG messages to NOTICE would
>> be nice, but I don't know how to design or implement that. This is
>> much easier, and I find it quite useful.
>>
>> I have it PGC_SUSET because it does send some tiny amount of
>> information about the blocking process (the PID) to the blocked
>> process. That is probably too paranoid, because the PID can be seen
>> by anyone in the pg_locks table anyway.
>>
>> Do you think this is useful and generally implemented in the correct
>> way? If so, I'll try to write some sgml documentation for it.
>>
>
>
> Providing the details of lock wait to the client is good. I fell this
> message
> is useful for the cases where User/administrator is trying to perform some
> SQL operations.
>
> I also feel that, adding a GUC variable for these logs to show it to user
> may not be good. Changing the existing GUC may be better.
>

I don't think it would be a good idea to refactor the existing GUC
(log_lock_waits) to accomplish this.

There would have to be four states, log only, notice only, both log and
notice, and neither. But non-superusers can't be allowed to change the
log flag, only the notice flag. It is probably possible to implement that,
but it seems complicated both to implement, and to explain/document. I
think that adding another GUC is better than greatly complicating an
existing one.

What do you think of Jim Nasby's idea of making a settable level, rather
just on or off?

Thanks,

Jeff