Re: Request: pg_cancel_backend variant that handles 'idle in transaction' sessions

On Wed, Nov 4, 2015 at 11:15 AM, Pavel Stehule <pavel(dot)stehule(at)gmail(dot)com> wrote:
>
>
> 2015-11-04 18:11 GMT+01:00 Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>:
>>
>> Merlin Moncure <mmoncure(at)gmail(dot)com> writes:
>> >> Yes, and that is what I meant. I have two problems with
>> >> transaction_idle_timeout (as opposed to transaction_timeout):
>> >>
>> >> A) It's more complex. Unsophisticated administrators may not
>> >> understand or set it properly
>> >>
>> >> B) There is no way to enforce an upper bound on transaction time with
>> >> that setting. A pathological application could keep a transaction
>> >> open forever without running into any timeouts -- that's a dealbreaker
>> >> for me.
>> >>
>> >> From my point of view the purpose of the setting should be to protect
>> >> you from any single actor from doing things that damage the database.
>> >> 'idle in transaction' happens to be one obvious way, but upper bound
>> >> on transaction time protects you in general way.
>>
>> > Note, having both settings would work too.
>>
>> I'd vote for just transaction_timeout. The way our timeout manager
>> logic works, that should be more efficient, as the timeout would only
>> have to be established once at transaction start, not every time the
>> main command loop iterates.
>
>
> I cannot to say, so transaction_timeout is not useful, but it cannot be
> effective solution for some mentioned issues. With larger data you cannot to
> set transaction_timeout less than few hours.

sure. note however any process can manually opt in to a longer timeout.

merlin