[Pgpool-general] num_init_children and max_pool

[Tatsuo Ishii]

> On Mon, May 9, 2011 at 7:07 PM, Tatsuo Ishii <ishii at sraoss.co.jp> wrote:
> 
>> >> You mean connection to backend? Yes, anytime child re-open the
>> >> connection automatically if neccessary (and possible). It's a
>> >> transparent processing for clients.
>> >>
>> >>
>> > In that case, and still with the objective of minimizing the memory
>> > footprint,  does this seem like a good compromise:
>> >
>> > child_life_time = 1800
>> > connection_life_time = 300
>> >
>> > If the server is idle, the child will exit / respawn after 30 min, but
>> after
>> > 5 min it will drop the connection to the backend, meaning that an idle
>> child
>> > would only be connected to the backend 1/6th of the time. Is that
>> correct?
>>
>> Yes, correct.
>>
>> > Do you see any potential issues with such a setup?
>>
>> If a client keeps on connecting to pgpool-II, the above settings will
>> not work since they are only activated after the client disconnects to
>> pgpool-II. If you want to force idle clients be disconnected, you
>> could turn on this:
>>
>> # If client_idle_limit is n (n > 0), the client is forced to be
>> # disconnected whenever after n seconds idle (even inside an explicit
>> # transactions!)
>> # 0 means no disconnect.
>> client_idle_limit = 0
>>
>>
> What is considered idle?

No packet has been sent from the client last client_idle_limit
seconds. This does not include the case while the client sent a query
and wait for the result. For example, waiting for a lock is acquired.

> I guess the "even inside an explicit transaction"
> scares me, as "inside a transaction" doesn't seem "idle".

Why? A sloppy client always can just send "BEGIN" then do
nothing. Idle is idle, right?
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese: http://www.sraoss.co.jp