[pgpool-general: 5723] Re: PGpool Configuration

[Tatsuo Ishii]

Occasionally close connections from jemeter to Pgpool-II.  I am not
sure if you can do it with jmeter though.

And set this:

child_max_connections = 100

will restart Pgpool-II child process every 100 times connections to
Pgpool-II and it will free memory used by Pgpool-II child process.

Best regards,
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp

> Team,
> 
> We have upgraded the latest version. but still the memory leak issue not
> resolved. We are doing the performance testing using jmeter with 4k
> concurrent users for more than a day where the memory is getting filled and
> request has been slow down with errors.
> 
> Any possible solutions?
> 
> 
> On Thu, Sep 7, 2017 at 9:24 AM, Tatsuo Ishii <ishii at sraoss.co.jp> wrote:
> 
>> >>From the above setup we are trying hit the concurrent users of 4k... But
>> at
>> > some point of time PGpool load got increased and memory is fully occupied
>> > even if i increase the Memory full memory got occupied. This resulting
>> > requests slow down and request per second throughput and failures.
>>
>> Sounds like an old problem in Pgpool-II 3.6.2. You should upgrade to
>> the latest version: 3.6.6.
>> >
>> > But without PGpool directly connecting to Master works fine without
>> errors
>> > and much better throughput.
>> >
>> > We would like to scale 20k concurrent users with PGPool setup. We would
>> > like to know whether PGpool is enterprise standard to work on.. Else we
>> > have to look for other solutions.
>>
>> But that requires PostgreSQL to accept 20k concurrent connections or
>> 40k connections (num_init_children * max_pool), which I have never
>> seen in the field. Probably you should test out with PostgreSQL
>> whether it works with 20k concurrent connections first.
>>
>> Best regards,
>> --
>> Tatsuo Ishii
>> SRA OSS, Inc. Japan
>> English: http://www.sraoss.co.jp/index_en.php
>> Japanese:http://www.sraoss.co.jp
>>
> 
> 
> 
> -- 
> 
> Regards,
> Rishi