[pgpool-hackers: 3170] Re: New feature proposal: On demand scale up and scale down of child processes
m.usama at gmail.com
Thu Dec 6 22:43:51 JST 2018
On Thu, Dec 6, 2018 at 5:40 AM Tatsuo Ishii <ishii at sraoss.co.jp> wrote:
> > Hi
> > Currently Pgpool-II spawns the num_init_children number of child
> > at the time of startup. This is the standard approach we follow in
> > Pgpool-II since its inception. But this approach is not very optimal for
> > the setups with variable load requirements. For example for the setups
> > where we get instantaneous load spikes let say only for small amount of
> > time, we still have to set the num_init_children in Pgpool-II to the
> > maximum value, and most of the time the majority of the the child
> > would be sitting idle and consuming the system resources.
> > Like consider a case of some sports website which gets huge number of
> > when some sport event is happening, but for rest of the time it gets very
> > few visitors. But with the current available options when designing the
> > system using the Pgpool-II we have to set the num_init_children to the
> > value what we expect to get at the maximum peak time and that number of
> > children will be spawned by Pgpool-II at startup, which is a waste of
> > resources.
> > So my proposal is that instead of spawning all the num_init_children
> > of child process at startup, we scale up and scale down the number of
> > processes as per the demand. And only keep may be 10 percent of extra
> > child processes at any time.
> > I believe this approach will improve the load management and performance
> > Pgpool-II
> > What are your thought and suggestions on this.
> Do you have a performance measurement on this?
I am running some tests to gauge the actual performance impact we get
because of extra idle child process,
I will share the results on this thread.
> Best regards,
> Tatsuo Ishii
> SRA OSS, Inc. Japan
> English: http://www.sraoss.co.jp/index_en.php
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the pgpool-hackers