Re: apache parked processes using CPU

[Jan Schaumann <jschauma%netmeister.org@localhost>]

Martin Husemann <martin%duskware.de@localhost> wrote:
> On Sun, Nov 13, 2022 at 10:59:09AM -0500, Jan Schaumann wrote:
> > Hey,
> > 
> > About two days ago, my apache-2.4.54 running on my
> > NetBSD-9.3 amd64 VPS started leaving multiple
> > processes in 'parked' state, utilizing a lot of CPU:
> > 
> >   PID USERNAME PRI NICE   SIZE   RES STATE      TIME WCPU    CPU COMMAND
> > 21048 nobody    42    0   316M   14M parked     5:28 7.32%  7.32% httpd
> > 17677 nobody    42    0   318M   14M parked     5:25 6.69%  6.69% httpd
> > 16398 nobody    41    0   319M   17M parked    18:53 5.03%  5.03% httpd
> >   829 nobody    41    0   320M   18M parked    18:59 4.83%  4.83% httpd
> > 21512 nobody    42    0   322M   19M parked    18:49 4.74%  4.74% httpd
> > 23308 nobody    42    0   322M   19M parked    18:51 1.61%  1.61% httpd
> 
> Why do you think they eat CPU while being parked?

The system got notably slower to respond.

> I would expect them to eat lots of CPU for tiny spots of time (when answering
> requests) and being parked for most of the time (when waiting for more work
> to come in).

I see many connections are left in TIME_WAIT, so
perhaps the system load average goes up because apache
keeps each process parked while waiting for more work
on these connections, but why would that behavior
suddenly show up like that when it never did before?

Trying to tune the time apache waits before it
disconnects an idle connection seems to help, but I'm
again confused what would have caused this to be
necessary all of a sudden.

-Jan