On Tue, 21 Nov 2000, Michal Kara wrote:
> > > My initial reaction (given my old fart Unix status) is that I'd
> > > recommend using cpp to generate your pmlogger config files, rather than
> > > adding this functionality to pmlogger (this functionality certainly is
> > > not part of pmlogger at the moment).
>
> Or m4... It is not a bad idea.
I personally find m4 to be a pain, but others don't of course.
>
> > In the simple case, simply cat all the configs you want together
> > and pipe them straight into pmlogger (it reads it's config on stdin
> > if -c is not given). But there are some cpp features that may be
> > useful, e.g. conditional constructs, macros, etc.
> >
> > So you could use a simple script like this:
> > ...
>
> I think it would be more complicated if you want to run more pmloggers,
> you'd
> have to do something like:
>
> file=$1
> shift
> cpp $file | pmlogger $*
>
> I will probably write my solution using makefile, it seems better for me.
>From what I can gather, you should use a Makefile to drive cpp in the
/var/pcp/config/pmlogger directory.
>
> BTW, I have two more features I miss in pmlogger:
>
> Ability to simply restart pmlogger for given machine. Maybe it is already
> there(?). Currently I have to go to pmlc, connect to loggers and determine
> which
> connects which machine. But it is just a suggestion.
I think all of this pmlogger management stuff you need is already
available. Check out the man page for pmlogger_check(1). This is an
extensible cron based pcp archive log management infrastructure driven
by the control file in /var/pcp/config/pmlogger/control. Suitable/example
cron entries are in /var/pcp/config/pmlogger/crontab.
>
> Second feature is to be able to tell pmlogger to behave more like cron -
> when
> you have metric A gathered every 30 seconds and metric B every 60 seconds.
> Currently, it can happen that the metric A is gathered on seconds 10 and 40
> and
> metric B on second 15 which results in more requests and sometimes the pmda
> must
> do its work twice. The reason I need this is that pmcd on the machines I
> monitor
> currently uses 5-10% of the CPU and I think this feature would lower the load.
>
That is far too much CPU time for such a low fetch rate. Can you send me your
pmlogger configs and/or gprof /var/pcp/pmdas/linux/pmdalinux (this is the
daemon form of the linux PMDA dso, so you can profile it. You'll need to
edit /var/pcp/config/pmcd/pmcd.conf to change from DSO to daemon).
thanks
-- Mark
|