Hi, Martin -
> >>pmlognow -s -p job-1000-begin- -d /var/log/pcp/pmlogger/HOSTNAME -c
> >>/etc/pcp/pmlogger/pmlogger-config.ubccr
> >>[...]
> >(Assuming -h local: ?)
>
> For now, yes, that was all we needed. But other functionality could be
> added later on.
OK, just pointing out the inconsistency with other pcp client tools.
> >Have you tried something like:
> >
> >% echo 'log mandatory on once {foo.bar foo.baz}' | pmlc $PMLOGGER_PORT #
> >or -P
>
> [...] Since we wanted to have just one config file for our
> logging, it would have required some scripting anyway to start up
> another logger, parse our logger config, convert all the directives to
> "log once" and then feed that to pmlc, it seemed like I might as well
> just use the api directly. [...]
So you're starting from a preexisting pmlogger configuration file, but
one that doesn't say "log once ..." on all of its metrics? Perhaps
then consider instead a change to pmlogger proper that overrides all
the sampling intervals, and runs just one loop. That way you preserve
the full pmlogger language (which you can't parse with line-scanning
heuristics properly anyway), and all the other parametrization
pmlogger already has.
> [...] I think we would have a use case for doing a "log now" of a
> set of metrics defined by an alias for diagnosis of different
> classes of application that might be running.
(A list of metric name aliases would make the pmlc approach even
simpler.)
- FChE
|