----- Original Message -----
> Hi -
>
> > PMDAs can export many metrics, some may be expensive others not.
> > [...] It exports many other metrics, and enabling every single
> > expensive metric that might make sense for it to export, simply
> > because pminfo connected and asked for a value (of an unrelated
> > metric) would not be sensible.
>
> I don't think anyone was suggesting an all-or-none sort of collection
> behavior. I only said on-demand, referring to the metrics actually
> requested by a client (suffering PM_ERR_VALUE or PM_ERR_AGAIN or
> PM_ERR_PMDANOTREADY or whatever for that first pmFetch).
Its common for someone to run something like "pminfo -f network" to get
a high level look, then browse through the result, maybe iterating on
further with pminfo to something more specific, or switching to pmchart
or another tool for a different view.
pminfo and pmprobe are the classic short-lived-clients cases that should
show why this would be a problem in practice (the PMDAs do not know what
the nature of the connecting clients are, and nor should they).
cheers.
--
Nathan
|