pcp
[Top] [All Lists]

Re: pmcd.openfds Metric (was: [pcp] NSS/NSPR Testing Status)

To: "Frank Ch. Eigler" <fche@xxxxxxxxxx>
Subject: Re: pmcd.openfds Metric (was: [pcp] NSS/NSPR Testing Status)
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Tue, 4 Dec 2012 03:45:25 -0500 (EST)
Cc: Dave Brolley <brolley@xxxxxxxxxx>, pcp@xxxxxxxxxxx, Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
In-reply-to: <20121203212150.GJ27600@xxxxxxxxxx>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>

----- Original Message -----
> > ...
> > So provided you have closure of all the places open() and friends
> > are called (as the current code does) [...]
> 
> Do we really?

Not exactly.  There's critical places in pmcd (client connect, and on
agent start) which are instrumented.  AIUI (*my* recollection of the
original intent) was to detect long-term fd leaks cheaply, which this
simplistic approach does still manage to do (thanks to recycling of
lower fd slots after close).

> Third-party PMDA DSO's are unknown, and so may be some
> of the fd operations used by various libraries our own DSO's link in
> (e.g. the new systemd one! :-).

If they/pmcd leak fds it will show up by watching (logging) this metric
over long periods, because the high water mark gets bumped ever upward
rather than reaching a steady state.

cheers.

--
Nathan

<Prev in Thread] Current Thread [Next in Thread>