pcp
[Top] [All Lists]

Re: [pcp] automatic derived metrics slowing down remote pcp clients, esp

To: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Subject: Re: [pcp] automatic derived metrics slowing down remote pcp clients, esp. pmlogconf
From: "Frank Ch. Eigler" <fche@xxxxxxxxxx>
Date: Sun, 15 May 2016 07:05:53 -0400
Cc: pcp developers <pcp@xxxxxxxxxxx>, mgoodwin@xxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <57382F72.1080401@xxxxxxxxxxxxxxxx>
References: <20160514193945.GC1418@xxxxxxxxxx> <5737DC21.6030509@xxxxxxxxxxxxxxxx> <20160515024019.GF1418@xxxxxxxxxx> <57382F72.1080401@xxxxxxxxxxxxxxxx>
User-agent: Mutt/1.4.2.2i
Hi, Ken -

On Sun, May 15, 2016 at 06:12:34PM +1000, Ken McDonell wrote:
> >>and (worse) (b) generate _exactly_ the same pmlogger configuration
> >>file as the last time pmlogconf was run (for this host).
> >
> >Heh, but how would we know, unless we run it? :-)
> 
> Why not simple rate limit? e.g. run pmlogconf for host X if (a) 
> configfile does not exist, or (b) pmlogconf not run in the last XYZ for 
> host X ... where XYZ is "10mins" or "hour" or "day" or "week" or ... 
> some config option

Yeah, considered that, but that can only be a heuristic.  If a
sysadmin had good reason to believe that those files will have a fixed
lifetime of correctness, then she could probably eschew pmlogconf
entirely and just supply a hand-made conf file rotated at that
lifetime.  If on the other hand the network contains unpredictable
changes, then any particular lifetime heuristic would risk being
out-of-date.


> >Maybe save pminfo metric-lists from previous pmlogconf runs, and
> >assume that if the metric-list now is the same, reuse the previous
> >pmlogger.config.  So basically use the PMNS as a cache validation key.
> >Though we'd also need to hash in the pmlogconf input files, should
> >they change.  pmlogconf could implement such a cache internally.
> 
> Seems like a bit of a sledge hammer ... collecting, saving and comparing 
> the "metric-lists" may be more work than you're trying to save

Could it be just one PMNS_NAMES PDU (pminfo -h HOST)?  Hm, maybe not,
if pmprobe's value-fetch-attempt fails.  OK.

> if it is _all_ available metrics from host X then PMDAs with a
> dynamic PMNS may trip this up.

You mean if the list varied from pmlogconf cache-hit pminfo time to
pmlogconf cache-miss pmprobe*85 time?  Yes, that could be a problem.
Though the problem would be similar if the list varied between
pmlogconf time and pmlogger time.


- FChE

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