pcp
[Top] [All Lists]

Re: [pcp] pmlc access control, was Re: PCP Updates: qa fallout from ipv6

To: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Subject: Re: [pcp] pmlc access control, was Re: PCP Updates: qa fallout from ipv6/unix sockets for pmlogger and pmlc
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Thu, 20 Feb 2014 16:50:35 -0500 (EST)
Cc: pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <53067159.9050409@xxxxxxxxxxxxxxxx>
References: <52FE5058.4030702@xxxxxxxxxx> <757832688.10280462.1392753861578.JavaMail.zimbra@xxxxxxxxxx> <896174788.10421447.1392770006295.JavaMail.zimbra@xxxxxxxxxx> <5304D039.9010708@xxxxxxxxxx> <1347098955.12246278.1392874951684.JavaMail.zimbra@xxxxxxxxxx> <530612EC.8020206@xxxxxxxxxx> <y0meh2xmtb9.fsf_-_@xxxxxxxx> <53067159.9050409@xxxxxxxxxxxxxxxx>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Thread-index: 5LWBXSvUkBHCP14cbFDwo6RBpKoBzw==
Thread-topic: pmlc access control, was Re: PCP Updates: qa fallout from ipv6/unix sockets for pmlogger and pmlc

----- Original Message -----
> The status quo just seems wrong ... anything that can change the state
> of pmlogger should not be in the "enquire" class.  Shall I fix this?

Yes please!  :)

> Locking down pmlc is unlikely to have any negative fallout.
> 
> I believe pmlc use is very rare, probably because I have not managed to
> convince anyone else that the "value add" proposition for pmlc in the
> following scenario is real:

The other scenario, where I've seen pmlc used (as in: really used, not
a hypothetical scenario) is to query/check "is pmlogger really logging
the things its meant to be" - so extending to the "is pmlogger running
at all" checks.

> * pmlogger configured with shallow metric coverage and long sample intervals
> * pmie looking for badness
> <badness happens>
> * pmie rule fires to ...
>       use pmlc to increase metric depth and/or reduce sample interval
>       wait a while
>       use pmlc to return pmlogger to default configuration

*nod* - have never seen this used in the wild FWIW.  But, high volume
event metric data could make a stronger case for this model, which is
potentially much more of a firehose than regular sampled data - so an
on/off logging switch is more compelling.

> Even for this scenario, one could use a second pmlogger instance and
> later merge the two archives with pmlogextract, avoiding pmlc altogether.

*nod* - its all a bit manual though, whereas the above is more neatly
automated (eg we could enable that sort of thing in a default install
if some common situation came up, whereas this latter model would be
possibly a little more awkward to integrate).

cheers.

--
Nathan

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