pcp
[Top] [All Lists]

Re: [pcp] Valid metric names

To: Martins Innus <minnus@xxxxxxxxxxx>
Subject: Re: [pcp] Valid metric names
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Wed, 18 Mar 2015 00:17:43 -0400 (EDT)
Cc: pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <550848E7.4040800@xxxxxxxxxxx>
References: <550848E7.4040800@xxxxxxxxxxx>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Thread-index: /QWHX2f2aphFL+ubqv+xkA6UBrk+4Q==
Thread-topic: Valid metric names
Hi Martins,

----- Original Message -----
> Hi,
>      We are running into an issue with possibly having to change some
> metric names and want to make sure we proceeed correctly, if we need to
> at all.
> 
>      The perfevent pmda that Joe wrote gets names for the metrics and
> exposes them as libpfm4 has named them.  These can include some
> non-alpha numeric characters.  We've seen things like colons, dashes,
> equal signs, etc.

The one area I know of is pmie(1) - minus and equals are tokens in the
pmie language, so it will split these metrics names into multiple, eg
"foo.bar-mumble.fratz" would indicate subtraction between two different
metrics.

> If not, it would probably help pmda authors to have checks for valid
> names in the appropriate spots.

*nod* - certainly a utility API for verifying validity and maybe an API
for converting a metric name into something 'legal'.  The dynamic names
APIs could certainly acquire calls to check too, yeah.

  If this is done though, we have ~ 20TB
> of archives that these metrics are a member of.  How to handle those
> with newer versions of pcp tools that may check for valid metric names?
> 
> Thanks for any guidance.

pmlogrewrite(1) may be able to help there, if you need to do this.

cheers.

--
Nathan

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