pcp
[Top] [All Lists]

Re: [pcp] new pmdas and some questions

To: pcp@xxxxxxxxxxx
Subject: Re: [pcp] new pmdas and some questions
From: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Thu, 27 Jun 2013 07:09:40 +1000
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <51CB3E26.5060502@xxxxxxxxxxx>
References: <51CB3E26.5060502@xxxxxxxxxxx>
User-agent: Mozilla/5.0 (X11; Linux i686; rv:17.0) Gecko/20130510 Thunderbird/17.0.6
On 27/06/13 05:16, Martins Innus wrote:
Hello,

     We are in the process of writing a few pmdas and assuming all goes
well would make them publicly available. ...

That would be both welcome and encouraged.

> I have a few questions on the
best way of going about this.  We have written and are currently testing
pmdas for the Panasas file system ( https://www.panasas.com/ ) and for
NVML ( https://developer.nvidia.com/nvidia-management-library-nvml ).
Can we reserve PMID's for these so we don't conflict with anything else?

Best to get the PMDA domain number assigned (which gives you a range of PMIDs and instance domain IDs to be managed as you see fit within your
PMDA).

I have tentatively assigned
PANASAS         118
NVML            119
in my tree ... pending your confirmation these will flow up stream to the official PCP tree in due course.

     When complete, would we host these ourselves for download or is
there interest in incorporating them into pcp?  I ask because for
instance we use the Infiniband PMDA which is not distributed by default
with PCP anymore.  Is the expectation that niche pmdas like this would
live outside the main tree?  They both require 3rd party APIs in order
to build/test/run.

This is a per-PMDA decision.

In the tree means you can leverage the existing packaging framework and QA infrastructure, and you're more likely to receive code review feedback. We can accommodate conditional builds for PMDAs based on the presence of 3rd party artifacts where such a dependency exists.

But some PMDAs are not of general interest and remain outside the tree.

Independent of that decision, I would strongly encourage you to solicit feedback from the list at the point where you map the available metrics on to the PCP namespace (PMNS) and decide on the metric and instance domain choices ... this is historically the place where most PMDA implementation mistakes are made. After this the PMDA APIs have been very stable for a long time and there is plenty of example code to use as a template.

Good luck.

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