pcp
[Top] [All Lists]

Re: [pcp] Compressing .meta file also?

To: Mark Goodwin <mgoodwin@xxxxxxxxxx>
Subject: Re: [pcp] Compressing .meta file also?
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Thu, 12 May 2016 21:20:23 -0400 (EDT)
Cc: pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <82ff1613-795a-e04d-00fa-7a10e26677a5@xxxxxxxxxx>
References: <1478926031.47431716.1463099651741.JavaMail.zimbra@xxxxxxxxxx> <82ff1613-795a-e04d-00fa-7a10e26677a5@xxxxxxxxxx>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Thread-index: MO1Yur3LDcrjmrONpCCQXPK5w4HY3A==
Thread-topic: Compressing .meta file also?

----- Original Message -----
> On 05/13/2016 10:34 AM, Nathan Scott wrote:
> > Hi Ken,
> >
> > I've seen two cases this week of people starting to get sufficiently
> > large archive .meta files that compression has been requested, or at
> > least considered (both times from the well known proc-indom issue).
> 
> The proc-indom issue is in large part due to the external instance names
> for proc metrics - these include the pid and command *and* all arguments.
> e.g. to see what I mean: run pminfo -f proc.psinfo.stime | grep chrome
> 
> I propose we shorten this to just the pid and cmd name, without all the args
> - apps can also fetch proc.psinfo.psargs  if they need the entire command
> line.

Yep, that's a good point - I like that approach.

> Not sure how we'd accommodate downrev/remote clients, but it'd certainly
> help reduce the metadata size in archives containing proc metrics.

Not sure we can (nor have to?) accommodate that - I guess we could have a
back-compat option to pmdaproc to generate the old-style instance names?

cheers.

--
Nathan

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