pcp
[Top] [All Lists]

Re: [pcp] pcp updates - derived metrics and pmlogger

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: [pcp] pcp updates - derived metrics and pmlogger
From: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Tue, 11 May 2010 09:30:01 +1000
Cc: pcp@xxxxxxxxxxx
In-reply-to: <1107765550.5261273525622330.JavaMail.root@xxxxxxxxxxxxxxxxxx>
References: <1107765550.5261273525622330.JavaMail.root@xxxxxxxxxxxxxxxxxx>
Reply-to: kenj@xxxxxxxxxxxxxxxx
On Tue, 2010-05-11 at 07:07 +1000, Nathan Scott wrote:
> ----- "Ken McDonell" <kenj@xxxxxxxxxxxxxxxx> wrote:
> 
> > This allows pmlogger to work with derived metrics in the
> > configuration
> > file.
> 
> Thanks Ken.
> 
> > There remains an issue for pmchart and the additional tab launched
> > after
> > recording is finished (pminfo inherits the setting of
> > $PCP_DERIVED_CONFIG and the operands for the derived metrics are not
> > in
> 
> Not sure which pminfo invocation you mean there, pretty sure this is
> pmchart itself thats being affected this way...?

My mistake, I mis-remebered this comment from the buzilla comments ...
        pmchart doesn't run pminfo at the end of the recording, rather
        it attempts to open a new tab with archive and view config from
        the recording session.

I'm not sure what the right approach is here.  Having pmchart unset
$PCP_DERIVED_CONFIG in this particular case seems wrong, although even
in the case where the archive contains none of the derived metrics, the
same warning dialogs are produced ... I'm not sure what pmchart is doing
here that involves exploring metrics outside those named in the config
file?

It appears the pmchart may do no more than create a new context, causing
bind_expr() to be called which is where the error messages come from.

I'm considering moving these messages from pmprintf() to debug guarded
by -D derive because an error here can be silently and safely ignored
_unless_ there is a subsequent attempt to access the value or metadata
of the derived metric ... at which time there will be an error
indication, just not as specific as these error messages.  In the
pmchart case we'd just go down the "silently and safely ignored" path.

Thoughts?

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