pcp
[Top] [All Lists]

Re: [pcp] /prod/<pid>/stat additions

To: Martins Innus <minnus@xxxxxxxxxxx>
Subject: Re: [pcp] /prod/<pid>/stat additions
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Mon, 22 Sep 2014 23:59:28 -0400 (EDT)
Cc: pcp developers <pcp@xxxxxxxxxxx>
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <54205ED8.2090301@xxxxxxxxxxx>
References: <54205ED8.2090301@xxxxxxxxxxx>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Thread-index: yvoIWaieAkGcxuzJG+ewRec1LtgwdA==
Thread-topic: /prod/<pid>/stat additions

----- Original Message -----
> Hi,
>      I filled in missing fields from stat for the linux_proc_pmda:
> 

Looks good Martins.  I've merged & tested - there was a handful of
QA test failures, but those are now resolved (please review commit
942ad80b if you have time).

> For most of the existing metrics from the stat file, there is 1-1
> correspondence between the item id and its position in the buffer that
> is read from proc.  But 3 existing metrics( ttyname, wchan_symbol,
> psargs) use ids that do not correspond to items in the buffer.  So for
> the metrics i added, I just started after these and coded in the
> appropriate offset in the fetch.

This is fine.

> I assume messing with existing ids would break lots of existing archives,
> etc?

That's correct.

> Also, I only needed delayacct_blkio_ticks for my purposes, but did the
> others just for completeness.  Slightly off topic, but am I

Great, thanks.  Since the PMDA does the conversion from ticks to msecs,
I tweaked the metric name to reflect that ... other than that, its all
looking good to me.

> understanding correctly that delayacct_blkio_ticks would be a measure of
> per process IO wait?

I've not looked into the kernel code behind that metric before, perhaps
someone else can provide some insight here?

cheers.

--
Nathan

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