pcp
[Top] [All Lists]

Re: [pcp] linux pmda badness

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: [pcp] linux pmda badness
From: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Wed, 21 Jan 2015 12:21:17 +1100
Cc: PCP <pcp@xxxxxxxxxxx>
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <331211176.14228104.1421801734447.JavaMail.zimbra@xxxxxxxxxx>
References: <54B444A8.8070005@xxxxxxxxxxxxxxxx> <331211176.14228104.1421801734447.JavaMail.zimbra@xxxxxxxxxx>
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.4.0
On 21/01/15 11:55, Nathan Scott wrote:
> ...
> FWLIW, the metric help text is also dynamic for these, and has more detailed
> info (wasn't used for the metric names themselves due to PMNS name rules).

We could map the help text to valid PMNS names pretty simply it would appear, 
except for the wrinkle below.

> I think an extension or analogous interfaces to the pmdaCache API series that
> will allow allocation of item numbers in the range 2^10-1 is needed here - we
> have hit this in a couple of places now.

Without some extra semantics from somewhere, even this is doomed ... on my 
local machine there are duplicates in the help text ...

kenj@bozo:~/src/pcp$ pminfo -T kernel.percpu.interrupts | sed -e '/Help:/d' -e 
'/^$/d' -e '/^kernel\./d'  | sort | uniq -c | awk '$1 != 1 {print}'
      2 IO-APIC-edge pata_atiixp
      7 PCI-MSI-edge xhci_hcd

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