pcp
[Top] [All Lists]

Re: [pcp] pcp updates - nfs4 request metrics

To: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Subject: Re: [pcp] pcp updates - nfs4 request metrics
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Tue, 7 Oct 2014 02:44:51 -0400 (EDT)
Cc: pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <54337DAE.6000107@xxxxxxxxxxxxxxxx>
References: <54337DAE.6000107@xxxxxxxxxxxxxxxx>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Thread-index: wgdO+r9soEgiRPvkblaWT8phMGq7SA==
Thread-topic: pcp updates - nfs4 request metrics
Hi Ken,

----- Original Message -----
> qa/232 was failing on Centos 7.
> 
> There are NR_RPC4_SVR_COUNTERS instances, numbered 1 (not 0) to
> NR_RPC4_SVR_COUNTERS, but the values are stored in [0] ...
> [NR_RPC4_SVR_COUNTERS-1] of proc_net_rpc->server.reqcounts4[].
> 
> It would be _really_ good to have someone who knows the nfs code in the linux
> pmda to review these changes, and if they are correct to check that similar
> mismatches are not happening for the other nfs/rpc metrics.

I recently spent several hours pouring over the NFS kernel code to
get a handle on these metrics (adding the nfs4.1 ops support), as
well as the userspace code in nfsstat(1).  Is the test failing with
latest dev?  (its not failing here - can you send through the .bad
file you saw?  thanks!)  Was this commit in place for this QA run:

commit b690dcb8049963a6c7516cc5c89312dd9095840a
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date:   Fri Sep 26 10:44:01 2014 +1000

    Update qa/232 to handle recent nfs4.1 indom changes


The odd non-zero indexing is a quirk of the NFS v4 code, and does
not affect the earlier NFS variants.  I was fairly confident that
the pmdalinux code was correct when I left it, so I'm keen to get
more of a handle on things from that .bad file.

cheers.

--
Nathan

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