Hi Mark,
----- Original Message -----
> Changes committed to git://oss.sgi.com/markgw/pcp/pcp.git dev
>
> This fixes a bug, repro'd by: service pcp restart; pminfo -f hinv.cpu
> (I'll write qa for it tomorrow).
I can't seem to reproduce this - the above works for me every time.
Your fix doesn't look like its in quite the right place too - might
be better to add that special case be in with the other cases inside
linux_fetch() before the refresh call... ?
[even if I just start pmcd by hand with -f (no other client fetches
possible that might perturb the result), its still all good. Odd.]
> Nathan, kernel.percpu.interrupts
> seems to have it's own cpu instance domain management - should it
> be shared with proc_stat (and proc_cpuinfo)?
Hmmm, not sure what you mean there - the interrupts metrics do use
CPU_INDOM currently (force the CLUSTER_STAT refresh in linux_fetch
too). Can you elaborate on what you meant there?
cheers.
ps: I'll be in the office tomorrow if you've been trying to get me on
IRC, sorry - school holidays, child-minding duty today.
--
Nathan
|