| To: | Ken McDonell <kenj@xxxxxxxxxxxxxxxx>, pcp@xxxxxxxxxxx |
|---|---|
| Subject: | Re: [pcp] pcp updates: fix kernel.pernode.cpu metrics on single CPU systems |
| From: | Mark Goodwin <mgoodwin@xxxxxxxxxx> |
| Date: | Mon, 05 Jan 2015 15:21:47 +1100 |
| Delivered-to: | pcp@xxxxxxxxxxx |
| In-reply-to: | <54A5E591.3010309@xxxxxxxxxx> |
| References: | <54A4EE5F.2000108@xxxxxxxxxx> <54A5AD0C.9010105@xxxxxxxxxxxxxxxx> <54A5E591.3010309@xxxxxxxxxx> |
| User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.3.0 |
On 01/02/2015 11:25 AM, Mark Goodwin wrote: [...] Any thoughts on how we might test cpu hotplug/unplug? Do we need to even support it? I suspect that if the per-cpu and per-node indoms need to be dynamic then this isn't the only codxe that will need some attention ... just following up here after a bit more investigation - this code needs a cleanup and re-work. Firstly, the proc_stat and proc_cpuinfo per-cpu arrays shoudl be merged (and similarly for the various per-node metrics), and secondly, cpu hotplug/unplug is supported in qemu-kvm-0.12.1.2-2.250.el6 and later, see https://bugzilla.redhat.com/show_bug.cgi?id=562886 and probably also in some h/w architectures. I'll post a BZ for this work since this is a fairly intrusive change to the most commonly fetched metrics in the linux PMDA, and making the cpu and node indoms dynamic most likely has ramifications elsewhere.. Regard -- Mark |
| <Prev in Thread] | Current Thread | [Next in Thread> |
|---|---|---|
| ||
| Previous by Date: | [Bug 1100] libpcp interp.c crash on archive scan, bugzilla-daemon |
|---|---|
| Next by Date: | Re: [pcp] Fetching proc data from a remote host, Nathan Scott |
| Previous by Thread: | Re: [pcp] pcp updates: fix kernel.pernode.cpu metrics on single CPU systems, Mark Goodwin |
| Next by Thread: | [Bug 1100] New: libpcp interp.c crash on archive scan, bugzilla-daemon |
| Indexes: | [Date] [Thread] [Top] [All Lists] |