pcp
[Top] [All Lists]

Re: [pcp] pernode metrics still wrong on 1 CPU systems

To: Nathan Scott <nathans@xxxxxxxxxx>, Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Subject: Re: [pcp] pernode metrics still wrong on 1 CPU systems
From: Mark Goodwin <goodwinos@xxxxxxxxx>
Date: Thu, 22 Jan 2015 10:05:55 +1100
Cc: PCP <pcp@xxxxxxxxxxx>
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <27495938.14845022.1421880746472.JavaMail.zimbra@xxxxxxxxxx>
References: <54C00816.10808@xxxxxxxxxxxxxxxx> <54C02474.9000407@xxxxxxxxx> <27495938.14845022.1421880746472.JavaMail.zimbra@xxxxxxxxxx>
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.3.0
On 01/22/2015 09:52 AM, Nathan Scott wrote:


----- Original Message -----
[...]
Mark, what additional info would help diagnose/fix this?

what is vm04 (kernel non-smp? uname -a, which distro)?

ls -l /sys/devices/system/node
ls -l /sys/devices/system/node/node0/cpu*

As previously noted, I'm merging and partly re-writing proc_cpuinfo.[ch]
and proc_stat.[ch], including the pernode metrics. I wont get time to work
on this at all until Saturday. A shorter term fix might be possible with
the above requested info ..

I have one machine here reporting no NUMA nodes, if that's the trigger?

yes it is. And nnode == 0 may well be behind the valgrind issues that Ken
reported, since it'll be trying to allocate an array of size 0 (and then
assuming the array is at least one element on single CPU systems, hence
overrunning the array).


$ pminfo -f hinv.nnode

hinv.nnode
     value 0

$ ls /sys/devices/system/node
ls: cannot access /sys/devices/system/node: No such file or directory

$ uname -a
Linux verge 3.2.0-4-rt-686-pae #1 SMP PREEMPT RT Debian 3.2.41-2 i686 GNU/Linux

We'll need a fix before Saturday...

So that's a 32bit debian VM then. I don't have one handy, are there
any accessible out there, anyone? I need to poke around in /sys to
see what's going on. Either debian doesn't have the same /sys stuff
as fedora and other kernels, or it doesn't "fake" a single numa node
on single cpu systems.

Regards
-- Mark

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