Hello Mark,
Thanks for your prompt response.
If it is any help, this was from PCP 3.6.9 on a production machine
running CentOS 5.5 x86_64.
My laptop, running Ubuntu 12.04, with PCP 3.5.11 works correctly.
Regards
Chandana
On 31/12/12 08:29, Mark Goodwin wrote:
On 12/30/2012 05:41 PM, Chandana De Silva wrote:
pmval -r -i 3008 proc.psinfo.stime
Hi Chandana, it's not working for me either, e.g. :
$ while true; do date >/dev/null; done &
[3] 3266
$ pmval -r -i 3266 proc.psinfo.stime
metric: proc.psinfo.stime
host: localhost
semantics: cumulative counter
units: millisec
samples: all
3266
18446744073696892616
18446744073696875616
18446744073696859616
18446744073696843616
18446744073696828616
18446744073696813616
I'll check the code - the metric item number is supposed to correspond
to the field number (counting from 0) in /proc/PID/stat. Maybe that has
come to grief or something or there is an error or overflow in the
conversion from jiffies to milliseconds.
$ pminfo -m proc.psinfo.stime
proc.psinfo.stime PMID: 3.8.14
----------^^
$ awk '{print $15}' /proc/3266/stat
18952
$ awk '{print $15}' /proc/3266/stat
19015
So the exported values fro mthe kernel look sane, but the values reported
by pmval look suspiciously like they've wrapped :
$ bc -q
obase=16
18446744073696813616
FFFFFFFFFF3DA230
I'll investigate some more ...
Regards
-- Mark
|