Ken,
On 1/28/2015 4:52 PM, Ken McDonell wrote:
But is it valid to assume that, as a separate case, pmcd should continue
to function if a pmda gets "killed" in some other way? OOM killer, some
other error?
Yes. Unless of course the OOM condition is so extreme to get pmcd
killed.
The design point was that pmcd should continue to operate under
extreme circumstances, and pmdas coming and going (for whatever
reason) falls within that mantra.
If you have a counter example, I'd be interested to hear about it.
Yes, this would be my example posted that showed a backtrace with a
problem in AgentsAttributes:
http://oss.sgi.com/archives/pcp/2015-01/msg00149.html
I muddled the issue because it seemed like a pmie problem. But
basically the following occurs for me:
>killall -v pmdaproc
Killed pmdaproc(23682) with signal 15
>pmval pmcd.agent.status
pmval: pmLookupDesc: IPC protocol failure
>pmval hinv.ncpu
pmval: pmLookupDesc: IPC protocol failure
>pminfo hinv.ncpu
Error: hinv.ncpu: Broken pipe
The only thing that brings it back to life is a "pminfo", "pminfo proc",
or if a pmlogger instance is logging a proc metric.
Martins
|