[Pacemaker] Debian packages, OCFS2, high CPU load

Stefan Förster cite+pacemaker at incertum.net
Mon Nov 30 18:51:19 UTC 2009


* Stefan Förster <cite+pacemaker at incertum.net>:
> * Dejan Muhamedagic <dejanmm at fastmail.fm>:
>> On Fri, Nov 27, 2009 at 01:05:41PM +0100, Stefan Förster wrote:
>>> With Debian, apart from some minor glitches (path to controld.pcmk,
>>> old udev, old kernel) everything went well, but as soon as I commit
>>> the configuration containing the O2CB resources, both nodes become
>>> unresponsive, cluster communication fails and corosync (which was
>>> started as "aisexec") is at about 100% CPU.
>> 
>> corosync runs as corosync. aisexec is from the older openais
>> (0.8x).
> 
> With the Debian packages from http://people.debian.org/~madkiss/ha/,
> openais contains "/usr/sbin/aisexec", which is a shellscript calling:
> 
> export COROSYNC_DEFAULT_CONFIG_IFACE="openaisserviceenableexperimental:corosync_parser"
> corosync "$@"
> 
> The Debian openais package also contains /usr/lib/lcrso/service_ckpt.lcrso
> which isn't loaded without the above environemnt settings. Amongst
> others, it contains:
> 
> /usr/lib/lcrso/service_msg.lcrso
> /usr/lib/lcrso/service_lck.lcrso
> /usr/lib/lcrso/service_clm.lcrso
> /usr/lib/lcrso/service_evt.lcrso
> /usr/lib/lcrso/openaisserviceenable.lcrso
> /usr/lib/lcrso/service_ckpt.lcrso
> /usr/lib/lcrso/service_amf.lcrso
> /usr/lib/lcrso/service_tmr.lcrso
> 
>> Otherwise, perhaps you found a bug. See if it's reproducible
>> without o2cb.
> 
> I'm unsure on how to do this. Perhaps simply using another service
> which relies on CKPT would trigger that bug?

I could reproduce the problem: The behaivour arises as soon as
Pacemaker stops DLM for the first time - it seems it's not related to
o2cb at all. As soon as the DLM resource is stopped, the CPU usage of
corosync is at 100%.

Anything else I can do to aid in debugging this?


Cheers
Stefan




More information about the Pacemaker mailing list