As I currently only have three nodes, I'm running in the non-recommended
configuration where I've got both VMs and ceph running on the same
hosts; I added the stanza below to the upstart jobs for ceph-mon,
ceph-mds and ceph-osd, and it much improved the contention between ceph
and the VMs. I'm not entirely certain that I'm doing things correctly as
far as cgroups are concerned, but it does at least appear to achieve the
desired result.
post-start script
set -e
#CGROUP=ceph
[ -f /etc/default/ceph ] && . /etc/default/ceph
if [ -n "$CGROUP" ] ; then
pid=$(status $UPSTART_JOB id=$id cluster=${cluster:-ceph} |
head -n1 | egrep -oi '([0-9]+)$')
for subsystem in $(mount -t cgroup | awk '{ print $3 }' | uniq)
; do
if [ ! -d $subsystem/$CGROUP ] ; then
mkdir -p $subsystem/$CGROUP || true
fi
if [ -w $subsystem/$CGROUP/tasks ] ; then
echo $pid > $subsystem/$CGROUP/tasks || true
fi
done
fi
end script
--
Martin Rudat
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
