Ceph

I created a cluster with ceph-deploy on a set of Debian 7 VMs for testing. Each VM has a physical device/partition at /dev/vda1 formatted with XFS and mounted to /data.

From the admin node, I ran "ceph-deploy osd prepare osd[1-4]:/data" and "ceph-deploy osd activate osd[1-4]:/data". No errors were reported.

- ceph status showed 0 up, 0 in and ceph-osd was not running on any of the nodes
- running /etc/init.d/ceph start osd on the osd node would not produce any output
- symlinks were properly created to /data in /var/lib/ceph/osd

I found the problem in ceph_common.sh, which uses find -type d to look for local configurations. Adding -L causes find to return the type on the other end of a symlink allowing the symlink to /data to be picked up with find.

--- ceph_common.sh 2013-07-02 05:53:39.628550081 0000
++ /opt/ceph/lib/ceph/ceph_common.sh 2013-07-02 02:03:46.012964001 0000
@ -136,7 +136,7 @
get_local_daemon_list() {
type=$1
if [ d "/var/lib/ceph/$type" ]; then
for i in `find L /var/lib/ceph/$type -mindepth 1 -maxdepth 1 -type d -printf '%f\n'`; do
for i in `find /var/lib/ceph/$type -mindepth 1 -maxdepth 1 -type d -printf '%f\n'`; do
if [ -e "/var/lib/ceph/$type/$i/sysvinit" ]; then
id=`echo $i | sed 's/[^]*-//'`
local="$local $type.$id"