Description of problem:

The /usr/bin/ceilometer-agent-compute program has a hard requirement on the python module
python-ceilometerclient:

sudo runuser -s /bin/bash ceilometer -c '/usr/bin/ceilometer-agent-compute --logfile /var/log/ceilometer/compute.log '
Traceback (most recent call last):
  File "/usr/bin/ceilometer-agent-compute", line 6, in <module>
    from ceilometer.cli import agent_compute
  File "/usr/lib/python2.6/site-packages/ceilometer/cli.py", line 33, in <module>
    from ceilometer.alarm import service as alarm_service
  File "/usr/lib/python2.6/site-packages/ceilometer/alarm/service.py", line 23, in <module>
    from ceilometerclient import client as ceiloclient
ImportError: No module named ceilometerclient

This fails to start via the typical service command, but the command is *recorded* as
successful, thus cluttering up /var/lock/subsys & /var/run/ceilometer/*.pid:

$ sudo service openstack-ceilometer-compute start ; echo $?
Starting openstack-ceilometer-compute:                     [  OK  ]
0
$ sudo service openstack-ceilometer-compute status ; echo $?
openstack-ceilometer-compute dead but pid file exists
1

That's because on line 34 of /etc/init.d/openstack-ceilometer-compute:

34    daemon --user ceilometer --pidfile $pidfile "$exec --logfile $logfile &>/dev/null & echo \$! > $pidfile"

The command *always* returns 0, since the final command, echo $! returns 0. So the 
retval check at line 37 "succeeds".

Version-Release number of selected component (if applicable):

openstack-ceilometer-compute-2014.1-1.el6.noarch

How reproducible:

Always, IF python-cielometerclient is not installed otherwise (puppet-openstack modules may do this).

Steps to Reproduce:
1. rpm -e openstack-ceilometer-compute python-ceilometerclient
2. yum install -y openstack-ceilometer-compute
3. service openstack-ceilometer-compute start
4. service openstack-ceilometer-compute status

Actual results:

$ service openstack-ceilometer-compute status ; echo $?
openstack-ceilometer-compute dead but pid file exists
1

Expected results:

openstack-ceilometer (pid  12345) is running...

Additional info:

It looks like *all* of the /etc/init.d/openstack-* files use the construct:

daemon --user SERVICE--pidfile $pidfile "$exec &>/dev/null & echo \$! > $pidfile"

which is always going to return 0 from the final echo. Why is the pidfile being
created twice? Does the --pidfile option not work?

Proposed solution is to add a:

Requires: python-ceilometerclient

to %package compute

or, arguably, to python-ceilometer, which is the owner of:

$ rpm -qf /usr/lib/python2.6/site-packages/ceilometer/alarm/service.py
python-ceilometer-2014.1-1.el6.noarch