ceph-osd charm does not error when all configured devices are missing

Partial log of charm startup:

2015-02-20 11:11:06 INFO juju-log mon:32: Making dir /var/lib/charm/ceph-osd root:root 555
2015-02-20 11:11:06 INFO mon-relation-changed update-alternatives: using /var/lib/charm/ceph-osd/ceph.conf to provide /etc/ceph/ceph.conf (ceph.conf) in auto mode
2015-02-20 11:11:06 INFO mon-relation-changed creating /var/lib/ceph/bootstrap-osd/ceph.keyring
2015-02-20 11:11:06 INFO mon-relation-changed added entity client.bootstrap-osd auth auth(auid = 18446744073709551615 key=AQAeFOdUSOBDERAA9hbzzoKgdKlAW1ZYJQy1fQ== with 0 caps)
2015-02-20 11:11:06 INFO juju-log mon:32: Path /dev/sdd does not exist - bailing
2015-02-20 11:11:06 INFO juju-log mon:32: Path /dev/sde does not exist - bailing

Regardless of the number of devices/directories configured, if > 0 locations fail due to problems, the charm should report an error.

I think this is a good candidate for unit status output - a unit could display a BLOCKED message with a suitable - 'No usable storage devices found' message.

moved to gerrit
https://review.openstack.org/#/c/292194/

There are two conflicting ideas here:

A) If ANY osd-device fails or is not found, BLOCK. (Comment #1)

vs.

B) If ALL osd-devices fail or are not found, BLOCK. (Comment #2)

Given that the behavior and documentation of osd-devices is that of a whitelist and not an explicit list of required devices, idea B) is most compatible with the established and expected configuration and behavior.

A change for A) was proposed and merged, then reverted. If A) is desired, that is a certain shift in default behavior and it should be discussed thoroughly via cross-team meeting.

This bug is actually resolved, the ceph-osd charm will switch into a blocked state if no devices are detected for the given whitelist.

This does not apply to the ceph charm (however that has been superceeded by ceph-mon, so I'm proposing not to fix ceph for this problem).