Bug #4031
closedceph-disk-activate hardcodes journal path, ignores configuration
0%
Description
I'm having my ceph.conf configured to store journals in a different place, like:
[osd]
osd journal = /var/lib/ceph/journal/$cluster-$id
(/var/lib/ceph/journal is on /, which is mounted on an RAID0 SSD array)
However, ceph-disk-activate has this hardcoded in mkfs():
'--osd-data', path,
'--osd-journal', os.path.join(path, 'journal'),
Looking up values via either ceph-osd --show-config-value or ceph-conf --lookup should be trivial and should be used instead.
Updated by Sage Weil about 11 years ago
- Status changed from New to Resolved
commit:813e9fe2b4291a1c1922ef78f031daa9b78fe53b
Updated by Sage Weil about 11 years ago
- Status changed from Resolved to Won't Fix
Or not.. I had to revert this.
The problem is that ceph-disk-activate mounts the osd in a temporary location and needs to specify the paths to make things work. If it mounted it in the correct location things would be a bit racy, but possibly okay... in the meantime, though, ceph-disk-* dictate those locations for keyring and journal.
Of course, this doesn't prevent you from putting the journals at /var/lib/ceph/journal/*... you just need to pass that path to the ceph-disk-prepare command and it will update the symlink in the osd data dir accordingly. Basically, don't adjust the osd data or osd journal config options, but do this via ceph-disk-prepare.
Updated by Faidon Liambotis about 11 years ago
When I experienced this (probably a few days before reporting it) there was no support for passing the journal path, it was either a separate block device or a partition on the OSD device. I see this is implemented now, so this is perfectly workaroundable now.
That's not in bobtail though, sadly. I guess it's too invasive to backport?
Updated by Sage Weil about 11 years ago
We may backport it later, but first we'd like to get ceph-deploy into a "done" state.