OSP10 ships RHSC2 and stands up RHSC2 ceph clusters but it also must support an external Ceph cluster which could be running RHCS1.3. To support this configuration overcloud nodes which are ceph clients need "rbd default features = 1" in the [client] section of their ceph.conf. OSPd doesn't have an option to configure this so this BZ requests that be an option in THT and that puppet-ceph support this option as user configurable.
raising request for pm_ack+ for blocker? bug
After applying the changes from the following two upstream patches:
and adding the following to my Heat environment file:
I was able to get a deploy with the following ceph.conf on my compute node (a ceph client):
[root@overcloud-novacompute-0 ~]# cat /etc/ceph/ceph.conf
osd_pool_default_min_size = 1
auth_service_required = cephx
mon_initial_members = overcloud-controller-0,overcloud-controller-1,overcloud-controller-2
fsid = 63089b2a-930d-11e6-aba5-525400330666
cluster_network = 192.168.1.26/24
auth_supported = cephx
auth_cluster_required = cephx
mon_host = 172.16.1.23,172.16.1.16,172.16.1.24
auth_client_required = cephx
public_network = 172.16.1.18/24
rbd_default_features = 1
Next step is to get the change merged upstream.
Alternatively, as a workaround measure, the overcloud can be deployed with an environment file like the following:
Updating doctext as comment #3 indicates that the fix is supported. The user just needs to use the extra Heat form comment #3.
In OSP10, compatibility with Ceph1.3 is only supported if the Ceph cluster is external. Marking this bug as depends on 1384079 because it prevents OSP10 from working with an external Ceph cluster.
This is not a blocker for OSP10 since the user may simply use an environment file like the following:
Marking this bug as POST so that testing may be done using the above. The proposed upstream fix linked from upstream LP bug 1633552 may be used for OSP11 (clonned this to 1388515 targetted for OSP11 for that purposed).
Should this be documented (in which case, this should be assigned and in the documentation component). If not, please add the TestOnly keyword.
Yes, this should be documented. Apologies, I thought adding doctext and ticking requires_doc_text was sufficient. I have assigned it to the documentation component.
Moving to 'NEW' to be triaged as resources allow.
FYI on an additional verification: QE has verified that using the extra parms from comment #6 is necessary when using OSP10 with an external Red Hat Ceph Server of version 1.3. See this BZ:
Thus, it's important that this be included in the OSP10 w/ Ceph docs.
FYI: upstream documentation on using OSP10 with an extenral Ceph 1.3 server is in review at https://review.openstack.org/#/c/397825/ .
This was vetted by me. Don, not sure if you merged this already. If not, please feel free to do so.