Function Purpose: Description of problem: Support for deploying manila with CephFS-with-NFS (via ganesha gateway) was added in OSP13 but this was only for deployments where Director installs the Ceph daemons and the Ganesha daemon. While this meets the needs of some of our customers, some would also like to be able to deploy manila such that it references an externally deployed Ceph cluster. There are several possible ways that this need might be met. For example there is ongoing work to deploy the ceph daemons and ganesha via rook and kubernetes, but we don't have a concrete timeline for that work and it would not be something we could backport. Alternatively, we may be able to modify the current TripleO heat templates and ceph-ansible playbooks so that if an external cluster is available we can reference it instead of installing the daemons ourselves. There are two variations of this last approach -- one where only the ceph daemons are external and we still deploy ganesha, and one where ganesha is also external. An important consideration for all these possibilities is that ganesha is in the data path for share service and cannot today run active-active. That is why when we introduced support for Cephfs-via-NFS in OSP13 we ran ganesha on controller nodes as part of the pacemaker cluster there, and that need drove the choice to lead with support only for Director-integrated deployment of ganesha and ceph daemons. So this work may split into three phases: 1) see if we can keep pacemaker control of ganesha for service availability but allow the ceph daemons themselves to be externally deployed. 2) for deployments that can manage ganesha availability themselves, allow ceph daemons and ganesha to be externally deployed. This scenario would likely always involve a Support Exception so that Red Hat is not held accountable for failure in the data path. 3) longer term, work with Storage BU on the rook based deployment of external ceph daemons and ganesha service, where even if Director triggers the deployment of the ceph-ganesha infrastructure at the same time that it deploys the overcloud, it is technically external to OpenStack itself and where the HA for ganesha service is no longer maintained by pacemaker cluster in OpenStack.
verified