We are seeing some clouds where physically, compute/storage nodes are racked unevenly across racks, and while the compute service may be okay with the imbalance of hypervisors in each availability zone, ceph performance can be severely impacted by an imbalance of spindles/nodes.
The status message of ceph-osd notes how many OSDs are available from each unit:
i.e.
message: Unit is ready (14 OSD)
message: Unit is ready (8 OSD)
message: Unit is ready (14 OSD)
message: Unit is ready (8 OSD)
message: Unit is ready (14 OSD)
message: Unit is ready (2 OSD)
juju-lint should aggregate number of OSDS per zone and report the number of OSDS per zone and WARN if they are not identical, and ERROR if they are off by 20% or more.
We are seeing some clouds where physically, compute/storage nodes are racked unevenly across racks, and while the compute service may be okay with the imbalance of hypervisors in each availability zone, ceph performance can be severely impacted by an imbalance of spindles/nodes.
The status message of ceph-osd notes how many OSDs are available from each unit:
i.e.
juju-lint should aggregate number of OSDS per zone and report the number of OSDS per zone and WARN if they are not identical, and ERROR if they are off by 20% or more.