The cephfs driver has support for multiple cephfs fileystems on cluster.
But the capacity information for each cephfs fileystems is directly
related to the amount of storage space available on the cluster. Thus
the capacity for each cephfs filesystem is the same. When you configure
multiple backends against the same cluster, the scheduler will treat
both backends equally, which will result in one backend getting all of
the share creation requests and the other backend getting none.
To fix this have the driver report the allocated_capacity_gb along with
the provisioned_capacity_gb, which is the sum of the shares assigned to
that backend. Then if you enable thin_provisioning and thin_logic the
capacity weigher will use the least allocated cephfs backend as a higher
candidate for share creates. Which will result in share creations being
split between the cephfs backends.
This information is now being cached, and a new configuration option
named ``cephfs_cached_allocated_capacity_update_interval`` was added to
the driver, so that OpenStack operators can define how long they would.
like for this information to be persisted. It defaults to 60 seconds.
Co-Authored-By: Carlos da Silva <ces.eduardo98@gmail.com>
Closes-Bug: #2049538
Change-Id: I141a5db9cf66327f68f1fa4c7f2bb72135171e43
d3d86bde80