Glance and Cinder appear to be using different defaults for the rbd_chunk_size while they should not or operations which involve snapshotting/cloning images will fail. This is especially important when the Ceph cluster is at the Pacific level, where such a thing is enforced with the cluster returning error if it isn't, see https://review.opendev.org/c/openstack/cinder/+/804265 While recent versions of Cinder are trying to workaround this by adopting whichever chunk size is in use for the source image, we should configure the services to use by the same value and eventually make it configurable for performance reasons. puppet-glance and puppet-cinder support this already https://github.com/openstack/puppet-glance/blob/master/manifests/backend/multistore/rbd.pp#L32 https://github.com/openstack/puppet-cinder/blob/master/manifests/backend/rbd.pp#L62
There's a discussion in #rhos-storage that suggests this BZ should be an RFE, and coordinated with the cinder squad to address any questions such as whether there's a performance impact on changing the chunk size for all newly created cinder volumes. The change might be suitable for a future release, but may not be suitable for slipping into an OSP-16 maintenance release.