| *** markvoelker has joined #openstack-tc | 01:54 | |
| *** markvoelker has quit IRC | 01:59 | |
| *** markvoelker has joined #openstack-tc | 02:29 | |
| *** markvoelker has quit IRC | 02:34 | |
| *** markvoelker has joined #openstack-tc | 04:30 | |
| *** evrardjp has quit IRC | 04:33 | |
| *** evrardjp has joined #openstack-tc | 04:33 | |
| *** markvoelker has quit IRC | 04:35 | |
| *** markvoelker has joined #openstack-tc | 06:05 | |
| *** markvoelker has quit IRC | 06:09 | |
| *** markvoelker has joined #openstack-tc | 06:19 | |
| *** markvoelker has quit IRC | 06:29 | |
| *** markvoelker has joined #openstack-tc | 06:30 | |
| *** markvoelker has quit IRC | 06:34 | |
| *** tosky has joined #openstack-tc | 08:29 | |
| *** markvoelker has joined #openstack-tc | 08:31 | |
| *** markvoelker has quit IRC | 08:35 | |
| *** slaweq has joined #openstack-tc | 09:36 | |
| *** slaweq has quit IRC | 09:44 | |
| *** markvoelker has joined #openstack-tc | 10:05 | |
| *** markvoelker has quit IRC | 10:09 | |
| *** slaweq has joined #openstack-tc | 10:12 | |
| tbarron | clarkb: zaneb yeah the rook case with cinder is interesting. When running rook with cloud provider VMs rather than bare metal it normally can't | 10:48 |
|---|---|---|
| tbarron | access real disks, so being able to get at raw block devices is advantageous since it can use them as if they are disk drives | 10:49 |
| tbarron | e.g. use volusmes from Amazon Elastic Block Service as backing store for Ceph OSDs rather than spinning or solid state drives | 10:50 |
| tbarron | k8s PVs traditionally weren't suited for this because PVs backed by block devices used "file mode" (and still do by default) | 10:52 |
| tbarron | when presented with a block device k8s would inspect it for a filesystem and if none was present make a node-local filesystem (ext4 or xfs) | 10:53 |
| tbarron | the node-local aspect impled no RWS | 10:53 |
| tbarron | However now there is raw block volume support https://kubernetes.io/docs/concepts/storage/persistent-volumes/#raw-block-volume-support | 10:54 |
| tbarron | which is not useful for most of the regular applications running on k8s (which want a filesystem bind mounted into pods) but which i good for | 10:55 |
| tbarron | booting VMs in pods (kubevirt) *and* for emulating disk drives with raw block devices, as when a k8s cluster wants to build and mqanage its own software defined file system li9ke Ceph | 10:56 |
| tbarron | In the k8s on OpenStack case this could be a way for each K8s cluster to run their own Ceph without using the Ceph cluster (if there is one) that backs OpenStack | 10:58 |
| tbarron | One interesting quesion in this scenario is the nature of the underlying Cinder block storage. | 10:59 |
| tbarron | E.g. if Cinder is backed by Ceph RBD you end up with a Ceph-on-Ceph data path. | 10:59 |
| tbarron | Assuming default Ceph RADOS replication at both layers (2 replicas) that means each write mushrooms to require 9 synchronous round trips. | 11:00 |
| tbarron | So to do the kind of thing Rook-Ceph does on EBS it might be better to have Cinder backed by iscsi or some such -- still indirection in the data path, but less of it. | 11:02 |
| tbarron | s/RWS/RWX/ above | 11:02 |
| tbarron | clarkb: zaneb: so yeah, if you want to run a cluster-specific software-defined file system in a k8s cluster running on openstack cinder-csi with | 11:04 |
| tbarron | (non-default) raw block mode is a better fit than manila-csi | 11:04 |
| tbarron | But I'm not sure this is a starter-kit use case? | 11:05 |
| *** markvoelker has joined #openstack-tc | 12:05 | |
| *** markvoelker has quit IRC | 12:10 | |
| tbarron | mnaser: thanks for https://groups.google.com/a/kubernetes.io/g/steering/c/VGjECFmk-cQ !! | 12:35 |
| fungi | wow, they don't even allow you to read that without logging in | 13:54 |
| *** markvoelker has joined #openstack-tc | 14:06 | |
| *** markvoelker has quit IRC | 14:11 | |
| *** markvoelker has joined #openstack-tc | 14:54 | |
| *** markvoelker has quit IRC | 14:59 | |
| tbarron | :( | 15:30 |
| *** markvoelker has joined #openstack-tc | 16:55 | |
| *** markvoelker has quit IRC | 16:59 | |
| *** markvoelker has joined #openstack-tc | 18:56 | |
| *** markvoelker has quit IRC | 19:01 | |
| *** markvoelker has joined #openstack-tc | 21:55 | |
| *** markvoelker has quit IRC | 22:00 | |
| *** tosky has quit IRC | 22:43 | |
| *** slaweq has quit IRC | 22:44 | |
| *** tkajinam has joined #openstack-tc | 22:58 | |
| *** slaweq has joined #openstack-tc | 23:00 | |
| *** slaweq has quit IRC | 23:05 | |
| *** slaweq has joined #openstack-tc | 23:10 | |
| *** slaweq has quit IRC | 23:14 | |
| *** tetsuro has joined #openstack-tc | 23:57 | |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!