Open Stack

This question is focused on Tripleo overcloud nodes meant to handle block storage or object storage rather than the regular control and compute nodes.

Basically I want to get peoples thoughts on how much manipulation of the underlying storage devices we should be expecting to do if we want standalone overcloud nodes to provide block and object storage via their local disks, i.e. not via NFS/gluster/ceph etc

Consider an overcloud node which will be used for providing object storage (swift) from its local disks:
IIUC swift really just cares that for each disk you want to use for storage:
a) it has a partition
b) that the partition has a filesystem on it
c) that the partition is mounted under /srv/node

Given tripleo is taking ownership of installing the operating system on these nodes, how much responsibility should tripleo take for getting the above steps done? In the case that this machine just came in off the truck, all of those steps would need to be done prior to the system being a usable part of the overcloud.
If we don't want tripleo dealing with this stuff right now, e.g. its eventually going to be done by ironic e.g. https://blueprints.launchpad.net/ironic/+spec/utility-ramdisk, then what is the best process today? Is it that someone does a bunch of work on these machines before we start the tripleo deployment process? Presumably we would at least need to be able to feed Heat a list of partitions which we then mount under /srv/node and update fstab accordingly so the changes stick?

[Right now we skip all of a)-c) above and just have swift using a folder, /srv/node/d1 (doesn't this want to be under /mnt/state?), to store all its content.]


Now consider an overcloud node which will be used for providing block storage (cinder) from its local disks:
IIUC the cinder LVM driver is the preferred option when accessing local storage. In this case cinder really just cares that for each disk you want to use for storage it is added to a specific volume group. [Assuming we're not going to allow people to create disk partitions and then select particular ones]. We would then presumably need to include the appropriate filter options in lvm.conf so the selected devices get correctly scanned by lvm at startup?

[Right now we do all this for a dummy loopback device which gets created under /mnt/state/var/lib/cinder/ and whose size you can set via the Heat template: https://github.com/openstack/tripleo-image-elements/blob/master/elements/cinder-volume/os-refresh-config/post-configure.d/72-cinder-resize-volume-groups ]

Thanks
Charles