Timofei, thanks for bringing up the resource providers work; it is
absolutely trying to solve the problem highlighted in this post.
Antonio, comments inline.
On 08/08/2016 05:22 AM, Antonio Messina wrote:
2016-08-08 10:52 GMT+02:00 Timofei Durakov <tdura...@mirantis.com>:
Hi,
so for this moment we have 2 options:
- somehow check every time that nodes are on shared storage(now it's done
over ssh)
- define that shared storage explicitly()
The possible workaround for the first option is to use RPC for checking that
temporarily created test file exists(more RPC calls).
This sounds like a sane idea as well.
For the second option(I think the most preferable) there is a bp [1] that
introduce abstraction that allows set up such shared storages.
Please note that "shared storage" is a relation between two compute
nodes: "do we share the same storage?". In a complex enough cloud you
might have nodes with shared storage and "standalone" nodes, so you
cannot use a global config option.
The linked blueprint seems to go in the right direction (although I
could only make sense of it after reading the related spec page:
http://specs.openstack.org/openstack/nova-specs/specs/newton/approved/generic-resource-pools.html)
Please do let us know if you have questions regarding the proposed
implementation in the spec above. We are actively working on this code
and our goal is to have the resource tracker in Nova writing inventory
and allocation records by the Newton release, with integrations of the
Nova scheduler and conductor with the placement API coming in Ocata.
Best,
-jay
Alternatively we could use an host aggregate property, like
"shared_storage=<label>". All nodes with the same label also share the
same storage. But I personally like the resource-pool idea better.
my 2 cents.
.a.
_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
