[
https://issues.apache.org/jira/browse/CLOUDSTACK-8629?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14625875#comment-14625875
]
Wido den Hollander commented on CLOUDSTACK-8629:
------------------------------------------------
We should probably refactor the Heartbeat stuff a bit. It's now very
NFS-centric. We should come up with a more general layout which we can
implement for each storage pool type.
Eg, canHeartBeat(), if that's true the pool is capable of writing heartbeats
and we let it do that.
This way we can implement it for NFS and RBD, but also for any other future
storage pool.
In case of RBD we have the choice of creating a RBD device for each Host and
write the heartbeat to that, OR go lower and write directly to RADOS. The last
is the best I think, but the objects it write will be less visible to admins
though. But that shouldn't really matter since the objects will be just a
couple of bytes in size.
> Use Ceph RBD storage pool for writing HA heartbeats
> ---------------------------------------------------
>
> Key: CLOUDSTACK-8629
> URL: https://issues.apache.org/jira/browse/CLOUDSTACK-8629
> Project: CloudStack
> Issue Type: Improvement
> Security Level: Public(Anyone can view this level - this is the
> default.)
> Components: KVM
> Reporter: Wido den Hollander
> Assignee: Wido den Hollander
> Fix For: Future
>
>
> Just like NFS we should write a heartbeat for each Instance to RADOS.
> Each hosts could write a simple object like: <pool>/<host ip>
> They simply write the timestamp to the object encoded in JSON.
> Other hosts can read that object and see if the host wrote the timestamp
> lately. If it did it means that it is still up and running and Fencing is not
> needed or required.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
