Hi Neil,
Very cool. Glad that i asked about the "lesser DomUs being
offlined" thingie because, as you can see, i was taking it as
a completely different beast then what you had in mind.
Vote approval.
PS: For others in the community who might have similar questions such as
mine, is there a way to deposit a more detailed description of the
project somewhere on the community site? I didn't see a URL for this
proposed project anywhere in this e-mail thread, which leads me to
believe that this project exists only in this thread so far.
Thanks,
-ashu
Neil Garthwaite wrote:
> Hi Ashu,
>
> Concerning migration it would be triggered by a "clrg switch" command,
> although migration or live migration would be resource specific. By this I
> mean that when a "DomU resource" is registered with OHAC, the switchover
> method will be determined, i.e. migration, live migration or just failover
> (stop/relocate/start). However whatever method is chosen, the appropriate
> method would be triggered by a "clrg switch" and implemented by the HA-Xen
> agent.
>
> There are some restrictions for migration, in that the DomU's virtual block
> devices (VBD) need to be accessible by both nodes that are participating in
> such a migration. One could think of "accessible" in terms of a global file
> system. If a Zpool/ZFS was being used by a DomU's VBD list, then that
> Zpool/ZFS would only be available to both nodes participating in a migration
> and not accessible. In this regard, it's my belief that migration or live
> migration would not be possible for a DomU using a Zpool/ZFS VBD. However, my
> belief my change once we are engaged with the Xen community.
>
> Concerning placing a DomU offline. For those familiar with Solaris Cluster,
> this is akin to RG_affinities and in particular a strong negative affinity.
> For those not familiar with this terminology please see
> http://blogs.sun.com/SC/entry/how_does_sun_cluster_decide for an informative
> article about affinities.
>
> With the above in mind, "Allow for important DomUs where lesser DomUs are
> "offlined" means that using HA-Xen with OHAC and assuming a 2-node cluster,
> under normal operations both nodes could be running several domUs. However,
> some DomUs maybe more important than others and in the event of a node
> failure it maybe desirable to favor an important DomU over another DomU.
>
> Utilizing RG_affinities it would be possible to offline a less important DomU
> in favor or an important DomU to free up system resources so that upon
> failover the important DomU maintains a specific service level.
>
> In summary, this is an important feature and one that already exists within
> Solaris Cluster and therefore OHAC. However, in my opinion perhaps the
> greatest strength of RG_affinities is the knowledge that one could utilize
> all nodes thereby not having any idle nodes with the comfort of knowing that
> in the event of a node failure, OHAC + HA-Xen will favor important DomUs over
> lesser DomUs when using RG_affinities.
>
> Many thanks for your support with the remainder.
>
> Regards
> Neil
> --
>
> This message posted from opensolaris.org
>
> _______________________________________________
> ha-clusters-discuss mailing list
> ha-clusters-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/ha-clusters-discuss
