[Linux-HA] Xen EVMS-HA mini-howto and locking mechanism remark

[Tijl Van den Broeck]

Robert & Jo,

Thank you very much for this information, it clarifies a lot.
To create a failsafe environment, in which virtual machines cannot
ever be double booted.
Either a shared cluster filesystem (such as OCFS2) is required as
backend (in which OCFS2 provides locking) or a Xen-vm resource for HA
with a shared raw I/O backend device, in which heartbeat fully
controls and prevents the duplicates of a Xen-vm. This is in anyhow
the great pointer I was looking for.

I'll try to keep the lists and Xensource wiki updated of any progress made.

greetings

Tijl Van den Broeck



On 1/4/07, Robert Wipfel <rawipfel at novell.com> wrote:
> Tijl,
>
> Jo De Baer has been adding some information about shared
> cluster containers, for OCFS2 and Xen, at this Wiki
> http://wiki.novell.com/index.php/SUSE_Linux_Enterprise_Server#High_Availability_Storage_Foundation
> ==
> First preview release - November 15 2007 : download here
> This preview release shows how to integrate a EVMS2 shared cluster container
> ==
>
> In response to your question about Xen using EVMS cluster containers...
> EVMS itself doesn't do any locking that would protect a container on one node
> from concurrent access by a different node. Cluster containers work by
> masking access - the EVMS devnodes should only be visible on one node at
> a time - as a function of the failover manager managing EVMS to mask /
> unmask the devnodes across nodes (for private containers). Shared cluster
> containers are visible on all nodes, with the expectation highler layer
> software (e.g. a cluster file system) is sharing and coordinating access
> to that shared storage.
>
> Whether or not the same Xen VM is active on more than one cluster
> node at a time, is also provided by Heartbeat's management of Xen VMs
> as cluster resources - it's Heartbeat's job to ensure the same resource
> (VM) isn't running in two places at the same time.
>
> (Xen knows whether the same block device is being accessed by more
> than one VM on the same server - and protects against that).
>
> Hth,
> Robert
>
>