Hi,
On Wed, May 20, 2009 at 06:21:47PM +0100, John Leedham wrote:
> Hi,
>
> I'm looking at deploying a configuration in which we have two ethernet
> interfaces per node, one 10Gig and one 1Gig on different networks (lets
> say 10.0.0.* and 10.0.1.*).
>
> This cluster will be an active/active pair of file servers, and in the
> case where one machines 10Gig interface goes down, we wish the sharead
> disk resources to failover, unless both 10Gig interfaces are down, in
> which case we'd like the resources to continue running somewhere over
> 1Gig.
>
> So a given machine might have
>
> 10Gig - 10.0.0.1, gw 10.0.0.254
> 1Gig - 10.0.1.1, gw 10.0.1.254
>
> The difficulty appears to be that if we lose a 10Gig link (due to a NIC
> or cable failure), the machine can still ping the 10Gig gateway (since the
> network routing in the company means that the gateways on those networks
> can route to each other), and therefore will not notice the link is down.
>
> Any thoughts as to the best way to address this and get the desired
> functionality? Whilst there is information about monitoring two different
> networks, I've had a look at the pingd options and there doesn't appear to
> be anything to control hop counts or similar. I'd thought of using an
> agent to actively monitor the particular 10Gig interface (i.e. check ethX
> is up or down), but I haven't found a suitable agent.
>
> NB. In the particular scenario we are in, we aren't using floating IP
> addresses, so the IP's of the interfaces don't need to move (the file
> system client knows to try and access the resource via another IP in the
> case of one of the interface failures, and this is preferable to moving
> the IP over as well as the disks)
I don't think there's a ready made solution for this.
You can create your own OCF RA, which would just monitor the
interfaces and set some attribute (crm_attribute or "crm node
status-attr"), then set location constraints depending on the
value of that attribute. Start with the Dummy OCF RA and just
expand the monitor action.
For example, the attribute could have values from the set
{0,1,10,11}. You could assign scores depending for the node
depending on that attribute's value (as with pingd).
Thanks,
Dejan
> Many thanks,
>
> John
>
> _______________________________________________
> Linux-HA mailing list
> [email protected]
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
