On 02/24/2011 11:24 AM, Digimer wrote:
> On 02/24/2011 10:44 AM, Richard Allen wrote:
>> Hi all
>>
>> I notice in the Release Notes for RHEL6 that many changes have been made
>> to the Cluster Suite (HA Addon) but I am unable to find any mention of
>> how the new suite does heartbeat.
>> In previous versions the Cluster could only do heartbeats (node
>> intercommunication) on one network link and for redundancy the only
>> option was to use bonded network devices.
>> There was a way to add a second heartbeat using altnode directives in
>> the XML config file but that always felt a bit hackish and was only
>> limited to only one altnode, giving two heartbeat paths.
>>
>> So I would like to ask how RHEL6 does this. If I have nodes with 4 10Gb
>> NIC's, one connected to an admin network, another to a Database network
>> and one to the Application network and the last one connected directly
>> to the other node with a crossover cable, can the cluster now use all
>> possible paths to communicate to the other nodes or will one of those
>> paths become a single point of failure in the cluster?
>>
>> I'm used to using Clusters like HP's ServiceGuard where I can easily
>> define which links to use as heartbeat. It can even use a serial
>> connection (in a two node cluster) as a additional heartbeat and I have
>> always felt this is quite a big limitation in Red Hat's cluster suite up
>> to RHEL6 atleast.
>>
>> Thanks in advance
>> Richard.
> Hi Richard,
>
> Can I assume that you are talking about High Availability in general,
> as opposed to Heartbeat specifically? If not, the rest won't be too
> relevant.
>
> As you know, the 'altnode' parameter is how you assign a second link.
> This is still the case (as is bonding to get more links, but that
> requires common subnets which you don't have).
>
> Corosync is used as the cluster communication layer (as opposed to
> openais from RHEL 5.x). It supports one or two interfaces for "totem"
> communication. If the main fails, the second link will be used
> automatically. However, when the main is restored, totem must be
> manually moved back to the original link.
>
> So in short; as it was in 5, so it is in 6. That said, the 'altname'
> is perfectly valid way of removing that SPF. :)
>
Thanks for the reply.
I was reading up on this and I noticed something new in the cman(5) man
page. Quote:
Multi-home configuration
It is quite common to use multiple ethernet adapters for cluster nodes, so
they will toler-
ate the failure of one link. A common way to do this is to use ethernet
bonding. Alterna-
tively you can get corosync to run in redundant ring mode by specifying an
’altname’ for
the node. This is an alternative name by which the node is known, that
resolves to another
IP address used on the other ethernet adapter(s). You can optionally specify
a different
port and/or multicast address for each altname in use. Up to 9 altnames (10
interfaces in
total) can be used.
Note that if you are using the DLM with cman/corosync then you MUST tell it
to use SCTP as
it’s communications protocol as TCP does not support multihoming.
So I can use up to 9 altnames now? Is true, it would be fantastic :)
--
Rikki. -- RHCE, RHCX, HP-UX Certified Administrator.
-- Solaris 7 Certified Systems and Network Administrator.
Bell Labs Unix -- Reach out and grep someone.
Those who do not understand Unix are condemned to reinvent it, poorly.
--
Linux-cluster mailing list
Linux-cluster@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/linux-cluster
