[Openais] Redundant ring not recovering after issuing the command corosync-cfgtool -r

Darren Thompson darrent at akurit.com.au
Tue Apr 13 15:36:49 PDT 2010


Question.
Is it better to use card bonding and a single ring or unbonded  
interfaces and dual rings?

Sent from my iPhone

On 14/04/2010, at 4:04 AM, Steven Dake <sdake at redhat.com> wrote:

> On Tue, 2010-04-13 at 19:31 +0100, Tom Pride wrote:
>> Just to clarify, when I ifdown eth1 corosync does detect a failure  
>> and
>> it does mark the ring as faulty.  Are you saying that when I use ifup
>> corosync can't work out that the interface is back up and
>> communications can resume when I run corosync-cfgtool -r ?  Would I
>> therefore get a different result if I introduced the failure by
>> physically unplugging the cat5 from the server and then physically
>> reconnecting the cat5?  What about if I shut down the port on the
>> switch it is connected to?
>>
>
> Yes this is correct.  You should see proper operation if the network
> link is lost normally (ie the nic fails, the link fails, the switch  
> port
> fails, the switch fails).
>
> When an interface is ifdowned, it sends a special event to corosync,
> which corosync captures and causes special behavior to occur (the
> binding to 127.0.0.1).  Pulling a network cable doesn't cause this  
> same
> event to occur.  This rebind behavior is incompatible with redundant
> ring.
>
> Regards
> -steve
>
>> On Tue, Apr 13, 2010 at 6:33 PM, Steven Dake <sdake at redhat.com>  
>> wrote:
>>        On Tue, 2010-04-13 at 17:04 +0100, Tom Pride wrote:
>>> Hi Steve,
>>>
>>> Thanks for the suggestion but that didn't work.  I'm not
>>        sure if you
>>> read my entire post or not, but the two redundant rings that
>>        I have
>>> configured, both work without a problem until I introduce a
>>        fault by
>>> shutting down eth1 on one of the nodes.  This then causes
>>        the cluster
>>> to mark ringid 0 as FAULTY.  When I then reactivate eth1 and
>>        both
>>> nodes can once again ping each other over the network, I
>>        then run


More information about the Openais mailing list