You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@qpid.apache.org by Jason Stelzer <ja...@gmail.com> on 2010/10/19 16:56:25 UTC
Options for clustering/multicast
I am setting up a 2 node cluster using the C++ qpidd 0.5 on fedora 12.
This is primarily for failover.
I'm in a situation where one of our hardware vendors is not handling
multicast correctly.
As an interim solution, is it possible to cluster via unicast? My
understanding is that the corosync library uses multicast and that
qpidd relies on that working correctly. This seems to imply that
unicast is not possible, but I figured I'd ask and see if I get a
definitive answer one way or the other.
--
J.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:users-subscribe@qpid.apache.org
Re: Options for clustering/multicast
Posted by Carl Trieloff <cc...@redhat.com>.
There is an cman option today for that. I expect that in due course this
option
will be added for pacemarker also.
In the current cluster with the cman option specified in config, the
broker actively
checks quorum.
Carl.
On 10/19/2010 01:04 PM, Clark O'Brien wrote:
> Could you use an external cluster manager like Pacemaker (http://www.clusterlabs.org/#2) to detect a failed node and move traffic to other nodes. Pacemaker could also start a recovery instance. Pacemaker is available from EPEL-compiled versions of Linux.
>
> I would really like to hear opinions on this strategy especially from guys with networking expertise. We are building messaging in the cloud and this is a strong contender.
>
>
>
>
>
>
>
>
> --- On Tue, 10/19/10, Jason Stelzer<ja...@gmail.com> wrote:
>
>
>> From: Jason Stelzer<ja...@gmail.com>
>> Subject: Re: Options for clustering/multicast
>> To: "Alan Conway"<ac...@redhat.com>
>> Cc: users@qpid.apache.org
>> Date: Tuesday, October 19, 2010, 10:57 AM
>> What if I setup a gre tunnel between
>> the 2 hosts with the mtu set low
>> enough that it doesn't fragment the packets? At that point
>> I could
>> configure corosync to multicast to the virtual interface
>> and work
>> around the dodgy switch.
>>
>> On Tue, Oct 19, 2010 at 11:19 AM, Alan Conway<ac...@redhat.com>
>> wrote:
>>
>>> On 10/19/2010 10:56 AM, Jason Stelzer wrote:
>>>
>>>> I am setting up a 2 node cluster using the C++
>>>>
>> qpidd 0.5 on fedora 12.
>>
>>>> This is primarily for failover.
>>>>
>>>> I'm in a situation where one of our hardware
>>>>
>> vendors is not handling
>>
>>>> multicast correctly.
>>>>
>>>> As an interim solution, is it possible to cluster
>>>>
>> via unicast? My
>>
>>>> understanding is that the corosync library uses
>>>>
>> multicast and that
>>
>>>> qpidd relies on that working correctly. This seems
>>>>
>> to imply that
>>
>>>> unicast is not possible, but I figured I'd ask and
>>>>
>> see if I get a
>>
>>>> definitive answer one way or the other.
>>>>
>>>>
>>> No, there's not currently a way to run clustering over
>>>
>> unicast.
>>
>>>
>>
>>
>> --
>> J.
>>
>> ---------------------------------------------------------------------
>> Apache Qpid - AMQP Messaging Implementation
>> Project: http://qpid.apache.org
>> Use/Interact: mailto:users-subscribe@qpid.apache.org
>>
>>
>>
>
>
>
> ---------------------------------------------------------------------
> Apache Qpid - AMQP Messaging Implementation
> Project: http://qpid.apache.org
> Use/Interact: mailto:users-subscribe@qpid.apache.org
>
>
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:users-subscribe@qpid.apache.org
Re: Options for clustering/multicast
Posted by Clark O'Brien <cl...@yahoo.com>.
Could you use an external cluster manager like Pacemaker (http://www.clusterlabs.org/#2) to detect a failed node and move traffic to other nodes. Pacemaker could also start a recovery instance. Pacemaker is available from EPEL-compiled versions of Linux.
I would really like to hear opinions on this strategy especially from guys with networking expertise. We are building messaging in the cloud and this is a strong contender.
--- On Tue, 10/19/10, Jason Stelzer <ja...@gmail.com> wrote:
> From: Jason Stelzer <ja...@gmail.com>
> Subject: Re: Options for clustering/multicast
> To: "Alan Conway" <ac...@redhat.com>
> Cc: users@qpid.apache.org
> Date: Tuesday, October 19, 2010, 10:57 AM
> What if I setup a gre tunnel between
> the 2 hosts with the mtu set low
> enough that it doesn't fragment the packets? At that point
> I could
> configure corosync to multicast to the virtual interface
> and work
> around the dodgy switch.
>
> On Tue, Oct 19, 2010 at 11:19 AM, Alan Conway <ac...@redhat.com>
> wrote:
> > On 10/19/2010 10:56 AM, Jason Stelzer wrote:
> >>
> >> I am setting up a 2 node cluster using the C++
> qpidd 0.5 on fedora 12.
> >> This is primarily for failover.
> >>
> >> I'm in a situation where one of our hardware
> vendors is not handling
> >> multicast correctly.
> >>
> >> As an interim solution, is it possible to cluster
> via unicast? My
> >> understanding is that the corosync library uses
> multicast and that
> >> qpidd relies on that working correctly. This seems
> to imply that
> >> unicast is not possible, but I figured I'd ask and
> see if I get a
> >> definitive answer one way or the other.
> >>
> >
> > No, there's not currently a way to run clustering over
> unicast.
> >
>
>
>
> --
> J.
>
> ---------------------------------------------------------------------
> Apache Qpid - AMQP Messaging Implementation
> Project: http://qpid.apache.org
> Use/Interact: mailto:users-subscribe@qpid.apache.org
>
>
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:users-subscribe@qpid.apache.org
Re: Options for clustering/multicast
Posted by Jason Stelzer <ja...@gmail.com>.
What if I setup a gre tunnel between the 2 hosts with the mtu set low
enough that it doesn't fragment the packets? At that point I could
configure corosync to multicast to the virtual interface and work
around the dodgy switch.
On Tue, Oct 19, 2010 at 11:19 AM, Alan Conway <ac...@redhat.com> wrote:
> On 10/19/2010 10:56 AM, Jason Stelzer wrote:
>>
>> I am setting up a 2 node cluster using the C++ qpidd 0.5 on fedora 12.
>> This is primarily for failover.
>>
>> I'm in a situation where one of our hardware vendors is not handling
>> multicast correctly.
>>
>> As an interim solution, is it possible to cluster via unicast? My
>> understanding is that the corosync library uses multicast and that
>> qpidd relies on that working correctly. This seems to imply that
>> unicast is not possible, but I figured I'd ask and see if I get a
>> definitive answer one way or the other.
>>
>
> No, there's not currently a way to run clustering over unicast.
>
--
J.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:users-subscribe@qpid.apache.org
Re: Options for clustering/multicast
Posted by Alan Conway <ac...@redhat.com>.
On 10/19/2010 10:56 AM, Jason Stelzer wrote:
> I am setting up a 2 node cluster using the C++ qpidd 0.5 on fedora 12.
> This is primarily for failover.
>
> I'm in a situation where one of our hardware vendors is not handling
> multicast correctly.
>
> As an interim solution, is it possible to cluster via unicast? My
> understanding is that the corosync library uses multicast and that
> qpidd relies on that working correctly. This seems to imply that
> unicast is not possible, but I figured I'd ask and see if I get a
> definitive answer one way or the other.
>
No, there's not currently a way to run clustering over unicast.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:users-subscribe@qpid.apache.org