Open Stack

But you could use a VIP for each rabbit cluster and use a TCP load balancer. Correct?

Subbu

On Feb 1, 2014, at 5:19 AM, Belmiro Moreira <moreira.belmiro.email.lists at gmail.com> wrote:

> Hi,
> to describe our experience,
> we are using RabbitMQ in active/active with mirror queues.
> We have 4 cells and each cell has 3 rabbitMQ servers with more than 2000 compute nodes in total.
> 
> Until now we didn’t have any particular issue with this configuration (we had a network partition but easily fixed).
> 
> The problem with cells is that between cells is only possible to define one rabbit host.
> There is an open bug: https://bugs.launchpad.net/nova/+bug/1178541
> 
> Belmiro
> 
> 
> 
> On Jan 31, 2014, at 18:48 , Allamaraju, Subbu <subbu at subbu.org> wrote:
> 
>> Alexander,
>> 
>> Thanks for sharing your experience. We've been using RabbitMQ active-active behind a VIP/TCP LB. Just wanted to check if the HA Guide's recommendation is valid.
>> 
>> Subbu
>> 
>> On Jan 31, 2014, at 1:27 AM, Papaspyrou, Alexander <papaspyrou at adesso-mobile.de> wrote:
>> 
>>> Subbu,
>>> 
>>> we are running on RMQ 3.2 with server-side HA policy on all OpenStack queues.
>>> 
>>> Even with two Rabbit servers dispersed over two sufficiently distant data centers (different subnets, different locations, connected via fibre over a number of routers), and besides network partitions here and there (which RMQ fixes automatically most of the time, if configured properly), our setup runs like a charm. Hardware is negligible; RMQ almost never goes beyond a GB or so of memory usage, and the CPU is usually bored to death.
>>> 
>>> We found the DRBD setup much more flaky and rather cumbersome to setup, and frankly, I never understood why people happen to take the dark alley Linux-HA. If you want, you can put ldirectord in front of the two boxes to balance the load (we did that, although from a performance perspective, this is not necessary). ldirectord also detects whether one of the boxes is out to lunch (which never happened so far), and reroutes the traffic automatically.
>>> 
>>> To be really sure, put a corosync/pacemaker-managed failover (virtual) IP in front of the two boxes, and run corosync/pacemaker/ldirectord/rabbitmq on each box, with properly configured VIP transportation – that’s where I’d invest the effort to dive into the details of Linux HA glory.
>>> 
>>> Regards,
>>> Alexander
>>> -- 
>>> adesso mobile solutions GmbH
>>> Dr.-Ing. Alexander Papaspyrou
>>> Senior System Architect
>>> IT Operations
>>> 
>>> Freie-Vogel-Str. 391 | 44269 Dortmund
>>> T +49 231 930 66480 | F +49 231 930 9317 | M +49 172  209 4739
>>> Mail: papaspyrou at adesso-mobile.de | Web: www.adesso-mobile.de | Mobil-Web: mobil.adesso-mobile.de
>>> 
>>> Vertretungsberechtigte Geschäftsführer: Dr. Josef Brewing, Frank Dobelmann
>>> Registergericht: Amtsgericht Dortmund
>>> Registernummer: HRB 13763
>>> Umsatzsteuer-Identifikationsnummer gemäß § 27 a Umsatzsteuergesetz: DE201541832
>>> 
>>> 
>>> 
>>> 
>>> Am 20.01.2014 um 05:33 schrieb Allamaraju, Subbu <subbu at subbu.org>:
>>> 
>>>> OpenStack HA guide (http://docs.openstack.org/high-availability-guide/content/s-rabbitmq.html) says that Pacemaker/DRBD approach is preferred over  active-active mirrored queues. Details are sparse in the guide. Is anyone aware of any data/issues first hand?
>>>> 
>>>> Thanks for any pointers.
>>>> 
>>>> Subbu
>>>> _______________________________________________
>>>> OpenStack-operators mailing list
>>>> OpenStack-operators at lists.openstack.org
>>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>>> 
>> 
>> 
>> _______________________________________________
>> OpenStack-operators mailing list
>> OpenStack-operators at lists.openstack.org
>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>