You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@qpid.apache.org by Alan Conway <ac...@redhat.com> on 2012/01/25 23:32:26 UTC
Progress on new HA approach
For anyone who's interested: This work is coming along on branch qpid-3603-2
(note qpid-3603 is obsolete)
At this point we have
- replication of configuration (queues, exchanges bindings)
- replication of messages
- clients directed to primary broker, excluded from backups.
- client failover (tested for c++ and python clients at this point)
There is still work to do to support broker failover and integration with a
cluster resource manager (e.g. rgmanager.)
I have updated the design doc to better reflect the current state of affairs:
http://svn.apache.org/viewvc/qpid/branches/qpid-3603-2/qpid/cpp/design_docs/new-ha-design.txt?view=markup
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: Progress on new HA approach
Posted by Alan Conway <ac...@redhat.com>.
On 01/26/2012 05:50 AM, Rob Godfrey wrote:
> Looks interesting...
>
> how do you ensure transactional atomicity with this approach (e.g. on the
> primary broker in a transaction I receive from one queue, do work and then
> publish to a second queue... if the queues are replicating independently is
> it possible to guarantee that in the backup broker you do not have the
> condition that the dequeue was replicated but the enqueue was not?)
Currently we don't, it's mentioned in the section on limitations. I think the
solution will be to add extra transactional "event" messages (like the position
event messages that are sent now) and to annotate messages with headers
providing transactional info. It'll be a significant piece of work but I think
its doable.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: Progress on new HA approach
Posted by Rob Godfrey <ro...@gmail.com>.
Looks interesting...
how do you ensure transactional atomicity with this approach (e.g. on the
primary broker in a transaction I receive from one queue, do work and then
publish to a second queue... if the queues are replicating independently is
it possible to guarantee that in the backup broker you do not have the
condition that the dequeue was replicated but the enqueue was not?)
Cheers,
Rob
On 25 January 2012 23:32, Alan Conway <ac...@redhat.com> wrote:
> For anyone who's interested: This work is coming along on branch
> qpid-3603-2 (note qpid-3603 is obsolete)
> At this point we have
> - replication of configuration (queues, exchanges bindings)
> - replication of messages
> - clients directed to primary broker, excluded from backups.
> - client failover (tested for c++ and python clients at this point)
>
> There is still work to do to support broker failover and integration with
> a cluster resource manager (e.g. rgmanager.)
>
> I have updated the design doc to better reflect the current state of
> affairs: http://svn.apache.org/viewvc/**qpid/branches/qpid-3603-2/**
> qpid/cpp/design_docs/new-ha-**design.txt?view=markup<http://svn.apache.org/viewvc/qpid/branches/qpid-3603-2/qpid/cpp/design_docs/new-ha-design.txt?view=markup>
>
>
> ------------------------------**------------------------------**---------
> Apache Qpid - AMQP Messaging Implementation
> Project: http://qpid.apache.org
> Use/Interact: mailto:dev-subscribe@qpid.**apache.org<de...@qpid.apache.org>
>
>