[ovs-discuss] OVN scalability and HA

Andy Zhou azhou at ovn.org
Thu Jul 14 21:28:39 UTC 2016


On Thu, Jul 14, 2016 at 9:34 AM, Alexey I. Froloff <raorn at raorn.name> wrote:

> Greetings,
>
> We're about to use OVN in a large production environment :-)
>
> Well, not really "production" production, but having fun testing
> it, looking if we can use it in such way.  While functionality is
> good for us, two question appear - scalability and availability.
>
> What are current OVN (as of master branch) limitations in
> hypervisors in one segment (one SBDB), lswitches in one segment,
> lports in one lswitch, etc?
>
> About clustering.  Since ovn-northd does not support clustering
> by itself, it is possible to create active-passive HA cluster
> ising third-party clustering software.  NBDB and SBDB are
> replicated using ovsdb mechanisms and clustering software ensures
> that there are one copy of ovn-northd daemon is running.  Is this
> correct?  Can ovn-controller's automatically switch to different
> database server?
>

There may be multiple possible ways to deploy OVN in a production
environment, an OVN release probably should be able support at least
one, and probably document other use cases OVN can support.

In the long run, we'd like etcd to be actual database for NBDB and SBDB, but
the work is not going to land in 2.6.0

For 2.6.0, OVSDB server already has the ability to replicate transactions
from
an active server to an stand-by server. contributed by Mario Cabrera from
hpe.com.
We plan to make use of this feature for 2.6.0 release.

This is what I think may work at a high level,  ovn-northd and
ovn-controller should be told
about both active and backup servers and only connect to the active server
initially.
When the connections to the active ovsdb server are lost, they will try to
establish connections
to the back up server, and confirm that the back up server is now an active
server, then continue
to issue transactions from that point on.

We would like the default OVN HA clustering solution not require additional
software, in order
to reduce overall system complexity.

The above functions have not been flushed out nor implemented.  Feedbacks
are welcome.


> What clustering/HA/scalability features are planned for 2.6.0
> release?
>

Would above plan, or something similar be sufficient?


>
> Thank you in advance.
>
> --
> Regards,    --
> Sir Raorn.   --- https://plus.google.com/+AlexeyFroloff
>
> _______________________________________________
> discuss mailing list
> discuss at openvswitch.org
> http://openvswitch.org/mailman/listinfo/discuss
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20160714/18f8faec/attachment-0002.html>


More information about the discuss mailing list