[ovs-discuss] On OVNDB-HA using pacemaker two issues discussed

Andy Zhou azhou at ovn.org
Wed Dec 7 21:04:03 UTC 2016


On Tue, Dec 6, 2016 at 7:28 PM, 姜尚0387 <ligs at dtdream.com> wrote:

> I am learning OVNDB-HA using pacemaker, I feel this is a good design.
>
> At the same time I have two questions and would like to discuss:
>
>   1,  we have two resources, one is the master node of the OVNDB, one is
> VIP.
>
>        Pacemaker resource constraints are dependencies, not symbiotic, so
> we must consider who depends on who.
>
>        http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html/
> Pacemaker_Explained/s-resource-colocation.html
>
>        If the master node OVNDB dependence VIP,  VIP or VIP network down,
> OVNDB master node will follow the VIP migration, But the OVNDB process of
> the master node is dead and there is no migration.
>
>        If the VIP is dependent on the OVNDB master process,  When the
> OVNDB master process down, all will migrate. And VIP or VIP network down
> Indicating that the node is down, all will migration.
>
>        Is it better to configure the resource constraints ?
>
>   pcs constraint order ovndb_servers-master then VirtualIP
>   pcs constraint colocation add VirtualIP with master ovndb_servers-master score=INFINITY
>
>
> OVSDB-HA is mainly designed to provide high availability against host
crashes. In case ovsdb server crashes, it can restart when --monitor option
is specified.


>   2,  OVN nodes in a total of three processes, the South to the database,
> north to the database, northd process.
>
>        But only in the OCF script to start and monitor the database
> process, there is no northd process.
>
>          How to ensure high reliability northd process?
>
>          My idea is OCF script at the same time start and monitor northd
> process, because I think these three processes is together.
>

I agree that the northd HA is not well addressed with current OCF script.
However, I am not sure it can be addressed by simply extending current OCF
scripts
In theory, the northd can run on another machine, accessing ovsdb-server
via the virtual IP.  May be this is more a packaging issue.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openvswitch.org/pipermail/ovs-discuss/attachments/20161207/e3b3a1d6/attachment.html>


More information about the discuss mailing list