We are from the NETVIRT project of OpenDaylight Controller and we are involved
driving the "L3VPN-as-a-Service" subproject within the same.
This service is triggered/realized by using Neutron Routers and/or Neutron
BGPVPN feature orchestrations.
We had an opportunity to learn few failures on production deployments of ODL
(3-node), where we
figured out that such problems were due to improper serialization of workflows
scaled across the 3-nodes of the ODL cluster.
In order to serialize such workflows (few eg):
VIP movement in the cloud between VM ports etc ),
we were contemplating to generate and store Event Objects into a Queue with
the backend for
the queue being MD-SAL Datastore (Operational Shard).
While we understand that OP Datastores are used for only State-information
management inside ODL, we wanted an
Event-Serialization mechanism that will spawn across the ODL cluster and
consumers servicing such events in that order
provides for consistent runtime-behaviour across the nodes of the ODL cluster.
Can you please let us know if this approach of wrapping the Op Datastores with
a Queuing framework, and
using the same for Event-Serialization is wise to pursue?
Are there other implementations within ODL that use Event Queues (that are
usable across a 3-node cluster) , and
if so can you please point us to them ?
controller-dev mailing list