ODL Notes
ODL clustering has no notion of discovery. There is a flat file (akka.conf) on each VM that contains info about its peers.
Basically the two main limitations are (a) the total number of nodes has to be odd (due to how voting works) and (b) if you add additional nodes, you need to update the akka.conf on all the existing nodes as well and bounce them. Make sure to bring up the old ones first to make sure one of those becomes the "shard leader" (otherwise I'd worry that one of the new ones would think it's the leader even though it has no data yet, and "replicate" its lack of data to the others).
ODL clustering was designed for high availability really - not for scaling. So that's why it doesn’t address scaling in any real way. There are some improvements in clustering planned for the next ODL release (Oxygen), the details for the improvements are not yet known.
ConfigScaleOut API Definition
Information on the ConfigScaleOut can be found on readthedocs:
2 Comments
James MacNider
Hi Randa Maher, the SDN-C project is similarly limited by the design of ODL when it comes to horizontal scaling, and it's been a topic of some discussion lately due to the platform maturity requirements in Beijing. Given that APP-C and SDN-C share a common platform we should be trying to work closely together on a number of these requirements so we can avoid duplicating effort. I'm not sure if you and Dan Timoney are already coordinating, but I'll try to attend the APP-C weekly calls over the next while to help bridge the projects. You can see the table of the SDN-C planned commitments put together by Dan Timoney for these requirements here.
Randa Maher
Hi James,
Working together would be great! Aaron Hay did speak with Dan Timoney and much of what is noted on this page is based on discussions with him. Looking forward to your participation and collaboration!