#opendaylight-meeting: MD-SAL Interest Call

Meeting summary

1. agenda (colindixon, 16:12:23)
   1. https://wiki.opendaylight.org/view/MD-SAL_Weekly_Call this is where the agenda and notes will be (colindixon, 16:12:45)
   
   
2. status update (colindixon, 16:16:12)
   1. https://git.opendaylight.org/gerrit/#/c/9496/ is the introduction of the writer implementation (rovarga_, 16:31:13)
   2. https://gist.github.com/3a4ac3907c45049c6789 (flaviof, 16:35:55)
   
   
3. clustering demo (colindixon, 16:45:17)
   1. there won’t be a recording because we don’t have the host (colindixon, 16:46:50)
   2. moiz shows the config in “configuration/initial/module-shards.conf” (colindixon, 16:47:45)
   3. it includes information about the shards and the shard leaders in JSON format (colindixon, 16:48:07)
   4. https://git.opendaylight.org/gerrit/#/c/9677/ this patch has the code which has this file (colindixon, 16:49:32)
   5. demos that data is available from any node even though it’s stored only on one node (colindixon, 16:50:21)
   6. demos with toaster that changing the toaster config on one node and showing it’s represented on all nodes in the cluster (colindixon, 16:51:57)
   7. then he shuts down the leader of the toaster shard (colindixon, 16:52:15)
   8. shows that the data is still there (colindixon, 16:52:21)
   9. the leader has now changed to one of the up nodes (colindixon, 16:52:59)
   10. this means we can still change the values even after the old leader has failed (colindixon, 16:53:47)
   11. in addition, we have persistence from this (colindixon, 16:54:08)
   
   
4. questions for the demo (colindixon, 16:54:41)
   1. colindixon asks “this looks like it’s working, is that right?” (colindixon, 16:54:56)
   2. answer is that there are still some places where things need to be tweaked, for example sending of snapshots of state between nodes vs. just updates (colindixon, 16:55:38)
   3. robert varga asks how serialization is working for them (colindixon, 16:55:54)
   4. answer is that they’re using protocol buffers for serialization and they’re benchmarking it, but the scaling seems like it could use a bit of work (colindixon, 16:56:58)
   
   
5. clustering RPC demo (colindixon, 16:58:15)
   1. starts one mininet instance pointing to one controller instance in the cluster (colindixon, 16:58:59)
   2. shows that you can make the RPC call from a different instance in the cluster and it will appropriately go to the right instance (colindixon, 17:01:01)
   3. the demo actually shows doing a RESTCONF operation to the flow service, which then turns it into an RPC to the OpenFlow plugin, which installs the flow (colindixon, 17:02:14)
   4. Ramkumar asks what changes are needed from the OpenFlow plugin (colindixon, 17:03:02)
   5. answer is none by default, they will need to change things if they want to take advantage of the clustering. i.e., recover from a failed instance cleanly (colindixon, 17:03:55)
   6. there is a question about how the remote RPCs are routed to the appropriate cluster instance (colindixon, 17:06:39)
   7. answer is that you shouldn’t have to worry about that unless you want to really carefully control it (colindixon, 17:07:19)

Action items

1. (none)

People present (lines said)

1. colindixon (43)
2. tbachman (5)
3. odl_meetbot (4)
4. edwarnicke (2)
5. dmm (2)
6. flaviof (1)
7. rovarga_ (1)