#opendaylight-meeting: clustering hackers

Meeting summary

1. agenda bashing (colindixon, 15:08:19)
   1. https://git.opendaylight.org/gerrit/#/c/45912/ rovarga asks TomP to look into this patch to merge it before merge conflicts start to come up (colindixon, 15:09:30)
   2. colindixon and TomP give a once-through and find a few minor things, TomP says he's likely to give the benefit of the doubt around which parents each pom uses (colindixon, 15:13:49)
   
   
2. moving to frontend/backend separation (bug 5280) (colindixon, 15:22:13)
   1. https://git.opendaylight.org/gerrit/#/q/message:5280 patches (colindixon, 15:22:16)
   2. TomP asks about how well it is tested (colindixon, 15:22:26)
   3. rovarga says that the first thing to do is to get the unit tests to run (colindixon, 15:22:55)
   4. Muthu and Ashutosh say that a lot of testing (beyond very high scale) is already in CSIT (colindixon, 15:23:20)
   5. the very high scale seems to have issues which are hard to figure out if they are the data store or the OpenFlow plugin (colindixon, 15:24:34)
   6. rovarga also says that his take is that there should be at least one test case on BGP that would hopefully go back to passing (colindixon, 15:25:01)
   7. TomP and rovarga discuss cherry-picking the patches back to Boron, and when to flip the switch to use the new back end (colindixon, 15:35:48)
   8. TomP seems to think flipping the switch in SR2 is more sane right now givent that SR1 is ~3 weeks out (colindixon, 15:36:12)
   9. rovarga says the key point is that the switch-over shoudl really be a config option, not a code option, so no special build shoudl be required (colindixon, 15:39:24)
   
   
3. carbon planning (colindixon, 15:44:35)
   1. there is some work on sharding to be able to use pluggable back-ends with Erricsson folks interested in using cassandra to back some shards (colindixon, 15:45:13)
   2. it looks like the plan is augment the existing models with RPCs that would then call back into the backing data store without going through the MD-SAL data store (colindixon, 15:46:01)
   3. rovarga says he'd really like to see it go through the conceptual data tree APIs instead of just using their own RPCs to get the current benefits of things, e.g., visibility to RESTCONF (colindixon, 15:46:37)
   4. ashutosh says that he agrees, but that at some point if you need performance, it might be worth a trade-off where you lose visibility to RESTCONF and data sharing in exchange for performance (colindixon, 15:50:17)
   5. TomP asks if this is upstreamed, Ashutosh says the discussion is happening in the open, TomP asks what project this is looking to modify and if the modifications are happening upstream or not (colindixon, 15:51:09)
   6. ashutosh says that the bascially, they don't have confidence that the conceptual data tree API will give them the flexibility they need to deliver the performance and scale they need, so they're focusing on bypassing things for now (colindixon, 15:55:46)
   7. ashutosh points out for instance that needing to replicate all the data you need on all nodes where you might want a notificaiton would be real problem for scale (colindixon, 15:56:38)
   8. rovarga, TomP and ashutosh talk about how the conceptual data tree would work and how it might work for the given use case, ashutosh asks questions around if there is still a shard leader (colindixon, 15:57:51)
   9. ashutosh says that in the long-run if the conceptual data tree becomes real, they'll move to it, but now they want to do something *now* and the conceptual data tree isn't yet done (colindixon, 16:00:25)
   10. muthu asks if this stuff was part of the summit presentations, rovarga says yes (colindixon, 16:02:21)
   
   
4. jumping to MD-SAL call (colindixon, 16:02:53)
   1. since I'm the only one taking notes, the notes might not reflect things after that (colindixon, 16:03:06)

Action items

1. (none)

People present (lines said)

1. colindixon (30)
2. odl_meetbot (4)