#opendaylight-meeting: OpenDaylight-OPNFV interlock meeting

Meeting started by dneary at 15:02:54 UTC (full logs).

Meeting summary

    1. goal of meeting is to introduce OPNFV and ODL communities (dfarrell07, 15:05:07)
    2. also going to talk about an issue we ran into recently, details of how it was fixed (dfarrell07, 15:05:26)
    3. edwarnicke (edwarnicke, 15:06:00)
    4. dfarrell07 (dfarrell07, 15:06:10)
    5. regXboi (Ryan Moats) (regXboi, 15:06:10)
    6. Daniel Smith, Ericsson (dneary, 15:06:13)
    7. tbachman (tbachman, 15:06:16)
    8. Prem, Ericsson (Prem, 15:06:21)
    9. shague Red hat, odl ovsdb project (shague, 15:06:28)
    10. ebrjohn Brady Johnson, Ericsson, SFC Project Lead (ebrjohn, 15:06:29)
    11. Uli Kleber, Huawei (dneary, 15:06:42)
    12. Brady Johnson, Ericsson (dneary, 15:06:51)
    13. Frank Brockners, Cisco (frankbrockners, 15:06:55)
    14. flaviof Red Hat, Boston (flaviof, 15:07:06)
    15. trozet, Red Hat (trozet, 15:07:35)
    16. From OPNFV BGS: trozet, frankbrockners (dneary, 15:09:10)
    17. trozet discovered most recent issue, gives overview (dfarrell07, 15:09:10)
    18. From OVSDB project in OpenDaylight: flaviof, shague (dneary, 15:09:30)
    19. tl;dr is that there was a timing issue involving feature loading in ODL, trozet made changes in quickstack Puppet mod to work around it (dfarrell07, 15:10:11)
    20. trozet changed quickstack to wait for odl to have bundled features loaded before having neutron connect to odl (flaviof, 15:10:42)
    21. there's another issue that ODL is sending 200 OKs when it shouldn't (details in open bugs) (dfarrell07, 15:11:47)
    22. I actually argue that "another issue" is not an issue - it is things working by design (regXboi, 15:13:11)
    23. trozet waits for the net-virt feature to show bundle active in order to work around issue #1 (flaviof, 15:13:42)
    24. From the ODL Controller: Ryan Moats, Ed Warnicke (dneary, 15:14:13)
    25. From ODL Integration and ODL Puppet mod, Daniel Farrell (dfarrell07) (dfarrell07, 15:14:39)
    26. correction: Ed Warnicke is from Controller, Ryan Moats is a general gadfly (regXboi, 15:14:44)
    27. discussion about making sure we're waiting on the right "things are loaded and gtg"-type event from ODL (dfarrell07, 15:15:35)
    28. question on ODL start order from dneary (dfarrell07, 15:18:54)
    29. 1. Neutron -> ODL ML2 agent (OpenStack side) (dneary, 15:19:49)
    30. neutron in odl uses whiteboard pattern so an other odl bundle -- ovsdb being one of them -- can get neutron events (flaviof, 15:20:10)
    31. 2. ML2 agent -> Neutron service (part of Controller) (dneary, 15:20:11)
    32. 3. Controller SB services get "offered" Neutron request, if all say "yes" when offered, the controller responds "OK 200" (dneary, 15:20:57)
    33. 4. SB services then commit to handling the request SB asynchronously (dneary, 15:21:22)
    34. flaviof: "At this point, there are no SB modules, no modules say no, therefore request is accepted, but can't be handled" (dneary, 15:22:03)
    35. trozet: my scenario is ODL is fully up, 2 OVS nodes ( one on compute, one one network), if you remove br-int from network, ODL will still add vxlan tunnels/flows to br-int on compute and act like its creating networks, even though the other side of the tunnel is missing (dneary, 15:35:48)
    36. trozet: if you remove br-int from compute side nova will error out when you try to bring up the instance, but this doesnt happen on the other side (dneary, 15:36:03)
    37. trozet: if ODL gets a request from Neutron, and it is missing its br, it should error back on the REST call imo (dneary, 15:36:18)
    38. Discussion of when it's appropriate to be synchronous in ML2 responses and whether it's possible to do async notifications via (for example) OpenStack's MQ (dneary, 15:40:34)
    39. regXboi proposes the implementation that uses config and operational states (flaviof, 15:45:45)
    40. Some action items for OpenDaylight: If there are no SB modules available, return a fail on any NB requests (dneary, 15:51:58)
    41. Some debate about whether to have a fire/notify model or a 2 pass approach ("request then check") but general opinion expressed by edwarnicke that there's a mismatch between Neutron's synchronous request model & a back-end async model (dneary, 15:53:48)
    42. question from OPNFV: "we need a mostly-working deployment in the next few weeks. possible at all?" answer from shague: "trozet has a working setup today. it doesn't address all of the details described on today's call, but it does the basics" (dfarrell07, 16:00:07)

Meeting ended at 16:05:08 UTC (full logs).

Action items

  1. (none)

People present (lines said)

  1. dneary (34)
  2. dfarrell07 (26)
  3. tbachman (13)
  4. regXboi (11)
  5. flaviof (9)
  6. odl_meetbot (7)
  7. shague (5)
  8. trozet (5)
  9. frankbrockners (4)
  10. Prem (1)
  11. ebrjohn (1)
  12. edwarnicke (1)
  13. mlemay (1)

Generated by MeetBot 0.1.4.