#opendaylight-ovsdb: ovsdb_weekly_call

Meeting started by tbachmanAfk at 19:03:19 UTC (full logs).

Meeting summary

    1. https://meetings.opendaylight.org/opendaylight-ovsdb/2015/ovsdb_weekly_call/opendaylight-ovsdb-ovsdb_weekly_call.2015-02-24-19.59.html Last recorded meeting minutes (tbachman, 19:04:56)

  1. agenda (tbachman, 19:05:01)
    1. topics: status and Trello (tbachman, 19:06:08)

  2. Trello (tbachman, 19:06:09)
    1. flaviof says there have been problems with the bulk tests — not sure why it’s failing yet (tbachman, 19:11:10)
    2. https://gist.github.com/a223d79b648c51128223 Output when bulk tests fail (tbachman, 19:12:00)
    3. shague says the tempest tests are running now, so now the debugging of those tests can begin, because we can actually see the results (tbachman, 19:12:21)
    4. flaviof says we need to use the latest ODL rather than Helium ODL (tbachman, 19:12:31)
    5. shague asks about the report that gets generated — does the latest gerrit allow the report to be generated (tbachman, 19:12:55)
    6. flaviof says he’s not sure — he needs to circle back with mestery on that (tbachman, 19:13:05)
    7. flaviof says it’s basically knowing where the file is and having JJB putting in the artifact of the build (tbachman, 19:13:18)
    8. LuisGomez_ asks if there’s already automation in place to deploy devstack (tbachman, 19:13:53)
    9. flaviof says they have it - just needs to be modfiied to support what the integration test folks need (tbachman, 19:14:07)
    10. flaviof says it uses an unstable openstack with a stable ODL, which needs to be reverted (tbachman, 19:14:21)
    11. flaviof says there’s a caveat in that it’s just a single host test — so, no multinode testing, which is something that the OVSDB project is interested in (tbachman, 19:14:46)
    12. flaviof has been talking with dfarrell07 re: his efforts on this same task (devstack use) (tbachman, 19:15:02)
    13. mlemay says there’s two types of reports he’s been looking into: the openstack one and another one (tbachman, 19:15:48)
    14. mohnish says one of his colleagues is trying to reproduce the tempest tests, and was wondering if the instructions have been sent out as to how to reproduce this? (tbachman, 19:16:45)
    15. flaviof says there are two ways. The easiest is using the vagrant VM that dave_tucker and mestery created and use; with that vagrant file, you can stamp out a VM, run the stack, and then the tempest tests (tbachman, 19:17:28)
    16. https://etherpad.openstack.org/p/odl-virtual-ci-hackfest hackfest on running temptests (flaviof, 19:17:55)
    17. mlemay says that his other reporting is simply vagrant + OVS and have a DUT (device under test) structure to simply test basic OVSDB operations (crud bridge, tunnels, etc.) (tbachman, 19:18:26)
    18. edwarnicke says there’s one more feature that he’s likely to add for the MD-SAL southbound (tbachman, 19:18:44)
    19. edwarnicke says the mapping of the OVSDB stuff back to the flow model is challenging, so he’s adding something that if a flownode comes up for a particular OVSDB managed bridge, you get an instance identifier that points to the flownode — he’s thinking of doing this as a separate feature (tbachman, 19:19:29)
    20. repenno asks with M3 approaching, he sent an email to the list about CRUD for interfaces, bridges, etc. and some folks said they were working on that. This functionality is needed by SFC project (tbachman, 19:21:28)
    21. shague says that OVSDB doesn’t have an API freeze, but offset 0 projects do (tbachman, 19:22:47)
    22. colindixon says he believes we don’t need to have the YANG model nailed down until a couple of weeks (tbachman, 19:23:35)
    23. edwarnicke says most of the model is picking up extensions, rather than changes (tbachman, 19:23:57)
    24. repenno says he’s wondering if folks are working on the CRUD APIs; the APIs that he’s looking for needs to be callable from SFC (tbachman, 19:26:07)
    25. SharadMishra says he provided the patches and added repenno as a reviewer (tbachman, 19:27:43)
    26. repenno will look at the patch and provide feedback (tbachman, 19:28:01)
    27. shague says edwarnicke has done most of the bridge side — bridge, port, interface (tbachman, 19:28:37)
    28. edwarnicke says for bridges, there’s a patch out; there are patches out that bring in the datapath information and protocols (tbachman, 19:29:15)
    29. edwarnicke says right now, datapath is coming back from the library API as a set, and he didn’t think the datapath was multi-valued in the schema (tbachman, 19:31:20)
    30. edwarnicke asks if that’s a bug in the library, or if that’s actually part of the schema (tbachman, 19:31:37)
    31. shague says it may be — they haven’t used the datapath much (tbachman, 19:31:48)
    32. repenno asks if there’s an API exposed that SFC can use (tbachman, 19:33:38)
    33. edwarnicke says repenno is probably looking for the overlay API; that’s next after the CRUD is working (tbachman, 19:33:53)
    34. repenno asks how he creates bridges, ports, etc. and do that on OVS nodes (tbachman, 19:34:22)
    35. edwarnicke says if you want to create those, you can use the stuff in the ovsdb.yang model directly (tbachman, 19:34:38)
    36. edwarnicke says you can write to the datastore directly (tbachman, 19:34:49)
    37. repenno says then you’ll have a multi-writer issue (tbachman, 19:35:02)
    38. edwarnicke says it’s intended to allow multiple applications to ask for things (tbachman, 19:35:13)
    39. edwarnicke says this should be reflected in the operational datastore (tbachman, 19:36:14)
    40. shague says there’s a southbound shim and southbound factory methods (tbachman, 19:37:24)
    41. repenno says that’s what he’s referring to — the shim (tbachman, 19:37:40)
    42. repenno says SFC needs something like “create this interface”, “create this bridge”, or “create this port" (tbachman, 19:38:01)
    43. repenno says the question is whether this is a datastore access or something that calls in to OVSDB (tbachman, 19:38:24)
    44. edwarnicke asks if repenno’s asking about convenience methods or RPCs (tbachman, 19:38:39)
    45. repenno says whatever works (tbachman, 19:38:49)
    46. repenno says if OVSDB creates an API, then that API may have extra sanitization for accessing the data store; if SFC does it, then it has to create the extra sanitization to access the data store (tbachman, 19:39:52)
    47. mohnish asks if some portion of the SB yang models are going to go through the network, and some will be non-network (tbachman, 19:40:41)
    48. mohnish asks if the network needs to be notified of the se changes (tbachman, 19:41:36)
    49. edwarnicke says anyone can subscribe to the datastore to get norifications for the changes (tbachman, 19:41:46)
    50. mohnish asks if netvirt is listening to those changes (tbachman, 19:42:02)
    51. mohnish asks if groupbasedpolicy is going to use the neutron APIs and will use the OVSDB APIs, and stil manage the openflow APIs (tbachman, 19:43:32)
    52. edwarnicke says exactly — it would manage the flows directly, OVSDB SB to manage bridges, etc. (tbachman, 19:43:45)
    53. edwarnicke says he believes SFC is also managing the flows directly as well (tbachman, 19:43:56)
    54. repenno says that’s correct (tbachman, 19:44:10)
    55. shague says the documentation needs to be started by this Thursday (tbachman, 19:44:44)
    56. colindixon says the documentation in Lithium is attempting to get a more uniform agreement on what each document type is for — may require some massaging in order to adhere to the new outlines (tbachman, 19:45:34)
    57. shague says this may be just taking a lot of the documentation from helium and putting it in the new format (tbachman, 19:46:14)
    58. shague says they’ve gotten rid of the majority of the AD-SAL pieces (tbachman, 19:47:01)
    59. shague says they will then start looking at the tempest tests pretty heavily (tbachman, 19:47:13)
    60. shague says right now, the net-virt doesn’t come down through the MD-SAL; if they have time, they’ll look into whether that can be migrated as well (tbachman, 19:48:39)
    61. mohnish says if net-virt is the top-most application, then the consumer is net-virt itself; are we going to push that information into a data store? (tbachman, 19:49:27)
    62. shague says with the existing plugin, you get all of OVSDB (tbachman, 19:49:45)
    63. edwarnicke says the idea of the overlay model is that no single domain of control is consistent (tbachman, 19:50:05)
    64. mohnish says he’s thinking from the net-virt application itself — asks if there’s work going on there as well (tbachman, 19:50:50)
    65. shague says the net-virt application is only going to work with OVS specifically (tbachman, 19:51:02)
    66. flaviof asks if mohnish was asking about the L3 forwarding (tbachman, 19:52:09)
    67. mohnish says yes — understood that some of the tempest tests were failing b/c of L3; some of his team-members were looking at these failures as well (tbachman, 19:52:46)
    68. flaviof says for L3 forwarding, it’s less about OVSDB, and more about openflow (tbachman, 19:54:06)
    69. flaviof says there’s a question of where that router’s intelligence lives — Linux IP stack, or OpenDaylight (tbachman, 19:55:03)
    70. flaviof says L3 forwarding is using Nicira extensions to flow-mods (tbachman, 19:55:26)
    71. mohnish says when the APIs are invoked, the expectation is that ODL would be implementing something equivalent to DVR; the question is the user wants to manage the bridges and routers through opendaylight? (tbachman, 19:56:53)
    72. flaviof says as of now it’s a DVR replacement; there’s a hybrid approach where ODL fills in OpenStack’s DVR shortcomings and vise-versa (tbachman, 19:57:25)
    73. flaviof says there’s a distributed way of replying to ARPs; but there’s no way of doing an ICMP reply to TTL expirations (tbachman, 19:58:01)
    74. mohnish says this needs more discussion — asks if shague and flaviof would want a separate call for this (tbachman, 19:58:39)
    75. flaviof says we can, but there currently isn’t enough cycles to do this for Lithium (tbachman, 19:58:59)
    76. Swami says we can have a discussion, but as flaviof said, for Lithium, there’s no focus on DVR on par wtih neutron; would like to be included in any such discussion tho (tbachman, 19:59:53)


Meeting ended at 20:00:15 UTC (full logs).

Action items

  1. (none)


People present (lines said)

  1. tbachman (97)
  2. odl_meetbot (9)
  3. flaviof (7)
  4. tbachmanAfk (5)
  5. tbachman_ (4)
  6. shague (2)
  7. mlemay (2)
  8. phrobb (1)


Generated by MeetBot 0.1.4.