#opendaylight-meeting: kernel projects

Meeting started by rgoulding at 16:01:54 UTC (full logs).

Meeting summary

  1. agenda bashing (rgoulding, 16:02:10)
    1. merged/modified patches in the last week (rgoulding, 16:03:52)

  2. How to have a single shard for operational and config, but with/without persistence? (rgoulding, 16:05:00)
    1. this topic comes from last week but no RH folks were on the call (rgoulding, 16:05:09)
    2. https://lists.opendaylight.org/pipermail/controller-dev/2017-October/013958.html (rgoulding, 16:07:38)
    3. skitt says there are a bunch of a places in netvirt that tries to manipulate config and oper simulateously which is pretty bad (rgoulding, 16:08:20)
    4. skitt this probably needs to be fixed in the downstream (rgoulding, 16:08:32)
    5. the solution is to split the writes up (rgoulding, 16:08:49)
    6. skitt if you don’t care about scaleout it tends to be easier to know where your leaders are and direct requests towards that specific node (rgoulding, 16:09:26)
    7. this can be done through configuring the module shards and is considered a deployment concern (rgoulding, 16:10:48)
    8. the way the term shard is used in ODL vs other distributed systems; you can usually use per key-range shard vs per yang tree (rgoulding, 16:11:28)
    9. rovarga the trouble is basically that you are comparing databases that arekey/value pairs that can be represented by a table. the problem is…. how do you shard a tree? (rgoulding, 16:12:05)
    10. part of the CDT design is that you can look at IID prefix to determine nested trees. thus you can shard the tree by hierarchy (rgoulding, 16:12:55)
    11. rather than using hashing / key-ranges (rgoulding, 16:13:08)
    12. for that to work you need some sort of consistent key hashing (rgoulding, 16:13:27)
    13. specifically in the case of lists (rgoulding, 16:13:40)
    14. this can be done but is not something we have done at this time (rgoulding, 16:13:47)
    15. can have a “shard” per list-entry item (rgoulding, 16:14:06)
    16. you cannot currently have a shard for a range of keys within a list right now (rgoulding, 16:14:18)
    17. that is a question of designing and implementing it rather than architecture and design (rgoulding, 16:14:30)
    18. it can be done but hasnt (rgoulding, 16:14:33)
    19. skitt where is the mapping done between shards and the tree? (rgoulding, 16:15:13)
    20. CDS takes care of module-based shards. MDSAL component is used for prefix-based shards in DOM Broker impl or somewhere? (rgoulding, 16:15:38)
    21. it is called ShardedDomDataTree (rgoulding, 16:16:15)
    22. https://wiki.opendaylight.org/view/MD-SAL:Boron:Conceptual_Data_Tree (rovarga_, 16:18:54)

  3. merged/pending patches in last 7 days (rgoulding, 16:20:35)
    1. rovarga many changes in yang-model-api (rgoulding, 16:21:12)
    2. one in particular is Bug-9265 (rgoulding, 16:21:40)
    3. this is done to ensure that leaf does not have a null value, but rather Empty value (rgoulding, 16:21:49)
    4. there will be work needed in mdsal mostly, and other downstreams probably don’t have to worry so much (rgoulding, 16:22:10)
    5. rovarga will communicate via email some of the changes as they near completion (rgoulding, 16:22:40)
    6. https://git.opendaylight.org/gerrit/63329 (rgoulding, 16:24:27)
    7. https://bugs.opendaylight.org/show_bug.cgi?id=9189 (rgoulding, 16:26:07)
    8. this may only affect integration/distribution (rgoulding, 16:29:15)
    9. this is just a convenience and can probably be done in integration/distribution or the existing code might be migrated off CSS? (rgoulding, 16:33:39)
    10. waiting for a “writable” local data store (rgoulding, 16:34:02)
    11. combo of RPC + blueprint could be used. or “.cfg” file using ManagedService or something (rgoulding, 16:35:59)
    12. https://wiki.opendaylight.org/view/Using_Blueprint#Application_configuration (rgoulding, 16:38:34)

  4. DiagStatus Expose Dataastore health to applications via infrautils.diagstatus (rgoulding, 16:39:46)
    1. https://lists.opendaylight.org/pipermail/controller-dev/2017-October/thread.html (rgoulding, 16:39:49)
    2. look at things related to “Expose Dataastore health to applications via infrautils.diagstatus" (rgoulding, 16:39:59)
    3. varying philosophies on “mature” vs “incubation” stuff here (rgoulding, 16:40:26)
    4. tpantelis raises the piont that this diagstatus mechanism is a small bundle (rgoulding, 16:41:19)
    5. it is a small component and completely new (rgoulding, 16:41:27)
    6. rovarga raises the point… “are we going to do stable features ever?" (rgoulding, 16:41:48)
    7. rovarga says that if you pull it into the DS, you cannot call the controller feature a “Stable Feature" (rgoulding, 16:42:08)
    8. since stable features require dependencies to be stable as well (rgoulding, 16:42:23)
    9. if it is just JMX, it can live anywhere (rgoulding, 16:42:32)
    10. for example, why can’t we use YANG model DTO(s) there instead of hand-written ones (rgoulding, 16:42:53)
    11. rovarga raises the point that this goes against the desire to do stable features (rgoulding, 16:43:06)
    12. rovarga asks whether this is still happening and whether it is completely shot from the radar (rgoulding, 16:43:24)
    13. can we have infrautils become “mature” then do the change? (rgoulding, 16:44:26)
    14. if infrautils just depends on odlparent, it needs to use custom DTOs instead of modeled DTOs (rgoulding, 16:47:11)
    15. this means that data becomes somewhat siloed (rgoulding, 16:49:24)
    16. skitt: this comes down to inability to make these small projects like lldp in Nitrogen etc. (rgoulding, 16:56:13)
    17. becomes a hierarchy question (rgoulding, 16:56:26)
    18. it is something needed to query th eTSC (rgoulding, 16:57:20)
    19. i.e., how do we have smaller scoped projects in ODL (rgoulding, 16:57:28)
    20. then something that would have ended up in yangtools/infratutils is to develop it in a consumer, have tests, the API has been used extensively, then you pull it out and it becomes its own small project (rgoulding, 16:58:07)
    21. that could be an ideal solution for diagstatus etc. (rgoulding, 16:58:19)
    22. put into infrautils using JMX API (rgoulding, 17:01:56)
    23. separate bundle or something (rgoulding, 17:02:08)
    24. https://wiki.opendaylight.org/view/ODL_Root_Parent:Mature_Graduation_Proposal (rovarga_, 17:05:23)


Meeting ended at 17:09:22 UTC (full logs).

Action items

  1. (none)


People present (lines said)

  1. rgoulding (70)
  2. odl_meetbot (6)
  3. rovarga_ (2)
  4. vrpolak (1)


Generated by MeetBot 0.1.4.