=========================================
#opendaylight-group-policy: GBP datastore
=========================================


Meeting started by alagalah at 14:02:29 UTC.  The full logs are
available at
http://meetings.opendaylight.org/opendaylight-group-policy/2014/gbp_datastore/opendaylight-group-policy-gbp_datastore.2014-05-19-14.02.log.html
.



Meeting summary
---------------

* agenda bashing  (regXboi, 14:03:18)
  * only agenda item I have is new DC numbers at link below  (regXboi,
    14:03:36)
  * LINK: https://wiki.opendaylight.org/view/Group_Policy:Scaling
    (regXboi, 14:03:46)
  * LINK:
    https://wiki.opendaylight.org/view/OpenDaylight_Controller:MD-SAL:Architecture:Clustered_Data_Store
    (raghu67, 14:05:26)
  * raghu67 asks about what we're calling resolvers in the use cases
    (tbachman, 14:09:36)
  * regXboi says the renderer takes all the particulars of the model,
    and renders it into the particular configuration language needed by
    the resolver  (tbachman, 14:10:45)
  * the resolve operation should be a local operation in most cases.
    (tbachman, 14:10:55)
  * the only time it isn't local is when the operation is completely out
    of scope, in which case it goes back to the renderer for resolution.
    If not resolvable by the renderer, this leads to an exception path
    (tbachman, 14:11:36)
  * regXboi believes we need the inventory of resolvers for the
    exception path -- hopes it's part of ODL's inventory.  (tbachman,
    14:12:18)
  * This leads to the question of how big the inventory in ODL can
    be/gets  (tbachman, 14:12:37)
  * Moiz continues data store presentation  (tbachman, 14:22:15)
  * regXboi wants to make sure that we have architectural white space
    for auditing  (tbachman, 14:22:39)
  * Moiz says we can either use logging or a separate monitoring
    application for this.  (tbachman, 14:23:02)
  * regXboi claims that logging is not auditing  (tbachman, 14:23:21)
  * in the case of audit, want to know who did it, what authority that
    had to do it, and arguments provided so that a reply can be
    performed.  (tbachman, 14:24:52)
  * jumping down to call diagrams  (regXboi, 14:28:12)
  * LINK:
    https://wiki.opendaylight.org/view/OpenDaylight_Controller:MD-SAL:Architecture:Clustered_Data_Store#Creating_a_new_transaction
    (regXboi, 14:28:36)
  * raghu67 asks whether we need to support distributed transactions
    across shards. alagalah says he doesn't believe we do.  (tbachman,
    14:31:45)
  * alagalah asks whether the operations are synchronous for all shards
    (tbachman, 14:32:09)
  * Moiz says that it's asynchronous, with the "ask" getting a future
    which can be called later  (tbachman, 14:32:51)
  * readams notes that there currently aren't any semantics for things
    like rollbacks in the current data store APIs  (tbachman, 14:35:13)
  * readams noticed that the change notifications API might be a bit
    broken, when considering a multi-thread context. For example, are
    notifications guaranteed to be dispatched from a single thread?
    (tbachman, 14:41:05)
  * raghu67 notes that some interfaces pass entire subtrees, which can
    be expensive.  (tbachman, 14:41:43)
  * current commits are from a single thread, which may create
    performance issues later on  (regXboi, 14:41:59)
  * point that failed transaction objects are expected to be thrown away
    and a new transaction object allocated  (regXboi, 14:42:32)
  * discussion of better documentation and examples (and fixed code) and
    possibly fix API...  (regXboi, 14:50:22)
  * no meeting next week due to the holidy  (tbachman, 15:02:48)



Meeting ended at 15:04:53 UTC.



People present (lines said)
---------------------------

* tbachman (19)
* regXboi (12)
* alagalah (6)
* odl_meetbot (4)
* raghu67 (2)



Generated by `MeetBot`_ 0.1.4