#opendaylight-meeting: MD-SAL interest call
Meeting started by colindixon at 16:04:24 UTC
(full logs).
Meeting summary
- agenda (colindixon, 16:05:49)
- Stephen Kitt's e-mail on what happened this
week in code (colindixon,
16:06:17)
- Real world NETCONF device in Beryllium
(colindixon,
16:06:26)
- Jersey Upgrade: Bug 4502
odl-tsdr-cassandra-persistence depends on non-odlparent guava
version (colindixon,
16:06:40)
- jersey upgrade (colindixon, 16:06:48)
- rgouiding found a bug with guava/cassadra
versions in TSDR making upgrading jersery hard, he's working on
it (colindixon,
16:07:32)
- version missmatch (on guava) when installing
the feature, may require upgrading the cassandra plugin (colindixon,
16:07:59)
- it's delaying the jersey upgrade (colindixon,
16:08:14)
- stephen kitt's weekly summary of code changes (colindixon, 16:10:57)
- https://lists.opendaylight.org/pipermail/dev/2015-October/001333.html
(colindixon,
16:11:19)
- two things for YANG tools seem maybe
relevant (colindixon,
16:11:37)
- 1. Leaf references no longer have a default
value, as per RFC 6020. (colindixon,
16:11:53)
- 2. Mis-matched module/submodule revisions no
longer cause a NullPointerException (but you should still fix the
revisions!). (colindixon,
16:12:17)
- rovarga says that (1) above was removing
pre-hydrogen code that didn't completely work and thus people
couldn't have been using it (colindixon,
16:13:34)
- real world NETCONF devices aren't mounting in Berryllium (colindixon, 16:15:09)
- ghall says that he's working with a lot of
people who are using ODL to mount NETCONF devices to call RPCs, etc.
and write apps using that (colindixon,
16:15:54)
- ghall is running into issues that real devices
that claim to support NETCONF don't always follow the spec
perfectly, and even when they do sometimes interpret the spec the
same way that ODL does (colindixon,
16:16:30)
- ghall says he'd like to see us be tolerant of
non-strict yang (colindixon,
16:16:45)
- https://lists.opendaylight.org/pipermail/netconf-dev/2015-October/000066.html
thread on the netconf-dev list raising this issue (colindixon,
16:17:12)
- https://bugs.opendaylight.org/show_bug.cgi?id=4492
this is the related bug (colindixon,
16:19:06)
- ghall and maros seem to be saying having a
passthrough to get the data back from a NETCONF device without
parsing it (at least strictly) assuming we can mount it (colindixon,
16:20:05)
- rovarga asks how this would work for things
other than RESTCONF, e.g., reading from Binding Aware types
(colindixon,
16:22:53)
- rovarga notes that there are individual ways to
to fix up individual models and that might be a better overall way
to approach things (colindixon,
16:23:28)
- https://git.opendaylight.org/gerrit/#/c/28511/
a (partial) patch to start fixing some of the issues (colindixon,
16:25:18)
- ghall is looking for a way to mount
semi-compliant devices in the Beryllium time-frame, e.g., we should
try to parse and deliver what data we can (maybe dropping just the
yang which is invalid) (colindixon,
16:26:42)
- ghall says that in this case and many other
cases this is clearly a bug with vendor code, in the idea world we
would file the bug, get the bug fixed, and then end-users would
upgrade before testing or using it with opendaylight (colindixon,
16:29:35)
- ghall says in the real world, that's not going
to happen and we should figure out how much we want to inconvenience
end-users who happen to have devices that aren't totally supporting
NETCONF to the same strictness that OpenDaylight does (colindixon,
16:30:41)
- ghall says in practice this is causing him
additional problems with values on a device being set outside their
range in the YANG file (colindixon,
16:31:15)
- colindixon suggests that a good litmus test
would be to ask "are we unnecessarily punishing end-users for their
vendors poor behavior?" (colindixon,
16:33:37)
- if the answer is yes, then we should do our
best to avoid doing that (colindixon,
16:33:51)
- it seems like at least allowing a user to
choose (globally, per-model, per-something-else?) whether to allow
for sloppy vs. strict enforcement so that they can do what a
manually constructed NETCONF call could do would be a maybe a
reasonable approach (colindixon,
16:37:33)
- ghall notes that right now in his day-to-day
development effort he's able to do things with ncclient and get his
work done there in a way he can't with NETCONF mounts in ODL
(colindixon,
16:38:21)
- the discussion seems to revolve around the
issue that if we allow sloppy mounting for devices, that potentially
risks how apps and users perceive the system by compromising the
strictness which the MD-SAL currently provides (colindixon,
16:39:27)
- ttkacik and rovarga ask if we allow a
get-config to work despite not parsing strictly, will we allow a
edit-config to allow for them to pass the same data back?
(colindixon,
16:41:07)
- ghall says that he doesn't quite understand why
we're even parsing when we make RESTCONF calls to YANG mounts
(colindixon,
16:41:28)
- ttkacik says that's not quite true, you have to
parse it to provide RESTCONF features, e.g., converting to
JSON (colindixon,
16:41:56)
- ghall asks if we were asking for XML and XML
only, could we allow for just a passthrough, ttkacik says yes, but
that might confuse users by not allowing for RESTCONF JSON
(colindixon,
16:42:40)
- colindixon says his take is that we're really
talking about having a way to do a sloppy mount over NETCONF and
provide the information to anyone who wants that it's a sloppy mount
and could provide a reasonable trade-off between device compatibly
and user/app expectations (colindixon,
16:46:23)
- rovarga says that this should all be doable
today in the NETCONF project if we wanted to, he'd want to make sure
that we track what things we've violated in the process of doing a
sloppy mount (colindixon,
16:47:01)
- colindixon disagrees that we necessarily need a
list of violations to have it be useful, a simple sloppy bit would
be enough, but he's willing to agree to disagree (colindixon,
16:50:45)
- ghall says he'd like to work on this at the
hackfest 11/9 and 11/10 (colindixon,
16:52:57)
- rovarga asks if he can move BUG 4492 to the
netconf project instead of yangtools, his take is that the issues
raised can be fixed there (colindixon,
16:53:27)
- ghall says sure "as longs as we're happy
returning a supposedly mounted, but useless NETCONF devices on an
NPE" (colindixon,
16:54:33)
- ghall says basically there are still bugs in
YANG tools to at least provide more meaningful error messages
(colindixon,
16:56:11)
- ACTION: ghall to
rework his patch to allow for better errors and/or better mounting
in YANG tools (colindixon,
16:59:14)
- hackfest (colindixon, 16:59:18)
- ghall asks who from the MD-SAL core team will
be there (colindixon,
16:59:33)
- none planning to come now (colindixon,
16:59:45)
- colindixon asks what would help them attend,
rovarga asks for a concrete agenda and work items (colindixon,
17:00:07)
- ACTION: phrobb to
work with rovarga and ghall and others to develop a better
agenda (colindixon,
17:00:31)
Meeting ended at 17:00:34 UTC
(full logs).
Action items
- ghall to rework his patch to allow for better errors and/or better mounting in YANG tools
- phrobb to work with rovarga and ghall and others to develop a better agenda
People present (lines said)
- colindixon (54)
- odl_meetbot (4)
- phrobb_ (1)
Generated by MeetBot 0.1.4.