#opendaylight-meeting: kernel projects
Meeting started by rgoulding at 16:58:43 UTC
(full logs).
Meeting summary
- agenda bashing (rgoulding, 16:58:53)
- https://wiki.opendaylight.org/view/ODL_Root_Parent:BigBumpOfJan2018
(rgoulding,
16:59:00)
- AsyncWriteTransaction submit() with
@CheckReturnValue (rgoulding,
16:59:07)
- BigBumpOfJan2018 (rgoulding, 17:04:07)
- https://wiki.opendaylight.org/view/ODL_Root_Parent:BigBumpOfJan2018
(rgoulding,
17:04:17)
- kernel patches have been updated already
(rgoulding,
17:07:17)
- many of the others including netvirt,
infrautils, genius, etc. have been updated to use 3.0.2 (rgoulding,
17:07:32)
- more testing is needed (rgoulding,
17:07:37)
- other patches need to be ammended to utilize
3.0.2 instead of 3.0.2-SNAPSHOT (rgoulding,
17:08:26)
- skitt, if CSIT passes and results aren’t too
scary, perhaps that is enough to get releng to merge them?
(rgoulding,
17:10:56)
- the initial plan was to kick everyone out and
then bring back in (rgoulding,
17:11:08)
- Thursday TSC call will make the call whether to
go/no-go (rgoulding,
17:11:16)
- rovarga says it might be better to just do it
one by one (rgoulding,
17:12:22)
- i.e., wait until one merge job finishes and
publishes artifacts, then recheck the next, and so-on (rgoulding,
17:12:42)
- Will 3.0.2 be the final version? (rgoulding,
17:18:41)
- the opinion is that this is the version we will
release with unless we find something that requries a new
revision (rgoulding,
17:18:56)
- if we do need a new version, we should be able
to do this easier by just bumping the version to 3.0.3 in
integration/distribution and inheriting the newer version in the
runtime since we use version ranges (rgoulding,
17:19:46)
- rovarga if this transition is smooth and done
in a reasonable timeframe, then we can discuss whether it makes
sense to roll out 2.0.1 of yangtools to get the fixes in that
(rgoulding,
17:21:01)
- if not, we will just punt to SR1 and be
done (rgoulding,
17:21:19)
- vorburger states that the timeline is
tight (rgoulding,
17:31:44)
- LuisGomez says that there will need to be a
patch to remove non-tested patches (rgoulding,
17:32:21)
- vorburger is there a fallback plan?
(rgoulding,
17:32:35)
- skitt the fallback would be odlparent 2.0.6 and
what would be the case in yangtools? (rgoulding,
17:32:54)
- rovarga only option would be to use yangtools
1.2.1 which was used in Nitrogen SR1 (rgoulding,
17:33:14)
- skitt what would the consequences of that be to
end users in oxygen? (rgoulding,
17:33:36)
- rovarga there is a clustering bug reported in
live environments that could not be fixed without yangtools
2.0.0 (rgoulding,
17:34:00)
- related to running clusters across
timezones (rgoulding,
17:34:07)
- it would be painful for downstreams
(rgoulding,
17:34:18)
- the expectation is that fallout from this will
be minor, since a lot of unit tests are passing (rgoulding,
17:35:33)
- vorburger states that it now takes much much
longer to make changes in odlparent now to get changes in
(rgoulding,
17:37:50)
- rovarga you either get isolation or
integratability (rgoulding,
17:38:02)
- there isn’t really a middle ground (rgoulding,
17:38:07)
- rovarga at some point the core needs to stop
churning releases and pushing on downstreams to track them
(rgoulding,
17:38:55)
- vorburger doesn’t have a solution persay, but
he does wnat to point out that it takes a long time to fix even
small things. Things have slowed down to integrate things in
odlparent (rgoulding,
17:39:23)
- vorburger asks if there are better
solutions? (rgoulding,
17:39:31)
- skitt there is friction that arises since
upgrades all have different constraints. for example, we could fix
bugs quickly in karaf and only bump affected downstream projects.
But since we have everything bunched together, it is more involved
to adapt. The advantage to version ranges is that we could achieve
a lot of this much more quickly (rgoulding,
17:41:37)
- perhaps we will need to split odlparent up
somehow (rgoulding,
17:41:45)
- skitt regarding development velocity there is
some rule / best practice (written down somewhere) stating we
develop things in a downstream and then once it becomes more useful
we push it up. if you stick to that then assuming nothing upstream
blocks you, then you can still have good velcoity. Starting things
where they will ultimately end up (i.e., odlparent), that will slow
you down. (rgoulding,
17:43:14)
- the idea is to start local, then move
upwards (rgoulding,
17:43:33)
- rovarga not sure if we will be able to make
version ranges completely work (rgoulding,
17:43:59)
- rovaraga what if we bump to karaf to 4.1.4 in
odlparent 3.0.3, which also bumps jetty then what happens in the
various situations? (rgoulding,
17:44:25)
- i.e., jetty causes incompatibilities but karaf
doesnt (rgoulding,
17:44:33)
- other projects have stated they arent really
doing semver (netty) (rgoulding,
17:45:03)
- some upstreams don’t necessarily comply to
non-breaking API changes, whcih makes it particularly hard to
upgrade (rgoulding,
17:45:36)
- there is a lot of work that could be done on
the odlparent side that could be done to pre-validate, but that
might be more useful in int/dist (rgoulding,
17:45:51)
- skitt for checkstyle, we should have checkstyle
snippets to get to the point where only certain parts can be
enforced in downstreams as opposted to all or none (rgoulding,
17:46:25)
- skitt temporarily, we could have different
projects using different versions of odlparent (rgoulding,
17:46:48)
- i.e., if there is somethign missing in the
karaf configuration (rgoulding,
17:47:09)
- skitt points out that this means being very
careful about what is merged to odlparent (rgoulding,
17:47:43)
- rovarga, which runs contratry to quicker
velocity desires (rgoulding,
17:47:59)
- run continuous deployment will allow us to see
an automated scripted release with each patch. We don’t have all
the scripting or build infra to get there yet. (rgoulding,
17:48:38)
- the distribution-check does some of this, but
it depends on autorelease (rgoulding,
17:49:06)
- LuisGomez something we are missing for sure is
an autorelease job that builds all the projects with latest version
of odlparent/yangtools to produce some distribution that we can pass
to CSIT (rgoulding,
17:52:21)
- https://jira.opendaylight.org/projects/RELENG/issues/RELENG-69
(LuisGomez,
17:52:43)
- rovarga it should be spinning a release for
each project in the dependency order, and running CSIT against the
staged artifacts, and if the projects CSIT passes, then it should be
good to release (rgoulding,
17:53:19)
- rovarga the thing is that if it is autorelease,
then it takes half a day and any sort of glitch or failure is a show
stopper for seeing useful results (rgoulding,
17:54:25)
- LuisGomez this may be too expensive though (per
project) (rgoulding,
17:56:09)
- LuisGomez there will be a lot of infra load.
Maybe it makes more sense to do it with one particular case (i.e.,
netvirt which has a lot of CSIT) then we have some confidence of
whether the change is good (rgoulding,
17:56:55)
- AsyncWriteTransaction submit() with @CheckReturnValue (rgoulding, 17:57:46)
- https://git.opendaylight.org/gerrit/#/c/66360/
(rgoulding,
17:57:59)
- https://git.opendaylight.org/gerrit/#/c/66361/
(rgoulding,
17:58:11)
- tpantelis this is an API change and we are
beyond API freeze now (rgoulding,
17:58:45)
- API freeze is not only an ODL internal thing,
but also the downstreams that we don’t know about (rgoulding,
17:59:47)
- while it gives us a feeling of what it will do
andaffect in our own release, it doesn’t address the fact that we
dont’ know what is happening downstream (rgoulding,
18:00:06)
- slf4j code quality checks with checkstyle-odl-checks VS findbugs-slf4j.. because of Eclipse IDE integration issue, I propose to: (rgoulding, 18:02:04)
- https://git.opendaylight.org/gerrit/#/c/66449/
(rgoulding,
18:02:16)
- compare the checks that are provided by
both (rgoulding,
18:03:37)
- rovarga doesn’t want to get rid of the
checkstyle ones if they add anything on top of findbugs one
(rgoulding,
18:04:56)
- we can keep in cs too (rgoulding,
18:05:02)
- vorburger is much less motivated to get the CS
patches in (rgoulding,
18:05:16)
- FYI error-prone now adopted in infrautils, with custom tuned config. If time, may try to use infrautils parent e.g. in genius. (rgoulding, 18:07:25)
- vorburger spent time to tune a config for this
and has switched infrautils to enforce (rgoulding,
18:07:54)
- rovarga enabling error-prone breaks some valid
code (rgoulding,
18:10:12)
- rovarga something with type
inferences/generics (rgoulding,
18:10:20)
- vorburger this project is very active and what
might have broken in the past might be fixed now (rgoulding,
18:10:41)
Meeting ended at 18:13:03 UTC
(full logs).
Action items
- (none)
People present (lines said)
- rgoulding (78)
- odl_meetbot (3)
- LuisGomez (1)
- vorburger (1)
Generated by MeetBot 0.1.4.