#opendaylight-meeting: tws
Meeting started by colindixon at 18:02:56 UTC
(full logs).
Meeting summary
- agenda bashing (colindixon, 18:03:01)
- https://wiki.opendaylight.org/index.php?title=Tech_Work_Stream:Main&oldid=40524#Upcoming_Meeting_Agendas
the agenda for today (colindixon,
18:04:31)
- today we'll cover boron planning (colindixon,
18:06:02)
- two topics: (i) fast-phased approach for
shorter releases and (ii) advisory group input on boron (colindixon,
18:06:25)
- advisory group input for Boron (colindixon, 18:07:09)
- https://lists.opendaylight.org/pipermail/tsc/2016-January/004565.html
(colindixon,
18:07:36)
- colindixon notes that there are (in his mind) 3
bins of asks: (1) is new features at the ~project-level, (2) another
is ODL-wide issues, (3) is sort of meta-issues around network
architectures etc. (colindixon,
18:15:27)
- examples of (1) are things like BGP add-path
and/or aggregated logs/alrsm (colindixon,
18:15:54)
- examples of (2) are things like [easier]
upgrades between releases (colindixon,
18:17:50)
- (3) is stuff like better defining the role of
ODL as VNF manager vs. controller vs. NFV orchestrator (colindixon,
18:22:02)
- easier upgrades (colindixon, 18:22:11)
- easier upgrades were something that was the top
prioirty coming out of the advisory group (colindixon,
18:22:52)
- specifically, not for third-party apps, but for
the components released as part of a major ODL releases it should be
easier for users to upgrade than what they're currently doing which
is starting from scratch every new release (colindixon,
18:26:48)
- edwarnicke says that he thinks upgrades are
complicated and it's going to take us a few tries to get it right,
thus having a shorter release cycle is likely to give us more tries
to get it right in the same period of time (colindixon,
18:28:07)
- https://docs.google.com/presentation/d/1IKI8DV3oqb1snJLCaK6lLvvCXCnpEolQ6Xp_Qlk6E8M/edit
(edwarnicke,
18:29:09)
- colindixon and rovarga agree with this within
caveats of how fast you can actually go, how much you can actually
do in a given time window, etc. (colindixon,
18:29:55)
- skitt and vishnoianil comment that if it's true
that most users are only upgrading every year (or every other
release) that releasing faster might not help (colindixon,
18:31:10)
- they say this is true both because faster
releases won't help them, but also because debugging upgrades
requires input from users who would also need to test upgrades
faster, which they might or might not (colindixon,
18:32:26)
- uri worries faster releases will potentially
signal instability (colindixon,
18:35:20)
- edwarnicke presents his fast-phased approch
with the diagram about the different branches (see slide 2 of the
linked google doc above) (colindixon,
18:35:53)
- colindixon says other than the raw overhead
(which skitt and rovarga pointed out), there is the issue of how
many differnet branches a project will have to simultaneously
maintain (colindixon,
18:36:56)
- edwarnicke says it's all trade-offs, colindixon
agrees (colindixon,
18:37:07)
- edwarnicke points out the primary advantage of
this approach is that everyone is building against (likely stable)
release artifacts so that people aren't running into
SNAPSHOT-integrated transient breakages, but also don't have to wait
for 6 months for new released artifacts (colindixon,
18:38:40)
- skitt says in the end you'll see a piepline
that is 6 months long, but emits a release every 2 months
(colindixon,
18:39:00)
- edwarnicke, vishnoianil, skitt, etc. talk about
the merits of faster releases, edwarnicke says that he finds that
testing and other things are easier when you do less long releases
more often (colindixon,
18:43:43)
- edwarnicke says there's only so much trouble
developers can get in in 2 months (colindixon,
18:44:06)
- vishnoianil says this would need to support
projects that wouldn't have any changes in a 2 month period, that
should be faciliated (colindixon,
18:44:38)
- planning how fast-phased would work (colindixon, 18:47:31)
- colindixon says one good part of this is that
it starts offf with kernel having a 2 month release, protocols have
a 4 month release, and apps have a 6 month release, which is a
reasonable way to get our feet wet without risking everything
(colindixon,
18:49:43)
- colindixon ask kernel project devs how crazy
this looks to them (colindixon,
18:49:54)
- edwarnicke points out that one advantage would
be that we could maybe eliminate milestones other than cutting RCs
some distance before release, e.g., we no longer need API freeze
since that's really for downstream projects (colindixon,
18:51:17)
- rovarga says that he doesn't see the milestones
as mosty for downstream projects, they're also for internal project
discipline (colindixon,
18:52:32)
- rovarga also says that this needs to be fleshed
out a lot more before we can commit to it (colindixon,
18:52:49)
- zxiiro asks how many branches everyone will
need, edwarnicke, he, and colindixon agree there will be three
branches per project (between stable and integration branches),
that's in addition to branches for older supported releases
(colindixon,
18:54:35)
- rovarga says we'd need to see this diagram for
two full cycles (~12 months) before we'd understand what's going
on (colindixon,
18:56:49)
- skitt, rovarga and others argue that we should
abandon text-named versions, e.g., -Lithium or -Lithium-SR1,
etc. (colindixon,
18:57:27)
- skitt aks if we would consider shipping every 2
months for ourselves, but only some of those are targeted for
"customers" (colindixon,
18:59:19)
- colindixon says he had thought about that too,
which is that we might make only some releases public, e.g., 2 per
year (colindixon,
18:59:52)
- LuisGomez had brought this idea up earlier,
which is how many versions do we support in the past, are we
supporting releases for a year or for N releases, if it's just 2
releases does this mean we only support things for 2 months?
(colindixon,
19:01:29)
- abhijitkumbhare asks how OpenStack does this,
vishnoianil and colindixon say they have shorter project dependency
chains and their APIs are more stable/ossified so they have fewer
issues with integration (colindixon,
19:02:13)
- abhijitkumbhare also asks how we fail out if
this if it turns out to have been a bad idea (colindixon,
19:02:31)
- vishnoianil points out deprecation and removal
will be faster if we keep it the same which could be worse for users
since they'd have to worry about changes every 4 months instead of
every 6-12 in the past (colindixon,
19:03:23)
- edwarnicke says that we can lead by example by
showing how to use integration branches (colindixon,
19:04:25)
- colindixon says that's part of it, wall-clock
times being shortened down by a factor of 3 still matters
(colindixon,
19:04:51)
- next steps (colindixon, 19:06:26)
- colindixon will start a thread on the mailing
list (colindixon,
19:06:38)
- we could delay boron start to get it right and
start it this way (colindixon,
19:06:49)
- we could try to do a shadow version of this
alongside a more "normal" boron release (colindixon,
19:07:15)
- edwarnicke proposes other ideas which is that
we could have just kernel projects try to opt in for boron and what
that mgiht mean (colindixon,
19:09:23)
Meeting ended at 19:09:26 UTC
(full logs).
Action items
- (none)
People present (lines said)
- colindixon (52)
- odl_meetbot (4)
- rovarga (2)
- abhijitkumbhare (2)
- edwarnicke (1)
Generated by MeetBot 0.1.4.