[openstack-dev] [Neutron][Nova] Neutron mid-cycle summary report
martin.hickey at ie.ibm.com
Mon Aug 29 16:18:04 UTC 2016
Not a bother. Great to have the Neutrinos in Cork ! :-)
From: "Armando M." <armamig at gmail.com>
To: "OpenStack Development Mailing List (not for usage questions)"
<openstack-dev at lists.openstack.org>
Cc: Martin Hickey/Ireland/IBM at IBMIE
Date: 27/08/2016 01:14
Subject: [Neutron][Nova] Neutron mid-cycle summary report
For those of you who couldn't join in person, please find a few notes below
to capture some of the highlights of the event.
I would like to thank everyone one who helped me put this report together,
and everyone who helped make this mid-cycle a fruitful one.
I would also like to thank IBM, and the individual organizers who made
everything go smoothly. In particular Martin, who put up with our moody
requests: thanks Martin!!
Feel free to reach out/add if something is unclear, incorrect or
We touched on these topics (as initially proposed on
Keystone v3 and project-id adoption:
dasm and amotoki have been working to making the Neutron server
process project-id correctly . Looking at the spec , we
are half way through having completed the DB migration, being
Keystone v3 complaint, and having updated the client bindings
HenryG, dougwig and kevinbenton worked out a plan to get
the common_db_mixin into neutron-lib. Because of the risk of
regression, this is being deferred until Ocata opens up.
However, simpler changes like the he model_base move to lib was
agreed on and merged.
A plan to provide test support was discussed. The current
strategy involves providing test base classes in lib (this
reverses the stance conveyed in Austin). The usual steps
involved require to making public the currently private
classes, ensure the lib's copies are up-to-date with core
neutron, and deprecate the ones located in Neutron.
rtheis and armax worked on having networking-ovn test
periodically against neutron-lib [1,2,3].
A tool (tools/migration_report.sh) helps project team determine
the level of dependency they have with Neutron. It should be
improved to report the exact offending imports.
Right now neutron-lib 0.4.0 is released and available in
Objects and hitless upgrades:
Ihar gave the team an overview and status update 
There was a fruitful discussion that hopefully set the way
forward for Ocata. The discussed plan was to start Ocata with
the expectation that no new contract scripts are landing in
Ocata, and to revisit the requirement later if for some reason
we see any issue with applying the requirement in practice.
Some work was done to deliver necessary objects for
push-notifications. Patches up for review. Some review cycles
were spent to work on landing patches moving model definitions
rtheis gave an update to the team on the state of the
transition. Core resources commands are all available through
OSC; QoS, Metering and *-aaS are still not converted.
There is some confusion on how to tackle openstacksdk support.
We discussed the future goal of python binding of Networking
API. OSC uses OpenStack SDK for network commands and Neutron
OSC plugin uses python bindings from python-neutronclient. A
question is to which project developers who add new features
implement, both, openstack SDK or python-neutronclient? There
was no conclusion at the mid-cycle. It is not specific to
neutron. Similar situation can happen for nova, cinder and
other projects and we need to raise it to the community.
Ocata is going to be the first release where the neutronclient
CLI is officially deprecated. It may take us more than the
usual two cycles to remove it altogether, but that's a signal
to developer and users to seriously develop against OSC, and
report bugs against OSC.
Several pending contributions into osc-lib.
An update is available on [1,2]
armax was bug deputy for the week of the mid-cycle; nothing
critical showed up in the gate, however pluggable ipam 
switch merged, which might have some unexpected repercussions
down the road.
A number of bugs older than a year were made expirable .
kevinbenton and armax devised a strategy and started working on
 to ensure DB retriable errors are no longer handled at the
The gate witnessed phantom resets that turned out to be related
kevinbenton, jlibosva, armax, jschwarz talked about the Neutron
job queue configuration; it would be desirable to start gating
on rally, and fullstack jobs but this can be done once Ocata
opens up. Testing L3 HA in the multinode configuration is
something that needs tackling too. Ideally we'd move away from
testing legacy routing in the gate to give resources to other
jlibosva should follow up with patches to make
fullstack/functional job "bullet proof";
armax to make a case for check queue only voting status for
Stadium wide efforts:
There was a discussion around what our API definition is.
Specifically, whether we consider supported queries part of it,
and if so, if we plan to enforce this aspect in the future.
Code inspection showed that it’s hard to determine query
parameters supported by plugins right now. We will look into
non-invasive ways to collect this data in Ocata. Armax to
follow up on , which he is using to break ground.
API reference clean up sprint has been announced and started.
The guideline on the API reference was prepared . Some
lead-by-example patches have been posted and merged .
armax, johnthetubaguy and carl_baldin participated in a couple
of discussions about live migration and multiple port binding.
There is a common understanding of the problem and how to
improve the resiliency of the boot/live migration operation.
Some thoughts are being captured on [1,2]. The plan is to
ensure all the legwork is complete ahead of the summit so that
the team can switch into execution mode during the Ocata
There are areas where Nova talks to Neutron that can be
improved and the team identified/discussed a few, the periodic
polls being an example. Providing the ability to aggregate more
data in a single API was also contemplated.
armax, johnthetubaguy, ihrachys talked about next steps in
order to allow os-vif to access a versioned object of vif
details as stored by Neutron.
carl_baldwin was able to make some progress on  to abort
build if Nova tries to schedule a vm to a segment where the IP
is not usable.
Newton blueprints (in no particular order):
push notifications: discussed future steps on how to implement
compare and swap updates to reduce some of the races
experienced when dealing with the Neutron API.
Spent some time tying up some final loose ends on the
service subnets capability. Patch  was the final one
of the effort. Need to get the release note written and
mlavalle implemented a Generic Resource Pools client 
leveraged in .
carl_baldwin and kevinbenton had a very productive
discussion about how to add standard attributes to the
networksegments table to complete its transition to being
a first class API resource. kevinbenton was able to
provide an example and we're working to wrap that up now
carl_baldwin discussed the patch  to enable creating
and deleting segments with the ML2 plugin with
kevinbenton. This may be the last hurdle for routed
networks. Wrapping this up before feature freeze would be
of great help.
jlibosva and armax talked about how to deal with failures
during OVS trunk setup. That resulted in jlibosva working
on on OVSDB transaction manager to handle nested
armax and tidwellr are working on completing the trunk
kevinbenton is rebasing/working on the linuxbridge driver
OVN driver coming along as well.
Many Neutron resources have status/admin_state_up attributes,
however their meaning is not consistent across. This makes
difficult to address use cases like . We should start
painting a picture of what exactly each attribute mean for each
resource in order to make sure we can approach to cases like
 more informed.
-------------- next part --------------
An HTML attachment was scrubbed...
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 105 bytes
Desc: not available
More information about the OpenStack-dev