[neutron] October 2021 - PTG Summary
katonalala at gmail.com
Mon Oct 25 15:50:09 UTC 2021
Below is my summary of the Neutron sessions during last week's PTG.
Here is the etherpad with the agenda:
# Day 1 (Monday Oct 18.)
## Xena retrospective (I just list here the topics, not every discussion
* Good things:
** Less meetings
** New features
** CI jobs have again meaningful names
** Housekeeping - new payload style callbacks switch finished
** Healthy team - we have people from various companies, doing reviews and
* Bad / not so good / change it
** CI stability could be better
** lack of people involved in the CI meetings and CI issues
** is there a chance to decrease the number of jobs/backends to be
* Actions coming out from the above:
** Having video based team meetings every few weeks.
** Video meeting with screen sharing for CI meeting every few weeks.
** ping people before the meetings
## Should we move some parts of neutron devstack plugin to devstack repo?
Slaweq brought this topic to rationalize devstack plugin code between
neutron and devstack.
As a rule of thumb: "what is used/tested by jobs from other repositories
(like tempest) should be in devstack repo and not in neutron's devstack
The bad part is that we should loose review velocity if there's nobody to
review in devstack.
As a consequence some code should be moved back from devstack to neutron's
devstack plugin (for drivers that are not that common now like linuxbridge)
## ML2/OVS -> ML2/OVN migration - presentation (continued on Friday)
We got an overview from Jakub (jlibosva) of how the migration tool works
now, what is finished and which are the possible directions to improve it.
* Problematic parts:
** steps that require the tool to touch the db (no API currently)
** no rollback
** no CI (at least upstream)
** no devstack based tools (related to the previous one)
** New job for ovs-to-ovn migration
*** Now start with tripleo based job and go for a devstack based one
** Don't create new APIs for job type change and other operations during
the migration as it seems there is no other usecase for them.
** Add developer documentation for the migration (preparation, steps,
** add automatic rollback
# Day 2 (Tuesday Oct. 19.)
## tempest_neutron_plugin - a RedHat downstream tests
We got an overview of these tests from Eduardo Olivares.
* tests requiring advance images
** move them to neutron-tempest-plugin with special tag and run them
** use ubuntu cloud minimal image
* tests requiring multinode topology + capturing traffic on hosts/computes
** These tests use a node to run tempest (tripleo based deployment)
** these tests should be changed to run in upstream CI
* Tests that restart nodes / services / containers
** better not to run them in upstream CI
** check if it is possible to migrate some to tobiko (
* sriov tests (macvtap, direct and direct-physical ports)
** No possibility today to run these upstream.
## Edge session (https://etherpad.opendev.org/p/ecg-ptg-october-2021 )
* Thursday we had a session together with the Designate team where we
discussed possible action items related to edge use cases.
## RBAC together with Nova
* https://etherpad.opendev.org/p/nova-yoga-ptg l108. - l137.
# Day 3 (Wednesday Oct. 20.)
## Job reorganization recap
* We need fresh data to see where we still have pain points.
* Ongoing CI rationalization items:
** Use Ubuntu minimal image where possible.
** Serialize the run of tests with advanced images (reduce memory usage):
## ryu / os-ken
Recently we realized that ryu has some activity to fix bugs that can hit
Neutron as well (as we forked ryu as os-ken).
* regularly checking ryu bugs and fixes (sync on team meeting)
* Do not return to ryu as os-ken is adopted for Openstack workflow with CI,
## Nova-Neutron cross-project session
* OVN live-migration (https://etherpad.opendev.org/p/ovn_live_migration )
* RBAC continued (https://etherpad.opendev.org/p/nova-yoga-ptg l144 - l165)
* libvirt virt driver does not wait for network-vif-plugged event during
** document which backend sends what event and when
** workaround in nova for telling whether we should wait for an event (
** long-term solution has to be Neutron providing event information thru
the port binding information
** Improved testing of move operations
*** Nova will keep one ML2/OVS multinode job running
** Supporting ports with QoS min bw and min pps in multi segment networks
with multiple physnets
*** finish the ongoing QoS min pps support
*** extend placement to support any-trait queries
## Interop with Martin Kopec
* Check which tests should be included to refstack
## Quick introduction of the Tobiko test tool
We got an overview of Tobiko from Frederico Ressi, and how it can be used
to test scenarios which are destructive.
* Neutron already has a periodic job with tobiko.
# Day 4 (Thursday Oct. 21.)
## OVN and BGP
** not just for neutron-dynamic-routing but for networking-bgpvpn as well.
** ovn-bgp-agent is another backend which could be behind
neutron-dynamic-routing, and uses FRR.
** In the future it will be a Neutron stadium project
** need more tests, integration with Neutron and better documentation.
* Make neutron-dynamic-routing work with OVN
** Frickler and bcafarel checked if the current jobs and tests can work
** more investigation is necessary if it works as is now or some change is
** Redhat and Ericsson help in the maintenance of neutron-dynamic-routing.
## Multi segments and provider networks on a host, with 1 subnet per segment
* different from the usual routed network use case where we have one
segment available in each rack
* WIP patch which was never finished:
## Use openstacksdk as client for Heat
* Now Heat uses python-neutronclient for python bindings / client (not the
CLI part which is deprecated but the python bindings)
** We collected a buch of projects over than heat which uses
python-neutronclient: Nova, Horizon, Rally....
*** Write mail about it to have the attention of the community
*** Have a migration plan and document it.
*** Have a slot during the team meeting to track progress.
## Designate session
The topics coming from Designate team were really helpful to translate the
requirements from edge session to action items for Neutron.
** dns resolver is not specified for a subnet, (RFE for OVN as it works
only for OVS currently)
** initiate a cross-project documentation effort for setting up an edge site
** initiate cross-project job for testing an edge site
# Day 5 (Friday Oct. 22.)
## Continued discussion of neutron-dynamic-routing's future
* Who will review new features (Both Redhat and Ericsson will increase
review attention for this project)
* Fixing the scheduler:
** New scheduler whith which the user can select which host/dragent to
* OVN and neutron-dynamic-routing
** add the missing features to the OVN gap list (
## RBAC discussion continued with TC
## ML2/OVS -> ML2/OVN migration - continue
We continued the discussion from Monday, on how to integrate the ovs-ovn
migration tool to Neutron CI.
## Team photo:
I would like to thank you for the great discussions during last week, it
was really useful and productive week. Based on these we have some plan
what we will work on in the Yoga cycle.
Lajos Katona (lajoskatona)
Ericsson Software Technologies
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the openstack-discuss