[kuryr][ptg] Yoga cycle PTG summary

Maysa De Macedo Souza mdemaced at redhat.com
Sat Oct 30 12:05:57 UTC 2021


I would like to thank you for the great discussion and engagement during
the PTG sessions.
It was a really productive week. Here follows the summary of the Kuryr
sessions during the PTG:

*Day 1 - Tuesday*
Etherpad: https://etherpad.opendev.org/p/kuryr-yoga-ptg

   - Xena retrospective:
      - We celebrated the contributions from many Outreachy applicants and
      the interns.
         - The addition of new features, such as:
         - The Reconciliation mechanisms that enforces the matching of a
         Kubernetes Service with a load-balancer.
         - The setting of Octavia Listener timeout based on annotation of
         the Kubernetes Service.
      - Improvements on the logs to facilitate debuggability and
      - Usage of Kubeadm to facilitate configuration of Kubernetes cluster
      with Devstack.
      - Many documentation improvements.

   - Stable branches:
      - We have many stable branches which are not highly maintained anymore
            - action items: EOL the branches till Stein
         - As we currently release with OpenStack, it was proposed that we
         release with Kubernetes
            - action items: Keep following the OpenStack release, but make
            sure that when there is a Kubernetes release it's updated in Kuryr.

   - Future of the CI:
      - action items:
         - Move all the gates to use CRI-O and have one with Docker.
         - Attempt to remove the Amphora dependency for the API
         load-balancer when OVN Octavia driver is available.
         - Move the jobs that use OVN to voting.

   - Rate limit requests:
      - Possibility to limit the requests to Octavia during the
      reconciliation between load-balancers and Kubernetes Service .
      - Possibility to limit the amount of requests sent to Neutron

*Day 2 - Wednesday*
Etherpad: https://etherpad.opendev.org/p/kuryr-yoga-ptg

   - Merge of Kuryr lib into Kuryr-kubernetes:
      - action items: analyze the feasibility of moving the needed pieces
      into Kuryr-kubernetes and sync with other projects that use it.

   - Different log levels:
      - action items:
         - clean up some unnecessary debugging messages and move the level
         of some to warning or info.

   - Feasibility of reducing the current approach of one Network per
   Namespace, to one Network per cluster to reduce resource usage:
      - action items:
      - check if there is a plan to implement Network cascade delete with
         the Neutron team
         -  attempt to create a PoC with one Pod Network per Cluster

   - Possibility of creating external reachable services without Amphora.
   - Discussion about the open Kubernetes proposal of notifying the user
   when a Network Policy is actually enforced.
   - Drop the liveness checks for certain scenarios:
      - The controller and CNI restarts has helped with certain cornes
      cases, but it can decrease the availability of Kuryr if
restarted many times
         - action items:
            - drop restarts on unavailability of Neutron/k8s API
            - enforce restarts when all the Kuryr watches, handlers and
            drivers are not running

   - Drop usage of Flask debug servers
      - Currently many servers created by Kuryr like Health, Prometheus
      exporter and Kuryr daemon are based on flask and could likely be moved to
      another type of server.
         - action item: Move the kuryr-cni to kuryr-daemon communication to
         use gRPC

Don't exitate to contact us either on IRC or email if you have any

Maysa Macedo.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20211030/5d76958e/attachment.htm>

More information about the openstack-discuss mailing list