Migrating the project’s container images linked via k8s.gcr.io from gcr.io/google-containers to gcr.io/k8s-artifacts-prod has been rolled back because we discovered a number of undocumented internal references the hard way. The infrastructure team will be cleaning things up and retrying this migration sometime next month.
Project Zoom accounts have been locked down more: owners will need to use the Zoom app, and attendees will have to log in if they follow a web link. Please remember to update your Zoom client, and follow the Zoom guidelines.
Next Deadline: Start Enhancements Tracking, April 14
The 1.19 release cycle has officially started, and has a schedule. We start tracking enhancements tommorrow, so update those KEPs; you have until May 5th to figure out what your SIG will work on this cycle.
The Release team is also selecting shadows for the various roles, so if you’ve always wanted to participate, here’s your chance. Deadline Tuesday.
Version 1.18 has its first update, v1.18.1, that fixes a bunch of bugs.
A quiet week in PRs, one big set of improvements this week was to the NodeInfo type and surrounding code. This PR was the big one, a package swap that will need updating in most current scheduler PRs as well as things linking against the scheduler code. This was joined by #90030 which did further cleanups to the API and made it more idiomatic Go.
And then a client-oriented fix, avoiding fallback to full collection list on watch failures if there is a client-side timeout. As a separate server-side fix, #89881 improves the behavior of watches during unplanned Etcd failovers.
- Allow hard-failing if kube-proxy is unable to bind to a port
- Restore the ability to
kubectl apply --prunewithout -n flag
- Fix creating internal Google load balancers on clusters with more than 1000 nodes in a zone
- Dry run support for
Original Source: http://lwkd.info/2020/20200414