Kubernetes Test-Infra has updated their docs, particularly around writing and maintaining test jobs, such as Prow Job Examples and Prow Configuration. At this point, all new jobs should be implemented via Prow annotations instead of updating config.yaml. The easy legacy jobs have already been migrated by the team.
Next Deadline: Enhancements Tracking, Alpha.1 July 16th
Enhancements tracking starts today, so at this point any feature updates you have should be reflected in a KEP update and in the Enhancements repo. Use this time to figure out what you’re going to complete before Code Freeze on Aug. 29th. Enhancements freeze will be July 30th.
Patch release 1.15.1 will come out on July 18th, following the patch releases for the other branches that came out July 8th. Note that 1.12.10, which was released 7/8, was the last patch release for version 1.12; if you are on Kubernetes 1.12, you need to upgrade to a later version soon, or find a vendor willing to batchpatch fixes for you.
Etcd2 support is long since removed, and so now the official Etcd images will no longer include the 2->3 migration tools. If you still need these tools, you can use an older version of the image or run them some other way. This is unlikely to catch most dev clusters but if you still have Etcd2 in a dark corner somewhere, no time like the present to upgrade it.
Admission webhooks have graduated to v1! And with this comes a bunch of changes from beta to GA, mostly changes to the default values of fields to encourage safer usage of webhooks. To shamelessly quote from the release notes:
failurePolicydefault changed from
matchPolicydefault changed from
timeoutdefault changed from
sideEffectsdefault value is removed and the field made required for v1.
admissionReviewVersionsdefault value is removed and the field made required for v1 (supported versions for AdmissionReview are
namefield for specified webhooks must be unique for
ValidatingWebhookConfigurationobjects created via
The v1beta1 versions will still be available until 1.19 as per normal deprecation cycles.
A small change but should be a nice performance improvement for the scheduler, using the new Event API. This should be a transparent change, unless you have custom integrations with the scheduler subsystem.
And finally our awesome code of conduct team has finalized their official group charter. This formalizes the mission and scope of the committee as well as detailing how it operates. I would strongly encourage all community members to give it a quick read so they are familiarized with this team.
We had an extremely busy week:
- Out-of-tree Scheduler Plugins are now a thing! This implements part of the scheduling framework
- Update go to 1.12.7, which as usual has the potential to break a lot of stuff. Watch your testgrids!
- e2e tests will stop using global command line flags so that jobs and suites can decide which flags to use
- Stop using HTTP compression when it’s not helpful
- Kubeadm adds support for Dual Network
- IPVS has increased the log level for its talky termination messages
upgrade diffcontacts the original cluster
- APIVersion gets added to all internal authorization requests, as it always should have been
- SIG-scalability created a microbenchmark for pod deployment
- API Aggregation Server will stop pinging the REST service root if it’s not used
- The sample-apiserver template includes an OpenAPI spec generator
- Add delete and replace to
- Switch the namespace controller to the metadata client
- Stop asking for AWS instances by name, use ID instead
- HPA can scale to/from zero pods
- APIserver gets a shutdown delay
- Be smarter about picking a Windows homedir
kubectl -wadds an event type column