Closed pohly closed 3 weeks ago
Is my understanding correct that this change will be usable right after merging, i.e. without a formal kind release?
Is my understanding correct that this change will be usable right after merging, i.e. without a formal kind release?
yes
@pohly @onsi this output looks weird, Will run 2144 of 7407 specs
but it Ran 875 of 7407 Specs
/home/prow/go/src/k8s.io/kubernetes/_output/bin/ginkgo --nodes=25 --poll-progress-after=60m --poll-progress-interval=5m --source-root=/home/prow/go/src/k8s.io/kubernetes/cluster/.. --timeout=24h --flake-attempts=1 --no-color /home/prow/go/src/k8s.io/kubernetes/_output/bin/e2e.test -- --kubeconfig=/root/.kube/kind-test-config --host=https://127.0.0.1:41971/ --provider=skeleton --gce-project= --gce-zone= --gce-region= --gce-multizone=false --gke-cluster= --kube-master= --cluster-tag= --cloud-config-file= --repo-root=/home/prow/go/src/k8s.io/kubernetes/cluster/.. --node-instance-group= --prefix=e2e --network=e2e --node-tag= --master-tag= --docker-config-file= --dns-domain=cluster.local --prepull-images=false --report-complete-ginkgo --report-complete-junit --provider=skeleton --num-nodes=2 --ginkgo.focus=. '--ginkgo.skip=\[Serial\]|\[Slow\]|\[Disruptive\]|\[Flaky\]|\[Feature:.+\]|PodSecurityPolicy|LoadBalancer|load.balancer|Simple.pod.should.support.exec.through.an.HTTP.proxy|subPath.should.support.existing' --ginkgo.label-filter= --report-dir=/logs/artifacts --disable-log-dump=true
Running Suite: Kubernetes e2e suite - /home/prow/go/src/k8s.io/kubernetes/_output/bin
=====================================================================================
Random Seed: 1713281512 - will randomize all specs
Will run 2144 of 7407 specs
Running in parallel across 25 processes
SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS
... snipped ...
Ran 875 of 7407 Specs in 924.114 seconds
SUCCESS! -- 875 Passed | 0 Failed | 0 Pending | 6532 Skipped
@aojea i don’t know if y’all also call Skip()
in some specs at runtime. That would explain the discrepancy.
We do a lot of runtime skipping. I think that explains it.
one last comment and we are good to go https://github.com/kubernetes-sigs/kind/pull/3582/files#r1635882841
/lgtm /approve
/cc: @BenTheElder
[APPROVALNOTIFIER] This PR is APPROVED
This pull-request has been approved by: aojea, pohly
The full list of commands accepted by this bot can be found here.
The pull request process is described here
we should also think in phasing this: 1.31 start the transition to labels and deprecation focus and skip , warn on test-infra to use filters
1.32 phase out and forbid focus and skip in test-infra with a. linter
The label filter query is more expressive (logical operations) and readable (no regexp unless absolutely required). Such a query can be combined with focus + skip, but in practice a single label filter can replace both of those and is easier to understand.
Kubernetes has supported ginkgo v2 and thus --label-filter since v1.25.0. This makes it safe to pass that command line flag unconditionally when invoking the E2E suite. However, actual labels were only added in Kubernetes 1.29.
Note: I have not tested this yet. If we are confident that it'll work, then we can merge and I'll start using it in a small job (most likely DRA) once it is available.