kubernetes

mirror of https://github.com/k3s-io/kubernetes.git synced 2026-07-17 02:06:23 +00:00

Files

Ed Bartosh 891a905031 test/localupcluster: stop all components before starting replacements

Modify() was replacing components one at a time: stop X, start X, stop Y,
start Y, ... in version-skew order (apiserver last on downgrade). This
caused a crash during downgrade: KCM-1.35 started against the still-
running apiserver-1.36, passed its /healthz, and then immediately lose
its connection when apiserver-1.36 was killed by the localupcluster.
KCM-1.35 would reconnect to the not-yet-ready apiserver-1.35, hit a
403 RBAC error during controller initialization, and exit — because that
initialization phase does not retry on RBAC errors.

Fix by splitting Modify() into two phases:

  Phase 1 — stop all components to be replaced, in reverse startup order
  (kube-proxy down to apiserver), so dependent components release their
  connections before the apiserver is stopped.

  Phase 2 — start all replacement components in standard startup order
  (apiserver first), so each component connects to a fully-ready apiserver.

2026-04-23 08:27:36 +02:00

coredra_test.go

Add partitionable devices upgrade/downgrade test.

2026-03-09 14:09:36 +00:00

devicetaints_test.go

DRA E2E: disarm footgun in PodExternal

2026-02-17 16:35:36 +01:00

extendedresources_test.go

Merge pull request #136655 from bitoku/partitionable-devices

2026-03-09 23:55:21 +05:30

partitionabledevices_test.go

Add partitionable devices upgrade/downgrade test.

2026-03-09 14:09:36 +00:00

README.md

DRA upgrade/downgrade: rewrite as Go unit test

2026-01-16 07:54:51 +01:00

resourceclaimstatus_test.go

DRA upgrade/downgrade: rewrite as Go unit test

2026-01-16 07:54:51 +01:00

run.sh

…

upgradedowngrade_test.go

test/localupcluster: stop all components before starting replacements

2026-04-23 08:27:36 +02:00

README.md

This directory contains a testsuite with automatic upgrade/downgrade tests for DRA. Conceptually this is like an integration test, in the sense that it starts/stops cluster components and runs tests against them. It has its own directory because it needs to be started differently than other integration tests or unit tests, which makes it more like an E2E suite.

The difference is that it starts Kubernetes components by running the actual binaries, relying on local-up-cluster.sh for the logic and configuration steps. local-up-cluster.sh needs additional permissions and preparations on the host.

To run it:

Make sure that hack/local-up-cluster.sh works:
- sudo must work
- Set env variables as necessary for your environment.
Ensure that /var/lib/kubelet/plugins, /var/lib/kubelet/plugins_registry, and /var/run/cdi are writable.
Build binaries with make.
Export KUBERNETES_SERVER_BIN_DIR=$(pwd)/_output/local/bin/linux/amd64 (or whatever is your GOOS/GOARCH and output directory).
Optional: export KUBERNETES_SERVER_CACHE_DIR=$(pwd)/_output/local/bin/linx/amd64/cache-dir to reuse downloaded release binaries across test invocations.
Optional: set ARTIFACTS to store component log files persistently. Otherwise a test tmp directory is used.

Invoke as a Go test (no need for the ginkgo CLI), for example:

  go test -v -count=1 -timeout=1h ./test/e2e_dra
  dlv test ./test/e2e_dra -- -test.v
  make test KUBE_TIMEOUT=-timeout=1h WHAT=test/e2e_dra FULL_LOG=true KUBE_TEST_ARGS="-count=1"

make test instead of make test-integration is intentional: local-up-cluster.sh itself wants to start etcd. -count=1 ensures that test runs each time it is invoked. -v/-test.v/FULL_LOG=true make the test output visible while the test runs.

To simplify starting from scratch, ./test/e2e_dra/run.sh cleans up, sets permissions, and then invokes whatever command is specified on the command line:

 ./test/e2e_dra/run.sh go test ./test/e2e_dra