mirror of
https://github.com/k3s-io/kubernetes.git
synced 2026-01-24 05:59:02 +00:00
3eadd1a9ea
A number of race conditions exist when pods are terminated early in
their lifecycle because components in the kubelet need to know "no
running containers" or "containers can't be started from now on" but
were relying on outdated state.
Only the pod worker knows whether containers are being started for
a given pod, which is required to know when a pod is "terminated"
(no running containers, none coming). Move that responsibility and
podKiller function into the pod workers, and have everything that
was killing the pod go into the UpdatePod loop. Split syncPod into
three phases - setup, terminate containers, and cleanup pod - and
have transitions between those methods be visible to other
components. After this change, to kill a pod you tell the pod worker
to UpdatePod({UpdateType: SyncPodKill, Pod: pod}).
Several places in the kubelet were incorrect about whether they
were handling terminating (should stop running, might have
containers) or terminated (no running containers) pods. The pod worker
exposes methods that allow other loops to know when to set up or tear
down resources based on the state of the pod - these methods remove
the possibility of race conditions by ensuring a single component is
responsible for knowing each pod's allowed state and other components
simply delegate to checking whether they are in the window by UID.
Removing containers now no longer blocks final pod deletion in the
API server and are handled as background cleanup. Node shutdown
no longer marks pods as failed as they can be restarted in the
next step.
See https://docs.google.com/document/d/1Pic5TPntdJnYfIpBeZndDelM-AbS4FN9H2GTLFhoJ04/edit# for details
test/e2e
This is home to e2e tests used for presubmit, periodic, and postsubmit jobs.
Some of these jobs are merge-blocking, some are release-blocking.
e2e test ownership
All e2e tests must adhere to the following policies:
- the test must be owned by one and only one SIG
- the test must live in/underneath a sig-owned package matching pattern:
test/e2e/[{subpath}/]{sig}/..., e.g.test/e2e/auth- all tests owned by sig-authtest/e2e/common/storage- all testscommonto cluster-level and node-level e2e tests, owned by sig-nodetest/e2e/upgrade/apps- all tests used inupgradetesting, owned by sig-apps
- each sig-owned package should have an OWNERS file defining relevant approvers and labels for the owning sig, e.g.
# test/e2e/node/OWNERS
# See the OWNERS docs at https://go.k8s.io/owners
approvers:
- alice
- bob
- cynthia
emeritus_approvers:
- dave
reviewers:
- sig-node-reviewers
labels:
- sig/node
- packages that use
{subpath}should have animports.gofile importing sig-owned packages (for ginkgo's benefit), e.g.
// test/e2e/common/imports.go
package common
import (
// ensure these packages are scanned by ginkgo for e2e tests
_ "k8s.io/kubernetes/test/e2e/common/network"
_ "k8s.io/kubernetes/test/e2e/common/node"
_ "k8s.io/kubernetes/test/e2e/common/storage"
)
- test ownership must be declared via a top-level SIGDescribe call defined in the sig-owned package, e.g.
// test/e2e/lifecycle/framework.go
package lifecycle
import "github.com/onsi/ginkgo"
// SIGDescribe annotates the test with the SIG label.
func SIGDescribe(text string, body func()) bool {
return ginkgo.Describe("[sig-cluster-lifecycle] "+text, body)
}
// test/e2e/lifecycle/bootstrap/bootstrap_signer.go
package bootstrap
import (
"github.com/onsi/ginkgo"
"k8s.io/kubernetes/test/e2e/lifecycle"
)
var _ = lifecycle.SIGDescribe("[Feature:BootstrapTokens]", func() {
/* ... */
ginkgo.It("should sign the new added bootstrap tokens", func() {
/* ... */
})
/* etc */
})
These polices are enforced:
- via the merge-blocking presubmit job
pull-kubernetes-verify - which ends up running
hack/verify-e2e-test-ownership.sh - which can also be run via
make verify WHAT=e2e-test-ownership