If a CRI error occurs during the terminating phase after a pod is
force deleted (API or static) then the housekeeping loop will not
deliver updates to the pod worker which prevents the pod's state
machine from progressing. The pod will remain in the terminating
phase but no further attempts to terminate or cleanup will occur
until the kubelet is restarted.
The pod worker now maintains a store of the pods state that it is
attempting to reconcile and uses that to resync unknown pods when
SyncKnownPods() is invoked, so that failures in sync methods for
unknown pods no longer hang forever.
The pod worker's store tracks desired updates and the last update
applied on podSyncStatuses. Each goroutine now synchronizes to
acquire the next work item, context, and whether the pod can start.
This synchronization moves the pending update to the stored last
update, which will ensure third parties accessing pod worker state
don't see updates before the pod worker begins synchronizing them.
As a consequence, the update channel becomes a simple notifier
(struct{}) so that SyncKnownPods can coordinate with the pod worker
to create a synthetic pending update for unknown pods (i.e. no one
besides the pod worker has data about those pods). Otherwise the
pending update info would be hidden inside the channel.
In order to properly track pending updates, we have to be very
careful not to mix RunningPods (which are calculated from the
container runtime and are missing all spec info) and config-
sourced pods. Update the pod worker to avoid using ToAPIPod()
and instead require the pod worker to directly use
update.Options.Pod or update.Options.RunningPod for the
correct methods. Add a new SyncTerminatingRuntimePod to prevent
accidental invocations of runtime only pod data.
Finally, fix SyncKnownPods to replay the last valid update for
undesired pods which drives the pod state machine towards
termination, and alter HandlePodCleanups to:
- terminate runtime pods that aren't known to the pod worker
- launch admitted pods that aren't known to the pod worker
Any started pods receive a replay until they reach the finished
state, and then are removed from the pod worker. When a desired
pod is detected as not being in the worker, the usual cause is
that the pod was deleted and recreated with the same UID (almost
always a static pod since API UID reuse is statistically
unlikely). This simplifies the previous restartable pod support.
We are careful to filter for active pods (those not already
terminal or those which have been previously rejected by
admission). We also force a refresh of the runtime cache to
ensure we don't see an older version of the state.
Future changes will allow other components that need to view the
pod worker's actual state (not the desired state the podManager
represents) to retrieve that info from the pod worker.
Several bugs in pod lifecycle have been undetectable at runtime
because the kubelet does not clearly describe the number of pods
in use. To better report, add the following metrics:
kubelet_desired_pods: Pods the pod manager sees
kubelet_active_pods: "Admitted" pods that gate new pods
kubelet_mirror_pods: Mirror pods the kubelet is tracking
kubelet_working_pods: Breakdown of pods from the last sync in
each phase, orphaned state, and static or not
kubelet_restarted_pods_total: A counter for pods that saw a
CREATE before the previous pod with the same UID was finished
kubelet_orphaned_runtime_pods_total: A counter for pods detected
at runtime that were not known to the kubelet. Will be
populated at Kubelet startup and should never be incremented
after.
Add a metric check to our e2e tests that verifies the values are
captured correctly during a serial test, and then verify them in
detail in unit tests.
Adds 23 series to the kubelet /metrics endpoint.
1. Scheduler bug-fix + scheduler-focussed E2E tests
2. Add cgroup v2 support for in-place pod resize
3. Enable full E2E pod resize test for containerd>=1.6.9 and EventedPLEG related changes.
Co-Authored-By: Vinay Kulkarni <vskibum@gmail.com>
1. Core Kubelet changes to implement In-place Pod Vertical Scaling.
2. E2E tests for In-place Pod Vertical Scaling.
3. Refactor kubelet code and add missing tests (Derek's kubelet review)
4. Add a new hash over container fields without Resources field to allow feature gate toggling without restarting containers not using the feature.
5. Fix corner-case where resize A->B->A gets ignored
6. Add cgroup v2 support to pod resize E2E test.
KEP: /enhancements/keps/sig-node/1287-in-place-update-pod-resources
Co-authored-by: Chen Wang <Chen.Wang1@ibm.com>
Currently, there are some unit tests that are failing on Windows due to
various reasons:
- On Windows, consecutive time.Now() calls may return the same timestamp, which would cause
the TestFreeSpaceRemoveByLeastRecentlyUsed test to flake.
- tests in kuberuntime_container_windows_test.go fail on Nodes that have fewer than 3 CPUs,
expecting the CPU max set to be more than 100% of available CPUs, which is not possible.
- calls in summary_windows_test.go are missing context.
- filterTerminatedContainerInfoAndAssembleByPodCgroupKey will filter and group container
information by the Pod cgroup key, if it exists. However, we don't have cgroups on Windows,
thus we can't make the same assertions.
Added new gRPC call 'ListPodSanboxMetrics' which would return additional
container stats currently supported by cAdvisor, but outside the scope
of /stats/summary api. Added new types to support metric exporting of
prometheus, including Metric and other subfields. Added fake runtime
changes associated with the CRI changes.
The path module has a few different functions:
Clean, Split, Join, Ext, Dir, Base, IsAbs. These functions do not
take into account the OS-specific path separator, meaning that they
won't behave as intended on Windows.
For example, Dir is supposed to return all but the last element of the
path. For the path "C:\some\dir\somewhere", it is supposed to return
"C:\some\dir\", however, it returns ".".
Instead of these functions, the ones in filepath should be used instead.
Currently, there are some unit tests that are failing on Windows due to
various reasons:
- config options not supported on Windows.
- files not closed, which means that they cannot be removed / renamed.
- paths not properly joined (filepath.Join should be used).
- time.Now() is not as precise on Windows, which means that 2
consecutive calls may return the same timestamp.
- different error messages on Windows.
- files have \r\n line endings on Windows.
- /tmp directory being used, which might not exist on Windows. Instead,
the OS-specific Temp directory should be used.
- the default value for Kubelet's EvictionHard field was containing
OS-specific fields. This is now moved, the field is now set during
Kubelet's initialization, after the config file is read.
Align the behavior of HTTP-based lifecycle handlers and HTTP-based
probers, converging on the probers implementation. This fixes multiple
deficiencies in the current implementation of lifecycle handlers
surrounding what functionality is available.
The functionality is gated by the features.ConsistentHTTPGetHandlers feature gate.
Some of the unit tests cannot pass on Windows due to various reasons:
- fsnotify does not have a Windows implementation.
- Proxy Mode IPVS not supported on Windows.
- Seccomp not supported on Windows.
- VolumeMode=Block is not supported on Windows.
- iSCSI volumes are mounted differently on Windows, and iscsiadm is a
Linux utility.
There is a corner case when blocking Pod termination via a lifecycle
preStop hook, for example by using this StateFulSet:
```yaml
apiVersion: apps/v1
kind: StatefulSet
metadata:
name: web
spec:
selector:
matchLabels:
app: ubi
serviceName: "ubi"
replicas: 1
template:
metadata:
labels:
app: ubi
spec:
terminationGracePeriodSeconds: 1000
containers:
- name: ubi
image: ubuntu:22.04
command: ['sh', '-c', 'echo The app is running! && sleep 360000']
ports:
- containerPort: 80
name: web
lifecycle:
preStop:
exec:
command:
- /bin/sh
- -c
- 'echo aaa; trap : TERM INT; sleep infinity & wait'
```
After creation, downscaling, forced deletion and upscaling of the
replica like this:
```
> kubectl apply -f sts.yml
> kubectl scale sts web --replicas=0
> kubectl delete pod web-0 --grace-period=0 --force
> kubectl scale sts web --replicas=1
```
We will end up having two pods running by the container runtime, while
the API only reports one:
```
> kubectl get pods
NAME READY STATUS RESTARTS AGE
web-0 1/1 Running 0 92s
```
```
> sudo crictl pods
POD ID CREATED STATE NAME NAMESPACE ATTEMPT RUNTIME
e05bb7dbb7e44 12 minutes ago Ready web-0 default 0 (default)
d90088614c73b 12 minutes ago Ready web-0 default 0 (default)
```
When now running `kubectl exec -it web-0 -- ps -ef`, there is a random chance that we hit the wrong
container reporting the lifecycle command `/bin/sh -c echo aaa; trap : TERM INT; sleep infinity & wait`.
This is caused by the container lookup via its name (and no podUID) at:
02109414e8/pkg/kubelet/kubelet_pods.go (L1905-L1914)
And more specifiy by the conversion of the pod result map to a slice in `GetPods`:
02109414e8/pkg/kubelet/kuberuntime/kuberuntime_manager.go (L407-L411)
We now solve that unexpected behavior by tracking the creation time of
the pod and sorting the result based on that. This will cause to always
match the most recently created pod.
Signed-off-by: Sascha Grunert <sgrunert@redhat.com>
