This commit makes the job controller re-honor exponential backoff for
failed pods. Before this commit, the controller created pods without any
backoff. This is a regression because the controller used to
create pods with an exponential backoff delay before (10s, 20s, 40s ...).
The issue occurs only when the JobTrackingWithFinalizers feature is
enabled (which is enabled by default right now). With this feature, we
get an extra pod update event when the finalizer of a failed pod is
removed.
Note that the pod failure detection and new pod creation happen in the
same reconcile loop so the 2nd pod is created immediately after the 1st
pod fails. The backoff is only applied on 2nd pod failure, which means
that the 3rd pod created 10s after the 2nd pod, 4th pod is created 20s
after the 3rd pod and so on.
This commit fixes a few bugs:
1. Right now, each time `uncounted != nil` and the job does not see a
_new_ failure, `forget` is set to true and the job is removed from the
queue. Which means that this condition is also triggered each time the
finalizer for a failed pod is removed and `NumRequeues` is reset, which
results in a backoff of 0s.
2. Updates `updatePod` to only apply backoff when we see a particular
pod failed for the first time. This is necessary to ensure that the
controller does not apply backoff when it sees a pod update event
for finalizer removal of a failed pod.
3. If `JobsReadyPods` feature is enabled and backoff is 0s, the job is
now enqueued after `podUpdateBatchPeriod` seconds, instead of 0s. The
unit test for this check also had a few bugs:
- `DefaultJobBackOff` is overwritten to 0 in certain unit tests,
which meant that `DefaultJobBackOff` was considered to be 0,
effectively not running any meaningful checks.
- `JobsReadyPods` was not enabled for test cases that ran tests
which required the feature gate to be enabled.
- The check for expected and actual backoff had incorrect
calculations.
The fake client doesn't guarantee that the informer cache is updated.
If it's not up-to-date, the controller always tries to set the
StartTime, leading to a broken test.
Change-Id: I71f26d46ea44beff88f0d03517985348654aec95
* Wait for Pods to finish before considering Failed
Limit behavior to feature gates PodDisruptionConditions and
JobPodFailurePolicy and jobs with a podFailurePolicy.
Change-Id: I926391cc2521b389c8e52962afb0d4a6a845ab8f
* Remove check for unsheduled terminating pod
Change-Id: I3dc05bb4ea3738604f01bf8cb5fc8cc0f6ea54ec
in case its status is False or Unknown.
In case the status of the pre-existing condition is true we ignore the new
condition. If there is no pre-existing failed condition, then append
the new failed condition as before.
Also, make the condition comparisons less hacky by ignoring timestamp fields
in tests.
In some rare race conditions, the job controller might create new pods after the job is declared finished.
Change-Id: I8a00429c8845463259cd7f82bb3c241d0011583c
- Lock feature gate to true and schedule for deletion in 1.26
- Remove checks on feature gate
- Graduate E2E test to Conformance
Change-Id: I6814819d318edaed5c86dae4055f4b050a4d39fd
Add the UIDs of Pods for which we are removing finalizers to an in-memory cache.
The controller removes UIDs from the cache as Pod updates or deletes come in.
This avoids double counting finished Pods when Pod updates arrive after Job status updates.
https://github.com/kubernetes/kubernetes/issues/105200
Doing a GET right before retrying has 2 problems:
- It can masquerade conflicts
- It adds an additional delay
As for retries, we are better of going through the sync backoff.
In the case of conflict, we know that there was a Job update that would trigger another sync, so there is no need to do a rate limited requeue.
When doing partial updates for uncountedTerminatedPods, the controller might have removed UIDs for Pods which still had finalizers.
Also make more space by removing UIDs that don't have finalizers at the beginning of the sync.
Through Job.status.uncountedPodUIDs and a Pod finalizer
An annotation marks if a job should be tracked with new behavior
A separate work queue is used to remove finalizers from orphan pods.
Change-Id: I1862e930257a9d1f7f1b2b0a526ed15bc8c248ad
