Part of https://github.com/kubernetes/kubernetes/pull/63040 is the
assumption that scheduler cache updates must happen before equivalence
cache updates for any given informer event.
The reason for this is that the equivalence cache implementation checks
the main cache for staleness while holding the equiv. cache write lock.
case 1: If an informer invalidates an equiv. cache entry before the
staleness check, then we know that the main cache update completed.
case 2: If an informer blocks trying to grab the equiv. cache lock, then
invalidation will occur right after the potentially stale update is
written.
This patch adds a note to places where we invalidate the equivalence
cache so that hopefully nobody violates this invariant.
Automatic merge from submit-queue (batch tested with PRs 62937, 63105, 63031, 63174). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Revert "Revert "Revert revert of equivalence class hash calculation i…
…n scheduler""
This reverts commit 4386751b5d.
**What this PR does / why we need it**:
This re-introduces the change from https://github.com/kubernetes/kubernetes/pull/58555 which changes how the scheduler computes equivalence classes of pods. I believe we have fixed the flakiness observed previously (https://github.com/kubernetes/kubernetes/issues/61512, https://github.com/kubernetes/kubernetes/issues/62921). I have run the test in question a few dozen times without a failure.
```bash
make test-integration WHAT="./test/integration/scheduler" KUBE_TEST_ARGS="-run TestPreemptionStarvation" GOFLAGS="-v"
```
/ref https://github.com/kubernetes/kubernetes/issues/58222
**Special notes for your reviewer**:
I had to resolve several merge conflicts. I think I resolved them correctly, but keep an eye out for anything silly.
**Release note**:
```release-note
NONE
```
/sig scheduling
Automatic merge from submit-queue (batch tested with PRs 59592, 62308, 62523, 62635, 62243). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Separate pod priority from preemption
**What this PR does / why we need it**:
Users request to split priority and preemption feature gate so they can use priority separately.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes#62068
**Special notes for your reviewer**:
~~I kept use `ENABLE_POD_PRIORITY` as ENV name for gce cluster scripts for backward compatibility reason. Please let me know if other approach is preffered.~~
~~This is a potential **break change** as existing clusters will be affected, we may need to include this in 1.11 maybe?~~
TODO: update this doc https://kubernetes.io/docs/concepts/configuration/pod-priority-preemption/
[Update] Usage: in config file for scheduler:
```yaml
apiVersion: componentconfig/v1alpha1
kind: KubeSchedulerConfiguration
...
disablePreemption: true
```
**Release note**:
```release-note
Split PodPriority and PodPreemption feature gate
```
Automatic merge from submit-queue (batch tested with PRs 61918, 62180, 62198). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Use provided node object in volume binding predicate
**What this PR does / why we need it**:
Autoscaler creates fake node objects, so we should use the provided node object instead of looking up the node from the informer.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes#62178
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 60919, 60953, 61085, 61083, 60971). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Sched cache resync
**What this PR does / why we need it**: Scheduler cache comparer
A debug tool that collects resources from api server and compares it
with the scheduler cache. It currently only compares the node list, but
it should be easy to extend. The compare is triggered by signal USER2,
by doing
kill -12 ${SCHED_PID}
The compare result goes to scheduler log.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Towards #60860
**Special notes for your reviewer**: @bsalamat
**Release note**:
```release-note
None
```
Pods in scheduler cache contains both the scheduled pods and those not
scheduled yet in scheduling queue. This commit adds the second group of
pods into consideration while comparing the cache.
A debug tool that collects resources from api server and compares it
with the scheduler cache. It currently only compares the node list, but
it should be easy to extend. The compare is triggered by signal USER2,
by doing
kill -12 ${SCHED_PID}
The compare result goes to scheduler log.
Towards #60860
Equivalence cache for CheckNodeConditionPred becomes invalid when
Node.Spec.Unschedulable changes. This can happen even if
Node.Status.Conditions does not change, so move the logic around.
This logic is covered by integration test
"test/integration/scheduler".TestUnschedulableNodes but equivalence
cache is currently skipped when test pods have no OwnerReference.
Add benchmark for equivalence hashing.
Change equivalence hash function.
This changes the equivalence class hashing function to use as inputs all
the Pod fields which are read by FitPredicates. Before we used a
combination of OwnerReference and PersistentVolumeClaim info, which was
a close approximation. The new method ensures that hashing remains
correct regardless of controller behavior.
The PVCSet field can be removed from equivalencePod because it is
implicitly included in the Volume list.
Tests are now broken.
Move equivalence class hash code.
This moves the equivalence hashing code from
algorithm/predicates/utils.go to core/equivalence_cache.go.
In the process, making the hashing function and hashing function factory
both injectable dependencies is removed.
Fix equivalence cache hash tests.
Co-authored-by: Jonathan Basseri <misterikkit@google.com>
Co-authored-by: Harry Zhang <resouer@gmail.com>
Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Fixes volume predicate handler for equiv class
**What this PR does / why we need it**:
Per discussion in #58797 , we are missing some predicate handler in factory.go.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes#58797
Ref #58222
**Special notes for your reviewer**:
Kindly ping @msau42
**Release note**:
```release-note
Fixes volume predicate handler for equiv class
```
Automatic merge from submit-queue (batch tested with PRs 59010, 59212, 59281, 59014, 59297). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Replace nominateNodeName annotation with PodStatus.NominatedNodeName
**What this PR does / why we need it**:
Replaces nominateNodeName annotation with PodStatus.NominatedNodeName in scheudler's logic. We don't expect any logic/behavior changes.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
ref #57471
/sig scheduling
cc: @k82cn @aveshagarwal @resouer
This moves the equivalence hashing code from
algorithm/predicates/utils.go to core/equivalence_cache.go.
In the process, making the hashing function and hashing function factory
both injectable dependencies is removed.
Equivalence cache for CheckNodeConditionPred becomes invalid when
Node.Spec.Unschedulable changes. This can happen even if
Node.Status.Conditions does not change, so move the logic around.
This logic is covered by integration test
"test/integration/scheduler".TestUnschedulableNodes but equivalence
cache is currently skipped when test pods have no OwnerReference.
This moves plugin/pkg/scheduler to pkg/scheduler and
plugin/cmd/kube-scheduler to cmd/kube-scheduler.
Bulk of the work was done with gomvpkg, except for kube-scheduler main
package.
