- snapshot equivalence cache generation numbers before snapshotting the
scheduler cache
- skip update when generation does not match live generation
- keep the node and increment its generation to invalidate it instead of
deletion
- use predicates order ID as key to improve performance
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>.
Improve scheduler's performance by eliminating sorting of nodes by their score
**What this PR does / why we need it**:
Profiling scheduler, I noticed that scheduler spends a significant amount of time in sorting the nodes after we score them to find nodes with the highest score. Finding nodes with the highest score does not need sorting the array. This PR replaces the sort with a linear scan.
Eliminating the sort results in over 10% improvement in throughput of the scheduler.
Before (3 runs for 5000 nodes, scheduling 1000 pods in a cluster running 2000 pods):
BenchmarkScheduling/5000Nodes/2000Pods-12 1000 20682552 ns/op
BenchmarkScheduling/5000Nodes/2000Pods-12 1000 20464729 ns/op
BenchmarkScheduling/5000Nodes/2000Pods-12 1000 21188906 ns/op
After:
BenchmarkScheduling/5000Nodes/2000Pods-12 1000 18485866 ns/op
BenchmarkScheduling/5000Nodes/2000Pods-12 1000 18457749 ns/op
BenchmarkScheduling/5000Nodes/2000Pods-12 1000 18418200 ns/op
**Release note**:
```release-note
Improve scheduler's performance by eliminating sorting of nodes by their score.
```
This moves the equivalence cache implementation out of the 'core'
package and into k8s.io/kubernetes/pkg/scheduler/core/equivalence.
Separating the equiv. cache from the genericScheduler implementation
make their interaction points easier to follow, and prevents us from
accidentally accessing unexported fields.
Automatic merge from submit-queue (batch tested with PRs 64142, 64426, 62910, 63942, 64548). 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>.
scheduler: further cleanup of equivalence cache
**What this PR does / why we need it**:
This improves comments and simplifies some names/logic in equivalence_cache.go, as well as changing the order of some items in the file.
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
/kind cleanup
Automatic merge from submit-queue (batch tested with PRs 64252, 64307, 64163, 64378, 64179). 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>.
Remove unused parameter (pod) in `pkg/scheduler/core/generic_scheduler`
**What this PR does / why we need it**:
Remove unused parameter (pod) in `pkg/scheduler/core/generic_scheduler`
**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
```
The purpose of this map is to combine two predicate results before
writing to the equivalence cache. However, the branch that combines
results is unreachable.
1. Combining results happens in the second iteration of the outer loop.
2. There is only a second iteration when podsAdded is true.
3. We skip equiv. cache when podsAdded is true.
Because the scheduler takes a snapshot of cache data at the start of
each scheduling cycle, updates to the equivalence cache should be
skipped if there was a cache update during the cycle.
If the current NodeInfo becomes stale while we evaluate predicates, we
will not write any results into the equivalence cache. We will still use
the results for the current scheduling cycle, though.
Automatic merge from submit-queue (batch tested with PRs 60898, 60912, 60753, 61002, 60796). 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 of equivalence class hash calculation in scheduler
**What this PR does / why we need it**:
NOTE: This is a revert revert of https://github.com/kubernetes/kubernetes/pull/58555
But since the original PR has been changed, I have to copy the original changes and resend this new PR. See: https://github.com/kubernetes/kubernetes/pull/58555#issuecomment-364345972
And I kept @misterikkit 's change as the first commit (by co-author feature of github) in the history.
We decide to do revert revert because #58989 has been fixed, which should help to improve the time consumed by integration test.
**But** we should still pay attention to integration tests to see if there's frequent timeout happen.
**Special notes for your reviewer**:
**Release note**:
```release-note
Improve equivalence class hash calculation in scheduler
```
Add verb and preemption for scheduler extender
Update bazel
Use simple preemption in extender
Use node name instead of v1.Node
Fix support method
Fix preemption dup
Remove uneeded logics
Remove nodeInfo from param to extender
Update bazel for scheduler types
Mock extender cache with nodeInfo
Add nodeInfo as extender cache
Choose node name or node based on cache flag
Always return meta victims in result
Automatic merge from submit-queue (batch tested with PRs 60106, 59510, 60263, 60063, 59088). 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>.
Reuse the `min*Nodes` slices in order to save GC time
**What this PR does / why we need it**:
Reuse the `min*Nodes` slices to save GC time when executing `pickOneNodeForPreemption`.
**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#59748
**Special notes for your reviewer**:
**Release note**:
```release-note
None
```
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>.
Format some import statements in scheduler pkg
**What this PR does / why we need it**:
As the title says, apply `goimports` on some files under `pkg/scheduler` pkg.
**Release note**:
```release-note
NONE
```
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>.
Avoid race condition when updating equivalence cache
**What this PR does / why we need it**:
Lock the ecache to update the ecache on each predicate running, to avoid race condition.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fix#58507
**Special notes for your reviewer**:
None
**Release note**:
```release-note
None
```
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.
Wei Huang