diff options
| author | Kubernetes Submit Queue <k8s-merge-robot@users.noreply.github.com> | 2017-10-06 13:05:39 -0700 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2017-10-06 13:05:39 -0700 |
| commit | 9979d3736ebd90ea694a0dafb7c659cfabb3fad9 (patch) | |
| tree | 4093368f22e52608ff837315f546346a1ba57555 | |
| parent | aab1ceb21501beef8b692ac265dfc65e8150e354 (diff) | |
| parent | 4b70550e0ea44b3c1e7fd2c8bf7bcfbbb55e3319 (diff) | |
Merge pull request #1162 from dashpole/priority_eviction
Automatic merge from submit-queue.
Update priority eviction docs
After discussion with @dchen1107 and @bsalamat I think it would be simpler to start with a tiered eviction sorting rather than by using a function. See https://github.com/kubernetes/community/pull/846#pullrequestreview-66920328 for some of the rationale.
This PR makes two changes:
1. Changes the release at which changes take effect to 1.9 (since implementation missed 1.8)
2. Changes the strategy from (usage > requests, func(priority, usage - requests)) to (usage > requests, priority, usage - requests)
cc @dchen1107 @derekwaynecarr @vishh
| -rw-r--r-- | contributors/design-proposals/node/kubelet-eviction.md | 15 |
1 files changed, 6 insertions, 9 deletions
diff --git a/contributors/design-proposals/node/kubelet-eviction.md b/contributors/design-proposals/node/kubelet-eviction.md index 4dd78861..0b627b97 100644 --- a/contributors/design-proposals/node/kubelet-eviction.md +++ b/contributors/design-proposals/node/kubelet-eviction.md @@ -248,15 +248,12 @@ Priority, and are the largest consumers of the starved resource relative to their scheduling request. It will target pods whose usage of the starved resource exceeds its requests. -Of those pods, it will rank by a function of priority, and usage - requests. -Roughly speaking, if a pod has twice the priority of another pod, it will -recieve half the penalty for usage above requests. If system daemons are -exceeding their allocation (see [Strategy Caveat](strategy-caveat) below), -and all pods are using less than their requests, then it will evict a pod -whose usage is less than requests, based on the function of priority, and -usage - requests. - -Prior to v1.8: +Of those pods, it will rank by priority, then usage - requests. If system +daemons are exceeding their allocation (see [Strategy Caveat](strategy-caveat) below), +and all pods are using less than their requests, then it must evict a pod +whose usage is less than requests, based on priority, then usage - requests. + +Prior to v1.9: The `kubelet` will implement a default eviction strategy oriented around the pod quality of service class. |