kubernetes/kubernetes

Cache NodeInfo for Kubelet admission

Open

#132,858 创建于 2025年7月10日

在 GitHub 查看
 (13 评论) (0 反应) (3 负责人)Go (122,268 star) (43,066 fork)batch import
help wantedkind/cleanupsig/nodetriage/accepted

描述

Constructing a new schedulerframework.NodeInfo is a fairly heavy operation: iterate through every resource in every container 3 times (desired, actual & allocated) in every pod and add them all up. The Kubelet does this on every admission request, but also every time it retries a pending resize.

Fortunately the NodeInfo is designed to be cached, which is what the scheduler does. We should consider caching it in the Kubelet too. To keep the cache up-to-date, it will need to be updated:

  • when a pod is added
  • when a pod's allocation changes (resized)
  • when a pod becomes inactive (see kubelet.filterOutInactivePods)
  • when a pods status resources change (resize is actuated)

/sig node /kind cleanup

/cc @natasha41575

贡献者指南

技术栈
go
领域
backendinfrastructureperformance
议题类型
performance
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
5
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
over 1 week
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
blocked
清晰度议题是否清楚说明期望改动、验收标准和下一步。
clear
前置要求
GoKubernetesKubelet
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
10
研究方向
Examine the current NodeInfo construction in the Kubelet admission path, specifically the schedulerframework.NodeInfo. Look at the kubelet's pod addition, resize, and status change handlers. The scheduler already caches NodeInfo, so study that pattern. Also review kubelet.filterOutInactivePods to understand when pods become inactive. Engage with assignees (natasha41575 and others) to ensure the cache design aligns with ongoing work.
Cache NodeInfo for Kubelet admission · kubernetes/kubernetes#132858 | Good First Issue