Skip to content

feat: add leader election function to the scheduler extension - #87

Open
coldzerofear wants to merge 8 commits into
mainfrom
feat/dry-run
Open

feat: add leader election function to the scheduler extension#87
coldzerofear wants to merge 8 commits into
mainfrom
feat/dry-run

Conversation

@coldzerofear

Copy link
Copy Markdown
Owner

No description provided.

coldzerofear and others added 8 commits July 21, 2026 01:24
Cluster Autoscaler PR kubernetes/autoscaler#9786 把 HTTP scheduler
extender 接进了扩容仿真链路,使 CA 能感知 extender 管理的 vGPU 资源。
vgpu-manager 作为标准 extender 正是受益方,但当前 Filter 无法被仿真
安全调用。

核心阻塞:Filter 把「可行性判断」与「乐观预分配提交」耦合在一起,
命中节点后会 PatchPodPreAllocatedMetadata 真改 Pod、写 podLister
缓存并发 Event,而 CA 仿真必须只读且会跨 nodegroup 多轮调用。

本文档给出改造设计:抽出只读的 feasibleNodes() 作为共享可行性核心
(allocator.Allocate 本就在 DeepCopy 上纯计算),提交段内聚为
commitPreAllocation() 仅真实路径调用;新增独立只读端点
/scheduler/filter-dryrun 供 CA 配置 filterVerb 指向,不占串行锁。
scale-from-0 依赖模板节点容量暴露的标准化,划入 Phase 2。

含方案选型对比、详细改动点、CA 侧配置样例、影响面与测试计划。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: coldzerofear <coldzerofear@outlook.com>
Signed-off-by: coldzerofear <coldzerofear@outlook.com>
Signed-off-by: coldzerofear <coldzerofear@outlook.com>
Signed-off-by: coldzerofear <coldzerofear@outlook.com>
Signed-off-by: coldzerofear <coldzerofear@outlook.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant