GangCache Design

.github/workflows/ci.yaml

@@ -117,8 +117,7 @@ jobs:
           platforms: linux/amd64
           push: true
           pull: true
-          build-args: |
-            MODULE=${{ matrix.target }}
+          file: docker/${{ matrix.target }}.dockerfile
           labels: |
             org.opencontainers.image.title=${{ matrix.target }}
             org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }}

.goreleaser.yaml

@@ -5,13 +5,13 @@ before:
   hooks:
     - go mod tidy
 builds:
-  - env:
+  - id: koord-runtime-proxy
+    env:
       - CGO_ENABLED=0
     goos:
       - linux
     goarch:
       - amd64
-    id: koord-runtime-proxy
     main: ./cmd/koord-runtime-proxy
     binary: koord-runtime-proxy
     ldflags:
@@ -20,13 +20,13 @@ builds:
       - -X github.com/koordinator-sh/koordinator/pkg/version.buildDate={{ .Date }}
       - -X github.com/koordinator-sh/koordinator/pkg/version.gitCommit={{ .Commit }}
       - -X github.com/koordinator-sh/koordinator/pkg/version.gitTreeState=clean
-  - env:
+  - id: koord-manager
+    env:
       - CGO_ENABLED=0
     goos:
       - linux
     goarch:
       - amd64
-    id: koord-manager
     main: ./cmd/koord-manager
     binary: koord-manager
     ldflags:
@@ -35,13 +35,13 @@ builds:
       - -X github.com/koordinator-sh/koordinator/pkg/version.buildDate={{ .Date }}
       - -X github.com/koordinator-sh/koordinator/pkg/version.gitCommit={{ .Commit }}
       - -X github.com/koordinator-sh/koordinator/pkg/version.gitTreeState=clean
-  - env:
+  - id: koord-scheduler
+    env:
       - CGO_ENABLED=0
     goos:
       - linux
     goarch:
       - amd64
-    id: koord-scheduler
     main: ./cmd/koord-scheduler
     binary: koord-scheduler
     ldflags:
@@ -50,13 +50,13 @@ builds:
       - -X github.com/koordinator-sh/koordinator/pkg/version.buildDate={{ .Date }}
       - -X github.com/koordinator-sh/koordinator/pkg/version.gitCommit={{ .Commit }}
       - -X github.com/koordinator-sh/koordinator/pkg/version.gitTreeState=clean
-  - env:
-      - CGO_ENABLED=0
+  - id: koordlet
+    env:
+      - CGO_ENABLED=1
     goos:
       - linux
     goarch:
       - amd64
-    id: koordlet
     main: ./cmd/koordlet
     binary: koordlet
     ldflags:
@@ -87,10 +87,11 @@ changelog:
       - '^chore:'
       - '^feat(deps):'
 dockers:
-  - image_templates:
+  - id: koord-manager
+    image_templates:
       - "ghcr.io/{{.ProjectName}}/koord-manager:{{ .Version }}"
       - "registry.cn-beijing.aliyuncs.com/{{.ProjectName}}/koord-manager:{{ .Version }}"
-    dockerfile: .goreleaser/Dockerfile
+    dockerfile: .goreleaser/koord-manager.dockerfile
     build_flag_templates:
       - "--pull"
       - "--label=org.opencontainers.image.title=koord-manager"
@@ -99,16 +100,15 @@ dockers:
       - "--label=org.opencontainers.image.revision={{.FullCommit}}"
       - "--label=org.opencontainers.image.version={{.Version}}"
       - "--label=org.opencontainers.image.licenses=Apache-2.0"
-      - "--build-arg=MODULE=koord-manager"
-    id: koord-manager
     ids:
       - koord-manager
     goos: linux
     goarch: amd64
-  - image_templates:
+  - id: koordlet
+    image_templates:
       - "ghcr.io/{{.ProjectName}}/koordlet:{{ .Version }}"
       - "registry.cn-beijing.aliyuncs.com/{{.ProjectName}}/koordlet:{{ .Version }}"
-    dockerfile: .goreleaser/Dockerfile
+    dockerfile: .goreleaser/koordlet.dockerfile
     build_flag_templates:
       - "--pull"
       - "--label=org.opencontainers.image.title=koordlet"
@@ -117,16 +117,15 @@ dockers:
       - "--label=org.opencontainers.image.revision={{.FullCommit}}"
       - "--label=org.opencontainers.image.version={{.Version}}"
       - "--label=org.opencontainers.image.licenses=Apache-2.0"
-      - "--build-arg=MODULE=koordlet"
-    id: koordlet
     ids:
       - koordlet
     goos: linux
     goarch: amd64
-  - image_templates:
+  - id: koord-scheduler
+    image_templates:
       - "ghcr.io/{{.ProjectName}}/koord-scheduler:{{ .Version }}"
       - "registry.cn-beijing.aliyuncs.com/{{.ProjectName}}/koord-scheduler:{{ .Version }}"
-    dockerfile: .goreleaser/Dockerfile
+    dockerfile: .goreleaser/koord-scheduler.dockerfile
     build_flag_templates:
       - "--pull"
       - "--label=org.opencontainers.image.title=koord-scheduler"
@@ -135,8 +134,6 @@ dockers:
       - "--label=org.opencontainers.image.revision={{.FullCommit}}"
       - "--label=org.opencontainers.image.version={{.Version}}"
       - "--label=org.opencontainers.image.licenses=Apache-2.0"
-      - "--build-arg=MODULE=koord-scheduler"
-    id: koord-scheduler
     ids:
       - koord-scheduler
     goos: linux

.goreleaser/koord-manager.dockerfile

@@ -0,0 +1,4 @@
+FROM gcr.io/distroless/static:latest
+WORKDIR /
+COPY koord-manager .
+ENTRYPOINT ["/koord-manager"]

.goreleaser/koord-scheduler.dockerfile

@@ -0,0 +1,4 @@
+FROM gcr.io/distroless/static:latest
+WORKDIR /
+COPY koord-scheduler .
+ENTRYPOINT ["/koord-scheduler"]

.goreleaser/koordlet.dockerfile

@@ -0,0 +1,4 @@
+FROM nvidia/cuda:11.6.1-base-ubuntu20.04
+WORKDIR /
+COPY koordlet .
+ENTRYPOINT ["/koordlet"]

Makefile

@@ -114,15 +114,15 @@ docker-build: test docker-build-koordlet docker-build-koord-manager docker-build
 
 .PHONY: docker-build-koordlet
 docker-build-koordlet: ## Build docker image with the koordlet.
-	docker build --build-arg MODULE=koordlet -t ${KOORDLET_IMG} .
+	docker build --pull -t ${KOORDLET_IMG} -f docker/koordlet.dockerfile .
 
 .PHONY: docker-build-koord-manager
 docker-build-koord-manager: ## Build docker image with the koord-manager.
-	docker build --build-arg MODULE=koord-manager -t ${KOORD_MANAGER_IMG} .
+	docker build --pull -t ${KOORD_MANAGER_IMG} -f docker/koord-manager.dockerfile .
 
 .PHONY: docker-build-koord-scheduler
 docker-build-koord-scheduler: ## Build docker image with the scheduler.
-	docker build --build-arg MODULE=koord-scheduler -t ${KOORD_SCHEDULER_IMG} .
+	docker build --pull -t ${KOORD_SCHEDULER_IMG} -f docker/koord-scheduler.dockerfile .
 
 .PHONY: docker-push
 docker-push: docker-push-koordlet docker-push-koord-manager docker-push-koord-scheduler

README-zh_CN.md

@@ -0,0 +1,65 @@
+<h1 align="center">
+  <p align="center">Koordinator</p>
+  <a href="https://koordinator.sh"><img src="https://github.com/koordinator-sh/koordinator/raw/main/docs/images/koordinator-logo.jpeg" alt="Koordinator"></a>
+</h1>
+
+[![License](https://img.shields.io/github/license/koordinator-sh/koordinator.svg?color=4EB1BA&style=flat-square)](https://opensource.org/licenses/Apache-2.0)
+[![GitHub release](https://img.shields.io/github/v/release/koordinator-sh/koordinator.svg?style=flat-square)](https://github.com/koordinator-sh/koordinator/releases/latest)
+[![CI](https://img.shields.io/github/workflow/status/koordinator-sh/koordinator/CI?label=CI&logo=github&style=flat-square)](https://github.com/koordinator-sh/koordinator/actions/workflows/ci.yaml)
+[![Go Report Card](https://goreportcard.com/badge/github.com/koordinator-sh/koordinator?style=flat-square)](https://goreportcard.com/report/github.com/koordinator-sh/koordinator)
+[![codecov](https://img.shields.io/codecov/c/github/koordinator-sh/koordinator?logo=codecov&style=flat-square)](https://codecov.io/github/koordinator-sh/koordinator)
+[![PRs Welcome](https://badgen.net/badge/PRs/welcome/green?icon=https://api.iconify.design/octicon:git-pull-request.svg?color=white&style=flat-square)](CONTRIBUTING.md)
+[![Slack](https://badgen.net/badge/slack/join/4A154B?icon=slack&style=flat-square)](https://join.slack.com/t/koordinator-sh/shared_invite/zt-1756qoub4-Cn4~esfdlfAPsD7cwO2NzA)
+
+
+[English](./README.md) | 简体中文
+
+
+
+## 介绍
+
+Koordinator 基于 QoS 机制，支持 Kubernetes 上多种工作负载的混部调度。它旨在提高工作负载的运行时效率和可靠性（包括延迟敏感型负载和批处理任务），简化资源相关的配置调优，增加 Pod 部署密度以提高资源利用率。
+
+Koordinator 通过提供如下功能来增强用户在 Kubernetes 上管理工作负载的体验：
+
+- 精心设计的 Priority 和 QoS 机制，支持在一个集群或者一个节点上混部不同的工作负载。
+- 采用应用画像机制（application profiling mechanism），支持超卖资源，以实现在满足 QoS 保障的前提下实现高资源利用率。
+- 细粒度的资源编排和隔离机制以提高工作负载（包括延迟敏感型负载和批处理任务）的效率。
+- 灵活的任务调度机制以支持特定领域（如大数据、AI、音频和视频）的工作负载。
+- 一套支持监控、故障排除、运维的工具集。
+
+## 快速开始
+
+你可以在 [Koordinator website](https://koordinator.sh/docs) 查看到完整的文档集。
+
+- 安装/升级 Koordinator [最新版本](https://koordinator.sh/docs/installation)
+- 参考[最佳实践](https://koordinator.sh/docs/best-practices/colocation-of-spark-jobs)，里面有一些关于运行混部工作负载的示例。
+
+## 行为守则
+
+Koordinator 社区遵照[行为守则](CODE_OF_CONDUCT.md)。我们鼓励每个人在参与之前先读一下它。
+
+为了营造一个开放和热情的环境，我们作为贡献者和维护者承诺：无论年龄、体型、残疾、种族、经验水平、教育程度、社会经济地位、国籍、个人外貌、种族、宗教或性认同和性取向如何，参与我们的项目和社区的每个人都不会受到骚扰。
+
+## 贡献
+
+我们非常欢迎每一位社区同学共同参与 Koordinator 的建设，你可以从 [CONTRIBUTING.md](CONTRIBUTING.md) 手册开始。
+
+## 成员
+
+我们鼓励所有贡献者成为成员。我们的目标是发展一个由贡献者、审阅者和代码所有者组成的活跃、健康的社区。在我们的[社区成员](docs/community/community-membership.md)页面，详细了解我们的成员要求和责任。
+
+## 社区
+
+你可以通过如下途径联系到项目维护者：
+
+- [Slack](https://join.slack.com/t/koordinator-sh/shared_invite/zt-1756qoub4-Cn4~esfdlfAPsD7cwO2NzA)
+- 钉钉( Chinese ): 搜索群ID `33383887`或者扫描二维码加入
+
+<div>
+  <img src="https://github.com/koordinator-sh/koordinator/raw/main/docs/images/dingtalk.png" width="300" alt="Dingtalk QRCode">
+</div>
+
+## License
+
+Koordinator is licensed under the Apache License, Version 2.0. See [LICENSE](./LICENSE) for the full license text.

README.md

@@ -11,6 +11,7 @@
 [![PRs Welcome](https://badgen.net/badge/PRs/welcome/green?icon=https://api.iconify.design/octicon:git-pull-request.svg?color=white&style=flat-square)](CONTRIBUTING.md)
 [![Slack](https://badgen.net/badge/slack/join/4A154B?icon=slack&style=flat-square)](https://join.slack.com/t/koordinator-sh/shared_invite/zt-1756qoub4-Cn4~esfdlfAPsD7cwO2NzA)
 
+English | [简体中文](./README-zh_CN.md)
 ## Introduction
 
 Koordinator is a QoS based scheduling system for hybrid orchestration workloads on Kubernetes. It aims to improve the
@@ -42,7 +43,7 @@ before participating.
 
 In the interest of fostering an open and welcoming environment, we as contributors and maintainers pledge to making
 participation in our project and our community a harassment-free experience for everyone, regardless of age, body size,
-disability, ethnicity, gender identity and expression, level of experience, education, socio-economic status,
+disability, ethnicity, level of experience, education, socio-economic status,
 nationality, personal appearance, race, religion, or sexual identity and orientation.
 
 ## Contributing

apis/extension/node.go

@@ -20,6 +20,8 @@ import (
 	"encoding/json"
 
 	"k8s.io/apimachinery/pkg/types"
+
+	schedulingconfig "github.com/koordinator-sh/koordinator/apis/scheduling/config"
 )
 
 const (
@@ -30,6 +32,22 @@ const (
 	// AnnotationNodeCPUSharedPools describes the CPU Shared Pool defined by Koordinator.
 	// The shared pool is mainly used by Koordinator LS Pods or K8s Burstable Pods.
 	AnnotationNodeCPUSharedPools = NodeDomainPrefix + "/cpu-shared-pools"
+
+	// LabelNodeCPUBindPolicy constrains how to bind CPU logical CPUs when scheduling.
+	LabelNodeCPUBindPolicy = NodeDomainPrefix + "/cpu-bind-policy"
+	// LabelNodeNUMAAllocateStrategy indicates how to choose satisfied NUMA Nodes when scheduling.
+	LabelNodeNUMAAllocateStrategy = NodeDomainPrefix + "/numa-allocate-strategy"
+)
+
+const (
+	// NodeCPUBindPolicyFullPCPUsOnly requires that the scheduler must allocate full physical cores.
+	// Equivalent to kubelet CPU manager policy option full-pcpus-only=true.
+	NodeCPUBindPolicyFullPCPUsOnly = "FullPCPUsOnly"
+)
+
+const (
+	NodeNUMAAllocateStrategyLeastAllocated = string(schedulingconfig.NUMALeastAllocated)
+	NodeNUMAAllocateStrategyMostAllocated  = string(schedulingconfig.NUMAMostAllocated)
 )
 
 type CPUTopology struct {
@@ -77,3 +95,16 @@ func GetPodCPUAllocs(annotations map[string]string) (PodCPUAllocs, error) {
 	}
 	return allocs, nil
 }
+
+func GetNodeCPUSharePools(nodeTopoAnnotations map[string]string) ([]CPUSharedPool, error) {
+	var cpuSharePools []CPUSharedPool
+	data, ok := nodeTopoAnnotations[AnnotationNodeCPUSharedPools]
+	if !ok {
+		return cpuSharePools, nil
+	}
+	err := json.Unmarshal([]byte(data), &cpuSharePools)
+	if err != nil {
+		return nil, err
+	}
+	return cpuSharePools, nil
+}

apis/extension/pod.go

@@ -27,7 +27,7 @@ import (
 const (
 	AnnotationPodCPUBurst = DomainPrefix + "cpuBurst"
 
-	AnnotationPodMemoryQoS = DomainPrefix + "memoryQoS"
+	AnnotationPodMemoryQoS = DomainPrefix + "memoryQOS"
 )
 
 func GetPodCPUBurstConfig(pod *corev1.Pod) (*slov1aplhpa1.CPUBurstConfig, error) {
@@ -47,15 +47,15 @@ func GetPodCPUBurstConfig(pod *corev1.Pod) (*slov1aplhpa1.CPUBurstConfig, error)
 	return &cpuBurst, nil
 }
 
-func GetPodMemoryQoSConfig(pod *corev1.Pod) (*slov1aplhpa1.PodMemoryQoSConfig, error) {
+func GetPodMemoryQoSConfig(pod *corev1.Pod) (*slov1aplhpa1.PodMemoryQOSConfig, error) {
 	if pod == nil || pod.Annotations == nil {
 		return nil, nil
 	}
 	value, exist := pod.Annotations[AnnotationPodMemoryQoS]
 	if !exist {
 		return nil, nil
 	}
-	cfg := slov1aplhpa1.PodMemoryQoSConfig{}
+	cfg := slov1aplhpa1.PodMemoryQOSConfig{}
 	err := json.Unmarshal([]byte(value), &cfg)
 	if err != nil {
 		return nil, err