上一篇《k8s调度器 kube-scheduler 源码解析》大概介绍一调度器的内容，提到扩展点的插件这个概念，下面我们看看如何开发一个自定义调度器。
本文源码托管在 https://github.com/cfanbo/sample-scheduler。
插件机制
在Kubernetes调度器中，共有两种插件机制，分别为 in-tree 和 out-of-tree。
- In-tree插件（内建插件）：这些插件是作为Kubernetes核心组件的一部分直接编译和交付的。它们与Kubernetes的源代码一起维护，并与Kubernetes版本保持同步。这些插件以静态库形式打包到kube-scheduler二进制文件中，因此在使用时不需要单独安装和配置。一些常见的in-tree插件包括默认的调度算法、Packed Scheduling等。
- Out-of-tree插件（外部插件）：这些插件是作为独立项目开发和维护的，它们与Kubernetes核心代码分开，并且可以单独部署和更新。本质上，out-of-tree插件是基于Kubernetes的调度器扩展点进行开发的。这些插件以独立的二进制文件形式存在，并通过自定义的方式与kube-scheduler进行集成。为了使用out-of-tree插件，您需要单独安装和配置它们，并在kube-scheduler的配置中指定它们。
可以看到 in-tree 插件与Kubernetes的核心代码一起进行维护和发展，而 out-of-tree插件可以单独开发并out-of-tree插件以独立的二进制文件部署。因此 out-of-tree插件具有更大的灵活性，可以根据需求进行自定义和扩展，而 in-tree 插件受限于Kubernetes核心代码的功能和限制。
对于版本升级in-tree插件与Kubernetes版本保持同步，而out-of-tree插件可以单独进行版本升级或兼容。
总的来说，in-tree 插件是Kubernetes的一部分，可以直接使用和部署，而 out-of-tree 插件则提供了更多的灵活性和定制化能力，但需要单独安装和配置。
一般开发都是采用out-of-tree 这各机制。
扩展点
下图显示了一个 Pod 的调度上下文以及调度框架公开的扩展点。
一个插件可以在多个扩展点处注册执行，以执行更复杂或有状态的任务。
[图片上传失败...(image-abf318-1690960748676)]
对于每个扩展点的介绍，可参考上面给出的官方文档，这里不再做介绍。
我们下面开发的是一个 Pod Scheduling Context 部分的 Filter调度器插件，插件的功能非常的简单，就是检查一个 Node 节点是否存在 cpu=true标签，如果存在此标签则可以节点有效，否则节点视为无效，不参与Pod调度。
插件实现
要实现一个调度器插件必须满足两个条件：
- 必须实现对应扩展点插件接口
- 将此插件在调度框架中进行注册。
不同扩展点可以启用不同的插件。
插件实现
每个扩展点的插件都必须要实现其相应的接口，所有的接口定义在 https://github.com/kubernetes/kubernetes/blob/v1.27.3/pkg/scheduler/framework/interface.go。
// Plugin is the parent type for all the scheduling framework plugins.
type Plugin interface {
Name() string
}
// FilterPlugin is an interface for Filter plugins. These plugins are called at the
// filter extension point for filtering out hosts that cannot run a pod.
// This concept used to be called 'predicate' in the original scheduler.
// These plugins should return "Success", "Unschedulable" or "Error" in Status.code.
// However, the scheduler accepts other valid codes as well.
// Anything other than "Success" will lead to exclusion of the given host from
// running the pod.
// 这个是我们要实现的插件
type FilterPlugin interface {
Plugin
// Filter is called by the scheduling framework.
// All FilterPlugins should return "Success" to declare that
// the given node fits the pod. If Filter doesn't return "Success",
// it will return "Unschedulable", "UnschedulableAndUnresolvable" or "Error".
// For the node being evaluated, Filter plugins should look at the passed
// nodeInfo reference for this particular node's information (e.g., pods
// considered to be running on the node) instead of looking it up in the
// NodeInfoSnapshot because we don't guarantee that they will be the same.
// For example, during preemption, we may pass a copy of the original
// nodeInfo object that has some pods removed from it to evaluate the
// possibility of preempting them to schedule the target pod.
Filter(ctx context.Context, state *CycleState, pod *v1.Pod, nodeInfo *NodeInfo) *Status
}
// PreEnqueuePlugin is an interface that must be implemented by "PreEnqueue" plugins.
// These plugins are called prior to adding Pods to activeQ.
// Note: an preEnqueue plugin is expected to be lightweight and efficient, so it's not expected to
// involve expensive calls like accessing external endpoints; otherwise it'd block other
// Pods' enqueuing in event handlers.
type PreEnqueuePlugin interface {
Plugin
// PreEnqueue is called prior to adding Pods to activeQ.
PreEnqueue(ctx context.Context, p *v1.Pod) *Status
}
// LessFunc is the function to sort pod info
type LessFunc func(podInfo1, podInfo2 *QueuedPodInfo) bool
// QueueSortPlugin is an interface that must be implemented by "QueueSort" plugins.
// These plugins are used to sort pods in the scheduling queue. Only one queue sort
// plugin may be enabled at a time.
type QueueSortPlugin interface {
Plugin
// Less are used to sort pods in the scheduling queue.
Less(*QueuedPodInfo, *QueuedPodInfo) bool
}
要实现 PreEnqueue扩展点的插件必须实现 PreEnqueuePlugin接口，而如何实现QueueSort扩展点插件的话，同需要实现 QueueSortPlugin接口，在这里实现 Filter 插件接口。
k8s 默认已有一个调度器 default-scheduler, 现在我们自定义一个调度器 sample-scheduler。
image.png
对于默认调度器它是以 staticPod 方式部署（左图），其yaml定义文件一般为控制面主机的 /etc/kubernetes/manifests/kube-scheduler.yaml 文件 ，而对于自定义调度器一般以Pod的形式部署（右图）。这样一个集群里可以有多个调度器，然后在编写Pod的时候通过 spec.schedulerName 指定当前Pod使用的调度器。
官方给出了一些插件实现的示例，插件编写参考 https://github.com/kubernetes/kubernetes/tree/v1.27.3/pkg/scheduler/framework/plugins/examples
// pkg/plugin/myplugin.go
package plugin
import (
"context"
"k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/runtime"
"k8s.io/kubernetes/pkg/scheduler/framework"
"log"
)
// Name is the name of the plugin used in the plugin registry and configurations.
const Name = "sample"
// Sort is a plugin that implements QoS class based sorting.
type sample struct{}
var _ framework.FilterPlugin = &sample{}
var _ framework.PreScorePlugin = &sample{}
// New initializes a new plugin and returns it.
func New(_ runtime.Object, _ framework.Handle) (framework.Plugin, error) {
return &sample{}, nil
}
// Name returns name of the plugin.
func (pl *sample) Name() string {
return Name
}
func (pl *sample) Filter(ctx context.Context, state *framework.CycleState, pod *v1.Pod, nodeInfo *framework.NodeInfo) *framework.Status {
log.Printf("filter pod: %v, node: %v", pod.Name, nodeInfo)
log.Println(state)
// 排除没有cpu=true标签的节点
if nodeInfo.Node().Labels["cpu"] != "true" {
return framework.NewStatus(framework.Unschedulable, "Node: "+nodeInfo.Node().Name)
}
return framework.NewStatus(framework.Success, "Node: "+nodeInfo.Node().Name)
}
func (pl *sample) PreScore(ctx context.Context, state *framework.CycleState, pod *v1.Pod, nodes []*v1.Node) *framework.Status {
log.Println(nodes)
return framework.NewStatus(framework.Success, "Node: "+pod.Name)
}
实现了Filter 和 PreScore 两类插件，不过本方只演示Filter。
通过 app.NewSchedulerCommand() 注册自定义插件，提供插件的名称和构造函数即可（参考 https://github.com/kubernetes-sigs/scheduler-plugins/blob/master/cmd/scheduler/main.go）。
// cmd/scheduler/main.go 插件调用入口
package main
import (
"os"
"github.com/cfanbo/sample/pkg/plugin"
"k8s.io/kubernetes/cmd/kube-scheduler/app"
)
func main() {
command := app.NewSchedulerCommand(
app.WithPlugin(plugin.Name, plugin.New),
)
if err := command.Execute(); err != nil {
os.Exit(1)
}
}
插件编译
将应用编译成二进制文件(这里是arm64)
➜ GOOS=linux GOARCH=arm64 go build -ldflags '-X k8s.io/component-base/version.gitVersion=$(VERSION) -w' -o bin/sample-scheduler cmd/scheduler/main.go
其命令用法参考 bin/sample-scheduler -h 了解
下面调度器的执行也可以在本机执行，不过由于本机已经有了一个调度器，可能存在一些冲突的情况，这里为了方便直接使用Pod方式进行调度器的部署。
制作镜像
Dockerfile 内容
FROM --platform=$TARGETPLATFORM ubuntu:20.04
WORKDIR .
COPY bin/sample-scheduler /usr/local/bin
CMD ["sample-scheduler"]
制作镜像，参考 https://blog.haohtml.com/archives/31052
在生产中要尽量使用体积最小的基础镜像，这里为了方便直接使用了 ubuntu:20.04 镜像，镜像大小有些大
docker buildx build --platform linux/arm64 -t cfanbo/sample-scheduler:v0.0.1
将生成的镜像上传到远程仓库，以便后面将通过Pod进行部署。
docker push cfanbo/sample-scheduler:v0.0.1
这里环境为 arm64 架构
插件部署
插件功能开发完后，剩下就是如何部署的问题了。
要想让插件运行，必须先将插件在调度框架中进行注册，这个操作是通过编写KubeSchedulerConfiguration 配置文件来定制 kube-scheduler 的操作实现的。
本文我们将调度器插件以Pod 的方式运行。
插件在容器里运行时，需要指定一个调度器配置文件，这个文件内容是一个 KubeSchedulerConfiguration 对象，而这个对象内容我们可以通过 volume 这种方式挂载到容器里，插件运行时指定这个配置文件就可可以了。
首先创建一个 ConfigMap 对象，其内容就是我们需要的 KubeSchedulerConfiguration 配置，在容器里再通过首先通过 volume 来挂载到本地目录，最后通过 --config 来指定这个配置文件。
这里整个配置通过一个yaml 文件实现了，很方便。
# sample-scheduler.yaml
kind: ClusterRole
apiVersion: rbac.authorization.k8s.io/v1
metadata:
name: sample-scheduler-clusterrole
rules:
- apiGroups:
- ""
resources:
- namespaces
verbs:
- create
- get
- list
- apiGroups:
- ""
resources:
- endpoints
- events
verbs:
- create
- get
- update
- apiGroups:
- ""
resources:
- nodes
verbs:
- get
- list
- watch
- apiGroups:
- ""
resources:
- pods
verbs:
- delete
- get
- list
- watch
- update
- apiGroups:
- ""
resources:
- bindings
- pods/binding
verbs:
- create
- apiGroups:
- ""
resources:
- pods/status
verbs:
- patch
- update
- apiGroups:
- ""
resources:
- replicationcontrollers
- services
verbs:
- get
- list
- watch
- apiGroups:
- apps
- extensions
resources:
- replicasets
verbs:
- get
- list
- watch
- apiGroups:
- apps
resources:
- statefulsets
verbs:
- get
- list
- watch
- apiGroups:
- policy
resources:
- poddisruptionbudgets
verbs:
- get
- list
- watch
- apiGroups:
- ""
resources:
- persistentvolumeclaims
- persistentvolumes
verbs:
- get
- list
- watch
- apiGroups:
- ""
resources:
- configmaps
verbs:
- get
- list
- watch
- apiGroups:
- "storage.k8s.io"
resources:
- storageclasses
- csinodes
- csistoragecapacities
- csidrivers
verbs:
- get
- list
- watch
- apiGroups:
- "coordination.k8s.io"
resources:
- leases
verbs:
- create
- get
- list
- update
- apiGroups:
- "events.k8s.io"
resources:
- events
verbs:
- create
- patch
- update
---
apiVersion: v1
kind: ServiceAccount
metadata:
name: sample-scheduler-sa
namespace: kube-system
---
kind: ClusterRoleBinding
apiVersion: rbac.authorization.k8s.io/v1
metadata:
name: sample-scheduler-clusterrolebinding
namespace: kube-system
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: sample-scheduler-clusterrole
subjects:
- kind: ServiceAccount
name: sample-scheduler-sa
namespace: kube-system
---
apiVersion: v1
kind: ConfigMap
metadata:
name: scheduler-config
namespace: kube-system
data:
scheduler-config.yaml: |
apiVersion: kubescheduler.config.k8s.io/v1
kind: KubeSchedulerConfiguration
leaderElection:
leaderElect: false
leaseDuration: 15s
renewDeadline: 10s
resourceName: sample-scheduler
resourceNamespace: kube-system
retryPeriod: 2s
profiles:
- schedulerName: sample-scheduler
plugins:
filter:
enabled:
- name: sample
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: sample-scheduler
namespace: kube-system
labels:
component: sample-scheduler
spec:
selector:
matchLabels:
component: sample-scheduler
template:
metadata:
labels:
component: sample-scheduler
spec:
serviceAccountName: sample-scheduler-sa
priorityClassName: system-cluster-critical
volumes:
- name: scheduler-config
configMap:
name: scheduler-config
containers:
- name: scheduler
image: cfanbo/sample-scheduler:v0.0.1
imagePullPolicy: IfNotPresent
command:
- sample-scheduler
- --config=/etc/kubernetes/scheduler-config.yaml
- --v=3
volumeMounts:
- name: scheduler-config
mountPath: /etc/kubernetes
这个yaml 文件共完成以下几件事：
- 通过
ConfigMap声明一个KubeSchedulerConfiguration配置 - 创建一个
Deployment对象，其中容器镜像cfanbo/sample-scheduler:v0.0.1是前面我们开发的插件应用，对于调度器插件配置通过volume的方式存储到容器里/etc/kubernetes/scheduler-config.yaml，应用启动时指定此配置文件；这里为了调试方便指定了日志--v=3等级 - 创建一个
ClusterRole，指定不同资源的访问权限 - 创建一个
ServiceAccount - 声明一个
ClusterRoleBinding对象，绑定ClusterRole和ServiceAccount两者的关系
安装插件
➜ kubectl apply -f sample-scheduler.yaml</pre>
此时插件以pod的形式运行（命令空间为 `kube-system` ）。
<pre class="EnlighterJSRAW" data-enlighter-language="generic" data-enlighter-theme="" data-enlighter-highlight="" data-enlighter-linenumbers="" data-enlighter-lineoffset="" data-enlighter-title="" data-enlighter-group="">➜ kubectl get pod -n kube-system --selector=component=sample-scheduler
NAME READY STATUS RESTARTS AGE
sample-scheduler-85cd75d775-jq4c7 1/1 Running 0 5m50s
查看pod 容器进程启动参数
➜ kubectl exec -it -n kube-system pod/sample-scheduler-85cd75d775-jq4c7 -- ps -auxww
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
root 1 4.9 0.6 764160 55128 ? Ssl 07:41 0:05 sample-scheduler --config=/etc/kubernetes/scheduler-config.yaml --v=3
root 39 0.0 0.0 5472 2284 pts/0 Rs+ 07:43 0:00 ps -auxww
可以看到插件进程，同时指定了两个参数。
新开一个终端，持续观察调度器插件的输出日志
➜ kubectl logs -n kube-system -f sample-scheduler-85cd75d775-jq4c7
插件测试
正常情况下，这时Pod是无法被正常调度的，因为我们插件对这个调度行为进行了干预。
无法调度
创建一个pod，并指定调度器为 sample-scheduler
# test-scheduler.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: test-scheduler
spec:
selector:
matchLabels:
app: test-scheduler
template:
metadata:
labels:
app: test-scheduler
spec:
schedulerName: sample-scheduler # 指定使用的调度器，不指定使用默认的default-scheduler
containers:
- image: nginx:1.23-alpine
imagePullPolicy: IfNotPresent
name: nginx
ports:
- containerPort: 80
➜ kubectl apply -f test-scheduler.yaml
由于插件主要实现对节点筛选，排除那些不能运行该 Pod 的节点，运行时将检查节点是否存在 cpu=true 标签，如果不存在这个label标签，则说明此节点无法通过预选阶段，后面的调度与绑定步骤就不可能执行。
我们看一下这个Pod能否被调度成功
➜ kubectl get pods --selector=app=test-scheduler
NAME READY STATUS RESTARTS AGE
test-scheduler-78c89768cf-5d9ct 0/1 Pending 0 11m
可以看到一直处于 Pending 状态，说明一直无法被调度，我们再看一下这个Pod描述信息
➜ kubectl describe pod test-scheduler-78c89768cf-5d9ct
Name: test-scheduler-78c89768cf-5d9ct
Namespace: default
Priority: 0
Service Account: default
Node: <none>
Labels: app=test-scheduler
pod-template-hash=78c89768cf
Annotations: <none>
Status: Pending
IP:
IPs: <none>
Controlled By: ReplicaSet/test-scheduler-78c89768cf-5d9ct
Containers:
nginx:
Image: nginx:1.23-alpine
Port: 80/TCP
Host Port: 0/TCP
Environment: <none>
Mounts:
/var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-gptlh (ro)
Volumes:
kube-api-access-gptlh:
Type: Projected (a volume that contains injected data from multiple sources)
TokenExpirationSeconds: 3607
ConfigMapName: kube-root-ca.crt
ConfigMapOptional: <nil>
DownwardAPI: true
QoS Class: BestEffort
Node-Selectors: <none>
Tolerations: node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
Events: <none>
此时 Node 和 Event 字段均为空。
为什么这样呢，是不是我们插件发挥作用了呢？我们看下调度器插件日志
# 收到一个刚刚创建的 Pod 日志
I0731 09:14:44.663628 1 eventhandlers.go:118] "Add event for unscheduled pod" pod="default/test-scheduler-78c89768cf-5d9ct"
# 现在开始调度 Pod
I0731 09:14:44.663787 1 schedule_one.go:80] "Attempting to schedule pod" pod="default/test-scheduler-78c89768cf-5d9ct"
# 这里是我们程序里的调试日志，正好对应两个 log.Println()
2023/07/31 09:14:44 filter pod: test-scheduler-78c89768cf-5d9ct, node: &NodeInfo{Pods:[calico-apiserver-f654d8896-c97v9 calico-node-94q9p csi-node-driver-7bjxx nginx-984448cf6-45nrp nginx-984448cf6-zx67q ingress-nginx-controller-8c4c57cd9-n4lvm coredns-7bdc4cb885-m9sbg coredns-7bdc4cb885-npz8p etcd-k8s kube-apiserver-k8s kube-controller-manager-k8s kube-proxy-fxz6j kube-scheduler-k8s controller-7948676b95-b2zfd speaker-p779d minio-operator-67c694f5f6-b4s7h], RequestedResource:&framework.Resource{MilliCPU:1150, Memory:614465536, EphemeralStorage:524288000, AllowedPodNumber:0, ScalarResources:map[v1.ResourceName]int64(nil)}, NonZeroRequest: &framework.Resource{MilliCPU:2050, Memory:3131047936, EphemeralStorage:0, AllowedPodNumber:0, ScalarResources:map[v1.ResourceName]int64(nil)}, UsedPort: framework.HostPortInfo{"0.0.0.0":map[framework.ProtocolPort]struct {}{framework.ProtocolPort{Protocol:"TCP", Port:7472}:struct {}{}, framework.ProtocolPort{Protocol:"TCP", Port:7946}:struct {}{}, framework.ProtocolPort{Protocol:"UDP", Port:7946}:struct {}{}}}, AllocatableResource:&framework.Resource{MilliCPU:4000, Memory:8081461248, EphemeralStorage:907082144291, AllowedPodNumber:110, ScalarResources:map[v1.ResourceName]int64{"hugepages-1Gi":0, "hugepages-2Mi":0, "hugepages-32Mi":0, "hugepages-64Ki":0}}}
2023/07/31 09:14:44 &{{{0 0} {[] {} 0x40003a8e80} map[PreFilterNodePorts:0x4000192280 PreFilterNodeResourcesFit:0x4000192288 PreFilterPodTopologySpread:0x40001922c8 PreFilterVolumeRestrictions:0x40001922a0 VolumeBinding:0x40001922c0 kubernetes.io/pods-to-activate:0x4000192248] 4} false map[InterPodAffinity:{} NodeAffinity:{} VolumeBinding:{} VolumeZone:{}] map[]}
2023/07/31 09:14:44 filter pod: test-scheduler-78c89768cf-5d9ct, node: &NodeInfo{Pods:[calico-apiserver-f654d8896-flfhb calico-kube-controllers-789dc4c76b-h29t5 calico-node-cgm8v calico-typha-5794d6dbd8-7gz5n csi-node-driver-jsxcz nginx-984448cf6-jcwfs nginx-984448cf6-jcwp2 nginx-984448cf6-r6vmg nginx-deployment-7554c7bd74-s2kfn kube-proxy-hgscf sample-scheduler-85cd75d775-jq4c7 speaker-lvp5l minio console-6bdf84b844-vzg72 minio-operator-67c694f5f6-g2bll tigera-operator-549d4f9bdb-txh45], RequestedResource:&framework.Resource{MilliCPU:200, Memory:268435456, EphemeralStorage:524288000, AllowedPodNumber:0, ScalarResources:map[v1.ResourceName]int64(nil)}, NonZeroRequest: &framework.Resource{MilliCPU:1800, Memory:3623878656, EphemeralStorage:0, AllowedPodNumber:0, ScalarResources:map[v1.ResourceName]int64(nil)}, UsedPort: framework.HostPortInfo{"0.0.0.0":map[framework.ProtocolPort]struct {}{framework.ProtocolPort{Protocol:"TCP", Port:5473}:struct {}{}, framework.ProtocolPort{Protocol:"TCP", Port:7472}:struct {}{}, framework.ProtocolPort{Protocol:"TCP", Port:7946}:struct {}{}, framework.ProtocolPort{Protocol:"UDP", Port:7946}:struct {}{}}}, AllocatableResource:&framework.Resource{MilliCPU:4000, Memory:8081461248, EphemeralStorage:907082144291, AllowedPodNumber:110, ScalarResources:map[v1.ResourceName]int64{"hugepages-1Gi":0, "hugepages-2Mi":0, "hugepages-32Mi":0, "hugepages-64Ki":0}}}
2023/07/31 09:14:44 &{{{0 0} {[] {} 0x40007be050} map[] 0} false map[InterPodAffinity:{} NodeAffinity:{} VolumeBinding:{} VolumeZone:{}] map[]}
I0731 09:14:44.665942 1 schedule_one.go:867] "Unable to schedule pod; no fit; waiting" pod="default/test-scheduler-78c89768cf-5d9ct" err="0/4 nodes are available: 1 Node: k8s, 1 Node: node1, 1 node(s) had untolerated taint {node.kubernetes.io/unreachable: }, 1 node(s) were unschedulable. preemption: 0/4 nodes are available: 2 No preemption victims found for incoming pod, 2 Preemption is not helpful for scheduling.."
# 调度失败结果
2023/07/31 09:14:44.666174 1 schedule_one.go:943] "Updating pod condition" pod="default/test-scheduler-78c89768cf-5d9ct" conditionType=PodScheduled conditionStatus=False conditionReason="Unschedulable"
当前环境共四个节点, 但只有 k8s 和 node1 两个节点可以正常使用。
➜ kubectl get node
NAME STATUS ROLES AGE VERSION
k8s Ready control-plane 65d v1.27.1
node1 Ready <none> 64d v1.27.2
node2 NotReady <none> 49d v1.27.2
node3 NotReady,SchedulingDisabled <none> 44d v1.27.2
从日志输出中可以看到此时插件已经起作用了，这正是我们想要的结果，用来干预Pod的调度行为，目前为止一切符合预期。
恢复调度
下面我们实现让Pod 恢复正常调度的效果，我们给 node1添加一个 cpu=true 标签
➜ kubectl label nodes node1 cpu=true
node/node1 labeled
➜ kubectl get nodes -l=cpu=true
NAME STATUS ROLES AGE VERSION
node1 Ready <none> 64d v1.27.2
再次观察插件日志
# 调度Pod
I0731 09:24:16.616059 1 schedule_one.go:80] "Attempting to schedule pod" pod="default/test-scheduler-78c89768cf-5d9ct"
# 打印日志
2023/07/31 09:24:16 filter pod: test-scheduler-78c89768cf-5d9ct, node: &NodeInfo{Pods:[calico-apiserver-f654d8896-c97v9 calico-node-94q9p csi-node-driver-7bjxx nginx-984448cf6-45nrp nginx-984448cf6-zx67q ingress-nginx-controller-8c4c57cd9-n4lvm coredns-7bdc4cb885-m9sbg coredns-7bdc4cb885-npz8p etcd-k8s kube-apiserver-k8s kube-controller-manager-k8s kube-proxy-fxz6j kube-scheduler-k8s controller-7948676b95-b2zfd speaker-p779d minio-operator-67c694f5f6-b4s7h], RequestedResource:&framework.Resource{MilliCPU:1150, Memory:614465536, EphemeralStorage:524288000, AllowedPodNumber:0, ScalarResources:map[v1.ResourceName]int64(nil)}, NonZeroRequest: &framework.Resource{MilliCPU:2050, Memory:3131047936, EphemeralStorage:0, AllowedPodNumber:0, ScalarResources:map[v1.ResourceName]int64(nil)}, UsedPort: framework.HostPortInfo{"0.0.0.0":map[framework.ProtocolPort]struct {}{framework.ProtocolPort{Protocol:"TCP", Port:7472}:struct {}{}, framework.ProtocolPort{Protocol:"TCP", Port:7946}:struct {}{}, framework.ProtocolPort{Protocol:"UDP", Port:7946}:struct {}{}}}, AllocatableResource:&framework.Resource{MilliCPU:4000, Memory:8081461248, EphemeralStorage:907082144291, AllowedPodNumber:110, ScalarResources:map[v1.ResourceName]int64{"hugepages-1Gi":0, "hugepages-2Mi":0, "hugepages-32Mi":0, "hugepages-64Ki":0}}}
2023/07/31 09:24:16 &{{{0 0} {[] {} 0x4000a0ee80} map[PreFilterNodePorts:0x400087c110 PreFilterNodeResourcesFit:0x400087c120 PreFilterPodTopologySpread:0x400087c158 PreFilterVolumeRestrictions:0x400087c130 VolumeBinding:0x400087c148 kubernetes.io/pods-to-activate:0x400087c0d0] 4} true map[InterPodAffinity:{} NodeAffinity:{} VolumeBinding:{} VolumeZone:{}] map[]}
2023/07/31 09:24:16 filter pod: test-scheduler-78c89768cf-5d9ct, node: &NodeInfo{Pods:[calico-apiserver-f654d8896-flfhb calico-kube-controllers-789dc4c76b-h29t5 calico-node-cgm8v calico-typha-5794d6dbd8-7gz5n csi-node-driver-jsxcz nginx-984448cf6-jcwfs nginx-984448cf6-jcwp2 nginx-984448cf6-r6vmg nginx-deployment-7554c7bd74-s2kfn kube-proxy-hgscf sample-scheduler-85cd75d775-jq4c7 speaker-lvp5l minio console-6bdf84b844-vzg72 minio-operator-67c694f5f6-g2bll tigera-operator-549d4f9bdb-txh45], RequestedResource:&framework.Resource{MilliCPU:200, Memory:268435456, EphemeralStorage:524288000, AllowedPodNumber:0, ScalarResources:map[v1.ResourceName]int64(nil)}, NonZeroRequest: &framework.Resource{MilliCPU:1800, Memory:3623878656, EphemeralStorage:0, AllowedPodNumber:0, ScalarResources:map[v1.ResourceName]int64(nil)}, UsedPort: framework.HostPortInfo{"0.0.0.0":map[framework.ProtocolPort]struct {}{framework.ProtocolPort{Protocol:"TCP", Port:5473}:struct {}{}, framework.ProtocolPort{Protocol:"TCP", Port:7472}:struct {}{}, framework.ProtocolPort{Protocol:"TCP", Port:7946}:struct {}{}, framework.ProtocolPort{Protocol:"UDP", Port:7946}:struct {}{}}}, AllocatableResource:&framework.Resource{MilliCPU:4000, Memory:8081461248, EphemeralStorage:907082144291, AllowedPodNumber:110, ScalarResources:map[v1.ResourceName]int64{"hugepages-1Gi":0, "hugepages-2Mi":0, "hugepages-32Mi":0, "hugepages-64Ki":0}}}
2023/07/31 09:24:16 &{{{0 0} {[] {} 0x4000a0f2b0} map[] 0} true map[InterPodAffinity:{} NodeAffinity:{} VolumeBinding:{} VolumeZone:{}] map[]}
# 绑定pod与node
I0731 09:24:16.619544 1 default_binder.go:53] "Attempting to bind pod to node" pod="default/test-scheduler-78c89768cf-5d9ct" node="node1"
I0731 09:24:16.640234 1 eventhandlers.go:161] "Delete event for unscheduled pod" pod="default/test-scheduler-78c89768cf-5d9ct"
I0731 09:24:16.642564 1 schedule_one.go:252] "Successfully bound pod to node" pod="default/test-scheduler-78c89768cf-5d9ct" node="node1" evaluatedNodes=4 feasibleNodes=1
I0731 09:24:16.643173 1 eventhandlers.go:186] "Add event for scheduled pod" pod="default/test-scheduler-78c89768cf-5d9ct"
从日志来看，应该是调度成功了，我们再根据pod的状态确认一下
➜ kubectl get pods --selector=app=test-scheduler
NAME READY STATUS RESTARTS AGE
test-scheduler-78c89768cf-5d9ct 1/1 Running 0 10m
此时Pod状态由 Pending 变成了Running ，表示确实调度成功了。
我们再看看此时的Pod描述信息
➜ kubectl describe pod test-scheduler-78c89768cf-5d9ct
Name: test-scheduler-78c89768cf-5d9ct
Namespace: default
Priority: 0
Service Account: default
Node: node1/192.168.0.205
Start Time: Mon, 31 Jul 2023 17:24:16 +0800
Labels: app=test-scheduler
pod-template-hash=78c89768cf
Annotations: cni.projectcalico.org/containerID: 33e3ffc74e4b2fa15cae210c65d3d4be6a8eadc431e7201185ffa1b1a29cc51d
cni.projectcalico.org/podIP: 10.244.166.182/32
cni.projectcalico.org/podIPs: 10.244.166.182/32
Status: Running
IP: 10.244.166.182
IPs:
IP: 10.244.166.182
Controlled By: ReplicaSet/test-scheduler-78c89768cf
Containers:
nginx:
Container ID: docker://97896d4c4fec2bae294d02125562bc29d769911c7e47e5f4020b1de24ce9c367
Image: nginx:1.23-alpine
Image ID: docker://sha256:510900496a6c312a512d8f4ba0c69586e0fbd540955d65869b6010174362c313
Port: 80/TCP
Host Port: 0/TCP
State: Running
Started: Mon, 31 Jul 2023 17:24:18 +0800
Ready: True
Restart Count: 0
Environment: <none>
Mounts:
/var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-wmh9d (ro)
Conditions:
Type Status
Initialized True
Ready True
ContainersReady True
PodScheduled True
Volumes:
kube-api-access-wmh9d:
Type: Projected (a volume that contains injected data from multiple sources)
TokenExpirationSeconds: 3607
ConfigMapName: kube-root-ca.crt
ConfigMapOptional: <nil>
DownwardAPI: true
QoS Class: BestEffort
Node-Selectors: <none>
Tolerations: node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Warning FailedScheduling 14m sample-scheduler 0/4 nodes are available: 1 Node: k8s, 1 Node: node1, 1 node(s) had untolerated taint {node.kubernetes.io/unreachable: }, 1 node(s) were unschedulable. preemption: 0/4 nodes are available: 2 No preemption victims found for incoming pod, 2 Preemption is not helpful for scheduling..
Warning FailedScheduling 9m27s sample-scheduler 0/4 nodes are available: 1 Node: k8s, 1 Node: node1, 1 node(s) had untolerated taint {node.kubernetes.io/unreachable: }, 1 node(s) were unschedulable. preemption: 0/4 nodes are available: 2 No preemption victims found for incoming pod, 2 Preemption is not helpful for scheduling..
Normal Scheduled 5m9s sample-scheduler Successfully assigned default/test-scheduler-78c89768cf-5d9ct to node1
Normal Pulled 5m8s kubelet Container image "nginx:1.23-alpine" already present on machine
Normal Created 5m8s kubelet Created container nginx
Normal Started 5m8s kubelet Started container nginx
从 Event 字段可以看到我们给 node1 添加 label 标签前后事件日志信息。
到此我们整个插件开发工作基本完成了。
总结
当前示例非常的简单，主要是为了让大家方便理解。对于开发什么样的插件，只需要看对应的插件接口就可以了。然后在配置文件里在合适的扩展点启用即可。
对于自定义调度器的实现，是在main.go文件里通过
app.NewSchedulerCommand(
app.WithPlugin(plugin.Name, plugin.New),
)
来注册自定义调度器插件，然后再通过--config 指定插件是否启用以及启用的扩展点。
本文调度器是以Pod 方式运行，并在容器里挂载一个配置 /etc/kubernetes/scheduler-config.yaml ，也可以直接修改集群 kube-scheduler 的配置文件添加一个新调度器配置来实现，不过由于对集群侵入太大了，个人不推荐这种用法。
常见问题
这里关于k8s 依赖的地方，全部使用 replace 方式才运行起来。这里全部replace到k8s源码的staging目录下了。
[图片上传失败...(image-96b2d9-1690960748676)]
如何不使用 replace 的话，会提示下载k8s依赖出错
k8s.io/kubernetes@v1.20.2 requires
// k8s.io/api@v0.0.0: reading https://goproxy.io/k8s.io/api/@v/v0.0.0.mod: 404 Not Found
// server response: not found: k8s.io/api@v0.0.0: invalid version: unknown revision v0.0.0
在这个问题上卡了好久好久，不理解为什么为这样？哪怕指定了版本号也不行
以下是我的 go.mod 内容
module github.com/cfanbo/sample
go 1.20
require (
github.com/spf13/cobra v1.6.0
k8s.io/kubernetes v0.0.0
)
require (
github.com/Azure/go-ansiterm v0.0.0-20210617225240-d185dfc1b5a1 // indirect
github.com/NYTimes/gziphandler v1.1.1 // indirect
github.com/antlr/antlr4/runtime/Go/antlr v1.4.10 // indirect
github.com/asaskevich/govalidator v0.0.0-20190424111038-f61b66f89f4a // indirect
github.com/beorn7/perks v1.0.1 // indirect
github.com/blang/semver/v4 v4.0.0 // indirect
github.com/cenkalti/backoff/v4 v4.1.3 // indirect
github.com/cespare/xxhash/v2 v2.1.2 // indirect
github.com/coreos/go-semver v0.3.0 // indirect
github.com/coreos/go-systemd/v22 v22.4.0 // indirect
github.com/davecgh/go-spew v1.1.1 // indirect
github.com/docker/distribution v2.8.1+incompatible // indirect
github.com/emicklei/go-restful/v3 v3.9.0 // indirect
github.com/evanphx/json-patch v4.12.0+incompatible // indirect
github.com/felixge/httpsnoop v1.0.3 // indirect
github.com/fsnotify/fsnotify v1.6.0 // indirect
github.com/go-logr/logr v1.2.3 // indirect
github.com/go-logr/stdr v1.2.2 // indirect
github.com/go-openapi/jsonpointer v0.19.6 // indirect
github.com/go-openapi/jsonreference v0.20.1 // indirect
github.com/go-openapi/swag v0.22.3 // indirect
github.com/gogo/protobuf v1.3.2 // indirect
github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da // indirect
github.com/golang/protobuf v1.5.3 // indirect
github.com/google/cel-go v0.12.6 // indirect
github.com/google/gnostic v0.5.7-v3refs // indirect
github.com/google/go-cmp v0.5.9 // indirect
github.com/google/gofuzz v1.1.0 // indirect
github.com/google/uuid v1.3.0 // indirect
github.com/grpc-ecosystem/go-grpc-prometheus v1.2.0 // indirect
github.com/grpc-ecosystem/grpc-gateway/v2 v2.7.0 // indirect
github.com/imdario/mergo v0.3.6 // indirect
github.com/inconshreveable/mousetrap v1.0.1 // indirect
github.com/josharian/intern v1.0.0 // indirect
github.com/json-iterator/go v1.1.12 // indirect
github.com/mailru/easyjson v0.7.7 // indirect
github.com/matttproud/golang_protobuf_extensions v1.0.2 // indirect
github.com/mitchellh/mapstructure v1.4.1 // indirect
github.com/moby/sys/mountinfo v0.6.2 // indirect
github.com/moby/term v0.0.0-20221205130635-1aeaba878587 // indirect
github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd // indirect
github.com/modern-go/reflect2 v1.0.2 // indirect
github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822 // indirect
github.com/opencontainers/go-digest v1.0.0 // indirect
github.com/opencontainers/selinux v1.10.0 // indirect
github.com/pkg/errors v0.9.1 // indirect
github.com/prometheus/client_golang v1.14.0 // indirect
github.com/prometheus/client_model v0.3.0 // indirect
github.com/prometheus/common v0.37.0 // indirect
github.com/prometheus/procfs v0.8.0 // indirect
github.com/spf13/pflag v1.0.5 // indirect
github.com/stoewer/go-strcase v1.2.0 // indirect
go.etcd.io/etcd/api/v3 v3.5.7 // indirect
go.etcd.io/etcd/client/pkg/v3 v3.5.7 // indirect
go.etcd.io/etcd/client/v3 v3.5.7 // indirect
go.opentelemetry.io/contrib/instrumentation/google.golang.org/grpc/otelgrpc v0.35.0 // indirect
go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.35.1 // indirect
go.opentelemetry.io/otel v1.10.0 // indirect
go.opentelemetry.io/otel/exporters/otlp/internal/retry v1.10.0 // indirect
go.opentelemetry.io/otel/exporters/otlp/otlptrace v1.10.0 // indirect
go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracegrpc v1.10.0 // indirect
go.opentelemetry.io/otel/metric v0.31.0 // indirect
go.opentelemetry.io/otel/sdk v1.10.0 // indirect
go.opentelemetry.io/otel/trace v1.10.0 // indirect
go.opentelemetry.io/proto/otlp v0.19.0 // indirect
go.uber.org/atomic v1.7.0 // indirect
go.uber.org/multierr v1.6.0 // indirect
go.uber.org/zap v1.19.0 // indirect
golang.org/x/crypto v0.1.0 // indirect
golang.org/x/net v0.8.0 // indirect
golang.org/x/oauth2 v0.0.0-20220223155221-ee480838109b // indirect
golang.org/x/sync v0.1.0 // indirect
golang.org/x/sys v0.6.0 // indirect
golang.org/x/term v0.6.0 // indirect
golang.org/x/text v0.8.0 // indirect
golang.org/x/time v0.0.0-20220210224613-90d013bbcef8 // indirect
google.golang.org/appengine v1.6.7 // indirect
google.golang.org/genproto v0.0.0-20220502173005-c8bf987b8c21 // indirect
google.golang.org/grpc v1.51.0 // indirect
google.golang.org/protobuf v1.28.1 // indirect
gopkg.in/inf.v0 v0.9.1 // indirect
gopkg.in/natefinch/lumberjack.v2 v2.0.0 // indirect
gopkg.in/yaml.v2 v2.4.0 // indirect
gopkg.in/yaml.v3 v3.0.1 // indirect
k8s.io/api v0.0.0 // indirect
k8s.io/apimachinery v0.0.0 // indirect
k8s.io/apiserver v0.0.0 // indirect
k8s.io/client-go v0.0.0 // indirect
k8s.io/cloud-provider v0.0.0 // indirect
k8s.io/component-base v0.0.0 // indirect
k8s.io/component-helpers v0.0.0 // indirect
k8s.io/controller-manager v0.0.0 // indirect
k8s.io/csi-translation-lib v0.0.0 // indirect
k8s.io/dynamic-resource-allocation v0.0.0 // indirect
k8s.io/klog/v2 v2.90.1 // indirect
k8s.io/kms v0.0.0 // indirect
k8s.io/kube-openapi v0.0.0-20230501164219-8b0f38b5fd1f // indirect
k8s.io/kube-scheduler v0.0.0 // indirect
k8s.io/kubelet v0.0.0 // indirect
k8s.io/mount-utils v0.0.0 // indirect
k8s.io/utils v0.0.0-20230209194617-a36077c30491 // indirect
sigs.k8s.io/apiserver-network-proxy/konnectivity-client v0.1.2 // indirect
sigs.k8s.io/json v0.0.0-20221116044647-bc3834ca7abd // indirect
sigs.k8s.io/structured-merge-diff/v4 v4.2.3 // indirect
sigs.k8s.io/yaml v1.3.0 // indirect
)
// 使用本地的 k8s 源码路径替换
replace (
k8s.io/api => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/api
k8s.io/apiextensions-apiserver => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/apiextensions-apiserver
k8s.io/apimachinery => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/apimachinery
k8s.io/apiserver => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/apiserver
k8s.io/cli-runtime => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/cli-runtime
k8s.io/client-go => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/client-go
k8s.io/cloud-provider => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/cloud-provider
k8s.io/cluster-bootstrap => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/cluster-bootstrap
k8s.io/code-generator => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/code-generator
k8s.io/component-base => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/component-base
k8s.io/component-helpers => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/component-helpers
k8s.io/controller-manager => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/controller-manager
k8s.io/cri-api => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/cri-api
k8s.io/csi-translation-lib => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/csi-translation-lib
k8s.io/dynamic-resource-allocation => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/dynamic-resource-allocation
k8s.io/kms => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/kms
k8s.io/kube-aggregator => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/kube-aggregator
k8s.io/kube-controller-manager => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/kube-controller-manager
k8s.io/kube-proxy => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/kube-proxy
k8s.io/kube-scheduler => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/kube-scheduler
k8s.io/kubectl => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/kubectl
k8s.io/kubelet => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/kubelet
k8s.io/kubernetes => /Users/sxf/workspace/kubernetes
k8s.io/legacy-cloud-providers => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/legacy-cloud-providers
k8s.io/metrics => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/metrics
k8s.io/mount-utils => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/mount-utils
k8s.io/pod-security-admission => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/pod-security-admission
k8s.io/sample-apiserver => /Users/sxf/workspace/kubernetes/staging/src/k8s.io/sample-apiserver
)
这里有些依赖是没有用的，真正开发的时候直接使用 go mod tidy 清理一下即可。
参考资料
- https://github.com/kubernetes-sigs/scheduler-plugins
- https://github.com/kubernetes-sigs/scheduler-plugins/blob/master/doc/install.md
- https://github.com/kubernetes-sigs/scheduler-plugins/blob/master/doc/develop.md
- https://github.com/kubernetes-sigs/scheduler-plugins/blob/master/cmd/scheduler/main.go
- https://kubernetes.io/zh-cn/docs/concepts/scheduling-eviction/scheduling-framework/
- https://kubernetes.io/zh-cn/docs/reference/scheduling/config/
- https://www.qikqiak.com/post/custom-kube-scheduler/
- https://github.com/kubernetes/enhancements/tree/master/keps/sig-scheduling/624-scheduling-framework#custom-scheduler-plugins-out-of-tree
- https://www.xiexianbin.cn/kubernetes/scheduler/creating-a-k8s-scheduler-plugin/index.html
- https://mp.weixin.qq.com/s/FGzwDsrjCNesiNbYc3kLcA
- https://github.com/kelseyhightower/kubernetes-the-hard-way/blob/master/docs/08-bootstrapping-kubernetes-controllers.md#configure-the-kubernetes-scheduler
- https://developer.ibm.com/articles/creating-a-custom-kube-scheduler/
- https://www.bilibili.com/video/BV1tT4y1o77z/
