目录
[Metrics Server](#Metrics Server)
[安装Metrics Server](#安装Metrics Server)
[查看metrics-server Pod状态](#查看metrics-server Pod状态)
[测试 API 可用性](#测试 API 可用性)
[获取 Service 的访问地址](#获取 Service 的访问地址)
[安装压测工具（例如 hey 或 ab）](#安装压测工具（例如 hey 或 ab）)
[多指标 HPA（CPU + 内存）](#多指标 HPA（CPU + 内存）)
HPA概述
概念：
Kubernetes 的 Horizontal Pod Autoscaler (HPA) 是一个自动化的控制回路，其核心特点是能根据观察到的指标（如 CPU 利用率）自动调整 Pod 的副本数量。
核心机制与工作原理：
HPA 作为一个控制回路，周期性（默认 15 秒）地运行，其工作流程如下：
查询指标：从 metrics-server（资源指标）或自定义 API（自定义/外部指标）获取目标工作负载（如 Deployment）下所有 Pod 的指标数据。
计算期望副本数：根据所有 Pod 指标的平均值与设定的目标值进行比较，计算出一个期望的副本数。计算公式为：期望副本数 = ceil当前副本数 \* (当前指标平均值 / 期望指标值)。
执行扩缩容：将 spec.replicas 调整为计算出的期望副本数，但结果会受到 minReplicas 和 maxReplicas 的限制。
扩缩容特点：
快速扩容，谨慎缩容。默认值：扩容稳定窗口为 0 秒（立即扩容），缩容稳定窗口为 300 秒（5分钟）。窗口期内，HPA 会持续观察并记录多个期望副本数，缩容时，会从窗口内的所有期望值中选择最大的一个副本数，以避免因指标瞬时下降而误缩容。扩容时，则会选择最小的一个副本数，避免因指标瞬时飙升而过度扩容。
Metrics Server
Metrics Server是 Kubernetes 集群的"资源数据采集中心"，专门负责收集所有 Node 和 Pod 的 CPU、内存实时使用数据，并通过标准 API 提供给 HPA、kubectl top 等组件使用。
安装Metrics Server
[root@k8s-master01 metrics]# kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml
serviceaccount/metrics-server created
clusterrole.rbac.authorization.k8s.io/system:aggregated-metrics-reader created
clusterrole.rbac.authorization.k8s.io/system:metrics-server created
rolebinding.rbac.authorization.k8s.io/metrics-server-auth-reader created
clusterrolebinding.rbac.authorization.k8s.io/metrics-server:system:auth-delegator created
clusterrolebinding.rbac.authorization.k8s.io/system:metrics-server created
service/metrics-server created
deployment.apps/metrics-server created
apiservice.apiregistration.k8s.io/v1beta1.metrics.k8s.io created
[root@k8s-master01 metrics]#查看metrics-server Pod状态
READY（就绪）：显示 1/1 表示容器已启动且健康检查通过，可以正常提供服务。如果显示 0/1，说明启动失败或健康检查未通过。
STATUS（状态）：必须是 Running 才算正常。
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-84c7d8c58-cw4fn 0/1 ImagePullBackOff 0 87s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-84c7d8c58-cw4fn 0/1 ImagePullBackOff 0 87s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-84c7d8c58-cw4fn 0/1 ErrImagePull 0 107s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-84c7d8c58-cw4fn 0/1 ImagePullBackOff 0 113s
[root@k8s-master01 metrics]#从结果看，不是1/1 Running，说明不正常。
经检查发现镜像下载不下来，编辑镜像地址
[root@k8s-master01 metrics]# kubectl edit deployment metrics-server -n kube-system
deployment.apps/metrics-server edited
[root@k8s-master01 metrics]# 把原来的镜像
registry.k8s.io/metrics-server/metrics-server:v0.8.1
修改为如下镜像地址
registry.cn-hangzhou.aliyuncs.com/google_containers/metrics-server:v0.8.1
保存退出（和vi命令保存方法一致）
查看metrics-server Pod状态
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-7cdb4f6f76-vh96m 0/1 Running 0 51s
metrics-server-84c7d8c58-bx7dh 0/1 ImagePullBackOff 0 5m48s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-7cdb4f6f76-vh96m 0/1 Running 0 52s
metrics-server-84c7d8c58-bx7dh 0/1 ImagePullBackOff 0 5m49s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-7cdb4f6f76-vh96m 0/1 Running 0 53s
metrics-server-84c7d8c58-bx7dh 0/1 ImagePullBackOff 0 5m50s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-7cdb4f6f76-vh96m 0/1 Running 0 54s
metrics-server-84c7d8c58-bx7dh 0/1 ImagePullBackOff 0 5m51s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-7cdb4f6f76-vh96m 0/1 Running 0 55s
metrics-server-84c7d8c58-bx7dh 0/1 ImagePullBackOff 0 5m52s
[root@k8s-master01 metrics]# kubectl delete pod metrics-server-84c7d8c58-bx7dh -n kube-system
pod "metrics-server-84c7d8c58-bx7dh" deleted
[root@k8s-master01 metrics]# kubectl delete rs metrics-server-84c7d8c58 -n kube-system
replicaset.apps "metrics-server-84c7d8c58" deleted
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-7cdb4f6f76-vh96m 0/1 Running 0 3m51s看到0/1 Running，说明未就绪
排查问题
[root@k8s-master01 metrics]# kubectl describe pod metrics-server-7cdb4f6f76-vh96m -n kube-system
Name: metrics-server-7cdb4f6f76-vh96m
Namespace: kube-system
Priority: 2000000000
Priority Class Name: system-cluster-critical
Service Account: metrics-server
Node: k8s-node01/192.168.204.102
Start Time: Sun, 14 Jun 2026 00:18:53 +0800
Labels: k8s-app=metrics-server
pod-template-hash=7cdb4f6f76
Annotations: cni.projectcalico.org/containerID: 2254575903eb886a815025558f386b5310c595692be2921f8f5cab844afc8cc8
cni.projectcalico.org/podIP: 10.244.85.247/32
cni.projectcalico.org/podIPs: 10.244.85.247/32
Status: Running
IP: 10.244.85.247
IPs:
IP: 10.244.85.247
Controlled By: ReplicaSet/metrics-server-7cdb4f6f76
Containers:
metrics-server:
Container ID: docker://074410d0608036ec8bb65949128b6cae70890f8cc7bf5be4b196dd246958b602
Image: registry.cn-hangzhou.aliyuncs.com/google_containers/metrics-server:v0.8.1
Image ID: docker-pullable://registry.cn-hangzhou.aliyuncs.com/google_containers/metrics-server@sha256:24301f788bfea02750c8a4a522c0bdf36ab83edf99780bc843ec23a5706610a0
Port: 10250/TCP
Host Port: 0/TCP
SeccompProfile: RuntimeDefault
Args:
--cert-dir=/tmp
--secure-port=10250
--kubelet-preferred-address-types=InternalIP,ExternalIP,Hostname
--kubelet-use-node-status-port
--metric-resolution=15s
State: Running
Started: Sun, 14 Jun 2026 00:18:54 +0800
Ready: False
Restart Count: 0
Requests:
cpu: 100m
memory: 200Mi
Liveness: http-get https://:https/livez delay=0s timeout=1s period=10s #success=1 #failure=3
Readiness: http-get https://:https/readyz delay=20s timeout=1s period=10s #success=1 #failure=3
Environment: <none>
Mounts:
/tmp from tmp-dir (rw)
/var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-qtpcs (ro)
Conditions:
Type Status
PodReadyToStartContainers True
Initialized True
Ready False
ContainersReady False
PodScheduled True
Volumes:
tmp-dir:
Type: EmptyDir (a temporary directory that shares a pod's lifetime)
Medium:
SizeLimit: <unset>
kube-api-access-qtpcs:
Type: Projected (a volume that contains injected data from multiple sources)
TokenExpirationSeconds: 3607
ConfigMapName: kube-root-ca.crt
ConfigMapOptional: <nil>
DownwardAPI: true
QoS Class: Burstable
Node-Selectors: kubernetes.io/os=linux
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
---- ------ ---- ---- -------
Normal Scheduled 5m default-scheduler Successfully assigned kube-system/metrics-server-7cdb4f6f76-vh96m to k8s-node01
Normal Pulled 5m kubelet Container image "registry.cn-hangzhou.aliyuncs.com/google_containers/metrics-server:v0.8.1" already present on machine
Normal Created 5m kubelet Created container metrics-server
Normal Started 4m59s kubelet Started container metrics-server
Warning Unhealthy 90s (x22 over 4m40s) kubelet Readiness probe failed: HTTP probe failed with statuscode: 500
[root@k8s-master01 metrics]#发现Readiness probe failed: HTTP probe failed with statuscode: 500问题，说明Kubernetes 对 Metrics Server 的健康检查（/readyz 接口）发出了请求，但 Metrics Server 返回500错误。
编辑metrics-server部署资源配置
[root@k8s-master01 metrics]# kubectl edit deployment metrics-server -n kube-system在 spec.template.spec.containers0.args 中添加一行 --kubelet-insecure-tls。
修改前
添加内容
- --kubelet-insecure-tls修改后
等待一会后，查询pod状态，为 1/1 Running，说明正常可用。
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-678895c74f-vztdq 0/1 Running 0 16s
metrics-server-7cdb4f6f76-vh96m 0/1 Running 0 8m44s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-678895c74f-vztdq 1/1 Running 0 2m21s
[root@k8s-master01 metrics]#测试 API 可用性
预期结果：kubectl top 能返回 CPU/内存数据（即使数值为 0）。若无数据，请检查 Metrics Server Pod 日志。
# 查看节点CPU/内存数据
[root@k8s-master01 metrics]# kubectl top nodes
NAME CPU(cores) CPU% MEMORY(bytes) MEMORY%
k8s-master01 172m 4% 2078Mi 58%
k8s-node01 88m 2% 1138Mi 32%
k8s-node02 98m 2% 1284Mi 36%
# 看所有PodsCPU/内存数据
[root@k8s-master01 metrics]# kubectl top pods -A
NAMESPACE NAME CPU(cores) MEMORY(bytes)
kube-system calico-kube-controllers-7fbd86d5c5-k2nd8 1m 58Mi
kube-system calico-node-qfqqc 30m 201Mi
kube-system calico-node-vz7zx 24m 192Mi
kube-system calico-node-z6p7k 28m 198Mi
kube-system calico-typha-669c48c58c-sf7km 3m 62Mi
kube-system coredns-855c4dd65d-l4rcc 2m 17Mi
kube-system coredns-855c4dd65d-qrjlk 2m 60Mi
kube-system etcd-k8s-master01 23m 227Mi
kube-system kube-apiserver-k8s-master01 40m 540Mi
kube-system kube-controller-manager-k8s-master01 12m 130Mi
kube-system kube-proxy-pwdvj 1m 79Mi
kube-system kube-proxy-wt7gl 1m 80Mi
kube-system kube-proxy-xc6gt 1m 80Mi
kube-system kube-scheduler-k8s-master01 3m 70Mi
kube-system metrics-server-678895c74f-vztdq 3m 20Mi
metallb-system controller-cd8c9874-5vl4c 3m 63Mi
metallb-system speaker-6vmp7 4m 64Mi
metallb-system speaker-pqr96 4m 65Mi
metallb-system speaker-pztv2 4m 64Mi
[root@k8s-master01 metrics]#设置资源限制
编辑metrics-server部署资源配置
[root@k8s-master01 metrics]# kubectl edit deployment metrics-server -n kube-system设置资源限制是为了防止容器过度消耗节点资源，保证系统稳定性。
修改前
在resources中添加
limits:
cpu: 200m
memory: 400Mi修改后
保存
查看pods状态
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-678895c74f-vztdq 1/1 Running 0 13m
metrics-server-7b97b84c9c-wthbc 0/1 Running 0 21s
[root@k8s-master01 metrics]# kubectl get pods -n kube-system | grep metrics-server
metrics-server-7b97b84c9c-wthbc 1/1 Running 0 67s
[root@k8s-master01 metrics]#
[root@k8s-master01 metrics]# cd ~
[root@k8s-master01 ~]# mkdir hpa
[root@k8s-master01 ~]# cp nginx-deployment.yaml nginx-service.yaml hpa
[root@k8s-master01 ~]# cd hpa
[root@k8s-master01 hpa]# ls
nginx-deployment.yaml nginx-service.yaml修改nginx-deployment.yaml
[root@k8s-master01 hpa]# vi nginx-deployment.yaml
在image的下一行添加
resources:
requests:
cpu: 100m # 0.1 核
memory: 128Mi
limits:
cpu: 200m
memory: 256Mi修改后
修改后完整的nginx-deployment.yaml如下
apiVersion: apps/v1 #与k8s集群版本有关，使用 kubectl api-versions 即可查看当前集群支持的版本
kind: Deployment #该配置的类型，使用的是 Deployment
metadata: #译名为元数据，即 Deployment 的一些基本属性和信息
name: nginx-deployment #Deployment 的名称
labels: #标签，可以灵活定位一个或多个资源，其中key和value均可自定义，可以定义多组，目前不需要理解
app: nginx #为该Deployment设置key为app，value为nginx的标签
spec: #这是关于该Deployment的描述，可以理解为你期待该Deployment在k8s中如何使用
replicas: 1 #使用该Deployment创建一个应用程序实例
selector: #标签选择器，与上面的标签共同作用，目前不需要理解
matchLabels: #选择包含标签app:nginx的资源
app: nginx
template: #这是选择或创建的Pod的模板
metadata: #Pod的元数据
labels: #Pod的标签，上面的selector即选择包含标签app:nginx的Pod
app: nginx
spec: #期望Pod实现的功能（即在pod中部署）
containers: #生成container，与docker中的container是同一种
- name: nginx #container的名称
image: nginx:alpine #使用镜像nginx:alpine创建container，该container默认80端口可访问
resources:
requests:
cpu: 100m # 0.1 核
memory: 128Mi
limits:
cpu: 200m
memory: 256Mi应用nginx-deployment.yaml
[root@k8s-master01 hpa]# kubectl apply -f nginx-deployment.yaml
deployment.apps/nginx-deployment created
[root@k8s-master01 hpa]# kubectl rollout status deployment nginx-deployment
deployment "nginx-deployment" successfully rolled out
[root@k8s-master01 hpa]#验证资源设置
[root@k8s-master01 hpa]# kubectl describe pod -l app=nginx | grep -A5 "Requests\|Limits"
Limits:
cpu: 200m
memory: 256Mi
Requests:
cpu: 100m
memory: 128Mi
Environment: <none>
Mounts:
/var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-72h5l (ro)
[root@k8s-master01 hpa]#同时执行 kubectl top pod 观察基准资源使用。
[root@k8s-master01 hpa]# kubectl top pod
NAME CPU(cores) MEMORY(bytes)
nginx-deployment-678cb5c9fd-t7s2b 0m 12Mi创建nginx-service.yaml
vi nginx-service.yaml内容如下
apiVersion: v1
kind: Service
metadata:
name: nginx-service #Service 的名称
labels: #Service 自己的标签
app: nginx #为该 Service 设置 key 为 app，value 为 nginx 的标签
spec: #这是关于该 Service 的定义，描述了 Service 如何选择 Pod，如何被访问
selector: #标签选择器
app: nginx #选择包含标签 app:nginx 的 Pod
ports:
- name: nginx-port #端口的名字
protocol: TCP #协议类型 TCP/UDP
port: 80 #集群内的其他容器组可通过 80 端口访问 Service
nodePort: 32600 #通过任意节点的 32600 端口访问 Service
targetPort: 80 #将请求转发到匹配 Pod 的 80 端口
type: NodePort #Serive的类型，ClusterIP/NodePort/LoaderBalancer应用nginx-service.yaml
[root@k8s-master01 hpa]# kubectl apply -f nginx-service.yaml
service/nginx-service created
[root@k8s-master01 hpa]#创建HorizontalPodAutoscaler
基于 CPU 使用率伸缩是最常用的生产策略。创建 HPA 对象，目标为 CPU 使用率 50%，副本数范围 1~10。可使用命令行创建创建或者使用YAML文件创建，这里使用YAML文件创建。
使用命令行创建（快速验证）
kubectl autoscale deployment nginx-deployment --cpu-percent=50 --min=1 --max=10这里不操作命令行创建，仅作为了解
使用 YAML 文件（更利于版本控制）
创建 hpa-nginx.yaml：
vi hpa-nginx.yaml 内容如下
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
name: nginx-hpa
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: nginx-deployment
minReplicas: 1
maxReplicas: 10
metrics:
- type: Resource
resource:
name: cpu
target:
type: Utilization
averageUtilization: 50应用
[root@k8s-master01 hpa]# kubectl apply -f hpa-nginx.yaml
horizontalpodautoscaler.autoscaling/nginx-hpa created
[root@k8s-master01 hpa]#查看 HPA 状态
kubectl get hpa
kubectl describe hpa nginx-hpa此时 TARGETS 列应显示 unknown 或 0%/50%（因为无负载）。
操作过程
[root@k8s-master01 hpa]# kubectl get hpa
NAME REFERENCE TARGETS MINPODS MAXPODS REPLICAS AGE
nginx-hpa Deployment/nginx-deployment cpu: 0%/50% 1 10 1 31s
[root@k8s-master01 hpa]# kubectl describe hpa nginx-hpa
Name: nginx-hpa
Namespace: default
Labels: <none>
Annotations: <none>
CreationTimestamp: Sun, 14 Jun 2026 00:56:08 +0800
Reference: Deployment/nginx-deployment
Metrics: ( current / target )
resource cpu on pods (as a percentage of request): 0% (0) / 50%
Min replicas: 1
Max replicas: 10
Deployment pods: 1 current / 1 desired
Conditions:
Type Status Reason Message
---- ------ ------ -------
AbleToScale True ScaleDownStabilized recent recommendations were higher than current one, applying the highest recent recommendation
ScalingActive True ValidMetricFound the HPA was able to successfully calculate a replica count from cpu resource utilization (percentage of request)
ScalingLimited False DesiredWithinRange the desired count is within the acceptable range
Events: <none>
[root@k8s-master01 hpa]#压力测试与自动伸缩验证
模拟生产流量，观察 HPA 如何增加 Pod 数量并最终恢复。
获取 Service 的访问地址
NODE_IP=$(kubectl get nodes -o jsonpath='{.items[0].status.addresses[?(@.type=="InternalIP")].address}')
NODE_PORT=32600
echo "http://$NODE_IP:$NODE_PORT"操作过程
[root@k8s-master01 hpa]# NODE_IP=$(kubectl get nodes -o jsonpath='{.items[0].status.addresses[?(@.type=="InternalIP")].address}')
NODE_PORT=32600
echo "http://$NODE_IP:$NODE_PORT"
http://192.168.204.101:32600
[root@k8s-master01 hpa]#安装压测工具（例如 hey 或 ab）
以 hey （轻量并发工具）为例
# 安装go
yum install epel-release -y
yum update -y
yum install golang -y
go version
go env -w GOPROXY=https://goproxy.cn,direct
# 安装hey
go install github.com/rakyll/hey@latest部分操作过程如下
[root@k8s-master01 hpa]# yum install golang -y
Last metadata expiration check: 1:39:20 ago on Sat Jun 13 23:32:42 2026.
Dependencies resolved.
===================================================================================================
Package Architecture Version Repository Size
===================================================================================================
Installing:
golang x86_64 1.26.3-1.el9_8 appstream 1.4 M
Installing dependencies:
cpp x86_64 11.5.0-14.el9 appstream 11 M
gcc x86_64 11.5.0-14.el9 appstream 32 M
glibc-devel x86_64 2.34-270.el9_8 appstream 37 k
glibc-headers x86_64 2.34-270.el9_8 appstream 444 k
golang-bin x86_64 1.26.3-1.el9_8 appstream 44 M
golang-race x86_64 1.26.3-1.el9_8 appstream 1.7 M
golang-src noarch 1.26.3-1.el9_8 appstream 11 M
kernel-headers x86_64 5.14.0-687.15.1.el9_8 appstream 2.5 M
libmpc x86_64 1.2.1-4.el9 appstream 61 k
libpkgconf x86_64 1.7.3-10.el9 baseos 35 k
libxcrypt-devel x86_64 4.4.18-3.el9 appstream 28 k
make x86_64 1:4.3-8.el9 baseos 529 k
openssl-devel x86_64 1:3.5.5-4.el9_8 appstream 3.4 M
pkgconf x86_64 1.7.3-10.el9 baseos 40 k
pkgconf-m4 noarch 1.7.3-10.el9 baseos 14 k
pkgconf-pkg-config x86_64 1.7.3-10.el9 baseos 10 k
Transaction Summary
===================================================================================================
Install 17 Packages
Total download size: 108 M
Installed size: 337 M
Downloading Packages:
(1/17): libpkgconf-1.7.3-10.el9.x86_64.rpm 49 kB/s | 35 kB 00:00
(2/17): pkgconf-1.7.3-10.el9.x86_64.rpm 53 kB/s | 40 kB 00:00
(3/17): pkgconf-m4-1.7.3-10.el9.noarch.rpm 105 kB/s | 14 kB 00:00
(4/17): make-4.3-8.el9.x86_64.rpm 614 kB/s | 529 kB 00:00
(5/17): pkgconf-pkg-config-1.7.3-10.el9.x86_64.rpm 60 kB/s | 10 kB 00:00
(6/17): glibc-devel-2.34-270.el9_8.x86_64.rpm 97 kB/s | 37 kB 00:00
(7/17): glibc-headers-2.34-270.el9_8.x86_64.rpm 1.1 MB/s | 444 kB 00:00
(8/17): golang-1.26.3-1.el9_8.x86_64.rpm 938 kB/s | 1.4 MB 00:01
(9/17): cpp-11.5.0-14.el9.x86_64.rpm 1.3 MB/s | 11 MB 00:08
(10/17): golang-race-1.26.3-1.el9_8.x86_64.rpm 702 kB/s | 1.7 MB 00:02
(11/17): gcc-11.5.0-14.el9.x86_64.rpm 1.8 MB/s | 32 MB 00:18
(12/17): kernel-headers-5.14.0-687.15.1.el9_8.x86_64.rpm 869 kB/s | 2.5 MB 00:02
(13/17): libmpc-1.2.1-4.el9.x86_64.rpm 108 kB/s | 61 kB 00:00
(14/17): libxcrypt-devel-4.4.18-3.el9.x86_64.rpm 34 kB/s | 28 kB 00:00
(15/17): openssl-devel-3.5.5-4.el9_8.x86_64.rpm 1.5 MB/s | 3.4 MB 00:02
(16/17): golang-bin-1.26.3-1.el9_8.x86_64.rpm 1.9 MB/s | 44 MB 00:23
(17/17): golang-src-1.26.3-1.el9_8.noarch.rpm 697 kB/s | 11 MB 00:15
---------------------------------------------------------------------------------------------------
Total 3.9 MB/s | 108 MB 00:27
Running transaction check
Transaction check succeeded.
Running transaction test
Transaction test succeeded.
Running transaction
Running scriptlet: golang-1.26.3-1.el9_8.x86_64 1/1
Preparing : 1/1
Installing : libmpc-1.2.1-4.el9.x86_64 1/17
Installing : cpp-11.5.0-14.el9.x86_64 2/17
Installing : kernel-headers-5.14.0-687.15.1.el9_8.x86_64 3/17
Installing : golang-src-1.26.3-1.el9_8.noarch 4/17
Installing : glibc-headers-2.34-270.el9_8.x86_64 5/17
Installing : pkgconf-m4-1.7.3-10.el9.noarch 6/17
Installing : make-1:4.3-8.el9.x86_64 7/17
Installing : libpkgconf-1.7.3-10.el9.x86_64 8/17
Installing : pkgconf-1.7.3-10.el9.x86_64 9/17
Installing : pkgconf-pkg-config-1.7.3-10.el9.x86_64 10/17
Installing : glibc-devel-2.34-270.el9_8.x86_64 11/17
Installing : libxcrypt-devel-4.4.18-3.el9.x86_64 12/17
Installing : gcc-11.5.0-14.el9.x86_64 13/17
Installing : openssl-devel-1:3.5.5-4.el9_8.x86_64 14/17
Installing : golang-bin-1.26.3-1.el9_8.x86_64 15/17
Installing : golang-race-1.26.3-1.el9_8.x86_64 16/17
Installing : golang-1.26.3-1.el9_8.x86_64 17/17
Running scriptlet: golang-1.26.3-1.el9_8.x86_64 17/17
Verifying : libpkgconf-1.7.3-10.el9.x86_64 1/17
Verifying : make-1:4.3-8.el9.x86_64 2/17
Verifying : pkgconf-1.7.3-10.el9.x86_64 3/17
Verifying : pkgconf-m4-1.7.3-10.el9.noarch 4/17
Verifying : pkgconf-pkg-config-1.7.3-10.el9.x86_64 5/17
Verifying : cpp-11.5.0-14.el9.x86_64 6/17
Verifying : gcc-11.5.0-14.el9.x86_64 7/17
Verifying : glibc-devel-2.34-270.el9_8.x86_64 8/17
Verifying : glibc-headers-2.34-270.el9_8.x86_64 9/17
Verifying : golang-1.26.3-1.el9_8.x86_64 10/17
Verifying : golang-bin-1.26.3-1.el9_8.x86_64 11/17
Verifying : golang-race-1.26.3-1.el9_8.x86_64 12/17
Verifying : golang-src-1.26.3-1.el9_8.noarch 13/17
Verifying : kernel-headers-5.14.0-687.15.1.el9_8.x86_64 14/17
Verifying : libmpc-1.2.1-4.el9.x86_64 15/17
Verifying : libxcrypt-devel-4.4.18-3.el9.x86_64 16/17
Verifying : openssl-devel-1:3.5.5-4.el9_8.x86_64 17/17
Installed:
cpp-11.5.0-14.el9.x86_64 gcc-11.5.0-14.el9.x86_64
glibc-devel-2.34-270.el9_8.x86_64 glibc-headers-2.34-270.el9_8.x86_64
golang-1.26.3-1.el9_8.x86_64 golang-bin-1.26.3-1.el9_8.x86_64
golang-race-1.26.3-1.el9_8.x86_64 golang-src-1.26.3-1.el9_8.noarch
kernel-headers-5.14.0-687.15.1.el9_8.x86_64 libmpc-1.2.1-4.el9.x86_64
libpkgconf-1.7.3-10.el9.x86_64 libxcrypt-devel-4.4.18-3.el9.x86_64
make-1:4.3-8.el9.x86_64 openssl-devel-1:3.5.5-4.el9_8.x86_64
pkgconf-1.7.3-10.el9.x86_64 pkgconf-m4-1.7.3-10.el9.noarch
pkgconf-pkg-config-1.7.3-10.el9.x86_64
Complete!
[root@k8s-master01 hpa]# go version
go version go1.26.3 (Red Hat 1.26.3-1.el9_8) linux/amd64
[root@k8s-master01 hpa]# go env -w GOPROXY=https://goproxy.cn,direct
[root@k8s-master01 hpa]# go install github.com/rakyll/hey@latest
go: downloading github.com/rakyll/hey v0.1.5
go: downloading golang.org/x/net v0.48.0
go: downloading golang.org/x/text v0.33.0
[root@k8s-master01 hpa]#发起持续负载
向 Nginx Service 发送请求，50 个并发连接（即 50 个客户端线程同时发送请求），保持 QPS 约 2000，总共发送 20 万个请求
hey -n 200000 -c 50 -q 2000 http://$NODE_IP:$NODE_PORT/操作过程
[root@k8s-master01 hpa]# hey -n 200000 -c 50 -q 2000 http://$NODE_IP:$NODE_PORT/
-bash: hey: command not found
[root@k8s-master01 hpa]# ls ~
231 deploy-all.sh hpa
232 deploy-full.sh metallb-config.yaml
anaconda-ks.cfg envoyproxy nginx-deployment.yaml
calico.yaml envoyproxy.yaml nginx-httproute.yaml
certificate.yaml fix-k8s-cert.sh nginx-service.yaml
cluster-issuer.yaml gateway.yaml redirect-httproute.yaml
cri-dockerd gateway.yaml.old tls.crt
crtbak gatewayclass.yaml tls.key
delete-all.sh go worker_fix.sh
[root@k8s-master01 hpa]# ls ~/go/
bin pkg
[root@k8s-master01 hpa]# ls ~/go/bin/
hey
[root@k8s-master01 hpa]#
[root@k8s-master01 hpa]# ~/go/bin/hey -n 200000 -c 50 -q 2000 http://$NODE_IP:$NODE_PORT/注意：hep命令在~/go/bin目录下
发起持续负载过程如下
[root@k8s-master01 hpa]# ~/go/bin/hey -n 200000 -c 50 -q 2000 http://$NODE_IP:$NODE_PORT/
Summary:
Total: 59.2321 secs
Slowest: 0.4976 secs
Fastest: 0.0002 secs
Average: 0.0140 secs
Requests/sec: 3376.5488
Total data: 179200000 bytes
Size/request: 896 bytes
Response time histogram:
0.000 [1] |
0.050 [176109] |■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
0.100 [20891] |■■■■■
0.149 [2514] |■
0.199 [286] |
0.249 [112] |
0.299 [24] |
0.348 [17] |
0.398 [5] |
0.448 [15] |
0.498 [26] |
Latency distribution:
10%% in 0.0008 secs
25%% in 0.0015 secs
50%% in 0.0031 secs
75%% in 0.0071 secs
90%% in 0.0689 secs
95%% in 0.0859 secs
99%% in 0.1034 secs
Details (average, fastest, slowest):
DNS+dialup: 0.0000 secs, 0.0000 secs, 0.0027 secs
DNS-lookup: 0.0000 secs, 0.0000 secs, 0.0000 secs
req write: 0.0000 secs, 0.0000 secs, 0.0261 secs
resp wait: 0.0133 secs, 0.0002 secs, 0.4975 secs
resp read: 0.0006 secs, 0.0000 secs, 0.3090 secs
Status code distribution:
[200] 200000 responses
[root@k8s-master01 hpa]#观察伸缩过程（另开终端）
发起持续负载后，立即打开另外的终端，实时监控 HPA 状态
预期现象：TARGETS 从 0%/50% 上升到 >50%，REPLICAS 开始增加（1→2→4... 直至满足负载或达到 max=10）。
[root@k8s-master01 ~]# kubectl get hpa -w
NAME REFERENCE TARGETS MINPODS MAXPODS REPLICAS AGE
nginx-hpa Deployment/nginx-deployment cpu: 0%/50% 1 10 1 18m
nginx-hpa Deployment/nginx-deployment cpu: 41%/50% 1 10 1 20m
nginx-hpa Deployment/nginx-deployment cpu: 200%/50% 1 10 1 21m
nginx-hpa Deployment/nginx-deployment cpu: 201%/50% 1 10 4 21m
nginx-hpa Deployment/nginx-deployment cpu: 194%/50% 1 10 4 21m一段时间后观察到如下数据
[root@k8s-master01 ~]# kubectl get hpa -w
NAME REFERENCE TARGETS MINPODS MAXPODS REPLICAS AGE
nginx-hpa Deployment/nginx-deployment cpu: 0%/50% 1 10 1 18m
nginx-hpa Deployment/nginx-deployment cpu: 41%/50% 1 10 1 20m
nginx-hpa Deployment/nginx-deployment cpu: 200%/50% 1 10 1 21m
nginx-hpa Deployment/nginx-deployment cpu: 201%/50% 1 10 4 21m
nginx-hpa Deployment/nginx-deployment cpu: 194%/50% 1 10 4 21m
nginx-hpa Deployment/nginx-deployment cpu: 113%/50% 1 10 8 21m
nginx-hpa Deployment/nginx-deployment cpu: 0%/50% 1 10 10 22m
nginx-hpa Deployment/nginx-deployment cpu: 0%/50% 1 10 10 26m
nginx-hpa Deployment/nginx-deployment cpu: 0%/50% 1 10 9 26m
nginx-hpa Deployment/nginx-deployment cpu: 0%/50% 1 10 2 27m
nginx-hpa Deployment/nginx-deployment cpu: 0%/50% 1 10 1 27m同时持续观察Pod数量变化（另开终端）
[root@k8s-master01 ~]# kubectl get pods -l app=nginx -w
NAME READY STATUS RESTARTS AGE
nginx-deployment-678cb5c9fd-47wwb 1/1 Running 0 76s
nginx-deployment-678cb5c9fd-82tdw 1/1 Running 0 45s
nginx-deployment-678cb5c9fd-bb7j7 1/1 Running 0 30s
nginx-deployment-678cb5c9fd-ddm86 1/1 Running 0 75s
nginx-deployment-678cb5c9fd-lz8vd 1/1 Running 0 45s
nginx-deployment-678cb5c9fd-n6l7h 1/1 Running 0 45s
nginx-deployment-678cb5c9fd-t7s2b 1/1 Running 0 29m
nginx-deployment-678cb5c9fd-tpskp 1/1 Running 0 30s
nginx-deployment-678cb5c9fd-wgkm4 1/1 Running 0 45s
nginx-deployment-678cb5c9fd-zqppl 1/1 Running 0 75s观察 CPU 使用率（另开终端）
[root@k8s-master01 ~]# kubectl top pods -l app=nginx
NAME CPU(cores) MEMORY(bytes)
nginx-deployment-678cb5c9fd-47wwb 0m 6Mi
nginx-deployment-678cb5c9fd-82tdw 0m 4Mi
nginx-deployment-678cb5c9fd-bb7j7 0m 4Mi
nginx-deployment-678cb5c9fd-ddm86 0m 11Mi
nginx-deployment-678cb5c9fd-lz8vd 0m 4Mi
nginx-deployment-678cb5c9fd-n6l7h 0m 4Mi
nginx-deployment-678cb5c9fd-t7s2b 0m 12Mi
nginx-deployment-678cb5c9fd-tpskp 0m 4Mi
nginx-deployment-678cb5c9fd-wgkm4 0m 4Mi
nginx-deployment-678cb5c9fd-zqppl 0m 4Mi
[root@k8s-master01 ~]#Metrics Server 采集延迟会导致伸缩延迟，生产可接受 1-2 分钟。
[root@k8s-master01 ~]# kubectl get pods -l app=nginx -w
NAME READY STATUS RESTARTS AGE
nginx-deployment-678cb5c9fd-47wwb 1/1 Running 0 76s
nginx-deployment-678cb5c9fd-82tdw 1/1 Running 0 45s
nginx-deployment-678cb5c9fd-bb7j7 1/1 Running 0 30s
nginx-deployment-678cb5c9fd-ddm86 1/1 Running 0 75s
nginx-deployment-678cb5c9fd-lz8vd 1/1 Running 0 45s
nginx-deployment-678cb5c9fd-n6l7h 1/1 Running 0 45s
nginx-deployment-678cb5c9fd-t7s2b 1/1 Running 0 29m
nginx-deployment-678cb5c9fd-tpskp 1/1 Running 0 30s
nginx-deployment-678cb5c9fd-wgkm4 1/1 Running 0 45s
nginx-deployment-678cb5c9fd-zqppl 1/1 Running 0 75s
nginx-deployment-678cb5c9fd-wgkm4 1/1 Terminating 0 5m
nginx-deployment-678cb5c9fd-wgkm4 1/1 Terminating 0 5m
nginx-deployment-678cb5c9fd-wgkm4 0/1 Completed 0 5m1s
nginx-deployment-678cb5c9fd-wgkm4 0/1 Completed 0 5m1s
nginx-deployment-678cb5c9fd-wgkm4 0/1 Completed 0 5m1s
nginx-deployment-678cb5c9fd-ddm86 1/1 Terminating 0 5m45s
nginx-deployment-678cb5c9fd-bb7j7 1/1 Terminating 0 5m
nginx-deployment-678cb5c9fd-lz8vd 1/1 Terminating 0 5m15s
nginx-deployment-678cb5c9fd-82tdw 1/1 Terminating 0 5m15s
nginx-deployment-678cb5c9fd-t7s2b 1/1 Terminating 0 33m
nginx-deployment-678cb5c9fd-n6l7h 1/1 Terminating 0 5m15s
nginx-deployment-678cb5c9fd-zqppl 1/1 Terminating 0 5m45s
nginx-deployment-678cb5c9fd-n6l7h 1/1 Terminating 0 5m15s
nginx-deployment-678cb5c9fd-ddm86 1/1 Terminating 0 5m45s
nginx-deployment-678cb5c9fd-zqppl 1/1 Terminating 0 5m45s
nginx-deployment-678cb5c9fd-t7s2b 1/1 Terminating 0 33m
nginx-deployment-678cb5c9fd-bb7j7 1/1 Terminating 0 5m1s
nginx-deployment-678cb5c9fd-lz8vd 1/1 Terminating 0 5m16s
nginx-deployment-678cb5c9fd-82tdw 1/1 Terminating 0 5m16s
nginx-deployment-678cb5c9fd-n6l7h 0/1 Completed 0 5m16s
nginx-deployment-678cb5c9fd-ddm86 0/1 Completed 0 5m46s
nginx-deployment-678cb5c9fd-zqppl 0/1 Completed 0 5m46s
nginx-deployment-678cb5c9fd-t7s2b 0/1 Completed 0 33m
nginx-deployment-678cb5c9fd-bb7j7 0/1 Completed 0 5m1s
nginx-deployment-678cb5c9fd-lz8vd 0/1 Completed 0 5m16s
nginx-deployment-678cb5c9fd-82tdw 0/1 Completed 0 5m16s
nginx-deployment-678cb5c9fd-t7s2b 0/1 Completed 0 33m
nginx-deployment-678cb5c9fd-t7s2b 0/1 Completed 0 33m
nginx-deployment-678cb5c9fd-n6l7h 0/1 Completed 0 5m17s
nginx-deployment-678cb5c9fd-ddm86 0/1 Completed 0 5m47s
nginx-deployment-678cb5c9fd-ddm86 0/1 Completed 0 5m47s
nginx-deployment-678cb5c9fd-ddm86 0/1 Completed 0 5m47s
nginx-deployment-678cb5c9fd-n6l7h 0/1 Completed 0 5m17s
nginx-deployment-678cb5c9fd-n6l7h 0/1 Completed 0 5m17s
nginx-deployment-678cb5c9fd-zqppl 0/1 Completed 0 5m47s
nginx-deployment-678cb5c9fd-zqppl 0/1 Completed 0 5m47s
nginx-deployment-678cb5c9fd-zqppl 0/1 Completed 0 5m47s
nginx-deployment-678cb5c9fd-bb7j7 0/1 Completed 0 5m2s
nginx-deployment-678cb5c9fd-bb7j7 0/1 Completed 0 5m2s
nginx-deployment-678cb5c9fd-bb7j7 0/1 Completed 0 5m2s
nginx-deployment-678cb5c9fd-82tdw 0/1 Completed 0 5m17s
nginx-deployment-678cb5c9fd-82tdw 0/1 Completed 0 5m17s
nginx-deployment-678cb5c9fd-82tdw 0/1 Completed 0 5m17s
nginx-deployment-678cb5c9fd-lz8vd 0/1 Completed 0 5m17s
nginx-deployment-678cb5c9fd-lz8vd 0/1 Completed 0 5m17s
nginx-deployment-678cb5c9fd-lz8vd 0/1 Completed 0 5m17s停止负载并观察缩容
停止压测命令（Ctrl+C）或者等发完所有请求自动停止负载。HPA 默认 5 分钟稳定窗口 后才会缩容（--horizontal-pod-autoscaler-downscale-stabilization）。
打印出HPA（nginx-hpa）事件日志，命令如下
kubectl describe hpa nginx-hpa | grep -A5 "Events"提示：A5表示显示匹配内容的后5行（After 5）
操作过程
[root@k8s-master01 hpa]# kubectl describe hpa nginx-hpa | grep -A5 "Events"
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal SuccessfulRescale 3m34s horizontal-pod-autoscaler New size: 4; reason: cpu resource utilization (percentage of request) above target
Normal SuccessfulRescale 3m3s horizontal-pod-autoscaler New size: 8; reason: cpu resource utilization (percentage of request) above target
Normal SuccessfulRescale 2m48s horizontal-pod-autoscaler New size: 10; reason: cpu resource utilization (percentage of request) above target
[root@k8s-master01 hpa]#|-------------------|-------------------------|------------------|
| 时间（事件发生顺序） | 扩容结果 | 触发原因 |
| 3分34秒前 | 扩容到 4 个 | CPU 使用率超过了设定的目标值 |
| 3分3秒前（间隔31秒） | 副本数从 4 扩容到 8 个 | CPU 使用率依然超过目标值 |
| 2分48秒前（间隔15秒） | 副本数从 8 扩容到 10 个 | CPU 使用率依然超过目标值 |
扩容非常积极：HPA 默认每 15 秒检查一次指标。第二次到第三次扩容只隔了 15 秒（正好是一个检查周期），说明压测流量（hey）特别猛，CPU 持续飙升，HPA 在全力追赶负载。
扩容步长很大：从 4→8（增加了 100%），再到 8→10。这符合 HPA 默认的扩容策略（每 15 秒允许翻倍或增加 4 个 Pod 取较大值），最终达到 maxReplicas（最大副本数）是 10。
谨慎缩容：需要等待5分钟后，再次查看hpa，副本数才降回 1（最小副本数）。
[root@k8s-master01 ~]# kubectl describe hpa nginx-hpa | grep -A15 "Events"
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal SuccessfulRescale 5m57s horizontal-pod-autoscaler New size: 4; reason: cpu resource utilization (percentage of request) above target
Normal SuccessfulRescale 5m26s horizontal-pod-autoscaler New size: 8; reason: cpu resource utilization (percentage of request) above target
Normal SuccessfulRescale 5m11s horizontal-pod-autoscaler New size: 10; reason: cpu resource utilization (percentage of request) above target
Normal SuccessfulRescale 25s horizontal-pod-autoscaler New size: 3; reason: All metrics below target
Normal SuccessfulRescale 10s horizontal-pod-autoscaler New size: 1; reason: All metrics below target生产级优化与进阶配置
生产环境不能仅依赖 CPU，还要考虑内存、自定义指标（如 QPS），并调整伸缩行为。
多指标 HPA（CPU + 内存）
编辑 hpa-nginx.yaml，增加内存指标：
vi hpa-nginx.yaml增加内容如下
metrics:
- type: Resource
resource:
name: cpu
target:
type: Utilization
averageUtilization: 60
- type: Resource
resource:
name: memory
target:
type: AverageValue
averageValue: 300Mi # 当内存超过 300Mi 时扩容注意：内存目标类型建议用 AverageValue（绝对值）而非 Utilization（因为内存使用率不随副本线性变化）。
调整伸缩行为（冷却时间、步长）
修改 hpa-nginx.yaml，使用 behavior 字段避免频繁伸缩：
behavior:
scaleDown:
stabilizationWindowSeconds: 300 # 缩容等待 5 分钟
policies:
- type: Percent
value: 10 # 每次缩容最多减少 10% 的副本数
periodSeconds: 60
scaleUp:
stabilizationWindowSeconds: 0 # 扩容无需等待
policies:
- type: Percent
value: 100 # 允许一次翻倍扩容
periodSeconds: 15
- type: Pods
value: 4
periodSeconds: 15
selectPolicy: Max完整的hpa-nginx.yaml示例如下
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
name: nginx-hpa
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: nginx-deployment
minReplicas: 1
maxReplicas: 10
metrics:
- type: Resource
resource:
name: cpu
target:
type: Utilization
averageUtilization: 60
- type: Resource
resource:
name: memory
target:
type: AverageValue
averageValue: 300Mi
behavior:
scaleDown:
stabilizationWindowSeconds: 300
policies:
- type: Percent
value: 10
periodSeconds: 60
scaleUp:
stabilizationWindowSeconds: 0
policies:
- type: Percent
value: 100
periodSeconds: 15
- type: Pods
value: 4
periodSeconds: 15
selectPolicy: Max接下来可以做的有：
关键指标监控：部署 Prometheus + Grafana，监控 HPA 决策、Pod 启动延迟、资源碎片。
设置警报：当 HPA 长时间达到 maxReplicas 且 CPU 仍高于目标时，触发告警。
完成！enjoy it！