在EKS集群运行Selenium测试遇CrashLoopBackOff问题求助

我将Java代码容器化后部署到EKS 1.24托管节点组集群，已构建Docker镜像并完成K8s部署，但Pod始终处于CrashLoopBackOff状态。工作节点挂载了卷可查看index.html，但测试执行时出现连接错误。

核心报错信息 #

Caused by: java.net.ConnectException: Failed to connect to localhost/[0:0:0:0:0:0:0:1]:4444
at okhttp3.internal.connection.RealConnection.connectSocket(RealConnection.java:242)
at okhttp3.internal.connection.RealConnection.connect(RealConnection.java:160)

补充环境信息 #

• 已部署Selenium Grid服务，通过ALB Ingress Controller暴露，可通过AWS提供的外部URL访问带单节点Chrome的Selenium Grid Hub
• Ingress地址：http://internal-k8s-devops-path-of-the-adddress
• 配置变量：
  • k8s_host: my-alb-ingress-address
  • selenium_grid_host: my-grid-hub-service-name:4444
  • MODULE: search-module.xml

Deployment配置文件（k8s-deployment.yaml） #

apiVersion: apps/v1
kind: Deployment
metadata:
  name: selenium-grid-k8s-autoscaler-deployment
  namespace: devops-java-selenium
  labels:
    app: selenium-grid-k8s-autoscaler
    name: selenium-grid-k8s-autoscaler
spec:
  replicas: 1
  selector:
    matchLabels:
      app: selenium-grid-k8s-autoscaler
  template:
    metadata:
      labels:
        app: selenium-grid-k8s-autoscaler
        name: selenium-grid-k8s-autoscaler
    spec:
      containers:
        - name: selenium-grid-k8s-autoscaler
          image: my-app-image
          env:
            - name: k8s_host
              value: my-alb-ingress-address
            - name: selenium_grid_host
              value: selenium-grid-hub:4444
            - name: MODULE
              value: search-module.xml
          volumeMounts:
              - mountPath: /usr/share/selenium/test-output
                name: devops-java-selenium-pv
      volumes:
        - name: devops-java-selenium-pv
          hostPath:
             path: /tmp/app

应用Pod日志 #

Jul 10, 2023 6:16:39 AM org.openqa.selenium.remote.DesiredCapabilities chrome
INFO: Using `new ChromeOptions()` is preferred to `DesiredCapabilities.chrome()`
Jul 10, 2023 6:16:39 AM org.openqa.selenium.remote.DesiredCapabilities chrome
INFO: Using `new ChromeOptions()` is preferred to `DesiredCapabilities.chrome()`
Jul 10, 2023 6:16:39 AM org.openqa.selenium.remote.DesiredCapabilities chrome
INFO: Using `new ChromeOptions()` is preferred to `DesiredCapabilities.chrome()`
Jul 10, 2023 6:16:39 AM org.openqa.selenium.remote.DesiredCapabilities chrome
INFO: Using `new ChromeOptions()` is preferred to `DesiredCapabilities.chrome()`

===============================================
search-module
Total tests run: 4, Failures: 0, Skips: 4
Configuration Failures: 4, Skips: 4
===============================================

排查步骤 #

1. 验证环境变量读取正确性

   • 进入应用Pod内部执行echo $selenium_grid_host，确认输出为selenium-grid-hub:4444而非localhost。如果Java代码硬编码了localhost:4444，需修改代码改为读取环境变量selenium_grid_host。

2. 检查Selenium Grid服务可达性

   • 在应用Pod内部执行curl selenium-grid-hub:4444/wd/hub/status，验证是否能连通Selenium Hub：
     • 若不通，确认Selenium Grid服务selenium-grid-hub是否和应用Pod在同一命名空间（devops-java-selenium），不在同一命名空间需使用跨命名访问格式selenium-grid-hub.<namespace>:4444。
     • 检查Selenium Hub服务的端口配置，确保暴露的是4444端口。
     • 确认Selenium Hub Pod本身运行正常，无Crash或重启情况。

3. 排查网络策略限制

   • 检查当前命名空间是否存在NetworkPolicy，确认没有限制应用Pod访问Selenium Grid服务4444端口的规则。

4. 验证hostPath卷挂载权限

   • 进入应用Pod执行ls -ld /usr/share/selenium/test-output，确认目录存在且容器进程有读写权限。必要时在Dockerfile中添加目录权限配置（如RUN mkdir -p /usr/share/selenium/test-output && chown -R <user>:<group> /usr/share/selenium/test-output）。

5. 查看Pod退出细节

   • 执行kubectl describe pod <pod-name> -n devops-java-selenium，查看Events中的退出原因，确认是否因4个测试配置失败导致进程退出，进而引发CrashLoopBackOff。

内容的提问来源于stack exchange，提问作者Gowmi