Azure容器实例组中容器频繁重启问题排查求助

Hey Ross, sorry to hear your Spring + React app is acting up on Azure Container Instances—local runs working fine but constant restarts are definitely frustrating. Let’s walk through some common culprits and how to dig into them:

Azure Container Instances come with default CPU/memory quotas that might not be enough for your app, especially during startup when Spring Boot is initializing and serving React static assets.

• Check your current ACI resource allocation with this CLI command:

  az container show --resource-group <your-resource-group> --name <your-container-group> --query 'containers[].resources'
  

• Try temporarily bumping up the CPU and memory (e.g., from 1 CPU/1GB RAM to 2 CPU/2GB RAM). If the restarts stop, you’ll know the issue was insufficient resources causing an OOM (Out of Memory) kill or startup timeout.

ACI will restart containers if it thinks they’ve failed to start properly. If your app takes a while to boot up (common with Spring Boot apps), tight health probe settings can trigger unnecessary restarts.

• Check if you have Liveness or Readiness Probes configured: If the initialDelaySeconds is too short (e.g., 10 seconds), the probe might mark the app as unhealthy before it’s fully started.
  • Adjust these values to match your app’s actual startup time (try 60+ seconds for initialDelaySeconds, and tweak periodSeconds/failureThreshold accordingly).
• Even without custom probes, ACI’s default checks might time out if your app’s startup takes longer than expected. Extending the startup timeout window in your ACI configuration can help.

Local vs. ACI environments often have subtle differences in environment variables or config files that can cause the app to crash silently (even if your app logs look "normal").

• Compare ACI’s environment variables to your local setup with this command:

  az container show --resource-group <your-resource-group> --name <your-container-group> --query 'containers[].environmentVariables'
  

  Pay close attention to database connections, API endpoints, or any sensitive configs that might be missing or incorrect in ACI.
• Verify that your Spring Boot application.properties/application.yml is being loaded correctly in the container—double-check file paths and ensure no critical configs are missing.

Even if your app logs look fine, the container might be exiting due to system-level issues that aren’t captured in your app’s logs.

• Grab the container’s exit code and recent logs:

  az container logs --resource-group <your-resource-group> --name <your-container-group> --container-name <your-container> --tail 100
  

  Common exit codes to watch for:
  • 1: App crashed due to an unhandled error
  • 137: Container was killed due to OOM
  • 143: Graceful termination, but ACI is configured to restart it
• Check ACI’s system logs in the Azure Portal: Navigate to your container group > Logs > Container Instance Logs. This will show you ACI’s own messages, like "Container restarted due to health probe failure" or "Insufficient memory to run container".

Mismatched port configurations or network restrictions can sometimes cause the app to crash, even if it’s not the most obvious culprit.

• Confirm that:
  • The EXPOSE port in your Dockerfile matches the port configured in ACI
  • Your Spring Boot server.port setting aligns with both of the above
• If you can exec into the container, run netstat -tulpn (if the tool is available) to verify the app is actually listening on the expected port.

It’s possible your Docker image is missing dependencies or React assets weren’t packaged correctly, leading to silent failures after startup.

• Double-check your Dockerfile: Ensure React’s build directory contents are copied into Spring Boot’s src/main/resources/static or target/classes/static folder (depending on your multi-stage build setup).
• Test the image locally: Run the exact same Docker image you’re using in ACI on your local machine, with the same environment variables. If it restarts locally too, you’ll know the issue is with the image itself, not ACI.

内容的提问来源于stack exchange，提问作者Ross