Azure实例缩容（Scale-in）引发5XX请求响应问题咨询

Absolutely, this is a super common pain point when scaling down Azure services — your guess is spot-on! When a scale-in operation triggers, if Azure terminates an instance before it finishes processing in-flight requests, the load balancer or gateway will throw a 502 Bad Gateway error because it can't get a valid response from the (now offline or shutting-down) instance.

Below are targeted fixes for different Azure services to resolve this:

1. For Azure App Service #

• Enable Graceful Shutdown: This feature gives your app a buffer period (default 120s, customizable) to wrap up ongoing requests after receiving a termination signal. You can turn it on via the Azure Portal: go to your App Service > Configuration > General Settings, or use this Azure CLI command:

  az webapp config set --name <your-app-name> --resource-group <your-rg> --graceful-shutdown-timeout 180
  

• Set Up Health Checks: Configure a health check endpoint (like /health that returns 200 OK) for your app. Azure will only remove the instance from the load balancer pool once the health check fails, ensuring no new requests are sent to it before shutdown. Find this in Portal > Monitoring > Health Check.

2. For Azure Kubernetes Service (AKS) #

• Configure Pod Graceful Termination: In your Deployment YAML, set terminationGracePeriodSeconds to give your app time to finish requests. Also, make sure your app listens for the SIGTERM signal to stop accepting new requests and wrap up existing work. Example snippet:

  spec:
    template:
      spec:
        containers:
        - name: your-app-container
          image: your-app-image
          terminationGracePeriodSeconds: 60
  

• Use Pod Disruption Budgets (PDB): Create a PDB to limit how many pods can be terminated at once, preventing sudden large-scale instance drops that overwhelm remaining instances. Example:

  apiVersion: policy/v1
  kind: PodDisruptionBudget
  metadata:
    name: your-app-pdb
  spec:
    minAvailable: 2
    selector:
      matchLabels:
        app: your-app
  

3. For Azure VM Scale Sets #

• Enable Connection Draining on Load Balancer: In the load balancer linked to your VMSS, turn on connection draining. This tells the load balancer to redirect ongoing connections to healthy instances before removing the scaling-down VM. Edit your load balancing rule in the Portal and enable this with a custom timeout.
• Add a Custom Termination Script: Use VMSS extensions to run a script that notifies the load balancer to stop sending traffic, waits for in-flight requests to finish, then shuts down the app and system cleanly.

If you can share more details about that "strange" behavior you mentioned, we can dig deeper — but the above fixes should cover most scale-in related 502 scenarios.

内容的提问来源于stack exchange，提问作者Zenka