Why doesn't ASP.NET gracefully shutdown during Linux App Service scale-down?

The behavior you’re seeing on Linux App Service scale-downs is a known quirk of the platform rather than an issue with your ASP.NET shutdown code. In Linux web apps, during scale-in operations the infrastructure may remove instances abruptly without giving them time to run their graceful shutdown routines (i.e. no time for your IHostedService.StopAsync or host lifetime cancellation callbacks to run) even if you’ve configured WEBSITES_CONTAINER_STOP_TIME_LIMIT or HostOptions.ShutdownTimeout.

Possible Workarounds

1. Custom Signal Handling in a Container:
   If you’re running your app in a custom container, you can intercept termination signals (SIGTERM, SIGINT) with an entrypoint script that delays the container’s exit long enough for your graceful shutdown code to run. For example:

   BASHCopied
   #!/bin/bash
   # entrypoint.sh
   function shutdown_handler() {
       echo "Shutdown signal received. Waiting for graceful shutdown..."
       # Wait for a period to allow the app to finish its shutdown
       sleep 120
   }
   
   # Trap the SIGTERM signal and execute handler
   trap shutdown_handler SIGTERM
   
   # Run the app (assumes dotnet YourApp.dll)
   dotnet YourApp.dll

   Then update your Dockerfile to use this script:

   DOCKERFILECopied
   FROM mcr.microsoft.com/dotnet/aspnet:6.0
   WORKDIR /app
   COPY . ./
   RUN chmod +x entrypoint.sh
   ENTRYPOINT ["./entrypoint.sh"]

   This “hacky” approach gives the container extra time after the SIGTERM before it is killed by the host, affording your application a chance to call StopAsync and run cancellation callbacks.

2. Consider a Custom Scale-in Strategy:
   If you’re not using a custom container, one option (though less elegant) is to architect your scaling strategy so that scale-down events can be managed more gracefully. For example, you might manually remove instances via a script or automation that first puts the instance into a drain state (allowing it to finish processing) before removing it from the load balancer.

Summary

Unfortunately, when scaling down Linux App Service Plans, the underlying platform can bypass graceful shutdown hooks that work during deployments or for manual stops. Using custom container signal handling (or similar orchestration workarounds) is currently the best “hacky” solution to ensure that your shutdown code runs as expected.