> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bunny.net/llms.txt
> Use this file to discover all available pages before exploring further.

# Graceful Shutdown

> Understand how Magic Containers handles container shutdown with SIGTERM signals and grace periods.

When Magic Containers needs to stop a container, whether during a rolling update, scaling down, or app deletion, it follows a graceful shutdown process to give your application time to clean up resources and complete in-flight requests.

## How it works

The shutdown process follows these steps:

1. **SIGTERM signal** - Magic Containers sends a `SIGTERM` signal to your container's main process (PID 1), notifying it to begin a graceful shutdown.
2. **Grace period** - Your application has a configurable grace period (default 30 seconds for new applications) to handle the signal, close connections, flush data, and exit cleanly.
3. **SIGKILL signal** - If the process is still running after the grace period, Magic Containers sends a `SIGKILL` signal to forcefully terminate the container.

<Info>
  The 30-second grace period matches Kubernetes' default termination behavior, allowing applications designed for Kubernetes to work seamlessly on Magic Containers.
</Info>

<Warning>
  Applications created before February 1st, 2026 have a 1-second grace period. Applications created after this date have a 30-second grace period. To check or update your application's grace period, use the `terminationGracePeriodSeconds` field via the API (`PUT /apps/APP_ID/`).
</Warning>

## Handling SIGTERM in your application

To ensure a clean shutdown, your application should:

* **Listen for SIGTERM** - Register a signal handler to catch the termination signal.
* **Stop accepting new requests** - Prevent new work from starting during shutdown.
* **Complete in-flight work** - Finish processing active requests or transactions.
* **Close connections** - Gracefully close database connections, file handles, and network sockets.
* **Flush data** - Write any buffered data to persistent storage.
* **Exit cleanly** - Terminate the process with a zero exit code.

## Tips

* **Keep shutdown fast** - While you have 30 seconds, aim to shut down as quickly as possible. Faster shutdowns reduce deployment times during rolling updates.
* **Use health checks** - Configure readiness health checks so traffic stops flowing to your container before the shutdown signal is sent.
* **Test your shutdown logic** - Send `SIGTERM` to your container locally to verify it handles the signal correctly and exits within the grace period.