Graceful shutdown implementation

[avlazarov]

👋

Right now, in-flight requests will get HTTP error 500 when sending TERM or INT signals to Falcon. On our K8s setup this would mean hundreds of client errors on each deploy (or any shutdown of pods running Falcon). We're trying to mitigate the issue by giving Falcon an artificial 60 seconds delay before sending the KILL signal. In these 60 seconds, we'll remove the server from the list of backends available to the reverse proxy. Waiting the entire 60 seconds, even if there are no in-flight request, is not ideal. We don't have a similar problem with Puma - it will stop accepting new requests after receiving a signal, then wait for in-flight requests to complete, and then shutdown.

Is there a better way to achieve a graceful shutdown in Falcon, via falcon serve/host?