Apache Druid

This is bad idea in my opinion. I dont know how big is your cluster, but in my experience autoscaling brokers would lead to further problems instead of solving the current ones.

on historicals you have `druid.processing.numThreads` and this is max what sum of `druid.broker.http.numConnections`  across all brokers  can be.
So with autoscaling brokers you end up with underutilization of cluster or overloading historicals.

at least would need some way to dynamically adjust those pool sizes so the historicals dont get oversubscribed

with regards to graceful shutdown, there are two configs which are aimed at allowing queries to finish processing on shutdown, `druid.server.http.unannouncePropagationDelay` which is how long after shtudown begins to wait until starting to shutdown the http server, and `druid.server.http.gracefulShutdownTimeout` which is how long the http server will wait to complete any in flight requests