Subgraphs autoscaling

[DGolubets]

One of our subgraph services has auto-scaling configured (in K8s).
When scaled down, Apollo Router logs errors like fetch_error=SubrequestHttpError { status_code: None, service: "model", reason: "connection closed before message completed" }.

Is there something I can do to handle this more gracefully?
Maybe Apollo Router should be aware of scaling, or the subgraph service should wait longer before terminating?

[Geal]

There is no support for service discovery in the router, but you can look into supporting graceful shutdown in the subgraphs (or in the load balancer in front of them). When a subgraph is going to stop, it should return Connection: close on all responses for a while and stop accepting new connections. Traffic to that subgraph instance will progressively die down and be dispatched to the other instances