Help troubleshooting RST from akka-http 10.1.12

Play Framework
1

I am using akka-http 10.1.12 via play 2.8.2. I am not using nginx or any other proxy in front of play - I have an AWS Application Load Balancer directly calling the play service. I’m trying to debug spikes of 502s that I am getting from the load balancer (when other services are calling this service). I have provided load balancer logs and tcpdump captures from the instance to AWS support and they say that the load balancer is receiving a slew of RST from the service/instance.

The relevant configs are as follows:

akka.http.server {
    max-connections = 4096
    backlog = 4096
    parsing {
      max-content-length = 4 MiB
    }
  }
}

play.server {
  pidfile.path = /dev/null

  http.idleTimeout = 65 seconds

  akka {
    requestTimeout = 60 seconds
  }
}

The load balancer idle timeout is set for 60 seconds.

An example packet capture from the instance which shows the service sending a bunch of RST:

pcapexample
pcapexample1481×725 278 KB

Any help is greatly appreciated.

2

In case this helps anyone in the future, our docker container was being killed (oom) and restarted - we did not realize this at first. It looks like the above RST are being sent from the OS (when the application is killed), not from akka-http.