I don't yet have a working bug-reproducing scenario (working on it), but under heavy load (that's slightly above 200 requests per second) spring actuator's health endpoint occasionally gives:
Error [reactor.netty.ReactorNetty$InternalNettyException: java.nio.channels.ClosedChannelException] for HTTP GET "/actuator/health", but ServerHttpResponse already committed (200 OK)
There's no additional stack trace or other error message in the output/logs.
The consequence is that the connection to the client is closed server-side before any response is sent out. When running in a pod in kubernetes, this causes the container to be rebooted. This is annoying and resource-wasting, when happening once for each container every (very) few minutes.
The project uses spring boot 2.2.2.RELEASE and spring cloud Hoxton.SR1, and is built using spring-webflux.
Any ideas why?