Best Practices for TCP Connection Management on EC2
Well this is a really crappy thing for AWS to mess around with, and then hide the announcement on a "best practices" page:
"With sixth-generation AWS Nitro (Nitro V6) instances, launched in June 2025 [c8, r8, etc], the default TCP connection tracking idle timeout changed from 432,000 seconds (5 days) to 350 seconds. Applications that hold idle connections open for long periods, such as [uhhh pretty much everything built on TCP - jm] may experience unexpected connection drops after migrating to these instances."
They go on to recommend that you "implement keepalives and connection lifecycle management", which is great fun if you don't control the code implementing your TCP-based network protocols. This is a very fundamental change for many protocols so it'll be fun dealing with it.
Kudos to Adam C in the ITC Slack for spotting this a while back.
Tags: networking protocols tcp idle-timeouts aws architecture nitro conntrack idle-connections