Connections are immediately saturated on long running servers

[anonymous_user_8ca9ab1b]

We’ve observed an issue where after 2-3 days of uptime our server will accept new connections and replicate the player controller for that connection but will then immediately always return false for Connection->IsNetReady(0) in UNetDriver::ServerReplicateActors causing our other actors replication to be permanently deferred. I’ve added additional logging and will update this ticket if I determine if the queuedbytes or sendbuffer are growing too large or are in some sort of bad state but I’d like to know if Epic have observed this before?

We are on 4.11.0 but will be moving to 4.12 in the near future.

[anonymous_user_8ca9ab1b]

My tickrate was too high and the elapsed time is stored in a float causing it to accumulate precision errors. I did attempt to change the elapsed time to a double and that fixed the replication issue but led to several other networking issues on servers that had run for more than a few days, probably due to accumulated precision errors in other parts of the code. Lowering the tickrate delays these issues for long enough for it to be a non issue for our project.