From my understanding the new replay server is trying to find the replay stream that matches most players (so if one stream "desyncs" it will not be used).
Furthermore: The old replay server was using blocking IO (do an operation on one thing and wait for it to finish) and causing permanent 100% CPU load on one core. Given the fact that we can not scale to more cores (Python does not offer multithreading) I am pretty sure some problems came from the problem that the server was simply overloaded with work. The new replay server is using asyncronous IO which hopefully will give us a big performance improvements.
Since we cannot desync the game on purpose, the only test is: Does it create working replays?Statistics: Posted by Brutus5000 — 11 Jul 2019, 09:39
]]>