Closed jkowalski closed 5 years ago
@jkowalski can we close this ticket now? I feel like we have acceptable performance at this stage?
Going to close this ticket, as I think we've reached out goals here for now.
Please feel free to reopen if we need to.
Currently when spinning up a large test fleet (10K GameServers) the Game Servers and pods are created relatively slowly (about 1 per second, measured on regional GKE cluster with 150 nodes * (8 cores + 8 GB RAM each)).
I think we're unnecessarily serializing something - in principle there should be no reason why this would be significantly slower than a deployment (which creates all pods almost instantaneously and they run in parallel).