While monitoring game sessions during a fleet scaling test, it appeared as though the total active player session count lagged behind the sum reported by individual game sessions.
Reloading the entire browser window also showed these numbers to be out of sync. This behavior appeared to coincide with a delay in the desired instances based on our autoscale settings. Is it possible I am missing an API call? Are the numbers reported by the dashboard unreliable?
We’re trying to make servers available quickly enough for an increasing player count to rarely have to wait for an instance to spin-up. I’m trying to identify any bottlenecks that could slow down the time to spin up an active instance.
Here’s another screenshot with out of sync values. This is after one of our server processes crashes and I believe the others were autoscaled into oblivion:
The delay on these number is somewhat alarming, especially if scaling rules are being applied based on #available player sessions, which isn’t being updated as a function of instances being scaled down. This, in theory, could scale down a whole fleet to zero if a simple scaling policy of “Scale down by 1 instance if available sessions > 20 for 1 minute”.