Linux fleets can't support as many processes as Windows - how to debug?

Hi there,

I made a question earlier, but perhaps it was on the wrong forum or something:

After investigating, it seems my actual question is - how do I debug what’s causing Linux fleets to fail to activate at higher process counts?

We’re able to get at least 30 processes of the server running on Windows, but any higher than 10 on Linux and the fleet will get stuck in Activating for a while (then error), presumably because something went wrong starting the server processes.

I was able to SSH into one of the errorred instances but there didn’t seem to be any game logs, so perhaps something went wrong very early in the servers starting up?

It’d be great if we’re able to fit more than 10 processes per machine on Linux, since the 2x cost savings are outweighed by 1/3 efficiency.

I can imagine that UE4 might well be less optimised on Linux, so fewer processes can fit in memory, but would be great to know what the problem is specifically :slight_smile:


Hi there @geordie

Try using a (much) larger instance type to activate. If this succeeds, then it is likely that the performance of the instance is being maxed out during the fleet activation.

You could also start the fleet using fewer concurrent processes, then when the fleet is active, Edit Fleet, and increase until you see the same problems, then back it down a bit. Don’t forget to bounce the fleet (scale it to zero and back up again) when you change the number of concurrent processes.

My theory hinges on the idea that if the processes are taking up too much time, then AuxProxy, the on-instance daemon that controls GameLift operations on the box can get too little time, which for instance could hold up Health checks to the degree that things start falling apart.

Let me know how it goes, and we will get you going again!

Al Murray :slight_smile: Solutions Architect, Amazon Game Tech

Thanks! Will try and get back to you.