For Agones: install Agones into a Kubernetes cluster via Helm, then define a Fleet resource specifying the game server container image, port config, and desired replica count; Agones manages GameServer pod lifecycle.
Integrate the Agones SDK into your game server binary: call sdk.Ready() when the server is ready to accept players, sdk.Allocate() to mark it as in-use, and sdk.Shutdown() when the match ends to trigger pod replacement.
For GameLift: package your server binary as a build artifact, upload it via the GameLift console or CLI, and create a Fleet specifying EC2 instance type, runtime configuration (launch path, parameters), and scaling limits.
Configure auto-scaling: in Agones use a FleetAutoscaler (buffer or webhook policy) to maintain a warm pool of Ready servers; in GameLift configure scaling policies based on available game sessions or CPU utilization.
Allocate servers through the matchmaker: Agones exposes a GameServerAllocation API that atomically claims a Ready server; GameLift uses CreateGameSession which triggers placement via queues and FlexMatch.
Monitor fleet health using Agones metrics (exported to Prometheus) or GameLift CloudWatch metrics; set alerts on low Ready server counts to trigger scale-out before the pool is exhausted.
Known gotchas
Agones requires Kubernetes node pool sizing to account for both the game server workload and Kubernetes system overhead; undersized nodes cause scheduling failures under load.
GameLift spot instances significantly reduce cost but introduce interruption risk; design your server to checkpoint state and use a mix of on-demand and spot capacity for the fleet.
Neither platform auto-scales to zero; maintain a minimum warm pool to avoid cold-start latency when players arrive in bursts after the fleet has scaled down.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp