CS2 moves to our own cloud
GameBling match servers now run on our own Kubernetes clusters on real public IPs — always on, self-healing, with the referee, the money HUD, and demo recording built into every room. The era of the gaming PC under the desk is over.
Every wagered CS2 match on GameBling runs on a server we control — that is the integrity model. The server is the referee: it whitelists exactly the staked players, reports every kill over a signed channel, records the demo, and settles the money. But "a server we control" has, until this week, meant something less glamorous than it sounds: a very fast gaming PC on a home connection, kept alive through a tunnel because residential internet does not hand out public addresses.
It worked, and it taught us everything about the plugin stack. It was also one power flicker away from voiding a live match. A wagering floor’s game host should have the same uptime posture as its ledger. So this week, CS2 match hosting moved to GameBling’s own cloud: our Kubernetes clusters, running on dedicated hardware in a datacenter, on real public IP addresses. No tunnel, no relay, no gaming PC. The machine that referees your money now lives in the same operational world as the money itself.
What a match room looks like now
Each CS2 room is a self-contained, self-healing workload. When a room starts, it assembles itself from the ground up: the dedicated server installs, the referee layer goes in — the tournament plugin that runs the whitelist and match flow — followed by our own plugins: the money HUD that paints your live wallet on screen, and the proximity-voice layer. The room patches itself together, binds to its public address, and reports for duty. If a CS2 update lands or something crashes, the room rebuilds itself the same way, unattended, in minutes.
Because the build recipe is code, "add another server" stopped being a hardware purchase. A second room is a copy of a dozen lines pointed at another port; a busy Saturday night fleet is that, repeated. Every room carries resource fences so a match can never starve the cluster’s control machinery, and the cluster watches each room’s health on its public address — the same address you connect to — so what the orchestrator sees is what players get.
Direct connections, real addresses
The quiet win is the network path. On the old host, RCON — the control channel the platform uses to configure matches and push wallet updates — traveled through a tunnel, and players connected through the same indirection. Every layer of indirection is a failure mode and a few milliseconds. Now the control plane talks straight to the server’s public IP, and so do you: the CONNECT command on your match ticket points at the actual machine in the actual datacenter. Fewer moving parts between your crosshair and the ledger.
The move also unlocked the features that shipped alongside it. Automatic clip detection wants demos on storage the platform can reach; the render pipeline we are building wants cluster jobs next to those demos; per-kill settlement wants an RCON path that never flakes mid-round. One infrastructure decision, several products.
The bigger picture
GameBling’s rule has always been that a game only carries wagers when we can prove its results — our servers, or the publisher’s official records. Moving our servers into our own cloud is that rule taken seriously at the infrastructure layer: provable results deserve production-grade referees. The old machine goes back to being what it was always meant to be — someone’s gaming PC — with a standing role as an emergency LAN fallback and our thanks for its service.
Matches are running on the cloud fleet now. You will not see any of this — which is exactly the point. The floor just works, at any hour, and the referee never sleeps.