Game Server Deploy
A cost-efficient, multi-game dedicated server platform on AWS Fargate with a local management UI and a fully serverless Discord bot. Servers only run — and only cost money — while somebody is playing; the rest of the time the account is silent except for a handful of near-free Lambdas.
What you get
- On-demand Fargate tasks per game (no persistent ECS service, no idle cost).
- EFS with per-game access points so each world is isolated yet persistent.
- Auto-DNS — a Lambda UPSERTs
{game}.{yourdomain}on task start and DELETEs it on stop. Optional ALB with ACM for HTTPS-fronted games. - Watchdog Lambda that stops servers after a configurable idle window.
- Local Nest.js + React dashboard to start/stop, edit config, stream logs, and track costs.
- Serverless Discord bot — two Node.js Lambdas plus DynamoDB and Secrets Manager handle every slash command; no 24/7 bot process.
Choose your path
The rest of the site is organised around three roles. Pick the one that matches what you're doing right now:
- Setup guide — from a blank AWS account to a running game server, in order.
- User guide — day-to-day operation: starting/stopping servers from the dashboard or Discord, reading the cost panel, checking logs.
- Maintainer guide — working on the code: monorepo layout, tests, lint, CI, release/deploy mechanics, load-bearing invariants not to break.
- Submodule guide — the
pattern of wrapping this repo as a git submodule inside a private parent
repo that holds
terraform.tfvars,server_config.json, state, and anything else secret.
Component reference
Deep-dives on each piece, for when the guides hand-wave past something:
- Architecture overview — the diagram below in full, with every arrow annotated.
- Terraform — every
.tffile, variables, outputs, and AWS services touched. - Management app — the
Nest.js API, React dashboard, and
@gsd/sharedlibrary. - Lambdas — the four Node.js Lambdas (interactions, followup, update-dns, watchdog).
High-level architecture
Run Terraform from your laptop (or a CI runner); after that the control plane is the dashboard, the Discord bot, or EventBridge schedules. No persistent server process — only the Fargate tasks themselves cost money, and only while they're running.
For the full breakdown (Discord bot internals, control-loop Lambdas, and
the /server-start sequence) see the
architecture overview.
Repository map
game-server-deploy/
├── app/ # Nest.js + React monorepo (npm workspaces)
│ └── packages/
│ ├── shared/ # @gsd/shared
│ ├── server/ # @gsd/server (Nest.js API)
│ ├── web/ # @gsd/web (React + Vite)
│ └── lambda/
│ ├── interactions/ # Discord Function URL entry point
│ ├── followup/ # async ECS work + Discord PATCH
│ ├── update-dns/ # Route 53 + ALB on task state change
│ └ ── watchdog/ # idle detection + auto-stop
├── terraform/ # all AWS infra (VPC, ECS, EFS, Lambdas, DDB...)
├── docs/ # this site
├── Dockerfile # containerised management app
├── docker-compose.yml
├── setup.sh # first-time bootstrap — Linux / macOS
├── setup.ps1 # first-time bootstrap — Windows (PowerShell)
└── README.md