docs: add infrastructure, build pipeline, and dev box sections to CLAUDE.md
All checks were successful
Build & Release / build (push) Successful in 15s

This commit is contained in:
Ryan Moon
2026-04-04 10:30:40 -05:00
parent b318345fdf
commit 7aa81c4e7c

112
CLAUDE.md
View File

@@ -63,3 +63,115 @@
- API routes are thin — validate with Zod, call a service, return result
- All financial events must be auditable (append-only audit records)
- JSON structured logging with request IDs on every log line
---
## Infrastructure
### Overview
LunarFront runs on DigitalOcean Kubernetes (DOKS). Each customer gets an isolated namespace, database, and Helm release managed by ArgoCD.
### Key Services
- **Cluster:** `lunarfront` DOKS cluster, region NYC
- **Registry:** `registry.digitalocean.com/lunarfront` (DOCR) — stores Docker images and Helm charts
- **Git:** `git.lunarfront.tech` — self-hosted Gitea, source of truth for code and charts
- **CI:** Gitea Actions — builds Docker images and Helm charts on push to `main`
- **CD:** ArgoCD at `argocd.lunarfront.tech` — deploys from `lunarfront-charts` repo
- **Database:** DO Managed PostgreSQL — one database per customer, plus manager DB
- **Cache/Queue:** DO Managed Valkey — shared across all customers (key-prefixed per customer)
- **Ingress:** nginx ingress controller with Cloudflare proxy in front
- **TLS:** cert-manager with Let's Encrypt (letsencrypt-prod cluster issuer)
- **DNS:** Cloudflare — wildcard `*.lunarfront.tech` → cluster LB IP `167.99.21.170`
### Node Pools
- `system` — 2x s-2vcpu-4gb, runs ingress, ArgoCD, manager, pgbouncer
- `customers` — autoscales 0→N, s-4vcpu-8gb, runs customer app pods (tainted `role=customer`)
- `dev` — autoscales 0→1, s-4vcpu-8gb, runs dev pod only (tainted `dedicated=dev:NoSchedule`)
### Repos
- `lunarfront-app` — main application code (this repo)
- `lunarfront-charts` — Helm charts and ArgoCD app definitions
- `lunarfront-infra` — Terraform for DO infrastructure (DOKS, managed DBs, registry, DNS)
- `lunarfront-manager` — internal ops tool for provisioning/deprovisioning customers
---
## Build & Deploy Pipeline
### How it works
1. Push code to `main` on `lunarfront-app`
2. Gitea Actions runs `.gitea/workflows/build.yml`:
- Builds `lunarfront-app` Docker image → pushes as `0.1.{run_number}`, `{sha}`, `latest`
- Builds `lunarfront-frontend` Docker image → same tags
- Packages Helm chart → pushes as `0.1.{run_number}` to DOCR OCI registry
3. ArgoCD image updater detects new image digests → updates customer deployments
4. New customer provisions always get the latest chart version (queried from DOCR at provision time)
5. Existing customers upgraded via `POST /customers/:slug/upgrade` or `POST /customers/upgrade-all` in the manager
### Versioning
- Version format: `0.1.{gitea_run_number}` — always incrementing, no git commit-back needed
- No version stored in git — source of truth is DOCR tags
- Chart version and app version are kept in sync
### Key files
- `Dockerfile` — backend image (bun runtime, runs `packages/backend/src/main.ts` directly)
- `Dockerfile.frontend` — frontend nginx image
- `chart/` — Helm chart for customer app deployments
- `.gitea/workflows/build.yml` — CI pipeline
- `.gitea/workflows/build-devpod.yml` — builds dev box image on Dockerfile.devpod changes
---
## Dev Box
### What it is
A persistent development pod running in the `dev` namespace on the cluster. Provides a full remote dev environment accessible from anywhere.
- **VS Code in browser:** `dev.lunarfront.tech` (Cloudflare Access protected, OTP to email)
- **SSH:** `ssh -p 2222 root@dev-ssh.lunarfront.tech`
- **Storage:** 100GB DO block storage PVC mounted at `/root` — everything in home dir persists
- **Image:** `registry.digitalocean.com/lunarfront/manager:devpod-latest`
- **Tools:** bun, Claude Code CLI, code-server, kubectl, helm, k9s, doctl, psql, redis-cli, git
### Managing the dev pod
```bash
# Scale up (provisions node automatically)
kubectl scale deployment dev -n dev --replicas=1
# Scale down (node auto-terminates after ~15 min)
kubectl scale deployment dev -n dev --replicas=0
```
### Running the app locally on the dev box (no containers)
The dev box runs the app as plain Bun processes, connecting to the same DO managed services as production.
**Required env vars** (create a `.env` file in the repo root or export in `.bashrc`):
```bash
DATABASE_URL=postgresql://... # DO managed postgres, lunarfront database
REDIS_URL=rediss://... # DO managed valkey
JWT_SECRET=... # any random hex string for local dev
PORT=8000
```
**Start the app:**
```bash
cd ~/lunarfront-app
bun run dev
```
Access the running backend at `dev.lunarfront.tech/proxy/8000/` in the browser (code-server proxy), or via SSH port forward:
```bash
ssh -p 2222 -L 8000:localhost:8000 root@dev-ssh.lunarfront.tech
```
**Run migrations against the dev database:**
```bash
bunx drizzle-kit migrate
```
### Workflow
1. Edit code in VS Code at `dev.lunarfront.tech` or via SSH
2. Run and test locally with `bun run dev` — app connects to DO managed postgres/valkey
3. Push to `main` → Gitea Actions builds and pushes new Docker image + Helm chart
4. ArgoCD deploys to the cluster automatically
5. Use manager at `manager.lunarfront.tech` to upgrade customer instances if needed