docs/AWS_DEPLOYMENT.md
Rama main · réplica de lectura del repositorio original.
# AWS Deployment (EC2 + Elastic IP + Cloudflare + HTTPS)
This deploys Seedcorn to a **single EC2 instance** — Postgres, gunicorn, and nginx all in
Docker on the box, behind an **elastic IP**, with DNS and TLS on **Cloudflare**. Cloudflare's
proxy terminates browser-facing HTTPS at its edge; nginx presents a **Cloudflare Origin
Certificate** to secure the Cloudflare→origin leg. Infrastructure is managed with **Terraform**.
It's sized for a small team now and has a clear path to ~3000 users (see
[Scaling](#scaling-to-3000-users)).
```
Cloudflare proxy (timeline.example.com → Elastic IP) — edge TLS, Full (strict)
│ HTTPS
▼
EC2 instance ── nginx (Origin Cert, SPA + reverse proxy) ── gunicorn (Django) ── Postgres
(containers, EBS volume)
```
## Prerequisites
- An **AWS account** and the AWS CLI configured (`aws configure`) — Terraform uses these creds.
- **Terraform ≥ 1.5** installed locally.
- An **SSH key pair** locally (e.g. `~/.ssh/id_ed25519[.pub]`).
- The **Cloudflare** domain (`timeline.example.com`) you control.
## 1. Provision the infrastructure (Terraform)
```bash
cd terraform
cp terraform.tfvars.example terraform.tfvars
# Edit terraform.tfvars: set allowed_ssh_cidr to YOUR_IP/32, confirm region/instance_type/key path.
terraform init
terraform apply
```
Terraform creates a VPC + public subnet + internet gateway, a security group (SSH from your IP,
80/443 open), an **Amazon Linux 2023** instance (Docker + Compose + git installed via cloud-init),
a **gp3 EBS root volume** (encrypted), and an **elastic IP**. Outputs include the IP and next steps:
```bash
terraform output public_ip # e.g. 52.x.x.x
terraform output next_steps
```
## 2. DNS (Cloudflare)
Create an **A record**: `timeline.example.com → <elastic IP>`.
> While testing, you can leave the record **DNS only** (grey cloud) to hit the origin directly.
> Once it's working, switch the proxy **on** (orange cloud) and set Cloudflare's SSL/TLS mode to
> **Full (strict)** so the edge validates the origin certificate.
>
> Note: Cloudflare's SSL/TLS encryption mode is **zone-wide**, so switching to **Full (strict)**
> affects *every* proxied hostname in the zone — make sure all of them present a valid origin
> cert first.
## 3. Deploy the app
```bash
ssh ec2-user@<elastic IP>
cd /opt/timeline
git clone https://github.com/vakahnke/seedcorn.git .
cp .env.example .env
```
Edit **`.env`** for production:
```bash
DJANGO_DEBUG=0
DJANGO_SECRET_KEY=<python -c "import secrets; print(secrets.token_urlsafe(64))">
POSTGRES_PASSWORD=<a strong password>
DATABASE_URL=postgres://timeline:<that password>@db:5432/timeline
DJANGO_ALLOWED_HOSTS=timeline.example.com
DJANGO_CSRF_TRUSTED_ORIGINS=https://timeline.example.com
DJANGO_CORS_ALLOWED_ORIGINS=
RUN_COLLECTSTATIC=1
DOMAIN=timeline.example.com
WEB_CONCURRENCY=3
# Account approval + email (new sign-ups stay inactive until you approve them)
REQUIRE_ACCOUNT_APPROVAL=1
SITE_URL=https://timeline.example.com
ACCOUNT_NOTIFY_EMAIL=you@gmail.com
EMAIL_HOST_USER=you@gmail.com
EMAIL_HOST_PASSWORD=your-16-char-app-password
# EMAIL_HOST/PORT/USE_TLS default to smtp.gmail.com/587/on; DEFAULT_FROM_EMAIL -> EMAIL_HOST_USER
```
> **Gmail SMTP:** use an **App Password** (Google Account → Security → 2-Step Verification →
> App passwords), not your normal password. Or use a transactional provider (SES, Postmark,
> Resend) — just point the `EMAIL_HOST_*` vars at its SMTP endpoint.
Install the origin certificate, then bring the stack up:
```bash
# In the Cloudflare dashboard: SSL/TLS → Origin Server → Create Certificate
# (defaults: RSA, hostnames *.example.com + example.com, 15-year validity).
# Copy the Origin Certificate and Private Key onto the box:
mkdir -p nginx/certs
# paste the Origin Certificate into nginx/certs/origin.pem
# paste the Private Key into nginx/certs/origin.key
docker compose -f docker-compose.prod.yml up -d --build
docker compose -f docker-compose.prod.yml exec backend python manage.py createsuperuser
```
`migrate` + `collectstatic` run automatically on backend start. nginx mounts
`nginx/certs/origin.pem` + `origin.key` read-only (both are gitignored). The app is now live at
**https://timeline.example.com** once the Cloudflare proxy is on (orange) with SSL/TLS mode
**Full (strict)**.
## 4. Backups
```bash
# manual
./deploy/backup-db.sh
# daily cron at 03:00
( crontab -l 2>/dev/null; echo "0 3 * * * cd /opt/timeline && ./deploy/backup-db.sh >> backups/backup.log 2>&1" ) | crontab -
```
Dumps go to `backups/` (gzipped, rotated `BACKUP_KEEP_DAYS`). Set `S3_BACKUP_BUCKET` in `.env`
(and attach an S3 IAM role to the instance) to copy them off-box. Restore:
```bash
gunzip -c backups/<file>.sql.gz | docker compose -f docker-compose.prod.yml exec -T db psql -U timeline timeline
```
## Updating
```bash
cd /opt/timeline && git pull
docker compose -f docker-compose.prod.yml up -d --build
```
There's no cert renewal to worry about — the Cloudflare Origin Certificate is valid for ~15 years.
---
## Sizing
The whole stack runs on one box (Postgres included). Approximate steady-state memory:
| Component | RAM |
|---|---|
| Amazon Linux 2023 | ~0.3 GB |
| Docker daemon | ~0.1 GB |
| PostgreSQL 16 | ~0.2 GB |
| gunicorn (3 workers × ~120 MB) | ~0.4 GB |
| nginx | ~0.05 GB |
| **Runtime total** | **~1.0–1.3 GB** |
| Building the frontend image on-box (vite/node, transient) | **+1–1.5 GB peak** |
**Recommendation (now — a team of 3, continued development):**
| Option | Instance | vCPU / RAM | Notes | ~On-demand* |
|---|---|---|---|---|
| **Recommended** | **t4g.medium** | 2 / 4 GiB | Comfortable runtime + on-box image builds + DB growth + headroom. **Default in Terraform.** | ~$24/mo |
| Budget | t4g.small | 2 / 2 GiB | Runs fine at runtime; add a **2 GB swapfile** to build images without OOM. | ~$12/mo |
| x86 | t3.medium / t3.small | same | Set `ami_architecture = "x86_64"`. Slightly pricier than Graviton. | ~$30 / $15 |
\* us-east-1, plus ~$2.4/mo for the 30 GiB gp3 volume. The elastic IP is free while attached.
**t4g (Graviton/ARM)** is the best value — every image here is multi-arch, so it just works.
Add swap on a small box:
```bash
sudo dd if=/dev/zero of=/swapfile bs=1M count=2048 && sudo chmod 600 /swapfile
sudo mkswap /swapfile && sudo swapon /swapfile
echo '/swapfile none swap sw 0 0' | sudo tee -a /etc/fstab
```
Tune `WEB_CONCURRENCY` (gunicorn workers) to the box: a rule of thumb is `2 × vCPU + 1`, but it's
memory-bound (~120 MB/worker), so `3` is right for a 2-vCPU/4-GiB box.
## Scaling to ~3000 users
3000 *registered* users is light for this CRUD app — the concern is concurrency and resilience,
not raw compute. When you outgrow the single box, evolve it in this order (the Terraform VPC is the
base to build on):
1. **Postgres → Amazon RDS** (`db.t4g.medium`, then `db.m6g.large`): managed backups, point-in-time
restore, easy vertical scaling. Point `DATABASE_URL` at the RDS endpoint and drop the `db`
container. This alone removes the DB from the app box's memory budget.
2. **Build images in CI → push to ECR**; instances *pull* (no on-box builds → smaller app boxes).
3. **Scale the web tier** — either one larger instance (`m6g.large`/`xlarge`, 8/16 GiB, 8–16 gunicorn
workers) **or** an **Application Load Balancer + Auto Scaling Group** of 2+ `t4g.large` app
instances. Terminate TLS at the **ALB with an ACM cert** (then the box-level origin-cert/nginx-TLS
is no longer needed — nginx just serves the SPA + proxies).
4. **CloudFront** in front of the SPA/assets for CDN caching (optional, helps global latency).
5. Consider **PgBouncer** and tuned `shared_buffers` / `max_connections` on the DB.
A reasonable 3000-user starting topology: **ALB + ASG (2× t4g.large) + RDS db.t4g.medium**, images
from ECR, TLS via ACM. Rough cost ~$200–400/mo depending on instance count and RDS size.
## Operational notes
- **Security group:** SSH is restricted to `allowed_ssh_cidr` (set it!); 80/443 are public. IMDSv2
is enforced and the EBS root volume is encrypted.
- **Logs:** `docker compose -f docker-compose.prod.yml logs -f backend` (gunicorn logs to stdout).
- **Admin:** `https://timeline.example.com/admin/` (after `createsuperuser`).
- **Health check:** `GET /api/health/` returns `200` (`{"status":"ok"}`) — use it for monitoring or
an ALB target-group health check later.
- **Approving accounts:** new sign-ups are inactive; you're emailed when one registers. Approve in
`…/admin/auth/user/` (tick **Active**, or the **"Approve & notify"** action) and the person is
emailed automatically. Set `REQUIRE_ACCOUNT_APPROVAL=0` to run an open instance.
- **Secrets:** `.env` lives only on the box (gitignored). For a team, consider AWS SSM Parameter
Store / Secrets Manager.