fetch_ml/configs/SECURITY.md
Jeremie Fraeys 86f9ae5a7e
docs(config): reorganize configuration structure and add documentation
Restructure configuration files for better organization:
- Add scheduler configuration examples (scheduler.yaml.example)
- Reorganize worker configs into subdirectories:
  - distributed/ - Multi-node cluster configurations
  - standalone/ - Single-node deployment configs
- Add environment-specific configs:
  - dev-local.yaml, docker-dev.yaml, docker-prod.yaml
  - homelab-secure.yaml, worker-prod.toml
- Add deployment configs for different security modes:
  - docker-standard.yaml, docker-hipaa.yaml, docker-dev.yaml

Add documentation:
- configs/README.md with configuration guidelines
- configs/SECURITY.md with security configuration best practices
2026-02-26 12:04:11 -05:00

130 lines
3.1 KiB
Markdown

# Security Guidelines for fetch_ml Distributed Mode
## Token Management
### Quick Start (Recommended)
```bash
# 1. Generate config with tokens
scheduler -init -config scheduler.yaml
# 2. Or generate a single token
scheduler -generate-token
```
### Generating Tokens
**Option 1: Initialize full config (recommended)**
```bash
# Generate config with 3 worker tokens
scheduler -init -config /etc/fetch_ml/scheduler.yaml
# Generate with more tokens
scheduler -init -config /etc/fetch_ml/scheduler.yaml -tokens 5
```
**Option 2: Generate single token**
```bash
# Generate one token
scheduler -generate-token
# Output: wkr_abc123...
```
**Option 3: Using OpenSSL**
```bash
openssl rand -hex 32
```
### Token Storage
- **NEVER commit tokens to git** — config files with real tokens are gitignored
- Store tokens in environment variables or secure secret management
- Use `.env` files locally (already gitignored)
- Rotate tokens periodically
### Config File Security
```
configs/
├── scheduler/scheduler.yaml # ⛔ NEVER commit with real tokens
├── scheduler/scheduler.yaml.example # ✅ Safe to commit (placeholders)
└── worker/distributed/worker.yaml # ⛔ NEVER commit with real tokens
```
All `*.yaml` files in `configs/` subdirectories are gitignored by default.
### Distribution Workflow
```bash
# On scheduler host:
$ scheduler -init -config /etc/fetch_ml/scheduler.yaml
Config generated: /etc/fetch_ml/scheduler.yaml
Generated 3 worker tokens. Copy the appropriate token to each worker's config.
=== Generated Worker Tokens ===
Copy these to your worker configs:
Worker: worker-01
Token: wkr_abc123...
Worker: worker-02
Token: wkr_def456...
# On each worker host - copy the appropriate token:
$ cat > /etc/fetch_ml/worker.yaml <<EOF
scheduler:
address: "scheduler-host:7777"
cert: "/etc/fetch_ml/scheduler.crt"
token: "wkr_abc123..." # Copy from above
EOF
```
## TLS Configuration
### Self-Signed Certs (Development)
```yaml
scheduler:
auto_generate_certs: true
cert_file: "/etc/fetch_ml/scheduler.crt"
key_file: "/etc/fetch_ml/scheduler.key"
```
Auto-generated certs are for development only. The scheduler prints the cert path on first run — distribute this to workers securely.
### Production TLS
Use proper certificates from your CA:
```yaml
scheduler:
auto_generate_certs: false
cert_file: "/etc/ssl/certs/fetch_ml.crt"
key_file: "/etc/ssl/private/fetch_ml.key"
```
## Network Security
- Scheduler bind address defaults to `0.0.0.0:7777` — firewall appropriately
- WebSocket connections use WSS with cert pinning (no CA chain required)
- Token authentication on every WebSocket connection
- Metrics endpoint (`/metrics`) has no auth — bind to localhost or add proxy auth
## Audit Logging
Enable audit logging to track job lifecycle:
```yaml
scheduler:
audit_log: "/var/log/fetch_ml/audit.log"
```
## Security Checklist
- [ ] Tokens generated via `scheduler -init` or `scheduler -generate-token`
- [ ] Config files with tokens NOT in git
- [ ] TLS certs distributed securely to workers
- [ ] Scheduler bind address firewalled
- [ ] Metrics endpoint protected (if exposed)
- [ ] Audit logging enabled