docs(deployments): single-site pattern

This commit is contained in:
2026-05-09 12:37:17 +00:00
parent 5a308fabf7
commit 73cf5ed350
+99
View File
@@ -0,0 +1,99 @@
# Single-site deployment
The default deployment pattern. Most customers start here.
## Topology
`R+1`: 1 router + 1 Cell + N cameras + optional operator workstation.
```
Internet (WAN)
┌──────────┐
│ Blocao │
│ Router │
└────┬─────┘
┌────────┴────────────┐
│ │
VLAN-10 VLAN-20
Cameras (N) Cell
192.168.10/24 192.168.20.10
├── cam-01 (PoE)
├── cam-02 (PoE)
├── cam-03 (WiFi 5GHz)
└── cam-04 (...)
```
## Hardware shopping list (4-cam example)
| Item | Model | Notes |
|---|---|---|
| Router | GL.iNet GL-MT6000 | 4×2.5GbE LAN + 1×2.5GbE WAN, WiFi 6E |
| Cell | Banana Pi BPI-W3 + 1TB NVMe + 4TB HDD | RK3588 with 6 TOPS NPU |
| Cameras | 4× Reolink RLC-820A or equivalent | 4K, ONVIF, PoE |
| PoE switch | Mikrotik CSS610-8P-2S+IN | 8 PoE+ ports + 2 SFP+ |
| UPS | Any 600W+ | Cell + router for 30min |
Approximate hardware cost: €1,500-2,500 depending on cameras chosen.
## Install procedure
1. **Prep at office**:
- Flash router with Blocao firmware (image from CI artifacts).
- Verify Balena fleet has the Cell stack ready.
- Pre-configure site ID and network parameters in the wizard if known.
2. **On site**:
- Mount cameras, run cabling.
- Connect: cameras → PoE switch → router LAN port (trunked VLAN), Cell → router LAN port, WAN cable to upstream router/modem.
- Power on router first, then Cell, then cameras.
3. **First boot**:
- From a laptop on VLAN-30 (or via Tailscale if pre-provisioned), open `http://blocao-router.local/`.
- Wizard runs. Installer goes through 6 steps (~15 minutes).
- Provisioning completes; redirects to SYNOPSIS.
4. **Camera onboarding**:
- From CAMS panel, click ADD CAMERA.
- Auto-discovery finds cameras in VLAN-10.
- For each: authenticate, force password rotation, test stream, name, save.
- Verify event flow in SYNOPSIS.
5. **Verification**:
- HEALTH panel shows all green.
- MQTT panel shows events flowing.
- Walk in front of a camera, confirm event in MQTT live tail.
6. **Handover**:
- Customer-side admin gets operator credentials.
- Walk through FORENSICS panel with sample data.
- Document case management workflow if applicable.
Estimated time on site: half a day for an experienced installer with prepped hardware.
## What customer sees
Day 1 after install: cameras recording, events being detected, console accessible from mgmt VLAN or via Tailscale.
Day 30: 30 days of footage retained, queries work across that range, evidence packs exportable for any case.
Day 90: typical first review meeting. Customer feedback informs config tweaks (zone definitions, retention overrides, model thresholds).
## Common gotchas
- **Camera vendor portal phone-home blocked**: cameras may show "cloud disconnected" in their own UI. This is intentional. Show the customer the DNS sinkhole stat for reassurance.
- **WiFi cameras drop**: 2.4GHz is congested, 5GHz has range issues. Use cabled cameras when possible; HaLow as the future option for distance.
- **WAN flaky**: hub bridge will queue events and reconnect. Customer doesn't need to do anything.
- **Wrong NTP**: chrony with NTS is the default. If a customer has a strict internal NTP server, configure it in the wizard step.
- **Operator forgets password**: recovery via console port on the router (physical access) or hub-side Keycloak admin reset.
## Limits of single-site
- One Cell ≈ 8-12 cameras at 1080p H.265 with full Frigate. Beyond that, scale to `R+2` (two Cells).
- Local storage limit: see [`../01-architecture/storage-retention.md`](../01-architecture/storage-retention.md).
- No cross-site queries (need hub).
- No central operator audit (need hub).
For single-site only, the customer doesn't need a hub subscription. Standalone mode (configured in wizard step 5) skips hub registration entirely.