# SyncPlayer (LAN Synchronized Video Playback)

Flask + Flask-SocketIO system for near frame-level synchronized playback across many browser-capable SoC displays on a local LAN.

## Key synchronization mechanism

### 1) Custom NTP-style time sync (client does **not** use its wall clock)

Client keeps a high-resolution monotonic clock using:

```js
client_ms = performance.timeOrigin + performance.now()
```

Every 10s the client sends `t1_ms` and the server replies with `{t1_ms, t2_ms}` where `t2_ms` is server time in ms.

Client receives the reply at `t3_ms` and calculates:

```text
rtt = t3 - t1
offset = t2 - (t1 + t3)/2
server_time ~= client_time + offset
```

Offset is smoothed to reduce jitter.

### 2) Scheduled absolute start

When an event triggers, the server schedules a future start time:

```text
start_time_ms = server_time_ms() + EVENT_LEAD_TIME_MS (default 5000)
```

Server sends each participating display:

- `video_url`
- `start_time_ms`
- `event_id`

Client preloads, pauses at 0, then uses a `setTimeout` + `requestAnimationFrame` loop to call `video.play()` *as close as possible* to `start_time_ms` using the corrected server clock.

### 3) Drift correction during playback

Every 2s client computes:

```text
expected = (serverNowMs - start_time_ms)/1000
drift = video.currentTime - expected
```

- if `|drift| > 15ms`: adjust `playbackRate` slightly (0.99 / 1.01)
- if `|drift| > 100ms`: hard seek to expected

This keeps the group synchronized on low-latency LANs.

## Project structure

```
/app
/templates
/static
/media
models.py
routes.py
sockets.py
udp_listener.py
sync.py
config.py
run.py
requirements.txt
```

## Setup

### 1) Create venv + install deps

Windows (cmd):

```bat
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
```

### 2) Initialize database

```bat
python run.py init-db
```

### 3) Run server

```bat
python run.py --host 0.0.0.0 --port 5000
```

Notes:
- On Windows / Python 3.14+, this runs in `threading` async mode by default (good enough for LAN testing). For highest scale/throughput, run on Linux and install `eventlet` or `gevent`.
- You can force an async mode with: `set ASYNC_MODE=threading` (Windows) or `export ASYNC_MODE=eventlet|gevent|threading` (Linux).

Open admin UI:

- `http://<server-ip>:5000/admin/`

Open display pages:

- `http://<server-ip>:5000/display/<public_id>`

## UDP trigger

Configure UDP Port + UDP Payload on an Event in the admin UI. The UDP listener binds ports automatically (refreshes config every 5s).

Send a UDP packet (example PowerShell):

```powershell
$udp = new-Object System.Net.Sockets.UdpClient
$bytes = [Text.Encoding]::UTF8.GetBytes("SHOW_START")
$udp.Send($bytes, $bytes.Length, "192.168.1.10", 7777) | Out-Null
$udp.Close()
```

### Firewall notes

- Allow TCP port (default 5000) for HTTP/WebSocket.
- Allow the configured UDP ports for triggers.

## HTTP URL trigger (LAN)

You can also trigger an event by opening a URL (useful for simple controllers without UDP):

- By id: `http://<server-ip>:5000/trigger/<event_id>`
- By name: `http://<server-ip>:5000/trigger_by_name/<event_name>`

Cooldown:
- Cooldown is enforced by default.
- To bypass cooldown (LAN only), add `?force=1`.

Example:

```
http://192.168.1.10:5000/trigger/1?force=1
```

## Idle image (no active event)

You can upload an **Idle Image** in the admin UI:

- `http://<server-ip>:5000/admin/idle-image`

When no triggered event is active, displays show this image fullscreen.

## Debug overlay

On display page:

- Press **F2** or **Ctrl+Shift+D**, or
- **Double click** anywhere

to toggle a small overlay showing Socket.IO connection state, offset/RTT, event id, drift and playbackRate.

## Display video cache warming

When a display page loads, it will fetch `/api/videos` and perform **background cache warming**:

- Default strategy: request the first **1MB** of each video via HTTP `Range` (`bytes=0-1048575`).
- Purpose: get the container + initial moov atoms into cache to reduce startup stutter.
- It is **best-effort**: browsers may evict cache under pressure, and some platforms ignore caching for large media.
- Cache warming automatically pauses once an event starts to avoid competing bandwidth during synchronized start.