ffmpegd Workers

ffmpegd is tvarr's distributed transcoding service.

What is ffmpegd?

ffmpegd (FFmpeg Daemon) is a standalone service that:

• Connects to tvarr's coordinator via gRPC
• Receives transcoding jobs
• Runs FFmpeg processes
• Streams output back to clients

Why Use Separate Workers?

Scale Horizontally

Run multiple workers on different machines:

tvarr (coordinator)
    │
    ├── ffmpegd (GPU server 1)
    ├── ffmpegd (GPU server 2)
    └── ffmpegd (CPU-only server)

Dedicated Hardware

Put workers on machines with:

• GPUs for hardware encoding
• More CPU cores for software encoding
• Better cooling for heavy workloads

Isolation

Keep transcoding separate from the main tvarr process:

• Transcoding crashes don't affect the coordinator
• Easier resource allocation
• Independent scaling

Running ffmpegd

Docker

docker run -d \
  --name ffmpegd \
  -e TVARR_COORDINATOR_URL=tvarr:9090 \
  -e TVARR_MAX_JOBS=4 \
  --device /dev/dri:/dev/dri \
  ghcr.io/jmylchreest/tvarr-transcoder:release

Docker Compose

services:
  ffmpegd:
    image: ghcr.io/jmylchreest/tvarr-transcoder:release
    environment:
      - TVARR_COORDINATOR_URL=tvarr:9090
      - TVARR_MAX_JOBS=4
      - TVARR_DAEMON_NAME=gpu-worker-1
    devices:
      - /dev/dri:/dev/dri

Configuration

Environment Variables

Variable	Default	Description
TVARR_COORDINATOR_URL	-	Coordinator gRPC address (required)
TVARR_DAEMON_NAME	hostname	Human-readable worker name
TVARR_MAX_JOBS	auto	Max concurrent transcoding jobs
TVARR_MAX_CPU_JOBS	auto	Max software encoding jobs
TVARR_MAX_GPU_JOBS	auto	Max hardware encoding jobs
TVARR_AUTH_TOKEN	-	Authentication token
TVARR_LOGGING_LEVEL	info	Log level

Auto-Detection

ffmpegd automatically detects:

• Available FFmpeg encoders/decoders
• GPU hardware and session limits
• CPU cores for software encoding

Monitoring Workers

View connected workers at Transcoders in the web UI:

• Worker status (online/offline)
• Current jobs
• Capabilities (encoders, GPUs)
• Resource usage

Job Distribution

The coordinator distributes jobs based on:

1. Encoder requirements - Does the worker have the needed encoder?
2. Available capacity - How many job slots are free?
3. Hardware preference - Prefer GPU over CPU when available

Standalone Mode

Run ffmpegd in standalone mode for testing:

tvarr-ffmpegd serve --standalone

This starts the daemon without connecting to a coordinator, useful for verifying FFmpeg detection.