mediasoup

mediasoup v3 is a Selective Forwarding Unit (SFU) designed to handle the media layer of real-time communication applications. It provides a low-level API that is signaling agnostic, meaning it does not mandate a specific signaling protocol.

Key characteristics include:

• Server-side: Available as a Node.js module or a Rust crate.
• Client-side: Provided as tiny TypeScript and C++ libraries.
• Capabilities: Supports both WebRTC and plain RTP input/output, multi-stream over a single ICE + DTLS transport, and is IPv6 ready.
• Media Handling: Supports Simulcast, SVC, congestion control, and data message exchange (WebRTC DataChannels, SCTP over UDP, or direct termination).

The mediasoup Rust implementation is divided into three distinct crates:

• mediasoup-sys: A low-level crate that wraps the C++ worker into Rust.
• mediasoup-types: A crate that defines and exposes the core mediasoup Rust types.
• mediasoup: The high-level crate that consumes mediasoup-sys and mediasoup-types to provide an idiomatic Rust API for users.

Dependency Rules:

• If changes are internal to the worker or mediasoup-sys, only mediasoup-sys strictly requires an update.
• Adding new APIs that the mediasoup crate must understand is considered a breaking change, as it may cause crashes or errors if unexpected data is encountered.

The webrtc-fuzzer-corpora repository provides libFuzzer corpus files designed for WebRTC developers to test their implementations.

• The corpora directory contains various subfolders with corpus files (including samples from the Chromium project).
• The reports directory contains fuzzer-generated crash and memory leak reports contributed by the community.

While add_sha1.sh only detects duplicates within the same subfolder, you can use a find command to reveal duplicate files across the entire repository structure by matching SHA1 sums.

find . \( ! -regex '.*/\..*' \) -type f -exec sha1sum {} + | sort | uniq -w40 -dD

Use the following NPM scripts to compile the TypeScript source code located in node/src into JavaScript in the node/lib directory:

• npm run typescript:build: Performs a one-time compilation.
• npm run typescript:watch: Compiles the code and watches for changes in the TypeScript files to trigger automatic recompilation.

mediasoup is intended for high-performance media applications where a low-level API is required to avoid constraints or assumptions.

Common Use Cases

• Group video chat applications.
• One-to-many or few-to-many real-time broadcasting.
• RTP streaming.

Core Features

• Low-level API: ECMAScript 6 for Node.js and idiomatic Rust.
• Transport: ICE / DTLS / RTP / RTCP over both UDP and TCP.
• Scalability: Support for Simulcast and SVC (Scalable Video Coding).
• Bandwidth Management: Sender and receiver bandwidth estimation with spatial/temporal layers distribution algorithms.
• Performance: Media worker threads/subprocesses are implemented in C++ on top of libuv.

To prevent redundant testing caused by duplicate files (files with different names but identical binary content), you can use the add_sha1.sh script. This script appends the SHA1 hashsum to the filenames in the target directory without touching file extensions.

Usage: Pass the target folder as the first argument.

Key behaviors:

• Renaming: It renames files to include their hash (e.g., file -> file-hash).
• Idempotency: It detects files that are already hashed by looking for the SHA1sum in the filename and marks them as OK.
• Duplicate Detection (Local): It identifies duplicate files within the same subfolder and labels them as DUPLICATE.

$ ./add_sha1.sh .

Corpus Files

Use the deps/webrtc-fuzzer-corpora/corpora directory to provide input to the fuzzer.

Because libFuzzer uses the first command-line directory parameter for both reading and storing newly generated corpus files, it is recommended to pass a dedicated directory like fuzzer/new-corpus as the first argument.

Crash Reports

When an issue is detected, the fuzzer generates a crash report file containing the input bytes. To store these reports in a specific directory (e.g., fuzzer/reports/), use the following CLI flag:

• -artifact_prefix=fuzzer/reports/

mediasoup uses the combination of Sender Reports (SR) and Receiver Reports (RR) to determine the quality of each link.

• Sender Reports: mediasoup locally generates these for the streams it sends and processes those received from producer endpoints.
• Receiver Reports: mediasoup locally generates these for the streams it receives and consumes those from every remote RTP receiver.

Quality information derived from these reports is exposed via the JavaScript API, allowing developers to monitor and react to the quality of each participant in the router.

Mediamosoup provides several Docker images for different environments:

• invoke docker: Builds a Linux Ubuntu image with fuzzer-capable clang++ and all dependencies.
• invoke docker-run: Runs the Ubuntu container with a bash session.
• invoke docker-alpine: Builds a Linux Alpine image with all dependencies.
• invoke docker-alpine-run: Runs the Alpine container with an ash session.
• invoke docker-386: Builds a 386 Linux Debian (32-bit) image.
• invoke docker-alpine-386: Runs the 386 Linux Debian (32-bit) container.

Important Note: Before running a container, you must clean previous installations by running invoke clean-all inside the worker directory.

cd worker
invoke clean-all
invoke docker
invoke docker-run

To build the mediasoup-worker binary, use the following scripts:

• npm run worker:build: Builds the binary.
• npm run worker:prebuild: Creates a prebuilt of the mediasoup-worker binary in the worker/prebuild folder.
• npm run worker:prebuild-name: Prints the name of the corresponding mediasoup-worker prebuild tar file.