![The LiveKit icon, the name of the repository and some sample code in the background.](https://raw.githubusercontent.com/livekit/livekit-cli/main/.github/banner_light.png)
This package includes command line utilities that interacts with LiveKit. It allows you to:
- Create access tokens
- Access LiveKit APIs, create, delete rooms, etc.
- Join a room as a participant, inspecting in-room events
- Start and manage Egress
- Perform load testing, efficiently simulating real-world load
brew install livekit-cli
curl -sSL https://get.livekit.io/cli | bash
Or download a precompiled binary for the latest release
winget install LiveKit.LiveKitCLI
Or download a precompiled binary for the latest release
This repo uses Git LFS for embedded video resources. Please ensure git-lfs is installed on your machine.
git clone https://github.com/livekit/livekit-cli
make install
See lk --help
for a complete list of subcommands.
When a default project is set up, you can omit url
, api-key
, and api-secret
when using the CLI.
You can also set up multiple projects, and temporarily switch the active project used with the --project
flag, or persistently using lk project set-default
.
lk project add --api-key <key> --api-secret <secret> <project_name>
lk project list
lk project set-default <project_name>
To publish a demo video as a participant's track, use the following.
lk room join --identity publisher --publish-demo <room_name>
This will publish the demo video track with simulcast, at 720p, 360p, and 180p.
You can publish your own audio/video files. These tracks files need to be encoded in supported codecs. Refer to encoding instructions
lk room join --identity publisher \
--publish <path/to/video.ivf> \
--publish <path/to/audio.ogg> \
--fps 23.98 \
<room_name>
This will publish the pre-encoded .ivf
and .ogg
files to the room, indicating video FPS of 23.98. Note that the FPS only affects the video; it's important to match video framerate with the source to prevent out of sync issues.
Note: For files uploaded via CLI, expect an initial delay before the video becomes visible to the remote viewer. This delay is attributed to the pre-encoded video's fixed keyframe intervals. Video encoded with LiveKit client SDKs do not have this delay.
It's possible to publish any source that FFmpeg supports (including live sources such as RTSP) by using it as a transcoder.
This is done by running FFmpeg in a separate process, encoding to a Unix socket. (not available on Windows).
lk
can then read transcoded data from the socket and publishing them to the room.
First run FFmpeg like this:
ffmpeg -i <video-file | rtsp://url> \
-c:v libx264 -bsf:v h264_mp4toannexb -b:v 2M -profile:v baseline -pix_fmt yuv420p \
-x264-params keyint=120 -max_delay 0 -bf 0 \
-listen 1 -f h264 unix:/tmp/myvideo.sock \
-c:a libopus -page_duration 20000 -vn \
-listen 1 -f opus unix:/tmp/myaudio.sock
This transcodes the input into H.264 baseline profile and Opus.
Then, run lk
like this:
lk room join --identity bot \
--publish h264:///tmp/myvideo.sock \
--publish opus:///tmp/myaudio.sock \
<room_name>
You should now see both video and audio tracks published to the room.
It's possible to publish from video streams coming over a TCP socket. lk
can act as a TCP client. For example, with a gstreamer pipeline ending in ! tcpserversink port=16400
and streaming H.264.
Run lk
like this:
lk room join --identity bot \
--publish h264:///127.0.0.1:16400 \
<room_name>
Using unix sockets, it's also possible to publish streams from your application. The tracks need to be encoded into a format that WebRTC clients could playback (VP8, H.264, and Opus).
Once you are writing to the socket, you could use ffplay
to test the stream.
ffplay -i unix:/tmp/myvideo.sock
Recording requires egress service to be set up first.
Example request.json files are located here.
# Start room composite (recording of room UI)
lk egress start --type room-composite <path/to/request.json>
# Start track composite (audio + video)
lk egress start --type track-composite <path/to/request.json>
# Start track egress (single audio or video track)
lk egress start --type track <path/to/request.json>
In order to speed up the development cycle of your recording templates, we provide a sub-command test-egress-template
that
helps you to validate your templates.
The command will spin up a few virtual publishers, and then simulate them joining your room It'll then open a browser to the template URL, with the correct parameters filled in.
Here's an example:
lk egress test-template \
--base-url http://localhost:3000 \
--room test-room --layout speaker --video-publishers 3
This command will launch a browser pointed at http://localhost:3000
, while simulating 3 publishers publishing to your livekit instance.
Load testing utility for LiveKit. This tool is quite versatile and is able to simulate various types of load.
Note: livekit-load-tester
has been renamed to sub-command lk load-test
This guide requires a LiveKit server instance to be set up. You can start a load tester with:
lk load-test \
--room test-room --video-publishers 8
This simulates 8 video publishers to the room, with no subscribers. Video tracks are published with simulcast, at 720p, 360p, and 180p.
To test audio capabilities in your app, you can also simulate simultaneous speakers to the room.
lk load-test \
--room test-room --audio-publishers 5
The above simulates 5 concurrent speakers, each playing back a pre-recorded audio sample at the same time. In a meeting, typically there's only one active speaker at a time, but this can be useful to test audio capabilities.
Generate a token so you can log into the room:
lk token create --join \
--room test-room --identity test-user
Head over to the example web client and paste in the token, you can see the simulated tracks published by the load tester.
Due to bandwidth limitations of your ISP, most of us wouldn't have sufficient bandwidth to be able to simulate 100s of users download/uploading from the internet.
We recommend running the load tester from a VM on a cloud instance, where there isn't a bandwidth constraint.
To make this simple, make
will generate a linux amd64 binary in bin/
. You can scp the binary to a server instance and run the test there.
Prior to running the load tests, it's important to ensure file descriptor limits have been set correctly. See Performance tuning docs.
On the machine that you are running the load tester, they would also need to be applied:
ulimit -n 65535
sysctl -w fs.file-max=2097152
sysctl -w net.core.somaxconn=65535
sysctl -w net.core.rmem_max=25165824
sysctl -w net.core.wmem_max=25165824
You can run the load tester on multiple machines, each simulating any number of publishers or subscribers.
LiveKit SFU's performance is measured by the amount of data sent to its subscribers.
Use this command to simulate a load test of 5 publishers, and 500 subscribers:
lk load-test \
--duration 1m \
--video-publishers 5 \
--subscribers 500
It'll print a report like the following. (this run was performed on a 16 core, 32GB memory VM)
Summary | Tester | Tracks | Bitrate | Latency | Total Dropped
| Sub 0 | 10/10 | 2.2mbps | 78.86829ms | 0 (0%)
| Sub 1 | 10/10 | 2.2mbps | 78.796542ms | 0 (0%)
| Sub 10 | 10/10 | 2.2mbps | 79.361718ms | 0 (0%)
| Sub 100 | 10/10 | 2.2mbps | 79.449831ms | 0 (0%)
| Sub 101 | 10/10 | 2.2mbps | 80.001104ms | 0 (0%)
| Sub 102 | 10/10 | 2.2mbps | 79.833373ms | 0 (0%)
...
| Sub 97 | 10/10 | 2.2mbps | 79.374331ms | 0 (0%)
| Sub 98 | 10/10 | 2.2mbps | 79.418816ms | 0 (0%)
| Sub 99 | 10/10 | 2.2mbps | 79.768568ms | 0 (0%)
| Total | 5000/5000 | 678.7mbps (1.4mbps avg) | 79.923769ms | 0 (0%)
You can customize various parameters of the test such as
--video-publishers
: number of video publishers--audio-publishers
: number of audio publishers--subscribers
: number of subscribers--video-resolution
: publishing video resolution. low, medium, high--no-simulcast
: disables simulcast--num-per-second
: number of testers to start each second--layout
: layout to simulate (speaker, 3x3, 4x4, or 5x5)--simulate-speakers
: randomly rotate publishers to speak
LiveKit Ecosystem | |
---|---|
Realtime SDKs | React Components · Browser · Swift Components · iOS/macOS/visionOS · Android · Flutter · React Native · Rust · Node.js · Python · Unity (web) · Unity (beta) |
Server APIs | Node.js · Golang · Ruby · Java/Kotlin · Python · Rust · PHP (community) |
Agents Frameworks | Python · Playground |
Services | LiveKit server · Egress · Ingress · SIP |
Resources | Docs · Example apps · Cloud · Self-hosting · CLI |