Skip to main content

Example: AWS EC2

This is an example configuration that shows how to deploy LiveCompositor to an AWS EC2 instance with Terraform configuration.

All examples are located in the github.com/membraneframework-labs/live_compositor_deployment repository:

  • project directory includes an example Membrane project that can consume multiple streams over RTMP and host the composed stream as an HLS playlist.
  • aws-ec2-terraform directory includes an example Terraform+Packer configuration for building an AMI (Amazon Machine Image) and deploying it to EC2.

Prerequisites​

  • Terraform
  • Packer
  • Elixir - required to build an example project
  • FFmpeg - used to send/receive streams from/to the compositor

CPU vs GPU rendering trade-off​

  • GPU+CPU - LiveCompositor uses wgpu (implementation of WebGPU standard written in Rust) for rendering. However, all decoding and encoding still happens on the CPU. When running on GPU the rendering cost should be negligible compared to the decoding/encoding.
  • CPU-only - When running on a CPU-only instance, all WebGPU code is emulated on the CPU. Unless your encoder quality is set very high, rendering will use most of the CPU processing time.

Actual price-to-performance can vary, but in general, CPU+GPU instances make more sense for fast encoder presets and complex rendering pipelines. However, CPU-only can be more optimal when using simple layouts and prioritizing quality over performance with slower preset.

How to deploy​

warning

The example configuration is using us-east-1 region. If you want to use a different one make sure to change it both in Packer and Terraform configuration. Specifically, if you use EC2 instances with GPU, you might only have them available in some regions.

CPU-only​

Go to aws-ec2-terraform/packer directory and run

packer build membrane.pkr.hcl

to build an AMI image with an example Membrane project. At the end of the process, the terminal will print the AMI ID, that will be needed in the next step (something like ami-0e18e9d7b8c037ec2).

The other pkr.hcl file in this directory (standalone.pkr.hcl) includes configuration for deploying just a standalone LiveCompositor instance, so you can also go that route, but the rest of this guide assumes you are using the provided Membrane project.

Open aws-ec2-terraform/main.tf, find aws_instance.demo_instance definition and update the ami field with the AMI ID from the previous step.

In aws-ec2-terraform directory run:

terraform apply

CPU+GPU​

Go to aws-ec2-terraform/packer directory and run

packer build -var "with-gpu=true" membrane.pkr.hcl

to build an AMI image with an example Membrane project. At the end of the process, the terminal will print the AMI ID that, will be needed in the next step (something like ami-0e18e9d7b8c037ec2).

Open aws-ec2-terraform/main.tf, find aws_instance.demo_instance definition and update the ami field with the AMI ID from the previous step.

In aws-ec2-terraform directory run:

terraform apply -var="with-gpu=true"
note

Instances with GPU like g4dn.xlarge are not available by default on AWS. You will need to request a quota increase from the AWS team to use them.

How to use​

After everything is deployed you can open your AWS dashboard and find the public IP of the newly deployed instance.

To test the service, run in separate terminals:

  • To receive the output stream 
    ffplay http://YOUR_INSTANCE_IP:9001/index.m3u8
  • To send an example input stream 
    ffmpeg -re -f lavfi -i testsrc
      -vf scale=1280:720 -vcodec libx264 \
      -profile:v baseline -preset fast -pix_fmt yuv420p \
      -f flv rtmp://YOUR_INSTANCE_IP:9000/app/stream_key
    • You can run this command multiple times with different paths instead of app/stream_key to connect multiple streams.