Cloudlet

The almost fast FaaS

Table of Contents

• Table of Contents
• Prerequisites
• Run Locally
  • Clone the project
  • Setup
  • Start the VMM
  • Run the API
  • Send the request using the CLI
• Architecture
• Config file

Prerequisites

Install the dependencies. On Debian/Ubuntu:

apt install build-essential cmake pkg-config libssl-dev flex bison libelf-dev iptables

Then, configure the Rust toolchain and install Just (only for dev environment):

rustup target add x86_64-unknown-linux-musl
cargo install just # optional

Finally, install the protobuf compiler.

Run Locally

Clone the project

git clone https://github.com/virt-do/cloudlet

Setup

Go to the project directory:

cd cloudlet

Create a TOML config file or update the existing one:

cat << EOF > src/cli/examples/config.toml
workload-name = "fibonacci"
language = "rust"
action = "prepare-and-run"

[server]
address = "localhost"
port = 50051

[build]
source-code-path = "$(readlink -f ./src/cli/examples/main.rs)"
release = true
EOF

Make sure to update the source-code-path to the path of the source code you want to run. Use an absolute path.

Here are more informations about each field

Start the VMM

[!WARNING] Make sure to replace CARGO_PATH environment variable with the path to your cargo binary

export CARGO_PATH=$(which cargo)

sudo -E capsh --keep=1 --user=$USER --inh=cap_net_admin --addamb=cap_net_admin -- -c  'RUST_BACKTRACE=1 '$CARGO_PATH' run --bin vmm -- grpc'

Run the API

cargo run --bin api

Send the request using the CLI

cargo run --bin cli -- run --config-path src/cli/examples/config.toml

[!NOTE] If it's your first time running the request, cloudlet will have to compile a kernel and an initramfs image. This will take a while, so make sure you do something else while you wait...

Architecture

Here is a simple sequence diagram of Cloudlet:

sequenceDiagram
    participant CLI
    participant API
    participant VMM
    participant Agent

    CLI->>API: HTTP Request /run
    API->>VMM: gRPC Request to create VM
    VMM->>Agent: Creation of the VM
    VMM->>Agent: gRPC Request to the agent
    Agent->>Agent: Build and run code
    Agent-->>VMM: Stream Response
    VMM-->>API: Stream Response
    API-->>CLI: HTTP Response

1. The CLI sends an HTTP request to the API which in turn sends a gRPC request to the VMM
2. The VMM then creates a VM
3. When a VM starts it boots on the agent which holds another gRPC server to handle requests
4. The agent then builds and runs the code
5. The response is streamed back to the VMM and then to the API and finally to the CLI.

Config file