Run Flower with the Deployment Engine

This how-to guide demonstrates how to set up and run Flower with the Deployment Engine using minimal configurations to illustrate the workflow. This is a complementary guide to the Run Flower using Docker guides.

In this how-to guide, you will:

  • Create a Flower App using the PyTorch template.

  • Start a Flower federation consisting of one SuperLink (« the server ») and two SuperNodes (« the clients »).

  • Run the Flower App on this federation.

The how-to guide should take less than 10 minutes to complete.

Prérequis

Before you start, make sure that:

  • The latest flwr CLI version is installed on your machine. Follow the installation instructions here.

  • This guide assumes all commands to be executed in the same machine in different terminals, each making use of the same Python environment.

  • This guide also assumes that you are familiar with the basic components in a Flower deployment (i.e. SuperLink and SuperNode), what their roles are and how they interact with each other. Please refer to the Architecture florale guide and the Flower Network Communication for an overview of what each component does and how they interact with each other.

Note

In a real deployment you would typically run the SuperLink the SuperNodes in different machines/servers from the one you develop your Flower app (i.e. from where you do flwr new and flwr run). The guide presented below is still valid for such scenarios but you will need to have setup a Python environment with the right set of dependencies for SuperLink and SuperNodes. An often easier way to achieve such deployments is by means of Docker. Check the Run Flower using Docker to gain a better understanding on how to do so.

Step 1: Create a Flower App

Although you could write a Flower app from scratch, it is often easier to start from one of the templates available via flwr new and then customize it to your use case. Create a new Flower app (PyTorch), and follow the instructions show upon executing flwr new:

$ flwr new my-project --framework PyTorch --username flower

🔨 Creating Flower App my-project...
🎊 Flower App creation successful.

To run your Flower App, use the following command:

        flwr run my-project

If you haven not installed all dependencies yet, follow these steps:

        cd my-project
        pip install -e .
        flwr run .

Note

You might want to update the torch and torchvision packages that come with the proejct to the latets released versions. Do so with: pip install -U torch torchvision.

If you decide to run the project with flwr run ., the Simulation Engine will be used. Continue to Step 2 to know how to instead use the Deployment Engine.

Astuce

Feel free to inspect the code using your favorite code editor before proceeding. Just open the my-project that was automatically created via flwr new. If you would like to get an overview of the code that was generated, take a look at the Démarrage rapide de PyTorch tutorial.

Step 2: Launch Flower Federation

In this section you will learn how to launch a SuperLink and connect two SuperNodes to it.

Start two Flower SuperNodes

In this step, you will launch two SuperNodes and connect them to the SuperLink. You will need two terminals for this step.

Note

Note that the values passed via the --node-config argument are specific to the behaviour of the ClientApp. If you inspect the code generated in the first step via flwr new, you’d see that the ClientApp is expecting a certain set of key-value pairs to be present in order to partition and load some data. Typically, your ClientApp wouldn’t partition a dataset, instead it would access the data directly available. In such cases you would write your ClientApp and make it receive, for example, the path to a directory of images.

  1. Terminal 1 Start the first SuperNode after activating your environment:

    $ flower-supernode \
     --insecure \
     --superlink 127.0.0.1:9092 \
     --clientappio-api-address 127.0.0.1:9094 \
     --node-config "partition-id=0 num-partitions=2"
    Understand the command

    • flower-supernode: Name of the SuperNode binary.

    • --insecure: This flag tells the SuperNode to operate in an insecure mode, allowing
      unencrypted communication. Refer to the Enable TLS connections guide to learn how to run your SuperNode with TLS.
    • --superlink 127.0.0.1:9092: Connect to the SuperLink’s Fleet API at the address
      127.0.0.1:9092. If you had launched the SuperLink in a different machine, you’d replace 127.0.0.1 with the public IP of that machine.
    • --clientappio-api-address 127.0.0.1:9094: Set the address and port number where the
      SuperNode is listening to communicate with the ClientApp.
    • --node-config "partition-id=0 num-partitions=2": The ClientApp code generated via the flwr new template expects those two key-value pairs to be defined at run time. Set the partition ID to 0 and the number of partitions to 2 for the SuperNode configuration.

  2. Terminal 2 Start the second SuperNode after activating your environment:

    $ flower-supernode \
     --insecure \
     --superlink 127.0.0.1:9092 \
     --clientappio-api-address 127.0.0.1:9095 \
     --node-config "partition-id=1 num-partitions=2"
    Understand the command

    • --clientappio-api-address 127.0.0.1:9095: Note that a different port is being used. This is only needed because you are running two SuperNodes on the same machine. Typically you would run one node per machine and therefore, the --clientappio-api-address could be omitted all together and left with its default value.

    • --node-config "partition-id=1 num-partitions=2"`: Note here we indicate a different partition-id. In this way, a ClientApp will use a different data partition depending on which SuperNode runs in.

Step 3: Run a Flower App on the Federation

At this point, you have launched two SuperNodes that are connected to the same SuperLink. The system is idling waiting for a Run to be submitted. Before you can run your Flower App through the federation we need a way to tell flwr run that the App is to be executed via the SuperLink we just started, instead of using the local Simulation Engine (the default). Doing this is easy: define a new federation section in the pyproject.toml, indicate the address of the SuperLink and pass a certificate (if any) or set the insecure flag (only when testing locally, real deployments require TLS).

  1. Open the pyproject.toml file and at the end add a new federation configuration:

    pyproject.toml
    [tool.flwr.federations.local-deployment]
address = "127.0.0.1:9093"
insecure = true

    Note

    You can customize the string that follows tool.flwr.federations. to fit your needs. However, please note that the string cannot contain a dot (.).

    In this example, local-deployment has been used. Just remember to replace local-deployment with your chosen name in both the tool.flwr.federations. string and the corresponding flwr run . command.

  2. In another terminal and with your Python environment activated, run the Flower App and follow the ServerApp logs to track the execution of the run:

    $ flwr run . local-deployment --stream

    If you want to rerun the project or test an updated version by making changes to the code, simply re-run the command above.

Step 4: Clean Up

Use the Ctrl+C command in each terminal to stop the respective processes.