Daemonize worker processes

Daemonize worker processes#

When running workflow applications, it’s helpful to create long-running processes that run at startup and are resilient to failure.

This guide shows you how to set up a systemd service to create long-running Prefect processes that poll for scheduled flow runs, including how to:

  • create a Linux user

  • install and configure Prefect

  • set up a systemd service for the Prefect worker or .serve process

Prerequisites#

  • An environment with a linux operating system with systemd and Python 3.9 or later.

  • A superuser account that can run sudo commands.

  • A Prefect Cloud account, or an instance of a Prefect server running on your network.

If using an AWS t2-micro EC2 instance with an AWS Linux image, you can install Python and pip with sudo yum install -y python3 python3-pip.

Steps#

A systemd service is ideal for running a long-lived process on a Linux VM or physical Linux server. You will use systemd and learn how to automatically start a Prefect worker or long-lived serve process when Linux starts. This approach provides resilience by automatically restarting the process if it crashes.

Step 1: Add a user#

Create a user account on your Linux system for the Prefect process. You can run a worker or serve process as root, but it’s best practice to create a dedicated user.

In a terminal, run:

sudo useradd -m prefect
sudo passwd prefect

When prompted, enter a password for the prefect account.

Next, log in to the prefect account by running:

sudo su prefect

Step 2: Install Prefect#

Run:

pip3 install prefect

This guide assumes you are installing Prefect globally, rather than a virtual environment. If running a systemd service in a virtual environment, change the ExecPath. For example, if using venv, change the ExecPath to target the prefect application in the bin subdirectory of your virtual environment.

Next, set up your environment so the Prefect client knows which server to connect to.

If connecting to Prefect Cloud, follow the instructions to obtain an API key, and then run the following:

prefect cloud login -k YOUR_API_KEY

When prompted, choose the Prefect workspace to log in to.

If connecting to a self-hosted Prefect server instance instead of a Prefect Cloud account, run the following command, substituting the IP address of your server:

prefect config set PREFECT_API_URL=http://your-prefect-server-IP:4200

Run the exit command to sign out of the prefect Linux account. This command switches you back to your sudo-enabled account where you can run the commands in the next section.

Step 3: Set up a systemd service#

See the section below if you are setting up a Prefect worker. Skip to the next section if you are setting up a Prefect .serve process.

Setting up a systemd service for a Prefect worker#

Move into the /etc/systemd/system folder and open a file for editing. We use the Vim text editor below.

cd /etc/systemd/system
sudo vim my-prefect-service.service

[Unit]
Description=Prefect worker

[Service]
User=prefect
WorkingDirectory=/home
ExecStart=prefect worker start --pool YOUR_WORK_POOL_NAME
Restart=always

[Install]
WantedBy=multi-user.target

Make sure you substitute your own work pool name.

Setting up a systemd service for .serve#

Copy your flow entrypoint Python file and any other files needed for your flow to run into the /home directory (or the directory of your choice).

Here’s a basic example flow:

from prefect import flow


@flow(log_prints=True)
def say_hi():
    print("Hello!")

if __name__=="__main__":
    say_hi.serve(name="served and daemonized deployment")

To make changes to your flow code without restarting your process, push your code to git-based cloud storage (GitHub, BitBucket, GitLab) and use flow.from_source().serve(), as in the example below.

if __name__ == "__main__":
    flow.from_source(
        source="https://github.com/org/repo.git",
        entrypoint="path/to/my_remote_flow_code_file.py:say_hi",
    ).serve(name="deployment-with-github-storage")

Make sure you substitute your own flow code entrypoint path.

If you change the flow entrypoint parameters, you must restart the process.

Move into the /etc/systemd/system folder and open a file for editing. This example below uses Vim.

cd /etc/systemd/system
sudo vim my-prefect-service.service

[Unit]
Description=Prefect serve

[Service]
User=prefect
WorkingDirectory=/home
ExecStart=python3 my_file.py
Restart=always

[Install]
WantedBy=multi-user.target

Step 4: Save, enable, and start the service#

To save the file and exit Vim, hit the escape key, type :wq!, then press the return key.

Next, make systemd aware of your new service by running:

sudo systemctl daemon-reload

Then, enable the service by running:

sudo systemctl enable my-prefect-service

This command ensures it runs when your system boots.

Next, start the service:

sudo systemctl start my-prefect-service

Run your deployment from the UI and check the logs on the Flow Runs page.

You can see if your daemonized Prefect worker or serve process is running, and the Prefect logs with systemctl status my-prefect-service.

You now have a systemd service that starts when your system boots, which will restart if it ever crashes.