Learn how to create deployments via the CLI, Python, or Terraform.
There are several ways to create deployments, each catering to different organizational needs.
serve
This is the simplest way to get started with deployments:
The serve
method creates a deployment from your flow and immediately begins listening for scheduled runs to execute.
Providing cron="* * * * *"
to .serve
associates a schedule with your flow so it will run every minute of every day.
For more configuration, you can create a deployment that uses a work pool. Reasons to create a work-pool based deployment include:
Work pools are popular with data platform teams because they allow you to manage infrastructure configuration across an organization.
Prefect offers two options for creating deployments with dynamic infrastructure:
deploy
To define a deployment with a Python script, use the flow.deploy
method.
Here’s an example of a deployment that uses a work pool and bakes the code into a Docker image.
To learn more about the deploy
method, see Deploy flows with Python.
If you’d rather take a declarative approach to defining a deployment through a YAML file, use a prefect.yaml
file.
Prefect provides an interactive CLI that walks you through creating a prefect.yaml
file:
The result is a prefect.yaml
file for deployment creation.
The file contains build
, push
, and pull
steps for building a Docker image, pushing code to a Docker registry, and pulling code at runtime.
Learn more about creating deployments with a YAML file in Define deployments with YAML.
Prefect also provides CI/CD options for automatically creating YAML-based deployments.
You can manage deployments with the Terraform provider for Prefect.
You can manage deployments with the Prefect API.
Choosing between deployment methods
For many cases, serve
is sufficient for scheduling and orchestration.
The work pool / worker paradigm via .deploy()
or prefect deploy
can be great for complex infrastructure requirements and isolated flow run environments.
You are not locked into one method and can combine approaches as needed.