Deploying Flows to Work Pools and Workers¶

In this guide, we will configure a deployment that uses a work pool for dynamically provisioned infrastructure.

All Prefect flow runs are tracked by the API. The API does not require prior registration of flows. With Prefect, you can call a flow locally or on a remote environment and it will be tracked.

A deployment turns your workflow into an application that can be interacted with and managed via the Prefect API. A deployment enables you to:

Schedule flow runs.
Specify event triggers for flow runs.
Assign one or more tags to organize your deployments and flow runs. You can use those tags as filters in the Prefect UI.
Assign custom parameter values for flow runs based on the deployment.
Create ad-hoc flow runs from the API or Prefect UI.
Upload flow files to a defined storage location for retrieval at run time.

Deployments created with .serve

A deployment created with the Python flow.serve method or the serve function runs flows in a subprocess on the same machine where the deployment is created. It does not use a work pool or worker.

Work pool-based deployments¶

A work pool-based deployment is useful when you want to dynamically scale the infrastructure where your flow code runs. Work pool-based deployments contain information about the infrastructure type and configuration for your workflow execution.

Work pool-based deployment infrastructure options include the following:

Process - runs flow in a subprocess. In most cases, you're better off using .serve.
Docker - runs flows in an ephemeral Docker container.
Kubernetes - runs flows as a Kubernetes Job.
Serverless Cloud Provider options - runs flows in a Docker container in a serverless cloud provider environment, such as AWS ECS, Azure Container Instance, Google Cloud Run, or Vertex AI.

The following diagram provides a high-level overview of the conceptual elements involved in defining a work-pool based deployment that is polled by a worker and executes a flow run based on that deployment.

%%{
  init: {
    'theme': 'base',
    'themeVariables': {
      'fontSize': '19px'
    }
  }
}%%

flowchart LR
    F("<div style='margin: 5px 10px 5px 5px;'>Flow Code</div>"):::yellow -.-> A("<div style='margin: 5px 10px 5px 5px;'>Deployment Definition</div>"):::gold
    subgraph Server ["<div style='width: 150px; text-align: center; margin-top: 5px;'>Prefect API</div>"]
        D("<div style='margin: 5px 10px 5px 5px;'>Deployment</div>"):::green
    end
    subgraph Remote Storage ["<div style='width: 160px; text-align: center; margin-top: 5px;'>Remote Storage</div>"]
        B("<div style='margin: 5px 6px 5px 5px;'>Flow</div>"):::yellow
    end
    subgraph Infrastructure ["<div style='width: 150px; text-align: center; margin-top: 5px;'>Infrastructure</div>"]
        G("<div style='margin: 5px 10px 5px 5px;'>Flow Run</div>"):::blue
    end

    A --> D
    D --> E("<div style='margin: 5px 10px 5px 5px;'>Worker</div>"):::red
    B -.-> E
    A -.-> B
    E -.-> G

    classDef gold fill:goldenrod,stroke:goldenrod,stroke-width:4px,color:black
    classDef yellow fill:gold,stroke:gold,stroke-width:4px,color:black
    classDef gray fill:lightgray,stroke:lightgray,stroke-width:4px
    classDef blue fill:blue,stroke:blue,stroke-width:4px,color:white
    classDef green fill:green,stroke:green,stroke-width:4px,color:white
    classDef red fill:red,stroke:red,stroke-width:4px,color:white
    classDef dkgray fill:darkgray,stroke:darkgray,stroke-width:4px,color:white

The work pool types above require a worker to be running on your infrastructure to poll a work pool for scheduled flow runs.

Additional work pool options available with Prefect Cloud

Prefect Cloud offers other flavors of work pools that don't require a worker:

Push Work Pools - serverless cloud options that don't require a worker because Prefect Cloud submits them to your serverless cloud infrastructure on your behalf. Prefect can auto-provision your cloud infrastructure for you and set it up to use your work pool.
Managed Execution Prefect Cloud submits and runs your deployment on serverless infrastructure. No cloud provider account required.

In this guide, we focus on deployments that require a worker.

Work pool-based deployments that use a worker also allow you to assign a work queue name to prioritize work and allow you to limit concurrent runs at the work pool level.

When creating a deployment that uses a work pool and worker, we must answer two basic questions:

What instructions does a worker need to set up an execution environment for our flow? For example, a flow may have Python package requirements, unique Kubernetes settings, or Docker networking configuration.
How should the flow code be accessed?

The tutorial shows how you can create a deployment with a long-running process using .serve and how to move to a work-pool-based deployment setup with .deploy. See the discussion of when you might want to move to work-pool-based deployments there.

Next, we'll explore how to use .deploy to create deployments with Python code. If you'd prefer to learn about using a YAML-based alternative for managing deployment configuration, skip to the later section on prefect.yaml.

Creating work pool-based deployments with `.deploy`¶

Automatically bake your code into a Docker image¶

You can create a deployment from Python code by calling the .deploy method on a flow.

buy.py

from prefect import flow


@flow(log_prints=True)
def buy():
    print("Buying securities")


if __name__ == "__main__":
    buy.deploy(
        name="my-code-baked-into-an-image-deployment", 
        work_pool_name="my-docker-pool", 
        image="my_registry/my_image:my_image_tag"
    )

Make sure you have the work pool created in the Prefect Cloud workspace you are authenticated to or on your running self-hosted server instance.
Then run the script to create a deployment (in future examples this step will be omitted for brevity):

python buy.py

You should see messages in your terminal that Docker is building your image. When the deployment build succeeds you will see helpful information in your terminal showing you how to start a worker for your deployment and how to run your deployment. Your deployment will be visible on the Deployments page in the UI.

By default, .deploy will build a Docker image with your flow code baked into it and push the image to the Docker Hub registry specified in the image argument`.

Authentication to Docker Hub

You need your environment to be authenticated to your Docker registry to push an image to it.

You can specify a registry other than Docker Hub by providing the full registry path in the image argument.

Warning

If building a Docker image, the environment in which you are creating the deployment needs to have Docker installed and running.

To avoid pushing to a registry, set push=False in the .deploy method.

if __name__ == "__main__":
    buy.deploy(
        name="my-code-baked-into-an-image-deployment", 
        work_pool_name="my-docker-pool", 
        image="my_registry/my_image:my_image_tag",
        push=False
    )

To avoid building an image, set build=False in the .deploy method.

if __name__ == "__main__":
    buy.deploy(
        name="my-code-baked-into-an-image-deployment", 
        work_pool_name="my-docker-pool", 
        image="discdiver/no-build-image:1.0",
        build=False
    )

The specified image will need to be available in your deployment's execution environment for your flow code to be accessible.

Prefect generates a Dockerfile for you that will build an image based off of one of Prefect's published images. The generated Dockerfile will copy the current directory into the Docker image and install any dependencies listed in a requirements.txt file.

Automatically build a custom Docker image with a local Dockerfile¶

If you want to use a custom Dockerfile, you can specify the path to the Dockerfile with the DeploymentImage class:

custom_dockerfile.py

from prefect import flow
from prefect.deployments import DeploymentImage


@flow(log_prints=True)
def buy():
    print("Selling securities")


if __name__ == "__main__":
    buy.deploy(
        name="my-custom-dockerfile-deployment", 
        work_pool_name="my-docker-pool", 
        image=DeploymentImage(
            name="my_image",
            tag="deploy-guide",
            dockerfile="Dockerfile"
    ),
    push=False
)

The DeploymentImage object allows for a great deal of image customization.

For example, you can install a private Python package from GCP's artifact registry like this:

Create a custom base Dockerfile.

FROM python:3.10

ARG AUTHED_ARTIFACT_REG_URL
COPY ./requirements.txt /requirements.txt

RUN pip install --extra-index-url ${AUTHED_ARTIFACT_REG_URL} -r /requirements.txt

Create our deployment by leveraging the DeploymentImage class.

private-package.py

from prefect import flow
from prefect.deployments.runner import DeploymentImage
from prefect.blocks.system import Secret
from my_private_package import do_something_cool


@flow(log_prints=True)
def my_flow():
    do_something_cool()


if __name__ == "__main__":
    artifact_reg_url: Secret = Secret.load("artifact-reg-url")

    my_flow.deploy(
        name="my-deployment",
        work_pool_name="k8s-demo",
        image=DeploymentImage(
            name="my-image",
            tag="test",
            dockerfile="Dockerfile",
            buildargs={"AUTHED_ARTIFACT_REG_URL": artifact_reg_url.get()},
        ),
    )

Note that we used a Prefect Secret block to load the URL configuration for the artifact registry above.

See all the optional keyword arguments for the DeploymentImage class here.

Default Docker namespace

You can set the PREFECT_DEFAULT_DOCKER_BUILD_NAMESPACE setting to append a default Docker namespace to all images you build with .deploy. This is great if you use a private registry to store your images.

To set a default Docker namespace for your current profile run:

prefect config set PREFECT_DEFAULT_DOCKER_BUILD_NAMESPACE=<docker-registry-url>/<organization-or-username>

Once set, you can omit the namespace from your image name when creating a deployment:

with_default_docker_namespace.py

if __name__ == "__main__":
    buy.deploy(
        name="my-code-baked-into-an-image-deployment", 
        work_pool_name="my-docker-pool", 
        image="my_image:my_image_tag"
    )

The above code will build an image with the format <docker-registry-url>/<organization-or-username>/my_image:my_image_tag when PREFECT_DEFAULT_DOCKER_BUILD_NAMESPACE is set.

While baking code into Docker images is a popular deployment option, many teams decide to store their workflow code in git-based storage, such as GitHub, Bitbucket, or Gitlab. Let's see how to do that next.

Store your code in git-based cloud storage¶

If you don't specify an image argument for .deploy, then you need to specify where to pull the flow code from at runtime with the from_source method.

Here's how we can pull our flow code from a GitHub repository.

git_storage.py

from prefect import flow

if __name__ == "__main__":
    flow.from_source(
        "https://github.com/my_github_account/my_repo/my_file.git",
        entrypoint="flows/no-image.py:hello_world",
    ).deploy(
        name="no-image-deployment",
        work_pool_name="my_pool",
        build=False
    )

The entrypoint is the path to the file the flow is located in and the function name, separated by a colon.

Alternatively, you could specify a git-based cloud storage URL for a Bitbucket or Gitlab repository.

Note

If you don't specify an image as part of your deployment creation, the image specified in the work pool will be used to run your flow.

After creating a deployment you might change your flow code. Generally, you can just push your code to GitHub, without rebuilding your deployment. The exception is if something that the server needs to know about changes, such as the flow entrypoint parameters. Rerunning the Python script with .deploy will update your deployment on the server with the new flow code.

If you need to provide additional configuration, such as specifying a private repository, you can provide a GitRepository object instead of a URL:

private_git_storage.py

from prefect import flow
from prefect.runner.storage import GitRepository
from prefect.blocks.system import Secret

if __name__ == "__main__":
    flow.from_source(
        source=GitRepository(
        url="https://github.com/org/private-repo.git",
        branch="dev",
        credentials={
            "access_token": Secret.load("github-access-token")
        }
    ),
    entrypoint="flows/no-image.py:hello_world",
    ).deploy(
        name="private-git-storage-deployment",
        work_pool_name="my_pool",
        build=False
    )

Note the use of the Secret block to load the GitHub access token. Alternatively, you could provide a username and password to the username and password fields of the credentials argument.

Store your code in cloud provider storage¶

Another option for flow code storage is any fsspec-supported storage location, such as AWS S3, GCP GCS, or Azure Blob Storage.

For example, you can pass the S3 bucket path to source.

s3_storage.py

from prefect import flow

if __name__ == "__main__":
    flow.from_source(
        source="s3://my-bucket/my-folder",
        entrypoint="flows.py:my_flow",
    ).deploy(
        name="deployment-from-aws-flow",
        work_pool_name="my_pool",
    )

In the example above your credentials will be auto-discovered from your deployment creation environment and credentials will need to be available in your runtime environment.

If you need additional configuration for your cloud-based storage - for example, with a private S3 Bucket - we recommend using a storage block. A storage block also ensures your credentials will be available in both your deployment creation environment and your execution environment.

Here's an example that uses an S3Bucket block from the prefect-aws library.

s3_storage_auth.py

from prefect import flow
from prefect_aws.s3 import S3Bucket

if __name__ == "__main__":
    flow.from_source(
        source=S3Bucket.load("my-code-storage"), entrypoint="my_file.py:my_flow"
    ).deploy(name="test-s3", work_pool_name="my_pool")

If you are familiar with the deployment creation mechanics with .serve, you will notice that .deploy is very similar. .deploy just requires a work pool name and has a number of parameters dealing with flow-code storage for Docker images.

Unlike .serve, if you don't specify an image to use for your flow, you must to specify where to pull the flow code from at runtime with the from_source method, whereas from_source is optional with .serve.

Additional configuration with `.deploy`¶

Our examples thus far have explored options for where to store flow code. Let's turn our attention to other deployment configuration options.

To pass parameters to your flow, you can use the parameters argument in the .deploy method. Just pass in a dictionary of key-value pairs.

pass_params.py

from prefect import flow

@flow
def hello_world(name: str):
    print(f"Hello, {name}!")

if __name__ == "__main__":
    hello_world.deploy(
        name="pass-params-deployment",
        work_pool_name="my_pool",
        parameters=dict(name="Prefect"),
        image="my_registry/my_image:my_image_tag",
    )

The job_variables parameter allows you to fine-tune the infrastructure settings for a deployment. The values passed in override default values in the specified work pool's base job template.

You can override environment variables, such as image_pull_policy and image, for a specific deployment with the job_variables argument.

job_var_image_pull.py

if __name__ == "__main__":
    get_repo_info.deploy(
        name="my-deployment-never-pull", 
        work_pool_name="my-docker-pool", 
        job_variables={"image_pull_policy": "Never"},
        image="my-image:my-tag"",
        push=False
    )

Similarly, you can override the environment variables specified in a work pool through the job_variables parameter:

job_var_env_vars.py

if __name__ == "__main__":
    get_repo_info.deploy(
        name="my-deployment-never-pull", 
        work_pool_name="my-docker-pool", 
        job_variables={"env": {"EXTRA_PIP_PACKAGES": "boto3"} },
        image="my-image:my-tag"",
        push=False
    )

The dictionary key "EXTRA_PIP_PACKAGES" denotes a special environment variable that Prefect will use to install additional Python packages at runtime. This approach is an alternative to building an image with a custom requirements.txt copied into it.

For more information on overriding job variables see this guide.

Working with multiple deployments with `deploy`¶

You can create multiple deployments from one or more Python files that use .deploy. These deployments can be managed independently of one another, allowing you to deploy the same flow with different configurations in the same codebase.

To create multiple work pool-based deployments at once you can use the deploy function, which is analogous to the serve function.

from prefect import deploy, flow

@flow(log_prints=True)
def buy():
    print("Buying securities")


if __name__ == "__main__":
    deploy(
        buy.to_deployment(name="dev-deploy", work_pool_name="my-dev-work-pool"),
        buy.to_deployment(name="prod-deploy", work_pool_name="my-prod-work-pool"),
        image="my-registry/my-image:dev",
        push=False,
    )

Note that in the example above we created two deployments from the same flow, but with different work pools. Alternatively, we could have created two deployments from different flows.

from prefect import deploy, flow

@flow(log_prints=True)
def buy():
    print("Buying securities.")

@flow(log_prints=True)
def sell():
    print("Selling securities.")


if __name__ == "__main__":
    deploy(
        buy.to_deployment(name="buy-deploy"),
        sell.to_deployment(name="sell-deploy"),
        work_pool_name="my-dev-work-pool"
        image="my-registry/my-image:dev",
        push=False,
    )

In the example above the code for both flows gets baked into the same image.

We can specify that one or more flows should be pulled from a remote location at runtime by using the from_source method. Here's an example of deploying two flows, one defined locally and one defined in a remote repository:

from prefect import deploy, flow


@flow(log_prints=True)
def local_flow():
    print("I'm a flow!")

if __name__ == "__main__":
    deploy(
        local_flow.to_deployment(name="example-deploy-local-flow"),
        flow.from_source(
            source="https://github.com/org/repo.git",
            entrypoint="flows.py:my_flow",
        ).to_deployment(
            name="example-deploy-remote-flow",
        ),
        work_pool_name="my-work-pool",
        image="my-registry/my-image:dev",
    )

You could pass any number of flows to the deploy function. This behavior is useful if using a monorepo approach to your workflows.

Creating work pool-based deployments with prefect.yaml¶

The prefect.yaml file is a YAML file describing base settings for your deployments, procedural steps for preparing deployments, and instructions for preparing the execution environment for a deployment run.

You can initialize your deployment configuration, which creates the prefect.yaml file, by running the CLI command prefect init in any directory or repository that stores your flow code.

Deployment configuration recipes

Prefect ships with many off-the-shelf "recipes" that allow you to get started with more structure within your prefect.yaml file; run prefect init to be prompted with available recipes in your installation. You can provide a recipe name in your initialization command with the --recipe flag, otherwise Prefect will attempt to guess an appropriate recipe based on the structure of your working directory (for example if you initialize within a git repository, Prefect will use the git recipe).

The prefect.yaml file contains deployment configuration for deployments created from this file, default instructions for how to build and push any necessary code artifacts (such as Docker images), and default instructions for pulling a deployment in remote execution environments (e.g., cloning a GitHub repository).

Any deployment configuration can be overridden via options available on the prefect deploy CLI command when creating a deployment.

prefect.yaml file flexibility

In older versions of Prefect, this file had to be in the root of your repository or project directory and named prefect.yaml. Now this file can be located in a directory outside the project or a subdirectory inside the project. It can be named differently, provided the filename ends in .yaml. You can even have multiple prefect.yaml files with the same name in different directories. By default, prefect deploy will use a prefect.yaml file in the project's root directory. To use a custom deployment configuration file, supply the new --prefect-file CLI argument when running the deploy command from the root of your project directory:

prefect deploy --prefect-file path/to/my_file.yaml

The base structure for prefect.yaml is as follows:

# generic metadata
prefect-version: null
name: null

# preparation steps
build: null
push: null

# runtime steps
pull: null

# deployment configurations
deployments:
- # base metadata
    name: null
    version: null
    tags: []
    description: null
    schedule: null

    # flow-specific fields
    entrypoint: null
    parameters: {}

    # infra-specific fields
    work_pool:
    name: null
    work_queue_name: null
    job_variables: {}

The metadata fields are always pre-populated for you. These fields are for bookkeeping purposes only. The other sections are pre-populated based on recipe; if no recipe is provided, Prefect will attempt to guess an appropriate one based on local configuration.

You can create deployments via the CLI command prefect deploy without ever needing to alter the deployments section of your prefect.yaml file — the prefect deploy command will help in deployment creation via interactive prompts. The prefect.yaml file facilitates version-controlling your deployment configuration and managing multiple deployments.

Deployment actions¶

Deployment actions defined in your prefect.yaml file control the lifecycle of the creation and execution of your deployments. The three actions available are build, push, and pull. pull is the only required deployment action — it is used to define how Prefect will pull your deployment in remote execution environments.

Each action is defined as a list of steps that are executing in sequence.

Each step has the following format:

section:
- prefect_package.path.to.importable.step:
    id: "step-id" # optional
    requires: "pip-installable-package-spec" # optional
    kwarg1: value
    kwarg2: more-values

Every step can optionally provide a requires field that Prefect will use to auto-install in the event that the step cannot be found in the current environment. Each step can also specify an id for the step which is used when referencing step outputs in later steps. The additional fields map directly onto Python keyword arguments to the step function. Within a given section, steps always run in the order that they are provided within the prefect.yaml file.

Deployment Instruction Overrides

build, push, and pull sections can all be overridden on a per-deployment basis by defining build, push, and pull fields within a deployment definition in the prefect.yaml file.

The prefect deploy command will use any build, push, or pull instructions provided in a deployment's definition in the prefect.yaml file.

This capability is useful with multiple deployments that require different deployment instructions.

The build action¶

The build section of prefect.yaml is where any necessary side effects for running your deployments are built - the most common type of side effect produced here is a Docker image. If you initialize with the docker recipe, you will be prompted to provide required information, such as image name and tag:

prefect init --recipe docker
>> image_name: < insert image name here >
>> tag: < insert image tag here >

Use --field to avoid the interactive experience

We recommend that you only initialize a recipe when you are first creating your deployment structure, and afterwards store your configuration files within version control. However, sometimes you may need to initialize programmatically and avoid the interactive prompts.
To do so, provide all required fields for your recipe using the --field flag:

prefect init --recipe docker \
    --field image_name=my-repo/my-image \
    --field tag=my-tag

build:
- prefect_docker.deployments.steps.build_docker_image:
    requires: prefect-docker>=0.3.0
    image_name: my-repo/my-image
    tag: my-tag
    dockerfile: auto
    push: true

Once you've confirmed that these fields are set to their desired values, this step will automatically build a Docker image with the provided name and tag and push it to the repository referenced by the image name.
As the prefect-docker package documentation notes, this step produces a few fields that can optionally be used in future steps or within prefect.yaml as template values.
It is best practice to use {{ image }} within prefect.yaml (specifically the work pool's job variables section) so that you don't risk having your build step and deployment specification get out of sync with hardcoded values.

Note

Note that in the build step example above, we relied on the prefect-docker package; in cases that deal with external services, additional packages are often required and will be auto-installed for you.

Pass output to downstream steps

Each deployment action can be composed of multiple steps. For example, if you wanted to build a Docker image tagged with the current commit hash, you could use the run_shell_script step and feed the output into the build_docker_image step:

build:
    - prefect.deployments.steps.run_shell_script:
        id: get-commit-hash
        script: git rev-parse --short HEAD
        stream_output: false
    - prefect_docker.deployments.steps.build_docker_image:
        requires: prefect-docker
        image_name: my-image
        image_tag: "{{ get-commit-hash.stdout }}"
        dockerfile: auto

Note that the id field is used in the run_shell_script step so that its output can be referenced in the next step.

The push action¶

The push section is most critical for situations in which code is not stored on persistent filesystems or in version control. In this scenario, code is often pushed and pulled from a Cloud storage bucket of some kind (e.g., S3, GCS, Azure Blobs, etc.). The push section allows users to specify and customize the logic for pushing this code repository to arbitrary remote locations.

For example, a user wishing to store their code in an S3 bucket and rely on default worker settings for its runtime environment could use the s3 recipe:

prefect init --recipe s3
>> bucket: < insert bucket name here >

Inspecting our newly created prefect.yaml file we find that the push and pull sections have been templated out for us as follows:

push:
- prefect_aws.deployments.steps.push_to_s3:
    id: push-code
    requires: prefect-aws>=0.3.0
    bucket: my-bucket
    folder: project-name
    credentials: null

pull:
- prefect_aws.deployments.steps.pull_from_s3:
    requires: prefect-aws>=0.3.0
    bucket: my-bucket
    folder: "{{ push-code.folder }}"
    credentials: null

The bucket has been populated with our provided value (which also could have been provided with the --field flag); note that the folder property of the push step is a template - the pull_from_s3 step outputs both a bucket value as well as a folder value that can be used to template downstream steps. Doing this helps you keep your steps consistent across edits.

As discussed above, if you are using blocks, the credentials section can be templated with a block reference for secure and dynamic credentials access:

push:
- prefect_aws.deployments.steps.push_to_s3:
    requires: prefect-aws>=0.3.0
    bucket: my-bucket
    folder: project-name
    credentials: "{{ prefect.blocks.aws-credentials.dev-credentials }}"

Anytime you run prefect deploy, this push section will be executed upon successful completion of your build section. For more information on the mechanics of steps, see below.

The pull action¶

The pull section is the most important section within the prefect.yaml file. It contains instructions for preparing your flows for a deployment run. These instructions will be executed each time a deployment created within this folder is run via a worker.

There are three main types of steps that typically show up in a pull section:

set_working_directory: this step simply sets the working directory for the process prior to importing your flow
git_clone: this step clones the provided repository on the provided branch
pull_from_{cloud}: this step pulls the working directory from a Cloud storage location (e.g., S3)

Use block and variable references

All block and variable references within your pull step will remain unresolved until runtime and will be pulled each time your deployment is run. This allows you to avoid storing sensitive information insecurely; it also allows you to manage certain types of configuration from the API and UI without having to rebuild your deployment every time.

Below is an example of how to use an existing GitHubCredentials block to clone a private GitHub repository:

pull:
    - prefect.deployments.steps.git_clone:
        repository: https://github.com/org/repo.git
        credentials: "{{ prefect.blocks.github-credentials.my-credentials }}"

Alternatively, you can specify a BitBucketCredentials or GitLabCredentials block to clone from Bitbucket or GitLab. In lieu of a credentials block, you can also provide a GitHub, GitLab, or Bitbucket token directly to the 'access_token` field. You can use a Secret block to do this securely:

pull:
    - prefect.deployments.steps.git_clone:
        repository: https://bitbucket.org/org/repo.git
        access_token: "{{ prefect.blocks.secret.bitbucket-token }}"

Utility steps¶

Utility steps can be used within a build, push, or pull action to assist in managing the deployment lifecycle:

run_shell_script allows for the execution of one or more shell commands in a subprocess, and returns the standard output and standard error of the script. This step is useful for scripts that require execution in a specific environment, or those which have specific input and output requirements.

Here is an example of retrieving the short Git commit hash of the current repository to use as a Docker image tag:

build:
    - prefect.deployments.steps.run_shell_script:
        id: get-commit-hash
        script: git rev-parse --short HEAD
        stream_output: false
    - prefect_docker.deployments.steps.build_docker_image:
        requires: prefect-docker>=0.3.0
        image_name: my-image
        tag: "{{ get-commit-hash.stdout }}"
        dockerfile: auto

Provided environment variables are not expanded by default

To expand environment variables in your shell script, set expand_env_vars: true in your run_shell_script step. For example:

- prefect.deployments.steps.run_shell_script:
    id: get-user
    script: echo $USER
    stream_output: true
    expand_env_vars: true

Without expand_env_vars: true, the above step would return a literal string $USER instead of the current user.

pip_install_requirements installs dependencies from a requirements.txt file within a specified directory.

Below is an example of installing dependencies from a requirements.txt file after cloning:

pull:
    - prefect.deployments.steps.git_clone:
        id: clone-step  # needed in order to be referenced in subsequent steps
        repository: https://github.com/org/repo.git
    - prefect.deployments.steps.pip_install_requirements:
        directory: {{ clone-step.directory }}  # `clone-step` is a user-provided `id` field
        requirements_file: requirements.txt

Below is an example that retrieves an access token from a 3rd party Key Vault and uses it in a private clone step:

pull:
- prefect.deployments.steps.run_shell_script:
    id: get-access-token
    script: az keyvault secret show --name <secret name> --vault-name <secret vault> --query "value" --output tsv
    stream_output: false
- prefect.deployments.steps.git_clone:
    repository: https://bitbucket.org/samples/deployments.git
    branch: master
    access_token: "{{ get-access-token.stdout }}"

You can also run custom steps by packaging them. In the example below, retrieve_secrets is a custom python module that has been packaged into the default working directory of a Docker image (which is /opt/prefect by default). main is the function entry point, which returns an access token (e.g. return {"access_token": access_token}) like the preceding example, but utilizing the Azure Python SDK for retrieval.

- retrieve_secrets.main:
    id: get-access-token
- prefect.deployments.steps.git_clone:
    repository: https://bitbucket.org/samples/deployments.git
    branch: master
    access_token: '{{ get-access-token.access_token }}'

Templating options¶

Values that you place within your prefect.yaml file can reference dynamic values in several different ways:

step outputs: every step of both build and push produce named fields such as image_name; you can reference these fields within prefect.yaml and prefect deploy will populate them with each call. References must be enclosed in double brackets and be of the form "{{ field_name }}"
blocks: Prefect blocks can also be referenced with the special syntax {{ prefect.blocks.block_type.block_slug }}. It is highly recommended that you use block references for any sensitive information (such as a GitHub access token or any credentials) to avoid hardcoding these values in plaintext
variables: Prefect variables can also be referenced with the special syntax {{ prefect.variables.variable_name }}. Variables can be used to reference non-sensitive, reusable pieces of information such as a default image name or a default work pool name.
environment variables: you can also reference environment variables with the special syntax {{ $MY_ENV_VAR }}. This is especially useful for referencing environment variables that are set at runtime.

As an example, consider the following prefect.yaml file:

build:
- prefect_docker.deployments.steps.build_docker_image:
    id: build-image
    requires: prefect-docker>=0.3.0
    image_name: my-repo/my-image
    tag: my-tag
    dockerfile: auto
    push: true

deployments:
- # base metadata
    name: null
    version: "{{ build-image.tag }}"
    tags:
        - "{{ $my_deployment_tag }}"
        - "{{ prefect.variables.some_common_tag }}"
    description: null
    schedule: null

    # flow-specific fields
    entrypoint: null
    parameters: {}

    # infra-specific fields
    work_pool:
        name: "my-k8s-work-pool"
        work_queue_name: null
        job_variables:
            image: "{{ build-image.image }}"
            cluster_config: "{{ prefect.blocks.kubernetes-cluster-config.my-favorite-config }}"

So long as our build steps produce fields called image_name and tag, every time we deploy a new version of our deployment, the {{ build-image.image }} variable will be dynamically populated with the relevant values.

Docker step

The most commonly used build step is prefect_docker.deployments.steps.build_docker_image which produces both the image_name and tag fields.

For an example, check out the deployments tutorial.

A prefect.yaml file can have multiple deployment configurations that control the behavior of several deployments. These deployments can be managed independently of one another, allowing you to deploy the same flow with different configurations in the same codebase.

Working with multiple deployments with prefect.yaml¶

Prefect supports multiple deployment declarations within the prefect.yaml file. This method of declaring multiple deployments allows the configuration for all deployments to be version controlled and deployed with a single command.

New deployment declarations can be added to the prefect.yaml file by adding a new entry to the deployments list. Each deployment declaration must have a unique name field which is used to select deployment declarations when using the prefect deploy command.

Warning

When using a prefect.yaml file that is in another directory or differently named, remember that the value for the deployment entrypoint must be relative to the root directory of the project.

For example, consider the following prefect.yaml file:

build: ...
push: ...
pull: ...

deployments:
- name: deployment-1
    entrypoint: flows/hello.py:my_flow
    parameters:
        number: 42,
        message: Don't panic!
    work_pool:
        name: my-process-work-pool
        work_queue_name: primary-queue

- name: deployment-2
    entrypoint: flows/goodbye.py:my_other_flow
    work_pool:
        name: my-process-work-pool
        work_queue_name: secondary-queue

- name: deployment-3
    entrypoint: flows/hello.py:yet_another_flow
    work_pool:
        name: my-docker-work-pool
        work_queue_name: tertiary-queue

This file has three deployment declarations, each referencing a different flow. Each deployment declaration has a unique name field and can be deployed individually by using the --name flag when deploying.

For example, to deploy deployment-1 you would run:

prefect deploy --name deployment-1

To deploy multiple deployments you can provide multiple --name flags:

prefect deploy --name deployment-1 --name deployment-2

To deploy multiple deployments with the same name, you can prefix the deployment name with its flow name:

prefect deploy --name my_flow/deployment-1 --name my_other_flow/deployment-1

To deploy all deployments you can use the --all flag:

prefect deploy --all

To deploy deployments that match a pattern you can run:

prefect deploy -n my-flow/* -n *dev/my-deployment -n dep*prod

The above command will deploy all deployments from the flow my-flow, all flows ending in dev with a deployment named my-deployment, and all deployments starting with dep and ending in prod.

CLI Options When Deploying Multiple Deployments

When deploying more than one deployment with a single prefect deploy command, any additional attributes provided via the CLI will be ignored.

To provide overrides to a deployment via the CLI, you must deploy that deployment individually.

Reusing configuration across deployments¶

Because a prefect.yaml file is a standard YAML file, you can use YAML aliases to reuse configuration across deployments.

This functionality is useful when multiple deployments need to share the work pool configuration, deployment actions, or other configurations.

You can declare a YAML alias by using the &{alias_name} syntax and insert that alias elsewhere in the file with the *{alias_name} syntax. When aliasing YAML maps, you can also override specific fields of the aliased map by using the <<: *{alias_name} syntax and adding additional fields below.

We recommend adding a definitions section to your prefect.yaml file at the same level as the deployments section to store your aliases.

For example, consider the following prefect.yaml file:

build: ...
push: ...
pull: ...

definitions:
    work_pools:
        my_docker_work_pool: &my_docker_work_pool
            name: my-docker-work-pool
            work_queue_name: default
            job_variables:
                image: "{{ build-image.image }}"
    schedules:
        every_ten_minutes: &every_10_minutes
            interval: 600
    actions:
        docker_build: &docker_build
            - prefect_docker.deployments.steps.build_docker_image: &docker_build_config
                id: build-image
                requires: prefect-docker>=0.3.0
                image_name: my-example-image
                tag: dev
                dockerfile: auto
                push: true

deployments:
- name: deployment-1
    entrypoint: flows/hello.py:my_flow
    schedule: *every_10_minutes
    parameters:
        number: 42,
        message: Don't panic!
    work_pool: *my_docker_work_pool
    build: *docker_build # Uses the full docker_build action with no overrides

- name: deployment-2
    entrypoint: flows/goodbye.py:my_other_flow
    work_pool: *my_docker_work_pool
    build:
        - prefect_docker.deployments.steps.build_docker_image:
            <<: *docker_build_config # Uses the docker_build_config alias and overrides the dockerfile field
            dockerfile: Dockerfile.custom

- name: deployment-3
    entrypoint: flows/hello.py:yet_another_flow
    schedule: *every_10_minutes
    work_pool:
        name: my-process-work-pool
        work_queue_name: primary-queue

In the above example, we are using YAML aliases to reuse work pool, schedule, and build configuration across multiple deployments:

deployment-1 and deployment-2 are using the same work pool configuration
deployment-1 and deployment-3 are using the same schedule
deployment-1 and deployment-2 are using the same build deployment action, but deployment-2 is overriding the dockerfile field to use a custom Dockerfile

Deployment declaration reference¶

Deployment fields¶

Below are fields that can be added to each deployment declaration.

Property	Description
`name`	The name to give to the created deployment. Used with the `prefect deploy` command to create or update specific deployments.
`version`	An optional version for the deployment.
`tags`	A list of strings to assign to the deployment as tags.
`description`	An optional description for the deployment.
`schedule`	An optional schedule to assign to the deployment. Fields for this section are documented in the Schedule Fields section.
`triggers`	An optional array of triggers to assign to the deployment
`entrypoint`	Required path to the `.py` file containing the flow you want to deploy (relative to the root directory of your development folder) combined with the name of the flow function. Should be in the format `path/to/file.py:flow_function_name`.
`parameters`	Optional default values to provide for the parameters of the deployed flow. Should be an object with key/value pairs.
`enforce_parameter_schema`	Boolean flag that determines whether the API should validate the parameters passed to a flow run against the parameter schema generated for the deployed flow.
`work_pool`	Information on where to schedule flow runs for the deployment. Fields for this section are documented in the Work Pool Fields section.

Schedule fields¶

Below are fields that can be added to a deployment declaration's schedule section.

Property	Description
`interval`	Number of seconds indicating the time between flow runs. Cannot be used in conjunction with `cron` or `rrule`.
`anchor_date`	Datetime string indicating the starting or "anchor" date to begin the schedule. If no `anchor_date` is supplied, the current UTC time is used. Can only be used with `interval`.
`timezone`	String name of a time zone, used to enforce localization behaviors like DST boundaries. See the IANA Time Zone Database for valid time zones.
`cron`	A valid cron string. Cannot be used in conjunction with `interval` or `rrule`.
`day_or`	Boolean indicating how croniter handles day and day_of_week entries. Must be used with `cron`. Defaults to `True`.
`rrule`	String representation of an RRule schedule. See the `rrulestr` examples for syntax. Cannot be used in conjunction with `interval` or `cron`.

For more information about schedules, see the Schedules concept doc.

Work pool fields¶

Below are fields that can be added to a deployment declaration's work_pool section.

Property	Description
`name`	The name of the work pool to schedule flow runs in for the deployment.
`work_queue_name`	The name of the work queue within the specified work pool to schedule flow runs in for the deployment. If not provided, the default queue for the specified work pool will be used.
`job_variables`	Values used to override the default values in the specified work pool's base job template. Maps directly to a created deployments `infra_overrides` attribute.

Deployment mechanics¶

Anytime you run prefect deploy in a directory that contains a prefect.yaml file, the following actions are taken in order:

The prefect.yaml file is loaded. First, the build section is loaded and all variable and block references are resolved. The steps are then run in the order provided.
Next, the push section is loaded and all variable and block references are resolved; the steps within this section are then run in the order provided
Next, the pull section is templated with any step outputs but is not run. Note that block references are not hydrated for security purposes - block references are always resolved at runtime
Next, all variable and block references are resolved with the deployment declaration. All flags provided via the prefect deploy CLI are then overlaid on the values loaded from the file.
The final step occurs when the fully realized deployment specification is registered with the Prefect API