prefect-azure

prefect-azure is a collection of Prefect integrations for orchestration workflows with Azure.

Getting Started

Installation

Install prefect-azure with pip

bash

pip install prefect-azure

To use Blob Storage:

bash

pip install "prefect-azure[blob_storage]"

To use Cosmos DB:

bash

pip install "prefect-azure[cosmos_db]"

To use ML Datastore:

bash

pip install "prefect-azure[ml_datastore]"

Managed identity authentication for the Prefect server database

prefect-azure provides a Prefect plugin that lets the Prefect server connect to Azure Database for PostgreSQL using a Microsoft Entra ID (managed identity) token instead of a password. When enabled, the server acquires a short-lived Entra token via DefaultAzureCredential and supplies it to asyncpg on every new connection, so tokens refresh automatically and no database password is stored anywhere.

Enable it on the process running prefect server start (install prefect-azure in that image/environment):

bash

# Enable the plugin system so Prefect loads the database hook
# (on Prefect < 3.7 use PREFECT_EXPERIMENTS_PLUGINS_ENABLED=true instead)
export PREFECT_PLUGINS_ENABLED=true

export PREFECT_INTEGRATIONS_AZURE_POSTGRES_MANAGED_IDENTITY_ENABLED=true
# Optional: select a specific user-assigned identity
export PREFECT_INTEGRATIONS_AZURE_POSTGRES_MANAGED_IDENTITY_CLIENT_ID=<client-id>

# Provide a password-less connection URL (the plugin supplies the token)
export PREFECT_API_DATABASE_CONNECTION_URL="postgresql+asyncpg://<entra-principal>@<host>:5432/<db>"

The Postgres server must have Microsoft Entra authentication enabled and the identity mapped to a database role (via pgaadauth). Locally, DefaultAzureCredential falls back to your az login identity, so the same configuration works for development.

Examples

Download a blob

python

from prefect import flow

from prefect_azure import AzureBlobStorageCredentials
from prefect_azure.blob_storage import blob_storage_download

@flow
def example_blob_storage_download_flow():
    connection_string = "connection_string"
    blob_storage_credentials = AzureBlobStorageCredentials(
        connection_string=connection_string,
    )
    data = blob_storage_download(
        blob="prefect.txt",
        container="prefect",
        azure_credentials=blob_storage_credentials,
    )
    return data

example_blob_storage_download_flow()

Use with_options to customize options on any existing task or flow:

python

custom_blob_storage_download_flow = example_blob_storage_download_flow.with_options(
    name="My custom task name",
    retries=2,
    retry_delay_seconds=10,
)

Run a command on an Azure container instance

python

from prefect import flow
from prefect_azure import AzureContainerInstanceCredentials
from prefect_azure.container_instance import AzureContainerInstanceJob


@flow
def container_instance_job_flow():
    aci_credentials = AzureContainerInstanceCredentials.load("MY_BLOCK_NAME")
    container_instance_job = AzureContainerInstanceJob(
        aci_credentials=aci_credentials,
        resource_group_name="azure_resource_group.example.name",
        subscription_id="<MY_AZURE_SUBSCRIPTION_ID>",
        command=["echo", "hello world"],
    )
    return container_instance_job.run()

Use Azure Container Instance as infrastructure

If we have a_flow_module.py:

python

from prefect import flow
from prefect.logging import get_run_logger

@flow
def log_hello_flow(name="Marvin"):
    logger = get_run_logger()
    logger.info(f"{name} said hello!")

if __name__ == "__main__":
    log_hello_flow()

We can run that flow using an Azure Container Instance, but first create the infrastructure block:

python

from prefect_azure import AzureContainerInstanceCredentials
from prefect_azure.container_instance import AzureContainerInstanceJob

container_instance_job = AzureContainerInstanceJob(
    aci_credentials=AzureContainerInstanceCredentials.load("MY_BLOCK_NAME"),
    resource_group_name="azure_resource_group.example.name",
    subscription_id="<MY_AZURE_SUBSCRIPTION_ID>",
)
container_instance_job.save("aci-dev")

Then, create the deployment either on the UI or through the CLI:

bash

prefect deployment build a_flow_module.py:log_hello_flow --name aci-dev -ib container-instance-job/aci-dev

Visit Prefect Deployments for more information about deployments.

Azure Container Instance Worker

The Azure Container Instance worker is an excellent way to run your workflows on Azure.

To get started, create an Azure Container Instances typed work pool:

prefect work-pool create -t azure-container-instance my-aci-work-pool

Then, run a worker that pulls jobs from the work pool:

prefect worker start -n my-aci-worker -p my-aci-work-pool

The worker should automatically read the work pool's type and start an Azure Container Instance worker.