rules_prefect

Bazel rules for deploying Prefect flows as Docker images to work pools (e.g. GCP Cloud Run).

Setup

Add to your MODULE.bazel:

bazel_dep(name = "rules_prefect", version = "0.1.0")

You must also have rules_python with a pip hub that includes prefect:

bazel_dep(name = "rules_python", version = "1.8.4")
bazel_dep(name = "aspect_rules_py", version = "1.8.4")

pip = use_extension("@rules_python//python/extensions:pip.bzl", "pip")
pip.parse(
    hub_name = "pip",
    python_version = "3.13",
    requirements_lock = "//:requirements.txt",
)
use_repo(pip, "pip")

Your requirements.txt must include prefect>=3.6.

Usage

load("@rules_prefect//prefect:defs.bzl", "prefect_deploy")

prefect_deploy(
    name = "deploy",
    flow_file = "flow.py",
    entrypoint = "flow.py:my_flow",
    image_repository = "us-central1-docker.pkg.dev/my-project/prefect",
    work_pool_name = "my-cloud-run-pool",
)

Then deploy with:

bazel run //:deploy

prefect_deploy Parameters

Parameter	Required	Default	Description
name	Yes	—	Bazel target name (typically "deploy")
flow_file	Yes	—	Python source file containing the flow
entrypoint	Yes	—	Prefect entrypoint ("file.py:function")
image_repository	Yes	—	Docker image repository URL
work_pool_name	No	None	Prefect work pool name
deployment_name	No	"main"	Deployment name
base_image	No	"prefecthq/prefect-client:3.6.18-python3.13"	Base Docker image
concurrency_limit	No	None	Max concurrent flow runs
cron	No	None	Cron schedule string
cpu	No	None	Cloud Run CPU setting
memory	No	None	Cloud Run memory setting
timeout	No	None	Timeout in seconds
deps	No	[]	Additional Bazel dependencies

GCP Cloud Run Configuration

Set these environment variables for GCP integration:

Variable	Description
GCP_PROJECT_ID	Overrides image_repository with {region}-docker.pkg.dev/{project}/prefect
GCP_REGION	GCP region (default: us-central1)
ENVIRONMENT	Set to local for local development, or your env name for remote deploys

Local Development

For local development without pushing images:

ENVIRONMENT=local bazel run //my/flow:deploy

This uses BUILD_WORKSPACE_DIRECTORY to mount your local source tree into the Docker container instead of baking it into the image.

Apple Silicon

On Apple Silicon Macs, Prefect's built-in Docker build produces arm64 images. Use --buildx_push for amd64 work pools:

bazel run //my/flow:deploy -- --buildx-push

License

Apache 2.0