Prefect¶

Prefect is a workflow automation system. QHub integrates Prefect with a feature flag as follows (in the top level):

prefect:
  enabled: true

You can also specify a particular image for Prefect with the image key as follows:

prefect:
  enabled: true
  image: prefecthq/prefect:0.14.22-python3.8

There are a bunch of components in getting Prefect working for you, here is a brief description of them:

Create a free Prefect cloud account here: https://cloud.prefect.io/
Create a Service Account and an API key for the same and add this to the CI secrets as TF_VAR_prefect_token:
- In GitHub: Set it in Secrets (https://docs.github.com/en/actions/reference/encrypted-secrets#creating-encrypted-secrets-for-a-repository)
- In GitLab: Set it as Variables (https://docs.gitlab.com/ee/ci/variables/#gitlab-cicd-variables)
Create a project in the Prefect Cloud Dashboard. Alternatively from CLI:

prefect create project 'your-prefect-project-name'

The TF_VAR_prefect_token API key is set as PREFECT__CLOUD__AGENT__AUTH_TOKEN environment variable in the agent. It’s used while deploying Prefect Agent so that it can connect to Prefect Cloud and query flows.

Prefect Cloud¶

Prefect Cloud is a fully hosted, production-ready backend for Prefect Core. Checkout prefect documentation to know more.

Prefect Agent¶

Prefect Agents is a lightweight processes for orchestrating flow runs. Agents run inside a user’s architecture, and are responsible for starting and monitoring flow runs. During operation the agent process queries the Prefect API for any scheduled flow runs, and allocates resources for them on their respective deployment platforms.

When you enable prefect via qhub-config.yml prefect agent is deployed on the QHub’s kubernetes cluster, which queries the Prefect Cloud for flow runs.

Agent configuration overrides¶

You can override your agent configuration without having to modify the helm files directly. The extra variable overrides makes this possible by changing the default values for the Agent chart according to the settings presented on your qhub-config.yaml file.

The current variables, originally available in the Agent helm chart that can be overridden include:

- IMAGE_PULL_SECRETS
- PREFECT__CLOUD__AGENT__LABELS
- JOB_MEM_REQUEST
- JOB_MEM_LIMIT
- JOB_CPU_REQUEST
- JOB_CPU_LIMIT
- IMAGE_PULL_POLICY

For example, if you just want to override the amount of CPU limits for each job, you would need to craft a declarative configuration, in you qhub-config.yaml file, as follows:

prefect:
 enabled: true
 overrides:
     prefect_agent:
       job:
         resources:
           limit:
             cpu: 4

Also, if you would like to include an extra variable to the agent environment configuration, that was not previously in the helm chart, you can do it by including it under the envVars field in the overrides block. For example, if you would like to add MY_VAR: "<value>" to you agent environment, you can do so by adding the following to your qhub-config

prefect:
 enabled: true
 overrides:
     envVars:
       MY_VAR: "<value>"

Adding secrets to your Agent configuration¶

Overrides also allow you to define extra secrets to pass through your agent configuration, for example, when using default secrets to automatically authenticate your flow with the listed service. In the Google cloud case, for GCP_CREDENTIALS context secret, you can do it by adding that specific key value pair into your configuration:

prefect:
 enabled: true
 overrides:
   secretEnvVars:
       PREFECT__CONTEXT__SECRETS__GCP_CREDENTIALS: '<Your value>'

This secret will then be stored as a kubernetes secret variable into you QHub secrets volume.

Flows¶

Prefect agent can only orchestrate your flows, you need an actual flow to run via prefect agent. The API for the same can be found in the prefect documentation Here is a simple example from their official doc:

from prefect import task, Task, Flow
import random

@task
def random_number():
    return random.randint(0, 100)

@task
def plus_one(x):
    return x + 1

with Flow('My Functional Flow') as flow:
    r = random_number()
    y = plus_one(x=r)

Storage¶

The Prefect Storage interface encapsulates logic for storing flows. Each storage unIt’s able to store multiple flows (with the constraint of name uniqueness within a given unit). The API documentation for the same can be found in the prefect documentation

Example: Creating, Building and Register Flow¶

Below is a complete example of

creating couple of workflows
Adding them to a storage (docker in this case)
Building and pushing that storage to a docker registry
Registering the flows to prefect cloud

import logging
import time

import prefect
from prefect import Flow, task
from prefect.storage import Docker


PREFECT_PROJECT = "your-prefect-project-name"
# Name of  your docker repository
DOCKER_REGISTRY = "quansight"


@task
def hello_world():
    """Sample flow to make sure prefect works."""
    logger = prefect.context.get("logger")
    logger.info("Running Hello World!")


def init_and_register_flows():
    storage = Docker(
        image_name="my-prefect-flows",
        image_tag=f"{time.time_ns()}",
        registry_url=DOCKER_REGISTRY,
        base_image="python:3.8",
        env_vars={
            "env_var": "value"
        },
    )
    prefect_flows = [
        Flow(
            name="hello-world-flow-1",
            tasks=[hello_world],
            storage=storage,
        ),
        Flow(
            name="hello-world-flow-2",
            tasks=[hello_world],
            storage=storage,
        ),
    ]

    add_flows_to_storage(storage, prefect_flows)
    storage.build()
    register_flows(prefect_flows, build=False)
    logging.info("Everything done.!")


def add_flows_to_storage(storage, flows):
    logging.info(f"Add flows: {flows} to storage: {storage}")
    flow_storage_paths = [storage.add_flow(flow) for flow in flows]
    logging.info(f"flows path in storage: {flow_storage_paths}")
    return storage


def register_flows(flows, **kwargs):
    logging.info(f"Registering flows: {flows}")
    registered_flows = [
        flow.register(project_name=PREFECT_PROJECT, **kwargs) for flow in flows
    ]
    logging.info(f"Registered flows: {registered_flows}")
    return registered_flows


def setup_logging(level=logging.INFO):
    """Setup standard logging format."""
    logging.basicConfig(
        level=level,
        format="%(asctime)s %(levelname)9s %(lineno)4s %(module)s: %(message)s"
    )


def main():
    setup_logging()
    init_and_register_flows()


if __name__ == "__main__":
    main()

Running your flows¶

Now that you have Prefect Agent running in QHub Kubernetes cluster, you can now run your flows from either of the two ways:

Triggering manually from the Prefect Cloud dashboard.
Running them on a schedule by adding a parameter to you flow. You can read more about it in the prefect docs.

ClearML

Traefik