block-cascade 2.5.4

Library for model training in multi-cloud environment.

cascade

Cascade is a library for submitting and managing jobs across multiple cloud environments. It is designed to integrate seamlessly into existing Prefect workflows or can be used as a standalone library.

Getting Started

Installation

poetry add block-cascade

or

pip install block-cascade

Example Usage

from block_cascade import remote
from block_cascade import GcpEnvironmentConfig, GcpMachineConfig, GcpResource

machine_config = GcpMachineConfig("n2-standard-4", 1)
environment_config = GcpEnvironmentConfig(
    project="example-project",
    region="us-west1",
    service_account=f"[e-mail]",
    image="us.gcr.io/example-project/cascade/cascade-test",
    network="projects/123456789123/global/networks/shared-vpc"
)
gcp_resource = GcpResource(
    chief=machine_config,
    environment=environment_config,
)

@remote(resource=gcp_resource)
def addition(a: int, b: int) -> int:
    return a + b

result = addition(1, 2)
assert result == 3

Configuration

Cascade supports defining different resource requirements via a configuration file titled either cascade.yaml or cascade.yml. This configuration file must be located in the working directory of the code execution to be discovered at runtime.

calculate:
  type: GcpResource
  chief:
    type: n1-standard-1
You can even define a default configuration that can be overridden by specific tasks to eliminate redundant definitions.

default:
    GcpResource:
        environment:
            project: example-project
            service_account: [e-mail]
            region: us-central-1
        chief:
            type: n1-standard-4

Authorization

Cascade requires authorization both to submit jobs to either GCP or Databricks and to stage picklied code to a cloud storage bucket. In the GCP example below, an authorization token is obtained via IAM by running the following command:

gcloud auth login --update-adc

No additional configuration is required in your application's code to use this token.

However, for authenticating to Databricks and AWS you will need to provide a token and secret key respectively. These can be passed directly to the DatabricksResource object or set as environment variables. The following example shows how to provide these values in the configuration file.

Persistent Resources in GCP

Cascade supports creating persistent resources in GCP. These resources can be reused across multiple tasks and will persist until deleted manually by the user. This can be useful for debugging tasks that involve large images that take a long time to be loaded onto a node or for reserving scarce resources like A100 GPUs.

You can create a persistent resource using the cascade CLI and suppling a cascade.yml with a configuration block that contains a persistent_resource_id field. This field will be used to identify the persistent resource when submitting tasks to it. It is recommended that you use the configuration file to define the resource as well as the tasks that will be submitted to it. This will ensure that the resource specified for your task is compatible with the shape of the persistent resource.

persistent-resource:
  type: GcpResource
  environment:
      project: example-project
      service_account: [e-mail]
      region: us-west1
      image: us.gcr.io/example-project/cascade/block-cascade
  chief:
      type: n1-standard-4
  persistent_resource_id: my-persistent-resource

create the persistent resource

cascade create-persistent-resource --config persistent-resource

You can then submit cascade tasks to this persistent resource

from block_cascade import remote


@remote(config_name="persistent-resource", job_name="hello-world")
def test_job():
    print("Hello World")


test_job()

Don't forget to delete the persistent resource when you are done with it

cascade delete-persistent-resource -i my-persistent-resource

Note: persistent resource ids can not be reused. If you delete a persistent resource, you will need to create a new one with a different id.

For Developers

Using hermit for managing Python

When developing cascade, you can optionally use hermit to manage the Python executable used by cascade. Together with using poetry to manage dependencies, this will ensure that your development environment is identical to other contributors. Follow the linked instructions for installing hermit and then you can create a virtualenv with Python@3.9 by running:

. ./bin/activate-hermit

Then, install the dependencies with poetry: poetry install