Skip to main content

michelangelo.uniflow.core.image_spec

Container image specifications for Uniflow tasks.

This module provides the ImageSpec dataclass for defining custom container images and build recipes for task execution environments. ImageSpec allows tasks to specify their runtime environment independently from the default workflow container.

Example:

Specifying a custom container image:

from michelangelo.uniflow.core.image_spec import ImageSpec
from michelangelo.uniflow.core.decorator import task

@task(
    config=RayTask(head_cpu=4),
    image_spec=ImageSpec(
        container_image="docker.io/myorg/ml-tools:v1.2.3",
        recipe="bazel://path/to:build_target"
    )
)
def train_model(data):
    # Runs in custom container with specific ML libraries
    pass

ImageSpec Objects​

@dataclass
class ImageSpec()

ImageSpec defines container image specifications for uniflow tasks.

Example usage:

@uniflow.task(
    config=RayTask(head_cpu=1),
    image_spec=ImageSpec(
        container_image="docker.io/library/examples:latest",
        recipe="bazel://path/to:build_target"
    )
)
def my_task():
    pass

container_image​

The container image name/tag to use for task execution

recipe​

Build recipe/target for reproducible image builds