Chain Flyte Entities

Data passing between tasks or workflows need not necessarily happen through parameters. In such a case, if you want to explicitly construct the dependency, flytekit provides a mechanism to chain Flyte entities using the >> operator.

Tasks

Let’s enforce an order for read() to happen after write(), and for write() to happen after create_bucket().

Note

To run the example locally, spin up the demo cluster using flytectl demo start.

Import the necessary dependencies.

import logging
from io import StringIO

import boto3
import pandas as pd
from flytekit import task, workflow

logger = logging.getLogger(__file__)

CSV_FILE = "iris.csv"
BUCKET_NAME = "chain-flyte-entities"


BOTO3_CLIENT = boto3.client(
    "s3",
    aws_access_key_id="minio",
    aws_secret_access_key="miniostorage",
    use_ssl=False,
    endpoint_url="http://minio.flyte:9000",
)

Create an s3 bucket. This task exists just for the sandbox case.

@task(cache=True, cache_version="1.0")
def create_bucket():
    try:
        BOTO3_CLIENT.create_bucket(Bucket=BUCKET_NAME)
    except BOTO3_CLIENT.exceptions.BucketAlreadyOwnedByYou:
        logger.info(f"Bucket {BUCKET_NAME} has already been created by you.")
        pass

Define a read() task to read from the s3 bucket.

@task
def read() -> pd.DataFrame:
    data = pd.read_csv(
        BOTO3_CLIENT.get_object(Bucket=BUCKET_NAME, Key=CSV_FILE)["Body"]
    )
    return data

Define a write() task to write the dataframe to a CSV file in the s3 bucket.

@task(cache=True, cache_version="1.0")
def write():
    df = pd.DataFrame(  # noqa : F841
        data={
            "sepal_length": [5.3],
            "sepal_width": [3.8],
            "petal_length": [0.1],
            "petal_width": [0.3],
            "species": ["setosa"],
        }
    )
    csv_buffer = StringIO()
    df.to_csv(csv_buffer)
    BOTO3_CLIENT.put_object(
        Body=csv_buffer.getvalue(), Bucket=BUCKET_NAME, Key=CSV_FILE
    )

We want to enforce an order here: create_bucket() followed by write() followed by read(). Since no data-passing happens between the tasks, use >> operator on the tasks.

@workflow
def chain_tasks_wf() -> pd.DataFrame:
    create_bucket_promise = create_bucket()
    write_promise = write()
    read_promise = read()

    create_bucket_promise >> write_promise
    write_promise >> read_promise

    return read_promise

SubWorkflows

Similar to tasks, you can chain subworkflows.

@workflow
def write_sub_workflow():
    write()


@workflow
def read_sub_workflow() -> pd.DataFrame:
    return read()

Use >> to chain the subworkflows.

@workflow
def chain_workflows_wf() -> pd.DataFrame:
    create_bucket_promise = create_bucket()
    write_sub_wf = write_sub_workflow()
    read_sub_wf = read_sub_workflow()

    create_bucket_promise >> write_sub_wf
    write_sub_wf >> read_sub_wf

    return read_sub_wf

Run the workflows locally.

if __name__ == "__main__":
    print(f"Running {__file__} main...")
    print(f"Running chain_tasks_wf()... {chain_tasks_wf()}")
    print(f"Running chain_workflows_wf()... {chain_workflows_wf()}")