Apache Airflow for ML orchestration in MLOps - Step-by-Step Execution

Process Flow - Apache Airflow for ML orchestration

Define DAG and tasks

↓

Schedule DAG run

↓

Trigger task 1: Data ingestion

↓

Trigger task 2: Data preprocessing

↓

Trigger task 3: Model training

↓

Trigger task 4: Model evaluation

↓

Trigger task 5: Model deployment

↓

DAG run complete

Airflow runs a Directed Acyclic Graph (DAG) of tasks in order, scheduling and executing each step of the ML workflow automatically.

Execution Sample

MLOps

from airflow import DAG
from airflow.operators.python import PythonOperator
from datetime import datetime

def ingest():
    print('Data ingested')

def preprocess():
    print('Data preprocessed')

with DAG('ml_pipeline', start_date=datetime(2024,1,1), schedule_interval='@daily') as dag:
    t1 = PythonOperator(task_id='ingest', python_callable=ingest)
    t2 = PythonOperator(task_id='preprocess', python_callable=preprocess)
    t1 >> t2

Defines a simple Airflow DAG with two tasks: ingest data and preprocess data, where preprocess runs after ingest.

Process Table

Step	Task Triggered	Task Status	Output	Next Task
1	ingest	running	Data ingested	preprocess
2	ingest	success	Data ingested	preprocess
3	preprocess	running	Data preprocessed	none
4	preprocess	success	Data preprocessed	DAG complete

💡 All tasks completed successfully, DAG run finished.

Status Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	After Step 4	Final
Task Status (ingest)	none	running	success	success	success	success
Task Status (preprocess)	none	none	none	running	success	success
DAG Run Status	none	running	running	running	running	success

Key Moments - 2 Insights

Why does the 'preprocess' task wait before running?

Because in the execution_table, 'preprocess' starts only after 'ingest' shows 'success' at Step 2, showing task dependencies.

What happens if a task fails during execution?

The DAG run stops or retries depending on configuration; here, all tasks succeed as shown by 'success' status in the table.

Visual Quiz - 3 Questions

Look at the execution_table, what is the status of 'ingest' at Step 2?

A running

B failed

C success

D queued

💡 Hint

Check the 'Task Status' column for 'ingest' at Step 2 in the execution_table.

At which step does the 'preprocess' task start running?

A Step 3

B Step 2

C Step 1

D Step 4

💡 Hint

Look for when 'preprocess' has status 'running' in the execution_table.

If the 'ingest' task failed at Step 2, what would happen to the DAG run?

A The DAG would continue to 'preprocess' anyway

B The DAG run would stop or retry 'ingest'

C The DAG would skip 'preprocess' and finish

D The DAG would restart from the beginning

💡 Hint

Refer to the key_moments explanation about task failure impact on DAG execution.

Concept Snapshot

Apache Airflow runs ML workflows as DAGs.
Define tasks and their order with operators.
Airflow schedules and runs tasks automatically.
Tasks run only after dependencies succeed.
Failures can stop or retry tasks.
Use PythonOperators for Python code tasks.

Full Transcript

Apache Airflow helps automate ML workflows by defining tasks in a DAG. Each task runs in order, waiting for dependencies to finish. For example, data ingestion runs first, then preprocessing. The execution table shows task status changes from running to success. If a task fails, the DAG run may stop or retry. This visual trace helps understand how Airflow manages ML pipelines step-by-step.