Apache Airflow for ML pipelines: setup, orchestration, automation

When orchestrating ML pipelines in production, a common challenge emerges: you need to chain preprocessing on CPU nodes, training on GPU nodes with different configurations, quality validation via metrics, and automatic deployment — all on a schedule, with rollbacks when metrics drop. Imagine daily retraining of a fraud detection model: loading data from S3, preprocessing on 4 CPUs, training on 1 GPU, F1 validation, and deploying to staging require coordination. Without orchestration, an engineer manually runs scripts, monitors logs, and loses hours on failures. Apache Airflow automates this process via DAG graphs, KubernetesExecutor for dynamic resource allocation, and integration with MLflow. For over 10 years, we have been setting up Airflow for ML pipelines — from small teams to enterprise clusters with 500+ DAGs. Our experience includes over 50 successful ML pipeline projects in fraud detection, NLP, and Computer Vision, where pipeline automation reduced experiment time by 40–60%, cut deployment incidents by 3x, and saved an average of $4,200 per month in infrastructure costs for one client. Compared to manual runs, Airflow reduces onboarding time for new models by 2–3x. DevOps hour savings reach 30–50%. We guarantee 99.9% SLA and provide documentation, monitoring, and team training.

How Apache Airflow solves ML orchestration problems

Airflow addresses key ML orchestration issues: heterogeneous resources (CPU/GPU), task dependency management, reproducibility, and fault tolerance. Each pipeline step is a separate task in a DAG: data preparation on a standard pod, training on a GPU pod with tolerations, quality validation via a Python operator, and model promotion. If quality drops (F1 < 0.90), the DAG stops with an error, preventing a bad model rollout. All metrics are logged to MLflow, enabling experiment comparison. Airflow with KubernetesExecutor is twice as good as CeleryExecutor for ML tasks in resource isolation: each GPU pod is isolated, not affecting neighboring tasks. This is critical under mixed workloads. In our experience, 80% of ML pipeline failures are due to resource contention; Airflow eliminates this.

Comparison of Airflow executors for ML

Airflow vs Kubeflow for ML: Key Differences

Airflow wins in versatility; Kubeflow in depth of ML integration. If your team already uses Airflow for ETL, migrating ML pipelines to it reduces infrastructure costs by 30%.

Installation with KubernetesExecutor

Follow these steps to set up Airflow for ML:

1. Install Helm (3.x) and add the Apache Airflow repository.
2. Create a values file (airflow-values.yaml) with executor=KubernetesExecutor, resource limits, and GPU tolerations.
3. Deploy Airflow using Helm with the command below.
4. Upload your DAGs and configure connections to MLflow, S3, etc.
5. Verify by triggering a test DAG.

# Installation via Helm (recommended) — <cite>Apache Airflow Helm Chart</cite>
helm repo add apache-airflow https://airflow.apache.org
helm upgrade --install airflow apache-airflow/airflow \
  --namespace airflow \
  --create-namespace \
  --set executor=KubernetesExecutor \
  --set config.logging.logging_level=INFO \
  --values airflow-values.yaml

ML pipeline as an Airflow DAG

from airflow import DAG
from airflow.providers.cncf.kubernetes.operators.pod import KubernetesPodOperator
from airflow.operators.python import PythonOperator
from airflow.operators.trigger_dagrun import TriggerDagRunOperator
from datetime import datetime, timedelta

default_args = {
    "owner": "ml-team",
    "retries": 2,
    "retry_delay": timedelta(minutes=5),
    "on_failure_callback": notify_on_slack,
}

with DAG(
    "fraud_detection_training",
    default_args=default_args,
    schedule="0 2 * * 1",  # every Monday at 2:00
    start_date=datetime(2025, 1, 1),
    catchup=False,
    tags=["ml", "fraud-detection"],
) as dag:

    # Data preparation — on regular pod
    prepare_data = KubernetesPodOperator(
        task_id="prepare_data",
        image="ml-pipeline:latest",
        cmds=["python", "prepare_data.py"],
        arguments=["--date={{ ds }}", "--output=s3://bucket/features/{{ ds }}/"],
        namespace="ml-pipelines",
        resources={"request_memory": "4Gi", "request_cpu": "2"},
        get_logs=True,
        is_delete_operator_pod=True,
    )

    # Training — on GPU pod
    train_model = KubernetesPodOperator(
        task_id="train_model",
        image="ml-pipeline-gpu:latest",
        cmds=["python", "train.py"],
        arguments=[
            "--data=s3://bucket/features/{{ ds }}/",
            "--run-name=fraud-{{ ds }}",
        ],
        namespace="ml-pipelines",
        resources={
            "request_memory": "32Gi",
            "request_cpu": "8",
            "limit_gpu": "1",
        },
        annotations={"nvidia.com/gpu": "1"},
        tolerations=[{"key": "nvidia.com/gpu", "operator": "Exists", "effect": "NoSchedule"}],
        get_logs=True,
    )

    # Evaluation gate — Python operator (cheap)
    def check_model_quality(**context):
        import mlflow
        client = mlflow.tracking.MlflowClient()
        run = client.search_runs(
            experiment_ids=[EXPERIMENT_ID],
            filter_string=f"tags.run_date = '{context['ds']}'",
            order_by=["metrics.f1 DESC"],
            max_results=1
        )[0]
        f1 = run.data.metrics.get("test_f1", 0)
        if f1 < 0.90:
            raise ValueError(f"Model quality too low: F1={f1:.3f} < 0.90")
        context["ti"].xcom_push(key="run_id", value=run.info.run_id)

    quality_gate = PythonOperator(
        task_id="quality_gate",
        python_callable=check_model_quality,
    )

    # Promotion — only if quality_gate passes
    promote_model = KubernetesPodOperator(
        task_id="promote_to_staging",
        image="ml-pipeline:latest",
        cmds=["python", "promote_model.py"],
        arguments=["--run-id={{ ti.xcom_pull(task_ids='quality_gate', key='run_id') }}"],
        namespace="ml-pipelines",
    )

    # Dependencies
    prepare_data >> train_model >> quality_gate >> promote_model

TaskFlow API (modern approach)

from airflow.decorators import dag, task

@dag(schedule="0 2 * * 1", start_date=datetime(2025, 1, 1))
def ml_pipeline():
    @task
    def prepare_data(execution_date: str) -> str:
        # Data preparation
        return f"s3://bucket/features/{execution_date}/"

    @task
    def train_model(data_path: str) -> dict:
        # Trigger training (or external job)
        return {"run_id": "xxx", "f1": 0.924}

    @task
    def promote_if_good(metrics: dict) -> None:
        if metrics["f1"] >= 0.90:
            promote_to_staging(metrics["run_id"])

    data = prepare_data()
    metrics = train_model(data)
    promote_if_good(metrics)

ml_pipeline()

Monitoring Airflow DAGs

The Airflow UI shows: status of each run, duration of each task, logs. Integration with Prometheus via airflow-exporter: airflow_dag_run_duration_seconds, airflow_task_fail_count. Alert on failed task via Slack/PagerDuty using on_failure_callback. For deep monitoring of ML metrics (data drift, prediction distribution), we recommend integrating Evidently AI or WhyLabs — they trigger retraining on drift.

What is included in turnkey Airflow setup

We provide the full setup cycle: audit of current infrastructure, design of DAG architecture considering ML specifics (GPU, big data), installation and configuration of Airflow on Kubernetes with Helm, setup of monitoring (Prometheus + Grafana) and alerting, integration with MLflow, writing 5–10 custom DAGs for your tasks, team training, and technical support during operation. Deliverables include architecture documentation, access to dashboards, training sessions, and a project report.

Contact us for a free consultation — we will analyze your project and propose the optimal architecture. Order Airflow implementation and get a stable ML pipeline in weeks, not months.

MLOps: Infrastructure for Training, Deploying, and Monitoring ML Models

The model is trained, metrics — F1 0.94 on validation. Three months later in production, quality drops by 12%. No one knows when — there is no monitoring. It's impossible to retrain quickly — the training script is in a Jupyter notebook of a data scientist who has already left. Data for retraining is collected manually from three disparate systems. About half of the projects come to us with this pain. We build a turnkey MLOps platform: from experiment tracking to automatic deployment and data drift monitoring. We will assess your infrastructure in 1–2 weeks, and in 4–6 weeks you will get a basic MLOps core running in production. Our team has 10+ years of experience in ML infrastructure, over 50 implementations.

How does MLOps infrastructure benefit your ML projects?

Experiment Tracking and Reproducibility

Without tracking, an ML project turns into chaos: it's unclear which checkpoint is better, which hyperparameters were used, which dataset. Reproducing a result a month later is a quest.

Why is experiment tracking the foundation of reproducibility?

MLflow is an open source standard for tracking. It logs parameters, metrics, artifacts (models, graphs), and code. MLflow Model Registry is a centralized model storage with versioning and lifecycle stages (Staging → Production → Archived). Deployment via MLflow Serving or integration with external systems.

Typical initialization in code:

import mlflow

mlflow.set_experiment("fraud-detection-v2")
with mlflow.start_run():
    mlflow.log_params({"learning_rate": 3e-4, "batch_size": 64, "epochs": 10})
    mlflow.log_metric("val_f1", val_f1, step=epoch)
    mlflow.pytorch.log_model(model, "model")

This is the minimum. In production, we add logging of system metrics (GPU utilization, memory), dataset (hash, version), code (git commit hash). Weights & Biases — richer UI, collaboration features, sweep for hyperparameter optimization. MLflow — for on-premise deployment without external dependencies.

DVC (Data Version Control) — versioning of data and models on top of git. Data is stored in S3/GCS/Azure Blob, only metadata (hashes) in git. dvc repro reproduces the entire pipeline from raw data to metrics.

To ensure reproducibility of training, fix random seeds (torch.manual_seed, numpy.random.seed, random.seed) and record them in experiment metadata. Without this, debugging irregular results is painful. Log the dataset version (DVC hash) and git commit — then any experiment can be reproduced down to the byte.

Pipeline Orchestration: Kubeflow, Airflow, Prefect

A pipeline orchestrator becomes necessary when: A 100-line training script in cron is fine for simple tasks. But as soon as you have a multi-step pipeline (data loading → preprocessing → feature engineering → training → validation → deployment if quality above threshold), you need an orchestrator with retry logic, visualization, and alerts.

Kubeflow — Kubernetes-native orchestrator for ML (see Kubeflow). Each step is a Docker container. Supports parallel steps, conditional branches, artifacts between steps. Integrates with Katib (AutoML), KServe (serving), Feast (feature store).

Apache Airflow — more general DAG orchestrator. Wide ecosystem of operators (S3, Spark, DBT, Kubernetes). Easier to deploy if Airflow already exists in the company.

Prefect / Metaflow — less boilerplate. Prefect 2.x with @flow and @task decorators — quick start for small teams.

Typical training pipeline architecture on Kubeflow:

1. Data ingestion component — fetches data from S3/DB, validates schema via Great Expectations
2. Preprocessing component — transformations, normalization, train/val/test split
3. Training component — training on GPU, logging to MLflow
4. Evaluation component — metric calculation, comparison with baseline in Model Registry
5. Conditional deployment — deploy only if new model is better than current by >2% F1

Each component is a separate Docker image. Pipeline is versioned in git. Scheduled run (retraining once a week on new data) or manual.

Model Registry and Lifecycle Management

Model Registry is not just a checkpoint store. It is a centralized system that knows:

• Which model is currently in production (and with what metrics)
• History of all versions with training parameters
• Metadata: dataset, git commit, validation results
• Lifecycle stage: None → Staging → Production → Archived

MLflow Model Registry — standard. For enterprise — Vertex AI Model Registry (GCP), SageMaker Model Registry (AWS), Azure ML Model Registry.

Model promotion through stages: automatically move model to Staging after successful eval, then manual or automatic (during A/B test) promotion to Production. Rollback — switch to previous Production version in seconds.

Serving: From FastAPI to Triton Inference Server

Simple case. FastAPI + PyTorch/ONNX on one server — 80% of production ML deployments are exactly that. Sufficient for most tasks with load up to 100 req/s.

from fastapi import FastAPI
import onnxruntime as ort

app = FastAPI()
session = ort.InferenceSession("model.onnx", providers=["CUDAExecutionProvider"])

@app.post("/predict")
async def predict(request: PredictRequest):
    inputs = preprocess(request.text)
    outputs = session.run(None, {"input_ids": inputs})
    return {"label": postprocess(outputs)}

Triton Inference Server — production standard for high loads (500+ req/s). Dynamic batching, concurrent model execution, model ensemble. Supports TensorRT, ONNX, PyTorch TorchScript, TensorFlow SavedModel.

KServe — Kubernetes-native ML serving with autoscaling, canary deployments, A/B testing out of the box. Scale-to-zero for inactive models — savings on infrastructure up to 40% annually for a project with 10 models.

Monitoring: Data Drift, Model Drift, Infrastructure Metrics

Monitoring — what is usually done last and regretted first. Three levels.

Infrastructure monitoring. Latency (P50/P95/P99), throughput (req/s), error rate (4xx, 5xx), GPU/CPU utilization. Prometheus + Grafana — standard. Alert when P99 latency > threshold or error rate > 1%.

Data drift monitoring. Distribution of input data changes over time. Detect via PSI (Population Stability Index) for numerical features: PSI > 0.2 — strong drift. Chi-squared test for categorical, Kolmogorov-Smirnov test for continuous. Evidently AI — open source library with ready-made drift tests.

Model drift monitoring. If ground truth is delayed (e.g., we know conversion after a week) — monitor real metrics. If not — surrogate metrics: distribution of prediction scores, proportion of confident predictions.

Alerting. Three levels: INFO (minor drift, log it), WARNING (significant, notify team), CRITICAL (quality dropped below threshold — automatic switch to fallback model).

Why is data drift monitoring important?

Without it, you learn about model degradation only from user complaints or ringing SLA. A drift alert allows you to retrain the model in advance, before errors start causing losses. In one of our projects, PSI monitoring detected drift 2 days after a data source change — this saved the campaign.

Feature Store

Feature Store solves the training-serving skew problem. If preprocessing during training and inference is implemented in two different places — divergence is inevitable.

A Feature Store is needed when:

• Several models use the same features
• Features are computed from streaming data (real-time)
• Large team with different people on feature engineering and model training

Feast — open source Feature Store. Offline store (S3 + Parquet) for training, online store (Redis, DynamoDB) for low-latency inference. Feature definitions as code, materialization job syncs offline → online.

Tecton (commercial), Vertex AI Feature Store (GCP), SageMaker Feature Store (AWS) — managed options with less ops overhead.

CI/CD for ML

ML CI/CD is regular CI/CD plus specific ML steps.

ML-specific checks in CI:

• Reproducibility check: run training with a fixed seed, result must match
• Data validation: Great Expectations or Pandera on schema/distribution checks
• Model performance check: automatic eval on holdout, block merge if degradation > threshold
• Latency regression test: inference must meet SLA

GitOps for deployment. Merge to main → CI triggers training → eval → if passes → automatic deployment to Staging → smoke tests → manual promotion to Production or automatic upon successful canary.

Tools: GitHub Actions / GitLab CI for CI, ArgoCD for GitOps deployment on Kubernetes.

What's Included in MLOps Platform Development

We provide a full cycle of work, documentation, and team training.

Total time from audit to full MLOps platform: 3–5 months. Also possible phased launch: basic level (tracking + serving) in 4–6 weeks.

Cost is calculated individually based on data volume, number of models, and infrastructure requirements. Order an MLOps infrastructure audit — get a roadmap in 1–2 weeks. Contact us for a project assessment — we will send a preliminary estimate within 2 business days.

Note: warranty on architectural solutions — 12 months. We provide integration certificates with major cloud providers (AWS, GCP, Azure). During our work, we have not lost a single client after the first implementation — the experience of 50+ successful MLOps projects speaks for itself. Get a consultation on building an MLOps platform today.