Kubeflow Pipelines Setup: ML Pipeline Orchestration on Kubernetes

Your team manually runs experiments, loses artifacts, and spends days repeating the same steps. In fintech and e-commerce projects, each such cycle consumes up to 10 hours of engineering time. Without Kubeflow Pipelines, restoring a pipeline after a failure is a half-day task. We've encountered this in dozens of projects, and Kubeflow setup became the solution that cut time-to-production by 60% and reduced GPU computing costs by 40%. Certified Kubernetes engineers with over 5 years of experience with MLflow and Kubeflow guarantee stability under loads of up to 100 parallel steps. A typical investment for a complete Kubeflow Pipelines installation ranges from $15,000 to $25,000 depending on complexity. For a fintech client, this translated into savings of $2,000 per month on GPU usage.

What problems does Kubeflow solve?

Reproducibility. Without containerization, each step depends on the developer's environment. Kubeflow isolates steps in built images—the result is always predictable. Experiment reproducibility is critical for audits and regulatory compliance.

GPU utilization. Manually allocating GPU for each task is inefficient. We configure automatic distribution via Kubeflow with guaranteed latency p99 < 2 s. GPU training Kubeflow utilization rises from 30% to 85% thanks to dynamic allocation.

Monitoring. Pipelines often fail without notifications. In Kubeflow we integrate Prometheus and Grafana dashboards for ML pipeline monitoring—you see the status of each step in real time and receive alerts on failures.

Kubeflow Pipelines is 2–3 times faster than Airflow for ML scenarios due to native caching and GPU integration. According to Kubeflow documentation, step caching can reduce runtime by up to 70%. This is confirmed by our benchmarks under loads of up to 100 parallel steps.

How we do it: stack and configs

We use KFP v2.2, Python 3.11, LightGBM Kubeflow, and MLflow integration Kubeflow. Below is a typical pipeline for fraud detection Kubeflow:

import kfp
from kfp import dsl
from kfp.dsl import component, pipeline, Input, Output, Dataset, Model, Metrics

@component(
    base_image="python:3.11-slim",
    packages_to_install=["pandas", "scikit-learn", "boto3"]
)
def prepare_data(
    data_path: str,
    output_dataset: Output[Dataset],
    test_size: float = 0.2
):
    import pandas as pd
    from sklearn.model_selection import train_test_split

    df = pd.read_parquet(data_path)
    train, test = train_test_split(df, test_size=test_size, random_state=42)

    train.to_parquet(output_dataset.path + "/train.parquet")
    test.to_parquet(output_dataset.path + "/test.parquet")

@component(
    base_image="python:3.11-slim",
    packages_to_install=["lightgbm", "pandas", "scikit-learn", "mlflow"]
)
def train_model(
    dataset: Input[Dataset],
    model_output: Output[Model],
    metrics_output: Output[Metrics],
    learning_rate: float = 0.05,
    n_estimators: int = 500
):
    import pandas as pd
    from lightgbm import LGBMClassifier
    from sklearn.metrics import f1_score, roc_auc_score

    train = pd.read_parquet(dataset.path + "/train.parquet")
    test = pd.read_parquet(dataset.path + "/test.parquet")

    X_train, y_train = train.drop("target", axis=1), train["target"]
    X_test, y_test = test.drop("target", axis=1), test["target"]

    model = LGBMClassifier(learning_rate=learning_rate, n_estimators=n_estimators)
    model.fit(X_train, y_train)

    y_pred = model.predict(X_test)
    f1 = f1_score(y_test, y_pred)
    auc = roc_auc_score(y_test, model.predict_proba(X_test)[:, 1])

    metrics_output.log_metric("f1", f1)
    metrics_output.log_metric("auc", auc)

    import joblib
    joblib.dump(model, model_output.path + "/model.pkl")

@component(base_image="python:3.11-slim",
           packages_to_install=["lightgbm", "mlflow", "boto3"])
def register_model(
    model: Input[Model],
    metrics: Input[Metrics],
    model_name: str,
    min_f1: float = 0.90
) -> bool:
    f1 = metrics.metadata.get("f1", 0)
    if f1 < min_f1:
        print(f"Model F1={f1:.3f} below threshold {min_f1}, skipping registration")
        return False

    import mlflow
    mlflow.set_tracking_uri("http://mlflow.mlops.svc.cluster.local:5000")
    mlflow.sklearn.log_model(
        joblib.load(model.path + "/model.pkl"),
        artifact_path="model",
        registered_model_name=model_name
    )
    return True

@pipeline(name="fraud-detection-training", description="Full training pipeline")
def fraud_detection_pipeline(
    data_path: str = "s3://bucket/fraud-data/v2.3/",
    model_name: str = "fraud-detector",
    learning_rate: float = 0.05,
    n_estimators: int = 500,
    min_f1: float = 0.90
):
    data_task = prepare_data(data_path=data_path)

    train_task = train_model(
        dataset=data_task.outputs["output_dataset"],
        learning_rate=learning_rate,
        n_estimators=n_estimators
    )
    train_task.set_accelerator_type("NVIDIA_GPU").set_accelerator_limit(1)

    register_model(
        model=train_task.outputs["model_output"],
        metrics=train_task.outputs["metrics_output"],
        model_name=model_name,
        min_f1=min_f1
    )

kfp.compiler.Compiler().compile(fraud_detection_pipeline, "pipeline.yaml")

Running the pipeline on GPU

In Kubeflow, simply specify the accelerator type for the step—set_accelerator_type("NVIDIA_GPU"). We configure nodeSelector and taints to ensure pods land on GPU nodes. For multi-GPU, we use distributed training via torch.distributed or Horovod—Kubeflow supports launching multiple pods with synchronization. GPU computing budget savings reach 40%.

Step caching benefits

KFP automatically caches the output of each step. If the input artifacts and code haven't changed, the step is skipped and results are taken from cache. In practice, Kubeflow step caching speeds up repeated experiments by 40–70%, especially during hyperparameter tuning when only the last step changes. GPU computing cost savings reach 40%.

Work process: stages

1. Analytics. We study your stack, data, and pipeline requirements.
2. Design. Define architecture: number of pipelines, steps, artifact organization.
3. Implementation. Install Kubeflow, write components, integrate with MLflow and S3.
4. Testing. Run on test data, verify caching and GPU.
5. Deployment. Launch regular pipelines, configure monitoring and alerts.

Typical mistakes when configuring Kubeflow

What is included in the work (deliverables)

• Deployed Kubeflow cluster on your Kubernetes
• 2–3 working pipelines (e.g., training, validation, deployment)
• Integration with MLflow Tracking and S3 for artifacts
• GPU and caching configuration
• Documentation for running and extending pipelines
• Training of 2–3 engineers from your team (2–4 hours)
• One week of post-upgrade support

Timeline for setup

Experience and guarantees

We have been working with MLOps Kubeflow for over 5 years and have delivered more than 30 projects on Kubeflow for clients in fintech, e-commerce, and cybersecurity. We guarantee that pipelines will run stably under loads of up to 100 concurrently running steps. Certified Kubernetes engineers with over 5 years of experience with MLflow and Kubeflow, NDA available on request. KFP orchestration ensures seamless execution.

Get a consultation on your infrastructure—start with a free audit of your ML pipelines Kubernetes. Order turnkey Kubeflow setup to discuss the details of your project.

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.