Skip to content

Latest commit

 

History

History
104 lines (76 loc) · 3.24 KB

QUICKSTART.md

File metadata and controls

104 lines (76 loc) · 3.24 KB

Quick Start Guide

Deploy the MLOps platform and serve your first model in 5 minutes (after infrastructure is ready).

Prerequisites

  • One of the cloud environments deployed (see Getting Started):
    • make deploy-aws — AWS EKS (~15-20 min)
    • make deploy-azure — Azure AKS (~15-25 min)
    • make deploy-gcp — GCP GKE (~15-25 min)
  • kubectl configured for your cluster (deploy scripts do this automatically)

Step 1: Access the Dashboards

Open separate terminals for each port-forward:

make port-forward-mlflow    # http://localhost:5000  — Experiment Tracking
make port-forward-argocd    # https://localhost:8080 — GitOps (admin / see terminal output)
make port-forward-grafana   # http://localhost:3000  — Monitoring (admin / make secrets-<cloud>)

Step 2: Deploy an Inference Service

make deploy-example
kubectl get inferenceservice -n mlops

Expected output:

NAME           URL                                       READY   AGE
sklearn-iris   http://sklearn-iris.mlops.example.com     True    2m

Step 3: Test the Model

# Port-forward the inference service (uses 8081 to avoid ArgoCD conflict)
kubectl port-forward svc/sklearn-iris-predictor 8081:80 -n mlops &

curl -X POST http://localhost:8081/v1/models/sklearn-iris:predict \
  -H "Content-Type: application/json" \
  -d '{"instances": [[5.1, 3.5, 1.4, 0.2], [6.2, 2.9, 4.3, 1.3]]}'

Expected response:

{"predictions": [0, 1]}

Step 4: Run a Training Pipeline

# Register the workflow template
kubectl apply -f pipelines/training/ml-training-workflow.yaml

# Submit a pipeline run (requires Argo CLI: brew install argo)
argo submit --from workflowtemplate/ml-training-pipeline -n argo

# Watch progress
argo watch -n argo @latest

# Or view in the Argo UI
make port-forward-argo-wf   # http://localhost:2746

Step 5: Explore

Dashboard URL What to See
MLflow http://localhost:5000 Experiments, model registry, metrics
ArgoCD https://localhost:8080 GitOps applications, sync status
Grafana http://localhost:3000 Platform metrics, pod health, cost dashboard
Argo Workflows http://localhost:2746 Pipeline runs, DAG visualization

Next Steps

Troubleshooting

# InferenceService not ready?
kubectl describe inferenceservice sklearn-iris -n mlops
kubectl logs -l serving.kserve.io/inferenceservice=sklearn-iris -n mlops

# KServe controller issues?
kubectl logs -l control-plane=kserve-controller-manager -n kserve

# Workflow failed?
argo logs -n argo @latest

Clean Up

kubectl delete inferenceservice sklearn-iris -n mlops
pkill -f "port-forward"
make destroy-aws    # or destroy-azure, destroy-gcp