Setup Production like MLflow on Kubernetes with RDS Postgres database
I am a student learning DevOps and Cloud Computing. My blogs and articles are primarily a platform for me to post whatever I am learning. My passion to explain things in simple words also makes me use this platform as a way to teach fellow learners if possible. I would love to receive feedback from most people coming across my articles.
MLflow is needed to bring structure, reproducibility, and traceability to the machine learning lifecycle. As models evolve through experiments, hyperparameter tuning, and retraining, MLflow provides a centralized way to track experiments, metrics, artifacts, and model versions. In production setups, it acts as a source of truth for what was trained, with which data and parameters, enabling reliable deployments, audits, and collaboration across teams.
Pre-requisites
AWS Account
psql cli
kubectl
helm
RDS Postgres DB setup
Pre-requisites
AWS Account
psql cli
Create an RDS database
Go to RDS console > Databases > Create new database > Easy create > PostgreSQL > master username: postgres > self managed password > Create
Enable access
Go to the database > modify > Connectivity > Select Publicly accessible
Setup DB
Connect to the database with
psql -h <rds-endpoint> -p 5432 -U postgres
CREATE DATABASE mlflow; CREATE USER mlflow_user WITH PASSWORD 'your_secure_password'; GRANT ALL PRIVILEGES ON DATABASE mlflow TO mlflow_user;
\c mlflow
GRANT USAGE, CREATE ON SCHEMA public TO mlflow_user;
Kubernetes cluster setup
Create and connect kubectl to your cluster.
I am using a kind cluster you can also follow the same
kind create cluster —name <cluster-name>
kubectl create ns mlflow
helm repo add community-charts https://community-charts.github.io/helm-charts
helm repo update
helm install mlflow community-charts/mlflow --namespace mlflow --set backendStore.databaseMigration=true --set backendStore.postgres.enabled=true --set backendStore.postgres.host=your-postgres-host --set backendStore.postgres.port=5432 --set backendStore.postgres.database=mlflow --set backendStore.postgres.user=mlflow_user --set backendStore.postgres.password=your_secure_password
Expose the pod
kubectl get pods -n mlflow
Wait for pod to be ready
kubectl port-forward -n mlflow pod/<pod-name> 5000:5000 —address 0.0.0.0
Access MLflow on localhost:7006
