MLflow Installation with PostgreSQL in EKS

Architecture

Kubernetes (EKS)
MLflow Tracking Server
PostgreSQL database
S3 bucket for artifacts
LoadBalancer / Ingress for access

1. Create PostgreSQL Database

You can use:

Amazon RDS PostgreSQL (Recommended)
PostgreSQL inside Kubernetes

Recommended: Amazon Web Services RDS PostgreSQL.

Example DB details:

DB_HOST=mydb.xxxxxx.ap-south-1.rds.amazonaws.com
DB_NAME=mlflow
DB_USER=mlflow
DB_PASSWORD=password123

Create database:

CREATE DATABASE mlflow;

2. Create S3 Bucket for MLflow Artifacts

aws s3 mb s3://mlflow-artifacts-prod

Example artifact path:

s3://mlflow-artifacts-prod

3. Create IAM Role for EKS Service Account (IRSA)

MLflow needs access to S3.

Create IAM policy:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "s3:*"
      ],
      "Resource": [
        "arn:aws:s3:::mlflow-artifacts-prod",
        "arn:aws:s3:::mlflow-artifacts-prod/*"
      ]
    }
  ]
}

Attach policy to Kubernetes service account using IRSA.

4. Create Kubernetes Namespace

kubectl create namespace mlflow

5. Create Kubernetes Secret

kubectl create secret generic mlflow-secret \
  --from-literal=DB_USER=mlflow \
  --from-literal=DB_PASSWORD=password123 \
  -n mlflow

6. MLflow Deployment YAML

apiVersion: apps/v1
kind: Deployment
metadata:
  name: mlflow
  namespace: mlflow
spec:
  replicas: 1
  selector:
    matchLabels:
      app: mlflow
  template:
    metadata:
      labels:
        app: mlflow
    spec:
      serviceAccountName: mlflow-sa

      containers:
      - name: mlflow
        image: ghcr.io/mlflow/mlflow:v2.12.1

        ports:
        - containerPort: 5000

        env:
        - name: AWS_DEFAULT_REGION
          value: ap-south-1

        - name: MLFLOW_S3_ENDPOINT_URL
          value: https://s3.amazonaws.com

        - name: DB_USER
          valueFrom:
            secretKeyRef:
              name: mlflow-secret
              key: DB_USER

        - name: DB_PASSWORD
          valueFrom:
            secretKeyRef:
              name: mlflow-secret
              key: DB_PASSWORD

        command:
        - /bin/sh
        - -c

        args:
        - |
          pip install psycopg2-binary boto3 && \
          mlflow server \
          --backend-store-uri postgresql://$(DB_USER):$(DB_PASSWORD)@mydb.xxxxxx.ap-south-1.rds.amazonaws.com:5432/mlflow \
          --default-artifact-root s3://mlflow-artifacts-prod \
          --host 0.0.0.0 \
          --port 5000

7. Service YAML

apiVersion: v1
kind: Service
metadata:
  name: mlflow-service
  namespace: mlflow
spec:
  type: LoadBalancer

  selector:
    app: mlflow

  ports:
  - protocol: TCP
    port: 80
    targetPort: 5000

8. Deploy MLflow

kubectl apply -f mlflow-deployment.yaml
kubectl apply -f mlflow-service.yaml

Check:

kubectl get pods -n mlflow
kubectl get svc -n mlflow

9. Access MLflow

http://EXTERNAL-IP

You should see:

MLflow Tracking UI

10. Configure Client

Install MLflow locally:

pip install mlflow

Set tracking URI:

export MLFLOW_TRACKING_URI=http://EXTERNAL-IP

Test:

mlflow experiments list