Installation & Setup

Understand Valohai's architecture and choose the right deployment model for your infrastructure

Valohai is a machine learning platform that runs on your infrastructure.

Valohai's architecture separates the application control plane from compute and data infrastructure. This guide covers the core components and deployment models for platform engineers setting up Valohai.

Deployment Models

Model	Application Layer	Compute & Data Layer	Best For
Hybrid	Valohai-managed (app.valohai.com)	Your infrastructure	Most teams. Teams who want managed platform updates and quick setup
Self-Hosted	Your infrastructure	Your infrastructure	Air-gapped environments, strict data residency, full control

Hybrid Deployment

The application runs on Valohai's infrastructure while compute and data remain in your environment. This ensures your code, data, and models never leave your control.

How it works:

• Users access Valohai at app.valohai.com

• Valohai communicates with your Redis queue machine to schedule jobs

• Workers in your infrastructure pick up and execute jobs

• All training data and artifacts stay in your storage

• Only high-level job metadata flow to Valohai's database

Requirements:

• Compute environment

  • Cloud account (AWS, Azure, GCP, Scaleway, Oracle)

  • on-premises infrastructure

  • (optional, for K8s workers) Kubernetes cluster

• Network access from app.valohai.com to your queue machine

• Object storage (S3, Azure Blob, GCS, or S3-compatible)

Self-Hosted Deployment

Both application and infrastructure run entirely within your environment. You control updates, access, and all network policies.

How it works:

• You host the Valohai application in your infrastructure

• Users access your instance at your domain

• All components communicate within your network

• Valohai provides application updates as Docker images

Requirements:

• Infrastructure for the application (bare metal, VM, or Kubernetes)

• PostgreSQL database

• Redis instance

• Object storage

Components

Component	Purpose	Specifications
Queue Machine	Manages job scheduling and routing	• Static VM or managed Redis service • 2 vCPUs, 4GB RAM minimum • DNS name • Runs Redis for job queue and short-term logs • TLS certificate for secure communication
Workers	Execute ML workloads	• Pull jobs from queue machine • Download inputs from object storage • Run code in Docker containers • Upload outputs to storage • Scale down when idle
Object Storage	Stores artifacts, models, and logs	• Git repository snapshots • Execution logs • Input datasets • Output artifacts (models, visualizations)

Worker Types

Valohai supports flexible worker infrastructure to match your environment:

Type	Description	Use Case
Autoscaled VMs	Cloud instances that scale based on demand	AWS, Azure, GCP, Scaleway, Oracle Cloud (OCI)
Kubernetes Pods	Containerized workers in K8s clusters	Cloud or on-premises Kubernetes
Static Machines	Physical or virtual machines	On-premises data centers, dedicated hardware
SLURM Nodes	HPC cluster integration	Existing SLURM-managed compute clusters

Supported Object Storage

• AWS S3

• Azure Blob Storage

• Google Cloud Storage

• Oracle Cloud Infrastructure (OCI) Object Storage

• S3-compatible storage (MinIO, NetApp, etc.)

Networking & Security

Hybrid Deployment

Source	Destination	Port	Purpose
app.valohai.com (34.248.245.191, 63.34.156.112)	Queue machine	63790	Job scheduling and queue management
Workers	Queue machine	63790	Pull jobs from Redis queue
Workers	app.valohai.com	443	Report status and metadata
Workers	Object storage	443	Download inputs, upload outputs
Workers	Internet	443	Pull Docker images and packages
Queue machine	Internet	80	Let's Encrypt certificate renewal (or use your own certificate)

Security principles:

• Code and data never leave your environment

• Workers never accept inbound connections from Valohai

• Only job metadata and logs are sent to Valohai's database

• All communication uses TLS encryption

Self-Hosted Deployment

Source	Destination	Port	Purpose
Users	Valohai app	443	Web interface access
Valohai app	PostgreSQL	5432	Database queries (internal only)
Valohai app	Queue machine	63790	Job scheduling
Workers	Queue machine	63790	Pull jobs from Redis queue
Workers	Valohai app	443	Report status and metadata
Workers	Object storage	443	Download inputs, upload outputs

💡 In self-hosted deployments, all network policies are under your control. The ports listed are defaults and can be customized during setup.

Choose Your Installation Path

I have AWS

• Hybrid: AWS Hybrid Deployment

• Self-hosted on EC2: AWS Self-Hosted EC2

• Self-hosted on EKS: AWS Self-Hosted EKS

I have Azure

• Hybrid (Marketplace): Azure Marketplace

• Hybrid (Manual): Azure Hybrid Deployment

I have GCP

• Hybrid: GCP Hybrid Deployment

I have Kubernetes

• Workers on any K8s: Kubernetes Workers

• OVH Cloud: OVH Kubernetes

• Oracle Cloud: Oracle Kubernetes

I have on-premises servers

• Linux workers: On-Premises Installation

I have OpenShift

• Self-hosted: OpenShift Self-Hosted

I have SLURM cluster

• SLURM integration: SLURM Cluster

What You'll Need

Every installation requires:

From Valohai:

• Queue address (e.g., yourcompany.vqueue.net)

  • You can also provision your own certificate

• Redis password (stored in your secret manager)

• For self-hosted: Docker images and license details

From your environment:

• Cloud account with appropriate permissions

• Object storage bucket

• Network connectivity between components

• For GPU workloads: GPU quota and drivers

Before you start:

• Review your organization's network and security policies

• Ensure you have necessary cloud permissions

• Decide on resource allocation (VM types, storage size)

• Plan your environment separation (dev/prod)

Getting Help

Contact [email protected] if you:

• Need help choosing the right deployment model

• Have specific security or compliance requirements

• Want guidance on resource sizing

• Encounter issues during installation