Skip to main content

High Availability (Multi-Node)

Full guide coming soon

This deployment model is available for production use. The complete step-by-step guide is being finalized. Contact us to plan a high-availability rollout and we will walk you through it.

Run DocAI Fabric across two or more nodes with shared storage, so that the platform keeps serving and processing documents even if a single node fails. This is the recommended model for production workloads.

For demos, evaluations, and small workloads, the simpler Standalone (Single Node) model is usually the better fit.

How it differs from Standalone

The application, OCR engines, and licensing work exactly as in the Standalone model. High Availability adds redundancy and horizontal scale:

AspectStandaloneHigh Availability
Nodes12 or more
StorageLocal folder on the hostShared storage reachable by every node
Job queue (Redis)On the same hostShared instance reachable by every node
Front doorDirect to the hostLoad balancer in front of the app nodes
Survives a node failureNoYes
ScaleTune workers on one hostAdd nodes

Architecture (overview)

                       ┌──────────────────┐
Users ───────► │ Load balancer │ (TLS)
└─────────┬─────────┘
┌──────────────┼──────────────┐
▼ ▼ ▼
┌──────────┐ ┌──────────┐ ┌──────────┐
│ app node │ │ app node │ … │ app node │
└────┬─────┘ └────┬─────┘ └────┬─────┘
└──────────────┼───────────────┘
┌─────────────┴─────────────┐
▼ ▼
┌──────────────┐ ┌──────────────────┐
│ shared Redis │ │ shared storage │
│ (job queue) │ │ (documents/state)│
└──────────────┘ └──────────────────┘

Every node runs the same application image. Because all nodes read and write the same shared storage and the same Redis queue, any node can serve a request or process any document.

What you will provide

A full checklist will accompany the complete guide. At a high level you will need:

  • Two or more nodes (virtual machines or servers) running Docker.
  • Shared storage that every node can mount (for example an NFS export), used as the document and state directory.
  • A shared Redis instance reachable by every node.
  • A load balancer in front of the app nodes, terminating TLS.
  • Outbound internet access for image pulls, licensing, and any cloud model providers, as in the Standalone model.

Backup and restore, updates, and log access follow the same principles as the Standalone model, applied to the shared storage and to each node. The complete guide will cover an automated update pipeline so new versions roll out to all nodes without manual steps.