Manticore Search on Microsoft Azure: DX1's Story

Source: Dev.to

TL;DR

• DX1 uses Manticore Search for customer and parts search with a fast type‑ahead UX
• Chosen for open‑source licensing and speed
• Deployed on Azure VMs running Ubuntu, aligned with DX1’s existing Azure footprint
• Handles 20 M+ parts; best type‑ahead performance requires indexes in memory
• Scales by upgrading VM memory or adding nodes to a Manticore cluster
• Day‑to‑day operations are low‑touch and low‑maintenance

Overview

DX1 runs Manticore Search as a fast, user‑facing search layer for a catalog of more than 20 million parts and customer records. The architecture is intentionally simple: Manticore runs on Ubuntu‑based Azure virtual machines (VMs) alongside the rest of DX1’s Azure infrastructure, delivering responsive type‑ahead while staying “low‑touch” operationally. As data and traffic grow, DX1 scales by upgrading VM sizes or adding more nodes.

“We use it for searching through customers and parts data, we have a type‑ahead functionality that our customers love.” – Damir Tresnjo, DX1

Why Manticore Search?

• Open source – no licensing fees and full control over the stack.
• Performance – millisecond‑level response times for autocomplete and fuzzy search.

“Open source and very fast.” – Damir

The combination of speed, flexibility, and cost‑effectiveness made Manticore a natural fit for DX1’s workload.

Deployment on Azure

DX1’s entire infrastructure lives on Microsoft Azure, so deploying Manticore on Azure VMs was the straightforward choice. The team runs Ubuntu on the VMs and installs Manticore directly, avoiding any Azure‑specific managed search services.

“We run everything on Azure, so we deployed Manticore there as well.”

Architecture Highlights

• VM‑based deployment – simple Linux VMs provide the needed flexibility.
• No managed services – avoids additional cost and complexity.

Performance & Scaling

Manticore has proven fast and stable even at the scale of 20 M+ parts.

“It performs very fast, we have over 20 million parts we search through.”

Memory Considerations

Type‑ahead performance benefits from keeping the index in RAM. When the index outgrows the available memory, the VM memory must be increased.

“It does need the database to be in memory for the type‑ahead performance. As soon as the index outgrows available memory, we need to upgrade the VM memory.”

Scaling Strategies

1. Scale‑up – increase RAM on existing VMs.
2. Scale‑out – add more VMs to a Manticore cluster.

“We can scale each VM or we can add more VMs to a Manticore cluster.”

Operations

DX1 describes Manticore as low‑touch and low‑maintenance.

“Low touch, low maintenance, most of the time it just runs.”

The setup relies solely on predictable VM operations, with no special Azure features required.

Recommendation

DX1 would recommend Manticore Search to any team seeking a fast, reliable, and cost‑effective search engine.

“Yes, I would recommend Manticore to anyone looking for a fast, reliable and cost‑effective search engine.”

Getting Started

For teams interested in a similar VM‑based setup on Azure:

• Assess dataset size, query patterns, and latency targets.
• Plan RAM headroom for type‑ahead indexes (scale‑up first).
• Consider clustering if you anticipate further growth (scale‑out).

If you’d like a quick architecture review, get in touch with the Manticore team. Share details about your dataset, query workload, and latency goals, and they’ll help you validate an approach and plan the next steps.

Further Reading

• Overview of fuzzy search and autocomplete: New fuzzy search and autocomplete (replace with actual URL)