MAAS Site Manager

During my time at Canonical, I worked extensively with the MAAS team. MAAS (Metal as a Service) is bare metal management software that can manage servers even without, for example, an operating system installed. The users are deeply technical IT professionals who manage large-scale infrastructure deployments.

I focused primarily on MAAS Site Manager, a companion application designed to address a significant pain point for users managing multiple data centers. While performance requirements dictate one MAAS installation per data center, this creates operational challenges: users lack unified oversight of their infrastructure, must replicate configurations and boot images across installations, and frequently struggle to ensure they’re performing actions on the correct MAAS instance.

MAAS Site Manager was created as the solution. A centralized management interface for multiple MAAS installations. I contributed to this project from its inception, helping shape it from initial concept through MVP delivery.

Geographic visualization challenge

One of the key features I designed for the MVP was the map visualization. Prior user research revealed that users needed geographic awareness of their MAAS installations to effectively manage distributed infrastructure. When troubleshooting errors, they required quick identification of affected regions, and when planning deployments, they needed visibility into data center locations across specific geographic areas.

In this research we also found that a map visualization would be the most intuitive solution for this geographic information. However, we also had a critical constraint from product management: the feature needed to function in air-gapped environments where internet connectivity is restricted or unavailable. This requirement was a significant technical and design challenge, as most mapping solutions depend on online services.

Design process and constraints

I began the project by creating several iterations of wireframes, addressing both visual and interaction design challenges. The map needed visual clarity for the specific use case while enabling users to interact with sites: selecting individual or multiple locations and performing actions on them.

The design had to accommodate potentially tens of thousands of sites, requiring scalable rendering solutions. Displaying 30,000 sites simultaneously would be neither technically performant nor visually useful to users. I needed to create user flows that were intuitive for site interaction and selection while ensuring the interface could scale effectively.

Throughout the design process, I conducted frequent feedback sessions with other designers to refine flows and visual approaches, and with developers to validate technical feasibility.

Technical roadblock and alternative solution

During one of these developer feedback session, we identified a significant technical challenge. Off-the-shelf open source map tiles (the building blocks of digital maps) require tens to hundreds of gigabytes of storage. Shipping this data with MAAS Site Manager would create an impractical burden for users, particularly since most self-host the application.

The development team proposed using an online map service like OpenStreetMaps to avoid hosting large datasets, but this would violate one of the core requirements, that it would work in an air-gapped environment. Beyond the technical constraint, I had additional concerns with this approach: limited control over visual appearance, excessive detail that could distract users from their primary tasks, and visual inconsistency with our brand identity.

Having recently worked with map data on a personal project (detecting potential new rock climbing areas using machine learning algorithms on satellite images), I knew there was a better solution. While the development team was hesitant to explore alternatives further, I decided to create a proof of concept demonstrating how to build a map with sufficient detail for our specific use case, custom styling that matched our brand, and dramatically reduced file size (tens of megabytes instead of tens of gigabytes).

I achieved this by customizing the tile building process (with tippecanoe), selecting only the map features necessary for our bare metal server management use case. This approach eliminated distracting details irrelevant to our use case while significantly reducing map tile size.

After developing the proof of concept, I convinced the development team that this approach was feasible. Using my prototype as a reference, they implemented a proper system for generating optimized maps for our application.

Results and validation

The result was a successful map feature in MAAS Site Manager that displays the appropriate level of detail, maintains visual consistency with our design system, and functions in air-gapped environments without requiring internet connectivity.

User testing with internal users validated the approach. Participants successfully completed tasks without requiring additional map detail, and the performance remained smooth due to the reduced file size. The map maintained visual consistency with the application’s styling and brand identity.

This project demonstrates how taking initiative to bridge the gap between design requirements and technical constraints can lead to solutions that meet user needs, business requirements, and brand standards simultaneously.