Urgent.News

What's breaking now, across thousands of outlets.

Tech

Mac at any scale

In a recent video, I unboxed hundreds of MacBook Pros to deploy them in datacenters. This idea struck a chord with viewers, as it may seem absurd at first, but there are practical reasons behind it. We need to build custom Mac racks, start macOS instances quickly, and deliver new Xcode versions shortly after Apple's releases. Our customers require Macs to build and test iOS apps, as the iOS Simulator only works on a Mac, and Apple mandates using a recent Xcode version for App Store releases.

While a Mac Mini would be more suitable for a datacenter due to its remote management capabilities, power, and cooling, ordering a few hundred Minis has proven challenging due to supply chain issues. Apple's inconsistent release of CPU models further complicates matters, as the M5 cores for Mac Mini didn't arrive promptly. To address these challenges, we created the MacBook-in-Datacenter project.

Each MacBook in our system is connected to an IP KVM for remote management, a crucial feature when dealing with laptops. Unlike Mac Mini, laptop firmware is designed with a screen in mind, making remote maintenance actions easier. All MacBooks receive external 10G Ethernet NICs for reliable connectivity, and we pay meticulous attention to cable management, employing custom-made holders to wick heat and guide it away from nearby machines.

However, our project has also presented several challenges. MagSafe cables have been unreliable, with infrequent disconnections, while HDMI cables often fall out when touched. To tackle FileVault disk encryption, we found that unlocking it remotely using SSH is not feasible as the prompt disappears after five minutes of inactivity. The only workaround is to physically reboot the MacBook by pressing the button and unlocking it in time.

Despite these challenges, we have made significant progress. With the upcoming M5 and M6 Mac Minis and Studios, management will be simpler, featuring built-in 10 Gb network cards, automatic boot-up, and automatic OS configuration via MDM. We are also developing custom shelves for the Minis to optimize cooling through airflow simulations.

Our Namespace instances run on virtual machines, and we have integrated multiple virtual machine engines to support Linux, Windows, and Mac instances efficiently. Our Mac solution relies on Apple's Virtualization framework, which is comparable to Firecracker in its architecture, allowing us to provide a unified platform for various operating systems.

Written by urgent.news from Lobsters's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at namespace.so →

More in Tech

Plan Mode Saves Your Plan Where Only One Agent Can Find It. Commit PLAN.md Instead

Book: RAG Pocket Guide The series: Pocket Guides for Developers — 8 short books on agents, RAG, prompts, observability, events, databases and system design — the whole series My project: Hermes IDE |…

  • Claude Code stores plans in a default location accessible only to the creating agent.
  • Committing PLAN.md to the repository enables multiple agents to access and build upon the plan.

My Mermaid renderer taught me that headless Chrome is a 300MB tax on every deploy

I needed flowcharts and sequence diagrams as images inside an agent pipeline — the kind of thing you sketch in Mermaid because nobody wants to draw boxes by hand.

  • Headless Chrome added 300MB to container size, causing slow cold starts and OOM-killed workers
  • Developer replaced headless Chrome with Mermaid endpoint, eliminating browser tax on deployments

More from Wednesday 4 November →