TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The Pragmatic Engineer published an interview with Maude Lemaire about scaling Slack and GitHub, drawing on her work at both companies. The discussion covers Slack’s WebSocket load testing and event-volume challenges, GitHub’s database queries and sharding constraints, and Lemaire’s approach to using AI tools.
The Pragmatic Engineer published an interview with Maude Lemaire, a principal engineer who worked on infrastructure at Slack and database scaling at GitHub, describing how the companies addressed performance challenges as their products grew. The discussion offers concrete examples of the trade-offs behind serving large, active user bases, including a Slack load test designed for 500,000 WebSocket connections and database queries that constrained GitHub’s ability to add shards.
At Slack, Lemaire led the backend performance infrastructure team during nearly eight years at the company, according to the source. One challenge discussed was preparing for a prospective customer with 500,000 users. A test built from headless Chrome sessions was estimated to cost $0.37 per instance for a five-hour session, making a full run prohibitively expensive. Lemaire instead built Koi Pond, a tool that used lightweight Slack-like clients written in Go to create WebSocket connections. The source says the straightforward browser-based approach would have cost about $150,000 per run.
The interview also describes how high-volume activity can stress Slack. A custom emoji posted in a large channel must be delivered to its members, so a burst of reactions can create heavy fan-out. The source says Slack limits custom emojis to 23 to help avoid this kind of load. A related problem, called an event flood in the interview, occurs when one client receives a sudden surge of events across many busy channels; that can overwhelm a power user’s client even if the service is handling activity across the wider system.
At GitHub, Lemaire worked on database scaling after joining in 2024, as Copilot adoption was accelerating, according to the report. The discussion focuses on GitHub’s large Ruby on Rails monolith and its use of Active Record. The source describes homepage loads that could issue around 100 SQL queries across three database clusters. It also says scatter queries—queries without a shard key that must run across multiple shards—made it difficult to add more than 16 database shards. These examples illustrate how application-level abstractions can hide the amount of work a request creates for databases.
The Cost of Scale at Slack and GitHub
The examples matter because infrastructure limits are often shaped by ordinary product behavior, not only by major outages or sudden traffic spikes. At Slack, a reaction or message can multiply into a large number of deliveries, while a single user connected to many busy channels can receive an unusually heavy stream of events. Tools such as Koi Pond let engineers test connection scale without paying to run a browser for every simulated user.
GitHub’s experience points to a different scaling risk: a database abstraction can make application code easier to write while making query costs less visible. When requests scatter across shards, adding more database partitions does not automatically solve the problem and may make some queries slower. For engineering teams, Lemaire’s account underscores the value of measuring real request patterns and checking how convenience in one layer affects capacity elsewhere.
The interview also discusses how Lemaire uses AI in engineering work. The source says she uses AI tools but personally verifies their output for foundational work, where errors could affect the systems built on top. That is her stated working approach, not a general performance finding or a claim that AI tools are reliable without review.
Products Worth Considering
DJI Mic 2 Wireless Microphone System (2 Transmitters, Receiver, Charging Case)
Kindle Paperwhite 16GB (2024) – 7 Glare-Free Display, Weeks Battery Life – Black
LG UltraGear 32GS60QC-B 32-inch Curved QHD Gaming Monitor (180Hz, 1ms, 1000R, AMD FreeSync, HDR10)
As an affiliate, we earn on qualifying purchases.
Lemaire’s Work Across Three Companies
Lemaire is the author of Refactoring at Scale: Regaining Control of Your Codebase, according to The Pragmatic Engineer. Before GitHub, she spent nearly eight years at Slack, including leading its backend performance infrastructure team. The interview places her work in the context of enterprise growth and the operational demands of a widely used collaboration product.
After joining GitHub in 2024, she faced scaling challenges in a large Rails application while the company’s Copilot product was taking off, the report says. The source also notes that Lemaire most recently worked on performance at Cursor, including time to first token. The episode is presented as a conversation about her experiences rather than a company announcement or a formal technical paper; its figures and examples are reported through that interview.
“The best competition is competition with yourself, knowing that you can do better, knowing that you can drive a better experience.”
— Maude Lemaire, as quoted in The Pragmatic Engineer’s interview
Details the Interview Leaves Open
The source material does not provide a publication date, complete episode transcript, or independent technical documentation for the specific metrics and examples. The figures—including the estimated $150,000 load-test cost and the 16-shard constraint—are presented in the interview’s account, and the source does not specify when each described system behavior was measured or whether it has since changed.
It is also unclear from the supplied material whether Slack’s 23-emoji limit remains in place today, how the companies’ architectures have evolved since Lemaire’s work, or what measurable performance gains resulted from the changes described. The account explains the problems and engineering responses but does not give comparative before-and-after service metrics.
Listen to the Full Interview
The source directs readers to the full episode and transcript through The Pragmatic Engineer, with listening options on YouTube, Apple, and Spotify. The transcript and episode timestamps may provide further detail on the examples, including the timing of the work and how the teams evaluated their systems.
No product launch, new infrastructure change, or follow-up milestone is announced in the supplied material. The next developments, if any, would depend on additional reporting or statements from Slack, GitHub, or Lemaire about how these systems have changed since the experiences discussed.
Key Questions
What is the news about Maude Lemaire?
The Pragmatic Engineer published an interview in which Lemaire discusses scaling work at Slack and GitHub, including performance testing, event delivery, and database constraints.
How did Slack test 500,000 user connections?
Lemaire built Koi Pond, a tool using lightweight clients written in Go to establish WebSocket connections. The approach avoided running a headless Chrome browser for every simulated Slack session, which the report estimated would have cost about $150,000 per five-hour run.
Why did GitHub’s database sharding become difficult?
The interview says many queries lacked a shard key and therefore had to run across shards. These scatter queries became slower as the number of shards increased, limiting GitHub’s ability to add more than 16 shards at the time described.
What does Lemaire say about using AI at work?
The source says Lemaire uses AI tools but personally verifies their output for foundational engineering work. It presents this as her approach, not as a claim that AI-generated work is automatically correct.
Source: newsletter.pragmaticengineer.com
