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. She describes Slack’s approach to load testing and event surges, GitHub’s database and sharding constraints, and why she personally verifies AI-generated work on foundational systems.
The Pragmatic Engineer has published an interview with Maude Lemaire, a principal engineer whose work has included infrastructure at Slack and database scaling at GitHub. The discussion details how Slack prepared for a customer with 500,000 users, why some GitHub database queries constrained sharding, and how Lemaire approaches AI-generated code in foundational engineering work.
At Slack, Lemaire led the backend performance infrastructure team during nearly eight years at the company, according to The Pragmatic Engineer. One challenge was preparing for a customer expected to bring 500,000 users onto the service. A load test using headless Chrome instances was estimated to cost about $150,000 per five-hour run. Lemaire instead built Koi Pond, a tool that uses lightweight clients written in Go to establish WebSocket connections, rather than running a full Slack session in each browser instance.
The interview also describes two ways activity can create concentrated load. Slack’s custom emoji limit is given as 23, in part to reduce the fan-out burden when a message prompts many users to react. Separately, an event flood can overwhelm an individual client connected to many busy channels. These examples show that scaling is not only a matter of adding capacity: the pattern and distribution of user activity matter too.
At GitHub, where Lemaire joined in 2024, she worked on database scaling in a large Ruby on Rails application. The report says the home page could trigger around 100 SQL queries across three database clusters. Many queries lacked a sharding key and therefore had to run across multiple shards. The resulting scatter-query problem, the report says, limited GitHub’s ability to add more than 16 database shards at the time described.
Scaling Beyond More Servers
The examples matter to engineering teams because they connect everyday product behavior and application abstractions to infrastructure costs. A user interface decision such as allowing many custom emoji reactions can translate into large numbers of network fan-outs. A database abstraction can make application code easier to write while obscuring the volume and reach of queries it generates.
The load-testing account also offers a practical lesson for teams preparing for large customers: realistic tests need not reproduce every feature of a production client if the immediate question is whether the system can sustain a target number of connections. Koi Pond’s lightweight WebSocket clients were built for that test, according to the report. The article does not establish that this approach suits every workload, but it illustrates how test design can affect both cost and usefulness.
For readers responsible for reliability, the interview frames performance as a product concern rather than a response only to competition or a crisis. It also highlights the trade-off between developer convenience and operational visibility, a recurring challenge in large systems.
Products Worth Considering
DJI Mic 2 Wireless Microphone System (2 Transmitters, Receiver, Charging Case)
Samsung Galaxy Tab A11 Wi-Fi Tablet, 8.7-Inch, 64GB, 4GB RAM, Gray (SM-X133, International Version)
DJI Osmo Mobile 7P Smartphone Gimbal Stabilizer with Mic Mini Transmitter
As an affiliate, we earn on qualifying purchases.
From Slack Infrastructure to GitHub Databases
Lemaire’s career provides the basis for the comparison. The report says she spent nearly eight years at Slack, leading backend performance infrastructure, before joining GitHub to work on database scaling. She later worked on performance at Cursor, including time to first token, according to the article. The interview is presented as a discussion of lessons from scaling large software products, rather than as an announcement of a new product or company policy.
The GitHub discussion focuses on the consequences of a large Rails monolith and its use of Active Record, an abstraction for database access. In the account, developers could make calls without seeing the full database cost. When queries did not include a primary or sharding key, they could become scatter queries across shard instances. As the report explains, adding shards does not automatically solve this problem: queries that must reach every shard can become slower as the number of shards grows.
The source also describes Lemaire’s view of performance through an earlier example from Rent the Runway. She recalled a homepage that could take up to 10 seconds to fully load because of a high-resolution hero image. Lemaire said she worked on speeding it up in her spare time, despite being told competition was not a concern. Her stated principle was that teams can improve the experience by competing with their own prior performance.
“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 by The Pragmatic Engineer
Limits of the Reported Examples
The published account does not specify the interview’s publication date, the exact duration or outcome of the 500,000-user load test, or whether the customer ultimately joined Slack at that scale. The $150,000 figure is an estimate for the proposed headless Chrome approach, not a reported bill or a general cost for load testing.
The report gives historical details about GitHub’s queries and shard limits, but does not say whether those constraints remain in place or describe the system’s current architecture. It also does not provide measurements showing how much Koi Pond reduced testing costs or how much performance improved after the work discussed. The examples should be read as accounts from Lemaire’s experience, not as a complete current audit of either company.
The interview mentions Lemaire’s use of AI tools, but the supplied material does not identify particular models, quantify productivity effects, or describe a formal company-wide policy. Her verification approach is presented as her own practice, especially for foundational work.
The Interview’s Practical Takeaways
The immediate next step for readers is to consult the full interview and transcript linked by The Pragmatic Engineer, which the publication says are available alongside episode timestamps. Those materials may provide more detail on the cases summarized in the report. No further product launch, company change, or follow-up milestone is announced in the supplied source.
For engineering teams, the issues raised point to questions to examine in their own systems: whether load tests model the resource they need to measure, whether high-volume events create uneven client load, and whether database abstractions conceal scatter queries. Lemaire’s comments also suggest a distinction between using AI for assistance and delegating decisions about foundational architecture. The report does not prescribe a universal method, and its examples leave open how each practice performs in other systems.
Key Questions
What is the news about Maude Lemaire?
The Pragmatic Engineer published an interview with Lemaire about scaling Slack and GitHub, based on her engineering work at both companies.
How did Slack approach testing for 500,000 users?
Lemaire built Koi Pond, a Go tool that created lightweight clients for WebSocket connections. The report says a proposed test using headless Chrome instances was estimated at about $150,000 for a five-hour run.
Why did GitHub’s database sharding face limits?
According to the report, queries without a sharding key had to run across multiple shards. This scatter-query pattern made adding shards less effective, and the article says it constrained GitHub to no more than 16 shards at the time discussed.
What does Lemaire say about using AI in engineering?
The report says she uses AI but personally verifies its output, with particular care for foundational work. It does not give a specific model, productivity measure, or company-wide AI policy.
Source: newsletter.pragmaticengineer.com
