Welcome to the new week!

If you somehow missed it, a reminder to check the last week’s webinar with Andrea Magnorsky about Bytesize Architecture Sessions! If you’re curious how to enhance your system design with a quick design sessions that’s something for you. I called it “a Pomodoro for design workshops”. See the recording!

Marten got a new, big, major release! Work on it lasted a few months, and most of the overhaul was made by Jeremy D. Miller. It brought a big improvements around performance, querying, resilience and projection handling. Read more in Jeremy’s blog post:

• Jeremy D. Miller - Marten 7.0 is Released!

Emmett, my young OSS child, is also spinning its wheels. We just got support for ESDB and integration tests. Check more in the release notes.

I’ve been asked if I actually sleep. I do, at least once per night! Jokes aside, we're too often trying to show ourselves as busy bees. Sometimes it's a 21st-century fetish, sometimes it's accidental. In both cases, this can be dangerous as we may be putting pressure on other people who follow us. I wrote a short article on that for you and myself not to forget about that.

• Oskar Dudycz - Keeping our overachieving freak on a leash

Too often, people tell me that I'm constantly busy while I'm working less than I did a few years ago. Knowing myself and how mental health is important, I should also look more at not leaving such false impressions as that can cause FOMO for others. If you’re a leader, you should also consider the consequences, e.g., sending emails in late hours, etc.

Moving on! Let’s do the roundtrip through the big tech blogs and case studies!

Uber introduced a new tool for load management called Cinnamon. They’re transitioning from announced in 2018 QALM. Cinnamon uses a specific algorithm, TCP-Vegas, to adjust on the fly how many requests Uber’s system can take and a control method known as a PID controller to decide which user requests get through first. This setup is meant to prevent the system from getting too bogged down during busy times by prioritising important requests.

• Uber - Cinnamon: Using Century Old Tech to Build a Mean Load Shedder

• Uber - Introducing QALM, Uber’s QoS Load Management Framework

However, while this new setup aims to keep things running smoothly, it makes the system more complex. For the engineers maintaining it, this complexity means figuring out why something went wrong or how to fix it might not be straightforward. While Cinnamon could make Uber's services more reliable for users, it also requires the engineers to navigate through a more complicated system to keep everything in check. This can be fine for the Uber scale, but remember that if you consider applying a similar approach to your scenario.

Michał Grabowski wrote on the risk of blindly copying successful architectures. We should take inspiration and analyse it, but always remember to compare our context and evaluate differences.

• Michał Grabowski - Do not copy successful architectures

Still, it’s worth knowing the directions and analysing the impact of the decisions. It’s also worth knowing the trends. One of them seems to be cell-based/cellular architecture.

Over two months ago, I went through a Slack case study on applying it. Now DoorDash did the same:

• Doordash Engineering - Staying in the Zone: How DoorDash used a service mesh to manage data transfer, reducing hops and cloud spend

DoorDash and Slack both adopted cell-based architectures, but their implementations and motivations diverge, reflecting their operational needs. The need for enhanced system resilience and reliability drove Slack's implementation of a cell-based architecture. Following a network outage, Slack introduced a manual traffic management feature that allows engineers to quickly redirect traffic away from affected areas, minimizing user impact.

DoorDash addressed the challenge of rising costs due to data moving between different data centres by setting up a system that automatically manages where data goes (based on Envoy and Service Mesh). Essentially, their tech ensures that data stays close to where it's being used, reducing the need to send data over long distances, which can get expensive. This setup helps DoorDash save money by avoiding unnecessary costs.

Both approaches have pros and cons. DoorDash introduced automation but also increased the complexity. Slack reduced complexity by semi-automation. While DoorDash automated its process to save costs, Slack's approach lets team members step in quickly to keep everything working right for their users.

Cell-based architecture is an intriguing technique and highlights the need to understand where your data is and how you use it. Without that you won’t be able to shard it and put into the cells.

Stripe wrote a general explanation of their financial storage: Ledger database. They built their own solution focused on storing and tracking financial data. Stripe's Ledger system, an auditable and immutable log. The setup is fine-tuned for their financial network, and handling complex money movements. That’s critical for maintaining high reliability while processing millions of transactions.

There’s an overlap between that and event stores. They both are focused on storing immutable data and its quality. Event Sourcing works well for processes that can be modelled as state machines, and Stripe’s Ledger also makes it a foundational piece. It also allows events notifications, auditability and enhanced tracing (which is crucial for fraud detection etc.).

However, this sophistication comes at a cost. The system's complexity might pose challenges for understanding and troubleshooting, especially given the unpredictable nature of financial services and the potential for errors from external banking partners. While Ledger significantly enhances Stripe's data integrity and operational efficiency, its intricate architecture requires a deep understanding of the technology and the financial domain to manage effectively.

That’s also the main difference between event stores and ledger databases. Event stores are general purpose databases, while ledger focused on financial data semantics. Going down the niche allows domain-specifics optimisation, but also limits the scope of usage. Still, that also proves that Event Sourcing is not as niche as it may seem. Read more:

• Stripe - Ledger: Stripe’s system for tracking and validating money movement

Continuing with the event-driven tooling, tradeoff and how semantics impact the real application, check out David Boyne’s talk, where he goes through the common mishaps and misconceptions around it:

• David Boyne - Beyond producers and consumers: Avoiding common issues with event-driven architectures

Don’t miss also great write-ups on how to document your software, both the current, past and the intended state:

• Jacqui Read - Decision Archeology: Using ADRs with Existing Products

• Mario Carrion - Using the C4 model to document Software Architectures

Check also other links!

Cheers

Oskar

p.s. I invite you to join the paid version of Architecture Weekly. It already contains the exclusive Discord channel for subscribers (and my GitHub sponsors), monthly webinars, etc. It is a vibrant space for knowledge sharing. Don’t wait to be a part of it!

p.s.2. Ukraine is still under brutal Russian invasion. A lot of Ukrainian people are hurt, without shelter and need help. You can help in various ways, for instance, directly helping refugees, spreading awareness, and putting pressure on your local government or companies. You can also support Ukraine by donating, e.g. to the Ukraine humanitarian organisation, Ambulances for Ukraine or Red Cross.

Architecture

• Uber - Cinnamon: Using Century Old Tech to Build a Mean Load Shedder

• Doordash Engineering - Staying in the Zone: How DoorDash used a service mesh to manage data transfer, reducing hops and cloud spend

• Pinterest - Running Unified PubSub Client in Production at Pinterest

• Jacqui Read - Decision Archeology: Using ADRs with Existing Products

• Mario Carrion - Using the C4 model to document Software Architectures

• Ryan Shriver - DDD: Design Bounded Contexts Miro Template

• David Boyne - Beyond producers and consumers: Avoiding common issues with event-driven architectures

• Nathaniel Schutta - Analyzing and Critiquing Architectures

• Michał Grabowski - Do not copy successful architectures

Databases

• Stripe - Ledger: Stripe’s system for tracking and validating money movement

Testing

• Meta - Automated Unit Test Improvement using Large Language Models as Meta

Frontend

• Nikita Prokopov - JavaScript Bloat in 2024

• Adam Fortuna - How we Increased Search Traffic by 20x in 4 Months with the Next.js App Router

• Rainer Hahnekamp - Local Change Detection in Angular

AI

• The Verge - Google pauses Gemini’s ability to generate AI images of people after diversity errors

Java

• Paul Bakker - How Netflix Really Uses Java

.NET

• Jeremy D. Miller - Marten 7.0 is Released!

• Martin Ullrich - Optimize for the Cloud – Lightning-speed .NET Container Apps

• Cezary Piątek - Mocking authorization tokens with WireMock.NET

• Aaron Stannard - .NET Performance Optimization: Deferred Allocations

WebAssembly

• Bytecode Alliance - Announcing Jco 1.0

Coding Life

• Oskar Dudycz - Keeping our overachieving freak on a leash

• Damian Płaza - Inverse Bruce Lee maneuver

• Vadim Kravcenko - Falsehoods Junior Developers believe about becoming Senior

Management

• Martin Fowler - Periodic Face-to-Face

Industry

• Chris Heilmann - Apple vs. PWAs – Go act now and help avoid this disaster

• Open Web Advocacy - Open Letter to Tim Cook, Sabotaging Web Apps Is Indefensible

• Open Collective Official Statement - OCF Dissolution

Security

• TechCrunch - A leaky database spilled 2FA codes for the world’s tech giants