🎉 Giảm 20% cho đơn hàng đầu tiên!✦🚚 Miễn phí vận chuyển cho đơn trên 500.000₫✦⚡ Flash Sale — Số lượng có hạn!✦🎁 Quà tặng kèm cho đơn từ 1.000.000₫✦🎉 Giảm 20% cho đơn hàng đầu tiên!✦🚚 Miễn phí vận chuyển cho đơn trên 500.000₫✦⚡ Flash Sale — Số lượng có hạn!✦🎁 Quà tặng kèm cho đơn từ 1.000.000₫✦
Breakscale: Don't Wait for Your App to Crash to Learn Where It Breaks — Break It First
Quay lại Blog

Breakscale: Don't Wait for Your App to Crash to Learn Where It Breaks — Break It First

16 tháng 11, 20267 phút đọcVũ Khôi
English
#English
Vũ Khôi

Vũ Khôi

Tác giả tại Wotech Electronics Blog

Illustration: the Breakscale interface on GitHub, a Client - API Server - Database diagram with measured metrics

Breakscale: Don't Wait for Your App to Crash to Learn Where It Breaks — Break It First

You vibecode a web app, test it with 10 users and it runs fine, then run ads the next day, traffic pours in, and it crashes — and you have no idea where it broke or what to fix. That is the problem Breakscale, an open-source system design simulator, wants to solve by doing the opposite: build the system, raise the user load until it breaks, and show you where it breaks and why. This article is based on a post by the Code Web Không Khó page on 16 November 2026; the features and figures all follow the post and its attached interface screenshot, and have not been independently verified.

Quick answer: Breakscale is an open-source system design simulator (MIT license) where every number is measured from real requests running through the system, not drawn to look nice. Users drag and drop 33 components such as a load balancer, cache, database, queue, worker, CDN, circuit breaker, or database sharding to assemble a system, then drag the traffic slider up until something turns red and watch why it breaks. The tool ships with 23 ready-made lessons, each teaching one failure mode, plus learning rebuilds of the architectures of Netflix, Spotify, Discord, Uber, Twitter/X, Stripe, and WhatsApp. Breakscale works on the web, inside VS Code, and through an MCP server for AI assistants — all three ways need no account.

System design is the work of designing a software system's architecture: which components take part, how they connect, and how the system holds up as user load grows. Simulation in Breakscale means real requests run through the system you build, and every number — latency, queue depth, error rate — is measured from that process, instead of being computed from average figures.

How is Breakscale different from traditional system design learning?

According to the post, system design learning has always been static diagrams and lines like "add a cache" or "add a queue" — after hearing them, you still do not understand why the system crashes. Breakscale does the reverse: it lets you build the system, then raise the user load until it breaks, and shows you where it breaks and why.

When you drag the traffic slider up, you see latency rise, queues fill up, the circuit breaker trip on its own, and the whole system crash under a flood of retries. Everything is measured from each real request running through the system, not multiplied from averages to look nice. According to the interface screenshot attached to the post, the project has 2.3 thousand stars on GitHub, an MIT license, belongs to Vercel's open-source software program, and once reached #9 "Repository Of The Day" on TrendShift.

What does the Retry Storm example in Breakscale show?

The most famous example the post gives is the Retry Storm. At first only 100 people send requests per second. But because the database is slow, users cannot wait and resend. After repeated resends, 100 requests become 348 requests hammering the database. The database is 99.9% busy while achieving exactly zero useful results, because by the time any request finishes, its sender has already left.

According to the post, you watch it once and remember it for life: why one slow database can bring down an entire system. This is an illustrative example from the simulator itself, as described by the post's author.

What do the 33 components and 23 lessons in Breakscale include?

You get 33 pieces to assemble a system: load balancer, cache, database, queue, worker, CDN, rate limiting, circuit breaker, read replica, database sharding, autoscaling, stream channel, WebSocket gateway, serverless function, bulkhead, and more. You drag them onto the board, wire them up, and you have a system.

You get 23 ready-made lessons, each teaching one failure mode:

Lesson Failure mode simulated
Single server As the database fills up, latency spikes
With load balancer 3 servers share the work, but still talk to 1 database
With cache Dropping the cache hit rate makes the database carry everything
Async worker Work piles up when workers cannot keep up, then drains gradually
Retry Storm Repeated resends raise the load and cause a crash
Rate limiting Serving a little less, but what is served stays fast
Circuit breaker Cutting off the faulty part, then recovering on its own
Database sharding 1 shard crashes while the average still looks fine
Autoscaling Requests dropped while new machines are starting up

There are also rebuilds of Netflix, Spotify, Discord, Uber, Twitter/X, Stripe, and WhatsApp, built from the engineering posts they have published. The post states clearly: these are learning versions, not internal secrets. You can also break things further: kill a machine, slow one down, inject an error rate, cut a wire, and watch how the failure spreads.

What are the 3 ways to use Breakscale, and do you need an account?

According to the post, all three ways need no account:

Way 1: on the web. Open breakscale.tech, pick an example on the left, drag the traffic slider up until something turns red. Your design stays in your browser unless you share it; share links are encrypted before being sent.

Way 2: inside VS Code. Install Breakscale in VS Code, open it and it runs right next to your code, usable offline. Saves stay in VS Code; to move to the web, send a link or a file.

Way 3: with AI. Breakscale is also an MCP server, so Claude, Copilot, Cursor, and Codex can all use it. You only need to tell the AI: "Read this repo and draw its architecture as Breakscale, and tell me where you are guessing," or: "Design a link-shortening website that handles 2,000 requests per second, and show me where it breaks." The AI builds it, runs the simulation, and returns measured numbers plus a link for you to open and inspect.

According to the post, you do not need to know how to code to use this tool. You only need to be able to describe: what does my system contain, and when traffic grows, which part breaks first.

FAQ

1. What is Breakscale? An open-source system design simulator: you build a system from drag-and-drop components, raise traffic until it breaks, and read measured numbers to understand why it broke.

2. Is Breakscale free? According to the post and the interface screenshot, the project is open source under the MIT license, and the web version needs no account creation.

3. What is a Retry Storm? The phenomenon where, when a system is slow, users resend requests many times, multiplying the requests hitting the database and crashing the whole system — in the Breakscale example: from 100 requests per second to 348, the database 99.9% busy with zero results.

4. Can Breakscale be used with AI? Yes. Breakscale is an MCP server, so AI assistants such as Claude, Copilot, Cursor, and Codex can use it: the AI builds the system, runs the simulation, and returns measured numbers.

5. Are the Netflix and Spotify rebuilds their real architectures? No. According to the post, they are learning rebuilds based on engineering posts the companies have published, not internal secrets.

Source: a post by the Facebook page Code Web Không Khó, 16 November 2026. The features, figures, and examples in this article follow the post and its attached interface screenshot, and have not been independently verified.