Hey everyone,

This week started with one of the biggest outages in recent years.

And no, it wasn’t Cloudflare this time 😆

Why the US remains at the center of the internet

The massive AWS outage affected numerous services including Amazon, Snapchat, Reddit, Signal, Roblox, Fortnite, Coinbase, Robinhood, and Vercel.

To say it was bad is an understatement. The outage exposed the internet's heavy dependence on a single cloud provider and affected millions of users worldwide.

According to Downdetector, the outage generated over 16 million user reports globally across more than 60 countries, with over 3,500 companies affected. The United States led with 6.3 million reports, followed by the UK with 1.5 million.

One platform that was not affected was X, with Elon Musk mocking AWS in several posts.

AWS identified the root cause as DNS resolution issues affecting DynamoDB API endpoints in the US-EAST-1 region, located in Northern Virginia.

The problem began at 11:49 PM PDT on October 19 (approximately 3 AM ET on October 20) when AWS experienced increased error rates for multiple services. Amazon determined that DNS could not properly resolve the location of DynamoDB's API endpoints, preventing applications from connecting to this critical database service. As DynamoDB failed, a cascade effect occurred, ultimately impacting 113 AWS services.

The company later clarified that "the root cause is an underlying internal subsystem responsible for monitoring the health of our network load balancers".

This incident underscores a critical vulnerability in modern internet infrastructure: the concentration of services within a handful of cloud providers. AWS holds approximately 30% of the global cloud market, and its US-EAST-1 region serves as a central control plane for AWS operations worldwide. The region's role as AWS's oldest and most critical hub means that failures there can trigger global disruptions, even for services architecturally designed with redundancy.

The outage serves as a stark reminder that true resilience requires distributing workloads across multiple regions and availability zones, rather than centralizing in a single location.

Cloudflare Workers gets 250% faster after public benchmarking controversy

Theo Browne recently called out Cloudflare's performance compared to Vercel, showing Workers running 3.5x slower on CPU-intensive tasks. Cloudflare's response? They fixed it and revealed some fascinating technical debt along the way.

Both Cloudflare and Vercel use V8 (Chrome's JavaScript engine), so why the massive performance gap? Turns out, it wasn't about CPU speed at all. Cloudflare's routing was creating a traffic jam by sending multiple requests to the same isolate, causing requests to queue rather than process slowly. Users weren't even being billed for the wait time.

The most embarrassing discovery: their garbage collector was using settings from 2017. One engineer (now their CTO) had configured it when Workers had just one developer, and nobody had touched it since. Changing this single configuration delivered a 25% performance boost instantly.

The investigation also revealed that OpenNext (the Next.js adapter) was unnecessarily copying entire data chunks. A 5MB response would be copied completely just to check its length, causing garbage collection to consume 25% of request time.

In perhaps the most surprising move, Cloudflare engineers discovered Node.js was 3x slower at trigonometric calculations due to a compiler flag. They submitted a pull request to Node.js to fix it, effectively making Vercel faster too. That's confidence.

The benchmark had issues too. The original comparison wasn't entirely fair:
- Cloudflare's version wasn't using force-dynamic mode
- React was running in development mode on Workers (production on Vercel)
- Network latency was being counted as CPU time

Results After Optimization:
✅ 250% overall performance improvement
✅ Matches or beats Vercel on all benchmarks except Next.js
✅ Changes already deployed globally to all Workers
✅ Existing applications got faster with zero code changes

This controversy demonstrates the power of public accountability in tech. A public benchmark forced Cloudflare to dig deep, uncovering seven-year-old technical debt, fixing memory leaks, and even improving V8 itself. The serverless wars might be heating up, but the real winners are developers using these platforms.

The worst crash so far, but markets have rebounded

Jim Cramer said to “buy crypto” 2 weeks ago. Well, that didn’t age well since the market crashed on 10th October.

Over $370 billion in market capitalization were erased within hours. Triggered by President Donald Trump’s unexpected announcement of 100% tariffs on Chinese imports, the geopolitical shockwave unleashed a chain reaction of liquidations across overleveraged positions on exchanges worldwide.

Total liquidations exceeded $19 billion. This event was nine times larger than any previous single-day liquidation event, surpassing even the sell-offs during the 2021 bull cycle corrections.

Some Top 20 cryptos fell to $0.001 on big exchanges like Binance. Some people managed to profit from it but these success stories are few and far between.

The bottom line is that this event was one of a kind in the crypto world, and that most cryptocurrencies rebounded to almost their initial level on the 12th October.

Have you noticed that AI generated website all use a purple color scheme? The main issue stems from defaults in TailwindCSS and the training data these models were exposed to.

Despite this quirk, vibe coding doesn't seem to be stopping anytime soon.

This coding approach is catching on fast. Startups and hobbyists love how it speeds up development and lets them focus on the creative side rather than sweating the technical details. Sure, it's not perfect and sometimes requires double-checking the output, but vibe coding's momentum is undeniable.

Hopefully nothing else breaks until next issue 🤞

Rares.