How half of the internet went down

#653 – November 23, 2025

No. It wasn't DNS.

Cloudflare outage on November 18
14 minutes by Matthew Prince

It wasn't the DNS. Cloudflare experienced a major network outage on November 18, 2025, lasting almost 6 hours. A database permission change caused a Bot Management feature file to double in size with duplicate entries. When this oversized file was distributed across Cloudflare's network, it exceeded software limits and caused widespread failures.

How Cursor, Intercom, and Cash App scale their databases
sponsored by PlanetScale

PlanetScale offers the fastest Postgres and MySQL available in the cloud. Blazing fast performance, ultra low-latency, unlimited IOPS, 99.999% uptime - all fully managed and available in AWS or GCP. Experience why companies like Cursor, Intercom, and Cash App choose PlanetScale to power their databases.

Bloom filters: The niche trick behind a 16× faster API
16 minutes by Mike Fisher

In this post Mike deep dives into how the team improved the P95 latency of an API endpoint from 5s to 0.3s using a niche little computer science trick called a bloom filter. He covers why the endpoint was slow, the options the team considered to make it fast and how they decided between them, and how it all works under the hood.

Start small, scale smart
5 minutes by Marcos F. Lobo

Incremental architecture is designed to grow with your code and system over time. This approach focuses on building small components that can be easily modified and extended. Key principles include organizing cross-functional teams first, using patterns only when needed, and following a simple loop: build small, ship, get feedback, and adjust.

Building a simple search engine that actually works
17 minutes by Michał Karbowiak

Michał explains how to build a custom search engine using your existing database instead of external services like Elasticsearch or Algolia. The system works by tokenizing content into searchable pieces, storing them with weights, and matching tokens during searches. It uses two simple database tables and multiple tokenizers for exact matches, partial matches, and typo handling. The search results are scored based on token weights, diversity, and document length to find the most relevant matches.

The internet is cool. Thank you, TCP
12 minutes by Moncef Abboud

TCP handles the unreliable internet to make apps work reliably. It manages packet loss, corruption, and reordering automatically so developers don't have to. TCP uses flow control to prevent overwhelming receivers and congestion control to prevent network collapse. The protocol establishes connections with a 3-way handshake and ensures data arrives in order through sequence numbers and acknowledgments.

And the most popular article from the last issue was:

newsletters