The startup's Postgres survival guide

#693 – September 06, 2026

practical lessons from running Postgres in a growing startup

The startup's Postgres survival guide
14 minutes by Alexander Belanger

Alexander shares practical lessons from running Postgres in a growing startup. He explains how to design schemas, write efficient queries, manage connections, handle migrations, and tune autovacuum. He also covers query planning, batching, partitioning, locking, and large data migrations. The goal is to help engineers keep Postgres fast, reliable, and stable as their applications scale.

Giving agents API access is hard
sponsored by Nango

Agent integrations become complicated at scale, not just because of the number of APIs involved, but because each provider implements auth differently. Even though OAuth is a standard, token rotation, expiry, errors, and breaking changes vary widely. This article shares lessons Nango learned from operating auth across 900+ APIs.

What is Orasort? Understanding Oracle’s sorting aglorithm in simple terms
4 minutes by Nadeem Shaikh

Oracle's Orasort algorithm compares 8 bytes at a time by converting string chunks into 64-bit integers, instead of the traditional one byte at a time approach. This reduces CPU cycles and makes sorting up to 5 times faster. After its patent expired in 2024, open-source databases like MySQL and PostgreSQL began adopting it, and cloud companies now benefit from lower computing costs.

Fornax's guide to ridiculously fast ethernet
28 minutes by Wim

Getting a 100 Gbps server to actually run at full speed takes serious tuning across the entire system. Key Linux network settings need adjusting, including TCP buffer sizes, congestion control, and memory limits. The network card also needs careful configuration through ethtool, covering ring buffers and interrupt timing. Critically, disabling the IOMMU in the BIOS often makes the biggest difference, yet almost no guides mention it.

Cores in space: The core memory module from a 1980 Spacelab computer
25 minutes by Ken Shirriff

Spacelab, a reusable lab carried in the Space Shuttle's cargo bay, used a French-built computer with 128 kilobytes of magnetic core memory instead of modern semiconductor chips. Core memory stores each bit in a tiny magnetic ring, which can hold data without power and resists radiation, making it ideal for space use. Built in 1980, this was an advanced late-era core memory system using high-density components and a design that boosted performance by removing the need for inhibit wires. Semiconductor memory eventually replaced it in the 1990s.

Parity before cutover
3 minutes by Alex O'Callaghan

Big rewrites held back from production let risk build up silently. Bugs pile up, fixes get missed, and two versions drift apart until a painful cutover reveals everything at once. Downgrading to an older Python version first made it possible to run one codebase in both old and new infrastructure, so each migration step changed only one thing. By the time the cloud switch happened, it was uneventful.

And the most popular article from the last issue was:

newsletters