Personal Infrastructure & Hardware

Why Your Home Server Really Dies — A Raspberry Pi Failure and What to Do Differently

A Pi 4 running 24/7 for years finally bricked overnight after a building blackout. The postmortem is a textbook lesson in how cheap storage silently fails and how small servers should actually be built.

• The failure was the microSD card: used as the root drive with no write-limiting measures, it developed growing corruption that finally stopped the kernel from booting.
• SD cards are rated for only a limited number of write cycles per cell — roughly 10,000 to 30,000 — far fewer than a solid-state drive, and every swap, log, and cache entry burns through them.
• The owner lost almost nothing because real data lived on external HDDs and the whole configuration was stored as code — but that discipline is what saved him.

Most people who build a home server on a Raspberry Pi make the same convenient mistake: boot the whole operating system off the microSD because it is already there, fits in the case, and costs five dollars. The trouble is that an OS is chatty. It writes swap when memory is tight, logs every minute, refreshes package caches, and keeps databases ticking. A Pi 4 running media servers, a photo library, and a torrent client is doing thousands of tiny writes every day. Each one degrades the flash cells a little, and over years the filesystem accumulates errors that fsck can paper over at first — but eventually a single bad write during a power glitch is the last straw. The power-outage detail in this story is not incidental: a sudden cut mid-write is how most SD-card failures make themselves obvious.

The rebuild that followed is where the engineering lesson lives. The owner moved swap, temporary files, and volatile caches off the SD entirely, put important data on redundant external drives, and kept the entire system configuration as a NixOS file that could be re-created anywhere. In other words: assume the boot drive is cheap and disposable, and never store anything you cannot replace on it. The server itself is no longer a treasure chest; it is a boot mechanism that points at real storage.

The story of the "death and rebirth of my home server" resonated widely on Hacker News because it is genuinely true for nearly every small self-hosted setup — the failure mode is the same whether you run Jellyfin, Immich, a DNS sinkhole, or a personal wiki. The practical takeaway is free: separate boot from data, minimize writes to the root card, and write down (as code) exactly how you built the machine, so the next blackout is just a Tuesday.