4Project2026-09-12

Homelab storage that survives a dead node

This is a sample entry. It shows the layout and is not a real write-up.

The idea

Keep one copy of everything that matters on three small servers in three places, so that losing any one of them is an inconvenience and not an event.

1

Mirror everything, everywhere

Simple to reason about and far too slow over a home uplink. The first full sync never finished.

2

Sync on a schedule

Finished, but a node that was offline at the wrong hour came back with stale data and quietly won.

3

Snapshots shipped in order

Almost. Ordering held, but a pulled network cable mid-transfer left a snapshot half applied.

4

Snapshots, resumable, verified before they count

A transfer only becomes the new truth after it is complete and checked. Pulling the cable now costs minutes, not data.

What I would skip next time

Try 2. A schedule hides the question of which copy is newer, and that question always comes back.

All projects