Tailscale traced a series of database corruption outages to a 16-year-old bug in SQLite's Write-Ahead Log (WAL), which caused a rare data race between checkpoint and write operations that could permanently lose data. The bug, present since SQLite 3.7.0 (2010), was extremely difficult to reproduce and required SQLite maintainers to build a new diagnostic tool — funded by Tailscale — to finally identify it. The issue was only triggered by Tailscale's non-standard practice of manually controlling the checkpointing process, highlighting the risks of operating reliable software outside its typical configuration.
