Tailscale traces 2025 outages to a 16-year-old SQLite write-ahead log flaw

Tailscale experienced intermittent outages beginning in August 2025, which were later linked to a 16-year-old flaw in SQLite’s write-ahead log (WAL) system. The issue arose because Tailscale’s control plane manually managed checkpointing to accelerate backups, a usage pattern that triggered a rare data race between checkpoint and write transactions. This race caused certain pages to be omitted from the main database file, corrupting the database and disrupting service.

With assistance from SQLite maintainers, Tailscale employed a newly built virtual file-system shim to log checkpoint activity and isolate the problem, now termed the WAL-Reset bug. SQLite has patched the defect, which had persisted since version 3.7.0 released in July 2010, and recommends users update to the corrected version. The incident highlights risks of operating reliable software in unconventional configurations.