Skip to content

CONCEPT Cited by 1 source

Split-brain storage

A failure mode where the same data lives in two or more stores simultaneously, with neither designated as authoritative. When the stores drift, there is no clean reconciliation path — the system silently serves inconsistent data.

Definition

Split-brain storage occurs when a system writes to multiple stores at the end of each operation, treating both as the source of truth. The invariant "both stores have the same data" is maintained optimistically but has no enforcement mechanism, no drift-detection alert, and no canonical reconciliation procedure.

Canonical instance

cdnjs (pre-2026): files lived in both Workers KV (at the edge) and a GitHub repository. The ingestion pipeline wrote to both at the end of every run. When they drifted — a version that processed cleanly, wrote to KV, but silently failed to land in GitHub — it served fine for weeks until someone noticed. There was no alert for that. There couldn't be, because nothing in the system knew the full pipeline state (Source: sources/2026-07-30-cloudflare-dogfooding-at-scale-migrating-cdnjs-to-cloudflares-developer-platform).

Resolution

Designate a single source of truth. cdnjs resolved this by making R2 the sole authoritative store and demoting KV to metadata-only.

Seen in

Last updated · 604 distilled / 1,840 read