Live12 DBEdge replication across regions
Live12 DB answers reads from regional edge endpoints so a lookup is served close to where the request originates. The trade-off with any geographically distributed read path is consistency: every edge has to converge on the same reference data, quickly, without a coordinator in the request path.
We treat the datasets as an append-only log of revisions. A change is written once, given a monotonic revision id, and replicated continuously to every edge node. Reads never block on replication — an edge always serves the latest revision it has, and the revision id makes it explicit which one that is.
Each region exposes its current revision through the gateway, so a client can tell whether two responses came from the same underlying state. For most reference data — country codes, currencies, calendar systems — revisions are infrequent and the whole fleet converges within seconds of a change being committed.
Because the data is the same everywhere, the Host header is the only
thing that selects a region; one artifact serves every node. That keeps operations
boring on purpose: there is no per-region content to drift, only a revision id to
compare.