That's no different than just having a single region serving your application with users that are different distances and latencies away. The one farther away can issue an UPDATE/DELETE at the same time as the closer user, and the closer one would win.
It's all a game of latencies and eventual consistency, so what Fly.io is just treating writes as an HTTP request that has to travel all the way to the primary original, triggered by error an the regional location rather than your application actively doing it.
Comments
That's no different than just having a single region serving your application with users that are different distances and latencies away. The one farther away can issue an UPDATE/DELETE at the same time as the closer user, and the closer one would win.
It's all a game of latencies and eventual consistency, so what Fly.io is just treating writes as an HTTP request that has to travel all the way to the primary original, triggered by error an the regional location rather than your application actively doing it.