The article mentions that "Tokyo" DC was taken offline by disconnecting the data center from the "New York" and "Paris" DCs. Then writes were performed, Tokyo was reconnected, and the writes replayed, effectively re-syncing the DCs.
My question is: how was consistency maintained? What mechanism prevented the same record from being changed both in Tokyo and in New York while the network was split? Or, what mechanism resolved the conflicts? I think the article might be glossing over some details here. I'm guessing that either one side of the split was offline (or at least read-only) during the outage or that the writes were made without guarantees.
My question is: how was consistency maintained? What mechanism prevented the same record from being changed both in Tokyo and in New York while the network was split? Or, what mechanism resolved the conflicts? I think the article might be glossing over some details here. I'm guessing that either one side of the split was offline (or at least read-only) during the outage or that the writes were made without guarantees.