Rebuild it
Build a hash ring with virtual nodes that copies each key to the next three servers, then repair a drifted copy with a Merkle tree.
Four prompts, in order. Each builds on the last. Write as you go; your notes stay on this device and work offline.
Build a hash ring with virtual nodes that copies each key to the next three servers, then repair a drifted copy with a Merkle tree.
Dynamo hands conflicting versions back to the application to merge. When is that a good deal, and when is it a trap?
With three copies, reads from two and writes to two, a network split leaves one copy alone. What can each side still do?
A database with one primary refuses writes when the primary is unreachable. Which Amazon features could live with that?