Database Replication

Used to decrease latency and cope with system failures.

Leader-Follower

Writes happen to one designated leader which propagates writes to all replicas. This can happen sync or async.

Sync writes have the downside that the writes have to be propagated to all replicas before the transaction finishes, which increases latency especially if a replica becomes unavailable.

Async writes have a better latency, but can become out of sync if the write propagation fails.

If a leader fails the leader-rollover mechanism selects a new leader from one of the replicas, which use consensus algorithms like Paxos.

Mutli-Leader

Works similar to Leader-Follower, but assigns multiple leader that are available to handle writes. These leaders tend to propagate changes asynchronously between each other.

Leaderless

Each replica is read and write changes and propagate changes to all other nodes. Mechanisms like read-repair can detect inconsistencies and fix them.

Name 3 Database-Replication strategies. ?

• Leader-Follower
• Multi-Leader
• Leaderless

What are the respective downsides of sync and async write propagation using the Leader-Follower database replication strategy? ?

• Async: Can become out of sync if write fails
• Sync: Higher latency because writes need to complete on each replica

Which algorithm is commonly used for consensus during a leader-rollover? ? Paxos