Copies guard against loss
Replication stores the same data on multiple independent devices so that losing one does not lose the data. Durability is the probability that committed data survives, and adding replicas raises it sharply because all replicas must fail at once to cause loss.
Independence is everything
Replicas only help if their failures are independent. Two copies on the same disk, same machine, or same power rail can die together. Strong durability spreads replicas across separate disks, racks, and even data center regions so a single fault cannot take them all.
Synchronous versus asynchronous
- Synchronous replication acknowledges a write only after enough replicas have it, so no acknowledged write is lost, at the cost of higher latency.
- Asynchronous replication acknowledges immediately and copies in the background, which is faster but can lose the most recent writes if the primary fails first.
Restoring the count
When a replica is lost, the system re replicates to restore the target number before another failure can cause data loss. Faster repair means a smaller window of vulnerability.
Key idea
Replication raises durability by keeping independent copies across separate failure domains, with synchronous writes preventing data loss and prompt re replication shrinking the window in which further failures could destroy the last copy.