Enable Fault Tolerance and Failure Detection

Handle temporary failures

Many distributed systems use a strict read/write quorum, where an operation must receive responses from a minimum number of replicas before it can proceed. If enough replicas are unavailable and the quorum cannot be satisfied, the operation fails, reducing availability. To maintain availability during such failures, the system can use a sloppy quorum.

With a sloppy quorum, the coordinator may use reachable substitute nodes beyond the usual $\text{n}$ replica owners. This can allow an operation to proceed when the normal replica set is unavailable, provided the configured response count can still be met. Substitution weakens the fixed-set overlap assumptions. Example: Consider a configuration where $\text{n = 3}$. If Node $\text{A}$ is unavailable during a write, the request is sent to the next healthy node, Node $\text{D}$. Node $\text{D}$ processes the request and stores a hint indicating the data belongs to $\text{A}$. Once $\text{A}$ recovers, $\text{D}$ forwards the data to $\text{A}$ and removes the local copy.