Towards Fast and Efficient Failure Handling for Paxos State Machines
Leander Jehl, Hein Meling · 2013
We investigate methods for handling failures in a Paxos State Machine and introduce Live Replacement, which quickly repairs failures at a low cost. Live Replacement enables a failed or disconnected replica to be replaced with a new one, with minimal disruption. Replacement does not rely on a special state machine command, and need not be ordered with respect to other commands. It is therefore independent from the state machine's progress. This enables Live Replacement to guarantee fast reaction to failures with minimal effect on the state machines operation.