Leader Election

A distributed system may require a process to act as a coordinator for performing functions such as initiating a computation, organizing a consensus on the status of partial computation, and the sequencing execution order of a set of tasks. Since processes are peers, no mechanism can distinguish a coordinating process from the others. Therefore, the participating processes should select one of the processes as the coordinator when needed. It is impossible to elect any peer as the leader unless there is a way to break the symmetry among the peers. The resolution of the symmetry‐breaking problem is possible using PIDs of the processes. The simplest solution to leader election requires a distributed system to be fully connected. Some algorithms can work on a ring‐based overlay network. Among these, the expanding‐subring algorithm is most interesting. However, we can find a spanning tree when network connectivity resembles a graph. The leader election in a tree network is equivalent to electing a node as the root. Ideally, the leader election problem is the same as distributed consensus problem. The processes must agree to a unique process to perform coordination activities. Therefore, it requires a solution to asynchronous communication when processes reside across clusters. Most practical solutions to distributed consensus have Paxos as the core protocol. Raft is a simpler alternative to Paxos.