Skip to content

Latest commit

 

History

History
53 lines (35 loc) · 2.41 KB

README.md

File metadata and controls

53 lines (35 loc) · 2.41 KB
Raw

Leader Election Algorithm

Build Status

The bully algorithm is one of methods in distributed computing for implementation of the leader election process for dynamically electing a coordinator by process ID number. The process with the highest number is selected as the coordinator.

Build

To run the algorithm you need a working installation of Erlang R15BXX (or later).

$ make get-deps
$ make compile-all

Usage

Configure timeout and nodes in the bully.config.

$ ./start.sh 1@127.0.0.1
$ ./start.sh 2@127.0.0.1
$ ./start.sh 3@127.0.0.1
$ ./start.sh 4@127.0.0.1

Then you can kill a node and see logs.

Description

Implementation consists of monitoring of the current leader and election a new one when the current is unavailable. The election of a new leader should be done as soon as possible. You have a distributed network from N nodes where each node described by an unique identity number.

Monitoring of the current leader

Each node sends ping every T seconds to the current leader and waits 4*T seconds for pong reply. On the waiting timeout the leader election process should be started.

Leader election process

  1. A node started the election starts sending announce election to all nodes with a higher identity number and waits for ok.
    • If no one ok after T seconds the node starts leading and sends new leader to all nodes.
    • If received ok the node starts waiting for new leader next T seconds. On new leader waiting timeout the node starts announcing election again.
  2. When receiving announce election a node replies with ok and starts the election process itself.
    • If the node received announce election is with the highest identity number it starts leading with sending new leader to all nodes.
  3. When receiving new leader a node saves a new leader and starts monitoring it.

Example

Example

Inspired by an article about a test task from aboutecho.com.

Copyright (C) 2014 Andrew Djoga, released under the MIT license.