Cloud Haskell nodes

Program for broadcasting messages based on random numbers.

Usage

You can run program with flag -h to see help info:

Usage: nodes --send-for S --wait-for S [--with-seed NUM] --nodes-conf PATH
  Cloud Haskell nodes

Available options:
  -h,--help                Show this help text
  --send-for S             Sending period, in seconds
  --wait-for S             Grace period, in seconds
  --with-seed NUM          Seed value for RNG
  --nodes-conf PATH        Path to config file with nodes' endpoints

Example command:

$ nodes --send-for 8 --wait-for 2 --nodes-conf ./nodes.conf

It means nodes will send messages to each other during 8 seconds. After that, during 2 seconds, nodes will print out results. ./nodes.conf is the path to nodes config file.

Nodes config file

Example of config file:

127.0.0.1:10300
127.0.0.1:10200
127.0.0.1:10201
127.0.0.1:10202

In this case program will try to start three last nodes, on 127.0.0.1:10200, 127.0.0.1:10201 and 127.0.0.1:10202. These are worker nodes. First endpoint is for special node (for special broadcasting commands).

Workflow

Each node contains two processes, sender and receiver. Sender generates random numbers and sends them (with timestamps) to all other nodes. Receiver receives such a messages and collects them. Example of message:

(1481616425.126990, 0.28400448513890497)

First element - time message was sent at (UTC-timestamp with milliseconds), second one - random number from range (0, 1].

After sending period expires, we send "stop message" to all nodes. Nodes stops sending messages and prepares result. Result is a tuple like this:

([0.28400448513890497, 0.39491548517890561, ...], 245.09809543677)

First element - list of the messages (without timestamps) sorted by sent time, second one - sum of the elements by formula:

m
Σ i * m(i)
i=1

where i - index of element in list of received messages, m(i) - i-th message from some node.

After preparing result node prints it out and dies. All nodes must die during grace period, otherwise program will killed.

Fault tolerance

If we can't start node on the given endpoint (for example, due permissions problem), program won't crash and all other nodes will be started.