Implementation of a decentralized network of multiple Nodes. One of the nodes has the role of a master and the nodes will elect the current master automatically.
Installation
You could use a virtual environment to avoid dependency problems:
python3 -m venv env
pip install --upgrade pip
source env/bin/activate
Install requirements:
pip install -r requirements
Manual
You can start one Node directly by using the node.py script.
python node.py
The first four arguments are address, port, broadcast address and broadcast port.
As a default the combination
'0.0.0.0' 8080 '255.255.255.255' 5595
will be used. The broadcast is used to make a handshake with existing nodes.
If a port is already in use a free port will be selected.
If you want, that instead of birthtime/time of initalization of one node the lowest port will be used for voting in quorum, you can add the --use_port_instead_of_life_time flag.
For testing issues, you can pass with
-m=i_am_master.py
and for slave
-s=i_am_slave.py
a path to python scripts which are executed if the node change his role.
With -d you get a debug log with more information about e.g. who received
a vote or number of connected notes.
A set A as new master
A add B to connected by handshake. Now are 1 nodes connected
A set B as new master
For better readability of logs you can give the node a nickname by set
-n=Your_Node_Name
This name will also be used for re-entering if a node dispatched and come back later.
If you want to dispatch your node enter 'q` in the command line. If you exit
by STRG + C then the node will be killed and marked as lost from existing Nodes
Example command:
python node.py 0.0.0.0 6170 -d -m=i_am_master.py -s=i_am_slave.py -n=B -d
Quorum algorithm
Each node calculates based on time or port metric his wanted master
and send constantly information about his vote to other nodes.
If a node detects changes (New Node or Lost/Dispatching Slave/Master)
it will change his voted master and change his transmitted information per
TCP socket. Also each Node constantly listen to incoming messages from other
nodes and evaluate most voted master.
So you can imagine the network as a constant election.
Base on own and recieving votes the new master is constantly
calculated by absolute majoritiy. If their is no absolute the largest part
will surivive. Others will dispatch. If more nodes lost than connected the node
will dispatch, because the network cannot build a majority any longer.
Tests
In test/ are unittest classes and integration_tests.py which contains
bigger integrations test especially for determination of a master in different
scenarios. More information about each case are in documentation.
pytest -v test/
pytest -v test/integration_tests.py
If you have questions, issues or feedback:
till.johanndeiter (at) web.de