Raft Testing Framework

This repository contains a framework for simulating communication between raft processes and a collection of test harnesses. Please refer to the MP description for more details.

Getting started

You can run the testing framework using python3.10 framework.py. (Note: the framework is tested using python3.9 and python3.10; it may not work on Python versions older than 3.7). The arguments to the tester are the number of processes and the command to run each individual process, e.g.:

python3.10 framework.py 5 ./raft

or

python3.10 framework.py 2 python3.10 pinger.py

It will create n processes using the command you specify: the pid of the process (0, ..., n-1) and the total number of processes. So the first command will run:

./raft 0 5
./raft 1 5
./raft 2 5
./raft 3 5
./raft 4 5

as separate processes and connect them up to the communication framework. Likewise, the second command will run:

python3.10 pinger.py 0 2
python3.10 pinger.py 1 2

The pinger.py script is a very simple user of the framework that simply sends a PING message to the next PID in a ring.

Running tests

The tests take a single argument, n, which is the number of Raft processes to use. E.g.: python3.10 raft_partiton_test.py 5. Currently enabled tests:

• raft_election_test: tests simple election of a Raft leader: waits for the processes to elect a leader and for each other process to become a follower by receiving AppendEntries
• raft_election_failure_test: after a leader is elected, stops the leader and waits for the remaining processes to elect a new leader
• raft_partition_test: waits for a leader to be elected, then partitions that leader off from the rest of the group. Once the remaining group elects a new leader, it repairs the partition and waits for the previous leader to catch up
• raft_simple_log_test: after a leader is elected, requests the leader to log a message and waits for it to be committed on all servers
• raft_log5_test: logs 5 messages in a sequence
• raft_log_leader_failure_test: logs 5 messages, fails a leader, waits for re-election and logs 5 more
• raft_log_follower_failure_test: logs 5 messages, fails a minority of followers and logs 5 more, then fails another follower and makes sure that commits don't happen without a majority
• raft_log_partition_test: logs 5 messages, partitions off leader, logs 5 more with new leader, repairs partition and waits for old leader to catch up