The labs of Raft consensus algorithm based on MadSim.

Some codes are derived from MIT 6.824 and PingCAP Talent Plan: Raft Lab. Thanks for their excellent work!

• Deterministic simulation: Catch a rare bug and then reproduce it at any time you want.
• Discrete event simulation: No time wasted on sleep. The full test can be completed in a few seconds.
• Async: The code is written in a fully async-style.

Read the instructions from MIT 6.824: Lab2, Lab3, Lab4.

Complete the code and pass all tests!

cargo test

To run a part of the tests or a specific test:

cargo test 2a
cargo test initial_election_2a

If a test fails, you will see a seed in the output:

---- raft::tests::initial_election_2a stdout ----
thread 'raft::tests::initial_election_2a' panicked at 'expected one leader, got none', src/raft/tester.rs:91:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
MADSIM_TEST_SEED=1629626496

Run the test again with the seed, and you will get exactly the same output:

MADSIM_TEST_SEED=1629626496 cargo test initial_election_2a

Enable logs to help debugging:

export RUST_LOG=madraft::raft=info

Run the test multiple times to make sure you solution can stably pass the test:

MADSIM_TEST_NUM=100 cargo test --release

Sometimes you may find that the test is not deterministic :(

Although the testing framework itself (MadSim) provides determinism, the entire system is not deterministic if your code introduces randomness.

Here are some tips to avoid randomness:

• Use madsim::rand::rng instead of rand::thread_rng to generate random numbers.
• Use futures::select_biased instead of futures::select macro.
• Do not iterate through a HashMap.

To make sure your code is deterministic, run your test with the following environment variable:

MADSIM_TEST_CHECK_DETERMINISTIC=1 cargo test

Your test will be run at least twice with the same seed. If any non-determinism is detected, it will panic as soon as possible.

Happy coding and Good luck!

Apache License 2.0