(c) 2021-2024 Espresso Systems.
HotShot
was developed by Espresso Systems.
HotShot is a BFT consensus protocol based off of HotStuff, with the addition of proof-of-stake and VRF committee elections.
DISCLAIMER: This software is provided "as is" and its security has not been externally audited. Use at your own risk.
Please see the rustdoc for API documentation, and the examples directory for usage.
nix develop
brew install cmake protobuf
apt-get install cmake protobuf
choco install cmake protoc
scoop bucket add extras
scoop install protobuf cmake
Once dependencies have been installed, to build everything:
just build
HotShot supports static linking for its examples:
# Nix-shell is optional but recommended
nix develop .#staticShell
just build
To test:
RUST_LOG=$ERROR_LOG_LEVEL RUST_LOG_FORMAT=$ERROR_LOG_FORMAT just test
RUST_LOG=$ERROR_LOG_LEVEL
: The basic levels of logging includewarn
,error
,info
.RUST_LOG_FORMAT=$ERROR_LOG_FORMAT
: The types of logging includefull
,json
, andcompact
.- Internally, the inclusion of the
--nocapture
flag indicates whether or not to output logs. - Internally, we run at
--test-threads=1
because the tests spawn up a lot of file handles, and unix based systems consistently run out of handles.
To stress test, run the ignored tests prefixed with test_stress
:
RUST_LOG=$ERROR_LOG_LEVEL RUST_LOG_FORMAT=$ERROR_LOG_FORMAT just run_test test_stress
To double check for UB:
nix develop .#correctnessShell
just careful
To test as if running on CI, one must limit the number of cores and ram to match github runners (2 core, 7 gig ram). To limit the ram, spin up a virtual machine or container with 7 gigs ram. To limit the core count when running tests:
ASYNC_STD_THREAD_COUNT=1 RUST_LOG=$ERROR_LOG_LEVEL RUST_LOG_FORMAT=$ERROR_LOG_FORMAT just tokio test
To use tokio-console, drop into the console shell:
nix develop .#consoleShell
Then, run an example.
On a separate terminal, also drop into the console shell and start tokio-console:
nix develop .#consoleShell -c tokio-console
This second window should now display task usage.
To view distributed logs with just the centralized server and one client, first edit the centralized_server/orchestrator
file to include have a threshold and num_nodes of 1.
Then open 3 terminals.
# Terminal 1
# Start the jaeger instance to view spans
docker run -d -p6831:6831/udp -p6832:6832/udp -p16686:16686 -p14268:14268 jaegertracing/all-in-one:latest
# Terminal 2
# Start the CDN
# Terminal 3
# Start the client
To generate usage stats:
- build the test suite
- find the executable containing the test of interest
- run profiling tools
The executable cargo
uses is shown in the output of cargo test
.
For example, to profile test_stress_dht_many_round
:
# bring profiling tooling like flamegraph and heaptrack into scope
nix develop .#perfShell
# show the executable we need run
# and build all test executables (required for subsequent steps)
cargo test --verbose --release --lib --bins --tests --benches --workspace -- --test-threads=1
# the output cargo test contains the tests path:
# Running `/home/jrestivo/work/crosscross/target/release/deps/counter-880b1ff53ee21dea test_stress --test-threads=1 --ignored`
# running 7 tests
# test test_stress_dht_many_rounds ... ok
# ...
# a more detailed alternative to flamegraph
# NOTE: only works on linux
heaptrack $(fd -I "counter*" -t x | rg release) --ignored -- test_stress_dht_many_round --nocapture
# palette provides memory statistics, omission will provide cpu cycle stats as colors
# NOTE: must be run as root on macos
flamegraph --palette=mem $(fd -I "counter*" -t x | rg release) --ignored -- test_stress_dht_one_round
# code coveragte statistics
cargo-llvm-cov llvm-cov --test=test_stress_dht_many_round --workspace --all-targets --release --html --output-path lcov.html
This will output:
heaptrack.counter-$HASH
which is viewable by heaptrack. This provides a plethora of useful statistics about memory and cpu cycles.flamegraph.svg
which is a (moderately) less detailed version of heaptrack.lcov.html
generates a summary of code coverage.
A debugging config file is provided for vscode and vscodium in .vscode/launch.json
. This is intended to be used with vadimcn/vscode-lldb but may work with other rust debuggers as well.
To bring lldb
into scope with nix, run nix develop .#debugShell
.
For espresso developers we have written up a description of our workflow here.
Choose an async runtime to use before launching a text editor. This may be done by setting the environment RUSTFLAGS. For example:
nvim # launch text editor of choice. We choose neovim in this example
unset RUSTFLAGS # Unset rustflags so we may continue to use the justfile. The justfile sets these particular config options
We support the CodeLLDB Debugger.
Install dap
and rust-tools
. Install the CodeLLDB debugger listed above.
Follow the instructions here to configure the adapter. To add our project-local configurations, run:
lua require('dap.ext.vscode').load_launchjs(nil, { ["codelldb"] = {"rust"} })
Finally, place a breakpoint and run :DapContinue
to begin debugging.
NOTE: Do NOT configure dap at all with rust-tools. Do it manually.
Install the extension and load the launch.json
file. Then run the desired test target.