Hase provides record-replay debugging suitable for all-ways-on in-production recording. It leverages intel processor trace and core dumps that can be recorded with little overhead. On top of that data it performs symbolic execution to recover states prior to the crash.

System design figure

Workshop presentation on Klee Workshop 2018

1. Setup virtual environment with python3.6 or pypy6

$ python3 -m venv venv
$ . venv/bin/activate

or for pypy:

$ pypy3 -m venv venv
$ . venv/bin/activate

2. Get the Intel processor-trace decoder library

3. Install project into the virtual environment

Make sure you pip is version >= 18.1 and setuptools >= 38.3:

$ ./venv/bin/pip install "pip>=18.1" "setuptools>=38.3"

$ ./venv/bin/pip install -e .

Trouble Shooting: New version of make may break the installation of pyvex, consult this upstream commit to resolve the issue manually.

Additionally pyqt5 is required and cannot be installed via pip.

4. Install test dependencies

$ python3 -m pip install -e '.[test]'

5. Testing examples

The integration test needs root.

make -C tests
sudo nosetests tests/test_record.py

The other tests work without root: Note that the test traces are stored via git-lfs

nosetests tests/test_replay.py

$ sudo ./bin/hase record <some crash program> <args>

Example crash:

$ sudo ./bin/hase record ./tests/bin/loopy
$ ls -la /var/lib/hase
.rw-rw-rw- 244 root  9 May  3:22 coredump.log
.rw-r--r--   4 root  9 May  3:22 hase-record.pid
.rw-r--r-- 41M root  9 May  3:22 loopy-20180509T022227.tar.gz

No crash:

$ sudo ./bin/hase record -- ls -al

Benchmarks require Pandas, which cannot be installed via pip3. Use https://pandas.pydata.org/pandas-docs/stable/install.html instead or install it using your system package manager.

To avoid breaking mypy use the following steps:

1. Create a feature branch:

$ git checkout -b <branch-name> 
$ git push origin <branch-name>

2. Make a pull request by visiting https://github.com/hase-project/hase/pull/new/<branch-name> or use hub:

$ git pull-request