Helps debug a non determinate test (or any flaky program) by running it until it exits with a non-zero exit code.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Helps debug a non determinate test (or any flaky program) by running it until it exits with a non-zero exit code.

deflake can be run on the command-line or imported as a module. See below for more details. At the command-line, run python --help for a list of options. Default maximum runs is 10. Default pool-size is 1.


$ pip install deflake

Example Running as Script

deflake will exit with 1 if any processes return non-zero.

# up to 25 times until failure
$ deflake "my_flaky_program arg1 arg2"
FAIL (run 4)
$ echo $?

Example as Python Class

In this case we use the Deflake class, trying to deflake ls. ls is pretty stable, so we don't expect a non-zero exit status. We should see PASS ten times (the default maximum runs) until deflake gives up. When running as a class, the run method runs the processes, and returns a list with the output from calling run.

>>> from deflake import Deflake
>>> d =Deflake("ls", quiet=True)
<deflake.Deflake object at 0x7fd1b40d76d0>
>>> results =


Use the pool-size or -p option to run your program in concurrent pools of processes. The default is no multiprocessing or a pool size of 1. Setting a pool-size to > 1 can make your deflaking faster, but it won't always work, depending on the program you are trying to deflake. If your program writes to a file, for example, then multiple processes may try to write to that file simultanously and render your deflaking, well, flaky. See below for a possible solution.

If the program you're deflaking writes to predetermined files (for example a log file), and you want to multiprocess using the pool-size option, deflake might report a failure due to multiple processes trying to write to the same file. In this case you can use the special #count# replacement token to change the name of the log file each process writes to. If you need to change the default #count# token to something else, use the --counter-token or -c option. Whatever replacement token you use, it will be replaced with the iterator used when looping through the processes. For example:

$ deflake --max-runs 5 --pool-size 4 'my_flaky_program --log-file log#count#.txt'

The previous run would generate::

  • log1.txt
  • log2.txt
  • log3.txt
  • log4.txt
  • log5.txt


To work on this package:

  1. Clone the repo.
  2. pip install -e .
  3. Test: python tests/
  4. Fix/add stuff.
  5. Test. $ pip install nose. Then, cd tests, and $ nosetests -vs tests/
  6. Increment version in
  7. Submit pull request
  8. Tag $ git tag -a x.x.x -m "Description"
  9. $ git push --tags

To submit to pypi, make sure your .pypirc is set up:

index-servers =


Also, increment the version in


python sdist upload -r pypi

Known Issues

  • Extra processes kicked off in Python3x