pytest-shard

Shards tests based on a hash of their test name enabling easy parallelism across machines, suitable for a wide variety of continuous integration services. Tests are split at the finest level of granularity, individual test cases, enabling parallelism even if all of your tests are in a single file (or even single parameterized test method).

Features

pytest-shard aims for simplicity. When installed, simply run:

$ pytest --shard-id=I --num-shards=N

where I is the index of this shard and N the total number of shards. For example, to split tests across two machines:

# On machine 1:
$ pytest --shard-id=0 --num-shards=2
# On machine 2:
$ pytest --shard-id=1 --num-shards=2

The intended use case is for continuous integration services that allow you to run jobs in parallel. For CircleCI, enable parallelism and then use:

pytest --shard-id=${CIRCLE_NODE_INDEX} --num-shards=${CIRCLE_NODE_TOTAL}

On Travis, you must define the environment variables explicitly, but can use a similar approach.

Alternatives

pytest-xdist allows you to parallelize tests across cores on a single machine, and can also schedule tests on a remote machine. I use pytest-shard to split tests across CI workers, and pytest-xdist to parallelize across CPU cores within each worker.

pytest-shard does not take into account the run time of tests, which can lead to suboptimal allocations. pytest-circleci-parallelized uses test run time, but can only split at the granularity of classes, and is specific to CircleCI.

Please open a PR if there are other promising alternatives that I have overlooked.