[wip] fix horovod installation

[akihironitta]

What does this PR do?

Follows up #12368 and closes #12314.

Does your PR introduce any breaking changes? If yes, please list them.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

[akihironitta]

Reinstalling it somehow fixes the error...🤔

result here

root@b43ef4616cac:/# horovodrun --check-build
Horovod v0.24.2:

Available Frameworks:
    [ ] TensorFlow
    [X] PyTorch
    [ ] MXNet

Available Controllers:
    [ ] MPI
    [ ] Gloo

Available Tensor Operations:
    [ ] NCCL
    [ ] DDL
    [ ] CCL
    [ ] MPI
    [ ] Gloo
root@b43ef4616cac:/# python -c "import horovod.torch"
Extension horovod.torch has not been built: /usr/local/lib/python3.7/dist-packages/horovod/torch/mpi_lib/_mpi_lib.cpython-37m-x86_64-linux-gnu.so not found
If this is not expected, reinstall Horovod with HOROVOD_WITH_PYTORCH=1 to debug the build error.
Warning! MPI libs are missing, but python applications are still available.
root@b43ef4616cac:/# pip uninstall -y horovod
...
root@b43ef4616cac:/# pip install --no-cache-dir horovod
...
root@b43ef4616cac:/# horovodrun --check-build
Horovod v0.24.2:

Available Frameworks:
    [ ] TensorFlow
    [X] PyTorch
    [ ] MXNet

Available Controllers:
    [ ] MPI
    [X] Gloo

Available Tensor Operations:
    [X] NCCL
    [ ] DDL
    [ ] CCL
    [ ] MPI
    [X] Gloo
root@b43ef4616cac:/# python -c "import horovod.torch"
root@b43ef4616cac:/#

[awaelchli]

@akihironitta at this point, anything that makes it work is good enough :))
Did you push the image? You may need to retrigger the GPU CI job to pick up the changes

[akihironitta]

Working on it!

[akihironitta]

The built image has been successfully pushed to the hub, and GPU CI doesn't have the issue #12314 anymore, but I'm seeing another issue...

...
tests/callbacks/test_pruning.py::test_pruning_callback_ddp_spawn FAILED  [ 12%]
...
tests/callbacks/test_quantization.py::test_quantization[True-True-average] /__w/_temp/6c6a26ed-fe5c-4e9d-83d9-a3a6a6b6dabb.sh: line 1:   343 Segmentation fault      (core dumped) python -m coverage run --source pytorch_lightning -m pytest pytorch_lightning tests --ignore tests/benchmarks -v --junitxml=/__w/1/a/test-results.xml --durations=50
##[error]Bash exited with code '139'.

https://dev.azure.com/PytorchLightning/pytorch-lightning/_build/results?buildId=61463&view=logs&j=3afc50db-e620-5b81-6016-870a6976ad29&t=8b07f9df-ad34-5ead-2e0d-a2d35ff7ad3a

[akihironitta]

UPDATE: Couldn't reproduce the above error with the exact same docker image in our cluster. Will try to debug it in the CI...

[awaelchli]

@akihironitta maybe we try to skip the test to see if it is this particular test itself causing the problem or something else.

[Borda]

seems I found the bug, it was my typo in the past missing check for correct build #12318 (comment)