compiler_gym.make("loop_tool-v0") produces a ServiceError in Collab

[Game4Move78]

🐛 Bug

Copying the GettingStarted example, there is a ServiceErorr.

To Reproduce

Steps to reproduce the behavior:

Here is a Collab example. gym.make("loop_tool-v0") throws the error.

Expected behavior

No error

Environment

Please fill in this checklist:

• CompilerGym: 0.2.3
• How you installed CompilerGym (conda, pip, source): pip
• OS: Collab
• Python version: 3.7
• Build command you used (if compiling from source):
• GCC/clang version (if compiling from source):
• Bazel version (if compiling from source):
• Versions of any other relevant libraries:

You may use the PyTorch
environment collection script
to generate most of this information. You can get the script and run it with:

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py

Additional context

[ChrisCummins]

Thanks for reporting! I confirm I can repro the problem. This also applies to the gcc-v0 environment. I will look into this.

https://colab.research.google.com/drive/1QXO_CPkaguCqwg_SXDlczP022lEGWizD?usp=sharing

Cheers,
Chris

[ChrisCummins]

Okay so there's a couple of problems going on here. The first is that the default colab runtime has CUDA, but does not have a GPU. This causes cuda initialization to fail. We can see this by running the loop_tool service binary directly in colab:

!/usr/local/lib/python3.7/dist-packages/compiler_gym/envs/loop_tool/service/compiler_gym-loop_tool-service

terminate called after throwing an instance of 'std::runtime_error'
  what():  assertion: result == CUDA_SUCCESS failed @ /tmp/pip-req-build-m1xnd0tv/src/backends/cuda/cuda.cpp:838 
error: CULIB(cuInit)(0) failed with error CUDA_ERROR_NO_DEVICE

If we add a GPU to the colab runtime (Runtime >> Change runtime type >> Hardware acceleartor >> GPU), the service runs properly:

!/usr/local/lib/python3.7/dist-packages/compiler_gym/envs/loop_tool/service/compiler_gym-loop_tool-service -v=1 --alsologtostderr

I0421 21:08:30.337343 140431714187136 create_and_run_compiler_gym_service.py:118] Service /tmp/compiler_gym-service-tcz8r_6b listening on 40409, PID = 223

Thanks @bwasti for figuring this one out!

However, there is a second problem, which is that launching python services on colab appears to be broken. I haven't yet figured out the cause, but this appears to be the problem behind loop_tool/GCC environments failing. I'll post here once I have an update.

Cheers,
Chris