Why do jobs in slurm freeze indefinitely when they are TensorFlow scripts?

Question

I am having this error when I use the slurm (http://slurm.schedmd.com/) workload manager. When I run some tensorflow python scripts, sometimes it results in an error (attached). It seems that it can't find cuda library installed but I am running scripts that do not require GPUs. Therefore, I find it very confusing why cuda would be an issue at all. Why is cuda installation an issue if I don't need it?

The only useful information I got from the slurm-job_id file was the following:

I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcublas.so locally
I tensorflow/stream_executor/dso_loader.cc:102] Couldn't open CUDA library libcudnn.so. LD_LIBRARY_PATH: /cm/shared/openmind/cuda/7.5/lib64:/cm/shared/openmind/cuda/7.5/lib
I tensorflow/stream_executor/cuda/cuda_dnn.cc:2092] Unable to load cuDNN DSO
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcufft.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcuda.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcurand.so locally
E tensorflow/stream_executor/cuda/cuda_driver.cc:491] failed call to cuInit: CUDA_ERROR_NO_DEVICE
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:153] retrieving CUDA diagnostic information for host: node047
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:160] hostname: node047
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:185] libcuda reported version is: Not found: was unable to find libcuda.so DSO loaded into this program
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:347] driver version file contents: """NVRM version: NVIDIA UNIX x86_64 Kernel Module  352.63  Sat Nov  7 21:25:42 PST 2015
GCC version:  gcc version 4.8.5 20150623 (Red Hat 4.8.5-4) (GCC)
"""
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:189] kernel reported version is: 352.63.0
I tensorflow/core/common_runtime/gpu/gpu_init.cc:81] No GPU devices available on machine.

I always thought that tensorflow would not require GPU. So I am assuming the last error saying there is not GPU is NOT causing the error (correct me if I am wrong).

I dont understand why I need the CUDA library. I am trying to run my jobs with GPU, why would I need the cuda library if my jobs are CPU jobs?

---

I tried logging into the node directly and starting tensorflow but I got no apparent error:

I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcublas.so locally
I tensorflow/stream_executor/dso_loader.cc:102] Couldn't open CUDA library libcudnn.so. LD_LIBRARY_PATH: /cm/shared/openmind/cuda/7.5/lib64:/cm/shared/openmind/cuda/7.5/lib
I tensorflow/stream_executor/cuda/cuda_dnn.cc:2092] Unable to load cuDNN DSO
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcufft.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcuda.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcurand.so locally

though I expected the error:

I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcublas.so locally
I tensorflow/stream_executor/dso_loader.cc:102] Couldn't open CUDA library libcudnn.so. LD_LIBRARY_PATH: /cm/shared/openmind/cuda/7.5/lib64:/cm/shared/openmind/cuda/7.5/lib
I tensorflow/stream_executor/cuda/cuda_dnn.cc:2092] Unable to load cuDNN DSO
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcufft.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcuda.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcurand.so locally
E tensorflow/stream_executor/cuda/cuda_driver.cc:491] failed call to cuInit: CUDA_ERROR_NO_DEVICE
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:153] retrieving CUDA diagnostic information for host: node047
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:160] hostname: node047
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:185] libcuda reported version is: Not found: was unable to find libcuda.so DSO loaded into this program
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:347] driver version file contents: """NVRM version: NVIDIA UNIX x86_64 Kernel Module  352.63  Sat Nov  7 21:25:42 PST 2015
GCC version:  gcc version 4.8.5 20150623 (Red Hat 4.8.5-4) (GCC)
"""
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:189] kernel reported version is: 352.63.0
I tensorflow/core/common_runtime/gpu/gpu_init.cc:81] No GPU devices available on machine.

---

I also made an official git issue in the tensorflow library:

https://github.com/tensorflow/tensorflow/issues/3632

Answer 1

There is some bug in tensorflow operating with slurm submitting via a batch job.

Currently I get around it by running srun on slurm.

It also appears in your case that you installed the GPU version of tensorflow and are running it on a machine that does not have a GPU. Which is causing another error in your case.