Impossible to test perfzero without docker

[shailensobhee]

Execute: python3 perfzero/lib/benchmark.py --git_repos="https://github.com/tensorflow/models.git;benchmark" --python_path=models --data_downloads="gs://tf-perf-imagenet-uswest1/tensorflow/cifar10_data/cifar-10-batches-bin" --benchmark_methods=official.r1.resnet.estimator_benchmark.Resnet50EstimatorBenchmarkReal.benchmark_graph_1_gpu

Error:

[...]
benchmarks/perfzero/lib/perfzero/utils.py", line 288, in run_commands
    cmd, exit_code, stdout))
Exception: "['gsutil', '-m', 'cp', '-r', '-n', 'gs://tf-performance/auth_tokens/benchmark_upload_gce.json', '<path>/benchmarks/perfzero/workspace']" failed with code:1 and stdout:
ServiceException: 401 Anonymous caller does not have storage.objects.get access to the Google Cloud Storage object.
CommandException: 1 file/object could not be transferred.

I did install the following before.

pip install gsutil
pip install cloud-tpu-client

Can you please run the CI tests?

[elgalu]

@shailensobhee did you solve it?

[vladfi1]

I got this error as well. You can get a bit further by authenticating with gsutil via gsutil config, but then you'll just get:

AccessDeniedException: 403 <email> does not have storage.objects.list access to the Google Cloud Storage bucket.
CommandException: 1 file/object could not be transferred.

The issue seems to be that the GCS bucket that hosts the cifar-10 dataset isn't publicly readable.