Open stauntonjr opened 1 year ago
I deleted pip.conf and am able to use pip everywhere again. I don't know what was in my pip.conf prior to that (I believe I was using it to register a private python package repository for a previous client), but so far I haven't run into any issues.
:fire: :bomb: This is a dangerous bug imho. My pip.conf figs are all gone. :bomb: :fire:
I am unable to figure out what code is doing it, I dont see anything in this repo that would do it.
It simply overwrites all your pip.conf with its idea of "good" settings. In my case this broke all my servers.
.... Meaning all of the carefully crafted configs that I have are all gone and I need to re-craft the ones that were not automated.
How high does NVIDIA stock price have to go before someone addresses this issue?
The adjustments in my pip.conf were made to work on highly sensitive government projects. Luckily the projects were already over when NVIDIA broke my system. But that doesn't mean it isn't a serious issue that NVIDIA needs to fix immediately.
I have been looking and I cant find it anywhere here. I also tried to reproduce the issue and cant ( I reinstalled a server from scratch) . But it had just happened to me, so it must have been one of the commands that I ran in the last couple hours. /me still trying to reproduce.
I also can't find where the package info is hosted to raise an issue (@DEKHTIARJonathan is listed as the pypi contact - is this in a private repo?)
One addition to this issue - it's unfortunate that nvidia-pyindex
automatically disables the (global) pip cache. I work in a - literally - remote environment with slow satellite internet and it took a while to figure out why pip
would re-download packages every single time.
This the command that sets up the new config (and overwrites the old one if it exists):
def create_pip_config_file(filepath, config_filedict):
try:
print("Creating dir:", os.path.dirname(filepath))
os.makedirs(os.path.dirname(filepath))
except FileExistsError:
pass
content = generate_configfile_header(config_filedict)
with open(filepath, 'w') as _file:
content += """[global]
no-cache-dir = true
index-url = https://pypi.org/simple
extra-index-url =
{extra_index_urls}
trusted-host =
{trusted_hosts}
"""
extra_index_urls = list()
trusted_hosts = list()
for nvidia_pypi_server in NVIDIA_PyPI_SERVERS:
extra_index_urls.append(_format_extra_index_url(nvidia_pypi_server))
trusted_hosts.append(_format_trusted_host(nvidia_pypi_server))
extra_index_urls[-1] = extra_index_urls[-1].rstrip('\n')
trusted_hosts[-1] = trusted_hosts[-1].rstrip('\n')
_file.write(
content.format(
extra_index_urls="".join(extra_index_urls),
trusted_hosts="".join(trusted_hosts),
)
)
which tracks with the original comment here. I think most people would be better/safer off just adding the urls to their pip config rather than doing this via a package.
System information
OS Platform and Distribution: Windows Version: Windows 11 Pro 22H2 22621.1992; WSL Version: 5.15.90.1-microsoft-standard-WSL2; WSL Linux Version: Ubuntu 22.04.1 LTS;
TensorFlow installed from (source or binary): None
TensorFlow version: None
Python version: Python 3.10.12 in Anaconda environment
Installed nvidia-pyindex using: pip inside conda env
Bazel version (if compiling from source): N/A
GCC/Compiler version (if compiling from source): N/A
CUDA/cuDNN version: Cuda compiler driver: cuda_11.5.r11.5/compiler.30672275_0 NVIDIA-SMI: 525.125.06
NVIDIA Driver Version: 536.67 "CUDA" (nvidia-smi): 12.2 cuda-profiler-api: 11.8.86 cuda-python: 11.8.2 cuda-version: 11.8 cudatoolkit: 11.8.0 cuDNN: N/A
GPU model and memory: RTX 3050 Laptop
Describe the problem
pip.conf was altered by nvidia-pyindex and now no pip commands work in any environment.
Provide the exact sequence of commands / steps that you executed before running into the problem
which makes changes to my pip.conf:
And now pip commands do not work:
This affects all other environments based on my path. I tried unsuccessfully to revert per https://github.com/NVIDIA/tensorflow/issues/19#issuecomment-816253523:
It seems it didn't know to look where my actual pip.conf file is, unfortunately. Is this normal? What am I missing? Is it related to WSL? How can I fix my pip? I think I now need to rewrite my pip.conf but I don't know what looked like before it was altered.
Any other info / logs N/A