When failing on NVMLError exception, bug in handling

[qwertAsc]

When failing on this line in smi.py with exception
nvmlDeviceGetSupportedMemoryClocks(handle)

following line fails with this error - "TypeError: list indices must be integers or slices, not str"
except NVMLError as err:
supportedClocks['Error'] = nvidia_smi.__handleError(err)

because supportedClocks defines as list

[rjzamora]

Thanks for raising an issue @qwertAsc - Can you provide a full reproducer here? How are you getting the handle?

For example, here is how I would expect someone to use nvmlDeviceGetSupportedMemoryClocks:

In [1]: import pynvml

In [2]: pynvml.nvmlInit()

In [3]: handle = pynvml.nvmlDeviceGetHandleByIndex(0)

In [4]: pynvml.nvmlDeviceGetSupportedMemoryClocks(handle)
Out[4]: [7001, 6501, 5001, 810, 405]

[qwertAsc]

Thanks @rjzamora
I am just calling the following
from pynvml.smi import nvidia_smi
nvidia_smi.getInstance().DeviceQuery()
and receiving this error
supportedClocks['Error'] = nvidia_smi.__handleError(err)
TypeError: list indices must be integers or slices, not str

when runnning your example I get the following
pynvml.nvmlDeviceGetSupportedMemoryClocks(handle)
File "/home/.../env/lib/python3.6/site-packages/pynvml/nvml.py", line 1135, in nvmlDeviceGetSupportedMemoryClocks
raise NVMLError(ret)
pynvml.nvml.NVMLError_NotSupported: Not Supported

[rjzamora]

Thanks for the info! Are you passing a query string to DeviceQuery (e.g. nvidia_smi.getInstance().DeviceQuery('memory.free') ?)

Also, can you specify the version of CUDA you are using and wheter you happen to be using MIG support?

[qwertAsc]

No, i am not passing any query string
I am using CUDA version 11.0
Also I am less bothered that the clock query fails, and more that except doesn't catch the error
(Not sure regarding MIG)

[danielbraun89]

have the same issue while running nvidia_smi.getInstance().DeviceQuery()

supportedClocks['Error'] = nvidia_smi.__handleError(err)
TypeError: list indices must be integers or slices, not str