Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

[DOC] dask-cuda-worker with --ucx-net-devices="auto" errors

See original GitHub issue

After installing the latest rapidsai 0.14 conda nightly packages, following the ucx-py blog post’s dask-cuda-worker startup instructions:

dask-cuda-worker ucx://10.10.10.10:8786 --enable-tcp-over-ucx --enable-nvlink --enable-infiniband --ucx-net-devices="auto"

results in an error about the ucx-net-devices param:

Usage: dask-cuda-worker [OPTIONS] [SCHEDULER] [PRELOAD_ARGV]...
Error: no such option: --ucx-net-devices=auto

Issue Analytics

State:
Created 3 years ago
Comments:7 (5 by maintainers)

Top GitHub Comments

1reaction

pentschevcommented, Apr 21, 2020

Yes, I’m sorry, I thought we were talking about renaming to --ucx-net-devices in dask-cuda, which is something Ben and I discussed in the past. I have now fixed the blog, thanks @randerzander for bringing this up to our attention.

I’ll leave this issue open as we intend to rename that to --ucx-net-devices in the future anyway.

0reactions

pentschevcommented, May 19, 2022

This is not relevant anymore since adding support for UCX >= 1.11.1, when we removed --ucx-net-devices. Closing.