Support detecting compute nodes of E3SM machines
chengzhuzhang opened this issue · comments
There has been issues raised that when running on compute nodes, analysis tools won't auto detect machines: e.g. E3SM-Project/zppy#406
The suggested solution for now is to manually specify machine name, e.x
[default]
machine = chrysalis
or if running in E3SM-Unified, to use the environment variable, e.x.
machine = os.environ["E3SMU_MACHINE"]
It might be viable to detect machines based on compute node names if there is a pattern. Though it can be challenging based on comment from @xylar: E3SM-Project/zppy#511 (comment)
It seems that we can start to add support for certain machines that has patterns in their compute nodes names.
Perfect, thanks! I will give it a try...
@chengzhuzhang, we have a backup option already, where we supposedly create a file when mache gets installed and read the machine name from it:
https://github.com/E3SM-Project/mache/blob/main/mache/discover.py#L59-L67
I wonder why that's not working.
The answer is that I clearly forgot to implement this on conda-forge.
@chengzhuzhang, we have a backup option already, where we supposedly create a file when mache gets installed and read the machine name from it: https://github.com/E3SM-Project/mache/blob/main/mache/discover.py#L59-L67
Smart!
This should be fixed in different ways by conda-forge/mache-feedstock#54 and #135. Hopefully, we really clobber this one once and for all...
(Sorry for getting a bit annoyed earlier. It's been a long, long day...)
Thank you for getting to it so quickly. (and the day is indeed too long, looking at current Berlin time...)