Replies: 1 comment
-
|
As you know, |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
One of our users was doing this within a Slurm job:
plan(cluster, workers=20). If the job ended up on a single node, that was fine. But when it ended up on multiple nodes, then one gets a warning or error because of oversubscribing the cores on the first node in the allocation, sincefuturetries to start 20 workers on that node. Of course this works perfectly ifworkersis omitted, which is the recommended workflow, but it has the potential to confuse a user not thinking about the fact that the number of nodes and number of cores per node may vary from job to job.I think it might be worth giving an explicit warning in such cases that passing a numeric value is not recommended, as doing so will attempt to start that many workers only on the first node. Currently the warning (or error depending on degree of oversubscription) talks about the number of workers relative to the number of cores, but without explicitly noting the context of multiple nodes.
Beta Was this translation helpful? Give feedback.
All reactions