Jobs dictionary does not list allocated cpus per node

[mrvn]

pyslurm version: 25.5.0

When I start a job with 3 cpus and run pyslurm.Jobs().load().to_dict() I get (for the job):

 'allocated_nodes': 'beo-203,node-4',
 'cpus': 3,

The question now is: Is that 2 cpus on beo-203 and 1 on node-4 or the reverse? The old jobs dictionary used to tell the cpu count for each node and that is sometimes important information.

[tazend]

Hi,

indeed, the "cpus" value here is job-specific, and is the total number of CPUs of the Job across all nodes.
You can however have a look at the get_resource_layout_per_node method on the Job instance.

It gives you back a view how Resources are distributed over each Node. There is the cpu_ids field, which should be a comma-separated string of CPU-IDs.
Since its a string, you'll need to do the counting yourselves by splitting on ,.

The whole return format of the method could be improved upon in general, though.

But essentially, the result you get back here has the same information like when executing scontrol show job <ID> -d for the per-node-view.

[mrvn]

Thanks, I will look into that call. Then I have to decide if that information is something that should still be included in the monitoring metrics. Can't imagine this will perform well if I query 1000 jobs every 5 seconds for their layout. Definetly will need some caching so the layout is only queried once when a job starts.