Information sources

The TUI is a thin presentation layer; every piece of data on screen comes from a standard Slurm CLI command. Knowing the underlying command makes it easy to verify a value or diagnose a discrepancy directly from the shell.

TUI element	Underlying command (run over SSH)	Notes
Job table (active)	squeue --me -o "%i|%j|%T|%M|%Z|%b" --noheader	Fields: id, name, state, time, work dir, GRES.
Job table (history)	sacct -X --format=JobID,JobName,State,Elapsed,WorkDir --units=M -n	Active jobs override history if both contain the same ID.
Cluster running / pending totals	squeue --noheader -o "%T"	Counted by state. Re-fetched every 30 s.
Pending reason / priority / QOS / submit time	squeue --me -t PENDING --noheader -o "%i|%r|%Q|%V|%q"	Populated only for PENDING jobs.
Queue rank	squeue -t PENDING --noheader -o "%Q|%i" --sort=-Q	1-based position among all pending jobs cluster-wide.
Job detail header / timing / log paths	scontrol show job <jobid>	Source of StdOut, StdErr, Command, SubmitTime, StartTime, EndTime, Partition, NodeList, TimeLimit, ReqTRES.
Live memory usage	sstat --format=MaxRSS -j <jobid>.batch --noheader	Only fetched for RUNNING jobs.
Per-GPU utilization	srun --jobid=<jobid> --overlap nvidia-smi --query-gpu=index,name,utilization.gpu,memory.used,memory.total --format=csv,noheader,nounits	Only attempted when ReqTRES declares GPUs.
Cluster capacity bars	sinfo -h -N -o "%N|%R|%T|%C|%e|%m|%G|%E"	One row per (node, partition). Rows are deduplicated on node name.
Partition table	sinfo -h -o "%R|%D|%C|%G|%m|%a"	Per-state node counts are filled in from the node walk above.
Log streaming	tail -n 50 -f <log_path> over a paramiko channel	The log path comes from scontrol; falls back to the configured pattern.
Batch script viewer	scontrol write batch_script <jobid> -	Read-only.
Cancel a job	scancel <jobid>	Triggered by c, gated by a confirmation modal.

If you ever doubt a value in the UI, run the corresponding command in an SSH shell — that is exactly what the TUI did, parsed, and rendered.