Xsinfo

Collect info about the current nodes and cores usage so that it can help one choose which and how much resource can be allocated for a job on SLURM.

Install

git clone https://github.com/FranckLejzerowicz/Xsinfo.git
cd Xsinfo
pip install -e .

*Note that python should be python3

Outputs

Two outputs:

Table

In a file located in ~/.xsinfo/YYYY-MM-DD.tsv containing the table returned by the following Slurm's sinfo command and expanded with cpu_load, memory_load (both in % of available CPUs and memory per node), as well as the number of allocated and idle CPUs per node:

sinfo --Node -h -O NodeList:10,Partition:10,StateLong:10,CPUsLoad:10,CPUsState:12,Sockets:4,Cores:4,Threads:4,Memory:12,FreeMem:12

Note: if Xsinfo is re-run the same day twice, it will not re-run this sinfo command. Instead, it will read the expanded ~/.xsinfo/YYYY-MM-DD.tsv file. If you need to re-run Xsinfo the same day if node usage changes that quick, please use the --refresh option (in ~/.xsinfo/ is only kept the latest table).

This table can be used by Xpbs - optionally - to allocate CPUs from idle nodes that have the right amount of memory available.

Stdout

Summaries are printed in stdout, including:

• "nodes per % of cpu load": nodes are binned per quartile of cpu load.
• "nodes per % of mem load": nodes are binned per quartile of memory load.

For the nodes binned in these two different ways are of shown:

• Their total number of CPUs (cpus)
• Their total available memory in GiB (mem(gb))
• Their average and standard deviation of available memory (av and ±)
• Their number (of nodes) (nodes)
• Their names (of nodes) (names)

Usage

Just run:

Xsinfo [OPTIONS]

• --torque: Use if your scheduler is Torque (and not Slurm), as the native qstats summary is a bit different and thus Xsinfo has to know that in order to properly parse out the node usage information [OPTION NOT YET WORKING!]
• --refresh: Force rewriting of the expanded node info collected today.
• --show: Will print the full cpu and memory info per nodes.

Options

Usage: Xsinfo [OPTIONS]

Options:
  --torque / --no-torque        Switch from Slurm to Torque  [default: no-
                                torque]
  --refresh, --no-refresh TEXT  Whether to update any sinfo snapshot file
                                written today in ~/.slurm.
  --version                     Show the version and exit.
  --help                        Show this message and exit.

Bug Reports

contact franck.lejzerowicz@gmail.com