[go: up one dir, main page]

Skip to content

kabouzeid/turm

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

68 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

turm

A text-based user interface (TUI) for the Slurm Workload Manager, which provides a convenient way to manage your cluster jobs.

turm demo

turm accepts the same options as squeue (see man squeue). Use turm --help to get a list of all available options.

Installation

turm is available on PyPI and crates.io:

# With pip.
pip install turm

# With pipx.
pipx install turm

# With uv.
uv tool install turm

# With cargo.
cargo install turm

# With wget. Make sure ~/.local/bin is in your $PATH.
wget https://github.com/kabouzeid/turm/releases/latest/download/turm-x86_64-unknown-linux-musl.tar.gz -O - | tar -xz -C ~/.local/bin/

The release page also contains precompiled binaries for Linux.

Shell Completion (optional)

Bash

In your .bashrc, add the following line:

eval "$(turm completion bash)"

Zsh

In your .zshrc, add the following line:

eval "$(turm completion zsh)"

Fish

In your config.fish or in a separate completions/turm.fish file, add the following line:

turm completion fish | source

How it works

turm obtains information about jobs by parsing the output of squeue. The reason for this is that squeue is available on all Slurm clusters, and running it periodically is not too expensive for the Slurm controller ( particularly when filtering by user). In contrast, Slurm's C API is unstable, and Slurm's REST API is not always available and can be costly for the Slurm controller. Another advantage is that we get free support for the exact same CLI flags as squeue, which users are already familiar with, for filtering and sorting the jobs.

Ressource usage

TL;DR: turmwatch -n2 squeue + tail -f slurm-log.out

Special care has been taken to ensure that turm is as lightweight as possible in terms of its impact on the Slurm controller and its file I/O operations. The job queue is updated every two seconds by running squeue. When there are many jobs in the queue, it is advisable to specify a single user to reduce the load on the Slurm controller (see squeue --user). turm updates the currently displayed log file on every inotify modify notification, and it only reads the newly appended lines after the initial read. However, since inotify notifications are not supported for remote file systems, such as NFS, turm also polls the file for newly appended bytes every two seconds.