sqtop

English | 中文

sqtop is an interactive terminal monitor for Slurm clusters. It gives ordinary HPC users a readable, continuously refreshed view of partitions, queues, users, nodes, and jobs without requiring slurmrestd, root access, or Slurm JSON output.

The project was previously published as s-top. The current executable, crate, and conda package name is sqtop.

Why sqtop

squeue, sinfo, and watch are reliable, but they are not ideal for answering scheduler questions quickly:

• Which partitions are saturated right now?
• How much of the queue belongs to me versus other users?
• Which users are consuming the most jobs and resources?
• Which jobs are running, pending, or stuck for a specific reason?
• What changed over the last few refresh cycles?

sqtop focuses on those questions and stays compatible with typical user-level Slurm environments.

Highlights

• Live partition overview with queue pressure, ownership split, and trend history
• Separate My Jobs, All Jobs, Users, Partition Detail, and Node Detail views
• Structured job detail and conservative scancel preview / result flows
• Text-based Slurm collection path designed for clusters where --json is unavailable
• Search, filtering, sorting, paging, and horizontal navigation for wide tables
• Mouse support for tabs, rows, modal actions, and sortable headers

Screenshots

View	Preview
Overview
My Jobs
All Jobs
Users
Partition Detail
Node Detail
Job Detail
Cancel Preview

Install

From crates.io

cargo install sqtop

From conda

conda install -c wubeizhongxinghua sqtop

If you want conda install sqtop to work without -c, add the channel once:

conda config --add channels wubeizhongxinghua
conda install sqtop

From source

cargo build --release
./target/release/sqtop

Quick start

Run the TUI:

sqtop

Collect one snapshot and print a summary:

sqtop --once

Use a faster refresh interval:

sqtop --interval 2

Dump raw and parsed collector output for troubleshooting:

sqtop --debug-dump

Typical use cases

• Check whether a partition is congested before submitting a job
• Compare your current queue footprint with other users on the cluster
• Inspect jobs on a specific partition or node
• Find pending jobs by reason, placement, or owner
• Review a selected job before canceling it

Views

Overview

The default page shows partition pressure, mine-versus-others usage, running-versus-pending counts, and rolling trends.

My Jobs

Focused view of the current user's active jobs, including resource footprint, placement, search, and cancel actions.

All Jobs

Cluster-wide active queue view with highlighting for the current user's jobs.

Users

Ranks active users by running jobs, pending jobs, total jobs, and resource footprint. The lower pane shows the selected user's active jobs.

Partition Detail

Adds partition-local trends, node-state distribution, node list, and jobs in the selected partition.

Node Detail

Shows jobs on a node with interactive user, state, where, and why filters.

Job Detail

Opens a structured detail modal for a selected job, grouped into identity, resources, scheduling, placement, and execution sections.

Cancel Preview and Result

Reviews eligible and blocked scancel targets before execution, then reports detailed per-job results afterward.

Keyboard and mouse

Keyboard

Key	Action	Scope
Tab / Shift-Tab	Switch top-level pages	Global
q	Go back from detail pages; quit from top-level pages	Global
j / k / Up / Down	Move selection or scroll	Lists and modals
Space / b	Page down / page up	Lists and modals
g / G	Jump to top / bottom	Lists and modals
Enter	Open the focused detail	Overview and lists
/	Start live search	Searchable views
s	Cycle sort key	Overview, Users, job lists
f	Cycle queue-state filter	Job lists
m	Toggle mine-only mode	Shared views
p	Pin or unpin the current partition	Overview and job views
[ / ]	Change selected node	Partition Detail
n	Open selected node	Partition Detail
u / w / y / c	Change or clear node filters	Node Detail
i	Open job detail	Job lists
x	Cancel the selected job	Job lists
X	Preview bulk cancel	Job lists
Left / Right	Move horizontal table viewport	Wide tables

Mouse

Interaction	Result
Click a tab	Switch page
Click a row	Select row
Double-click a row	Open detail
Click a sortable header	Sort by that column
Mouse wheel	Scroll the active list or modal
Click a modal action	Trigger that action
Click outside the job-detail modal	Close the modal

CLI options

Flag	Description
--interval <seconds>	Refresh interval. Default: 2.0
--user <name>	Override the identity used for Mine / Others
--all	Start on the All Jobs page
--no-all-jobs	Disable the All Jobs page
--theme <auto|dark|light>	Select the UI theme
--advanced-resources	Force advanced resource columns on
--no-advanced-resources	Hide advanced resource columns
--debug-dump	Print raw and parsed data, then exit
--once	Collect once, print a summary, then exit
--compact	Use a denser layout
--no-color	Disable color output

Data sources and compatibility

The live path is intentionally built around text-oriented Slurm commands:

• sinfo
• squeue
• scontrol show partition
• scontrol show node
• scontrol show job

sacct is used only for detail enrichment or optional historical views and is not required for the main TUI.

The collector is designed for user-level environments:

• no root privileges assumed
• no slurmrestd
• no requirement for squeue --json or sinfo --json
• graceful degradation when optional fields are unavailable

Project status

The repository is actively packaged and released through GitHub tags. For upcoming improvements, use the issue tracker and release history as the authoritative source instead of expecting a separate long-term roadmap document here.

Project layout

Path	Responsibility
src/collector/	Slurm command execution, timeout handling, cancellation, raw collection
src/model/	Parsers, normalized data structures, aggregation
src/app.rs	Application state, refresh orchestration, filtering, sorting, event handling
src/ui/	Rendering, themes, view composition, mouse hit testing
src/cli.rs	CLI parsing and current-user resolution
src/config.rs	Optional configuration support
recipe/	Conda recipe and build scripts
.github/workflows/	Release packaging and registry publishing

Limitations

• Availability of ReqTRES, AllocTRES, GRES, memory, and GPU fields depends on site configuration
• Pending jobs that are eligible for multiple partitions may appear in more than one partition-level pending aggregation
• Narrow terminals still require horizontal navigation on wide tables
• Trend rendering depends on terminal font support for Unicode symbols
• Conda packaging currently targets Linux x86_64
• The Linux conda package currently targets a glibc baseline of 2.17

Community and support

• Contributing
• Code of Conduct
• Support
• Security policy
• Citation
• Changelog

License

This project is distributed under the MIT license. See LICENSE.