slurm-dashboard 0.1.0

A lightweight web dashboard for monitoring Slurm HPC jobs

Slurm Dashboard

A lightweight, real-time web dashboard for monitoring Slurm HPC cluster jobs. View logs, track resource usage, analyze job patterns, and manage your workloads from a clean, modern interface.

Features

• Real-time job monitoring - Live log streaming with auto-refresh
• Multiple views - Table, Timeline, Pipeline DAG, Heatmap, and Insights
• Job management - Submit, cancel, and resubmit jobs with templates
• Smart log analysis - Automatic error detection and highlighting
• Resource tracking - CPU/GPU hours, memory efficiency, partition health
• Analytics dashboard - Job patterns, peak hours, failure analysis
• Batch operations - Select multiple jobs for cancel/resubmit/export
• Dark mode - Easy on the eyes during those late-night debugging sessions

Screenshots

Job List & Logs	Insights Dashboard
Heatmap View

Installation

pip install -e .

Requirements

• Python 3.8+
• Access to Slurm commands (squeue, sacct, scancel, sbatch)
• A web browser

Quick Start

# Basic usage - uses current user and default log location
slurm-dashboard

# Specify log directory
slurm-dashboard --log-root /path/to/your/logs

# Full options
slurm-dashboard --log-root ~/slurm-logs --user myusername --port 8080

Then open http://localhost:5000 in your browser.

Configuration

Command Line Options

Option	Default	Description
--host	127.0.0.1	Host to bind to
--port	5000	Port to bind to
--log-root	~/slurm-logs	Root directory for Slurm log files
--log-pattern	{name}/job.{stream}.{id}	Log file path pattern
--user	Current user	Slurm username to filter jobs
--refresh-cache	20	Cache refresh interval in seconds

Log File Patterns

The dashboard needs to know where your job logs are stored and how they're named. Use --log-pattern to match your cluster's configuration.

Pattern variables:

• {name} - Job name (from --job-name or script name)
• {id} - Slurm job ID
• {stream} - out for stdout, err for stderr

Common Cluster Configurations

Default Slurm (flat directory):

# Logs like: /home/user/slurm-12345.out, slurm-12345.err
slurm-dashboard --log-root ~ --log-pattern "slurm-{id}.{stream}"

Subdirectory per job name (default):

# Logs like: ~/slurm-logs/train_model/job.out.12345
slurm-dashboard --log-root ~/slurm-logs --log-pattern "{name}/job.{stream}.{id}"

Nested by job ID:

# Logs like: ~/logs/experiment/12345/stdout, ~/logs/experiment/12345/stderr
slurm-dashboard --log-root ~/logs --log-pattern "{name}/{id}/std{stream}"

Project-based structure:

# Logs like: /scratch/myproject/logs/job_12345.out
slurm-dashboard --log-root /scratch/myproject/logs --log-pattern "job_{id}.{stream}"

Date-based directories (use wildcards in log-root):

# For logs like: ~/logs/2024-01/train_12345.out
# Run from within the date directory or use a symlink
slurm-dashboard --log-root ~/logs/2024-01 --log-pattern "{name}_{id}.{stream}"

Setting Up Your Slurm Scripts

To work with the dashboard, configure your Slurm scripts to write logs to a consistent location:

#!/bin/bash
#SBATCH --job-name=train_model
#SBATCH --output=/home/%u/slurm-logs/%x/job.out.%j
#SBATCH --error=/home/%u/slurm-logs/%x/job.err.%j

# Your job commands here
python train.py

Slurm filename variables:

• %u - Username
• %x - Job name
• %j - Job ID
• %A - Array job ID
• %a - Array task ID

Array Jobs

For array jobs, you might want to include the array task ID:

#SBATCH --output=/home/%u/slurm-logs/%x/job.out.%A_%a
#SBATCH --error=/home/%u/slurm-logs/%x/job.err.%A_%a

Then use:

slurm-dashboard --log-pattern "{name}/job.{stream}.{id}"
# The dashboard handles array job IDs like 12345_0, 12345_1 automatically

Running on a Cluster

SSH Tunnel (Recommended)

If you're running the dashboard on a login node:

# On the cluster
slurm-dashboard --port 5000

# On your local machine (in another terminal)
ssh -L 5000:localhost:5000 [e-mail]

# Then open http://localhost:5000 in your browser

Binding to All Interfaces

To access from other machines on the same network (less secure):

slurm-dashboard --host 0.0.0.0 --port 5000

Note: Only do this on trusted networks. The dashboard can cancel and submit jobs.

Running in Background

# Using nohup
nohup slurm-dashboard --log-root ~/slurm-logs > dashboard.log 2>&1 &

# Using screen
screen -S dashboard
slurm-dashboard --log-root ~/slurm-logs
# Ctrl+A, D to detach

# Using tmux
tmux new -s dashboard
slurm-dashboard --log-root ~/slurm-logs
# Ctrl+B, D to detach

Views

Table View (Default)

• Running jobs with real-time status
• Recent jobs with expandable details
• Click any job to view its logs
• Quick filters for state, partition, GPU usage

Timeline View

• Gantt-style visualization of job history
• See job overlaps and gaps
• Filter by state or job name
• Zoom in/out for different time scales

Pipeline View

• DAG visualization of job dependencies
• Track pipeline progress
• See which jobs are blocking others

Heatmap View

• GitHub-style activity calendar
• Hourly submission patterns
• Success/failure rate visualization

Insights View

• Efficiency score (A-F grade)
• Resource usage metrics (CPU/GPU hours)
• Memory and time limit recommendations
• Failure pattern detection
• Peak hours analysis
• Top job types

Troubleshooting

"No recent logs found"

• Check that --log-root points to the correct directory
• Verify --log-pattern matches your log file naming
• Ensure log files exist and are readable

"sacct: command not found"

• The dashboard requires Slurm commands to be in your PATH
• Make sure you're running on a node with Slurm client tools installed

Jobs not showing up

• Check that --user matches the username used when submitting jobs
• Verify jobs are in sacct history: sacct -u $USER --starttime=now-7days

Log streaming not working

• Ensure the log files are accessible from the node running the dashboard
• Check file permissions on the log directory

Development

# Install in development mode
pip install -e ".[dev]"

# Run with debug mode
FLASK_DEBUG=1 slurm-dashboard

# Run tests
pytest

License

MIT