Command Cheatsheet

# Slurm Command Cheatsheet

## Job Submission

| Command | Description |
|---------|-------------|
| `sbatch script.sh` | Submit batch job |
| `sbatch --wrap="command"` | Submit one-liner |
| `sbatch --array=1-100%50 script.sh` | Submit job array (max 50 concurrent) |
| `sbatch --dependency=afterok:123 script.sh` | Submit with dependency |
| `sbatch --hold script.sh` | Submit in held state |

---

## Job Monitoring

| Command | Description |
|---------|-------------|
| `squeue --me` | My active jobs |
| `squeue` | All active jobs |
| `squeue -p gpu` | Jobs in GPU partition |
| `squeue --me --start` | Estimated start times |
| `squeue --me -o "%.8i %.9P %.20j %.2t %.10M %R"` | Custom format |
| `scontrol show job 123` | Full job details |
| `sacct -j 123` | Completed job info |
| `sacct -j 123 --format=JobID,Elapsed,MaxRSS,State,ExitCode` | Resource usage |
| `sacct --starttime=now-7days --state=FAILED` | Recent failures |

---

## Job Management

| Command | Description |
|---------|-------------|
| `scancel 123` | Cancel job |
| `scancel --me` | Cancel all my jobs |
| `scancel --me --state=PENDING` | Cancel only pending jobs |
| `scancel 123_[1-10]` | Cancel array tasks 1-10 |
| `scontrol hold 123` | Hold pending job |
| `scontrol release 123` | Release held job |
| `scontrol update JobId=123 TimeLimit=24:00:00` | Change time limit |
| `scontrol requeue 123` | Requeue job |

---

## Cluster Information

| Command | Description |
|---------|-------------|
| `sinfo` | Partition/node summary |
| `sinfo -N -l` | Detailed node list |
| `sinfo -p gpu -o "%N %G %c %m"` | GPU partition details |
| `sinfo -t idle` | Only idle nodes |
| `scontrol show partition` | Partition configuration |
| `scontrol show node node001` | Node details |
| `scontrol ping` | Controller health check |

---

## Accounting & Priority

| Command | Description |
|---------|-------------|
| `sacct` | Today's jobs |
| `sreport cluster utilization` | Cluster utilization |
| `sreport user TopUsage --tres=cpu TopCount=10` | Top CPU users |
| `sshare -l` | Fairshare values |
| `sprio` | Priority breakdown for pending jobs |
| `sdiag` | Scheduler diagnostics |

---

## Interactive Jobs

| Command | Description |
|---------|-------------|
| `srun --pty bash` | Interactive shell |
| `srun --pty -p gpu --gres=gpu:1 bash` | Interactive GPU shell |
| `srun --pty --x11 bash` | Interactive with X11 |
| `salloc --time=4:00:00` | Reserve resources |
| `srun hostname` | Run command on compute node |

---

## Admin Commands

| Command | Description |
|---------|-------------|
| `scontrol reconfigure` | Reload slurm.conf |
| `scontrol update NodeName=X State=DRAIN Reason="text"` | Drain node |
| `scontrol update NodeName=X State=RESUME` | Resume node |
| `scontrol create reservation ...` | Create reservation |
| `scontrol setdebug debug` | Increase log verbosity |
| `scontrol setdebug info` | Reset log verbosity |
| `sacctmgr show account` | List accounts |
| `sacctmgr show association tree` | Show account tree |
| `sacctmgr show qos` | List QOS |
| `sacctmgr add user name account=acct` | Add user |

---

## Common sbatch Options

```bash
#SBATCH --job-name=NAME          # Job name
#SBATCH --output=FILE            # Stdout (%j=jobid, %x=name, %a=arrayid)
#SBATCH --error=FILE             # Stderr
#SBATCH --partition=PART         # Partition
#SBATCH --time=HH:MM:SS         # Walltime
#SBATCH --ntasks=N               # Number of tasks (MPI ranks)
#SBATCH --cpus-per-task=N        # CPUs per task (threads)
#SBATCH --nodes=N                # Number of nodes
#SBATCH --mem=NG                 # Memory per node
#SBATCH --mem-per-cpu=NG         # Memory per CPU
#SBATCH --gres=gpu:N             # GPUs
#SBATCH --array=RANGE%LIMIT      # Job array
#SBATCH --dependency=TYPE:ID     # Dependency
#SBATCH --account=ACCT           # Account
#SBATCH --exclusive              # Exclusive node
#SBATCH --mail-type=END,FAIL     # Email notifications
#SBATCH --mail-user=EMAIL        # Email address
#SBATCH --constraint=FEATURE     # Node feature
```

Command Cheatsheet

References¶