ccancel - Cancel Jobs

ccancel terminates running or pending jobs or individual steps within jobs.

You can cancel jobs by job ID, cancel individual steps using the jobid.stepid format, or use filter conditions to cancel multiple jobs at once.

Command Syntax

ccancel [job_id[.step_id][,job_id[.step_id]...]] [OPTIONS]

Command Line Options

Job Selection

• job_id[.step_id][,job_id[.step_id]...]: Job ID(s) or step ID(s) to cancel (comma-separated list).
• Format for jobs: <job_id> or <job_id>,<job_id>,<job_id>...
• Format for steps: <job_id>.<step_id> or <job_id>.<step_id>,<job_id>.<step_id>...
• Mixed format: <job_id>,<job_id>.<step_id>,<job_id>.<step_id>...

Filter Options

• -n, --name string: Cancel jobs with the specified job name
• -u, --user string: Cancel jobs submitted by the specified user
• -A, --account string: Cancel jobs under the specified account
• -p, --partition string: Cancel jobs in the specified partition
• -t, --state string: Cancel jobs in the specified state. Valid states: PENDING (P), RUNNING (R), ALL (case-insensitive)
• -w, --nodes strings: Cancel jobs running on the specified nodes (comma-separated list)

Output Options

• --json: Output in JSON format

Miscellaneous

• -C, --config string: Configuration file path (default: /etc/crane/config.yaml)
• -h, --help: Display help information
• -v, --version: Display version number

Filter Rules

Important

At least one condition must be provided: either job ID(s) or at least one filter option.

When using multiple filters, jobs matching all specified conditions will be cancelled (AND logic).

Usage Examples

Cancel by Job ID

Cancel a single job:

ccancel 30686

Result:

Cancel multiple jobs:

ccancel 30686,30687,30688

Cancel by Job Name

Cancel all jobs named "test1":

ccancel -n test1

Result:

Cancel by Partition

Cancel all jobs in GPU partition:

ccancel -p GPU

Result:

Cancel by Node

Cancel all jobs running on crane02:

ccancel -w crane02

Cancel by State

Cancel all pending jobs:

ccancel -t Pending

Cancel all running jobs:

ccancel -t Running

Cancel all jobs (pending and running):

ccancel -t All

Cancel by Account

Cancel all jobs under PKU account:

ccancel -A PKU

Cancel by User

Cancel all jobs submitted by user ROOT:

ccancel -u ROOT

Combined Filters

Cancel all pending jobs in CPU partition:

ccancel -t Pending -p CPU

Cancel all jobs by user alice in GPU partition:

ccancel -u alice -p GPU

Cancel running jobs on specific nodes:

ccancel -t Running -w crane01,crane02

JSON Output

Get cancellation results in JSON format:

ccancel 30686 --json

Cancel with filters and JSON output:

ccancel -p GPU -t Pending --json

Examples Overview

Canceling Job Steps

CraneSched supports canceling individual steps within a job using the step ID format jobid.stepid. This allows you to terminate specific steps while keeping the parent job and other steps running.

Step Cancellation Syntax

Cancel a specific step:

ccancel 123.1      # Cancel step 1 of job 123

Cancel multiple steps:

ccancel 123.1,123.2,456.3

Cancel entire job (all steps):

ccancel 123        # Cancels job 123 and all its steps

Step Cancellation Behavior

When canceling steps:

• Single Step Cancellation: Canceling a specific step (jobid.stepid) only affects that step
• The parent job continues running
• Other steps in the same job are not affected

• Resources allocated to the step are released back to the parent job

• Full Job Cancellation: Canceling a job without specifying a step ID (jobid) cancels:

• All steps within the job
• The parent job itself
• All allocated resources are released

Step Cancellation Examples

Cancel only step 2 of job 100:

ccancel 100.2

Cancel steps 1 and 2 of job 100 (step 3 continues if exists):

ccancel 100.1,100.2

Mixed cancellation - entire job 100 and step 3 of job 200:

ccancel 100,200.3

Query steps before canceling:

# View all steps for a job
cqueue --step -j 123

# Cancel specific steps based on status
ccancel 123.2,123.4

Step Cancellation Permissions

• Regular Users: Can only cancel steps belonging to their own jobs
• Coordinators: Can cancel steps within jobs under their account
• Operators/Admins: Can cancel any step in the system

Behavior After Cancellation

After a job is cancelled:

1. Process Termination: If there are no other jobs from the user on the allocated nodes, the job scheduling system will terminate all user processes on those nodes

2. SSH Access Revocation: SSH access to the allocated nodes will be revoked for the user

3. Resource Release: All allocated resources (CPUs, memory, GPUs) are immediately released and become available for other jobs

4. Job State Update: The job state changes to CANCELLED in the job history

Permission Requirements

• Regular Users: Can only cancel their own jobs
• Coordinators: Can cancel jobs within their account
• Operators/Admins: Can cancel any job in the system

Important Notes

1. Immediate Effect: Job cancellation takes effect immediately. Running jobs are terminated without grace period by default

2. Multiple Jobs: You can cancel multiple jobs at once using comma-separated job IDs or filter conditions

3. No Confirmation: There is no confirmation prompt. Jobs are cancelled immediately upon command execution

4. State Filtering: Use -t to target specific job states to avoid accidentally cancelling jobs in unintended states

5. Job/Step ID Format: IDs must follow these formats with no spaces:

6. Jobs: <job_id> or <job_id>,<job_id>,<job_id>...
7. Steps: <job_id>.<step_id> or <job_id>.<step_id>,<job_id>.<step_id>...
8. Mixed: <job_id>,<job_id>.<step_id>...

Error Handling

Common errors:

• Invalid Job ID: Returns error if job ID doesn't exist or you don't have permission to cancel it
• No Matching Jobs: If filter conditions match no jobs, returns success with zero jobs cancelled
• Invalid State: State must be one of: PENDING, RUNNING, ALL (case-insensitive)

See Also

• cbatch - Submit batch jobs
• crun - Run interactive tasks
• calloc - Allocate interactive resources
• cqueue - View job queue
• cacct - View job accounting information
• ccontrol - Control jobs and system resources