Limiting maximum running jobs in a WorkGraph

Introduction

Imagine a workflow that needs to launch dozens of calculations. While this parallelism is powerful on a cluster, running them all at once on a local machine can overwhelm your system.

aiida-workgraph provides a simple way to control the number of concurrent jobs for a single workflow. This is controlled by a single attribute:

• WorkGraph.max_number_jobs: Sets the maximum number of child processes that this specific WorkGraph can run simultaneously.

from aiida import load_profile
from aiida_workgraph import task

load_profile()

Profile<uuid='2c6af82ef2b241408a4c27c7c32d5490' name='presto'>

Example

Let’s demonstrate this by wrapping an AiiDA CalcJob (ArithmeticAddCalculation) as a task and submitting several instances. We’ll set a concurrency cap to prevent more than two jobs from running at the same time.

from aiida.orm import Int, load_code
from aiida.calculations.arithmetic.add import ArithmeticAddCalculation

# First, wrap the CalcJob class into a reusable WorkGraph task
AddTask = task(ArithmeticAddCalculation)

code = load_code('add@localhost')


# Define the WorkGraph that launches multiple independent jobs
@task.graph
def many_adds(n: int, code):
    """Launches 'n' independent addition jobs."""
    for k in range(n):
        AddTask(
            x=Int(k),
            y=Int(1),
            code=code,
        )


# Build the graph for 5 jobs
wg = many_adds.build(n=5, code=code)

# Set the maximum number of concurrent jobs to 2
wg.max_number_jobs = 2

# Run the workflow
wg.run()

06/29/2026 09:09:51 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|continue_workgraph]: tasks ready to run: ArithmeticAddCalculation,ArithmeticAddCalculation1,ArithmeticAddCalculation2,ArithmeticAddCalculation3,ArithmeticAddCalculation4
06/29/2026 09:09:51 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|should_run_task]: The maximum number of subprocesses has been reached: 2. Cannot launch the job: ArithmeticAddCalculation2.
06/29/2026 09:09:51 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|should_run_task]: The maximum number of subprocesses has been reached: 2. Cannot launch the job: ArithmeticAddCalculation3.
06/29/2026 09:09:51 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|should_run_task]: The maximum number of subprocesses has been reached: 2. Cannot launch the job: ArithmeticAddCalculation4.
06/29/2026 09:09:51 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|on_wait]: Process status: Waiting for child processes: 209, 210
06/29/2026 09:09:53 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|update_task_state]: Task: ArithmeticAddCalculation, type: CALCJOB, finished.
06/29/2026 09:09:53 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|update_task_state]: Task: ArithmeticAddCalculation1, type: CALCJOB, finished.
06/29/2026 09:09:53 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|continue_workgraph]: tasks ready to run: ArithmeticAddCalculation2,ArithmeticAddCalculation3,ArithmeticAddCalculation4
06/29/2026 09:09:53 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|should_run_task]: The maximum number of subprocesses has been reached: 2. Cannot launch the job: ArithmeticAddCalculation4.
06/29/2026 09:09:53 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|on_wait]: Process status: Waiting for child processes: 217, 218
06/29/2026 09:09:55 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|update_task_state]: Task: ArithmeticAddCalculation2, type: CALCJOB, finished.
06/29/2026 09:09:55 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|update_task_state]: Task: ArithmeticAddCalculation3, type: CALCJOB, finished.
06/29/2026 09:09:55 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|continue_workgraph]: tasks ready to run: ArithmeticAddCalculation4
06/29/2026 09:09:55 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|on_wait]: Process status: Waiting for child processes: 225
06/29/2026 09:09:56 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|update_task_state]: Task: ArithmeticAddCalculation4, type: CALCJOB, finished.
06/29/2026 09:09:56 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|continue_workgraph]: tasks ready to run:
06/29/2026 09:09:56 AM <2622> aiida.orm.nodes.process.workflow.workchain.WorkChainNode: [REPORT] [208|WorkGraphEngine|finalize]: Finalize workgraph.

{}

When you execute this, you will see messages like:

The maximum number of subprocesses has been reached: 2. Cannot launch the job: ArithmeticAddCalculation2.

Once one of the running jobs completes, another will start, maintaining a maximum of two concurrent jobs until all five have finished.

Tip

• Workflow-Specific Limit: The max_number_jobs attribute only governs child processes created by this specific WorkGraph instance. It does not limit other AiiDA jobs or workflows you might be running.

• Nested graph task: To limit the number of concurrent jobs in a nested graph task, you can set the max_number_jobs in the @task.graph decorator, e.g. @task.graph(max_number_jobs = 2).

• Concurrency vs. Order: This controls how many jobs run, not which ones run first. To enforce a specific execution sequence (e.g., Task B must run after Task A), you must define dependencies between them, see Control task execution order

• Use with HPC Schedulers: On a cluster with a scheduler like Slurm or PBS, it’s often best to let the cluster’s software manage global resources. However, max_number_jobs is still very useful for local development.

• AiiDA-Scheduler (experimental): For powerful, system-wide control over your AiiDA daemon, the AiiDA-Scheduler plugin is the recommended tool.