All Products
Search

This Product

    DataWorks:PAI DLC node

    Document Center

    DataWorks:PAI DLC node

    Last Updated:Mar 26, 2026

    DataWorks PAI DLC nodes let you wrap existing Deep Learning Containers (DLC) jobs in a scheduling workflow — adding periodic triggers, upstream/downstream dependencies, and centralized monitoring.

    Prerequisites

    Before you begin, ensure that you have:

    Write task code

    On the PAI DLC node editing page, write the task code using one of two methods.

    Load an existing DLC job

    Search for a DLC job by name in PAI and load it. The node editor generates code from the job's existing configuration, which you can then edit.

    image
    If you don't have permission to load or create a job, follow the on-screen instructions to grant access. If no jobs are available, go to the PAI console to create one — see Create a training job, Create a training job: Python SDK, and Create a training job: command line.

    Write task code directly

    Write task code directly in the PAI DLC node editor. Use the dlc submit pytorchjob command to submit a PyTorchJob.

    Define variables in your code using the ${variable_name} format, then assign values in the Scheduling Parameters section under Properties on the right side of the page. This lets you pass parameters dynamically at scheduling time. For more information, see Sources and expressions of scheduling parameters.

    The following example submits a PyTorchJob with common configuration options:

    dlc submit pytorchjob \
    --name=test \
    --command='echo '\''hi'\''' \
    --workspace_id=309801 \
    --priority=1 \
    --workers=1 \
    --worker_image=<image> \
    --image_repo_username=<username> \
    --image_repo_password=<password> \
    --data_source_uris=oss://oss-cn-shenzhen.aliyuncs.com/::/mnt/data/:{mountType:jindo} \
    --worker_spec=ecs.g6.xlarge

    Replace the placeholders before running:

    Placeholder Description
    <image> Path of the container image for the worker
    <username> Username for authenticating against a private image registry
    <password> Password for authenticating against a private image registry

    Key parameters:

    Parameter Description
    --name Name of the DLC job. Use a variable name or the DataWorks node name.
    --command Command to run inside the container.
    --workspace_id ID of the PAI workspace where the job runs.
    --priority Job priority. Valid values: 1–9. 1 is the lowest priority; 9 is the highest.
    --workers Number of worker nodes. Set to more than 1 to run a distributed training job across multiple nodes.
    --worker_image Container image for each worker node.
    --image_repo_username / --image_repo_password Credentials for a private image registry. Omit for public images.
    --data_source_uris Mounts an Object Storage Service (OSS) data source into the container at the specified path. The example uses the jindo mount type.
    --worker_spec Instance type for each worker node, for example, ecs.g6.xlarge.

    Configure and run the node

    1. In the Run Configurations section, set Resource Group to a scheduling resource group that can reach your data source. For network connectivity options, see Network connectivity solutions.

    2. In the toolbar, click Run to run the node task immediately.

    Configure scheduling (optional)

    To run the node on a recurring schedule, configure its scheduling properties — including recurrence type, time zone, and upstream dependencies. See Configure scheduling properties for a node.

    Publish the node

    After the node is configured, publish it to the production scheduling environment. See Publish a node or workflow.

    Monitor scheduled runs

    After the node is published, auto-triggered task instances appear in the Operation Center. For a walkthrough, see Get started with Operation Center.

    What's next