Add tags to job created through invokation of endpoint deployment

Question

Hi there,

I want to use tags when invoking a batch endpoint pipeline component deployment, and I am using exactly this, as suggested here: https://learn.microsoft.com/en-us/answers/questions/5745760/how-to-set-pipeline-job-display-name-when-invoking?comment=answer-12567965&page=1#comment-2559071

ml_client.batch_endpoints.invoke(
    endpoint_name="pc-batch-endpoint-test",
    tags={
        "run_name": "pc-batch-endpoint-invocation-test",
        "source": "monthly-batch"
    }
)

It does not work. I cannot see the tags in the job overview inside azureml studio. I run this inside a notebook inside azureml studio. I am using azure-ai-ml 1.31.0.

What could be wrong?

Thanks a lot!

Answer 1

Hello BERGER, Matthias,

Welcome to Microsoft Q&A and Thank you for reaching out.

When invoking a batch endpoint pipeline deployment using the Python SDK call ml_client.batch_endpoints.invoke() with tags (for example, run_name or source), the invocation completes successfully, but the tags do not appear in the Azure ML Studio Job overview.

This behavior is expected for pipeline‑based batch endpoint deployments. The tags passed during invocation are accepted by the SDK but are not propagated to the underlying pipeline job metadata that is displayed in the Studio UI. This is a known platform limitation and not caused by incorrect code, SDK misuse, or the execution environment.

The tags are not visible because the batch_endpoints.invoke() method is designed as a convenience API for triggering batch endpoint runs.For pipeline component deployments, the invocation creates an internal pipeline job.

At present, job‑level metadata such as tags is not surfaced in the Azure ML Studio UI for these internally generated pipeline jobs.As a result, tags provided at invocation time do not appear in the job overview, even though the invocation itself succeeds.

Please check if these troubleshooting and Workarounds would help

1. Use descriptive endpoint or deployment naming – please include identifiers such as schedule name, batch frequency, or data source directly in the endpoint or deployment name.This information is reliably visible in the Studio UI.
2. Consider logging metadata inside the pipeline to emit run identifiers, source names, or batch context into logs or output artifacts.Logs and outputs are consistently available for pipeline jobs.
3. Please use tags for standalone jobs only -tags are fully supported and visible when submitting jobs directly using ml_client.jobs.create().This applies to standalone command jobs and batch jobs that are not invoked through a batch endpoint pipeline.
4. Use display names for clarity while tags are not surfaced, display names and pipeline structure still provide visibility into execution intent.

References:

Deploy pipelines with batch endpoints - Azure Machine Learning | Microsoft Learn

Create Jobs and Input Data for Batch Endpoints - Azure Machine Learning | Microsoft Learn

Thank you!

Answer 2

`wrong place

Answer 3

Tags for jobs created by invoking a batch endpoint pipeline component deployment are configured on the deployment side, not passed as tags in the invoke call.

In the available APIs, tags are modeled as part of the batch deployment configuration for pipeline component deployments:

• In PowerShell/ARM, IBatchPipelineComponentDeploymentConfiguration.Tag / BatchPipelineComponentDeploymentConfigurationTags represent "The tags which will be applied to the job." These tags are defined on the deployment configuration, not per-invocation.
• In the Java ARM SDK, BatchPipelineComponentDeploymentConfiguration.tags() returns "The tags which will be applied to the job." Again, this is a property of the deployment configuration.

The invoke examples for batch endpoints in the Python SDK only support specifying inputs (input / inputs) and do not show a tags parameter being honored for the created job. The documented way to influence the job is via data inputs, not tags, when calling ml_client.batch_endpoints.invoke.

Therefore, the behavior observed is expected: passing tags={...} to ml_client.batch_endpoints.invoke does not apply tags to the resulting job, so no tags appear in the job overview in Azure ML studio.

To have tags on jobs created from a pipeline component batch deployment, configure tags on the deployment configuration itself (for example, via the ARM/SDK layer where BatchPipelineComponentDeploymentConfiguration.tags / Tag is available). Those deployment-level tags are then applied to jobs created from that deployment.

---

References:

• How to operationalize a training pipeline with batch endpoints
• Create jobs and input data for batch endpoints
• BatchPipelineComponentDeploymentConfiguration Class
• IBatchPipelineComponentDeploymentConfiguration.Tag Property
• BatchPipelineComponentDeploymentConfigurationTags Class