Amazon SageMaker Pipelines: build a pipeline with Processing, Training, and RegisterModel steps using the SageMaker Python SDK

Steps

1. Set up a PipelineSession (from sagemaker.workflow.pipeline_context import PipelineSession) alongside your usual boto3/sagemaker.Session and execution role — pipeline step definitions use this session instead of the eager session so calls are deferred into the pipeline graph.
2. Define a Processing step: create a processor (e.g. SKLearnProcessor) with sagemaker_session=pipeline_session, call processor.run(inputs=[ProcessingInput(...)], outputs=[ProcessingOutput(output_name="train", ...), ...], code="preprocessing.py") to get step_args, then wrap it in ProcessingStep(name="...", step_args=step_args) from sagemaker.workflow.steps.
3. Define a Training step: configure an Estimator (e.g. XGBoost via sagemaker.estimator.Estimator) with sagemaker_session=pipeline_session, call estimator.fit(inputs={"train": TrainingInput(s3_data=step_process.properties.ProcessingOutputConfig.Outputs["train"].S3Output.S3Uri, ...)}) to get step_args, then wrap in TrainingStep(name="...", step_args=step_args) — note how the training input references the processing step's output property, wiring the DAG dependency automatically.
4. Define a model-registration step with sagemaker.workflow.step_collections.RegisterModel, passing the trained estimator, model_data=step_train.properties.ModelArtifacts.S3ModelArtifacts, content_types, response_types, inference_instances/transform_instances, and a model_package_group_name so runs accumulate into one Model Registry group.
5. Assemble everything with from sagemaker.workflow.pipeline import Pipeline; pipeline = Pipeline(name=..., parameters=[...], steps=[step_process, step_train, step_register]); optionally validate with json.loads(pipeline.definition()) before creating/starting the pipeline.
6. Optionally wrap the RegisterModel step in a ConditionStep (sagemaker.workflow.condition_step) so registration only runs if an evaluation metric (e.g. MSE from a model-evaluation Processing step) meets a threshold.