Automating and orchestrating ML pipelines

A Vertex AI Pipelines job defined with the KFP SDK v2 has caching enabled by default for all components. During a debugging session, a data scientist needs to force every step to re-execute for a single pipeline run without modifying the pipeline definition file. Which approach achieves this?

A retail company's ML workflow must: (1) run a Vertex AI custom training job, (2) execute a data validation step in BigQuery, and (3) trigger a legacy on-premises Hadoop MapReduce job over Cloud VPN before promoting the model, with complex branching logic depending on the Hadoop job's exit code. Which orchestration tool should manage this end-to-end workflow?

After evaluating a new model version, an ML engineer needs to designate it as the version served in production, replacing the previous production model, while keeping the old version available under its own identifier for rollback. Which action correctly accomplishes this using Vertex AI Model Registry?

In Vertex ML Metadata, a pipeline run produces a trained model artifact from a training dataset artifact. Which statement correctly describes how this relationship is represented in the lineage graph?

A Cloud Build pipeline compiles and pushes a new KFP pipeline template whenever code is merged to main. Before the new model version can be used for production predictions, an ML lead must manually review evaluation metrics and approve the promotion. Which design best implements this human-in-the-loop approval gate using Vertex AI capabilities?

New labeled training data is uploaded to a Cloud Storage bucket at unpredictable intervals, sometimes multiple times per day, sometimes not for a week. The team wants the retraining Vertex AI Pipeline to start automatically within minutes of new data arriving, without polling on a fixed schedule. Which architecture best satisfies this requirement?