You recently deployed a model to a Vertex AI endpoint and set up online serving in Vertex AI Feature Store. You have configured a daily batch ingestion job to update your featurestore. During the…

While scheduling an increase in nodes might help, it is a manual and less dynamic approach than autoscaling, which can adapt more precisely to fluctuating loads and potentially optimize costs by scaling down when demand is low.

Enabling autoscaling for the Vertex AI Feature Store's online serving nodes allows the system to automatically adjust the number of nodes based on real-time demand. This ensures that sufficient capacity is available to maintain low latency and high performance for online feature retrieval, even during peak loads caused by concurrent batch ingestion jobs.

Enabling autoscaling for the Vertex AI endpoint's prediction nodes addresses the scaling of the deployed machine learning model, not the performance of the Vertex AI Feature Store's online serving, which is the component experiencing high CPU and latency.

Increasing the `worker_count` for `ImportFeatureValues` would speed up the batch ingestion process itself, but it could potentially place more load on the feature store's backend during ingestion, possibly worsening online serving performance rather than improving it.