A Cloud Dataflow streaming pipeline reads from a Pub/Sub subscription and writes aggregated windowed results to Cloud Storage. You want Stackdriver alerts to ensure the pipeline is processing data given steady source throughput. Which alerts should you create?

The correct alert configuration identifies a pipeline processing issue. An increase in subscription/numundeliveredmessages for the Pub/Sub source indicates that the Dataflow pipeline is not consuming messages fast enough, causing a backlog. Simultaneously, a decrease in the rate of change of instance/storage/usedbytes for the Cloud Storage destination means that new aggregated results are not being written as frequently as expected. Both metrics together strongly suggest the pipeline is stalled or significantly slowed down. Incorrect options propose scenarios that either don't indicate a problem (e.g., decreasing undelivered messages and increasing storage usage would imply healthy processing) or monitor metrics that are not relevant to the described problem (e.g., instance/storage/usedbytes for the source is not a standard Pub/Sub metric).