GCP Professional Data Engineer Practice Question

Your company operates multiple Cloud Dataflow streaming pipelines across dev, staging, and prod projects. The SRE team wants to receive a PagerDuty notification when any production pipeline accumulates more than 10 minutes of system lag for at least 5 consecutive minutes, but they do not want alerts to fire after a job is drained or cancelled. Which Cloud Monitoring alerting configuration best satisfies these requirements?

Create a log-based metric that counts ERROR entries from Dataflow job logs and configure an uptime check alert when the metric count is non-zero for 5 minutes.
Set an alert on pubsub.googleapis.com/subscription/num_undelivered_messages aggregated across all streams and trigger when backlog exceeds the equivalent of 10 minutes of messages.
Create an alerting policy on metric dataflow.googleapis.com/job/current_system_lag for resource type dataflow_job, filter on label job_state="JOB_STATE_RUNNING", and fire when the maximum value exceeds 600 seconds for 5 minutes.
Alert on metric dataflow.googleapis.com/job/watermark with a condition that the minimum value is older than 600 seconds for 5 minutes; no additional label filters are required.

Report Issue

Answer Description

The metric that directly measures end-to-end delay in a streaming Dataflow job is dataflow.googleapis.com/job/current_system_lag. By creating an alerting policy on this metric with resource type dataflow_job, adding a filter that keeps only samples where the job_state label equals JOB_STATE_RUNNING, and setting a threshold of >600 seconds sustained for 5 minutes, the policy triggers only while the job is active and suppresses notifications once the job stops reporting. Using watermark metrics or Pub/Sub backlog does not capture Dataflow system lag directly, and relying on log-based or generic uptime checks would either miss the condition or generate false positives when a job ends.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.