Task 5.4: Monitoring and troubleshooting processes.

• AINFORMATION_SCHEMA.JOBS, summing total_bytes_billed by user_email

  Correct

  The JOBS view holds near-real-time metadata about each job, including the user who ran it and the bytes billed, for the last 180 days.

• BINFORMATION_SCHEMA.TABLE_STORAGE, grouped by the table owner

  Incorrect

  TABLE_STORAGE reports storage bytes per table, not query usage per user.

• CThe Cloud Billing export table, grouped by service description

  Incorrect

  Billing export breaks costs down by service and SKU, not by the user who ran each query.

• DAdmin Activity audit log entries for the BigQuery service

  Incorrect

  Admin Activity logs record configuration changes such as creating datasets, not the bytes each query billed.

INFORMATION_SCHEMA.JOBS (and JOBS_BY_USER, JOBS_BY_ORGANIZATION) is the place to analyze who ran what and what it cost: it records each job's user, query, bytes processed and billed, slot time and timing, which makes per-user cost reports a simple aggregation.

• AA daily review of the job graph in the Dataflow console

  Incorrect

  A daily look finds an hour-long delay far too late.

• BAn alert on the job's worker CPU utilization above 90%

  Incorrect

  CPU can stay normal while the job falls behind, for example when it waits on a slow sink.

• CA Cloud Monitoring alert on the job's system lag above two minutes

  Correct

  System lag reports the maximum lag across the pipeline in seconds, which is the measure of falling behind, and an alerting policy notifies someone when it crosses the threshold.

• DA log-based metric that counts worker log lines per minute

  Incorrect

  Log volume is not a measure of how far behind the pipeline is.

Dataflow publishes job metrics to Cloud Monitoring. For streaming freshness, system lag is the signal to alert on, with a threshold that matches the latency requirement; the Failed metric covers job failures.

• ACreate a budget for the project and connect it to a Pub/Sub topic

  Correct

  Budgets can send programmatic notifications to a Pub/Sub topic, which code can subscribe to.

• BSet a custom quota on the project's Dataflow job count

  Incorrect

  A quota caps one resource at all times; it is not tied to spend and does not react to the budget.

• CAsk Cloud Billing to stop charges at $2,000 automatically

  Incorrect

  Spend cap budgets cover only a few services, such as Cloud Run and the Gemini API, not Dataflow, and an alerts-only budget never stops spending by itself.

• DAdd more email recipients to the budget's alert settings

  Incorrect

  More emails still depend on a person reading and acting on them.

• EDeploy a function that subscribes to the topic and stops the jobs

  Correct

  A subscriber that receives the budget notifications can apply the team's response automatically.

Alerts-only Cloud Billing budgets alert on actual or forecast spend but do not stop spending, and spend cap budgets apply only to a short list of services that does not include Dataflow. Programmatic notifications publish budget updates to Pub/Sub, so a subscriber such as a Cloud Run function can respond automatically, for example by stopping jobs or disabling resources.

• AThe administrative resource charts in BigQuery monitoring

  Correct

  The resource charts show slot usage, job concurrency with running and pending jobs, job duration and bytes processed, filtered by reservation, project or user.

• BThe Cloud Billing reports page for the BigQuery service

  Incorrect

  Billing reports show cost, not slot usage or pending jobs.

• CThe query execution graph of one slow query

  Incorrect

  One query's graph shows its own stages, not usage across reservations.

• DThe audit logs of the reservation administration project

  Incorrect

  Audit logs record configuration changes and access, not slot usage over time.

BigQuery's administrative resource charts are the console view of capacity: slot usage including borrowed and idle slots, job concurrency, job duration and bytes processed, grouped and filtered by reservation, folder, project or user.

• AINFORMATION_SCHEMA.TABLE_STORAGE for the dataset

  Incorrect

  Storage metadata describes existing tables' bytes; it does not record who deleted a table.

• BThe dataset's time travel history for the table

  Incorrect

  Time travel can help restore the data, but it does not say who deleted it.

• CThe project's billing export to BigQuery

  Incorrect

  Billing data records charges, not the identity behind a deletion.

• DAdmin Activity audit logs in Cloud Logging

  Correct

  Admin Activity logs, which are always written, record actions that change resources or their metadata, including deletions, with the caller identity.

Cloud Audit Logs has several types. Admin Activity audit logs are always written, cannot be disabled, and record changes to resource configuration and metadata, such as deleting a table, with who did it and when.