Databricks ML Professional Practice Question 27: Table Update Job Trigger
Databricks Certified Machine Learning Professional · Test 3 · question 7 of 10
Labelled outcomes are appended to a Unity Catalog table at irregular intervals, sometimes twice a day, sometimes not for three days. A retraining job currently runs on an hourly cron schedule and exits immediately when it finds no new labels, which it does on most runs. The team wants retraining to begin promptly after new labels land, without paying for the empty runs.
What should they configure?
Show answer and explanation
Answer:
A table update trigger makes the job event-driven: Databricks watches the named Unity Catalog table and starts a run when it changes. That matches the workload's actual shape, irregular arrivals, so the job starts soon after labels land and does not start at all in the days when nothing arrives. Both requirements are met by the trigger type rather than by tuning an interval.
Option B moves in the wrong direction on cost. Twelve times as many polls means twelve times as many empty runs, each paying cluster start-up, to reduce worst-case latency from an hour to five minutes.
Option C is the most expensive option available. A continuous job keeps a run active permanently, which is designed for always-on streaming pipelines and is badly matched to work that happens a few times a week.
Option D removes the automation the team is trying to build, adds human latency of hours or overnight, and cannot be relied on at weekends.
Reference: Trigger types for Lakeflow Jobs