Skip to content
Cheat sheets

Data Engineer Professional · Domain 8: Debugging and Deploying · 10% of the exam

Task 8.2: Analyze the errors and remediate the failed job runs with job repairs and parameter overrides.

Repair runs, rerunning failed and dependent tasks, changing parameters for a repair or a run, and what a repair keeps from the original run.

Study it

  • Repairing failed runs and overriding parameters

    Lesson coming

Sample questions

Try each one before opening the answer. Every option is explained, with the Databricks documentation page that proves it.

Question 1 · choose 1

In a five-task job, the task load_gold appends rows to a gold table and failed halfway because of a bad lookup, after writing about half of its output. The engineer fixed the lookup and wants to repair the run so that only load_gold and its dependents run again, without leaving duplicate rows. What must be true before the repair?

  1. ANothing; repair resumes load_gold from the row where it stopped
  2. BThe job must be run again with Run now instead of repaired
  3. CThe successful tasks must be disabled before the repair
  4. Dload_gold writes idempotently, for example with MERGE or a scoped overwrite
Show the answer and why

  • ANothing; repair resumes load_gold from the row where it stopped

    Incorrect

    Repairs rerun tasks from the start, not from a saved position.

  • BThe job must be run again with Run now instead of repaired

    Incorrect

    Run now reruns all five tasks and still appends duplicates.

  • CThe successful tasks must be disabled before the repair

    Incorrect

    Repair already skips successful tasks; disabling them does not stop duplicates.

  • Dload_gold writes idempotently, for example with MERGE or a scoped overwrite

    Correct

    A repair reruns each unsuccessful task from the beginning, and jobs do not make tasks idempotent, so the task's own write must be safe.

Repair runs only the unsuccessful tasks and their dependents. Each rerun task starts from the beginning, so a task that wrote part of its output must use idempotent operations such as overwrite or merge.

Databricks documentation

Task 8.2Report a problem with this question

Question 2 · choose 1

A nightly multi-task run failed at its third task because the job parameter run_date was passed as 2026-13-01. The first two tasks succeeded and must not run again. How should the engineer rerun the failed work with run_date = 2026-10-30?

  1. AUse Run now with run_date = 2026-10-30
  2. BRepair run, with run_date = 2026-10-30 in the Repair dialog
  3. CHardcode the date in the third task's notebook and repair
  4. DRepair the run without changes, then fix the parameter later
Show the answer and why

  • AUse Run now with run_date = 2026-10-30

    Incorrect

    Run now starts a new run of every task, including the two that succeeded.

  • BRepair run, with run_date = 2026-10-30 in the Repair dialog

    Correct

    Parameters entered in the Repair dialog override existing values for the tasks that are rerun.

  • CHardcode the date in the third task's notebook and repair

    Incorrect

    Hardcoding breaks later runs and is not needed, because the repair can override parameters.

  • DRepair the run without changes, then fix the parameter later

    Incorrect

    The rerun task would get the same invalid date and fail again.

A repair reruns the unsuccessful tasks and their dependents. In the Repair job run dialog you can add or edit parameters, which override the existing values for that repair.

Databricks documentation

Task 8.2Report a problem with this question

Question 3 · choose 1

A job with a single notebook task failed. The Repair run button is not offered. How is the failed work rerun?

  1. AEnable repair in the job's advanced settings
  2. BTrigger the job again with Run now
  3. CAdd retries to the failed run
  4. DClone the run and edit it
Show the answer and why

  • AEnable repair in the job's advanced settings

    Incorrect

    There is no such setting; single-task jobs cannot be repaired.

  • BTrigger the job again with Run now

    Correct

    Repair works only for jobs that orchestrate two or more tasks.

  • CAdd retries to the failed run

    Incorrect

    Retries apply to future attempts, not a finished run.

  • DClone the run and edit it

    Incorrect

    Runs cannot be cloned and edited.

Repair is supported only for jobs with two or more tasks. To rerun a failed single-task job, trigger it again with Run now.

Databricks documentation

Task 8.2Report a problem with this question

Question 4 · choose 1

Tasks of a failed run shared the job cluster my_job_cluster. After a repair, the engineer sees a cluster named my_job_cluster_v1. Why?

  1. AThe repair upgraded the cluster to a newer Databricks Runtime
  2. BThe first cluster was deleted because its tasks failed
  3. CRepairs always run on serverless compute with a new name
  4. DRepairing tasks that shared a job cluster creates a new one
Show the answer and why

  • AThe repair upgraded the cluster to a newer Databricks Runtime

    Incorrect

    The name shows a new cluster, not an upgrade.

  • BThe first cluster was deleted because its tasks failed

    Incorrect

    The original remains visible with its settings.

  • CRepairs always run on serverless compute with a new name

    Incorrect

    Repairs use the tasks' configured compute.

  • DRepairing tasks that shared a job cluster creates a new one

    Correct

    The new cluster keeps the original run's cluster visible for comparison.

If tasks share a job cluster, a repair run creates a new job cluster, such as my_job_cluster_v1, so the clusters of the initial and repair runs can be told apart.

Databricks documentation

Task 8.2Report a problem with this question

Question 5 · choose 1

A run failed twice and succeeded on its second repair. The Runs tab shows a duration far longer than the final repair took. Why?

  1. ADuration includes only the time the run waited in the queue
  2. BThe duration of failed attempts is doubled in the Runs tab
  3. CIt spans from the first run's start to the latest repair's end
  4. DThe repair reran every task from scratch, including passed ones
Show the answer and why

  • ADuration includes only the time the run waited in the queue

    Incorrect

    It spans all attempts, not queue time alone.

  • BThe duration of failed attempts is doubled in the Runs tab

    Incorrect

    No such adjustment is made.

  • CIt spans from the first run's start to the latest repair's end

    Correct

    The displayed duration includes the original run and every repair.

  • DThe repair reran every task from scratch, including passed ones

    Incorrect

    Repair reruns only unsuccessful tasks and dependents.

For a repaired run, the Duration in the Runs tab spans from the start of the first run to the end of the latest repair run.

Databricks documentation

Task 8.2Report a problem with this question

Question 6 · choose 1

A task was disabled when the run happened, but the engineer now wants it to run as part of a repair through the API. How?

  1. AList the task in rerun_tasks in the repair request
  2. BNothing; a repair always runs tasks that were disabled
  3. CDelete the task and add it again before the repair
  4. DStart a new run of the job with Run now instead
Show the answer and why

  • AList the task in rerun_tasks in the repair request

    Correct

    Repairs use run state; listing a disabled task in rerun_tasks forces it to run.

  • BNothing; a repair always runs tasks that were disabled

    Incorrect

    Repairs decide by run state, so the task must be named.

  • CDelete the task and add it again before the repair

    Incorrect

    Recreating the task is not needed.

  • DStart a new run of the job with Run now instead

    Incorrect

    That reruns everything, not a repair.

Repairs use each task's run state to decide what to repair, not whether it is disabled. To force a disabled task to run in a repair, include it in rerun_tasks.

Databricks documentation

Task 8.2Report a problem with this question

Practice domain 8 →Practice all domains →