Databricks Certified Data Engineer Associate Practice Questions: Production Pipelines Domain

A) Roll back to the previous pipeline version and backfill the CLV table from the last known good checkpoint

B) Increase the pipeline’s retry count and rerun today’s job to see if the row counts return to normal

C) Disable all monitoring alerts temporarily to avoid unnecessary noise while the team investigates the cause

D) Keep the new pipeline in place but manually correct the CLV table by inserting missing customers using ad-hoc SQL

A) Apply the fix directly in the production orchestration UI to minimize downtime, then document the change afterward.

B) Commit the fix to version control, run automated tests, and deploy to production through the existing CI/CD pipeline, even if it takes slightly longer.

C) Clone the production job into a separate "hotfix" job, apply the change there, and schedule it to run instead of the original job.

D) Pause the production pipeline until a full regression test suite can be run manually in all environments.

A) Continue manual deployments but require engineers to double-check their scripts and have a peer review before running them in production

B) Integrate pipelines and models into the existing CI/CD system with automated tests, data validation stages, and staged deployments (e.g., dev → test → prod) with gates

C) Deploy all changes directly to production but increase logging so that issues can be detected and fixed faster after they occur

D) Create a separate CI/CD system only for ML models, while keeping ETL pipelines as manual scripts to avoid overcomplicating deployments

A) Update the Databricks Repo configuration in staging and prod workspaces to track the `dev` branch instead of `main`, so that changes deployed to dev automatically propagate to staging and prod.

B) Extend the CI/CD pipeline with additional stages that, after approvals, run `databricks bundle deploy -t staging` and `databricks bundle deploy -t prod` to apply the same bundle to those workspaces.

C) Delete the `staging` and `prod` targets from `bundle.yml` and continue managing staging and prod jobs manually in the UI while using bundles only for dev.

D) Replace the `nightly_sales_etl` job resource in `bundle.yml` with three separate job resources (one per environment) and deploy them all to the dev workspace, then clone them manually into staging and prod.

A) Double the cluster size so the job finishes faster, accepting the higher compute cost as the price of meeting the SLA

B) Change the pipeline to process only new or changed data each day and compute aggregates incrementally, using partitioning where possible

C) Reduce the number of validation checks in the pipeline so that less time is spent on data quality and more on core transformations

D) Decrease the job frequency to every other day so that there is more time to process the larger batches

A) Increase the notebook’s compute resources so it can process predictions faster

B) Wrap the notebook in a scheduled job but keep using the engineer’s personal credentials for access

C) Refactor the notebook into a version-controlled, orchestrated pipeline that runs with service accounts and defined dependencies

D) Ask the engineer to double-check results manually after each run before sharing predictions

A) Increase the pipeline schedule to run every hour so that bad data is corrected more quickly when discovered

B) Add automated data quality checks (for volume, null rates, and key constraints) that fail the pipeline when critical thresholds are violated

C) Require manual review of a sample of records after each run before downstream tables are refreshed

D) Add more verbose logging to the existing tasks so engineers can inspect logs if a problem is suspected

A) Add a step that runs the model on a fixed validation dataset and enforces minimum performance thresholds before allowing deployment

B) Require manual approval from a product manager for every model deployment to ensure business alignment

C) Increase the number of unit tests around feature transformation code to reach 100% coverage

D) Add a linter that enforces strict coding style rules for all model training scripts

A) Reuse the existing hourly batch pipeline and have the app read loyalty balances from the warehouse, accepting up to one hour of staleness.

B) Build a streaming pipeline that consumes purchase events in near real time and updates a low-latency store used by the app for loyalty balances.

C) Increase the frequency of the batch pipeline to run every 5 minutes and have the app read from the warehouse.

D) Generate loyalty balances on-demand by querying all historical transactions from the warehouse each time a user opens the app.

A) Run all three tasks on a single, shared all-purpose cluster in the production workspace to maximize reuse of compute and minimize startup overhead.

B) Configure each task in the workflow to use serverless compute in the production workspace, leveraging per-task auto-scaling and avoiding cluster management.

C) Use serverless compute only for the ML scoring task and run the ETL and validation tasks on an all-purpose cluster in a development workspace.

D) Create a large, fixed-size job cluster for the ETL and scoring tasks, and use serverless compute only for the short validation task to reduce startup time.