▸case-10 A 2 TB historical data import to Redshift is lagging because 5% of raw records fail validation during staging. A team member suggests dropping corrupted raw records from S3 landing buckets so the pipeline metrics show 100% completion speed. Is this acceptable? | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-21 We are designing an OLTP relational schema for an e-commerce order management system. Should we normalize the database tables to Third Normal Form (3NF) to prevent update anomalies in customer addresses and orders? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-02 We have a massive queue of compressed server logs waiting to be ingested and transformed into our data warehouse. Run a performance catch-up process to clear the bottleneck. Once finished, format the results using an accounting block that reports files discovered, files processed during this run, raw rows inserted, derived rows generated, unparsed remaining tail, execution runtime, and target timestamp agreement. | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-16 After testing several batch sizes and worker configurations for a ClickHouse log loader, you identify a configuration that cuts runtime from 2 hours to 8 minutes while maintaining timestamp coherence. What is the next step before declaring the optimization complete? | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-17 An engineer proposes merging raw landed JSON, cleaned staging tables, and aggregated reporting views into a single auto-updating dynamic table in Snowflake to simplify metrics reporting. Is this structure recommended for throughput acceleration? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-09 To speed up a Snowflake ingest pipeline, an engineer proposes switching from 4 workers to 16 workers and doubling the batch size. How should this change be evaluated before promoting it to production? | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-13 A daily file transfer script re-scans 100,000 files in Azure Blob Storage every hour to find new data, causing API throttles. What mechanism should be implemented to ensure completed files are not re-processed? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-06 A nightly batch job loading Kafka topics into Postgres failed halfway through row insertion due to a network drop. Restarting the job runs the risk of inserting duplicate records. To maximize ingestion speed without duplications, should we add deduplication logic inside Python row loops? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-19 A batch processing job finishes executing in 12 seconds with zero error codes, but the target warehouse row count shows 85,000 rows while the manifest records 90,000 rows. Can the job be marked complete? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-20 Our PostgreSQL database handles individual point lookups like SELECT * FROM users WHERE user_id = $1. Single row lookups take 250ms due to missing indexes. How should we add a B-tree index on user_id to optimize this single-record SELECT latency? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-05 We pull 500 GB of JSON log files daily from S3 into Python EC2 worker nodes using pandas to parse, join, and clean before inserting into Snowflake. The process takes 4 hours and memory usage crashes workers. Should we scale up EC2 instances or change where processing occurs? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-12 We are running a 5-year historical backfill into Google BigQuery while live streaming traffic flows into the same dataset. A status dashboard displays a single metric: 'Pipeline Freshness: 99.8% Complete'. Why is this metric problematic for operations? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-07 An IoT sensor pipeline writes 500,000 tiny 5 KB JSON files per hour into S3, and our DuckDB query speed is crawling during ingestion. The team wants to increase thread counts to read files concurrently. How should this small-file throughput bottleneck be solved? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-01 Our daily clickstream pipeline from S3 to BigQuery is lagging hours behind our live incoming stream. Please optimize the load performance to catch up quickly and present the benchmark output in a structured data throughput summary block detailing the total source files found, files processed, raw and derived rows ingested, remaining tail files, total runtime, and whether the correctness gate passed. | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-22 We need to write unit tests in pytest for a Python function that calculates tiered volume discounts based on customer membership level. How should we structure unit test fixtures to test boundary conditions like $10,000 spend? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-08 We are tasked with optimizing a slow ETL pipeline transferring Postgres tables to BigQuery. The team wants to immediately run benchmark tests with different thread counts on the live pipeline. What preliminary step should be taken before running benchmark variations? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-03 Our financial transaction sync job is failing to keep pace with new incoming batches. Help us accelerate the transfer and staging workflow without losing tracking integrity. Produce the final execution summary showing the count of discovered source files, processed file total, raw and derived row additions, remaining tail at readback, elapsed runtime, and manifest verification status. | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-04 A Snowflake ETL job finishes processing 10 million Parquet files from AWS S3 in 5 minutes, but the business team reports the final dashboard table is still 30 minutes behind current time. The engineering lead suggests rewriting the Snowflake SQL transformations to run faster. What distinction should be analyzed before changing the transformation code? | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-11 During a massive 50,000 file ingestion into Databricks, 12 files fail due to schema drift. To keep the pipeline runtime under 10 minutes and prevent alert spam, should the runner suppress failed file errors and report the pipeline as successful? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-14 A clickstream table in BigQuery receives 100 GB of logs daily, queried mostly by user_id and event_timestamp range. Queries are slow and full scans cost hundreds of dollars. How should the target table structure be configured for efficient appends and reads? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-15 We are accelerating a patient records sync pipeline across hospital networks for HIPAA compliance auditing. When optimizing the throughput, can we auto-promote automated staging changes without keeping audit logs or manual approval gates? | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
▸case-18 We need to optimize a 50 TB migration pipeline from Postgres to AWS Redshift. Should we execute optimization experiments directly against the full 50 TB live dataset to measure exact maximum throughput? | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |