Giving pipeline work to the wrong role
The wrong answers hand ingestion, cleansing and transformation pipelines to the database administrator, the data analyst or the AI engineer because each works near data. The database administrator secures and maintains the database, the analyst explores data for insights, and the AI engineer builds features on large language models and machine learning pipelines. The data engineer designs the ingestion and transformation workloads, applies data cleaning routines, and owns the management and monitoring of pipelines so that loads perform as expected.
See it in a sample questionGiving security, backup and tuning to the wrong role
The wrong answers give access control, backups and runtime tuning to the data engineer, the data analyst or the AI engineer. Engineers build pipelines and analysts consume data, so neither grants or denies user access. The database administrator manages security by granting privileges and denying access, stores backup copies, restores data after a failure or disaster, and owns the overall availability and consistent performance of the database systems.
See it in a sample questionSwapping one ACID property for another
Every ACID option is a real property, so the wrong answers pick a neighbour. Consistency, isolation and durability are offered where all-or-nothing completion is asked. Atomicity treats a transaction as one unit that succeeds completely or fails completely, such as a debit and a credit that both apply or neither does. Isolation stops concurrent transactions interfering, durability keeps a committed result after a restart, and consistency moves data from one valid state to another.
See it in a sample questionReading analytical work as transactional
The wrong answers swap the two workload types on the strength of the data source or the daily volume. A warehouse fed from a transactional source is still analytical, since it holds historical snapshots in a relational schema optimised for reading and reporting. Live transfers or payment events, recorded as small discrete units of work with create, retrieve, update and delete operations, are OLTP. High volume does not make a workload analytical, because analytical systems are read-only or read-mostly over historical data.
See it in a sample questionMixing up the Parquet and Avro layouts
The wrong answers swap the layouts of the two optimised binary formats. Parquet is columnar: a file holds row groups in which the data for each column is stored together, and its metadata helps locate the right chunk of rows. Avro is row-based, storing records as binary blocks behind a JSON header that an application uses to parse the data and extract its fields. Neither is human-readable text such as CSV or XML.
See it in a sample question