Database Administration Flashcards

An AI model serving system requires sub-millisecond feature lookups. Which database architecture best satisfies this latency requirement?

Answer: An in-memory key-value store (e.g., Redis) with pre-computed features

In-memory key-value stores like Redis serve pre-computed features in microseconds from RAM, making them the standard choice for low-latency online feature serving.

When designing a database schema for A/B testing AI model variants, which schema design correctly associates each prediction with its serving model version?

Answer: Create a foreign key from predictions.model_version_id to a model_versions table

A foreign key to a normalized model_versions table enforces referential integrity, enables joins for performance analysis, and avoids schema changes when new versions are added.

A PostgreSQL database supporting AI workloads shows high I/O wait due to frequent random reads. Which configuration change most directly addresses this for read-heavy analytical queries?

Answer: Increasing effective_cache_size and shared_buffers to better utilize RAM

Increasing shared_buffers allows PostgreSQL to cache more data in memory, and setting effective_cache_size correctly guides the planner to prefer index scans, both reducing random disk reads.

An AI application stores user interaction logs at 10,000 events per second. Which database design decision prevents write performance degradation as the table grows past 1 billion rows?

Answer: Declarative table partitioning by time range with automatic partition creation

Time-range partitioning keeps each partition small, so inserts target only the current partition, maintaining consistent write latency regardless of total historical data volume.

A fintech AI system must ensure that two database operations — debiting a source account and crediting a destination account — always succeed or fail together. Which database property guarantees this?

Answer: Atomicity

Atomicity ensures that all operations within a transaction are treated as a single unit — either all are committed or all are rolled back if any step fails.

An AI consultant is comparing PostgreSQL and MongoDB for storing semi-structured AI inference outputs that vary in schema across model versions. What is MongoDB's primary advantage in this scenario?

Answer: MongoDB's flexible document model accommodates varying schemas without migrations

MongoDB's schema-free document model allows different documents in the same collection to have different fields, making it well-suited for outputs that vary by model version.

Which database replication topology provides the highest read scalability for an AI system that reads feature data 1000x more than it writes?

Answer: Single primary with multiple read replicas and load-balanced reads

A single primary with multiple read replicas distributes the read workload across many nodes while centralizing writes on the primary to avoid conflicts.