A team is deciding how to organise objects within a lakehouse bucket. Which consideration is most relevant to query performance?
Select an answer to reveal the explanation.
Short Explanation
Partitioning and file size are what move the needle. Prune the partitions you don't need, and read a modest number of reasonably sized files rather than thousands of tiny ones - that's most of object-storage query tuning right there.
Full Explanation
The correct consideration is how data files are partitioned and sized, because partitioning lets an engine skip data that cannot satisfy a predicate, and file sizing determines how much per-object overhead the scan pays. Many very small files are the classic pathology: request overhead dominates and throughput collapses, which is why compaction exists.
The idea that key prefixes should use readable words rather than identifiers because engines parse prefixes to plan reads is incorrect, because engines resolve tables through catalog metadata rather than by interpreting key text. Readable prefixes help humans, not planners.
The idea that all objects for a table should be placed in a single prefix because object storage locks a prefix during writes is incorrect, because object storage does not lock prefixes. Concurrency is managed by the table format's commit protocol, not by storage-level locking.
The idea that object names should include the ingestion timestamp because the engine sorts by name before scanning is incorrect, because scan order is derived from metadata and the query plan, not from lexical ordering of names. Timestamps in names are an operational convenience at most.
For the exam, associate object-storage performance with partitioning, file size and compaction, and treat naming-based answers as distractors.