Data Skew
A non-uniform distribution of values, keys, or workload across categories or partitions that can distort estimates, parallelism, and resource utilization.
Equal row counts across partitions do not imply equal work. A few hot keys can carry most traffic, saturating some workers while others remain underused. In a database, the same skew can invalidate uniform-distribution assumptions and produce poor cardinality estimates.
Average partition size is therefore insufficient. Per-key traffic, histograms, percentiles, and shard-level CPU or I/O reveal the actual imbalance. Concurrency in distributed data systems and Oracle Database and PL/SQL show the same distribution problem at different layers.