Skip to content

concept: Data Skew

Jump to bottom
David Liu edited this page May 22, 2026 · 3 revisions

Data skew means the data is not distributed evenly across the distributions

  • some distributions finish their portion of a parallel query before others. Since the query can't complete until all distributions have finished processing, each query is only as fast as the slowest distribution.

  • 一个处理器核心承担的工作比其他处理器核心多得多，进而成为瓶颈

  • aka. low key cardinality

Clone this wiki locally