Help Center/ MapReduce Service/ Component Development Specifications/ Hudi/ Spark on Hudi Stream Scenario Operation Guide/ Determining the Table Type
Updated on 2026-02-25 GMT+08:00
View PDF
Share

Determining the Table Type

Table Type

Pros

Cons

Preferred Scenario

MOR
  • Fast write. Incremental data is written into log files.
  • Slow read. Log files and Parquet files are merged during read.
  • Table service maintenance operations including compaction, clean, and archive need to be performed periodically. Otherwise, the MOR tables will be read more and more slowly.
  • Spark/Flink (real-time migration task) stream scenario for real-time write.
  • In Spark micro-batch scenarios (CDM migration tasks and DataArts Studio scheduled Spark tasks), services require that data can be written within minutes, for example, within 5 minutes.

COW
  • Fast read. COW tables contain only Parquet files.
  • Compaction is not required. Only clean and archive are required.

Slow write. Incremental data and existing Parquet files are merged during write.
  • In Spark batch scenarios (CDM migration tasks and DataArts Studio scheduled Spark tasks), services have no requirements on write performance. Data can be written in minutes or hours.
  • insert overwrite scenario.