Updated on 2022-11-18 GMT+08:00

Constraints

Hudi 0.9.0 adds Spark SQL DDL and DML statements for using Hudi, making it easier for all users (including non-engineers or analysts) to access and operate Hudi.

Constraints

  • You can use Spark SQL to operate Hudi on the Hudi client.
  • You can use Spark SQL to operate Hudi in JDBCServer of Spark2x.
  • You cannot use Spark SQL to operate Hudi on the Spark2x client.
  • You cannot operate Hudi in the Hive or Hetu engine.
  • The default value of KeyGenerator in SQL is org.apache.hudi.keygen.ComplexKeyGenerator. Therefore, you need to set the KeyGenerator value to that of SQL when data is written in DataSource mode.
  • If you run the Show Partitions command for the first time, you need to run the msck repair command as prompted.
  • Spark SQL does not support incremental view query.
  • Spark SQL does not support ALTER CHANGE DATA TYPE.
  • To write data incrementally using SQL into a COW table created using DataSource, you need to run the ALTER TABLE table_name SET SERDEPROPERTIES(primaryKey='xx',preCombineField='xx') statement in SQL to reset two properties of the table. To write data incrementally using SQL into an MOR table created using DataSource, you need to specify a path to the one set for creating this MOR table.