hudi更新失败

问题复现

spark-sql --queue xxx--conf 'spark.serializer=org.apache.spark.serializer.KryoSerializer'   --conf 'spark.sql.extensions=org.apache.spark.sql.hudi.HoodieSparkSessionExtension'

构造数据

drop table hudi_test;
create table h2 (
  id int,
  name string,
  price double,
  ts long,
  dt string
 ) using hudi
partitioned by (dt)
options (
 primaryKey = 'id',
 type = 'cow'
);

insert into hudi_test select 1 as id, 'hudi1' as name, 97 as price, 99 as ts, '2021-05-05' as dt;
insert into hudi_test select 2 as id, 'hudi2' as name, 97 as price, 99 as ts, '2021-05-05' as dt;

删除正常

spark-sql> delete from hudi_test where id = 1;
22/09/03 22:32:07 WARN SparkSession$Builder: Using an existing SparkSession; some spark core configurations may not take effect.
14:43  WARN: Timeline-server-based markers are not supported for HDFS: base path hdfs://R2/projects/xxx/hive/dev_db/hudi_test.  Falling back to direct markers.
14:49  WARN: Timeline-server-based markers are not supported for HDFS: base path hdfs://R2/projects/xxx/hive/dev_db/hudi_test.  Falling back to direct markers.
14:49  WARN: Timeline-server-based markers are not supported for HDFS: base path hdfs://R2/projects/xxx/hive/dev_db/hudi_test.  Falling back to direct markers.
Time taken: 7.776 seconds

更新异常

 update hudi_test set name = 'leon' where id = 2;

异常日志

Caused by: org.apache.hudi.exception.HoodieException: (Part -) field not found in record. Acceptable fields were :[id, name, price, ts, dt]
    at org.apache.hudi.avro.HoodieAvroUtils.getNestedFieldVal(HoodieAvroUtils.java:485)

原因分析

没有指定preCombineField，可以删除，但无法更新。

个人感觉hudi还有比较多细节问题没解决，中英文档还不够丰富。

解决方案

建表时指定preCombineField，如下

drop table hudi_test;
create table hudi_test (
  id int,
  name string,
  price double,
  ts long,
  dt string
 ) using hudi
partitioned by (dt)
options (
 primaryKey = 'id',
 preCombineField = 'ts',
 type = 'cow'
);