MOR seems not working as expected very well in some corner case

[bithw1]

Describe the problem you faced

Hi，

I have a simple MOR table, defined as follows:

create table test_hudi_table_mor (
a int,
b int,
c int
) using hudi
 partitioned by (c)
 options (
  primaryKey = 'a',
  preCombineField = 'b',
  type = 'mor'
 );

Then, I do insert twice,

insert into test_hudi_table_mor  values(1,1,1);
insert into test_hudi_table_mor  values(2,2,2);

When I looked at the files created, I find two parquet files, no log files

Then, I do one update

 update test_hudi_table_mor  set b = 11 where a =1;

Then, I find a log file, so I would ask, why the two insert operations doesn't generate log files and only update generate, thanks.

Where the code in Hudi that I can refer to see why the behavior works like this.

[cshuo]

@bithw1 it depends on which index you're using, spark use SIMPLE index by default, which do not support file slice without base file, so base file will be generated for the first insert records. If you use index like hoodie.index.type = 'INMEMORY', you can see log files generated for the first two inserts.