[HIVE-22661] Compaction fails on non bucketed table with data loaded inpath

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 4.0.0-alpha-1
• Assignee: Ádám Szita
• Reporter: Ádám Szita
• Created: 18/Dec/19 14:53
• Updated: 17/Nov/22 08:51
• Resolved: 08/Jan/20 10:26

Description

Compaction cannot handle situations where:

• data was ingested with LOAD DATA INPATH
• this ingest method is run multiple times, and
  • with different number of files getting created in the delta directories

Therefore, for file/dir structures such as:

/warehouse/tablespace/managed/hive/comp3/delta_0000001_0000001_0000
/warehouse/tablespace/managed/hive/comp3/delta_0000001_0000001_0000/000000_0
/warehouse/tablespace/managed/hive/comp3/delta_0000001_0000001_0000/000001_0
/warehouse/tablespace/managed/hive/comp3/delta_0000002_0000002_0000
/warehouse/tablespace/managed/hive/comp3/delta_0000002_0000002_0000/000000_0
/warehouse/tablespace/managed/hive/comp3/delta_0000002_0000002_0000/000001_0
/warehouse/tablespace/managed/hive/comp3/delta_0000002_0000002_0000/000002_0 

Although the table is not bucketed, bucket is calculated from the (raw) files' names. Compaction in the above case will fail on delta1-1 not having data for 'bucket' 2.

Steps to repro using small dataset:

set tez.grouping.min-size=8;
set tez.grouping.max-size=8;
set mapreduce.input.fileinputformat.split.minsize=8;
set mapreduce.input.fileinputformat.split.minsize=8;

create external table comp0 (a string);
insert into comp0 values ("qwertyuiopasdfghjklzxcvbnm");
insert into comp0 values ("qwertyuiopasdfghjklzxcvbnm");

create external table comp1 stored as orc as select * from comp0;

insert into comp0 values ("qwertyuiopasdfghjklzxcvbnm");
create external table comp2 stored as orc as select * from comp0;

create table comp3 (a string);
load data inpath '/warehouse/tablespace/external/hive/comp1' into table comp3;
load data inpath '/warehouse/tablespace/external/hive/comp2' into table comp3;