Github user xuchuanyin commented on the issue: https://github.com/apache/carbondata/pull/2734 Hi, I've tested this PR in local machine and it works fine. Steps used to verify this: ``` 1. Use CarbonData 1.4.1-RC2 jar and start spark & JDBCServer & beeline 2. CREATE TABLE create table test_adpt_int (id int, name string, age int) stored by 'carbondata' TBLPROPERTIES('sort_columns'='id'); 3. CREATE DATAMAP create datamap dm_id on table test_adpt_int using 'bloomfilter' DMPROPERTIES('index_columns'='id'); 4. LOAD insert into table test_adpt_int values (1, 'name1', 10),(3, 'name3', 30),(5, 'name5', 50),(7, 'name7', 70),(9, 'name9', 90),(10, 'name10', 100); 5. QUERY select * from test_adpt_int where id = 6; select * from test_adpt_int where id = 5; 6. Use master code and apply current PR to generate jar and restart spark & JDBCServer & beeline 7. QUERY should work fine select * from test_adpt_int where id = 6; select * from test_adpt_int where id = 5; 8. LOAD again insert into table test_adpt_int values (1, 'name1', 10),(3, 'name3', 30),(5, 'name5', 50),(7, 'name7', 70),(9, 'name9', 90),(10, 'name10', 100); 9. QUERY again should work fine select * from test_adpt_int where id = 6; select * from test_adpt_int where id = 5; ``` besides, the bloom index folder looks like below: ![image](https://user-images.githubusercontent.com/10445758/45991871-53ab0600-c0b9-11e8-8320-38337b6eb23f.png) The segment generated in 1.4.1 do not have version info file while the segment generated in 1.5.0 has the version info. *Note:* in 1.5.0, we introduce the 'mergeShard' to merge the bloom index file.
---