Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
The following repo was provided by haozhu.
1. Create a Hbase table with 4 regions
create 'myhbase', 'cf1','cf2', {SPLITS => ['a', 'b', 'c']} put 'myhbase','a','cf1:col1','somedata' put 'myhbase','b','cf1:col2','somedata' put 'myhbase','c','cf2:col1','somedata'
One region has cf1.col1. One region has column family 'cf1', but does not have 'col1' under 'cf1'. One region has only column family 'cf2'. And last region is complete empty.
2. Prepare a csv file.
select * from dfs.tmp.`joinhbase.csv`; +-------------------+ | columns | +-------------------+ | ["1","somedata"] | | ["2","somedata"] | | ["3","somedata"] |
Now run the following query on drill 1.11.0-SNAPSHOT:
select cast(H.row_key as varchar(10)) as keyCol, CONVERT_FROM(H.cf1.col1, 'UTF8') as col1 from hbase.myhbase H JOIN dfs.tmp.`joinhbase.csv` C ON CONVERT_FROM(H.cf1.col1, 'UTF8')= C.columns[1] ;
The correct query result show be:
+---------+-----------+ | keyCol | col1 | +---------+-----------+ | a | somedata | | a | somedata | | a | somedata | +---------+-----------+
Turn off broadcast join, then we will see SchemaChangeException, or incorrect result randomly. By 'randomly', it means in the same session, the same query would hit SchemaChangeException in one run, while gets incorrect result in a second run.
alter session set `planner.enable_broadcast_join`=false;
select cast(H.row_key as varchar(10)) as keyCol, CONVERT_FROM(H.cf1.col1, 'UTF8') as col1 . . . . . . . . . . . . . . . . . .> from . . . . . . . . . . . . . . . . . .> hbase.myhbase H JOIN dfs.tmp.`joinhbase.csv` C . . . . . . . . . . . . . . . . . .> ON CONVERT_FROM(H.cf1.col1, 'UTF8')= C.columns[1] . . . . . . . . . . . . . . . . . .> ; Error: SYSTEM ERROR: SchemaChangeException: Hash join does not support schema changes
+---------+-------+ | keyCol | col1 | +---------+-------+ +---------+-------+ No rows selected (0.302 seconds)
Attachments
Issue Links
- is related to
-
-
- Resolved
-