[DRILL-3702] PartitionPruning hit ClassCastException in Interpreter when the pruning filter expression is of non-nullable type.

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.2.0
• Component/s: Query Planning & Optimization
• Assignee: Jinfeng Ni
• Reporter: Jinfeng Ni
• Created: 24/Aug/15 22:06
• Updated: 29/Sep/15 18:08
• Resolved: 26/Aug/15 22:10

Description

I have the following parquet table, created using partition by clause:

create table mypart (id, name) partition by (id) as select cast(n_regionkey as varchar(20)), n_name from cp.`tpch/nation.parquet`;

The generated parquet table consists of 5 files, each representing a partition:

0_0_1.parquet 0_0_2.parquet 0_0_3.parquet 0_0_4.parquet 0_0_5.parquet

For the following query, partition pruning works as expected:

select id, name from mypart where id  = '0' ;

00-01      Project(id=[$1], name=[$0])
00-02        Scan(groupscan=[ParquetGroupScan [entries=[ReadEntryWithPath [path=/tmp/mypart/0_0_1.parquet]], selectionRoot=file:/tmp/mypart, numFiles=1, columns=[`id`, `name`]]])

"selectionRoot" : "file:/tmp/mypart",
    "fileSet" : [ "/tmp/mypart/0_0_1.parquet" ],
    "cost" : 5.0

However, the following query would hit ClassCastException when PruneScanRule calls interpreter to evaluate the filtering condition, which happens to be non-nullable.

select id, name from mypart where concat(id,'')  = '0' ;

00-05              Project(id=[$1], name=[$0])
00-06                Scan(groupscan=[ParquetGroupScan [entries=[ReadEntryWithPath [path=file:/tmp/mypart]], selectionRoot=file:/tmp/mypart, numFiles=1, columns=[`id`, `name`]]])

"selectionRoot" : "file:/tmp/mypart",
    "fileSet" : [ "/tmp/mypart/0_0_1.parquet", "/tmp/mypart/0_0_4.parquet", "/tmp/mypart/0_0_5.parquet", "/tmp/mypart/0_0_2.parquet", "/tmp/mypart/0_0_3.parquet" ],
    "cost" : 25.0
  },

Here is the error for the ClassCastException, raised in Interpreter:

java.lang.ClassCastException: org.apache.drill.exec.expr.holders.BitHolder cannot be cast to org.apache.drill.exec.expr.holders.NullableBitHolder

The cause of the problem is that PruneScanRule assumes the output type of a filter condition is NullableBit, while in this case the filter condition is Bit type, which leads to ClassCastException.