[DRILL-5970] DrillParquetReader always builds the schema with "OPTIONAL" dataMode columns instead of "REQUIRED" ones

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.11.0
• Fix Version/s: Future
• Component/s: Execution - Codegen, Execution - Data Types, Storage - Parquet
• Assignee: Vitalii Diravka
• Reporter: Vitalii Diravka
• Reviewer: Salim Achouche
• Created: 15/Nov/17 17:41
• Updated: 30/Oct/18 11:37

Description

The root cause of the issue is that adding REQUIRED (not-nullable) data types to the container in the all MapWriters is not implemented.

It can lead to get invalid schema.

0: jdbc:drill:zk=local> CREATE TABLE dfs.tmp.bof_repro_1 as select * from (select CONVERT_FROM('["hello","hai"]','JSON') AS MYCOL, 'Bucket1' AS Bucket FROM (VALUES(1)));
SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further details.
+-----------+----------------------------+
| Fragment  | Number of records written  |
+-----------+----------------------------+
| 0_0       | 1                          |
+-----------+----------------------------+
1 row selected (2.376 seconds)

Run from Drill unit test framework (to see "data mode"):

@Test
  public void test() throws Exception {
    setColumnWidths(new int[] {25, 25});
    List<QueryDataBatch> queryDataBatches = testSqlWithResults("select * from dfs.tmp.bof_repro_1");
    printResult(queryDataBatches);
  }

1 row(s):
-------------------------------------------------------
| MYCOL<VARCHAR(REPEATED)> | Bucket<VARCHAR(OPTIONAL)>|
-------------------------------------------------------
| ["hello","hai"]          | Bucket1                  |
-------------------------------------------------------
Total record count: 1

vitalii@vitalii-pc:~/parquet-tools/parquet-mr/parquet-tools/target$ java -jar parquet-tools-1.6.0rc3-SNAPSHOT.jar schema /tmp/bof_repro_1/0_0_0.parquet 
message root {
  repeated binary MYCOL (UTF8);
  required binary Bucket (UTF8);
}

To simulate of obtaining the wrong result you can try the query with aggregation by using a new parquet reader (used by default for complex data types) and old parquet reader. False "Hash aggregate does not support schema changes" error will happen.

1) Create two parquet files.

0: jdbc:drill:schema=dfs> CREATE TABLE dfs.tmp.bof_repro_1 as select * from (select CONVERT_FROM('["hello","hai"]','JSON') AS MYCOL, 'Bucket1' AS Bucket FROM (VALUES(1)));
+-----------+----------------------------+
| Fragment  | Number of records written  |
+-----------+----------------------------+
| 0_0       | 1                          |
+-----------+----------------------------+
1 row selected (1.122 seconds)
0: jdbc:drill:schema=dfs> CREATE TABLE dfs.tmp.bof_repro_2 as select * from (select CONVERT_FROM('[]','JSON') AS MYCOL, 'Bucket1' AS Bucket FROM (VALUES(1)));
+-----------+----------------------------+
| Fragment  | Number of records written  |
+-----------+----------------------------+
| 0_0       | 1                          |
+-----------+----------------------------+
1 row selected (0.552 seconds)
0: jdbc:drill:schema=dfs> select * from dfs.tmp.bof_repro_2;

2) Copy the parquet files from bof_repro_1 to bof_repro_2.

[root@naravm1 ~]# hadoop fs -ls /tmp/bof_repro_1
Found 1 items
-rw-r--r--   3 mapr mapr        415 2017-07-25 11:46 /tmp/bof_repro_1/0_0_0.parquet
[root@naravm1 ~]# hadoop fs -ls /tmp/bof_repro_2
Found 1 items
-rw-r--r--   3 mapr mapr        368 2017-07-25 11:46 /tmp/bof_repro_2/0_0_0.parquet
[root@naravm1 ~]# hadoop fs -cp /tmp/bof_repro_1/0_0_0.parquet /tmp/bof_repro_2/0_0_1.parquet
[root@naravm1 ~]#

3) Query the table.

0: jdbc:drill:schema=dfs> ALTER SESSION SET  `planner.enable_streamagg`=false;
+-------+------------------------------------+
|  ok   |              summary               |
+-------+------------------------------------+
| true  | planner.enable_streamagg updated.  |
+-------+------------------------------------+
1 row selected (0.124 seconds)
0: jdbc:drill:schema=dfs> select * from dfs.tmp.bof_repro_2;
+------------------+----------+
|      MYCOL       |  Bucket  |
+------------------+----------+
| ["hello","hai"]  | Bucket1  |
| null             | Bucket1  |
+------------------+----------+
2 rows selected (0.247 seconds)
0: jdbc:drill:schema=dfs> select bucket, count(*) from dfs.tmp.bof_repro_2 group by bucket;
Error: UNSUPPORTED_OPERATION ERROR: Hash aggregate does not support schema changes

Fragment 0:0

[Error Id: 60f8ada3-5f00-4413-a676-4881fc8cb409 on naravm3:31010] (state=,code=0)

Issue Links

• links to: GitHub Pull Request #1047