[HIVE-8784] Querying partition does not work with JDO enabled against PostgreSQL

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.14.1
• Fix Version/s: 1.0.0
• Component/s: Metastore
• Assignee: Chaoyu Tang
• Reporter: Chaoyu Tang
• Created: 07/Nov/14 18:43
• Updated: 19/Feb/15 18:21
• Resolved: 11/Nov/14 23:47

Description

Querying a partition in PostgreSQL fails when using JDO (with hive.metastore.try.direct.sql=false) . Following is the reproduce example:

create table partition_test_multilevel (key string, value string) partitioned by (level1 string, level2 string, level3 string);

insert overwrite table partition_test_multilevel partition(level1='1111', level2='111', level3='11') select key, value from srcpart tablesample (11 rows);
insert overwrite table partition_test_multilevel partition(level1='1111', level2='222', level3='11') select key, value from srcpart tablesample (15 rows);
insert overwrite table partition_test_multilevel partition(level1='1111', level2='333', level3='11') select key, value from srcpart tablesample (20 rows);

select level1, level2, level3, count(*) from partition_test_multilevel where level2 <= '222' group by level1, level2, level3;

The query fails with following error:

      Caused by: org.apache.hadoop.hive.ql.parse.SemanticException: MetaException(message:Invocation of method "substring" on "StringExpression" requires argument 1 of type "NumericExpression")
	at org.apache.hadoop.hive.ql.optimizer.ppr.PartitionPruner.getPartitionsFromServer(PartitionPruner.java:392)
	at org.apache.hadoop.hive.ql.optimizer.ppr.PartitionPruner.prune(PartitionPruner.java:215)
	at org.apache.hadoop.hive.ql.optimizer.ppr.PartitionPruner.prune(PartitionPruner.java:139)
	at org.apache.hadoop.hive.ql.parse.ParseContext.getPrunedPartitions(ParseContext.java:619)
	at org.apache.hadoop.hive.ql.optimizer.pcr.PcrOpProcFactory$FilterPCR.process(PcrOpProcFactory.java:110)
	... 21 more

It is because the JDO pushdown filter generated for a query having inequality/between partition predicate uses DN indexOf function which is not working properly with postgresql (see http://www.datanucleus.org/servlet/jira/browse/NUCRDBMS-840)