[ https://issues.apache.org/jira/browse/DRILL-4783?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15380422#comment-15380422 ]
Chunhui Shi commented on DRILL-4783: ------------------------------------ Flatten operator should be able to initialize ValueVector appropriately even underlying operator (in this case it is Project to retrieve t1.`json`.`list`) could not initialize ValueVector correctly since without data underlying operator has no idea what t1.`json`.`list` is. I propose to fix it by modifying FlattenRecordBatch.java when there is ClassCastException thrown due to failing to cast ValueVector(NullInt in this example) to RepeatedValueVector, new a RepeatedMapVector instead. > Flatten on CONVERT_FROM fails with ClassCastException if resultset is empty > --------------------------------------------------------------------------- > > Key: DRILL-4783 > URL: https://issues.apache.org/jira/browse/DRILL-4783 > Project: Apache Drill > Issue Type: Bug > Reporter: Chunhui Shi > Assignee: Chunhui Shi > Priority: Critical > > Flatten failed to work on top of convert_from when the resultset is empty. > For a HBase table like this: > 0: jdbc:drill:zk=localhost:5181> select convert_from(t.address.cities,'json') > from hbase.`/tmp/flattentest` t; > +----------------------------------------------------------------------------------+ > | EXPR$0 > | > +----------------------------------------------------------------------------------+ > | {"list":[{"city":"SunnyVale"},{"city":"Palo Alto"},{"city":"Mountain > View"}]} | > | {"list":[{"city":"Seattle"},{"city":"Bellevue"},{"city":"Renton"}]} > | > | {"list":[{"city":"Minneapolis"},{"city":"Falcon Heights"},{"city":"San > Paul"}]} | > +----------------------------------------------------------------------------------+ > Flatten works when row_key is in (1,2,3) > 0: jdbc:drill:zk=localhost:5181> select flatten(t1.json.list) from (select > convert_from(t.address.cities,'json') json from hbase.`/tmp/flattentest` t > where row_key=1) t1; > +---------------------------+ > | EXPR$0 | > +---------------------------+ > | {"city":"SunnyVale"} | > | {"city":"Palo Alto"} | > | {"city":"Mountain View"} | > +---------------------------+ > But Flatten throws exception if the resultset is empty > 0: jdbc:drill:zk=localhost:5181> select flatten(t1.json.list) from (select > convert_from(t.address.cities,'json') json from hbase.`/tmp/flattentest` t > where row_key=4) t1; > Error: SYSTEM ERROR: ClassCastException: Cannot cast > org.apache.drill.exec.vector.NullableIntVector to > org.apache.drill.exec.vector.complex.RepeatedValueVector > Fragment 0:0 > [Error Id: 07fd0cab-d1e6-4259-bfec-ad80f02d93a2 on atsqa4-127.qa.lab:31010] > (state=,code=0) -- This message was sent by Atlassian JIRA (v6.3.4#6332)