[
https://issues.apache.org/jira/browse/HIVE-20524?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Matt McCline updated HIVE-20524:
--------------------------------
Attachment: HIVE-20524.04.patch
> Schema Evolution checking is broken in going from Hive version 2 to version 3
> for ALTER TABLE VARCHAR to DECIMAL
> ----------------------------------------------------------------------------------------------------------------
>
> Key: HIVE-20524
> URL: https://issues.apache.org/jira/browse/HIVE-20524
> Project: Hive
> Issue Type: Bug
> Components: Hive
> Reporter: Matt McCline
> Assignee: Matt McCline
> Priority: Critical
> Attachments: HIVE-20524.01.patch, HIVE-20524.02.patch,
> HIVE-20524.03.patch, HIVE-20524.04.patch
>
>
> Issue that started this JIRA:
> {code}
> create external table varchar_decimal (c1 varchar(25));
> alter table varchar_decimal change c1 c1 decimal(31,0);
> ERROR : FAILED: Execution Error, return code 1 from
> org.apache.hadoop.hive.ql.exec.DDLTask. Unable to alter table. The following
> columns have types incompatible with the existing columns in their respective
> positions :
> c1
> {code}
> There appear to be 2 issues here:
> 1) When hive.metastore.disallow.incompatible.col.type.changes is true (the
> default) we only allow StringFamily (STRING, CHAR, VARCHAR) conversion to a
> number that can hold the largest numbers. The theory being we don't want
> data loss you would get by converting the StringFamily field into integers,
> etc. In Hive version 2 the hierarchy of numbers had DECIMAL at the top. At
> some point during Hive version 2 we realized this was incorrect and put
> DOUBLE the top.
> However, the Hive2 Hive version 2 TypeInfoUtils.implicitConversion method
> allows StringFamily to either DOUBLE or DECIMAL conversion.
> The new org.apache.hadoop.hive.metastore.ColumnType class under Hive version
> 3 hive-standalone-metadata-server method checkColTypeChangeCompatible only
> allows DOUBLE.
> This JIRA fixes that problem.
> 2) Also, the checkColTypeChangeCompatible method lost a version 2 series bug
> fix that drops CHAR/VARCHAR (and DECIMAL I think) type decorations when
> checking for Schema Evolution compatibility. So, when that code is checking
> if a data type "varchar(25)" is StringFamily it fails because the "(25)"
> didn't get removed properly.
> This JIRA fixes issue #2 also.
> NOTE: Hive1 version 2 did undecoratedTypeName(oldType) and Hive2 version
> performed the logic in TypeInfoUtils.implicitConvertible on the
> PrimitiveCategory not the raw type string.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
