sqoop-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Greg Lindholm (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (SQOOP-3285) Hcatalog import is loosing precision on DECIMAL fields
Date Fri, 09 Feb 2018 21:27:00 GMT

    [ https://issues.apache.org/jira/browse/SQOOP-3285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16358981#comment-16358981
] 

Greg Lindholm edited comment on SQOOP-3285 at 2/9/18 9:26 PM:
--------------------------------------------------------------

Patch is attached [^0001-Fix-DECIMAL-percision-loss-with-Hcatalog-imports.patch]

A couple lines added to org.apache.sqoop.mapreduce.hcat.SqoopHCatImportHelper.convertNumberTypes()

The fix is for first check if the value is a BigDecimal or a BigInteger and if so use the
value to create the HiveDecimal directly to avoid converting it to a Double.

Here is the new logic to convert a Number into a HiveDecimal.
{code:java}
    } else if (hfsType == HCatFieldSchema.Type.DECIMAL) {
      if (n instanceof BigDecimal) {
        return HiveDecimal.create((BigDecimal)n);
      }
      if (n instanceof BigInteger) {
        return HiveDecimal.create((BigInteger)n);
      }
      BigDecimal bd = new BigDecimal(n.doubleValue(),
        MathContext.DECIMAL128);
      return HiveDecimal.create(bd);
    }

{code}
 


was (Author: glindholm):
Patch is attached [^0001-Fix-DECIMAL-percision-loss-with-Hcatalog-imports.patch]

A couple lines added to org.apache.sqoop.mapreduce.hcat.SqoopHCatImportHelper.convertNumberTypes()

The fix is for first check if the value is a BigDecimal or a BigInteger and if so use the
value to create the HiveDecimal directly to avoid converting it to a Double.

Here is the logic to new convert a Number into a HiveDecimal.
{code:java}
    } else if (hfsType == HCatFieldSchema.Type.DECIMAL) {
      if (n instanceof BigDecimal) {
        return HiveDecimal.create((BigDecimal)n);
      }
      if (n instanceof BigInteger) {
        return HiveDecimal.create((BigInteger)n);
      }
      BigDecimal bd = new BigDecimal(n.doubleValue(),
        MathContext.DECIMAL128);
      return HiveDecimal.create(bd);
    }

{code}
 

> Hcatalog import is loosing precision on DECIMAL fields
> ------------------------------------------------------
>
>                 Key: SQOOP-3285
>                 URL: https://issues.apache.org/jira/browse/SQOOP-3285
>             Project: Sqoop
>          Issue Type: Bug
>          Components: hive-integration
>    Affects Versions: 1.4.7
>            Reporter: Greg Lindholm
>            Priority: Major
>         Attachments: 0001-Fix-DECIMAL-percision-loss-with-Hcatalog-imports.patch
>
>
> Import with hcatalog for DECIMAL fields is loosing precision.
> The bug is in org.apache.sqoop.mapreduce.hcat.SqoopHCatImportHelper.convertNumberTypes()
> Internally the BigDecimal value is converted to a Double then back into a BigDecimal.
> As a test I imported from a MySQL db a DECIMAL(38,8) field:
> {code:java}
> create table sampledata1 ( n38 DECIMAL(38,8) );
> insert into sampledata1 ( n38 ) values ( -123456789012345678901234567890.12345678 )
> sqoop import \
>   --connect jdbc:mysql://10.210.144.22:3306/greg --username greg -P \
>   -m 1 \
>   --table sampledata1 \
>   --hcatalog-table sampledata1_orc2 \
>   --create-hcatalog-table \
>   --hcatalog-storage-stanza 'stored as orc'
> {code}
> The result in the hive table does not match the input.
> {noformat}
> Original:
> -123456789012345678901234567890.12345678
> Result:
> -123456789012345677877719597056.00000000{noformat}
> I get the same result with Orc, RCfile, and Avro files.
>  
> I believe I found the problem and will provide a patch shortly.
>  
>  



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message