You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@impala.apache.org by "Zoltán Borók-Nagy (JIRA)" <ji...@apache.org> on 2019/05/06 13:42:00 UTC

[jira] [Created] (IMPALA-8498) Write column index for floating types when NaN is not present

Zoltán Borók-Nagy created IMPALA-8498:
-----------------------------------------

             Summary: Write column index for floating types when NaN is not present
                 Key: IMPALA-8498
                 URL: https://issues.apache.org/jira/browse/IMPALA-8498
             Project: IMPALA
          Issue Type: Bug
            Reporter: Zoltán Borók-Nagy


IMPALA-7304 disabled column index writing for floating point columns until PARQUET-1222 is resolved.

PARQUET-1222 is responsible for defining a total order for floating values, but the problematic values are only the NaNs. Therefore we can write the column index if NaNs are not present in the data. Parquet-MR also does this, following the principles in [https://github.com/apache/parquet-format/blob/75eb7a7b84e6e62bfb09668b6d8d40b12597456e/src/main/thrift/parquet.thrift#L827-L834] 

Impala should follow this behavior, and also when storing zeroes, it should store -0.0 as minimum and +0.0 as maximum.



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