[
https://issues.apache.org/jira/browse/HIVE-10642?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14532917#comment-14532917
]
Sergio Peña commented on HIVE-10642:
------------------------------------
Hi [~dongc]
Here're some comments:
- Could you write a better name for ETIMESTAMP_CONVERTER_NEW? What if we use
this:
- Rename ETIMESTAMP_CONVERTER_NEW for ETIMESTAMP_CONVERTER for the new
timestamp
support. This name makes more sense for the new data type.
- Rename ETIMESTAMP_CONVERTER for ETIMESTAMP_INT96_CONVERTER for the old
timestamp
support. This will let developers know that this timestamp is a int96 data
type.
- Can we use a SWITCH statement instead of IF/ELSE? Just to make the code clean.
Note: you need to use TIMESTAMP_MILLIS on the case statement instead of
OriginalType.TIMESTAMP_MILLIS in order to compile.
And regarding the new support. Is this new timestamp data type making the
values less precise? I saw that {{01:01:01.111111111}} was changes to
{{01:01:01.111}}.
We need to confirm that this won't affect users already using the more precise
timestamp value.
> Update Timestamp type mapping to Parquet
> -----------------------------------------
>
> Key: HIVE-10642
> URL: https://issues.apache.org/jira/browse/HIVE-10642
> Project: Hive
> Issue Type: Sub-task
> Reporter: Dong Chen
> Assignee: Dong Chen
> Attachments: HIVE-10642-parquet.patch
>
>
> Timestamp in Hive was originally mapped to INT_96 in Parquet before. And
> inside Parquet it use Binary to save values.
> When working on HIVE-10255, I found predicate with timestamp does not work,
> since comparing Binary representative of Timestamp value can not give correct
> relationship.
> Meanwhile, in latest version Parquet, the timestamp is mapped to INT_64.
> Refer to https://github.com/Parquet/parquet-format/blob/master/LogicalTypes.md
> Therefore, this Jira will update the mapping of Timestamp type in Hive, and
> keep it backward compatible.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
