You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@drill.apache.org by "benj (JIRA)" <ji...@apache.org> on 2019/03/14 16:07:00 UTC
[jira] [Updated] (DRILL-7104) Change of data type when parquet with
multiple fragment
[ https://issues.apache.org/jira/browse/DRILL-7104?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
benj updated DRILL-7104:
------------------------
Description:
When creating a Parquet with a column filled only with "CAST(NULL AS VARCHAR)", if the parquet has several fragment, the type is read like INT instead of VARCHAR.
First, create +Parquet with only one fragment+ - all is fine (the type of "demo" is correct).
{code:java}
CREATE TABLE ....`nobug` AS
(SELECT CAST(NULL AS VARCHAR) AS demo
, md5(cast(rand() AS VARCHAR) AS jam
FROM ....`onebigfile` LIMIT 1000000));
+-----------+----------------------------+
| Fragment | Number of records written |
+-----------+----------------------------+
| 0_0 | 10000000 |
SELECT drilltypeof(demo) AS goodtype FROM ....`bug` LIMIT 1;
+--------------------+
| goodtype |
+--------------------+
| VARCHAR |
{code}
Second, create +Parquet with at least 2 fragments+ - the type of "demo" change to INT
{code:java}
CREATE TABLE ....`bug` AS
((SELECT CAST(NULL AS VARCHAR) AS demo
,md5(CAST(rand() AS VARCHAR)) AS jam
FROM ....`onebigfile` LIMIT 1000000)
UNION
(SELECT CAST(NULL AS VARCHAR) AS demo
,md5(CAST(rand() AS VARCHAR)) AS jam
FROM ....`onebigfile` LIMIT 1000000));
+-----------+----------------------------+
| Fragment | Number of records written |
+-----------+----------------------------+
| 1_1 | 1000276 |
| 1_0 | 999724 |
SELECT drilltypeof(demo) AS badtype FROM ....`bug` LIMIT 1;
+--------------------+
| badtype |
+--------------------+
| INT |{code}
The change of type is really terrible...
was:
When creating a Parquet with a column filled only with "CAST(NULL AS VARCHAR)", if the parquet has several fragment, the type is read like INT instead of VARCHAR.
First, create +Parquet with only one fragment+ - all is fine (the type of "demo" is correct).
{code:java}
CREATE TABLE ....`bug` AS
(SELECT CAST(NULL AS VARCHAR) AS demo
, md5(cast(rand() AS VARCHAR) AS jam
FROM ....`onebigfile` LIMIT 1000000));
+-----------+----------------------------+
| Fragment | Number of records written |
+-----------+----------------------------+
| 0_0 | 10000000 |
SELECT drilltypeof(demo) AS goodtype FROM ....`bug` LIMIT 1;
+--------------------+
| goodtype |
+--------------------+
| VARCHAR |
{code}
Second, create +Parquet with at least 2 fragments+ - the type of "demo" change to INT
{code:java}
CREATE TABLE ....`bug` AS
((SELECT CAST(NULL AS VARCHAR) AS demo
,md5(CAST(rand() AS VARCHAR)) AS jam
FROM ....`onebigfile` LIMIT 1000000)
UNION
(SELECT CAST(NULL AS VARCHAR) AS demo
,md5(CAST(rand() AS VARCHAR)) AS jam
FROM ....`onebigfile` LIMIT 1000000));
+-----------+----------------------------+
| Fragment | Number of records written |
+-----------+----------------------------+
| 1_1 | 1000276 |
| 1_0 | 999724 |
SELECT drilltypeof(demo) AS badtype FROM ....`bug` LIMIT 1;
+--------------------+
| badtype |
+--------------------+
| INT |{code}
The change of type is really terrible...
> Change of data type when parquet with multiple fragment
> -------------------------------------------------------
>
> Key: DRILL-7104
> URL: https://issues.apache.org/jira/browse/DRILL-7104
> Project: Apache Drill
> Issue Type: Bug
> Components: Storage - Parquet
> Affects Versions: 1.15.0
> Reporter: benj
> Priority: Major
>
> When creating a Parquet with a column filled only with "CAST(NULL AS VARCHAR)", if the parquet has several fragment, the type is read like INT instead of VARCHAR.
> First, create +Parquet with only one fragment+ - all is fine (the type of "demo" is correct).
> {code:java}
> CREATE TABLE ....`nobug` AS
> (SELECT CAST(NULL AS VARCHAR) AS demo
> , md5(cast(rand() AS VARCHAR) AS jam
> FROM ....`onebigfile` LIMIT 1000000));
> +-----------+----------------------------+
> | Fragment | Number of records written |
> +-----------+----------------------------+
> | 0_0 | 10000000 |
> SELECT drilltypeof(demo) AS goodtype FROM ....`bug` LIMIT 1;
> +--------------------+
> | goodtype |
> +--------------------+
> | VARCHAR |
> {code}
> Second, create +Parquet with at least 2 fragments+ - the type of "demo" change to INT
> {code:java}
> CREATE TABLE ....`bug` AS
> ((SELECT CAST(NULL AS VARCHAR) AS demo
> ,md5(CAST(rand() AS VARCHAR)) AS jam
> FROM ....`onebigfile` LIMIT 1000000)
> UNION
> (SELECT CAST(NULL AS VARCHAR) AS demo
> ,md5(CAST(rand() AS VARCHAR)) AS jam
> FROM ....`onebigfile` LIMIT 1000000));
> +-----------+----------------------------+
> | Fragment | Number of records written |
> +-----------+----------------------------+
> | 1_1 | 1000276 |
> | 1_0 | 999724 |
> SELECT drilltypeof(demo) AS badtype FROM ....`bug` LIMIT 1;
> +--------------------+
> | badtype |
> +--------------------+
> | INT |{code}
> The change of type is really terrible...
>
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)