alopukhov opened a new issue, #3835:
URL: https://github.com/apache/parquet-java/issues/3835
### Describe the bug, including details regarding any error messages,
version, and platform.
After updating to parquet 1.18.1 our team find out that parquet produced by
new writer can not be read by Hive 3.1.2 (which uses older parquet version).
Selecting simple parquet with double column show following error:
```
java.io.IOException: java.lang.IllegalStateException: Group type [message
spark_schema { required int64 id; required double val; } ] does not contain
requested field: required double val
```
Using older parquet library shows that new parquet's metadata ColumnOrder is
parsed as UNDEFINED (as expected, new order is not known until 1.18).
The problem is in Hive's class
https://github.com/apache/hive/blob/744271738e2d89f94156c41bc307efaa57a06b7e/ql/src/java/org/apache/hadoop/hive/ql/io/parquet/convert/HiveStructConverter.java#L110
It match columns by full schema.Type searching column with default column
order.
I assume that just bumping parquet version in hive deployment wouldn't solve
this issue. In such case hive can't read parquet produced by older reader.
There is a workaround for hive queries:
SET hive.fetch.task.conversion=none
Please consider making ieee 754 order as opt-in feature taking into account
the widespread use of Hive.
### Component(s)
Core
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
