Thank you for reporting this Bjørn.  I found this Drill ticket

https://issues.apache.org/jira/browse/DRILL-6670

which contains the following in the comments
This change aims at restoring original functionality - handling `TIMESTAMP_MICROS` as `INT64` with no logical type in both Parquet readers. It doesn't seem to make sense to do more since `TIMESTAMP_MICROS` is deprecated logical type as per Parquet [current documentation](https://github.com/apache/parquet-format/blob/master/LogicalTypes.md <https://github.com/apache/parquet-format/blob/master/LogicalTypes.md>).

on the parquet-format page I found
|TIMESTAMP_MICROS| is the deprecated ConvertedType counterpart of a |TIMESTAMP| logical type that is UTC normalized and has |MICROS| precision. Like the logical type counterpart, it must annotate an |int64|.

So I guess we won't ever add TIMESTAMP_MICROS but we would certainly want to support TIMESTAMP with MICROS precision (I don't know our current status there).

On 2021/10/06 17:27, Bjørn Jørgensen wrote:
Hi I did fill a bug in Apache spark for problems with datetime columns.

Looks like Apache Drill only implements TIMESTAMP_MILLIS in Parquet.
TIMESTAMP_MICROS is also Parquet standard but looks like the read path for
this type seems missing in Drill.

The bug report
https://issues.apache.org/jira/browse/SPARK-36934


Reply via email to