@Claudio Beretta​  you're likely hitting this Spark SQL bug:https://issues.apache.org/jira/browse/SPARK-4987

it's been resolved in Spark 1.3 and will be picked up by Databricks Cloud in the next few weeks.

thanks for reporting this. possible workaround is to convert to Unix epoch time using to_unixtime(), from_unixtime().

sorry for the inconvenience.

-Chris