[SPARK-30993] GenerateUnsafeRowJoiner corrupts the value if the datatype is UDF and its sql type has fixed length

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.3.0, 2.3.1, 2.3.2, 2.3.3, 2.3.4, 2.4.5, 3.0.0
• Fix Version/s: 2.4.6, 3.0.0
• Component/s: SQL
• Labels: correctness
• Assignee: Jungtaek Lim
• Reporter: Jungtaek Lim
• Created: 29/Feb/20 14:11
• Updated: 08/Jun/21 16:45
• Resolved: 02/Mar/20 19:26

Description

This is reported by user mailing list, though the mail thread is regarding suspect of the behavior of mapGroupsWithState.

https://lists.apache.org/thread.html/r08b44a7afac4e4c971633d30b4e5d11bd7c0d6e28180e03b874ea58b%40%3Cuser.spark.apache.org%3E

The actual culprit is, there're a couple of methods which don't handle UDT and it makes GenerateUnsafeRowJoiner to generate incorrect code. Specifically, the issue occurs when the sql type of UDT has fixed length - GenerateUnsafeRowJoiner has the logic to update the offset position for all variable-length data, and due to this bug, UDT field with fixed length is being treated as variable-length data and its value is modified.

Issue Links

• is duplicated by: SPARK-30986 Structured Streaming: mapGroupsWithState UDT serialization does not work
• links to: [Github] Pull Request #32824 (gengliangwang)
• links to: [Github] Pull Request #32824 (gengliangwang)
• links to: GitHub Pull Request #27747
• links to: GitHub Pull Request #27761