Malthe
Valued Contributor II

Disabling the setting worked for me on 16.4 (serverless):

spark.databricks.streaming.forEachBatch.optimized.fastPath.enabled=false

 But you have to wonder if this is something Databricks are aware of and are looking to fix.