Hi @tsam ,

Can you share few details:

• Which DBR is the job on?
• How many DEEP CLONEs you need to run in total?
• What is the parallelism of the for-each task?
• Are the cloned tables optimized (e.g. there is no "small file problem")?
• Can you share the Heap Histogram of the Driver (can be found in the Spark UI)

In parallel, a simple fix that I can suggest is to run it on the most recent DBR version.

Best regards,