I have a batch job that runs thousands of Deep Clone commands, it uses a ForEach task to run multiple Deep Clones in parallel. It was taking a very long time and I realized that the Driver was the main culprit since it was using up all of its memory a few minutes into the job. I increased the driver size and used a node type with a lot more memory. That improved performance significantly but the driver would still inevitably run out of memory and hit the same bottleneck, even now that it had 128GB of RAM.

You can see the incremental increase in memory utilization as the job progresses here:

By the end of the job, the Driver is using over 122GB of RAM, which seems excessive when all it's doing is running SQL Deep Clone commands without collecting any data.

What could cause so much bloat in this situation? And is there a way to avoid this from the start, or catch and remedy it during the job?