The error message is sometimes generic "possibly due to an OutOfMemoryError"

There is memory pressure indeed, but try to correlate those graph metrics with the Driver's STDOUT file content and check if the GC/FullGCs are able to work properly and reclaim memory. Also check if there are any other side effects, like CPU contention or anything that could cause or lead the driver to become unresponsive. If the Chauffeur can't heartbeat check with the Driver (because the heartbeat thread is blocked) then it would also lead to the same situation. Just sharing this as additional examples.

Is the Driver unresponsiveness triggered by your application? Check at which point of your application (transformation/action) it is leading the Driver to an unresponsiveness status.