Corda nodes store all flow checkpoints in memory
If there are too many flow checkpoints, the node crashes. Additionally, flows that fail due to memory errors do not necessarily reach the flow hospital
Upon restarting, the node tries to load all suspended flows into memory again. As a result, it will re-encounter a memory exception, and fail to start.
The node should be modified to offload suspended flows to the DB instead, and to guarantee that flows failing due to memory errors reach the flow hospital.