The following packages would be helpful:
From the way how local recovery works with incremental RocksDB checkpoints, I would not assume that it is the cause of the problem. In this particular case, the number of opened files on a local FS should not be higher than the number without local recovery. Maybe it is just a matter of the OS limit and the number of operators with a RocksDB backend running on the machine and the amount of files managed by all those RocksDB instances that simply exceed the limit. If you have an overview how many parallel operator instances with keyed state were running on the machine and assume some reasonable number of files per RocksDB instance and the limit configured in your OS, could that be the case?
Thanks for your help!