CEPH OSD latency increased on new added nodes if cluster uses customized bluestore_rocksdb_options

Issue

After adding new OSD nodes to existing cluster, latency and IOPS increased on these nodes:

• The global cluster latency increased
• Latency increased on new nodes only
• IOPS increased on those new nodes for no specific reason
• Latency goes back to normal after restart of the OSD processes, but then ramps up again until the next restart