Heya, We're occasionally seeing OSD aborts with something like this in the logs (this is on 17.2.7) bluestore(/var/lib/ceph/osd/ceph-70) operator() #2:61bc4388:::3c5b76bf-81e1-4a1a-bd94-7fee95fa35e2.25380786.89__shadow_.z1wu_6Nt7uzGSnV-cca7mQxgNVyKVUm_22:head# 0x6454c6b7ab00 exists in onode_map bluestore(/var/lib/ceph/osd/ceph-70) _txc_add_transaction error (39) Directory not empty not handled on operation 21 (op 1, counting from 0) _dump_transaction transaction dump: { "ops": [ { "op_num": 0, "op_name": "remove", "collection": "2.186_head", "oid": "#2:61800000::::head#" }, { "op_num": 1, "op_name": "rmcoll", "collection": "2.186_head" } ] } I see tracker issue #49900 which seems to have similar symptoms but is closed. The tracker has two Github PRs listed, #40237 and #40323 PR#40323 is merged, saying this might fix issue 49818, i.e. a different issue. PR#40237 however has been closed due to it being stale, and I wonder if tracker #49900 might have been closed erroneously? Could this be the case? Or might this be a regression? Thanks, peter.