While I’m using tgt+lvm as cinder backend, it’s possible that the volume goes into error-deleting after delete the volume.

Log shows

Stderr: 'device-mapper: remove ioctl failed: Device or resource busy\nCommand failed\n'

But how does it stay in ‘busy’ whereas I’ve already deasscoiated the volume from the instance?

tgtadm --lld iscsi --mode target --op show

Tells that it remains some connections on the lun.

Remove the target with

tgtadm --lld iscsi --mode target --op delete --force --tid x

Remove the lvm with

lvremove /dev/mapper/volumes-xxxxxx

Delete database record with

mysql -h controller -u cinder -p cinder_pass -e "update volumes set deleted = 1, deleted_at = now() where id = 'xxxxxxx'"

Finally, cinder list --all-tenant will show clean list :).