Currently the subtree lock cleaner only removes the locks when it detects that a namenode has died. This will not work in a scenario where the cluster is shutdown and some subtree locks are left behind. In such a case the flag set on a inode will be laziy removed, i.e., when an inode is read with STO flag belonging to a dead node then the flag is removed. However, this does not clean the "hdfs_on_going_subtree_ops" table. This jira fixes the lazily removal of locks in the "hdfs_on_going_subtree_ops" table.