After merging this PR (#4578), we can implement distributed construction of btree indexes, which can significantly shorten the index create time of very large datasets and make point query capabilities available on very large data sets.