RocksDB and LevelDB use a "log-structured merge tree" so you might want to use that as a search term. There is a section "Papers" at the bottom of the web page

RocksDB and LevelDB use a "log-structured merge tree" so you might want to use that as a search term. There is a section "Papers" at the bottom of the web page at http://leveldb.org/ http://leveldb.org/ . LMDB uses a B-tree variant, according to Wikipedia. Maybe it's a B+ tree as Wikipedia says, but keep your thinking cap on -- it's probably not in exactly the form you see described on Wikipedia. If you want to see a toy LSM-tree implementation, I have one at https://github.com/srh/nihdb https://github.com/srh/nihdb -- it was written as a starter Rust project, not as a fast LSM-tree engine. It's almost the simplest plausible LSM-tree. But you'll miss out on design considerations relating to concurrency and performance.