File directory structure and naming convention for storing columnar tables

Abstract

A database system stores a table as a set of column files in a columnar format in a manner that improves the write performance of the table and avoids use of separate metadata repository. In embodiments, each column file groups values into entity chunks indexed by an entity index. Each chunk includes a live value index that determines which rows in chunk has live values. New values are written to the column file by appending an updated copy of the entity chunk. The entity index to refer to the newly written chunk as the latest version. This approach avoids expensive in-place updating of individual column values and allows the update to be performed much more quickly. In embodiments, the database system encodes metadata such as table schema information using file naming and placement conventions in the file store, so that a centralized metadata repository is not required.