Datasets V2

Datasets V3 is the current repository structure Kart uses, and is a good starting point for learning about Datasets V2. Kart 0.10 continues to support Datasets V2, but all newly created repos will use Datasets V3.


A V2 dataset is exactly like a V3 dataset, except:

  • The folder that contains the entire dataset is called .sno-dataset instead of .table-dataset.

  • Attaching a particular path structure to the dataset within the path-structure.json meta item is not supported - instead, all V2 datasets use the same path structure, known as the legacy path structure.

Legacy path-structure

The legacy path structure information isn’t written to a path-structure.json file, but if it was, it would look as follows:

  "scheme": "msgpack/hash",
  "branches": 256,
  "levels": 2,
  "encoding": "hex"

This means that every feature path looks something like the following:


This example is for a feature with one primary key column only, and a primary key value of [77].

To generate the path to the file:

[77] -> MessagePack -> bytes([0x91, 0x4d]) -> SHA256 -> bytes([0x3c, 0x57, 0x8e, 0x75, ...]) -> hex encode first two bytes as a 2-level path -> 3c/57

To generate the filename:

[77] -> MessagePack -> bytes([0x91, 0x4d]) -> Base64 -> kU0=