created_at2020-02-10 15:14:40.671738
updated_at2020-03-11 11:32:43.092577
descriptionCore CQL Database functionality - a lightweight array-based database




Lightweight, growable, array-based storage solution, currently with the out-of-the-box storage types below (custom types also possible):

  • U64 (unsigned 64 integers)
  • F64 (64-bit floating point)
  • NullableF64 (nullable 64-bit floating point)
  • TinyText (255 char utf-8 strings)

The project works by treating the file system as an N dimensional array, removing the need to scan for items in order to find them. Currently the number of dimensions must be specified on create of the database, however each dimension (bar the last) may grow on demand.

The project was originally built with an eye on storing large volumes of relational time series data, however I am looking to explore other uses in my other projects.

Database is one indexed.

Project structure

The project is split into two core sub-projects, cql_db and cql_model, and a sub-project per supported type within the cql_storage_types folder.

The cql_db sub-project contains the core logic orchestrating the type specific logic, and the array-based logic allowing the whole thing to function as a database. It is dependent on cql_model and the U64 type.

The cql_model sub-project contains the interfaces consumed by cql_db and the storage types, and is referenced by all sub-projects.

The storage type specific projects contains type specific code used for read/writting specific types from a file, implementing the interfaces within the cql_model sub-project.

To use this project you'll need to import the cql_db sub-project and either each of the storage types that you with to use, or the cql_model and you own implementations of the traits within - should you wish to use other types.

Rustdocs (with examples) and crates for all published components can be found in the table below:

Repo link Crate Documentation Description
CQL Db crates.io docs.rs Core CQL database engine
CQL Model crates.io docs.rs Core CQL database models/interfaces
U64 crates.io docs.rs Unsigned 64-bit interger storage support
F64 unpublished unpublished 64-bit floating point storage support
NullableF64 crates.io docs.rs Nullable 64-bit floating point storage support
TinyText crates.io docs.rs 255 char utf-8 string storage support

Breaking changes

As this project is a data-storage solution acting upon the file system, some changes may alter the expected structure of the underlying data - preventing one version from correctly utilising a database created on a different version. Up until version 1.0.0, the minor version (middle number) will be incremented and the breaking version listed in the table below. If you need to upgrade between one of these versions, I'd suggest reading your entire database from the earlier version and the writing it to a new database with the new target version, I will try and improve this at somepoint. Please take care.

Crate Breaking version Description
cql_db 0.2.0 Changes made in the key files, and the database file itself. Commits 048e533 and 7dcaf7c

Getting started

To get started, pick a storage type(s) (examples use NullableF64), and add it as a dependency to your Cargo.toml, along with the core cql_db crate:

//... (any existing dependencies you may have)
cql_db = "^0.2"
cql_nullable_f64 = "^0.2"

You then need to create a folder where you want the database to live, and then try out the below:

use std::io;
use std::io::{ Cursor, SeekFrom, Seek };
use cql_nullable_f64::{ NullableF64, unpack_stream };

const N_VALUES_TO_READ: usize = 3;

pub fn example_cql() -> io::Result<()> {
    let base_point = [1];
    let value1 = Some(-1.6);
    let value3 = Some(5.4);

    // creates a one dimensional database, with a capacity of 3

    // writes Some(-1.6) to [0]

    // writes Some(5.4) to [2]
        &[base_point[0] + 2],

    let mut result: [Option<f64>; N_VALUES_TO_READ] = [None; N_VALUES_TO_READ];
    let mut stream = Cursor::new(Vec::new());

    // reads 3 points from [0] into `stream`
        &mut stream,
        N_VALUES_TO_READ as u64

    // returns to the start of the stream
    // unpacks the stream value by value into the result[]
    unpack_stream(&mut stream, N_VALUES_TO_READ, |idx, value| {
        result[idx] = value

    assert_eq!(result[0], value1);
    assert_eq!(result[1], None);
    assert_eq!(result[2], value3);

More examples can be found in the rustdocs.


Benchmarks supplied below for the U64 type and are fairly rudimentary (and rounded) and are there to give a rough idea of relative costs. Full benchmark code can be found in github and can be run with rustup run nightly cargo bench. Benchmarks for other types can be found in the the type's corresponding documentation.

Operation Database dimensions Mean time _unchecked (ns) Mean time (ns)
Single point read 1 2 450 (+/- 300) 7 500 (+/- 600)
Single point read 4 14 850 (+/- 1 000) 37 550 (+/- 2 300)
Single point write 1 2 800 (+/- 400) 7 700 (+/- 400)
Single point write 4 15 400 (+/- 2 500) 37 700 (+/- 3 000)
Stream read 1 point 1 2 500 (+/- 300) 10 000 (+/- 850)
Stream read 1 point 4 14 900 (+/- 600) 42 500 (+/- 6 500)
Stream read 50 000 points 1 27 650 000 (+/- 31 000) 27 630 000 (+/- 180 000)
Stream read 50 000 points 4 27 660 000 (+/- 1 200 000) 27 620 000 (+/- 480 000)


Licensed under either of

at your option.


Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Commit count: 420

cargo fmt