Incresql

A database supporting incremental updates of materialized views

Goals

The goal of the project is to explore and experiment with the incremental materialized views space.

Incremental materialized views can be thought of as fulfilling the same use cases as ksqlDB but with more standard sql in something that feels more like a standard relational database.

The use case's in mind are BI dashboarding/reporting as well as being able to support serving per-user data to user facing apps/services.

Project Status

Basic functionality is working with groupbys, joins etc. Project is abandoned, lessons have been learned...

Building/Running from source

To build and/or run incresql from source you will first need the rust toolchain installed, the install instructions can be found at https://rustup.rs/

Once this is done you can use the standard cargo commands to build/test/run.

  # Run Incresql(dev build)
  cargo run

  # Run Incresql(release build)
  cargo run --release

  # Test Incresql
  cargo test --workspace --tests

  # Build Incresql
  cargo build --release

Connecting

To connect to incresql first start it by running

  # Running from source
  cargo run --release

  # OR

  # Running from release
  ./incresql

And then in another terminal (tab) run(assuming you have the mysql client installed):

  mysql -h 127.0.0.1 -P3307

  mysql> select 1+2;

Developing

Before checking in all tests need to pass, the code needs to be formatted and lints need to pass.

  cargo test --workspace --tests
  cargo fmt --all
  cargo clippy --all

Benchmarking

To benchmark run the following command

  cargo run --release --example tpch

  # Or with a larger scale factor
  cargo run --release --example tpch -- -s 10

  # Or to skip the reset/loading phase and rerun with
  # the tables loaded from the previous run
  cargo run --release --example tpch -- --skipload

To manually run queries against the loaded benchmarking tables, run the following

cargo run  --release -- --directory target/benchmark_db

Related Work

Other similar projects are

• https://github.com/mit-pdos/noria - More academic in nature, focusing more on traditional transactional apps used prepared statements.
• https://github.com/MaterializeInc/materialize - Similar goals but is an in-memory system rather than leaning on rocksdb.