Pekka Enberg

This is a tracking issue for indexing related tasks: - https://github.com/penberg/limbo/issues/125 - https://github.com/penberg/limbo/issues/239 - https://github.com/penberg/limbo/issues/164

Indexing support for write path

1

Add support for updating an index in the write path.

Indexing lookup support

3

Add support for reading from indexes.

Page readahead

2

SQLite benefits from kernel readahead whereas we don't. Let's implement readahead in userspace.

Preparing a statement is slower than with SQLite

5

``` limbo/Prepare statement: 'SELECT 1' time: [1.4930 µs 1.4949 µs 1.4973 µs] thrpt: [667.86 Kelem/s 668.94 Kelem/s 669.81 Kelem/s] change: time: [-1.7708% +0.2245% +1.3879%] (p = 0.84 > 0.05) thrpt:...

Add `cargo-dist` packaging for `limbo` and `liblimbo_sqlite3.a`

1

Let's use https://github.com/axodotdev/cargo-dist to build, package, and release `limbo` the CLI and `liblimbo_sqlite3.a` the SQLite3 compatible library for Linux, Darwin, and Windows.

JSON function support

5

SQLite has a rich set of functions to work with JSON, let's implement it: https://github.com/penberg/limbo/blob/main/docs/sqlite-compat.md#json-functions

Simulator improvements

2

The simulator we have right now is pretty basic. Let's improve it as follows: - [ ] Multiple tenants with their own database, but using same I/O dispatcher. Verify that...

`ORDER BY` is much slower compared to SQLite

1

Limbo uses 14x the CPU: ``` penberg@vonneumann limbo % /usr/bin/time target/release/limbo testing/testing.db "SELECT id FROM users ORDER BY zipcode" 0,15 real 0,14 user 0,01 sys ``` in comparison to SQLite:...

Make I/O completion explicit

6

We currently rely on `run_once()` completing I/O. However, this is not guaranteed to be correct because I/O request can be reordered, for example, by the kernel. We should keep track...