Stoolap: High-performance embedded SQL database in pure Rust

by murat3ok • 7 months ago

113 41 7 months ago

Arcuru• 7 months ago

Time-Travel Queries: Query historical data at any point in time:

The example here looks like it may be storing the full history of transactions? Is that right? That's a pretty high cost to pay for something that's not touted as a marquee feature.

I'm working on a DB[1] that stores full transaction history but it's so that I can support decentralized synchronization. It's in service of my marquee feature so I need to pay the cost of storing history, but I'm surprised that Stoolap also seems to be doing it for a local embedded database.

[1] https://github.com/arcuru/eidetica

rich_sasha2• 7 months ago

I would imagine (but haven't looked at it at all) that it's a byproduct of an append only data format. Then having a historical PoV is cheap - you simply disregard changes after a certain time.

Append-only has many other benefits, including zero contention between many readers and (single) writers. In the vanilla version, writers still contend though.

hansvm• 7 months ago

I think their point is that system timestamps for that append-only format aren't good enough. You need logical timestamps corresponding to increasing transaction ids.

rich_sasha2• 7 months ago

Looks very interesting!

Some comparison to another embedded SQL DB, i.e. sqlite3, would be useful. How abusable is it? What tradeoffs are taken? Etc.

seg_lol• 7 months ago

    Initial release: Stoolap - A Modern Embedded SQL Database in Pure Rust
    
    Stoolap is a high-performance embedded SQL database featuring:
    
    Core Features:
    - Full ACID transactions with MVCC (READ COMMITTED & SNAPSHOT isolation)
    - Cost-based query optimizer with adaptive execution
    - Parallel query execution via Rayon
    - 101+ built-in functions (string, math, date/time, JSON, aggregate, window)
    - Multiple index types: B-tree, Hash, Bitmap (auto-selected or explicit)
    - Multi-column composite indexes
    - WAL + snapshots with crash recovery
    
    SQL Support:
    - JOINs (INNER, LEFT, RIGHT, FULL OUTER, CROSS)
    - Subqueries (scalar, IN, EXISTS, correlated)
    - Common Table Expressions (WITH and WITH RECURSIVE)
    - Window functions (ROW_NUMBER, RANK, LAG, LEAD, etc.)
    - ROLLUP, CUBE, GROUPING SETS
    - Temporal queries (AS OF TIMESTAMP/TRANSACTION)
    - Views, RETURNING clause, ON DUPLICATE KEY UPDATE
    
    104K lines of Rust | No C dependencies | Full documentation at stoolap.io

jdub3• 7 months ago

What an impressive release!

It makes me very curious.

Delivered to GitHub fully-formed: A grand total of 9 commits (mostly docs and CI fixes), all in the last 5 hours, and v0.1.0 released 3 hours ago.

No external database/storage-layer dependencies, so it's not "just" a CLI/server/parser wrapper around other libraries doing the "real work".

It appears to have a substantial test suite (76% code coverage, not skipping the scary bits), and great documentation.

There's a bit of context on https://github.com/stoolap but not much else about the author, project goals, relationship to other systems, e.g. it could be the data layer for something else.

(Interestingly, there's an archived stoolap-go repo with a very similar Go implementation of a columnar/hybrid database, so this is not the author's "first draft".)

forgotpwd163• 7 months ago

Can assume they worked on this last few months when they stopped development in the, now archived, Go attempt, but they scrapped the entire git history on publication. Still, even if consider heavy AI use, looks like they put quite the effort in this.

esafak• 7 months ago

I too am curious how to the first commit came about: https://github.com/stoolap/stoolap/commit/768eb836de0ff072b8...

Note to owner: CI is broken.

murat3ok• 7 months ago

The Go version was my first attempt. Hit some performance walls I couldn't solve cleanly, so I rewrote the whole thing in Rust over the past 6 months. Got about 5x speedup and the concurrency story is way better with ownership.

The git history thing honestly my commits were a mess after months of work. Dead ends, experiments, "fix fix fix" commits. Figured I'd start clean for release. In hindsight, probably should have kept the ugly history looks less suspicious than one big commit.

Goal is basically SQLite but with real MVCC and analytical features (window functions, parallel queries). Something you can embed but that doesn't choke on concurrent writes or complex queries.

Community kill me here but other side thank you for the positive comment here.

ctrust• 7 months ago

Very interesting. Roughly speaking, how does performance compare to SQLite?

jdub3• 7 months ago

Yay, glad you found the discussion (well, the good bits), and thanks for the answer. It's cool work!

Sytten• 7 months ago

In the same area, I am tracking the Rust rewrite of sqlite by Turso [1]. The big advantage is the file format compatibility.

[1] https://github.com/tursodatabase/turso

egorfine5• 7 months ago

It's not a rewrite of SQLite in Rust.

It's an entirely new project that happens to have some compatibility with one of the popular SQL databases, namely SQLite.

forgotpwd163• 7 months ago

From the devs themselves[0]:

Our goal is to build a reimplementation of SQLite from scratch, fully compatible at the language and file format level, with the same or higher reliability SQLite is known for, but with full memory safety and on a new, modern architecture.

And they call it rewrite in a recent followup post[1].

[0] https://turso.tech/blog/introducing-limbo-a-complete-rewrite...

[1] https://turso.tech/blog/we-will-rewrite-sqlite-and-we-are-go...

sudarshnachakra• 7 months ago

Does this support concurrent writers (unlike sqlite)? Quite an impressive feature set for a one-person project.

Also is this a single file DB? If so is the format stable?

edf13• 7 months ago

Sounds very interesting - I’ve used SQLite in a few Rust based projects where performance was the deciding factor… a perf comparison with this would be very useful

dash2• 7 months ago

I think the name is not good. It sounds like "stool app". Among other things, "stool" means poo.

duttish• 7 months ago

Yea, my first association was stool -> poo.

I've been trying to think of what other meaning they could have gone for but got nothing. Stoo lap? Sto olap?

bronlund3• 7 months ago

SQL Tool something something?

bronlund3• 7 months ago

SQL Transactional Objects OnLine Analytical Processing. My best guess so far.

kolektiv• 7 months ago

Another voice basically begging them to change the name here, yeah. It might be quite interesting as a tool, but please...

dominotw• 7 months ago

they are even highlighting a in green after stool to break the word into stool.

i am guessing its a joke?

DoctorOW• 7 months ago

Comments especially feel vibe coded. Not necessarily bad, just not something I would trust with prod data.

    /// Create a new empty row
    pub fn new() -> Self {
        Self { values: Vec::new() }
    }

Klonoar• 7 months ago

This particular bit doesn't scream vibe-coded to me at all.

In fact it looks like a generic comment I'd write and come back to later.

spoiler• 7 months ago

I generally like—and write—these types of doc comments myself. It just looks nicer in docs/intellisense.

I'm a big proponent of "everything public should have a doc comment," even if it's a short sentence. Doesn't hurt to have it. I never understood people who are allergic to comments.

The fact LLMs add comments is one of the few non-sloppy things they do, IMO

bronlund3• 7 months ago

Bold name choice.

skylurk• 7 months ago

I read it as stool lab...

Stoolap: we index your shit

JohnCClarke• 7 months ago

Excited for this! A couple of questions:

1. What is the resolution of timestamps (milli-, micro-, nano-seconds)? 2. Any plans for supporting large data BLOBs (e.g. PostgreSQL TOAST)? This would open up a lot of use cases and would be really interesting to make compatible with the in-memory emphasis for the atomic data types.

kekqqq• 7 months ago

The project is very new, with two days of unique days with commits and 11 commits in its history. I would bet it is vibecoded.

jdub3• 7 months ago

Don't let "AI" make you jump at shadows. Maybe, but probably not.

The first commit was pretty fully-formed, which without "AI" glasses on just means someone did a whole bunch of work before exposing/releasing their work.

andrewl• 7 months ago

As a big fan, and user, of SQLite, this looks like something to watch. And I agree with the comments about the unfortunate name. Just yesterday there was a post here about bad names for software:

https://news.ycombinator.com/item?id=46234806

kiliancs• 7 months ago

I would be interested in seeing numbers backing the high performance claims.

GlacierFox• 7 months ago

Stoolap? Sounds disgusting.

riku_iki• 7 months ago

Any benchmarks to compare to sqlite and pg?

Stoolap: High-performance embedded SQL database in pure Rust

Related Stories