Datalog in Rust
github.com
github.com
I'm in the middle of putting together a realtime strategy game using Differential Datalog[1] and Rust, with DDL managing the game's logic. Mostly as an excuse to expose myself to new ideas and engage in a whole lot of yak shaving.
I'll be interested in reading how this goes!
The narrator's interjections were a great touch. It's rare to see a post that is this technically deep but also so fun to read. The journey through optimizing the aliasing query felt like a detective story. We, the readers, were right there with you, groaning at the 50GB memory usage and cheering when you got it down to 5GB.
Fantastic work, both on the code and the prose.
It is slow going, partly since it is not a priority, partly because I suffer from second system syndrome. Mangle Rust should deal with any size data through getting and writing facts to disk via memory mapping. The golang implementation is in-memory.
This post is nice because it parses datalog and mentions the LSM tree, and much easier to follow than the data frog stuff.
There are very many datalog implementations in Rust (ascent, crepe) that use proc-macros. The downside is that they won't handle getting queries at runtime. For the static analysis use case where queries/programs are fixed, the proc macro approach might be better.
I'm encouraged by the other commenters sharing their recent Datalog projects. I am currently building a set of data quality pipelines for a legacy SQL database in preparation of a huge software migration.
We find Datalog much more useful in identifying and looking for data quality issues thatn SQL, as the queries can be incredibly readable when well-structured.
All of this is to say, I think Datalog 2.0's sparse attendance this year may be more indicative of the fact that it is a satellite workshop of an already-lesser-prestigious conference (itself not even the main event! That was ICLP!) rather than a lack of Datalog implementation excitement.
For what it's worth, none of what I'm saying is meant to rebut your high-level point that there is little novelty left in implementing raw Datalog engines. Of course I agree, the research space has moved far beyond that (arguably it did a while ago) and into more exotic problems involving things like streaming (HydroFlow), choice (Dusa), things that get closer to the general chase (e.g., Egglog's chase engine), etc. I don't think anyone disagrees that vanilla Datalog is boring, it's just that monotonic, chain-forward saturation (Horn clauses!) are a rich baseline with a well-understood engineering landscape (esp in the high-performance space) to build out more interesting theories (semirings, Z-sets, etc..).
If you were expecting European attendance, that would explain why it's different from the past.
It's the same in science circles, tourism and even business travel.
It's in Go, but you can apply most of it easily in other languages.
I'm so happy that modern languages like Rust, Zig, and Go support unicode/runes/graphemes natively. So many problem just disappear compared to Java, .net, C++, or scripting languages.
- https://github.com/frankmcsherry/blog/blob/master/posts/2025...
what is that Datalog improves over Prolog?
Just now I tried to skim the Wikipedia page of Datalog; the vague theory I'm getting from it, is that maybe Prolog has relatively poor performance, whereas Datalog dramatically improves performance (presumably allowing much bigger datasets and much more parallelized processing), at the cost of reducing expressiveness and features in some other important ways? (including making it no longer Turing-complete?) Is that what it's about, or am I completely missing the mark?
datalog on the other hands is more or less a relation db with a different syntax.
Datalog is a commuter car with a CVT. Prolog is an F1 car. Basically, it's not about improvement. It's about lobotomizing Prolog into something people won't blow their legs off with. Something that's also much easier to implement and embed in another application (though Prologs can be very easy to embed.)
If you're used to Prolog, you'll mostly just find Datalog to be claustrophobic. No call/3? No term/goal expansion? Datalog is basically designed to pull out the LCD featureset of Prolog for use as an interactive database search.
> Prolog has relatively poor performance, whereas Datalog dramatically improves performance
It's easier to write fast Datalog code but the ceiling is also way lower. Prolog can be written in a way to allow for concurrency, but that's an intermediate level task that requires understanding of your implementation. Guarded Horn Clauses and their derived languages[2] were developed to formalize some of that, but Japanese advancements over Prolog are extremely esoteric. Prolog performance really depends on the programmer and the implementation being used and where it's being used. Prolog, like a Lisp, can be used to generate native machine code from a DSL at compile-time.
If you understand how the underlying implementation of your Prolog works, and how to write code with the grain of your implementation, it's absolutely "fast enough". Unfortunately, that requires years of writing Prolog code with a single implementation. There's a lot of work on optimizing[3][4] prolog compilers out there, as well as some proprietary examples[5].
[1] - http://logicprogramming.stanford.edu/readings/ullman.pdf
[2] - https://www.ueda.info.waseda.ac.jp/AITEC_ICOT_ARCHIVES/ICOT/...
[3] - https://www.sciencedirect.com/science/article/pii/S074310669...
[4] - https://link.springer.com/content/pdf/10.1007/3-540-18024-9_...
[5] - https://sicstus.sics.se/
Although there’s no “ANSI SQL” equivalent standard across Datalog implementations, once you get a hang of the core idea it’s not too hard to understand another Datalog.
I started a Percival fork that compiles the Datalog to SQLite, if you want to check out how the two can express the same thing: https://percival.jake.tl/ (unfinished when it comes to aggregates and more advanced joins but the basic forms work okay). Logica is a much more serious / complete Datalog->SQL compiler written by a Google researcher that compiles to BigTable, DuckDB, and a few other SQL dialects (https://logica.dev/).
One area Datalog is an order of magnitude easier is when working with recursive queries / rules; this is possible in SQL but feels a bit like drinking playdough through a straw. Frank’s Materialize.com has a “WITH MUTUALLY RECURSIVE” SQL form (https://materialize.com/blog/recursion-in-materialize/) that’s much nicer than the ancient ANSI SQL recursive approach, we’re evaluating it for page load queries & data sync at Notion.
Feldera has a similar form for recursive views as well (https://www.feldera.com/blog/recursive-sql-queries-in-felder...). I like that Feldera lets you make each “rule” or subview its own statement rather than needing to pack everything into a single huge statement. Main downside I found when testing Feldera is that their SQL dialect has a bunch of limitations inherited from Apache Calcite, the Materialize SQL dialect tries very hard to be PostgresSQL compatible.
At Feldera, we're adding features to our SQL over time, by contributing them upstream to Calcite, making it better for everyone. Mihai Budiu, who is the author of the Feldera SQL compiler, is a Calcite committer.
Most of what I mean is places where Feldera/Calcite has slightly different syntax from Postgres for things. For example, Postgres syntax for cast to bigint is `some_expresion::bigint` although Postgres also supports ANSI SQL `CAST(some_expression AS bigint)`, most examples I find in the wild and in my own Postgres SQL use the Postgres special syntax. JSON syntax also differs; Feldera uses its own pretty elegant `VARIANT` type and `some_expression[key_expression]` to access properties, where Postgres calls this `json` or `jsonb`, and uses `some_expression->key_expression` to access properties. In those cases it's not like Feldera is wrong or lacks some support, but it's a bit harder to work with for me because I'm so used to Postgres syntax and I need to do some regex replace whenever I bring a query from Postgres over to Feldera.
Definitely not a deal-breaker, I am a Feldera enjoyer, but it does add some friction.
An interesting case of a user dealing with this problem: they use LLMs to mass migrate SparkSQL code over to Feldera (it's often json-related constructs as you also ran into). They then verify that both their original warehouse and Feldera compute the same results for the same inputs to ensure correctness.
Recent additions to ANSI SQL have added the capacity for recursion, so it's no longer completely impossible. But they have three big disadvantages:
1. They accidentally made SQL Turing-complete. Datalog queries, by contrast, are guaranteed to terminate.
2. They're still extremely clumsy to use.
3. Because of #1, they're often not implemented fully, so they are hard to rely on.
Last I checked, VMWare had moved away from differential datalog?
They switched from differential Datalog to differential SQL, I think because they realized Datalog is a really tough sell.
It is the only database/query engine that allows you to use the same SQL for both batch and streaming (with UDFs).
I have made an accessible version of a subset of Differential Dataflow (DBSP) in Python right here: https://github.com/brurucy/pydbsp
DBSP is so expressive that I have implemented a fully incremental dynamic datalog engine as a DBSP program.
Think of SQL/Datalog where the query can change in runtime, and the changes themselves (program diffs) are incrementally computed: https://github.com/brurucy/pydbsp/blob/master/notebooks/data...
Flink SQL also checks that box.
There has to be some change in the code, and they will not share the same semantics (and perhaps won't work when retractions/deletions also appear whilst streaming). And let's not even get to the leaky abstractions for good performance (watermarks et al).