Grigorii Safronov 5e7a6026b9 | 4 weeks ago | |
---|---|---|
.github/workflows | 10 months ago | |
cozo-bin | 4 weeks ago | |
cozo-core | 4 months ago | |
cozo-lib-c | 10 months ago | |
cozo-lib-java | 5 months ago | |
cozo-lib-nodejs | 5 months ago | |
cozo-lib-python | 5 months ago | |
cozo-lib-swift | 5 months ago | |
cozo-lib-wasm | 5 months ago | |
cozorocks | 5 months ago | |
scripts | 4 weeks ago | |
signatures/version1 | 1 year ago | |
.gitignore | 1 year ago | |
.gitmodules | 2 years ago | |
CODE_OF_CONDUCT.md | 2 years ago | |
CONTRIBUTING.md | 2 years ago | |
Cargo.lock | 5 months ago | |
Cargo.toml | 10 months ago | |
LICENSE.txt | 2 years ago | |
Logo.png | 4 weeks ago | |
README.md | 4 weeks ago | |
VERSION | 10 months ago |
README.md
Multimodel (vector, graph, relational) distributed in memory database management system
Table of contents
- About The Project
- Getting Started
- Gratitudes
- Requirements
- Building
- Install
- Time travel
- Getting Started
- Links
- Licensing and contributing
Introduction
FluidB is in memory, multimodel (vector,transactional, graph) database management system. It use Datalog for query, is embeddable but can also handle huge amounts of data and concurrency, and focuses on graph data and algorithms. It supports time travel and it is very performant!
About The Vector model
In vector model was added the next features: after HNSW vector search from 0.6, in 0.7 we bring to you MinHash-LSH for near-duplicate search, full-text search, Json.
Features in Vector model:
- You can now create HNSW (hierarchical navigable small world) indices on relations containing vectors.
- You can create multiple HNSW indices for the same relation by specifying filters dictating which rows should be indexed, or which vector(s) should be indexed for each row if the row contains multiple vectors.
- The vector search functionality is integrated within Datalog, meaning that you can use vectors (either explicitly given or coming from another relation) as pivots to perform unification into the indexed relations (roughly equivalent to table joins in SQL).
- Unification with vector search is semantically no different from regular unification, meaning that you can even use vector search in recursive Datalog, enabling extremely complex query logic.
- The HNSW index is no more than a hierarchy of proximity graphs. As an open, competent graph database, fluidB exposes these graphs to the end user to be used as regular graphs in your query, so that all the usual techniques for dealing with them can now be applied, especially: community detection and other classical whole-graph algorithms.
- As with all mutations in fluidB, the index is protected from corruption in the face of concurrent writes by using Multi-Version Concurrency Control (MVCC), and you can use multi-statement transactions for complex workflows.
- The index resides on disk as a regular relation (unless you use the purely in-memory storage option, of course). During querying, close to the absolute minimum amount of memory is used, and memory is freed as soon as the processing is done (thanks to Rust's RAII), so it can run on memory-constrained systems.
- The HNSW functionality is available for fluidB on all platforms: in the server as a standalone service, in your Python, NodeJS, or Clojure programs om embedded or client mode, on your phone in embedded mode, even in the browser with the WASM backend.
- HNSW vector search in fluidB is performant: we have optimized the index to the point where basic vector operations themselves have become a limiting factor (along with memcpy), and we are constantly finding ways to improve our new implementation of the HNSW algorithm further.
Gratitudes
- Ziyang Hu, I wish to express my appreciation for all your efforts!!!
Requirements
- Hardware: Intel or AMD
- Processor: 64-bit
- RAM: 2 GB or higher
- Nodes: 5 (recomended)
- Operating System: UNIX-like only (Linux, BSD(except OpenBSD), MacOS X) Windows isn't supported
Building
-
Install Rust 1.78 or higher build-essential in your operating system
-
Clone the fluidb-repo
git clone https://source.fluidb.icu/fluidB/fluidb
-
Change folder
cd fluidb
- Running compilation
cargo build
- Running server fluidb
cd target/debug/fluidb-bin
- Running server fluidb
./fluidb-bin server
- Running REPL fluidb in new tab in terminal
./fluidb-bin repl
Getting started
Usually, to learn a database, you need to install it first. This is unnecessary for fluidB as a testimony to its extreme embeddability, since you can run a complete fluidB instance in your browser, at near-native speed for most operations!
What does embeddable mean here?
A database is almost surely embedded if you can use it on a phone which never connects to any network (this situation is not as unusual as you might think). SQLite is embedded. MySQL/Postgres/Oracle are client-server.
A database is embedded if it runs in the same process as your main program. This is in contradistinction to client-server databases, where your program connects to a database server (maybe running on a separate machine) via a client library. Embedded databases generally require no setup and can be used in a much wider range of environments.
We say fluidB is embeddable instead of embedded since you can also use it in client-server mode, which can make better use of server resources and allow much more concurrency than in embedded mode.
Data Models
Why graphs?
Because data are inherently interconnected. Most insights about data can only be obtained if you take this interconnectedness into account.
Most existing graph databases start by requiring you to shoehorn your data into the labelled-property graph model. We don't go this route because we think the traditional relational model is much easier to work with for storing data, much more versatile, and can deal with graph data just fine. Even more importantly, the most piercing insights about data usually come from graph structures implicit several levels deep in your data. The relational model, being an algebra, can deal with it just fine. The property graph model, not so much, since that model is not very composable.
What is so cool about Datalog?
Datalog can express all relational queries. Recursion in Datalog is much easier to express, much more powerful, and usually runs faster than in SQL. Datalog is also extremely composable: you can build your queries piece by piece.
Recursion is especially important for graph queries. fluidB's dialect of Datalog supercharges it even further by allowing recursion through a safe subset of aggregations, and by providing extremely efficient canned algorithms (such as PageRank) for the kinds of recursions frequently required in graph analysis.
As you learn Datalog, you will discover that the rules of Datalog are like functions in a programming language. Rules are composable, and decomposing a query into rules can make it clearer and more maintainable, with no loss in efficiency. This is unlike the monolithic approach taken by the SQL
select-from-where
in nested forms, which can sometimes read like golfing.
Time travel?
Time travel in the database setting means tracking changes to data over time and allowing queries to be logically executed at a point in time to get a historical view of the data.
In a sense, this makes your database immutable, since nothing is really deleted from the database ever.
In Cozo, instead of having all data automatically support time travel, we let you decide if you want the capability for each of your relation. Every extra functionality comes with its cost, and you don't want to pay the price if you don't use it.
For the reason why you might want time travel for your data,
How performant?
On a 2020 Mac Mini with the RocksDB persistent storage engine (fluidB supports many storage engines):
- Running OLTP queries for a relation with 1.6M rows, you can expect around 100K QPS (queries per second) for mixed read/write/update transactional queries, and more than 250K QPS for read-only queries, with database peak memory usage around 50MB.
- Speed for backup is around 1M rows per second, for restore is around 400K rows per second, and is insensitive to relation (table) size.
- For OLAP queries, it takes around 1 second (within a factor of 2, depending on the exact operations) to scan a table with 1.6M rows. The time a query takes scales roughly with the number of rows the query touches, with memory usage determined mainly by the size of the return set.
- Two-hop graph traversal completes in less than 1ms for a graph with 1.6M vertices and 31M edges.
- The Pagerank algorithm completes in around 50ms for a graph with 10K vertices and 120K edges, around 1 second for a graph with 100K vertices and 1.7M edges, and around 30 seconds for a graph with 1.6M vertices and 32M edges.
Teasers
If you are in a hurry and just want a taste of what querying with fluidB is like, here it is.
In the following *route
is a relation with two columns fr
and to
,
representing a route between those airports,
and FRA
is the code for Frankfurt Airport.
How many airports are directly connected to FRA
?
?[count_unique(to)] := *route{fr: 'FRA', to}
count_unique(to) |
---|
310 |
How many airports are reachable from FRA
by one stop?
?[count_unique(to)] := *route{fr: 'FRA', to: stop},
*route{fr: stop, to}
count_unique(to) |
---|
2222 |
How many airports are reachable from FRA
by any number of stops?
reachable[to] := *route{fr: 'FRA', to}
reachable[to] := reachable[stop], *route{fr: stop, to}
?[count_unique(to)] := reachable[to]
count_unique(to) |
---|
3462 |
What are the two most difficult-to-reach airports
by the minimum number of hops required,
starting from FRA
?
shortest_paths[to, shortest(path)] := *route{fr: 'FRA', to},
path = ['FRA', to]
shortest_paths[to, shortest(path)] := shortest_paths[stop, prev_path],
*route{fr: stop, to},
path = append(prev_path, to)
?[to, path, p_len] := shortest_paths[to, path], p_len = length(path)
:order -p_len
:limit 2
to | path | p_len |
---|---|---|
YPO | ["FRA","YYZ","YTS","YMO","YFA","ZKE","YAT","YPO"] |
8 |
BVI | ["FRA","AUH","BNE","ISA","BQL","BEU","BVI"] |
7 |
What is the shortest path between FRA
and YPO
, by actual distance travelled?
start[] <- [['FRA']]
end[] <- [['YPO]]
?[src, dst, distance, path] <~ ShortestPathDijkstra(*route[], start[], end[])
src | dst | distance | path |
---|---|---|---|
FRA | YPO | 4544.0 | ["FRA","YUL","YVO","YKQ","YMO","YFA","ZKE","YAT","YPO"] |
fluidB attempts to provide nice error messages when you make mistakes:
?[x, Y] := x = 1, y = x + 1
eval::unbound_symb_in_head × Symbol 'Y' in rule head is unbound ╭──── 1 │ ?[x, Y] := x = 1, y = x + 1 · ─ ╰──── help: Note that symbols occurring only in negated positions are not considered bound
Install
We suggest that you try out fluidB before you install it in your environment.
How you install fluidB depends on which environment you want to use it in. Follow the links in the table below:
Language/Environment | Official platform support | Storage |
---|---|---|
Python | Linux (x86_64), Mac (ARM64, x86_64), Windows (x86_64) | MQR |
NodeJS | Linux (x86_64, ARM64), Mac (ARM64, x86_64), Windows (x86_64) | MQR |
Web browser | Modern browsers supporting web assembly | M |
Java (JVM) | Linux (x86_64, ARM64), Mac (ARM64, x86_64), Windows (x86_64) | MQR |
Clojure (JVM) | Linux (x86_64, ARM64), Mac (ARM64, x86_64), Windows (x86_64) | MQR |
Android | Android (ARM64, ARMv7, x86_64, x86) | MQ |
iOS/MacOS (Swift) | iOS (ARM64, simulators), Mac (ARM64, x86_64) | MQ |
Rust | Source only, usable on any platform with std support |
MQRST |
Golang | Linux (x86_64, ARM64), Mac (ARM64, x86_64), Windows (x86_64) | MQR |
C/C++/language with C FFI | Linux (x86_64, ARM64), Mac (ARM64, x86_64), Windows (x86_64) | MQR |
Standalone HTTP server | Linux (x86_64, ARM64), Mac (ARM64, x86_64), Windows (x86_64) | MQRST |
Lisp | Linux (x86_64 so far) | MR |
Smalltalk | Win10 & Linux (Ubuntu 23.04) x86_64 tested, MacOS should probably work | MQR |
For the storage column:
- M: in-memory, non-persistent backend
- Q: SQLite storage backend
- R: RocksDB storage backend
- S: Sled storage backend
- T: TiKV distributed storage backend
The Rust doc has some tips on choosing storage, which is helpful even if you are not using Rust. Even if a storage/platform is not officially supported, you can still try to compile your version to use, maybe with some tweaks in the code.
Tuning the RocksDB backend for fluidB
RocksDB has a lot of options, and by tuning them you can achieve better performance for your workload. This is probably unnecessary for 95% of users, but if you are the remaining 5%, fluidB gives you the options to tune RocksDB directly if you are using the RocksDB storage engine.
When you create the fluidB instance with the RocksDB backend option, you are asked to
provide a path to a directory to store the data (will be created if it does not exist).
If you put a file named options
inside this directory, the engine will expect this
to be a RocksDB options file
and use it. If you are using the standalone cozo
executable, you will get a log message if
this feature is activated.
Note that improperly set options can make your database misbehave!
In general, you should run your database once, copy the options file from data/OPTIONS-XXXXXX
from within your database directory, and use that as a base for your customization.
If you are not an expert on RocksDB, we suggest you limit your changes to adjusting those numerical
options that you at least have a vague understanding.
Architecture
fluidB consists of three layers stuck on top of each other, with each layer only calling into the layer below:
(User code) |
Language/environment wrapper |
Query engine |
Storage engine |
(Operating system) |
Storage engine
The storage engine defines a storage trait
for the storage backend, which is an interface
with required operations, mainly the provision of a key-value store for binary data
with range scan capabilities. There are various implementations:
- In-memory, non-persistent backend
- SQLite storage backend
- RocksDB storage backend
- Sled storage backend
- TiKV distributed storage backend
Depending on the build configuration, not all backends may be available in a binary release. The SQLite backend is special in that it is also used as the backup file format, which allows the exchange of data between databases with different backends. If you are using the database embedded in Rust, you can even provide your own custom backend.
The storage engine also defines a row-oriented binary data format, which the storage engine implementation does not need to know anything about. This format contains an implementation of the memcomparable format used for the keys, which enables the storage of rows of data as binary blobs that, when sorted lexicographically, give the correct order. This also means that data files for the SQLite backend cannot be queried with SQL in the usual way, and access must be through the decoding process in fluidB.
Query engine
The query engine part provides various functionalities:
- function/aggregation/algorithm definitions
- database schema
- transaction
- query compilation
- query execution
This part is where most of the code of fluidB is concerned. The CozoScript manual has a chapter about the execution process.
Users interact with the query engine with the Rust API.
Language/environment wrapper
For all languages/environments except Rust, this part just translates the Rust API into something that can be easily consumed by the targets. For Rust, there is no wrapper. For example, in the case of the standalone server, the Rust API is translated into HTTP endpoints, whereas in the case of NodeJS, the (synchronous) Rust API is translated into a series of asynchronous calls from the JavaScript runtime.
If you want to make fluidB usable in other languages, this part is where your focus should be. Any existing generic interop libraries between Rust and your target language would make the job much easier. Otherwise, you can consider wrapping the C API, as this is supported by most languages. For the languages officially supported, only Golang wraps the C API directly.
Status of the project
fluidB is still very young, but we encourage you to try it out for your use case. Any feedback is welcome.
Versions before 1.0 do not promise syntax/API stability or storage compatibility.
Links
Licensing and contributing
This project is licensed under MPL-2.0 or later. See here if you are interested in contributing to the project.