eyros (εύρος) is a multi-dimensional interval database.
The database is based on bkd and interval trees.
- high batch-write performance (expect 100,000s to 1,000,000s writes per second on modest hardware)
- designed for peer-to-peer distribution and query-driven sparse replication
- compiles to web assembly for use in the browser
- good for geospatial and time-series data
eyros operates on scalar (x) or interval (min,max) coordinates for each dimension. There are 2 operations: batched write (for inserting and deleting) and query by bounding box. All features that intersect the bounding box are returned in the query results.
This is an early release missing important features such as atomicity and concurrency. The data format is still in flux and will likely change in the future, requiring data migrations.
This example generates 800 random features in 3 dimensions: x, y, and time
with a u32 value payload. The x and y dimensions are both intervals with
a minimum and maximum f32 and time is a scalar f32.
After the data is written to the database, all features with an x interval
that overlaps (-0.5,0.3), a y interval that overlaps (-0.8,-0.5), and a
time scalar that is between 0.0 and 100.0 are printed to stdout.
use eyros::{DB,Row};
use rand::random;
use failure::Error;
use random_access_disk::RandomAccessDisk;
use std::path::PathBuf;
type P = ((f32,f32),(f32,f32),f32);
type V = u32;
fn main() -> Result<(),Error> {
let mut db: DB<_,_,((f32,f32),(f32,f32),f32),u32> = DB::open(storage)?;
let polygons: Vec<Row<P,V>> = (0..800).map(|_| {
let xmin: f32 = random::<f32>()*2.0-1.0;
let xmax: f32 = xmin + random::<f32>().powf(64.0)*(1.0-xmin);
let ymin: f32 = random::<f32>()*2.0-1.0;
let ymax: f32 = ymin + random::<f32>().powf(64.0)*(1.0-ymin);
let time: f32 = random::<f32>()*1000.0;
let value: u32 = random();
let point = ((xmin,xmax),(ymin,ymax),time);
Row::Insert(point, value)
}).collect();
db.batch(&polygons)?;
let bbox = ((-0.5,-0.8,0.0),(0.3,-0.5,100.0));
for result in db.query(&bbox)? {
println!("{:?}", result?);
}
Ok(())
}
fn storage(name:&str) -> Result<RandomAccessDisk,Error> {
let mut p = PathBuf::from("/tmp/eyros-db/");
p.push(name);
Ok(RandomAccessDisk::builder(p)
.auto_sync(false)
.build()?)
}The output from this program is of the form (coords, value, location):
$ cargo run --example polygons -q
(((-0.014986515, -0.014986515), (-0.5801666, -0.5801663), 45.314373), 1518966744, (0, 200))
(((-0.0892005, -0.015534878), (-0.65783, -0.65783), 3.6987066), 66257667, (0, 267))
(((0.1931547, 0.1931547), (-0.6388786, -0.60205233), 67.85113), 2744609531, (0, 496))
(((-0.28907382, -0.26248854), (-0.7761978, -0.77617484), 55.273056), 3622408505, (0, 651))
(((-0.080417514, -0.080417514), (-0.60076225, -0.5929384), 29.592216), 722871034, (0, 784))
(((0.14104307, 0.14104307), (-0.539363, -0.539363), 31.965792), 2866780128, (0, 933))
(((-0.12689173, -0.12689173), (-0.56708515, -0.56643564), 65.072), 1858542500, (0, 983))
(((-0.12520671, -0.1250745), (-0.6836084, -0.6836084), 93.58209), 3942792215, (0, 1019))
(((0.026417613, 0.026417613), (-0.786397, -0.786397), 61.52451), 1197187917, (0, 1102))
(((-0.18799019, -0.18799017), (-0.50418067, -0.50418067), 82.93134), 2811117540, (0, 1199))
(((-0.34033966, -0.34033966), (-0.53603613, -0.53603613), 91.07471), 302136936, (0, 1430))
(((-0.008744121, 0.54438573), (-0.73665094, -0.73665094), 69.67532), 719725479, (0, 1504))
(((-0.38071227, -0.38071224), (-0.75237143, -0.75237143), 72.245895), 2200140390, (0, 1628))
(((0.020396352, 0.020396352), (-0.7957357, -0.77274036), 40.785194), 2166765724, (0, 1708))
(((0.117452025, 0.117452025), (-0.7027955, -0.7026706), 82.033394), 2451987859, (0, 1886))
(((-0.11418259, -0.11418259), (-0.74327374, -0.74327374), 28.591274), 4283568770, (0, 1983))
(((-0.19130886, -0.19130856), (-0.7012402, -0.7012042), 2.1106005), 4226013993, (0, 2048))
(((-0.3000791, -0.3000791), (-0.7601782, -0.7601782), 24.528027), 2776778380, (0, 2349))
The coords and value are the values that were written earlier: in this case,
the coords are ((xmin,xmax),(ymin,ymax),time).
The location is used to quickly delete records without needing to perform
additional lookups. You'll need to keep the location around from the result of
a query when you intend to delete a record. Locations that begin with a 0 are
stored in the staging cache, so their location may change after the next write.
You can also mix and match scalar and interval values for each dimension.
An example of where these mixed types might be useful is storing geographic features to display on a map. Some of the features will be points and some will be lines or polygons which are contained in bounding boxes (intervals).
This example stores 2 dimensional points and regions in the same database, so that bounding box queries will return both types of features.
use eyros::{DB,Row,Mix,Mix2};
use rand::random;
use failure::Error;
use random_access_disk::RandomAccessDisk;
use std::path::PathBuf;
type P = Mix2<f32,f32>;
type V = u32;
fn main() -> Result<(),Error> {
let mut db: DB<_,_,P,V> = DB::open(storage)?;
let batch: Vec<Row<P,V>> = (0..1_000).map(|_| {
let value = random::<u32>();
if random::<f32>() > 0.5 {
let xmin: f32 = random::<f32>()*2.0-1.0;
let xmax: f32 = xmin + random::<f32>().powf(64.0)*(1.0-xmin);
let ymin: f32 = random::<f32>()*2.0-1.0;
let ymax: f32 = ymin + random::<f32>().powf(64.0)*(1.0-ymin);
Row::Insert(Mix2::new(
Mix::Interval(xmin,xmax),
Mix::Interval(ymin,ymax)
), value)
} else {
let x: f32 = random::<f32>()*2.0-1.0;
let y: f32 = random::<f32>()*2.0-1.0;
Row::Insert(Mix2::new(
Mix::Scalar(x),
Mix::Scalar(y)
), value)
}
}).collect();
db.batch(&batch)?;
let bbox = ((-0.5,-0.8),(0.3,-0.5));
for result in db.query(&bbox)? {
println!("{:?}", result?);
}
Ok(())
}
fn storage(name:&str) -> Result<RandomAccessDisk,Error> {
let mut p = PathBuf::from("/tmp/eyros-mix-db/");
p.push(name);
Ok(RandomAccessDisk::builder(p)
.auto_sync(false)
.build()?)
}license zero parity 7.0.0 and apache 2.0 (contributions)