PSOrvia

Demonstration – et musikbibliotek

100 000 numre, et indeks, fuldtekstsøgning og grupperede optællinger i ét kort Rust-program.

Hvad det gør

Programmet opretter en tabel med numre (titel, kunstner, år, længde) og et indeks på kunstneren, genererer 100 000 numre ud fra et fast seed, fylder en fuldtekst-tabel med en trigram-tokenizer og besvarer tre søgninger og én grupperet forespørgsel.

Trin

  1. Opret tabellen, indekset og fuldtekst-tabellen.
  2. Indsæt 100 000 numre i én transaktion med bundne parametre.
  3. Fyld fuldtekst-tabellen fra numrene i ét statement.
  4. Søg med MATCH, for eksempel en frase, to ord kombineret med AND og et ordpræfiks.
  5. Tæl numre pr. kunstner med GROUP BY.

Søgningen

// SPDX-License-Identifier: Apache-2.0
// Showcase: a music library of 100 000 tracks with full-text search.
// Usage: musiclib [tracks] [file]   (defaults: 100000 tracks, a temporary file)
use std::time::Instant;

use psob::encoding::key::Value;
use psob::sql::conn::Connection;
use psob::sql::eval::to_text;
use psob::sql::exec::Params;
use psob::{Database, OpenOptions};

const ADJ: [&str; 16] = [
    "Quiet", "Golden", "Broken", "Electric", "Hidden", "Silver", "Midnight", "Wild", "Slow", "Paper", "Distant",
    "Velvet", "Burning", "Frozen", "Lonely", "Bright",
];
const NOUN: [&str; 16] = [
    "River", "Garden", "Machine", "Harbor", "Letter", "Mirror", "Engine", "Forest", "Window", "Season", "Signal",
    "Island", "Morning", "Shadow", "Bridge", "Orchard",
];
const ARTIST: [&str; 12] = [
    "Anna Keller", "The Lanterns", "Mo Okafor", "Blue Harbor", "Ines Duarte", "Kite Season", "Sven Lindqvist",
    "Tomoko Ito", "Orchard Club", "Lena Voss", "North Wing", "Paper Moons",
];

struct Lcg(u64);
impl Lcg {
    fn next(&mut self, n: usize) -> usize {
        self.0 = self.0.wrapping_mul(6364136223846793005).wrapping_add(1442695040888963407);
        ((self.0 >> 33) as usize) % n
    }
}

fn text(s: &str) -> Value {
    Value::Text(s.to_string())
}

fn main() -> psob::Result<()> {
    let args: Vec<String> = std::env::args().collect();
    let n: usize = args.get(1).and_then(|s| s.parse().ok()).unwrap_or(100_000);
    let path = match args.get(2) {
        Some(p) => std::path::PathBuf::from(p),
        None => std::env::temp_dir().join(format!("psob-musiclib-{}.db", std::process::id())),
    };
    let _ = std::fs::remove_file(&path);
    let db = Database::open(&path, OpenOptions::default())?;
    let mut conn = Connection::with_mode(db, psob::sql::SqlMode::Sqlite)?;
    conn.execute("CREATE TABLE tracks(id INTEGER PRIMARY KEY, title TEXT NOT NULL, artist TEXT NOT NULL, year INT, seconds INT)")?;
    conn.execute("CREATE INDEX tracks_artist ON tracks(artist)")?;
    conn.execute("CREATE VIRTUAL TABLE search USING fts5(title, artist, tokenize = 'trigram remove_diacritics 1', content = '')")?;

    let t0 = Instant::now();
    let mut rng = Lcg(42);
    conn.execute("BEGIN")?;
    for i in 1..=n {
        let title = format!("{} {} {}", ADJ[rng.next(16)], NOUN[rng.next(16)], i);
        let artist = ARTIST[rng.next(12)];
        let p = Params {
            positional: vec![
                Value::Integer(i as i64),
                text(&title),
                text(artist),
                Value::Integer(1960 + rng.next(66) as i64),
                Value::Integer(120 + rng.next(300) as i64),
            ],
            named: vec![],
        };
        conn.execute_with("INSERT INTO tracks VALUES (?1, ?2, ?3, ?4, ?5)", &p)?;
    }
    conn.execute("INSERT INTO search(rowid, title, artist) SELECT id, title, artist FROM tracks")?;
    conn.execute("COMMIT")?;
    let load = t0.elapsed();

    let count = conn.query("SELECT count(*) FROM tracks")?;
    println!("tracks: {}", to_text(&count.rows[0][0]));

    for q in ["velvet bridge", "orchard AND kite", "froz"] {
        let t = Instant::now();
        let p = Params { positional: vec![text(q)], named: vec![] };
        let r = conn.query_with(
            "SELECT count(*) FROM search WHERE search MATCH ?1",
            &p,
        )?;
        let hits = to_text(&r.rows[0][0]);
        let top = conn.query_with(
            "SELECT t.title, t.artist FROM search JOIN tracks t ON t.id = search.rowid WHERE search MATCH ?1 ORDER BY t.id LIMIT 1",
            &p,
        )?;
        println!("search '{}': {} hits, first: {} / {}", q, hits, to_text(&top.rows[0][0]), to_text(&top.rows[0][1]));
        eprintln!("  ({:.1} ms)", t.elapsed().as_secs_f64() * 1000.0);
    }
    let by = conn.query("SELECT artist, count(*) AS n FROM tracks GROUP BY artist ORDER BY n DESC, artist LIMIT 3")?;
    for row in &by.rows {
        println!("artist {} has {} tracks", to_text(&row[0]), to_text(&row[1]));
    }
    eprintln!("loaded and indexed in {:.2} s", load.as_secs_f64());
    drop(conn);
    if args.get(2).is_none() {
        let _ = std::fs::remove_file(&path);
    }
    Ok(())
}

Kør det

Outputtet er det samme ved hver kørsel, fordi dataene kommer fra et fast seed. Tidsmålingerne går til fejloutputtet.

cd examples/rust
cargo run --release --bin musiclib

Output

tracks: 100000
search 'velvet bridge': 463 hits, first: Velvet Bridge 129 / Mo Okafor
search 'orchard AND kite': 509 hits, first: Golden Orchard 133 / Kite Season
search 'froz': 6299 hits, first: Frozen Bridge 6 / Sven Lindqvist
artist Paper Moons has 8448 tracks
artist Orchard Club has 8447 tracks
artist Sven Lindqvist has 8433 tracks

Tidsmålinger

Enkelt kørsel, ingen sammenligning med andre databaser, foreløbig.

Indsæt 100 000 numre og fyld fuldtekst-tabellen2,0 s
Søgning efter en frase (463 træffere)4,5 ms
Søgning efter to ord med AND (509 træffere)6,9 ms
Søgning efter et ordpræfiks (6299 træffere)11,4 ms

Fuldtekststøtten er i dag trigram-tokenizeren på indholdsløse tabeller. Andre tokenizere og rangeringsfunktioner er ikke bygget endnu.