vecdb-0.3.5 has been yanked.

vecdb

High-performance mutable persistent vectors built on rawdb.

It features:

Vec based API: push, update, truncate, delete by index
Multiple variants: raw, compressed, computed
Rollback via stamped change deltas
Sparse deletions with holes
Thread-safe with concurrent reads
Blazing fast (benchmark)
Persistence only on flush

It is not:

A key-value store (consider fjall or redb)
Suited for variable-sized types (String, Vec<T>, etc.)

Install

cargo add vecdb

Usage

use vecdb::{AnyStoredVec, Database, GenericStoredVec, RawVec, Result, Version};

fn main() -> Result<()> {
    // create
    let temp_dir = tempfile::TempDir::new()?;
    let db = Database::open(temp_dir.path())?;
    let mut vec: RawVec<usize, u64> = RawVec::import(&db, "vec", Version::ONE)?;

    // push
    for i in 0..1_000_000 {
        vec.push(i);
    }

    // flush
    vec.flush()?;
    db.flush()?;

    // read (sequential)
    let mut sum = 0u64;
    for value in vec.iter()? {
        sum = sum.wrapping_add(value);
    }

    // read (random)
    let indices: Vec<usize> = vec![500, 1000, 10];
    let reader = vec.create_reader();
    for i in indices {
        if let Ok(value) = vec.read_at(i, &reader) {
            sum = sum.wrapping_add(value);
        }
    }

    Ok(())
}

Constraints

Data must be fixed-size types: numbers, fixed arrays, structs with #[repr(C)].

Compression via Pcodec works for numeric types only.

When to use it

Need to store Vecs on disk
Append-only or append-mostly workloads
Need very high read speeds
Space-efficient storage for numeric data
Sparse deletions without reindexing
Rollback without full snapshots

Examples