An embedded JSON document database engine for the JVM, written in Kotlin.

Delva stores JSON documents of any size, indexes every key–value pair in them, and answers key–value and JSONPath searches over that index. It is a library, not a server: one process opens a directory and works with it directly. Documents are streamed in and out, so memory is bounded by configuration and never by document size: its tests insert, search and read back a 2 GiB document in a 256 MB heap.

Delva is Bulgarian — делва, the earthenware jar a household sinks into a cellar floor and seals for the winter.

Status: early development. Until 1.0, a minor version can change the API and the on-disk format incompatibly; see Compatibility.

Delva needs JDK 22 or newer, and depends only on the Kotlin standard library and kotlinx-coroutines. Its JPMS module name is app.oreshkov.delva. It is published to Maven Central:

dependencies {
    implementation("app.oreshkov.delva:delva:0.2.0")
}

import app.oreshkov.delva.*
import java.nio.file.Path

JsonStore.openOrCreate(Path.of("data/store")).use { store ->
    val blocking = store.blocking()

    val id = blocking.insert("""{"name":"ada","tags":["math","engines"],"score":42}""")

    blocking.search(KeyValueQuery("name", JsonQueryValue.Str("ada"))).use { result ->
        for (hit in result) println(hit.readJsonAsString())
    }

    // An element of an array value matches the array's member too.
    blocking.search(KeyValueQuery("tags", JsonQueryValue.Str("math"), ResultMode.CONTAINING_ELEMENT)).use { result ->
        for (hit in result) println(hit.node)
    }

    blocking.delete(id)
}

The guide covers the suspending functions that blocking() wraps, and everything else; the KDoc documents every public declaration.

• Key–value search over every object in every document, as whole documents or the containing objects, oldest or newest first, paged through cursors.
• JSONPath: all of RFC 9535 — select within one value, and searches that the index narrows down by the query's filters.
• Versions and keys: documents are immutable; replace swaps in a new version atomically, and an optional document key adds get, upsert and listing in key order.
• Types: name objects by JSON path as they are inserted or later, and filter searches on them.
• Readers beside the writer: other processes open a secondary that never locks or writes.
• Compare and diff two stored values, or one against JSON text, streamed with bounded memory.
• Crash consistency: an insert is completely visible after a crash or completely absent; verify scrubs the whole store for damage.
• Optional Unicode NFC normalization of every string, key and query.

• Guide: every feature with examples, limits and sizing, use from Java, and compatibility.
• Specification: the exact semantics and canonical byte forms, a draft until 1.0.
• Roadmap and changelog; cli-kit has its own.
• Development: building, the test suites and CI; releasing.

Bug reports and discussion of the semantics are welcome; see CONTRIBUTING.md. Report security problems — including anything that loses or corrupts data — privately, as SECURITY.md describes.

Apache License 2.0