Skip to content

[FEA] Lucene: provide a thin-JAR adapter for configured PyLucene HNSW #2619

Description

@nvzm123

Problem

Stock PyLucene wraps Lucene APIs but cannot directly construct cuvs-lucene's parameterized Lucene101AcceleratedHNSWCodec. Compiling a Java adapter at cuVS Bench runtime would put Java source under Python ownership and add an avoidable compiler prerequisite.

Proposed change

Add a small production adapter to the standard thin cuvs-lucene JAR. Its public no-argument constructor should read namespaced JVM properties for Lucene maxConn and beamWidth, validate them, snapshot them into AcceleratedHNSWParams, and delegate to the existing codec.

The adapter must be loaded explicitly by class name and must not be registered as a Lucene SPI provider, because SPI instances are cached and required properties would make enumeration unsafe. Add property-validation, repeated-construction, codec-name, thin-JAR-content, and fresh-JVM PyLucene coverage.

This is an additive API; existing codecs and persisted names remain unchanged. Related to #2470 and #2194.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions