Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
58 changes: 58 additions & 0 deletions .github/workflows/run-special-checks-sandbox.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,58 @@
name: "Run special checks: module lucene/sandbox"

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm add a CHANGES.txt entry? I think this can be safely backported to 10.x, after we bake for a week or two in main? It's sandbox, completely separate from everything else, experimental ...

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added!

It's sandbox, completely separate from everything else, experimental ...

+1 -- I don't think it should affect users who won't use it

on:
workflow_dispatch:

pull_request:
branches:
- '*'

push:
branches:
- 'main'
- 'branch_10x'

jobs:
faiss-tests:
name: tests for the Faiss codec (v${{ matrix.faiss-version }} with JDK ${{ matrix.java }} on ${{ matrix.os }})
timeout-minutes: 15

strategy:
matrix:
os: [ ubuntu-latest ]
java: [ '24' ]
faiss-version: [ '1.11.0' ]

runs-on: ${{ matrix.os }}

steps:
- name: Install Mamba
uses: conda-incubator/setup-miniconda@835234971496cad1653abb28a638a281cf32541f #v3.2.0
with:
miniforge-version: 'latest'
auto-activate-base: 'false'
activate-environment: 'faiss-env'
# TODO: Use only conda-forge if possible, see https://github.com/conda-forge/faiss-split-feedstock/pull/88
channels: 'pytorch,conda-forge'
conda-remove-defaults: 'true'

- name: Install Faiss
run: mamba install faiss-cpu=${{ matrix.faiss-version }}

- name: Checkout Lucene
uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2

- name: Prepare Lucene workspace
uses: ./.github/actions/prepare-for-build

- name: Run tests for Faiss codec
run: >
LD_LIBRARY_PATH=$CONDA_PREFIX/lib
./gradlew -p lucene/sandbox
-Dtests.faiss.run=true
test
--tests "org.apache.lucene.sandbox.codecs.faiss.*"

defaults:
run:
shell: bash -leo pipefail {0}
Original file line number Diff line number Diff line change
Expand Up @@ -147,6 +147,9 @@ buildOptions.addOption("tests.file.encoding", "Sets the default file.encoding on
])
})

buildOptions.addBooleanOption("tests.faiss.run", "Explicitly run tests for the Faiss codec.", false)
optionsInheritedAsProperties += ["tests.faiss.run"]

// TODO: do we still use these?
// Test data file used.
// [propName: 'tests.linedocsfile', value: 'europarl.lines.txt.gz', description: "Test data file path."],
Expand Down
2 changes: 2 additions & 0 deletions lucene/CHANGES.txt
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,8 @@ New Features
---------------------
* GITHUB#14097: Binary partitioning merge policy over float-valued vector field. (Mike Sokolov)

* GITHUB#14178: Add a Faiss-based vector format in the sandbox module. (Kaival Parikh)

Improvements
---------------------

Expand Down
3 changes: 3 additions & 0 deletions lucene/sandbox/src/java/module-info.java
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
requires org.apache.lucene.facet;

exports org.apache.lucene.payloads;
exports org.apache.lucene.sandbox.codecs.faiss;
exports org.apache.lucene.sandbox.codecs.idversion;
exports org.apache.lucene.sandbox.codecs.quantization;
exports org.apache.lucene.sandbox.document;
Expand All @@ -39,4 +40,6 @@

provides org.apache.lucene.codecs.PostingsFormat with
org.apache.lucene.sandbox.codecs.idversion.IDVersionPostingsFormat;
provides org.apache.lucene.codecs.KnnVectorsFormat with
org.apache.lucene.sandbox.codecs.faiss.FaissKnnVectorsFormat;
}
Original file line number Diff line number Diff line change
@@ -0,0 +1,120 @@
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.lucene.sandbox.codecs.faiss;

import static org.apache.lucene.util.hnsw.HnswGraphBuilder.DEFAULT_BEAM_WIDTH;
import static org.apache.lucene.util.hnsw.HnswGraphBuilder.DEFAULT_MAX_CONN;

import java.io.IOException;
import java.util.Locale;
import org.apache.lucene.codecs.KnnVectorsFormat;
import org.apache.lucene.codecs.KnnVectorsReader;
import org.apache.lucene.codecs.KnnVectorsWriter;
import org.apache.lucene.codecs.hnsw.FlatVectorScorerUtil;
import org.apache.lucene.codecs.hnsw.FlatVectorsFormat;
import org.apache.lucene.codecs.lucene99.Lucene99FlatVectorsFormat;
import org.apache.lucene.index.SegmentReadState;
import org.apache.lucene.index.SegmentWriteState;

/**
* A Faiss-based format to create and search vector indexes, using {@link LibFaissC} to interact
* with the native library.
*
* <p>The Faiss index is configured using its flexible <a
* href="https://github.com/facebookresearch/faiss/wiki/The-index-factory">index factory</a>, which
* allows creating arbitrary indexes by "describing" them. These indexes can be tuned by <a
* href="https://github.com/facebookresearch/faiss/wiki/Index-IO,-cloning-and-hyper-parameter-tuning">setting
* relevant parameters</a>.
*
* <p>A separate Faiss index is created per-segment, and uses the following files:
*
* <ul>
* <li><code>.faissm</code> (metadata file): stores field number, offset and length of actual
* Faiss index in data file.
* <li><code>.faissd</code> (data file): stores concatenated Faiss indexes for all fields.
* <li>All files required by {@link Lucene99FlatVectorsFormat} for storing raw vectors.
* </ul>
*
* <p>Note: Set the {@code $OMP_NUM_THREADS} environment variable to control <a
* href="https://github.com/facebookresearch/faiss/wiki/Threads-and-asynchronous-calls">internal
* threading</a>.
*
* <p>TODO: There is no guarantee of backwards compatibility!
*
* @lucene.experimental
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could you also add a sentence making it clear there is no promise of backwards compatibility here?

That said, does Faiss have any promise? If I write a Faiss HNSW graph with version X, and then upgrade Faiss to version X+1, can the HNSW graph be read/written?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I do see some references of backwards compatibility, and an old comment which says that newer versions of Faiss can read older indexes -- but I couldn't find documentation for it..

Further, we may change some internals of the codec making it incompatible with earlier versions -- but I'll add a comment saying there's no guarantee today, and a TODO to figure that out

*/
public final class FaissKnnVectorsFormat extends KnnVectorsFormat {
public static final String NAME = FaissKnnVectorsFormat.class.getSimpleName();
static final int VERSION_START = 0;
static final int VERSION_CURRENT = VERSION_START;
static final String META_CODEC_NAME = NAME + "Meta";
static final String DATA_CODEC_NAME = NAME + "Data";
static final String META_EXTENSION = "faissm";
static final String DATA_EXTENSION = "faissd";

private final String description;
private final String indexParams;
private final FlatVectorsFormat rawVectorsFormat;

/**
* Constructs an HNSW-based format using default {@code maxConn}={@value
* org.apache.lucene.util.hnsw.HnswGraphBuilder#DEFAULT_MAX_CONN} and {@code beamWidth}={@value
* org.apache.lucene.util.hnsw.HnswGraphBuilder#DEFAULT_BEAM_WIDTH}.
*/
public FaissKnnVectorsFormat() {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could you add a comment somewhere or maybe a README with rough high level instructions for how to build this? E.g. install Faiss 1.11.0 or newer, typically via pytorch or conda-forge channel from XYZ, then set this gradle property or so, then ./gradlew such-and-suchat root level, etc?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added to package-info.java as suggested below!

this(
String.format(Locale.ROOT, "IDMap,HNSW%d", DEFAULT_MAX_CONN),
String.format(Locale.ROOT, "efConstruction=%d", DEFAULT_BEAM_WIDTH));
}

/**
* Constructs a format using the specified index factory string and index parameters (see class
* docs for more information).
*
* @param description the index factory string to initialize Faiss indexes.
* @param indexParams the index params to set on Faiss indexes.
*/
public FaissKnnVectorsFormat(String description, String indexParams) {
super(NAME);
this.description = description;
this.indexParams = indexParams;
this.rawVectorsFormat =
new Lucene99FlatVectorsFormat(FlatVectorScorerUtil.getLucene99FlatVectorsScorer());
}

@Override
public KnnVectorsWriter fieldsWriter(SegmentWriteState state) throws IOException {
return new FaissKnnVectorsWriter(
description, indexParams, state, rawVectorsFormat.fieldsWriter(state));
}

@Override
public KnnVectorsReader fieldsReader(SegmentReadState state) throws IOException {
return new FaissKnnVectorsReader(state, rawVectorsFormat.fieldsReader(state));
}

@Override
public int getMaxDimensions(String fieldName) {
return DEFAULT_MAX_DIMENSIONS;
}

@Override
public String toString() {
return String.format(
Locale.ROOT, "%s(description=%s indexParams=%s)", NAME, description, indexParams);
}
}
Loading
Loading