[GitHub] [lucene] dweiss merged pull request #11993: Upgrade gradle to 7.6.
dweiss merged PR #11993: URL: https://github.com/apache/lucene/pull/11993 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] dweiss merged pull request #11994: Upgrade gradle to 7.6 (backport)
dweiss merged PR #11994: URL: https://github.com/apache/lucene/pull/11994 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] rmuir commented on pull request #11987: Make Decompressor release memory buffer
rmuir commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335235418 > @uschindler I think this issue just have a GC path of ThreadLocals. BUT, for instance in ES, when there is a 1000-shard-nodes, and normally one shard with 40 segments per shard, one opened segments would allocate one buffer with retained heap: 100KB, so this would use 4G resident heap memory, even some segments are rarely used. You can use less shards as a workaround for now. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] rmuir commented on pull request #11987: Make Decompressor release memory buffer
rmuir commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335237178 In fact there is no situation where thousands of shards makes sense on a single node. That's bad design. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] benwtrent commented on issue #11963: Improve vector quantization API
benwtrent commented on issue #11963: URL: https://github.com/apache/lucene/issues/11963#issuecomment-1335239951 @rmuir thank you! I currently have a major part of the refactor already written locally. But, it will be big and will need its own rounds of review to make sure its the way we want it. I want to try and get the PR up (once the size improvements is merged) ASAP -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] benwtrent commented on a diff in pull request #11860: GITHUB-11830 Better optimize storage for vector connections
benwtrent commented on code in PR #11860:
URL: https://github.com/apache/lucene/pull/11860#discussion_r1038154379
##
lucene/backward-codecs/src/java/org/apache/lucene/backward_codecs/lucene94/ExpandingVectorValues.java:
##
@@ -0,0 +1,49 @@
+/*
+ * 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.backward_codecs.lucene94;
+
+import java.io.IOException;
+import org.apache.lucene.index.FilterVectorValues;
+import org.apache.lucene.index.VectorValues;
+import org.apache.lucene.util.BytesRef;
+
+/** reads from byte-encoded data */
+public class ExpandingVectorValues extends FilterVectorValues {
+
+ private final float[] value;
+
+ /**
+ * Constructs ExpandingVectorValues with passed byte encoded VectorValues
iterator
+ *
+ * @param in the wrapped values
+ */
+ protected ExpandingVectorValues(VectorValues in) {
Review Comment:
@rmuir I don't think you are a madman. Apologies if my responses implied
that. Just a little frustrated (as this change took work and already has a
giant review).
I am working on the refactor now and want to get a PR for it ASAP after
this. I am 100% on your side that the current API is not optimal. So, I will be
splitting out the byte/float32 types. It will make the whole thing better.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
-
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] rmuir commented on pull request #11987: Make Decompressor release memory buffer
rmuir commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335269730 I will investigate the document API and try to make a proposal so that threadlocal is no longer needed. I'm really concerned about the merge case here causing regressions for folks that delete/update documents, I'll also try to run the benchmark myself for the unoptimized merge case. It requires a bit of time, as document api will impact 100% of lucene users so we need to get it right. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] luyuncheng commented on pull request #11987: Make Decompressor release memory buffer
luyuncheng commented on PR #11987:
URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335302726
> In fact there is no situation where thousands of shards makes sense on a
single node. That's bad design.
@rmuir i have another proposal:
what do you think of make `ByteBufferDataInput` as not fixed sized ( cause
you mentioned LUCENE-10627 do not introduce any other ByteBlockPool)
every time we call decompress and wrap decompressed BytesRef as
ByteArrayDataInput
the decompress can make a list ByteBuffer, and every time just copy
dictionary,
this can reduce memory copy from buffer and reduce ArrayUtil.grow's memory
copy,
as well as can solved buffer-heap-influence
simple code like following:
```
// Read blocks that intersect with the interval we need
ArrayList bufferList = new ArrayList<>();
while (offsetInBlock < offset + length) {
final int bytesToDecompress = Math.min(blockLength, offset + length
- offsetInBlock);
byte[] bytebuffer = new byte[dictLength + blockLength];
//Only copy dict
System.arraycopy(dictBuffer, 0, bytebuffer, 0, dictLength);
LZ4.decompress(in, bytesToDecompress, bytebuffer, dictLength);
offsetInBlock += blockLength;
bufferList.add(ByteBuffer.wrap(bytebuffer, dictLength,
bytesToDecompress).slice());
}
return = new ByteBuffersDataInput(bufferList);
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
-
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] jpountz commented on pull request #11987: Make Decompressor release memory buffer
jpountz commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335358751 I had opened a very similar PR to this one at https://github.com/apache/lucene/pull/137 which handled the merge case. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] jpountz commented on pull request #11987: Make Decompressor release memory buffer
jpountz commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335365755 I think I had not merged it because the follow-up discussion about removing thread locals had triggered naming/API concerns, but it should be a good incremental step and we could figure a way to remove threadlocals in Lucene 10 since it will require API changes? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] rmuir commented on pull request #11987: Make Decompressor release memory buffer
rmuir commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335391991 > I think I had not merged it because the follow-up discussion about removing thread locals had triggered naming/API concerns, but it should be a good incremental step and we could figure a way to remove threadlocals in Lucene 10 since it will require API changes? I don't think so, i tend to lean towards Uwe's thoughts here. I feel like this is "rush in another fix for 10,000 shards". We can investigate removing the threadlocals in lucene 10. Very possibly the change could be backported with deprecations for `IR.document(int)` (e.g. which would use threadlocal like now). At least to me, that's the ideal scenario. So you could migrate your lucene 9 code to no longer use deprecated methods and not have the threadlocals. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] jpountz commented on pull request #11987: Make Decompressor release memory buffer
jpountz commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335392646 Sounds good. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] jpountz closed pull request #137: LUCENE-9955: Reduced state of stored fields readers.
jpountz closed pull request #137: LUCENE-9955: Reduced state of stored fields readers. URL: https://github.com/apache/lucene/pull/137 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] rmuir commented on pull request #11987: Make Decompressor release memory buffer
rmuir commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335398330 its just gonna take me some time, i can't get something out there like today. for example nearly 100% of tests would be impacted :) It is fair, I will feel the same pain the users will. But it is some motivation to take care with the API and spend some time to do it well. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] luyuncheng commented on pull request #11987: Make Decompressor release memory buffer
luyuncheng commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335420862 Threadlocal just scale up the StoredFieldsReader's heap useage, BUT when one instance with only 10K segments would use 1G heap memory in `fieldsReaderOrig#buffer` without threadlocal. my idea is when the buffer is rarely used, it should be recycled -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] rmuir commented on pull request #11987: Make Decompressor release memory buffer
rmuir commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335444854 the idea is not to have one instance per segment. there would be zero instances. when you want to retrieve docs from indexreader, the user would call .getFieldsReader() or similar to create one. it would be eligible for GC afterwards. but it means we can reuse buffers per-search and per-merge rather than creating per-document garbage. basically it would work like every other part of lucene. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] rmuir commented on pull request #11987: Make Decompressor release memory buffer
rmuir commented on PR #11987: URL: https://github.com/apache/lucene/pull/11987#issuecomment-1335456484 and one idea i have is to try to prototype with the term vectors first (since both stored fields and term vectors have per-segment threadlocals that I'd like to remove). it is just less tests to fix, but still gives us a chance to look at the idea. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] dnhatn commented on pull request #11984: Add exponential growth to TimeLimitingBulkScorer
dnhatn commented on PR #11984: URL: https://github.com/apache/lucene/pull/11984#issuecomment-1335560505 Thanks @costin! -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] dnhatn merged pull request #11984: Add exponential growth to TimeLimitingBulkScorer
dnhatn merged PR #11984: URL: https://github.com/apache/lucene/pull/11984 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] dnhatn closed issue #11676: Can TimeLimitingBulkScorer exponentially grow the window size? [LUCENE-10640]
dnhatn closed issue #11676: Can TimeLimitingBulkScorer exponentially grow the window size? [LUCENE-10640] URL: https://github.com/apache/lucene/issues/11676 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org - To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
[GitHub] [lucene] jpountz commented on a diff in pull request #11860: GITHUB-11830 Better optimize storage for vector connections
jpountz commented on code in PR #11860:
URL: https://github.com/apache/lucene/pull/11860#discussion_r1038244235
##
lucene/backward-codecs/src/java/org/apache/lucene/backward_codecs/lucene94/ExpandingVectorValues.java:
##
@@ -0,0 +1,49 @@
+/*
+ * 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.backward_codecs.lucene94;
+
+import java.io.IOException;
+import org.apache.lucene.index.FilterVectorValues;
+import org.apache.lucene.index.VectorValues;
+import org.apache.lucene.util.BytesRef;
+
+/** reads from byte-encoded data */
+public class ExpandingVectorValues extends FilterVectorValues {
Review Comment:
@benwtrent Can you make it pkg-private?
##
lucene/core/src/java/org/apache/lucene/codecs/lucene95/Lucene95HnswVectorsFormat.java:
##
@@ -0,0 +1,192 @@
+/*
+ * 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.codecs.lucene95;
+
+import java.io.IOException;
+import org.apache.lucene.codecs.KnnVectorsFormat;
+import org.apache.lucene.codecs.KnnVectorsReader;
+import org.apache.lucene.codecs.KnnVectorsWriter;
+import org.apache.lucene.codecs.lucene90.IndexedDISI;
+import org.apache.lucene.index.SegmentReadState;
+import org.apache.lucene.index.SegmentWriteState;
+import org.apache.lucene.search.DocIdSetIterator;
+import org.apache.lucene.store.IndexOutput;
+import org.apache.lucene.util.hnsw.HnswGraph;
+
+/**
+ * Lucene 9.5 vector format, which encodes numeric vector values and an
optional associated graph
+ * connecting the documents having values. The graph is used to power HNSW
search. The format
+ * consists of three files:
+ *
+ * .vec (vector data) file
+ *
+ * For each field:
+ *
+ *
+ * Vector data ordered by field, document ordinal, and vector dimension.
When the
+ * vectorEncoding is BYTE, each sample is stored as a single byte. When
it is FLOAT32, each
+ * sample is stored as an IEEE float in little-endian byte order.
+ * DocIds encoded by {@link IndexedDISI#writeBitSet(DocIdSetIterator,
IndexOutput, byte)},
+ * note that only in sparse case
+ * OrdToDoc was encoded by {@link
org.apache.lucene.util.packed.DirectMonotonicWriter}, note
+ * that only in sparse case
+ *
+ *
+ * .vex (vector index)
+ *
+ * Stores graphs connecting the documents for each field organized as a
list of nodes' neighbours
+ * as following:
+ *
+ *
+ * For each level:
+ *
+ * For each node:
+ *
+ * [vint] the number of neighbor nodes
+ * array[vint] the delta encoded neighbor ordinals
+ *
+ *
+ * After all levels are encoded memory offsets for each node's neighbor
nodes encoded by
+ * {@link org.apache.lucene.util.packed.DirectMonotonicWriter} are
appened to the end of the
+ * file.
+ *
+ *
+ * .vem (vector metadata) file
+ *
+ * For each field:
+ *
+ *
+ * [int32] field number
+ * [int32] vector similarity function ordinal
+ * [vlong] offset to this field's vectors in the .vec file
+ * [vlong] length of this field's vectors, in bytes
+ * [vlong] offset to this field's index in the .vex file
+ * [vlong] length of this field's index data, in bytes
+ * [int] dimension of this field's vectors
+ * [int] the number of documents having values for this field
+ * [int8] if equals to -1, dense – all documents have values for
a field. If equals to
+ * 0, sparse – some documents missing val
[GitHub] [lucene] benwtrent commented on a diff in pull request #11860: GITHUB-11830 Better optimize storage for vector connections
benwtrent commented on code in PR #11860:
URL: https://github.com/apache/lucene/pull/11860#discussion_r1038415190
##
lucene/core/src/java/org/apache/lucene/codecs/lucene95/Lucene95HnswVectorsWriter.java:
##
@@ -0,0 +1,753 @@
+/*
+ * 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.codecs.lucene95;
+
+import static
org.apache.lucene.codecs.lucene95.Lucene95HnswVectorsFormat.DIRECT_MONOTONIC_BLOCK_SHIFT;
+import static org.apache.lucene.search.DocIdSetIterator.NO_MORE_DOCS;
+
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.nio.ByteOrder;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.List;
+import org.apache.lucene.codecs.CodecUtil;
+import org.apache.lucene.codecs.KnnFieldVectorsWriter;
+import org.apache.lucene.codecs.KnnVectorsWriter;
+import org.apache.lucene.codecs.lucene90.IndexedDISI;
+import org.apache.lucene.index.*;
+import org.apache.lucene.index.Sorter;
+import org.apache.lucene.search.DocIdSetIterator;
+import org.apache.lucene.store.IndexInput;
+import org.apache.lucene.store.IndexOutput;
+import org.apache.lucene.util.*;
+import org.apache.lucene.util.hnsw.HnswGraph;
+import org.apache.lucene.util.hnsw.HnswGraph.NodesIterator;
+import org.apache.lucene.util.hnsw.HnswGraphBuilder;
+import org.apache.lucene.util.hnsw.NeighborArray;
+import org.apache.lucene.util.hnsw.OnHeapHnswGraph;
+import org.apache.lucene.util.packed.DirectMonotonicWriter;
+
+/**
+ * Writes vector values and knn graphs to index segments.
+ *
+ * @lucene.experimental
+ */
+public final class Lucene95HnswVectorsWriter extends KnnVectorsWriter {
+
+ private final SegmentWriteState segmentWriteState;
+ private final IndexOutput meta, vectorData, vectorIndex;
+ private final int M;
+ private final int beamWidth;
+
+ private final List> fields = new ArrayList<>();
+ private boolean finished;
+
+ Lucene95HnswVectorsWriter(SegmentWriteState state, int M, int beamWidth)
throws IOException {
+this.M = M;
+this.beamWidth = beamWidth;
+segmentWriteState = state;
+String metaFileName =
+IndexFileNames.segmentFileName(
+state.segmentInfo.name, state.segmentSuffix,
Lucene95HnswVectorsFormat.META_EXTENSION);
+
+String vectorDataFileName =
+IndexFileNames.segmentFileName(
+state.segmentInfo.name,
+state.segmentSuffix,
+Lucene95HnswVectorsFormat.VECTOR_DATA_EXTENSION);
+
+String indexDataFileName =
+IndexFileNames.segmentFileName(
+state.segmentInfo.name,
+state.segmentSuffix,
+Lucene95HnswVectorsFormat.VECTOR_INDEX_EXTENSION);
+boolean success = false;
+try {
+ meta = state.directory.createOutput(metaFileName, state.context);
+ vectorData = state.directory.createOutput(vectorDataFileName,
state.context);
+ vectorIndex = state.directory.createOutput(indexDataFileName,
state.context);
+
+ CodecUtil.writeIndexHeader(
+ meta,
+ Lucene95HnswVectorsFormat.META_CODEC_NAME,
+ Lucene95HnswVectorsFormat.VERSION_CURRENT,
+ state.segmentInfo.getId(),
+ state.segmentSuffix);
+ CodecUtil.writeIndexHeader(
+ vectorData,
+ Lucene95HnswVectorsFormat.VECTOR_DATA_CODEC_NAME,
+ Lucene95HnswVectorsFormat.VERSION_CURRENT,
+ state.segmentInfo.getId(),
+ state.segmentSuffix);
+ CodecUtil.writeIndexHeader(
+ vectorIndex,
+ Lucene95HnswVectorsFormat.VECTOR_INDEX_CODEC_NAME,
+ Lucene95HnswVectorsFormat.VERSION_CURRENT,
+ state.segmentInfo.getId(),
+ state.segmentSuffix);
+ success = true;
+} finally {
+ if (success == false) {
+IOUtils.closeWhileHandlingException(this);
+ }
+}
+ }
+
+ @Override
+ public KnnFieldVectorsWriter addField(FieldInfo fieldInfo) throws
IOException {
+FieldWriter newField =
+FieldWriter.create(fieldInfo, M, beamWidth,
segmentWriteState.infoStream);
+fields.add(newField);
+return newField;
+ }
+
+ @Override
+ public void flush(int maxDoc, Sorter.DocMap sortMap) throws IOException {
+for (FieldWriter field : fields) {
+ if (s
[GitHub] [lucene] benwtrent commented on a diff in pull request #11860: GITHUB-11830 Better optimize storage for vector connections
benwtrent commented on code in PR #11860:
URL: https://github.com/apache/lucene/pull/11860#discussion_r1038420089
##
lucene/core/src/java/org/apache/lucene/util/hnsw/HnswGraph.java:
##
@@ -144,6 +145,10 @@ public NodesIterator(int size) {
this.size = size;
}
+public int[] copy() {
+ return ArrayUtil.copyOfSubArray(nodes, 0, size);
+}
Review Comment:
@jpountz we could, but then we lose the intrinsics. I did some performance
testing and found that using intrinsics here with ArrayUtil is way better and
without it flush time is increased significantly. Maybe it would be better for
the caller to provide an array and have the iterator populate it in a function?
Would you prefer:
```java
public void consume(int[] dest) {
System.arraycopy(nodes, 0, dest, 0, dest.length);
}
```
Or something? Or would you rather have the runtime cost on flush? (which
seems unjustified to me)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
-
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
