jtibshirani commented on a change in pull request #601:
URL: https://github.com/apache/lucene/pull/601#discussion_r784338686
##########
File path:
lucene/core/src/java/org/apache/lucene/codecs/lucene90/Lucene90HnswVectorsWriter.java
##########
@@ -145,6 +139,64 @@ public void writeField(FieldInfo fieldInfo,
KnnVectorsReader knnVectorsReader)
throw new IllegalArgumentException(
"Indexing an HNSW graph requires a random access vector values, got
" + vectors);
}
+
+ long vectorDataLength = vectorData.getFilePointer() - vectorDataOffset;
+ long vectorIndexLength = vectorIndex.getFilePointer() - vectorIndexOffset;
+ writeMeta(
+ fieldInfo,
+ vectorDataOffset,
+ vectorDataLength,
+ vectorIndexOffset,
+ vectorIndexLength,
+ count,
+ docIds);
+ writeGraphOffsets(meta, offsets);
+ }
+
+ @Override
+ public void mergeField(FieldInfo fieldInfo, MergeState mergeState) throws
IOException {
+ if (mergeState.infoStream.isEnabled("VV")) {
+ mergeState.infoStream.message("VV", "merging " + mergeState.segmentInfo);
+ }
+
+ writeVectorDataPadding();
+ long vectorDataOffset = vectorData.getFilePointer();
+
+ // write the merged vector data to a temporary file
+ VectorValues vectors = MergedVectorValues.mergeVectorValues(fieldInfo,
mergeState);
+ IndexOutput tempVectorData =
Review comment:
@mayya-sharipova I think that in order to open the index input, we need
to close the vector data file first. But then it would need to be reopened
again if there are more fields. This felt a bit messy/ fragile to me. Let me
know if you have a better idea.
> We could avoid writing a temporary file by overriding `merge` rathen than
`mergeField`, first write merged vectors for all fields, and then compute HNSW
graphs for all fields
Maybe I could also look into this in a follow-up. It relates to the question
of whether we should share logic between flushing and merging -- if we
specialized the merge strategy like this, it could be harder to share with
`writeField`?
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
