luyuncheng commented on issue #13798: URL: https://github.com/apache/lucene/issues/13798#issuecomment-4406666252
hi, @benwtrent @jpountz we try to use [#13190](https://github.com/apache/lucene/pull/13190) parallel forcemerge segments. and we found this issue comes from [#13475](https://github.com/apache/lucene/pull/13475) it occurs as followings stack: when parallel merge in docvalues: 1. clone merge state: https://github.com/benwtrent/lucene/blob/92eb78d6a23e01a0a00bb065a1b9933d167291fe/lucene/core/src/java/org/apache/lucene/index/SegmentMerger.java#L204-L210 ``` private void mergeDocValues( SegmentWriteState segmentWriteState, SegmentReadState segmentReadState) throws IOException { MergeState mergeState = mergeState(); try (DocValuesConsumer consumer = codec.docValuesFormat().fieldsConsumer(segmentWriteState)) { consumer.merge(mergeState); } } ``` 2. new instance `MergeState mergeState = mergeState();` would clone readers ``` if (docValuesProducers[i] != null) { docValuesProducers[i] = docValuesProducers[i].getMergeInstance(); } ``` getMergeInstance return new `Lucene90DocValuesProducer` https://github.com/benwtrent/lucene/blob/cd88a97a63a97dc857d692ec3eb73b229bb5f372/lucene/core/src/java/org/apache/lucene/codecs/lucene90/Lucene90DocValuesProducer.java#L168-L181 3. buffer reset new `Lucene90DocValuesProducer` it would do IndexInput data clone: `this.data = data.clone();` when IndexInput is `NIOFSIndexInput` it would do `BufferedIndexInput#clone()` when parallel get `MergeState mergeState = mergeState();` and buffer reset to `EMPTY_BYTEBUFFER` -- 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]
