viliam-durina commented on PR #14016: URL: https://github.com/apache/lucene/pull/14016#issuecomment-2510926970
Our use case is to speed-up indexing of larger segments. We want to build fewer segments, so it makes sense to build them on multiple cores. We build the segments directly, not building smaller segments first, and then merging. The `HnswGraphSearcher` is used internally when building the graph. When the graph is built concurrently, you need to copy the `neighborArray` while holding the lock, which is implemented in a private inner class within `HnswConcurrentMergeBuilder`. This PR moves that inner implementation to `HnswGraphBuilder` - this way the locking logic is better encapsulated there - if a lock is given, it will use the concurrent searcher using the lock. If there's no lock, it will use the unsynchronized version. I understand this PR is for a non-standard use of the API and you might not care, and that the current structure works for your case. But I think the new structure is more logical anyway. -- 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
