jimczi commented on PR #16684: URL: https://github.com/apache/lucene/pull/16684#issuecomment-5810855604
@uschindler you designed the merge-time read advice for vectors in #13920, so I'd like your opinion on the approach here before I take it further. The merge maps the `.fdt` a second time and advises that one sequential, instead of re-advising the mapping searches are on. Two things pushed me off `updateIOContext`: it changes the advice under the searchers for as long as the merge runs, and it never reaches the directory, so `DirectIODirectory` and anything else routing on context can't act on a merge that borrows a search reader. The second open is one mapping per stored-fields reader that gets merged, opened lazily, page cache shared so nothing is cached twice. It belongs to the reader that made it and is closed with it, merge instances only get clones. If it holds up, I'd like to do the same for vectors: let a merge get its own decision, and let the directory be the one making it. -- 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]
