jbewing commented on PR #5373: URL: https://github.com/apache/hbase/pull/5373#issuecomment-1766897059
> do we have a way to identify meta lookups, so that we could revert to the original seeking approach in those cases? We could probably do this (or approximate it), but the problem becomes passing that context all the way down to the StoreFileScanner level. My current thought is that since the "patch w/ lazy optimization" didn't benchmark well (I believe because the scanner caching is 10 for region location lookups on master), that the "lazy engage" approach isn't going to work super well in the long term since those settings can change. My thought now is to just disable this optimization when scanning RIV1 storefiles since this patch will always incur an extra-seek penalty for that data block encoding which isn't strictly necessary because RIV1 doesn't suffer the "seek from beginning of the block" penalty that the other data block encodings do. The meta table always uses RIV1 so this should prevent any regressions in meta behavior while making reverse scans 50% faster in other data block encodings where the penalty for seeking from the beginning of the block is more expensive. -- 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]
