yuanlihan commented on issue #11256: URL: https://github.com/apache/druid/issues/11256#issuecomment-841148193
@jihoonson > But I want to think about it more because we have no test coverage on the long term effect of this change. As this change will be enabled by default, I understand your concern. I tried to confirm the correctness based on the following facts: - the set of segments in each server holder is immutable in each cycle - the sampling method ensures(without mathematical proof) that getting k segments in a single pass is equal to getting k segments in k passes(1 segment per pass) > It would be great if you can provide some long term test results. Actually I have shipped this change into our on-premise cluster(which contains millions of segments) in Jan 2021 and it works well. In the past several months, the default cost-based balancer manages segment loading among 5 different historical tiers well. During this period, we expanded the cluster by adding new historical nodes and decommissioned historical nodes for server maintenance. It seems hard to provide visual long term function testing result about this change. -- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@druid.apache.org For additional commands, e-mail: commits-h...@druid.apache.org
