nickva commented on PR #4984: URL: https://github.com/apache/couchdb/pull/4984#issuecomment-1992166433
> Right. But is this a real problem here? The folding of Mango execution statistics hinges upon addition, which is a commutative and non-associative operation, which even lends itself to parallelization. The collection of the pieces happen locally without any interaction between the workers. Probably I am missing something here. It is if we want make any correctness guarantees. It doesn't have as much to do with commutativity and non-associativity of operations but which statistics should contribute to the result. We agreed to two things as far I recall: 1) We liked Bob's idea of counting only [contributing workers](https://github.com/apache/couchdb/pull/4958#issuecomment-1918072757) 2) based on your comment, not allowing uncertainty > @nickva I am afraid we cannot allow any uncertainty in the statistics if downstream consumers wanted to build uses cases on top of this mechanism that involves billing. And pointing out that having separate interleaved stats messages based on random arrival order from worker seems to conflict with the correctness goal. Towards the conclusion of request we could end up accounting for stats messages out-of-band so to speak, workers which did work that didn't end up contributing to the final response. However, as we discussed, the chance of this is probably low and maybe not worth worrying about. I am merely clarifying that it does impact correctness. -- 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: notifications-unsubscr...@couchdb.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
