swamirishi commented on PR #9553: URL: https://github.com/apache/ozone/pull/9553#issuecomment-3863881834
> > > ... it completely depends on the transactions that happen together in the same double buffer ... > > > > > > @swamirishi , I agree. I mean that the current code does not have such transactions and we are not going to support such code. > > > ... improves the efficiency ... > > > > > > Let's talk about the correctness before talking about efficiency. As you mentioned in the description, the previously implementation (quoted below) indeed has a bug. So this change is very risky. > > > ... It also misses a case where a putKey/deleteKey can get added even though a deleteRange has been executed in the next batch after the following continuousDeleteRange batch. ... > > It was not a bug but just inefficient. We are good as long as the order of operations are correct. For instance: [PUT K1, DELETE RANGE K5- K6, PUT K5 DELETE RANGE K1- K4] is equivalent to [DELETE RANGE K5-K6, PUT K5, DELETE RANGE K1-K4] So corresponding to the Single Key Op PUT K1 the previous implementation was only looking into the very next DeleteRange operation i.e. DELETE RANGE K5-K6 and was performing the first set of operations if there is no intersection and the implementation proposed in this PR looks into optimizing it further and peeks into all the Delete Range that come after i.e. after PUT K1 there are 2 DELETE Range K5-K6 and DELETE RANGE K1-K4 and K1 has an intersection with DELETE RANGE K1-K4 so the operation PUT K1 can be skipped even if it is performed it is not an issue but just redundant. So the issue was never around correctness. We had done it the earlier way since the DELETE RANGE was only being used for Snapshot CREATE and there was going to be no intersection of DELETE RANGE with any of the operation beyond one transaction since SNAPSHOT CREATE splits the double buffer batch. This changes proposed here is for more generic use cases like @aryangupta1998 's PR change which is also a safe 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. 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]
