+1 from me. The idea and solution is great!
Thanks. Duo Zhang <zhang...@apache.org> 于2021年12月18日周六 16:31写道: > There is a previous thread[1] to mention that the main development work on > HBASE-26233[2] is done. And now the testing work has been done too, so I > send this official vote to merge for HBASE-26233 back into master. > > For those who are not familiar with HBASE-26233, it is for re-implementing > the region replication framework, to decouple with the general replication > framework. You could see the design doc[3] for more details. I've also > opened a PR against master[4], PTAL if you have interest. > > You can see the final test result on HBASE-26487[5]. > In conclusion, there is no regression on performance when running PE. > Using the new tool introduced in HBASE-26540[6], the new implementation can > perform much better than the old implementation. When inserting 1M rows, > the max lag on secondary replica is 335ms, and the 99.9% lag is 3 only 3ms, > while the lag of the old implementation highly depends on the > replication.source.sleepforretries config but you should not set it to a > very small value as it will waste a lot of resources. > On IntegrationTestRegionReplicaReplication, both implementations can not > pass but the problems are not related to the changing part, so should not > be a blocker. Will file follow on issues to address these problems. > > The vote will open for at least 72 hours. > > Please vote: > +1: Merge the changes from HBASE-26233 to master > -1: Do not merge these changes because ... > > Thanks. > > 1. https://lists.apache.org/thread/t8w053kpy9pj1vg1x3c0kjjh4ongrxzy > 2. https://issues.apache.org/jira/browse/HBASE-26233 > 3. > > https://docs.google.com/document/d/1WPSBwRqIPgTrPRM2wtzAe-3XUvA9S7auWy3Fy8xuZLc/edit > 4. https://github.com/apache/hbase/pull/3891 > 5. https://issues.apache.org/jira/browse/HBASE-26487 > 6. https://issues.apache.org/jira/browse/HBASE-26540 >