You’re trying to go local quorum in one dc to local quorum in the other dc without losing any writes?
The easiest way to do this strictly correctly is to take the latency hit and do quorum while you run repair, then you can switch to local quorum on the other side. A few more notes inline > On Mar 15, 2020, at 10:15 PM, manish khandelwal > <manishkhandelwa...@gmail.com> wrote: > > > While switching over datacenters, there is a chance of mutation drop because > of which inconsistency may occur. > > To avoid inconsistency we can do following : > > Monitor and if require then run repair > > 1. Monitor tpstats in all nodes. If dropped message count is 0, it can > be inferred no mutation drop has occurred. > 2. Hints Replay - All hints have been replayed successfully. We need to > be sure there was no hints drop. > 3. If above monitoring suggests that no messages were dropped and hints > were replayed successfully across all nodes, then there is no need to run > repair. > > I would like to know > > 1. When hints replay fail will tpstats matrix record that as dropped message? No, failed hints are not dropped messages > 2. Is there any standard way to avoid repair while datacentere switch? Use three data centers and use quorum all the time --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@cassandra.apache.org For additional commands, e-mail: user-h...@cassandra.apache.org
