pinterest
consistency question for replication
hi, @newpoo @shuzhang1989 i have a question for rocksplicator replication recently. consider such case: if master node is down with the latest seq no for 100, and the latest seq no for slaves is 98, and the slave become master with latest seq no for 98. after a meanwhile, the master is up and become slave, and the new master's seq no is 101. according to seq no , the old master to be slave will pull data from the new master start with 100.
how can rocksplicator handle this case ? Is the older master will hold diff data with the new master for seq no 99 and 100 ？ thank you !
I'm interested in how rocksplicator will handle this scenario as well! For rocksdb-cloud from Rockset, it seems like they store rocksdb WAL in another distributed log system (like Kafka), so they can replay the events for these scenarios.
However, in the async replication style like rocksplicator, it does sound like that we may have consistency issues here if the master node crashes before these new events are received by the replicas or the replication workers.
