When a couchbase cluster is configured for memory first write which is the default where replication happens asynchronously after caller is acknowledged ,if a particular master shard fails is there any way to recover the data which were already there in memory but not asynchronously replicated yet ?
Does couchbase write to some sort of log file in addition to memory to recover on shard failure so that auto re-balancing can restore lost data ?
Or shard failure is prone to data loss on memory first approach?
Hopefully nodes failing in your cluster is not a situation that’s occurring frequently, but it can happen. There are some options for persistence and durability that you can use (with performance trade-offs) if you want to completely avoid any possibility of data loss during that (usually very short) period between being written to memory and being written to disk. You can pick and choose when to use and when not to use these options on a per-operation basis.