I'm genuinely curious what you mean here when you say high availability? Do you mean that during a distributed consensus failover timeframe, it is unacceptable for writes to be queued at the client?

It means that the system is still available when one node is dead.

When you say that there can't be sharding without more than one one node taking writes, I totally see where you're coming from. And I agree with you. Typically that means that there is a logical isolation between which data goes to which shard so this avoids conflicts too. Usually when people talk about "multimaster" I do not believe they're thinking of this particular scenario but I may be wrong. I'm not familiar but does postgres not shard?

There are two cases of multi-master.

- All nodes have all data (no sharding).

- No nodes have all data (sharding).