> 1. I need a read slave. Have you actually tested this need? It strikes me as unfair to assume a solution that works well for Galera is also necessary for Coc

> 1. I need a read slave. Have you actually tested this need? It strikes me as unfair to assume a solution that works well for Galera is also necessary for Cockroach. They have quite different architectures. CDB allows you to mix serializable and snapshot transactions. The latter should be quite efficient for read only workloads. You can create a user that only has only select privileges granted just like with most other SQL databases. I would not assume you need a 2nd read only daemon instance per node just based on your experience with another product. > 2. WAN sucks. ... If every site can accept data, then I have to assume that every site will remain in sync. This is not how CockroachDB works (or spanner style systems in general for that matter). You do not have to assume: CDB's replication is linearizable. You will not see things out of sync, and the scenario you describe will not happen. The worst you have to live with is the system becomes unavailable if you lose more nodes in a single replica group than can be tolerated with your quorum settings (ie, 1 out of 3 or 2 out of 5). > I'm not ragging on the product; it's a great good enough product for most use cases. I just more wanted to share my thoughts. If you can tolerate partial consistency, go for it. This is literally the opposite of the entire point of cockroach's design. You do not have to tolerate partial consistency. It's fully consistent. That's the whole point. I hope people reading along see more than your comment, otherwise they're going to walk away thinking cockroach is exactly the opposite of what it is.