discuss causal consistency trade-offs as well

[rkuhn]

In addition to the notes on strong consistency, we should also talk about the best that we can achieve without consensus: causal consistency.

This point was raised by Adrian Colyer:

On the one hand we know that causal consistency is the best we can do. On the other hand, I think it might be very easy to make a system that cannot achieve causal consistency. Consider the following local-first principle: “Always keep as much data local as possible, allowing only the minimum necessary to leave the device.” That sounds pretty good to me as a principle.

Now suppose at some point we release a summary sketch of the data on our local device - e.g., a count. Some user sees a total count across devices, and then wants to see the individual records. The total count was 7, but the user is only shown 3 records, because the local device that contributed a count of 4 is now disconnected. This user isn’t seeing a causally consistent view. They have unbounded eventual consistency instead.