3 ms·
This is less about cap theorem and more about immutable data stores. It does make me wonder about log append only stores. Why is it a linear log at all? That i
by pyrolistical 3y ago
This is less about cap theorem and more about immutable data stores.
It does make me wonder about log append only stores. Why is it a linear log at all? That is one of the problems OP is talking about when the partition heals and you need to merge two disparate logs.
But why? Why not use an order independent data structure like a set? Adding facts to two independent sets and then union the two sets results in the same set. I suspect the reason is, we haven’t found a good way to store sets. We only figured out how to store things linearly
- mjb 3y agoThe system described here (if I understand correctly) could store sets if care is taken to make the set writes commutative. The CRDT literature contains several solutions to this problem.
- lmm 3y ago> It does make me wonder about log append only stores. Why is it a linear log at all? That is one of the problems OP is talking about when the partition heals and you need to merge two disparate logs. > But why? Why not use an order independent data structure like a set? Adding facts to two independent sets and then union the two sets results in the same set. I suspect the reason is, we haven’t found a good way to store sets. We only figured out how to store things linearly No, the reason is the opposite. Storing sets is easy, but encoding everything we might want to store as a set is difficult. E.g. it's very common to want to store a list and want to be able to append items to the list and retrieve it in order. Doing that on top of a linear log store is trivial; doing it on top of a set store is hard if not impossible.