3 ms·

So where are you proposing the consensus be placed? A lot (I'd argue most) applications need consensus for them to be stable and reliable in the long term, so e

So where are you proposing the consensus be placed? A lot (I'd argue most) applications need consensus for them to be stable and reliable in the long term, so either you place it in the datastore or in the logic processing. Where else would you place it?

lmm 6y ago
I prefer to see the whole system as a succession of stream transformations (https://www.confluent.io/blog/turning-the-database-inside-out-with-apache-samza/ https://www.confluent.io/blog/turning-the-database-inside-ou...). If you view the sequence of input events as first-class and the "current state of the world" as derived, then a lot of problems go away. You need a datastore that can give you a consistent answer as to what order events occurred in, but it's a lot easier to make appending to an append-only list atomic than to make arbitrary state computations ACID. For downstream derived computations, basically you either make the causal relationship explicit, or accept that you have eventual consistency. The only case where you can have inconsistency is where you have a "diamond" in your computations (i.e. you compute B that's derived from A and C that's also derived from A, and you compute D that's derived from B and C). So you figure out the business implications and either accept it or eliminate the diamond (by computing (A, B) from A and (B,C) from (A, B) instead of computing B and C separately from A). You will also get inconsistency if you do some totally ad-hoc query that's not part of your existing pipelines, but usually that's the kind of reporting query that doesn't need to be 100% consistent; if you do need a consistent version of that then the best approach is to take a regular snapshot, which can be consistent. Basically you have a lot more precise control, you have causal relationships where you have explicit dependencies, so you have the level of consistency that you need for all your operational stuff, but you don't have a globally consistent realtime view of everything. It may take more work up front, but IME the notion of that kind of global consistency is a lie in a distributed world; even if you just have a basic webapp then you can't actually achieve the kind of consistency that that model pretends you have, because what the user is viewing in their browser (and then potentially making changes based on) at any given time is not necessarily the same as what's in the central database. To get back to the question, this means that if inconsistency is due to an actual bug, it's pretty easy to solve: fix the bug and then regenerate from the original events. If you don't understand why data is inconsistent then you can always look back to the source events and figure out what it should be.

lmm 6y ago

I prefer to see the whole system as a succession of stream transformations (https://www.confluent.io/blog/turning-the-database-inside-out-with-apache-samza/ https://www.confluent.io/blog/turning-the-database-inside-ou...). If you view the sequence of input events as first-class and the "current state of the world" as derived, then a lot of problems go away. You need a datastore that can give you a consistent answer as to what order events occurred in, but it's a lot easier to make appending to an append-only list atomic than to make arbitrary state computations ACID. For downstream derived computations, basically you either make the causal relationship explicit, or accept that you have eventual consistency. The only case where you can have inconsistency is where you have a "diamond" in your computations (i.e. you compute B that's derived from A and C that's also derived from A, and you compute D that's derived from B and C). So you figure out the business implications and either accept it or eliminate the diamond (by computing (A, B) from A and (B,C) from (A, B) instead of computing B and C separately from A). You will also get inconsistency if you do some totally ad-hoc query that's not part of your existing pipelines, but usually that's the kind of reporting query that doesn't need to be 100% consistent; if you do need a consistent version of that then the best approach is to take a regular snapshot, which can be consistent. Basically you have a lot more precise control, you have causal relationships where you have explicit dependencies, so you have the level of consistency that you need for all your operational stuff, but you don't have a globally consistent realtime view of everything. It may take more work up front, but IME the notion of that kind of global consistency is a lie in a distributed world; even if you just have a basic webapp then you can't actually achieve the kind of consistency that that model pretends you have, because what the user is viewing in their browser (and then potentially making changes based on) at any given time is not necessarily the same as what's in the central database. To get back to the question, this means that if inconsistency is due to an actual bug, it's pretty easy to solve: fix the bug and then regenerate from the original events. If you don't understand why data is inconsistent then you can always look back to the source events and figure out what it should be.