> For any anomaly in a stateful service, it is an anomaly only if clients can observe it one way or the other. Otherwise, we argue that it doesn’t matter at all.
Thanks for the comment! I stand by this claim; I would love to hear more about why an anomaly (any anomaly) matters if it can't possibly be observed by any client for stateful service.
The claim itself is fine because it is kind of tautological. An error that cannot be observed in any way is probably not an error. However, you have to prove that it can't be observed.
This is quite different than what the described system does though. The system flags errors that have been observed. The correct claim to go with what the article does would thus be:
> For any anomaly in a stateful service, it is an anomaly only if the current clients running their current operations happen to observe it. Otherwise, we argue that it doesn't matter at all.
The exact same argument applies to normal exceptions and errors. It might be the case that out of quadrillions of queries a day, just none of them hit the error path, and hence we won't observe any issues.
I can see that in this case, if it matters or not becomes somewhat subjective. It's a good point though!
Thanks for the comment! I stand by this claim; I would love to hear more about why an anomaly (any anomaly) matters if it can't possibly be observed by any client for stateful service.