> every request received by a non-failing node in the system must result in a response
I don't believe that encodes latency
Instead, here's consistency:
> any read operation that begins after a write operation completes must return that value, or the result of a later write operation
"after a write operation completes" feels like where latency kicks in? Because within that space you can play around with completing a write operation to get what you want.
In a real system you can wait for a partition event to resolve itself before replying which preserves consistency at the cost of latency (availability).
This is true, but PACELC states that even if there is no partition you still have a consistency vs latency tradeoff (because the processes that guarantee consistency eat up latency in form of network roundtrips)
Availability in CAP is purely a software decision, it's an algorithmic property. Roughly speaking, it is the decision to continue accepting requests during a partition, possibly sacrificing Consistency. If your refuse requests during a partition, you conserve Consistency, and you lose Availability. If you keep accepting requests during a partition, it's the opposite.
High latency is considered a partition in CAP. Any hardware error, any network trouble, any bug, crash, any unreachable machines is never an Availability issue.
I agree
> High latency is considered a partition in CAP.
Can you support this claim? I don't think this is true unless you're specifically talking about high latency between nodes rather than latency between the sender and the system.
Yes, that's what I meant. To clarify, latency between the sender and the system would still not be an availability issue, but it also wouldn't be a partition. It's just out of scope of the CAP theorem. The user might be on a bad mobile connection, but it doesn't impact the distributed system itself.
PACELC only says that even during normal operation you still need to make tradeoff between availability and latency whereas CAP only deals with a tradeoff during a partition event.