The "A" in CAP Theorem Is Completely Different from Infrastructure "Availability"
While learning about CAP theorem in distributed systems, I was confused by the meaning of "Availability". I discovered that infrastructure "availability" and CAP theorem "availability", despite using the same word, are entirely different concepts.
While learning about CAP theorem in distributed systems, I was confused by the term “Availability”. Following ACID and CAP “Consistency”, now it’s “Availability”.
Having had many opportunities to work with infrastructure, when I heard “availability”, I assumed it meant “making servers redundant so the system continues working even if one fails”. However, in the context of CAP theorem, it meant something completely different: “during network partition, whether to return an error or return stale data”.
I realized that despite using the same word “Availability”, it referred to entirely different concepts.
What I Noticed
While learning about CAP theorem, I noticed “Wait, doesn’t availability have two different meanings?” After investigating, they were indeed different.
Infrastructure Availability (What I Initially Thought)
This is about “whether servers are physically running”.
For example:
User → LB → Server A (running) Server B (stopped) ← Even if B is down, A can handle itThe goal is to prevent situations where users see “cannot connect” or “timeout”. This is about physical redundancy.
CAP Availability (What CAP Theorem Refers To)
This is about “how to behave during network partition”.
For example:
User → Server A (running) Server B (running) ← Both are running, but A-B communication is cutBoth servers are running. Users can connect to both. But communication between servers is cut.
In this situation:
- Prioritize consistency: Return an error (“I cannot guarantee data consistency right now, so I’m refusing the request”)
- Prioritize availability: Return stale data (“It might be slightly outdated, but here you go”)
CAP availability is about this choice.
Key Differences
My biggest discovery was these two points:
| Aspect | Infrastructure Availability | CAP Availability |
|---|---|---|
| Premise | None (establishing physical survival) | Infrastructure is running |
| ”Unavailable” state | Timeout, cannot connect | Error response (5xx) |
CAP availability assumes “infrastructure is running” - this was my biggest finding. Despite both using the word “availability”, the premises are completely different.
Summary
While learning about CAP theorem, I was confused by the meaning of “availability”. I discovered that infrastructure availability and CAP availability, despite using the same word, are problems at entirely different layers.
Specifically, CAP availability assumes “infrastructure is running”, and the behavior when unavailable differs (timeout vs 5xx error). Being aware of these two differences makes learning distributed systems much smoother.