BASE & Eventual Consistency — Junior¶

At junior level, focus on this question:

What does BASE give up compared to ACID, and why would any system choose that on purpose?

BASE, word by word¶

Word	Meaning	Contrast with ACID
Basically Available	The system responds to every request, even during a failure or partition — possibly with stale or approximate data.	ACID systems may block or reject a request rather than serve inconsistent data.
Soft state	The system's state may change over time even without new input, as replicas converge.	ACID systems have a definite, immediately-settled state after each commit.
Eventually consistent	If no new writes occur, all replicas will eventually return the same value.	ACID's Isolation guarantees a precise, immediate view — no "eventually" involved.

flowchart LR subgraph ACID["ACID system"] direction TB A1[Write] --> A2["Every subsequent read\nsees it immediately\n(possibly after blocking)"] end subgraph BASE["BASE system"] direction TB B1[Write to node A] --> B2["Some reads see it\nimmediately (node A)"] B1 -.propagates.-> B3["Other reads see stale data\nuntil propagation completes\n(node B, node C)"] end

Why choose this on purpose?¶

A single-node ACID database can't survive a network partition and stay both available and consistent (this is the CAP theorem — see CAP Theorem). Systems built to span multiple data centers or survive node failures without downtime (Cassandra, DynamoDB, S3, most CDNs) choose availability over strict consistency, and BASE describes the resulting contract with application developers: "you will always get an answer, but it might be a few moments out of date."

🎓 Takeaway: BASE isn't a lesser version of ACID — it's a different set of priorities for a different problem: systems that must keep serving traffic even when part of the network is unreachable, at the cost of a brief window where different replicas disagree.

Test yourself¶

Give one real system you've used (as a user, not an engineer) where you've personally noticed eventual consistency (e.g. a "like" count that takes a moment to update everywhere).
Why can't a single-node relational database "choose" BASE — what has to be true about a system's architecture before BASE becomes a relevant choice at all?
What does "soft state" mean for data you've written but hasn't yet propagated everywhere — is it wrong, or just not-yet-everywhere?

Continue to middle.md.