CAP Theorem and PACELC Explained in Plain English

Every distributed database gets a label: Cassandra is AP, HBase is CP, and PostgreSQL is CA. The letters are supposed to tell you what it gives up when the network breaks: AP keeps answering but may serve stale data, CP stays correct but may stop answering, CA assumes the network never breaks. And people repeat them like facts. Most of those labels are wrong, or at least too sloppy to be useful. People call MongoDB and Redis both CP, and both shipped default configurations that lost committed writes during a partition in independent testing (MongoDB, Redis). The label said one thing, the database did another. These labels all come from the CAP theorem. I'll walk through what CAP actually proves, why it doesn't give you the full picture, and why PACELC might be closer to the truth. By the end, I want you to look at those AP/CP stickers and wonder if they ever meant much at all. The CAP Theorem Read the full post