[twoodfin]

NDB is a specialized cluster database where all secondary indexes have to fit in the cluster's RAM(!).

That’s not that wild of a design principle: It’s been longstanding best practice to scale OLTP databases (and limit indexing) as to keep secondary indexes buffered in memory.

[johannes1234321]

To explain that statement: Historically ndb worked in a way that all data had to be in memory and it wouldn't touch the disk at all. For a while now data can be on disk, but all indexes afaik still have to be fully loaded into memory in startup.

The use cases are systems where you need "five nines" of uptime and fat responses. Coming out of Ericsson the classic area where it is used is in Telco (for instance "home location registers", the database recoding in which cell a mobile phone currently is, often use ndb) but there are different usages in Web (i.e. Session Store), "real time" information exchange (betting, gaming, ...) and so on.

It is not as easy to administer as a "normal" MySQL, but when deployed carefully it is powerful, fast and scalable (both locally as well as geo l-level)