Re:Feature Casualties of Large Databases
by jeffbee on Tuesday December 01, 2020 @08:37PM
(#25272314)
I think a lot of developers who think they have referential integrity and strong consistency requirements are mislead. I've seen the case where for example User A is leaving Company Z and the developers have, for some reason, rigged it so removing A's access to all documents in the world has to happen in one giant cross-shard transaction, because it's consistent which means it's correct. But ... that's crazy. There's a hundred equally-correct ways to do it with tombstones and ordering and without cross-shard or even cross-row transactions. If you give a developer something that looks like BigTable they'll learn to deal with ordering, instead of using atomicity as a crutch.
[ Parent
| on HN ]
Cool let's implement a distributed database when it turns out there's no reason to.
That's not crazy - most people dont have access to enough stuff to leave one or two datastores, a consistent approach is fine.
Choosing 100 equally correct ways when one simple way works is ???