2003 was when I picked PG as preferred. Since then I've continues to work with DBs and of course evaluated Maria at least every 18 months (or a client was paying me to do this). PG continues to win for nearly every case I throw at it. Like when NoSQL was a fad (why tho!?) but PG had native json, then jsonb types. I didn't even have to look at that silly Mongo, PG baba!
You have a problem if you're trying to index every column...
I'm curious how innodb solves this, given that the theory of indices pretty much demands write amplification by definition. Postgres' WA is worse?
Yeah - they indexed every single column.... also - what may not have worked for uber, will be just fine for 99% of usecases. How often do you work with that kind of load. What they moved to isn't so much mysql, rather mysql used as a key value store with 0 relational database usage. Not really a postgres issue, rather one of being too big for relational databases.