For the last decade or so, since I first became acquainted with databases, the HN crowd has said “Use PostgreSQL.” And every few weeks there’s a company blog post mentioning that they use MySQL. Why is this?
MySQL was basically the default database for developing web applications from 1998-2013. (It is the "M" in the LAMP stack) It gained this position by being free to use, reliable and stable. This began a virtuous cycle where more companies catered to MySQL users. Deploying, managing, and MySQL was easier since everyone catered to the audience which drove a virtuous cycle of more developers using MySQL.
A lot PostgreSQL's popularity can be traced to Heroku where it was the default choice for a database. Heroku made it even easier for developers to deploy their applications to the internet. Instead of having some janky build process you would just type "git push heroku master" and you changes would be live in a matter of minutes. This ease of deployment drove the virtuous cycle for PostgreSQL.
For the web it's unlikely that choosing either database would be a mistake. They both are great options. Going into the technical reasons why I believe PostgreSQL is a superior if you don't know what to choose would be a separate post in and of itself, but I've already gone on long enough.
I would say for anyone starting out and worrying about how large a single node can scale, Postgres can run well on aa 64/128core server, 2TB ram and 20TB ZRAID6 with a chain of read replicas. This can be done out of the box on Postgres without much issue and can get many businesses quite far, but once you go to the lots and lots of TB, or you have specific write latency, consistency, or other requirements, you have to evaluate multiple databases against your company's specific usage patterns and data, as no benchmark will give you a good idea.
With a bunch of read replicas you have to deal with eventual consistency when you get replication lag as well as the inability to scale writes.
If you shard out horizontally you get write scaleability, and read scalability with much less chance of replication delay. You like likely be reading from the master node most of the time. You also gain a bunch of operational flexibility, and smaller failure domains.
I am just curious, how does Vitess do that?
I personally advise people to just start with PG and if they encounter very unique requirements that for some reason PG can’t be tuned to - then do the switch.
As a developer I’ve used MySQL a lot, installed it myself in local environment and used it with fairly large tables. It’s very reliable and I’m generally comfortable with it. Once it’s running I don’t give it much thought frankly, so I’m not itching to change.