MySQL isn't in flux. It is getting better every single day because of the contributions from Percona, MariaDB, Facebook, Twitter and even Oracle. It's just that people have a nervousness because of the Oracle connection.
And PostgreSQL maybe solid but developers want a solution that horizontally scales. Until this is addressed it is going to find itself increasingly marginalised.
No it's not like MySQL which has MySQL Cluster a supported, well documented, OOTB, easy to use solution with lots of enterprise customers. Similarly Percona has a very, very impressive product with great support.
Instagram had to roll their own and Postgres XC, PGPool both have always seemed pretty sketchy. No official support, no notable customers, shocking complexity and everything just seems all over the place with documentation from say 2009 referencing PostgreSQL 8 e.g. PgPool beginner guide. They may be great solutions but do they really look like something that inspires confidence ?
Pretty sure Skype scales PG horizontally pretty darn well. PgPool is probably the least used, least recommended replication method out there. Streaming replication in 9, WAL shipping, and things like Londiste (Skype) are what real large production users use.
Have you ever used mysql cluster in a bigger deployment? I wonder if it got usable over the last ~4 years when I tried it. At the time it was buggy, continuously corrupting data and had no real support apart from a very niche IRC channel. I've still got some very basic bugs open against it without any resolution at all.
You are right. MySQL will be used for years and years. And it is getting better every day.
But MySQL is still behind Postgres in features and reliability. I have being bitten by MySQL shortcomings for a year. Last month was the limit of max 64 indexes per table. Before that it was the dozens of stupid limitations in the procedures and functions. Before that it was some stupid bug that only happens in master-master replication. I can go on for days ranting about MySQL.
Think SVN vs Git. SVN can get better everyday and it does get better, but incremental upgrades will never convert SVN to Git. You scale MySQL by putting Memcached in front of it. But this makes MySQL a fast read-only DB. You can't scale MySQL in a OLTP setup.
Postgres had some performance and scalability issues about 3 or 4 years ago. But the foundations were solid. Now Postgres has fixed those issues and the foundations are better than ever. You can scale Postgres in a OLTP setup and it will eat MySQL lunch every time.
And PostgreSQL maybe solid but developers want a solution that horizontally scales. Until this is addressed it is going to find itself increasingly marginalised.