I've always wondering why, in a world where PostgreSQL exists, would anyone use MySQL. New releases seems to only slightly close the gap between it and better RDBMSs while retaining quite a few kludges and ambiguities.

In the meantime PostgreSQL gets better at the same pace.

* Threaded connection model. Postgres still uses process-per-conn right? This is very painful in situations with high connection churn / fire-and-forget db conns, which are required in some circumstances (heavily sharded environments where keeping persistent connection pools between all app servers and db servers is not practical)

* Ecosystem and experienced talent pool. Among companies with the largest database fleets, a plurality (maybe even a majority) use MySQL, or at least started out on MySQL. This translates to a greater availability of engineers worldwide with experience using and automating giant sharded MySQL fleets, more exposure to MySQL's edge cases, more familiarity with very specific performance characteristics and tuning, etc.

* MySQL's pluggable storage engine API allows flexibility for interesting future possibilities. Facebook migrated part of their MySQL fleet's storage engine from InnoDB (btree-based) to MyRocks (LSM-based) and achieved incredible compression benefits as a result, which translates to huge cost savings, while still being much easier than moving to another DBMS altogether. And AFAIK Postgres does not have comparable compression levels to MyRocks at this time.

* Replication options. I'm sure Postgres will catch up here, but its logical replication support is still relatively new, whereas MySQL has always used logical replication and supports a lot of different topologies and configurations. (Although, I won't cite master-master support as many folks do... it is not a good setup in MySQL and has long been discouraged in the community.)

* InnoDB's use of a clustered index for primary key means that it performs better than a non-clustered storage solution for some (many?) workloads.

Don't get me wrong, Postgres is a wonderful database and absolutely has many qualities and features that MySQL lacks. But it's not the black-and-white situation that some make it out to be.

(Even though TokuDB is probably not the "best" choice for my workload, I love it for my text-heavy tables because its performance is still great and its on-disk compression ratios are superb.)

You could also ask why would anyone use Postgres? Either question is equally absurd. If MySQL and its forks work well enough for Google, Facebook, Uber, AirBNB, Alibaba and Linkedin, why should anyone take it on face value that Postgres is an unambiguously better choice?

That said I've toyed with switching in the past. Ten years ago, the MySQL query optimiser was fairly junk and I wanted to switch for that alone.[1] The main thing that held me back was that my application has hundreds—quite probably thousands—of SQL queries coded into it.

Despite "SQL" being a notional standard, even the most basic queries would need to be extensively reformatted to be compatible with Postgres. The effort required to rewrite all of these queries simply can't be justified, particularly because we simply don't have any problems with MariaDB. The query optimiser is a lot better now, such that I now rarely encounter situations where a sensible query doesn't yield an efficient execution plan.

[1] Between 2000 and 2007 I had spent a lot of time with MSSQL for a different project and I had grown to appreciate its excellent query optimiser and execution plan visualiser. Ten years on and I still haven't seen any F/OSS solution that does as good a job as SQL Server 2000 in this regard. The most recent version I've used is 2005 so it might be even better today. And believe me, I was not a fan of Microsoft in the 2000s.

https://eng.uber.com/mysql-migration/

The TL;DR is that PostgreSQL on-disk storage format leads to much higher IOPs for the same workload. For companies running OLTP at scale, the cost differences can be huge.

Their opinion was mostly "Uber could have just asked us the right way to do it" and "We believe Uber's new engineering leader needed something to do"

2. WordPress.

3. If you need something simple and basic, MySQL is much easier to get up and running.

In MySQL there are five steps: Install MySQL (creating root user at the same time), CREATE DATABASE X, USE DATABASE X, CREATE table Y, work.

If I need something more secure, I'll add a "Create User"/add permissions step in there.

For some reason, step 2-4 (create user and set up database/table) give me a headache in Postgres.

`su postgres` or whatever user it uses;

run `psql`;

`CREATE USER X IDENTIFIED BY PASSWORD;`

`CREATE DATABASE X OWNER X;'

Now X can connect there and add tables as he wishes.

Postgres onboarding coule be better, but it is naive to choose a tool you will use for years based on the first 5 minutes experience.

Googled that and not one result on the first page mentioned PostgreSQL. I don't think this approach is a common one at all. More to the point, I'm not sure i've ever seen that set of instructions anywhere, ever, for pgsql. Every tutorial and intro focuses on running 'createuser'.

Oh, yeah, and you'll need to be root to go in and futz with your pg_hba.conf file, then have permissions to write it out and restart the server (probably as root).

This 'createuser' approach hasn't been a '5 minute experience' - it's been the approach that is promoted as the default answer/approach almost everywhere I've ever looked for several years.

    \help CREATE USER

    \help CREATE DATABASE

I rarely come across tutorials for beginners where MySQL isn't treated like the only option, even if only by omission. I wouldn't be surprised if students only found out about the viability of different databases after a few years.

Don’t get me wrong, I prefer Postgres generally too, but I can definitely see wins to MySQL on ease-of-use. Especially the less standard but more forgiving SQL syntax and simple DESCRIBE commands. Yes the quirks add up for MySQL but Postgres has its share of quirks too.

It is optimized in MyISAM via table-level metadata, which is only possible because MyISAM doesn't use MVCC or support transactions.

In InnoDB, you can get an estimated row count from SHOW TABLE STATUS, but the estimate is based on table stats and can be wildly inaccurate.

I am a solid Postgres guy, but MySQL still has a better multi-master replication story. Confident that Postgres will eventually surpass it, possibly even in a later iteration of 10.

Now in a world where Postgres exists why would anyone use MongoDB is a valid question...

The version numbering was changed with 10

https://mariadb.com/kb/en/library/spatial-index/

At this point MySQL is playing catch-up to MariaDB.

MySQL was first in offering JSON with 5.7. It is still well ahead, for example, there is no JSON_TABLE, aggregate functions, native type or partial update in MariaDB. Some examples of the value here: https://mysqlserverteam.com/mysql-8-0-from-sql-tables-to-jso...

Both have always had GIS support, but have taken it in a different direction. With MySQL, the GIS is now based on Boost.Geometry and has Geography support.

w.r.t. metadata in transactions - I am not sure which feature you refer to here. The native innodb-based dictionary is a MySQL only feature.

For Window functions and CTEs, they were added at a similar time, but MySQL is on a ~3yr release cycle. MariaDB is on a ~1yr.

The same complaint that every single person brings up about MySQL when being compared to Postgres, including in this thread: DDL changes made during transactions cannot be rolled back.

Source: have code that works on both.

* The JSON datatype guarantees the content is valid JSON

* access to fields can be faster even without dynamic/virtual columns

* since 8.0 there are special optimizations for doing partial updates, which means that not the full JSON data has to be rewritten on small changes https://mysqlserverteam.com/partial-update-of-json-values/

I think Mariadb was there first. I think the format is very similar at the bit level, but the MySQL has nicer API with automatic returning it as text to apps rather than exposing the bits. I dislike json_extract and wish >> worked but it doesn't seem to on my percona server :(

Anyhow, you'll look back when you've got more distance from Oracle and realize that you don't need to try and convince yourself that MySQL was better quite so hard ;)

I wish we could be more civil and not resort to petty insults. Is there anything I wrote you can dispute with a source?

  * Proper native JSON support
  * Spatial support lags behind MySQL

I have also thought about using TokuDB but SuSE (13.2) didn't ship a module for it then. I have to check again in 42.3.

The only real issues with MariaDB is lack of proper JSON and spatial support which would be of great use to us. I'm talking about a real JSON data type with validation and means to query and maybe even replace JSON, not a mere alias for the LONGTEXT data type.

We intend to switch to Postgres on the long run. It has everything we want, moreso with TimescaleDB extensions which could proove very useful for storing time series data. We are currently using it for the spatial database.