Hacker News Reader

Hacker News

Comment by geenat | Hacker News Reader

Full thread

geenat

Bit of a nail in the coffin for me as of late.

* No async drivers.

* Total freezeup bug that only appears when MariaDB is under high load, since mid-2021 (10.6), not yet addressed.

* Postgres 15 outperforming MariaDB in both my own SELECT and INSERT benchmarks. Up to double performance for INSERT.

* Less and less compatibility with MySQL. Vitess dropping MariaDB.

* Postgres has a lighter setup, super fast compile in 2023. Standalone / portable Postgres is amazing: https://gist.github.com/gnat/cfe3754c3dc817c7fb8b2225ef4db62...

I do enjoy the MariaDB / MySQL tooling and syntax way more as a developer, and MaxScale etc. seems good, but there's only so much BS i'm willing to deal with when Postgres exists.

View on HN

Postgres was the first relational database I came into contact with, Interbase the second and Oracle the third.

Lately I've been writing a REST back end on top of MariaDB at work, and I swear it's going to drive me insane with its arbitrary limitations.

At least SQLite offers something in return that Postgres can't do.

> At least SQLite offers something in return that Postgres can't do.

If you're going to throw allegations around, at least get your facts straight. MariaDB and MySQL support complex replication topologies out of the box, including multi-master. Yes, it does come in useful if implemented correctly: we built a business application for a particular setup with many branches around the country, many of which have very unreliable network and must be able to continue working with a local database when that network is not available. It has been running for 4-5 years without issues.

They're also much easier to support: for example, upgrades are handled automatically — no need to dump and restore or resort to weird tricks; you can easily upgrade through ten major versions just by installing the latest version and restarting the daemon.

Some other features off the top of my mind: compression (zlib, lz4, lzo, zstd), versioned tables (an absolute life-saver if you need audit).

> we built a business application for a particular setup with many branches around the country, many of which have very unreliable network and must be able to continue working with a local database when that network is not available. It has been running for 4-5 years without issues.

How do you deal with reconciling after a connection loss? Is your client application read-only / the data sharded in a way that it doesn't matter for other branches if one branch is offline?

If you think that's bad, you weren't around for the MYiASM (I think it was.. days).

Yeah, for years, MySQL shipped with a default storage format that didn't enforce constraints.

To be fair,in those days MyISAM was pretty much the only DB engine that could keep up with the traffic growth rates of the early web, because it didn't enforce constraints. It was optimized for reads, not writes, and was fast enough to hold all state in one DB server for heavy traffic sites. Combined with stateless PHP and memory caching it was the best tool to use for this use case.

That it wasn't the best engine for all use cases and also corrupted data on disk occasionally is a different story. Once InnoDb was ready, you could just mix and match engines in one DB server as far as I remember and use referential constraints where needed.

Later MongoDB marketed itself as "web scale" due to no schema constraints at all. Fun times.

What kind of tooling does MariaDB/MySQL have that you miss when using Postgres? I'm not nearly as knowledgeable about the ecosystem as I am with Postgres.

I'm curious too, but my suspicion is that it's "the devil you know" at play here.

If you work with a tech for long enough you get the tooling "figured out" and things can feel like second nature. Maybe some alternative has equally "good" tooling (for however you define "good") but it may have slightly different quirks, strengths, and weaknesses. The fact that you have to figure that stuff out is itself a negative.

> What kind of tooling does MariaDB/MySQL have that you miss when using Postgres?

Easy upgrades? With MySQL it has never been more involved than running apt-get dist-upgrade/docker rm mysql && docker run mysql/kubectl deploy, depending on the stack. Downtimes of less than two minutes.

In contrast, PostgreSQL is such a hassle to upgrade that I only do it when forced to by an application dropping support for the current major version.

This 100%. I love PostgreSQL, but I rolled out MariaDB for my personal servers because I like not spending time doing weird things for upgrading the database software.

Not tooling, but the big one for me is the ability to reorder columns via ALTER statements.

Reply on news.ycombinator.com