Hacker News Reader

Comment by krakatau1 | Hacker News Reader

I don't have a lot of experience with Oracle but I can tell you that Postgres optimizer is shit compared to Db2 zOS or Db2 LUW.

When I worked in a large bank we tried to migrate core system from Db2 zOS to Postgres and it went nowhere. I was a in-house developer working with Postgres consultants and they were amazed by db2 performance in OLTP scenarios.

So if your organization is already spending cash on Oracle, Db2 or MSSQL, use them for superior performance. Migration off them is costly and risky process.

If your working at a startup there is absolutely no reason to choose anything but Postgres if you need relational.

View on HN

I've never understood why RDBMSes don't offer a low-level protocol where, rather than a SQL statement, you can just send over the exact query-plan AST you want the DB to prepare in some binary encoding. Then you could pre-compile your hot OLTP queries and heavy OLAP reports offline against your existing DB schema, using the same sort of techniques that got Lucene its optimized Levenstein-automata JVM bytecode. You could even tweak the resulting plan, op by op, before letting it go to the DB—as if you were doing final ASM tweaks on a game in the 90s.

In fact, my company wants to spare the expenses on Oracle. This is the main reason for migrating. We'll see how it works out. You are not the first one to point out postgre's optimizer. Is it really that bad?

YMMV. That depends on your workload, how your data is structured and a million other things. I know a few shops with heavy usage that couldn't be happier.

It may require handholding if you are not happy with the plans it is generating. Also the quality of the query planner results will depend a lot on how up to date the statistics are.

I see, thanks!

> Is it really that bad?

It's not that it's bad, but Oracle/MSSQL have had the benefit of decades of corporate muscle, researchers, and Fortune 500 clients to help pave the way.

Reply on news.ycombinator.com