osdir.com


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: CockroachDB optimizer


I’m not very impressed with this work. A DSL for query transformation rules has been tried in the past - all the way back to EXODUS[1] - but there are so many complexities to deal with that it is better to write your rules in a full programming language.

They don’t explicitly talk about relational algebra, which means that they are still thinking they can do their transformations on the SQL AST. Which - as I discovered around 1999, writing my 3rd SQL parser and 2nd optimizer - is not going to work because SQL’s scoping rules make everything too brittle.

Julian

[1] http://pages.cs.wisc.edu/~dewitt/includes/queryopt/sigmod87.pdf

> On Nov 9, 2018, at 7:16 AM, Michael Mior <mmior@xxxxxxxxxx> wrote:
> 
> They're not using Calcite. Their code base is in Go and they wrote their
> own in Go. If you read the blog post, they describe a DSL they built for
> expressing optimizer rules which is kind of nice although not terribly
> different from what we're doing with Calcite.
> 
> --
> Michael Mior
> mmior@xxxxxxxxxx
> 
> 
> Le ven. 9 nov. 2018 à 08:00, Stamatis Zampetakis <zabetak@xxxxxxxxx> a
> écrit :
> 
>> Thanks for sharing this Michael!
>> 
>> I had a look on their blog post but I didn't notice something really novel.
>> It gives me the impression that they re-implemented a Volcano-style
>> optimizer like the one of Calcite.
>> 
>> Out of curiosity, are they using Calcite, or they really built everything
>> from scratch?
>> 
>> Best,
>> Stamatis
>> 
>> Στις Παρ, 9 Νοε 2018 στις 2:55 π.μ., ο/η Michael Mior <mmior@xxxxxxxxxx>
>> έγραψε:
>> 
>>> The folks from Cockroch Labs just shared an interesting blog post on the
>>> development of their optimizer. Could be some interesting lessons in
>> their
>>> code base.
>>> 
>>> https://www.cockroachlabs.com/blog/building-cost-based-sql-optimizer/
>>> 
>>> 
>>> --
>>> Michael Mior
>>> mmior@xxxxxxxxxx
>>> 
>>