CODEWITHSUNDEEP
mysqlsqlgroup-byhavinghaving-clause
fredoverflow
fredoverflow

Reputation: 263240

Is the HAVING clause redundant?

The following two queries yield the exact same result:

select country, count(organization) as N
from ismember
group by country
having N > 50;

select * from (
  select country, count(organization) as N
  from ismember
  group by country) x
where N > 50;

Can every HAVING clause be replaced by a sub-query and a WHERE clause like this? Or are there situations where a HAVING clause is absolutely necessary/more powerful/more efficient/whatever?

Upvotes: 14

Views: 1402

Answers (6)

Tobias N. Sasse
Tobias N. Sasse

Reputation: 2557

Logically yes the result will be the same on the end. But performance might differ. The HAVING clause might lead the DB to change a different execution plan.

A note to the guys above (can't directly comment somehow) - the execution plan does not only depend on your query. It might also get adjusted by the DB depending on statistics, like table size etc on runtime. That said for DB2 at least...

Upvotes: 0

Eugen Rieck
Eugen Rieck

Reputation: 65304

There are 2 questions asked here: The answer to the first of which is yes: The resultset of a HAVING-laden query is identical to the resultset of the same query executed as a subquery, decorated with a WHERE clause.

The second question is about performance and expressivity - here we go heavily into implementation. On MySQL there is a thin red line, where the performance starts to drift apart: The moment the resultset of the inner query can no longer be held in memory. In this case, MySQL will create an on-disk representation of the inner query, then use the WHERE selector on it. This will not happen, if the HAVING clause is used, the disqualified group will be dropped from the result set.

This implies, that the higher the selectivity of the HAVING clause, the more performance relevance it has: Consider result set of a million rows of the inner query, that is reduce by the HAVING clause to 5 rows - it is very likely, that the result set of the inner query wouldn't be held in memory, but it is very likely, that the final result set would.

Edit

I had this once: The query selected the few outliers from a very evenly distributed table (Number of pieces produced on a physical machine in a workshop per day). I investigated because of the high IO-load.

Edit 2

Please keep in mind, that the query cache is not used for subqueries - IMHO a place development should focus more on - so the subquery pattern will not profit from the inner query being a cached result set.

Upvotes: 11

András Ottó
András Ottó

Reputation: 7695

I know, that you changed it from general to MySQL, but I would like to add here a (may usefull) note. With a little modification I tried your query in SQL Server 2008.

Just for anybody who wants more detail in it, the executionplan of the two query is even exactly the same in SQL Server 2008. So the optimalizer processing the two command on the same way with the same performance and estimations.

Upvotes: 1

Gert Arnold
Gert Arnold

Reputation: 109185

In Sql Server 2008 two similar queries have exactly the same execution plan:

enter image description here

I've also studied a lot of queries generated by Entity Framework (with SS 2008) and so far I never saw a query with a HAVING clause. Grouping queries with a condition on an aggregated result are always translated into a query with a sub query. I trust the ADO.Net team knows with they're doing...

Upvotes: 9

Rob Farley
Rob Farley

Reputation: 15849

The HAVING clause is very useful to avoid the added complexity of sub-queries. However, the two are logically equivalent and every HAVING clause can be rewritten using a sub-query as you have.

In case you're curious, you could also write every WHERE clause as a HAVING clause if you're prepared to take GROUP BY to the extreme.

Upvotes: 4

Vikdor
Vikdor

Reputation: 24134

IMHO, using the HAVING clause should be efficient because there would an additional pass on the worktable that contains the grouped results on top of which the filtering criteria is run, in the second case.

Upvotes: 0

Related Questions