Question

The following query hangs: (although subqueries perfomed separately are fine)

I don’t know how to make the explain table look ok. If someone tells me, I’ll clean it up.

select
sum(grades.points)) as p,  
from assignments 
left join grades using (assignmentID) 
where gradeID IN 

(select grades.gradeID 
from assignments 
left join grades using (assignmentID) 
where ... grades.date <= '1255503600' AND grades.date >= '984902400' 
group by     assignmentID order by grades.date DESC);

I think the problem is with the first grades table… the type ALL with that many rows seems to be the cause.. Everything is indexed.

I uploaded the table as an image. Couldn’t get the formatting right:
http://imgur.com/AjX34.png

A commenter wanted the full where clause:

explain extended select count(assignments.assignmentID) as asscount, sum(TRIM(TRAILING '-' FROM grades.points)) as p, sum(assignments.points) as t 
from assignments left join grades using (assignmentID) 
where gradeID IN 
(select grades.gradeID from assignments left join grades using (assignmentID) left join as_types on as_types.ID = assignments.type 
where assignments.classID = '7815' 
and (assignments.type = 30170 ) 
and grades.contactID = 7141 
and grades.points REGEXP '^[-]?[0-9]+[-]?' 
and grades.points != '-' 
and grades.points != '' 
and (grades.pointsposs IS NULL or grades.pointsposs = '') 
and grades.date <= '1255503600' 
AND grades.date >= '984902400' 
group by assignmentID 
order by grades.date DESC);

Answer 1

Suppose you use a Real Database (ie, any database except MySQL, but I’ll use Postgres as an example) to do this query :

SELECT * FROM ta WHERE aid IN (SELECT subquery)

a Real Database would look at the subquery and estimate its rowcount :

• If the rowcount is small (say, less than a few millions)

It would run the subquery, then build an in-memory hash of ids, which also makes them unique, which is a feature of IN().

Then, if the number of rows pulled from ta is a small part of ta, it would use a suitable index to pull the rows. Or, if a major part of the table is selected, it would just scan it entirely, and lookup each id in the hash, which is very fast.

• If however the subquery rowcount is quite large

The database would probably rewrite it as a merge JOIN, adding a Sort+Unique to the subquery.

However, you are using MySQL. In this case, it will not do any of this (it is gonna re-execute the subquery for each row of your table) so it will take 1000 years. Sorry.