Can I do a max(count(*)) in SQL?

Question

Here\'s my code:

    select yr,count(*)  from movie
join casting on casting.movieid=movie.id
join actor on casting.actorid = actor.id
where actor.name = \'Jo         


        

Answer 1

^{This question is old, but was referenced in a new question on dba.SE. I feel the best solutions haven't been provided, yet, so I am adding another one.}

First off, assuming referential integrity (typically enforced with foreign key constraints) you do not need to join to the table movie at all. That's dead freight in your query. All answers so far fail to point that out.

---

Can I do a max(count(*)) in SQL?

To answer the question in the title: Yes, in Postgres 8.4 (released 2009-07-01, before this question was asked) or later you can achieve that by nesting an aggregate function in a window function:

SELECT c.yr, count(*) AS ct, max(count(*)) OVER () AS max_ct
FROM   actor   a
JOIN   casting c ON c.actorid = a.id
WHERE  a.name = 'John Travolta'
GROUP  BY c.yr;

Consider the sequence of events in a SELECT query:

• Best way to get result count before LIMIT was applied

The (possible) downside: window functions do not aggregate rows. You get all rows left after the aggregate step. Useful in some queries, but not ideal for this one.

---

To get one row with the highest count, you can use ORDER BY ct LIMIT 1 like @wolph hinted:

SELECT c.yr, count(*) AS ct
FROM   actor   a
JOIN   casting c ON c.actorid = a.id
WHERE  a.name = 'John Travolta'
GROUP  BY c.yr
ORDER  BY ct DESC
LIMIT  1;

Using only basic SQL features available in any halfway decent RDBMS - the LIMIT implementation varies:

• SQL select elements where sum of field is less than N

Or you can get one row per group with the highest count with DISTINCT ON (only Postgres):

• Select first row in each GROUP BY group?

---

Answer

But you asked for:

... rows for which count(*) is max.

Possibly more than one. The most elegant solution is with the window function rank() in a subquery. Ryan provided a query but it can be simpler (details in my answer above):

SELECT yr, ct
FROM  (
   SELECT c.yr, count(*) AS ct, rank() OVER (ORDER BY count(*) DESC) AS rnk
   FROM   actor   a
   JOIN   casting c ON c.actorid = a.id
   WHERE  a.name = 'John Travolta'
   GROUP  BY c.yr
   ) sub
WHERE  rnk = 1;

All major RDBMS support window functions nowadays. Except MySQL and forks (MariaDB seems to have implemented them at last in version 10.2).