PostgreSQL GROUP BY different from MySQL

Question

I've been migrating some of my MySQL queries to PostgreSQL to use Heroku. Most of my queries work fine, but I keep having a similar recurring error when I use group by:

ERROR: column "XYZ" must appear in the GROUP BY clause or be used in an aggregate function

What am I doing wrong?

MySQL which works 100%

SELECT `availables`.*
FROM `availables`
INNER JOIN `rooms` ON `rooms`.id = `availables`.room_id
WHERE (rooms.hotel_id = 5056 AND availables.bookdate BETWEEN '2009-11-22' AND '2009-11-24')
GROUP BY availables.bookdate
ORDER BY availables.updated_at

PostgreSQL error

ActiveRecord::StatementInvalid: PGError: ERROR: column "availables.id" must appear in the GROUP BY clause or be used in an aggregate function: SELECT "availables".* FROM "availables" INNER JOIN "rooms" ON "rooms".id = "availables".room_id WHERE (rooms.hotel_id = 5056 AND availables.bookdate BETWEEN E'2009-10-21' AND E'2009-10-23') GROUP BY availables.bookdate ORDER BY availables.updated_at

Ruby code generating the SQL

expiration = Available.find(:all,
    :joins => [ :room ],
    :conditions => [ "rooms.hotel_id = ? AND availables.bookdate BETWEEN ? AND ?", hostel_id, date.to_s, (date+days-1).to_s ],
    :group => 'availables.bookdate',
    :order => 'availables.updated_at')

Expected Output (from working MySQL query):

+-----+-------+-------+------------+---------+---------------+---------------+
| id  | price | spots | bookdate   | room_id | created_at    | updated_at    |
+-----+-------+-------+------------+---------+---------------+---------------+
| 414 | 38.0  | 1     | 2009-11-22 | 1762    | 2009-11-20... | 2009-11-20... |
| 415 | 38.0  | 1     | 2009-11-23 | 1762    | 2009-11-20... | 2009-11-20... |
| 416 | 38.0  | 2     | 2009-11-24 | 1762    | 2009-11-20... | 2009-11-20... |
+-----+-------+-------+------------+---------+---------------+---------------+
3 rows in set

Answer 1

MySQL's totally non standards compliant GROUP BY can be emulated by Postgres' DISTINCT ON. Consider this:

MySQL:

SELECT a,b,c,d,e FROM table GROUP BY a

This delivers 1 row per value of a (which one, you don't really know). Well actually you can guess, because MySQL doesn't know about hash aggregates, so it will probably use a sort... but it will only sort on a, so the order of the rows could be random. Unless it uses a multicolumn index instead of sorting. Well, anyway, it's not specified by the query.

Postgres:

SELECT DISTINCT ON (a) a,b,c,d,e FROM table ORDER BY a,b,c

This delivers 1 row per value of a, this row will be the first one in the sort according to the ORDER BY specified by the query. Simple.

Note that here, it's not an aggregate I'm computing. So GROUP BY actually makes no sense. DISTINCT ON makes a lot more sense.

Rails is married to MySQL, so I'm not surprised that it generates SQL that doesn't work in Postgres.

Answer 2

PostgreSQL is more SQL compliant than MySQL. All fields - except computed field with aggregation function - in the output must be present in the GROUP BY clause.

Answer 3

MySQL's GROUP BY can be used without an aggregate function (which is contrary to the SQL standard), and returns the first row in the group (I don't know based on what criteria), while PostgreSQL must have an aggregate function (MAX, SUM, etc) on the column, on which the GROUP BY clause is issued.

Answer 4

Correct, the solution to fixing this is to use :select and to select each field that you wish to decorate the resulting object with and group by them.

Nasty - but it is how group by should work as opposed to how MySQL works with it by guessing what you mean if you don't stick fields in your group by.

Answer 5

If I remember correctly, in PostgreSQL you have to add every column you fetch from the table where the GROUP BY clause applies to the GROUP BY clause.

Answer 6

Not the prettiest solution, but changing the group parameter to output every column in model works in PostgreSQL:

expiration = Available.find(:all,
:joins => [ :room ],
:conditions => [ "rooms.hotel_id = ? AND availables.bookdate BETWEEN ? AND ?", hostel_id, date.to_s, (date+days-1).to_s ],
:group => Available.column_names.collect{|col| "availables.#{col}"},
:order => 'availables.updated_at')

Answer 7

According to MySQL's "Debuking GROUP BY Myths" http://dev.mysql.com/tech-resources/articles/debunking-group-by-myths.html. SQL (2003 version of the standard) doesn't requires columns referenced in the SELECT list of a query to also appear in the GROUP BY clause.

Answer 8

For others looking for a way to order by any field, including joined field, in postgresql, use a subquery:

SELECT * FROM(
SELECT DISTINCT ON(availables.bookdate) `availables`.* 
FROM `availables` INNER JOIN `rooms` ON `rooms`.id = `availables`.room_id 
WHERE (rooms.hotel_id = 5056 
AND availables.bookdate BETWEEN '2009-11-22' AND '2009-11-24')
) AS distinct_selected
ORDER BY availables.updated_at

or arel:

subquery = SomeRecord.select("distinct on(xx.id) xx.*, jointable.order_field")
      .where("").joins(")
result = SomeRecord.select("*").from("(#{subquery.to_sql}) AS distinct_selected").order(" xx.order_field ASC, jointable.order_field ASC")

Answer 9

I think that .uniq [1] will solve your problem.

[1] Available.select('...').uniq

Take a look at http://guides.rubyonrails.org/active_record_querying.html#selecting-specific-fields