SQL – Add Column If Match Else on Aggregate Data

Answer to question

SELECT DISTINCT ON (name, zonedistrict_id)
       ST_Union(geom) as geom, gid, name, zonedistrict_id, zonestyle_id, longname 
FROM   zones
GROUP  BY gid, name, zonedistrict_id, zonestyle_id, longname
ORDER  BY name, zonedistrict_id, zonestyle_id;

It depends on what you are actually trying to achieve and what version of Postgres you are using and the table definition of the underlying table. You chose not to disclose any of that.

This gives you one distinct row per (name, zonedistrict_id), but geom is only the aggregate over (gid, name, zonedistrict_id, zonestyle_id, longname)

In older versions you had to list every non-aggregated column of the SELECT list in the GROUP BY list. This changed somewhat with PostgreSQL 9.1. I quote the release notes:

Allow non-GROUP BY columns in the query target list when the primary key is specified in the GROUP BY clause (Peter Eisentraut)

My guess

However, I suspect you secretly want this:

SELECT DISTINCT ON (name, zonedistrict_id)
       ST_Union(geom) OVER (PARTITION BY name, zonedistrict_id) AS geom
     , gid, name, zonedistrict_id, zonestyle_id, longname 
FROM   zones
ORDER  BY name, zonedistrict_id, zonestyle_id;

Using a ST_Union(geom) as aggregate window function. So geom is actually the aggregate over (name, zonedistrict_id), which seems more likely.

This should be possible since the PostGis manual tells us about its aggregate functions:

... can be used just like any other sql aggregate function such as sum, average.

And SQL aggregate functions can be used as window functions.

The problem is that you are not joining your two tables based on band_id, as a result you are creating a Cartesian product. When you create this cartesian product, you are combining each row from your placing table to each row in the band table.

For example, if you have data similar to the following:

create table band
(
  band_id number
);

insert into band values (1);
insert into band values (2);
insert into band values (3);
insert into band values (4);
insert into band values (5);

create table placing
(
  band_id number,
  points number
);

insert into placing values (1, 10);
insert into placing values (1, 2);
insert into placing values (2, 1);
insert into placing values (3, 3);
insert into placing values (3, 89);
insert into placing values (4, 45);
insert into placing values (5, 63);

And you query the data without a join on the band_id column:

SELECT
  placing.points,
  band.band_id
FROM placing, band;

You are generating data similar to:

| POINTS | BAND_ID |
|--------|---------|
|     10 |       1 |
|      2 |       1 |
|      1 |       1 |
|      3 |       1 |
|     89 |       1 |
|     45 |       1 |
|     63 |       1 |

See Demo. As you can see band_id = 1 now has every single point value from the placing table even though your really only have points = 10 and points=2.

In order to get the correct result, you need to JOIN the two tables on the band_id column. Your query will be similar to the following:

SELECT
  SUM(p.points) AS sumpoints,
  b.band_id
FROM placing p
INNER JOIN band b
  on p.band_id = b.band_id
GROUP BY b.band_id
ORDER BY SUM(p.points) DESC;

See SQL Fiddle with Demo. This will give a result of:

| SUMPOINTS | BAND_ID |
|-----------|---------|
|        92 |       3 |
|        63 |       5 |
|        45 |       4 |
|        12 |       1 |
|         1 |       2 |