PostgreSQL: Order by multiple column weights - ruby-on-rails

Question

I'm using PostgreSQL 9.4. I've got a resources table which has the following columns:
id
name
provider
description
category
Let's say none of these columns is required (save for the id). I want resources to have a completion level, meaning that resources with NULL values for each column will be at 0% completion level.
Now, each column has a percentage weight. Let's say:
name: 40%
provider: 30%
description: 20%
category: 10%
So if a resource has a provider and a category, its completion level is at 60%.
These weight percentages could change at any time, so having a completion_level column which always has the value of the completion level will not work out (there could be million of resources). For example, at any moment, the percentage weight of description could decrease from 20% to 10% and category's from 10% to 20%. Maybe even other columns could be created and have their own weight.
The final objective is to be able to order resources by their completion levels.
I'm not sure how to approach this. I'm currently using Rails so almost all interaction with the database has been through the ORM, which I believe is not going to be much help in this case.
The only query I've found that somewhat resembles a solution (and not really) is to do something like the following:
SELECT * from resources
ORDER BY CASE name IS NOT NULL AND
provider IS NOT NULL AND
description is NOT NULL AND
category IS NOT NULL THEN 100
WHEN name is NULL AND provider IS NOT NULL...
However, there I must per mutate by every possible combination and that's pretty bad.

Answer 1

Add a weights table as in this SQL Fiddle:
PostgreSQL 9.6 Schema Setup:
CREATE TABLE resource_weights
( id int primary key check(id = 1)
, name numeric
, provider numeric
, description numeric
, category numeric);
INSERT INTO resource_weights
(id, name, provider, description, category)
VALUES
(1, .4, .3, .2, .1);
CREATE TABLE resources
( id int
, name varchar(50)
, provider varchar(50)
, description varchar(50)
, category varchar(50));
INSERT INTO resources
(id, name, provider, description, category)
VALUES
(1, 'abc', 'abc', 'abc', 'abc'),
(2, NULL, 'abc', 'abc', 'abc'),
(3, NULL, NULL, 'abc', 'abc'),
(4, NULL, 'abc', NULL, NULL);
Then calculate your weights at runtime like this
Query 1:
select r.*
, case when r.name is null then 0 else w.name end
+ case when r.provider is null then 0 else w.provider end
+ case when r.description is null then 0 else w.description end
+ case when r.category is null then 0 else w.category end weight
from resources r
cross join resource_weights w
order by weight desc
Results:
| id | name | provider | description | category | weight |
|----|--------|----------|-------------|----------|--------|
| 1 | abc | abc | abc | abc | 1 |
| 2 | (null) | abc | abc | abc | 0.6 |
| 3 | (null) | (null) | abc | abc | 0.3 |
| 4 | (null) | abc | (null) | (null) | 0.3 |

Answer 2

SQL's ORDER BY can order things by pretty much any expression; in particular, you can order by a sum. CASE is also fairly versatile (if somewhat verbose) and an expression so you can say things like:
case when name is not null then 40 else 0 end
which is more or less equivalent to name.nil?? 0 : 40 in Ruby.
Putting those together:
order by case when name is not null then 40 else 0 end
+ case when provider is not null then 30 else 0 end
+ case when description is not null then 20 else 0 end
+ case when category is not null then 10 else 0 end
Somewhat verbose but it'll do the right thing. Translating that into ActiveRecord is fairly easy:
query.order(Arel.sql(%q{
case when name is not null then 40 else 0 end
+ case when provider is not null then 30 else 0 end
+ case when description is not null then 20 else 0 end
+ case when category is not null then 10 else 0 end
}))
or in the other direction:
query.order(Arel.sql(%q{
case when name is not null then 40 else 0 end
+ case when provider is not null then 30 else 0 end
+ case when description is not null then 20 else 0 end
+ case when category is not null then 10 else 0 end
desc
}))
You'll need the Arel.sql call to avoid deprecation warnings in Rails 5.2+ as they don't want you to order(some_string) anymore, they just want you ordering by attributes unless you want to jump through some hoops to say that you really mean it.

Answer 3

Sum up weights like this:
SELECT * FROM resources
ORDER BY (CASE WHEN name IS NULL THEN 0 ELSE 40 END
+ CASE WHEN provider IS NULL THEN 0 ELSE 30 END
+ CASE WHEN description IS NULL THEN 0 ELSE 20 END
+ CASE WHEN category IS NULL THEN 0 ELSE 10 END) DESC;

Answer 4

This is how I would do it.
First: Weights
Since you say that the weights can chage from time to time, you have to create an structure to handle the changes. It could be a simple table. For this solution, it will be called weigths.
-- Table: weights
CREATE TABLE weights(id serial, table_nane text, column_name text, weight numeric(5,2));
id | table_name | column_name | weight
---+------------+--------------+--------
1 | resources | name | 40.00
2 | resources | provider | 30.00
3 | resources | description | 20.00
4 | resources | category | 10.00
So, when you need to change categories from 10 to 20 or/and description from 20 to 10, you update this structure.
Second: completion_level
Since you say that you could have millions of rows, it is ok to have completion_level column in the table resources; for efficiency purposes.
Making a query to get the completion_level works, you could have it in a view. But when you need the data fast and simple and you have MILLIONS of rows, it is better to set the data by "default" in a column or in another table.
When you have a view, every time you run it, it recreates the data. When you have it already on the table, it's fast and you don't have to recreate nothing, just query the data.
But how can you handle a completion_level? TRIGGERS
You would have to create a trigger for resources table. So, whenever you update or insert data, it will create the completion level.
First you add the column to the resources table
ALTER TABLE resources ADD COLUMN completion_level numeric(5,2);
And then you create the trigger:
CREATE OR REPLACE FUNCTION update_completion_level() RETURNS trigger AS $$
BEGIN
NEW.completion_level := (
CASE WHEN NEW.name IS NULL THEN 0
ELSE (SELECT weight FROM weights WHERE column_name='name') END
+ CASE WHEN NEW.provider IS NULL THEN 0
ELSE (SELECT weight FROM weights WHERE column_name='provider') END
+ CASE WHEN NEW.description IS NULL THEN 0
ELSE (SELECT weight FROM weights WHERE column_name='description') END
+ CASE WHEN NEW.category IS NULL THEN 0
ELSE (SELECT weight FROM weights WHERE column_name='category') END
);
RETURN NEW;
END $$ LANGUAGE plpgsql;
CREATE TRIGGER resources_completion_level
BEFORE INSERT OR UPDATE
ON resources
FOR EACH ROW
EXECUTE PROCEDURE update_completion_level();
NOTE: table weights has a column called table_name; it's just in case you want to expand this functionality to other tables. In that case, you should update the trigger and add AND table_name='resources' in the query.
With this trigger, every time you update or insert you would have your completion_level ready so getting this data would be a simple query on resources table ;)
Third: What about old data and updates on weights?
Since the trigger only works for update and inserts, what about old data? or what if I change the weights of the columns?
Well, for those cases you could use a function to recreate all completion_level for every row.
CREATE OR REPLACE FUNCTION update_resources_completion_level() RETURNS void AS $$
BEGIN
UPDATE resources set completion_level = (
CASE WHEN name IS NULL THEN 0
ELSE (SELECT weight FROM weights WHERE column_name='name') END
+ CASE WHEN provider IS NULL THEN 0
ELSE (SELECT weight FROM weights WHERE column_name='provider') END
+ CASE WHEN description IS NULL THEN 0
ELSE (SELECT weight FROM weights WHERE column_name='description') END
+ CASE WHEN category IS NULL THEN 0
ELSE (SELECT weight FROM weights WHERE column_name='category') END
);
END $$ LANGUAGE plpgsql;
So everytime you update the weights or to update the OLD data, you just run the function
SELECT update_resources_completion_level();
Finally: What if I add columns?
Well, you would have to insert the new column in the weights table and update the functions (trigger and update_resources_completion_level()). Once everything is set, you run the function update_resources_completion_level() to set all weights acording to the changes :D