Question

I have a data model like the following:

username | product1 | product2
-------------------------------
 harold     abc        qrs
 harold     abc        def   
 harold     def        abc
 kim        abc        def
 kim        lmn        qrs    
 ...

username | friend_username
---------------------------
 john       harold
 john       kim
 ...

I want to build a histogram of the most frequent product1 to product2 records there are, restricted to a given product1 id, and restricted only to friends of john. So something like:

What do friends of john link to for product1, when product1=’abc’:
Select all of john’s friends from the friends table. For each friend, count and group the number of records where product1 = ‘abc’, sort results in desc order:

Results:
abc -> def (2 instances)
abc -> qrs (1 instance)

I know we can do the following in a relational database, but there will be some threshold where this kind of query will start utilizing a lot of resources. Users might have a large number of friend records (500+). If this query is running 5 times every time a user loads a page, I’m worried I’ll run out of resources quickly.

Is there some other table I can introduce to my model to relieve the overhead of doing the above query everytime users want to see the histogram break down? All I can think of is to precompute the histograms when possible so that reads optimized.

Thanks for any ideas

Answer 1

Here’s your query:

  SELECT p.product2,
         COUNT(p.product2) AS num_product
    FROM PRODUCTS p
    JOIN FRIENDS f ON f.friend_username = p.username
                  AND f.username = 'john'
   WHERE p.product1 = 'abc'
GROUP BY p.product2
ORDER BY num_product DESC

To handle 5 products, use:

  SELECT p.product1,
         p.product2,
         COUNT(p.product2) AS num_product
    FROM PRODUCTS p
    JOIN FRIENDS f ON f.friend_username = p.username
                  AND f.username = 'john'
   WHERE p.product1 IN ('abc', 'def', 'ghi', 'jkl', 'mno')
GROUP BY p.product1, p.product2
ORDER BY num_product DESC

It’s pretty simple, and the more you can filter the records down, the faster it will run because of being a smaller dataset.

If this query is running 5 times every time a user loads a page, I’m worried I’ll run out of resources quickly.

My first question is why you’d run this query more than once per page. If it’s to cover more than one friend, the query I posted can be updated to expose counts for products on a per friend or user basis.

After that, I’d wonder if the query can be cached at all. How fresh do you really need the data to be – is 2 hours acceptable? How about 6 or 12… We’d all like the data to be instantaneous, but you need to weigh that against performance and make a decision.