为什么我的非零分母在 SQL 中给出“除以零”错误？

Question

First time asking a question here, thanks in advance. Also new to SQL so this may be a basic question.

I'm in charge of a weekly errors report for our product and trying to come up with a "smarter" query that'll only return cases where the number of errors represents a significant chunk of a customer's traffic (eg, don't show me 100,000 errors out of a customer's 50,000,000 sebsite visitors, but do show me 50,000 out of 200,000).

Where this gets difficult is, I want to differentiate mobile from desktop traffic, and return a customer's data when either one gets a high percentage of errors (for now let's say 20%).

Here's what I have so far:

WITH requests_summary AS (
    SELECT customer_id, column_x, column_y, column_z
        SUM(CASE WHEN event = 'error' AND device = 'mobile' THEN 1 ELSE 0 END) AS err_mbl,
        SUM(CASE WHEN event = 'traffic' AND device = 'mobile' THEN 1 ELSE 0 END) AS trf_mbl
        SUM(CASE WHEN event = 'error' AND device = 'desktop' THEN 1 ELSE 0 END) AS err_desk,
        SUM(CASE WHEN event = 'traffic' AND device = 'desktop' THEN 1 ELSE 0 END) AS trf_desk
    FROM "table"
        WHERE timestamp >= DATEADD(day,-1, GETDATE())
        GROUP BY 1,2,3,4
)

SELECT *
FROM requests_summary
WHERE 1.0 * err_mbl / trf_mbl > 0.2
OR 1.0 * err_desk / trf_desk > 0.2;

Now the problem is, I get a 'division by zero' error every time I run this query. So far I've:

• Commented out the desktop and mobile division conditions one at a time to verify they're both giving the error
• Tried a separate query for just the CASE clauses, one at a time, to verify the traffic on each platform is non-zero (it's not really possible for a whole day to go with no traffic, anyway)
• Replaced the trf_mbl and trf_desk denominators with (A) an arbitrary non-zero number and (B) a COUNT(*), both made the query work

So I'm thinking the issue is with my use of the WITH clause to create the temporary requests_summary table, but I'm still not sure how to resolve this. I've been tinkering with this for a good chunk of the past two days, but still no go. Can anyone offer guidance?

Answer 1

The issue is in your where clause where you are dividing values. in your where clause If either or both trf_mbl or trf_desk are equal to 0 you can get that error, meaning where there is not traffic event in that group ( customerid, x,y,z)

so you have to change your logic in the case there is no traffic event in that group.

Answer 2

Use 'having' clause in the subquery to check for 0 traffic

WITH requests_summary AS (
    SELECT customer_id, column_x, column_y, column_z
        SUM(CASE WHEN event = 'error' AND device = 'mobile' THEN 1 ELSE 0 END) AS err_mbl,
        SUM(CASE WHEN event = 'traffic' AND device = 'mobile' THEN 1 ELSE 0 END) AS trf_mbl
        SUM(CASE WHEN event = 'error' AND device = 'desktop' THEN 1 ELSE 0 END) AS err_desk,
        SUM(CASE WHEN event = 'traffic' AND device = 'desktop' THEN 1 ELSE 0 END) AS trf_desk
    FROM "table"
        WHERE timestamp >= DATEADD(day,-1, GETDATE())
        GROUP BY 1,2,3,4
        having trf_mbl>0 and trf_desk>0
)

SELECT *
FROM requests_summary
WHERE 1.0 * err_mbl / trf_mbl > 0.2
OR 1.0 * err_desk / trf_desk > 0.2;

Answer 3

I would simply rephrase the WHERE to be:

WHERE 1.0 * err_mbl > 0.2 * trf_mbl OR
      1.0 * err_desk > 0.2 * trf_desk

Voila. No division. No error.

Note: I am guessing that the 1.0 * is also not necessary.