不同的结果 UNNEST BigQuery

Question

I don't understand what is the differents between those queries:

SELECT event_timestamp ,user_pseudo_id, value.double_value as tax
FROM `bigquery-public-data.ga4_obfuscated_sample_ecommerce.events_*`, UNNEST(event_params) as event_params
WHERE event_name = "purchase" and event_params.key = "tax" 

The other query is:

SELECT event_timestamp ,user_pseudo_id, 
(SELECT value.double_value FROM UNNEST(event_params) WHERE key = "tax") as tax
FROM `bigquery-public-data.ga4_obfuscated_sample_ecommerce.events_*` 
WHERE event_name = "purchase"

In the first query, I get 5.242 registers and in the second 5.692. What is the mistake?

Thank you!

Answer 1

It depends on what you define as accurate. The reason you are getting a row count mismatches is because of the way the tax field is being handled. You can see this by running the following query to see the discrepancies:

with unnested as (
    SELECT event_timestamp ,user_pseudo_id, value.double_value as tax
    FROM `bigquery-public-data.ga4_obfuscated_sample_ecommerce.events_*`, UNNEST(event_params) as event_params
    WHERE event_name = "purchase" and event_params.key = "tax"
 ) 
SELECT events.event_timestamp ,events.user_pseudo_id, 
(SELECT value.double_value FROM UNNEST(event_params) WHERE key = "tax") as tax
FROM `bigquery-public-data.ga4_obfuscated_sample_ecommerce.events_*` events
LEFT JOIN unnested un
 on events.event_timestamp=un.event_timestamp
 and events.user_pseudo_id=un.user_pseudo_id
WHERE events.event_name = "purchase"
and un.event_timestamp is null
;

If you pick out a single record from that list and investigate with the two following queries:

    SELECT *
    FROM `bigquery-public-data.ga4_obfuscated_sample_ecommerce.events_*`, UNNEST(event_params) as event_params
    WHERE 1=1
    -- and event_name = "purchase" and event_params.key = "tax"
    and event_name = "purchase" and event_timestamp=1608955242902332 and user_pseudo_id='43627350.3807676886';


    SELECT 
    *,
    (SELECT value.double_value FROM UNNEST(event_params) WHERE key = "tax") as tax
    FROM `bigquery-public-data.ga4_obfuscated_sample_ecommerce.events_*` events
    WHERE events.event_name = "purchase"
    and event_timestamp=1608955242902332 and user_pseudo_id='43627350.3807676886'
    ;

The first query is filtering out the records without a tax field from your final set, while the second returns the records as having a null tax value. If the number registered is dependent on the presence of a value in the tax field the 5242 value is the correct number.