Sessions-per-User Distribution-table in Firebase

问题

This is a second post (a follow-up from my first post) on looking at distributions within Firebase Analytics Data. This time around, I want to create a user distribution table in BigQuery based on Firebase Session Data. The output should look like this:

I managed to create the following script to count on app_instance_id's:

#standardSQL
SELECT 
  COUNT(DISTINCT(CASE WHEN sess_id = 0 THEN app_instance_id END)) AS sess_count_0,
  COUNT(DISTINCT(CASE WHEN sess_id = 1 THEN app_instance_id END)) AS sess_count_1,
  COUNT(DISTINCT(CASE WHEN sess_id > 1 AND sess_id <= 5 THEN app_instance_id END)) AS sess_count_2BETWEEN5,
  COUNT(DISTINCT(CASE WHEN sess_id > 5 AND sess_id <= 10 THEN app_instance_id END)) AS sess_count_6BETWEEN10,
  COUNT(DISTINCT(CASE WHEN sess_id > 10 AND sess_id <= 30 THEN app_instance_id END)) AS sess_count_11BETWEEN30,
  COUNT(DISTINCT(CASE WHEN sess_id > 30 THEN app_instance_id END)) AS sess_count_PLUS31
FROM (SELECT *, SUM(session_start) OVER(PARTITION BY app_instance_id ORDER BY min_time) sess_id
      FROM (SELECT *, IF(previous IS null OR (min_time-previous)>(20*60*1000*1000),1, 0) session_start 
            FROM (SELECT *, LAG(max_time, 1) OVER(PARTITION BY app_instance_id ORDER BY max_time) previous
                  FROM (SELECT user_dim.app_info.app_instance_id, 
                  user_dim.device_info.mobile_model_name, 
                  user_dim.device_info.platform_version, 
                  (SELECT MIN(timestamp_micros) 
                  FROM UNNEST(event_dim)) min_time, 
                    (SELECT MAX(timestamp_micros) FROM UNNEST(event_dim)) max_time 
                    FROM `firebase-public-project.com_firebase_demo_IOS.app_events_*` 
                    WHERE (_TABLE_SUFFIX BETWEEN '20170701' AND '20170731')
                  )
            )
      )
)

Questions:

Considering Users (and not Sessions), I want to make 100 % sure whether I should still count on App Instances (and not the Session Ids) ?
Any thoughts on optimising this query Is there a more efficient way to aggregate all the distribution ranges with one query ?
Finally, I wanted to compare the overall total that I got from above with distinct users that triggered the session_start-event over the same period. I was hoping to see that it roughly would align, but it did not. Why is there such a big difference: 7688 vs 16310 (488+7343+4967+1956+1165+391) ? Where did my logic go wrong ?
```
#standardSQL
SELECT 
COUNT (DISTINCT user_dim.app_info.app_instance_id) as users
FROM `firebase-public-project.com_firebase_demo_IOS.app_events_*`, UNNEST(event_dim) AS event 
  WHERE (_TABLE_SUFFIX BETWEEN '20170701' AND '20170731')
  AND event.name = "session_start"
```

来源：https://stackoverflow.com/questions/48687226/sessions-per-user-distribution-table-in-firebase

标签

session

google-bigquery

firebase-analytics

bigquery-standard-sql