Question

I have table that records logins for users. I’m having trouble finding a query that will efficiently pull all unique users who have not had a login timestamp before a date of my choosing. I can easily find this information if I query per-user.

Fields are essentially:

 id, username, login, logout

I need a query that will find all usernames that have no logout timestamp in 2011-11. The fact that there are multiple join events for a single easier is what’s confusing me.

Answer 1

Exclude all users that had at least one logout at the date in question with a negated semi join (EXISTS) like this:

SELECT u.username
FROM   users u
WHERE  NOT EXISTS (
   SELECT *
   FROM   log
   WHERE  log.username = u.username
   AND    date(logout) = '2011-11-11')

This will not multiply rows from users. If 2011-11 is supposed to signify the month of November 2011:

...
   AND    logout >= '2011-11-01 0:0'
   AND    logout <  '2011-12-01 0:0'

---

Additional answer to question in comments

Flag the first login of users in the log

ALTER TABLE log ADD COLUMN first_log boolean; -- boolean flag

UPDATE log
     ,(SELECT username, min(login) AS min_log
       FROM   log
       GROUP  BY 1) x
SET    log.first_log = TRUE
WHERE  (log.username, log.login) = (x.username, x.min_log);

You can do that, but there are many more rows in table log than in table users. I would advise instead:

ALTER TABLE users ADD COLUMN first_log datetime;

UPDATE users u
     ,(SELECT username, min(login) AS min_log; -- datetime col
       FROM   log
       GROUP  BY 1) x
SET    u.first_log = x.min_log
WHERE  u.username = x.username