Investigate different domain_userid having the same domain_sessionid

[bogaert]

We found one case (a client using JS-tracker 1.5.2) where the number of distinct domain_sessionid is between 50 to 100% higher than the number of domain_userid || domain_sessionidx.

It appears that this is driven by different domain_userid (often on different domains) having the same domain_sessionid.

In SQL:

SELECT
domain_sessionid,
COUNT(DISTINCT domain_userid || domain_sessionidx) AS user_sessionidx,
COUNT(DISTINCT domain_userid)
FROM atomic.events
GROUP BY 1
ORDER BY 2 DESC
LIMIT 50

Needs to be investigated further.

[fblundun]

One idea we investigated: that the UUID generation was not generated sufficiently randomly. We use the standard Node UUID library. If it fails to find window.crypto.getRandomValues or window.msCrypto.getRandomValues, it will fall back on Math.random which is not guaranteed to return unique values.

This compatibility chart has information about which browsers support getRandomValues. Chrome for Android apparently does support it, but the problem of duplicated domain_sessionid sometimes appears for Chrome for Android. So falling back on Math.random doesn't explain this.

[paulboocock]

Closing in favour of #499