Machine count for Ubuntu version includes machines known to be running newer version

[Matthew Paul Thomas]

Suppose we get error reports from a machine running Ubuntu version N.

Later, we get error reports from the same machine running Ubuntu version N+1, N+2, or whatever.

Provided that we don't get further reports from it running Ubuntu version N (indicating a dual boot), it's safe to assume that it doesn't have N installed any more.

However, we don't remove it from the count of machines running N until 90 days has passed without a report. It would be more accurate to remove it from the count immediately.

Hypothesis: If we fix this, the October-to-January dip in <https://errors.ubuntu.com/?release=Ubuntu%2012.04&period=day> will disappear, because it was caused by 12.04 machines upgrading to 12.10 but still being counted as 12.04.

Similarly, the post-April plummet in <https://errors.ubuntu.com/?release=Ubuntu%2012.10&period=day> will disappear, because it was caused by 12.10 machines upgrading to 13.04 but still being counted as 12.10.

[Evan (ev)]

"Provided that we don't get further reports from it running Ubuntu version N"

How long do we wait? If we have a system in the list of systems seen in the past 90 days for Quantal and Raring, do we check to see if there were any crashes for that system in Quantal just for yesterday (we always work on the previous day)?

Or do we say that if said machine shows and Quantal crashes over the next week, we'll go back and repair the counts to include it in both the Quantal and Raring lists?

[Matthew Paul Thomas (mpt)]

The upgrade might have been part-way through yesterday. So you might have error reports for two versions of Ubuntu from the same machine yesterday, *without* it being a dual-boot.

Conversely, you might get reports from a machine for version N up to a given date, and only version N+1 for the next two weeks ... and then one for N again after that. It was a dual boot, but version N wasn't being used much (or was luckily avoiding errors).

So I don't think it's a matter of waiting, is it? If you get a report for version N+1, assume that the machine has upgraded. If you ever get another report for version N from the same machine, assume that it's a dual-boot. If they've actually downgraded, N+1 will fade out over 90 days.

This is all papering over bug 1046269 -- we're dividing by numbers of machines rather than by actual hours of use. That a machine is a dual-boot doesn't add more hours to the day in which the machine can experience errors. The machine I'm using right now is a triple-boot, but I seldom use the other installations.