On Thu, May 3, 2018 at 4:39 PM, Matthew Miller mattdm@fedoraproject.org wrote:
I'd like to talk about adding an randomized, rotating UUID to DNF and rpm-ostree updates which we could use to better count installed systems. The IP-based counting we currently use is deeply flawed, and we'd really benefit from getting more accurate numbers.
This would
let us make better strategic decisions based on our actual base of installed systems, and
help us demostrate growth and impact to various sponsors.
To be completely candid: this would be incredibly useful as I advocate for Fedora within Red Hat, but no one @redhat has directed me to ask for it. And its *not* just Red Hat — this would be useful in talking to our various hardware partners too and other potential sponsors.
Long ago, we had Smolt, but this was opt-in and didn't present a very complete picture. I'd like to propose we do something similar to what openSUSE describes here: https://en.opensuse.org/openSUSE:Statistics. That is: A completely random UUID used only for this purpose, and used for counting rather than tracking.
We could refresh the UUID periodically (like monthly) to ensure that long-term tracking wouldn't even be an option (while still distinguishing between short-lived cloud or test instances). And of course there would be some way to opt out.
Recently Canonical has implemented a much greater amount of data collection https://lists.ubuntu.com/archives/ubuntu-devel/2018-February/040139.html, and while maybe we might want to look at reviving an opt-in system, I don't think we need that right now. (Historically, we've had a lot of resistance to similar proposals as opt-out, and the value of opt-in is low.)
Basically, all I want is that UUID plus CPU arch plus ID, VERSION_ID, and VARIANT_ID from /etc/os-release. (And, I'd also like to start using VARIANT_ID for labs, spins, docker images, etc.)
What do you think? Does the suggested plan make sense?
Yes. I think we should go forward with this.
josh