Mike McGrath wrote:
I'm not sure how this would play with European data-privacy laws, especially in cases where one can tie an IP address to an individual.
One option here, mentioned in the Metrics page, is to use a public proxy server for the initial phone home. Obviously we wouldn't want to do a yum through the proxy server but some low-cost phone home (like retrieving the mirror list) would be valuable.
Probably the GUID is a bit of a red herring, since in the case there are millions of boxes it will be a ton of work to maintain the database of them and compare every log line against it, for limited hard information. Because boxes will typically download specific updated RPMs just the once, you can get an idea of the number of active boxes just by filtering on packages that have been updated for a while.
If at the time of a new release you just take care to make a changeless update to, say, fedora-release, which all installs will have, then you can see from the logs every box's first update action (because the new fedora-release will get pulled that time only). That gives you the (undercounted) number of new installs that had network connectivity, anyway.
-Andy
