Better download statistics covering repo downloads

This task tracks the implementation of better repo statistics. It is based on redirecting the mirrors to only a subset of the current mirrors – namely, those that can provide better stats.

Aa recap of our discussions from 2024-09-23

Problem definition: limited insights from Download stats, due to Windows skew

Overall issue: Our data on MariaDB Server downloads is not representative of actual MariaDB Server usage, which makes decision making difficult, for lack of data

Download stats by OS is heavily skewed to Windows - "95% of downloads represents 5% of usage" https://mariadb.org/statistics/#dlds-by-os

Conclusions of meeting Monty-Anna-Vicentiu-Kaj Mon 2024-09-23

• No further information of any reasonable value can be gained from existing download statistics
• Vicentiu: Perhaps the Windows share would move from 95% to 25%, if all tracking is turned on.
• Vicentiu: Turning it on is 1-2 weeks of work. Not easy because of how we work with mirrors, not easy to test.
• Q: Why did we not implement it completely 2021? A: Because it would have been more than 1-2 weeks in 2021.

Actions

• Vicentiu: Implement a proxy for RPM / DEB repositories, eliminate mirror raw urls. Work already partially done via deb.mariadb.org and rpm.mariadb.org. That will allow tracking individual package downloads, not just for bintars, msi, source, zips.
• Vicentiu: Assign Jira items properly. On 2024-12-17 we said "Create two Jira items, one for Vlad Radu, the other for Faustin. Faustin mentioned a potential third Jira task?" – this needs to be cleaned up
• Contents: Change download templates, convert traffic to store the information
• Get all proposed Jira items reviewed by Daniel from a user experience perspective