Investigate dependency proxy for packages performance

🔥 Problem

Compare the performance of the following scenarios:

1. CI job pulling all packages from an official registry directly vs through a dependency proxy (with a warm cache).
2. local machine (located far from the gitlab.com object storage) pulling all packages from an official registry directly vs through a dependency proxy (with a warm cache).

If usage (2.) (non CI) is larger that usage (1.) (CI), then a CDN approach (as in https://about.gitlab.com/blog/2022/10/25/gitlab-com-artifacts-cdn-change/) is valuable as it will help users to retrieve packages from closer locations. However, from my understanding, this brings no benefit in usage (1.).

🚒 Solution

1. Investigate the dependency proxy usage. In particular, analyze the CI usage vs the non CI usage.
   • We can use GCP IP ranges here to detect request coming from GCP. See https://www.gstatic.com/ipranges/goog.json.
2. Depending on the results of (1.), investigate if using a CDN is worth here.
3. If that's the case, investigate the amount of work to do.

🔮 Other considerations

The same analysis path/logic could be applied to the dependency proxy for container images.