Instead of longly talking about it, I just did it quickly. Seems like my theory used in reality is wrong because there's no significant improvement in terms of speed. Our jobs mostly depend on how quickly GitLab pulls down the images.
Latest master pipelines:
- 00:02:24
- 00:03:32
- 00:04:04
- 00:05:42
- 00:03:00
- 00:05:25
This MR's pipelines:
- 00:03:19
- 00:02:37