Not mentioned in the post is merge queuing, as discussed in Keeping master green at scale (Uber 2019) and probably used by every large tech company in some capacity. Assuming a suitably low test breakage rate for commits, you can essentially pipeline CI runs and achieve subquadratic CI costs.
We use merge queues as well. But “PR builds” - the ones that give feedback to developers on their pull requests are the biggest source of costs, in my experience. And those grow quadratically.
Not mentioned in the post is merge queuing, as discussed in Keeping master green at scale (Uber 2019) and probably used by every large tech company in some capacity. Assuming a suitably low test breakage rate for commits, you can essentially pipeline CI runs and achieve subquadratic CI costs.
We use merge queues as well. But “PR builds” - the ones that give feedback to developers on their pull requests are the biggest source of costs, in my experience. And those grow quadratically.