How much will Kubernetes “managed fields” feature cost the global economy over the next 10 years?
Kubernetes 1.16 introduced managed fields to help sort out server-side apply problems. These roughly 2x the amount of data to each
kubectl get -ojson. It only intended for software to read, you are forced to scroll past it and find the data you care about.
How much might this be costing us?
total cost = E (number of engineers) x M (number of manifests they inspect) x T (time managed fields add each inspecting) x C (cost of engineer time)
Lets invent some values:
- There are 25 million software engineers globally. 80k have starred the Kubernetes repository. So somewhere between the two. Shall we say 1–2 million Kubernetes engineers globally?
- Engineer each inspect 5-10 manifests a day, multiple this by 5 days per week then by 44 weeks, then by 10 years to get 11k to 22k inspections.
- Each inspection now takes 5–10 seconds longer.
- Engineers are paid between $100k and $200k a year. Lets double thatthat to get the actual cost to the company (e.g. the cost of HR, work gym, bananas in the office etc). We need the cost per second, so divide $200,000/44/5/8/60/60 would be $0.03 to $0.06 a second.
So at the low end:
1,000,000 engineers * 11,000 inspections * 5s * $0.03= $1.65 billion
And at the high end:
2,000,000 engineers * 22,000 inspections * 10s * $0.06= $26.4 billion
This discussion combines some invented numbers and doesn’t talk about the benefit of the feature at all. This feature is in fact, almost irrelevant, as this article is really intended to show how small changes can have massive impact if there are power multipliers involved.
kubectl get -ojson will not return managed fields anymore in v2.12, so only a problem for a couple of years, not ten years.