k8s.io
k8s.io copied to clipboard
move artifacts.k8s.io to AWS
We now have $3M AWS credits coming for next year, and we're still going to be tight on the kubernetes.io bill, with $600k credits coming to help get us through this year on top of the existing $3M GCP credits.
Looking at the bill, I estimate artifacts.k8s.io cost in excess of $29k for October 1-30th, based on the cloud networking cost for k8s-artifacts-prod (networking costs are for cloud CDN we have in front of the backing bucket, for k8s.gcr.io the costs are purely cloud storage in that same project). I think this is the single largest thing that we can possibly move to AWS with a quick turnaround.
This bucket / cloud CDN contains kOps binaries + some old crictl binaries, AFAICT.
We should look at hosting it out of AWS (s3?), in the simplest way we can quickly ship to help remove pressure on the kubernetes.io bill, then we can iterate to optimize on the AWS side.
I think this is roughly retargeting the load balancer at a 3XX redirect to an S3 bucket, later we can build a small architecture like registry.k8s.io to serve this to N regional buckets or something, but something simple should probably be fine to start.
We will also need to backfill the existing contents (rclone) and update the file promoter to promote here.

https://datastudio.google.com/c/u/0/reporting/14UWSuqD5ef9E4LnsCD9uJWTPv8MHOA3e/page/8fWn
If my estimate is right, this is $29,103.97 / $314,396.53 * 100 = 9.257% of the spend we can move, quickly.