I’m using droneci in our on-premise infra using Kubernetes runner and I’ve been using a registry mirror in my pipelines using the ECR plugin’s mirror parameter and everything is working as expected except that when when new builds is happening sometimes I get this error:
Normal Scheduled 16m default-scheduler Successfully assigned drone/drone-job-198609-9ot6qj9punizcsmwf-mcf9n to ip-10-0-12-34.ec2.internal Warning Failed 16m kubelet, ip-10-0-12-34.ec2.internal Failed to pull image "drone/controller:1": rpc error: code = Unknown desc = toomanyrequests: You have reached your pull rate limit. You may increase the limit by authenticating and upgrading: https://www.docker.com/increase-rate-limit Warning Failed 15m (x6 over 16m) kubelet, ip-10-0-12-34.ec2.internal Error: ImagePullBackOff
My guess is that the new pods created by drone in order to execute the build steps is being pulled from dockerhub every-single time a new runner pod is created and not utilizing the already downloaded images that’s supposedly cached on the k8s nodes itself? Note that There’s no scaling happened so no new nodes added. Is this the expected behavior? shouldn’t this pods get created using the the same images that old runners used?