Cluster provisioning is unreliable because of Docker Hub limits

[invidian]

CI fails sometimes for bare-metal with the following errors:

  Warning  Failed     16m (x4 over 17m)     kubelet            Failed to pull image "calico/cni:v3.16.4": rpc error: code = Unknown desc = Error response from daemon: toomanyrequests: You have reached your pull rate limit. You may increase the limit by authenticating and upgrading: https://www.docker.com/increase-rate-limit

We could implement #457 to resolve it.

[invidian]

Actually, we can only mirror images which are currently pulled from Docker Hub to Quay and defer #457 for now.

[invidian]

For future us, created this script to create multi-arch manifests on Quay for Calico:

#!/bin/bash

VERSION=v3.16.4
IMAGES="node cni kube-controllers pod2daemon-flexvol"
SOURCE_REG=quay.io/calico/
TARGET_REG=quay.io/kinvolk/calico-


for i in $IMAGES; do
  echo "docker manifest create ${TARGET_REG}${i}:${VERSION} \
    --amend ${SOURCE_REG}${i}:${VERSION}-amd64 \
    --amend ${SOURCE_REG}${i}:${VERSION}-arm64"

  echo "docker manifest annotate ${TARGET_REG}${i}:${VERSION} \
    ${SOURCE_REG}${i}:${VERSION}-amd64 --arch amd64"

  echo "docker manifest annotate ${TARGET_REG}${i}:${VERSION} \
    ${SOURCE_REG}${i}:${VERSION}-arm64 --arch arm64"

  echo "docker manifest push --purge ${TARGET_REG}${i}:${VERSION}"

  echo
done