From 0bf61789215f47941653755a431b8a1f9e51e0ba Mon Sep 17 00:00:00 2001
From: "Madhusudan.C.S" <madhusudancs@google.com>
Date: Wed, 21 Jun 2017 15:30:27 -0700
Subject: [PATCH] Add a spin-loop that waits for RBAC API's availability.

The very first thing that kubefed does when it comes up is run RBAC API
discovery. We believe this sometimes fail on new clusters and as
a result causes kubefed to assume that the RBAC API doesn't exist.
Therefore, we are applying this workaround for now to ensure that the
RBAC API is available before running kubefed.
---
 federation/cluster/federation-up.sh | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/federation/cluster/federation-up.sh b/federation/cluster/federation-up.sh
index 62efca9cdfc..01aaf05a564 100755
--- a/federation/cluster/federation-up.sh
+++ b/federation/cluster/federation-up.sh
@@ -81,6 +81,18 @@ function init() {
   kube::log::status "DNS_ZONE_NAME: \"${DNS_ZONE_NAME}\", DNS_PROVIDER: \"${DNS_PROVIDER}\""
   kube::log::status "Image: \"${kube_registry}/hyperkube-amd64:${kube_version}\""
 
+  # The very first thing that kubefed does when it comes up is run RBAC
+  # API discovery. We believe this sometimes fail on new clusters and as
+  # a result causes kubefed to assume that the RBAC API doesn't exist.
+  # Therefore, we are applying this workaround for now to ensure that the
+  # RBAC API is available before running kubefed.
+  timeout 1m bash <<EOF
+    while [[ ! "$(${KUBE_ROOT}/cluster/kubectl.sh api-versions)" =~ "rbac.authorization.k8s.io/" ]]; do
+      kube::log::status "Waiting for rbac.authorization.k8s.io API group to appear"
+      sleep 2
+    done
+EOF
+
   # Send INT after 20m and KILL 1m after that if process is still alive.
   timeout --signal=INT --kill-after=1m 20m \
       "${KUBE_ROOT}/federation/develop/kubefed.sh" init \