From 1e1d3e4b1aa699716de6df8c10164f32c7cb1cf6 Mon Sep 17 00:00:00 2001
From: John Bowdre <john@bowdre.net>
Date: Tue, 6 Dec 2022 16:42:44 -0600
Subject: [PATCH] use http check instead of ssh to detect when controlplane is
 ready

---
 terraform/scripts/initialize-controlplane.sh | 19 ++++++++++++++++---
 terraform/scripts/join-workers.sh            |  2 +-
 2 files changed, 17 insertions(+), 4 deletions(-)

diff --git a/terraform/scripts/initialize-controlplane.sh b/terraform/scripts/initialize-controlplane.sh
index 51823bf..8dbeee8 100644
--- a/terraform/scripts/initialize-controlplane.sh
+++ b/terraform/scripts/initialize-controlplane.sh
@@ -63,14 +63,27 @@ EOF
       exit 1
     fi
   fi
-  echo ">> Waiting up to 15 minutes for all nodes to be Ready..."
+  echo ">> Waiting up to 10 minutes for all control-plane nodes to be Ready..."
   python3 -m http.server &
   PROC_ID=$!
-  attempts_max=90
+  attempts_max=60
+  attempt=0
+  until [ "$(kubectl get nodes | grep -c ' control-plane ')" -eq "${K8S_CONTROLPLANE_COUNT}" ]; do
+    if [ ${attempt} -eq ${attempts_max} ]; then
+      echo ">> [ERROR] Timeout waiting for control-plane nodes to join! <<"
+      exit 1
+    fi
+    attempt=$((attempt+1))
+    sleep 10
+  done
+  echo ">> Continuing after $((attempt*10)) seconds."
+  touch .k8s-controlplane-success
+  echo ">> Waiting up to 10 minutes for all worker nodes to be Ready..."
+  attempts_max=60
   attempt=0
   until [ "$(kubectl get nodes | grep -c ' Ready ')" == "${K8S_NODE_COUNT}" ]; do
     if [ ${attempt} -eq ${attempts_max} ]; then
-      echo ">> [ERROR] Timeout waiting for cluster online! <<"
+      echo ">> [ERROR] Timeout waiting for worker nodes to join! <<"
       exit 1
     fi
     attempt=$((attempt+1))
diff --git a/terraform/scripts/join-workers.sh b/terraform/scripts/join-workers.sh
index 00c15ac..8e10b7e 100644
--- a/terraform/scripts/join-workers.sh
+++ b/terraform/scripts/join-workers.sh
@@ -17,7 +17,7 @@ echo ">> Continuing after $((attempt*10)) seconds."
 echo ">> Waiting up to 10 minutes for all control-plane nodes..."
 attempts_max=60
 attempt=0
-until [ "$(ssh -o StrictHostKeyChecking=no "${K8S_CONTROLPLANE_VIP}" kubectl get nodes | grep -c " control-plane ")" -eq "${K8S_CONTROLPLANE_COUNT}" ]; do
+until "$(wget http://${K8S_CONTROLPLANE_VIP}:8000/.k8s-controlplane-success)" 2>/dev/null; do
   if [ ${attempt} -eq ${attempts_max} ]; then
     echo ">> [ERROR] Timeout waiting for control-plane nodes! <<"
     exit 1