From 573dd3ec3ef04bc991833126bca3e6f1237c500b Mon Sep 17 00:00:00 2001
From: Stefan Reimer <stefan@zero-downtime.net>
Date: Wed, 13 Nov 2024 14:35:50 +0000
Subject: [PATCH] fix: minor upgrade flow fixes

---
 admin/kubezero.sh        |  2 +-
 admin/upgrade_cluster.sh | 14 ++++++++++----
 docs/nvidia-test.yaml    |  7 ++++++-
 3 files changed, 17 insertions(+), 6 deletions(-)

diff --git a/admin/kubezero.sh b/admin/kubezero.sh
index 1b0a68e3..cd41be02 100755
--- a/admin/kubezero.sh
+++ b/admin/kubezero.sh
@@ -64,7 +64,7 @@ render_kubeadm() {
     cat ${WORKDIR}/kubeadm/templates/${f}Configuration.yaml >> ${HOSTFS}/etc/kubernetes/kubeadm.yaml
   done
 
-  if [[ "$phase" =~ ^(bootstrap|restore)$ ]]; then
+  if [[ "$phase" =~ ^(bootstrap|join|restore)$ ]]; then
     cat ${WORKDIR}/kubeadm/templates/InitConfiguration.yaml >> ${HOSTFS}/etc/kubernetes/kubeadm.yaml
   fi
 
diff --git a/admin/upgrade_cluster.sh b/admin/upgrade_cluster.sh
index 0904bd57..b0b0f1f1 100755
--- a/admin/upgrade_cluster.sh
+++ b/admin/upgrade_cluster.sh
@@ -19,6 +19,11 @@ echo "Checking that all pods in kube-system are running ..."
 
 [ "$ARGOCD" == "True" ] && disable_argo
 
+# 1.30 fix for the missing kubeadm socket annotations
+for c in $(kubectl get nodes -l "node-role.kubernetes.io/control-plane=" | grep v1.29 | awk {'print $1}'); do
+  kubectl annotate node $c 'kubeadm.alpha.kubernetes.io/cri-socket=unix:///var/run/crio/crio.sock'
+done
+
 control_plane_upgrade kubeadm_upgrade
 
 echo "Control plane upgraded, <Return> to continue"
@@ -33,13 +38,14 @@ kubectl delete runtimeclass crio || true
 
 # upgrade modules
 #
-# Preload cilium images to running nodes
-all_nodes_upgrade "chroot /host crictl pull quay.io/cilium/cilium:v1.16.3"
+# Preload cilium images to running nodes, disabled till 1.31
+# all_nodes_upgrade "chroot /host crictl pull quay.io/cilium/cilium:v1.16.3; chroot /host crictl pull ghcr.io/k8snetworkplumbingwg/multus-cni:v3.9.3"
 
 control_plane_upgrade "apply_network, apply_addons, apply_storage, apply_operators"
 
-echo "Checking that all pods in kube-system are running ..."
-waitSystemPodsRunning
+# Disabled during 1.30 due to nvidia runtime deadlock
+#echo "Checking that all pods in kube-system are running ..."
+#waitSystemPodsRunning
 
 echo "Applying remaining KubeZero modules..."
 
diff --git a/docs/nvidia-test.yaml b/docs/nvidia-test.yaml
index 87000152..0c7e3f11 100644
--- a/docs/nvidia-test.yaml
+++ b/docs/nvidia-test.yaml
@@ -6,11 +6,12 @@ metadata:
 spec:
   runtimeClassName: nvidia
   containers:
+ #- image: nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0
   - image: nvidia/cuda:12.5.1-base-ubuntu22.04
     command:
     - "bash"
     - "-c"
-    - "sleep 3600"
+    - "nvidia-smi; sleep 3600"
     imagePullPolicy: IfNotPresent
     name: nvidia-test
     resources:
@@ -18,3 +19,7 @@ spec:
         nvidia.com/gpu: 1
       requests:
         memory: 1024Mi
+  tolerations:
+  - effect: NoSchedule
+    key: kubezero-workergroup
+    operator: Exists