Update master_node_install.sh

master_node_install.sh

@@ -10,6 +10,7 @@ set -Eeuo pipefail
 # - Installs required Arch packages
 # - Configures kernel modules and sysctl for Kubernetes
 # - Configures NetworkManager to ignore CNI interfaces
+# - Disables host nftables service to avoid breaking RKE2 service routing
 # - Installs RKE2 server
 # - Waits for Kubernetes and bundled RKE2 addons to become healthy
 # - Installs cert-manager
@@ -181,7 +182,7 @@ enable_support_services() {
   systemctl enable --now iscsid.service || true
 
   # Do NOT enable nftables.service here.
-  # On this Arch + RKE2 setup it broke service routing for the cluster IP range.
+  # On this Arch + RKE2 setup it can break service routing for cluster IPs.
   systemctl stop nftables.service >/dev/null 2>&1 || true
   systemctl disable nftables.service >/dev/null 2>&1 || true
   nft flush ruleset >/dev/null 2>&1 || true
@@ -287,15 +288,24 @@ wait_for_system_pods() {
   kubectl_rke2 get nodes -o wide || true
   kubectl_rke2 get pods -A || true
 
-  # Wait for Canal first because service routing depends on it.
+  log "Waiting for Canal (CNI) to be fully ready"
+
   waited=0
   until kubectl_rke2 -n kube-system get pods -l app=rke2-canal -o json 2>/dev/null | jq -e '
-    .items | length > 0 and all(.[]; .status.phase=="Running")
+    .items
+    | length > 0
+    and all(
+        .[];
+        (
+          (.status.containerStatuses // []) | length
+        ) > 0
+        and all(.status.containerStatuses[]; .ready == true)
+      )
   ' >/dev/null; do
     sleep 5
     waited=$((waited + 5))
     if (( waited % 30 == 0 )); then
-      warn "Canal not fully ready yet"
+      warn "Canal is not fully ready yet"
       kubectl_rke2 -n kube-system get pods -o wide || true
     fi
     if (( waited >= 900 )); then
@@ -307,11 +317,21 @@ wait_for_system_pods() {
   # Give kube-proxy and service routing a moment to settle.
   sleep 20
 
-  # Wait for bundled addons that Rancher depends on.
+  log "Waiting for bundled RKE2 addon deployments"
+
   waited=0
-  until kubectl_rke2 -n kube-system get deploy rke2-ingress-nginx-controller rke2-metrics-server rke2-snapshot-controller >/dev/null 2>&1; do
+  until kubectl_rke2 -n kube-system get deploy \
+    rke2-ingress-nginx-controller \
+    rke2-metrics-server \
+    rke2-snapshot-controller \
+    rke2-coredns-rke2-coredns >/dev/null 2>&1; do
     sleep 5
     waited=$((waited + 5))
+    if (( waited % 30 == 0 )); then
+      warn "Bundled addon deployments are not all present yet"
+      kubectl_rke2 -n kube-system get deploy || true
+      kubectl_rke2 -n kube-system get pods -o wide || true
+    fi
     if (( waited >= 900 )); then
       kubectl_rke2 -n kube-system get pods -o wide || true
       die "Timed out waiting for bundled RKE2 addon deployments"
@@ -321,9 +341,7 @@ wait_for_system_pods() {
   kubectl_rke2 -n kube-system rollout status deploy/rke2-ingress-nginx-controller --timeout=20m
   kubectl_rke2 -n kube-system rollout status deploy/rke2-metrics-server --timeout=20m
   kubectl_rke2 -n kube-system rollout status deploy/rke2-snapshot-controller --timeout=20m
-
+  kubectl_rke2 -n kube-system rollout status deploy/rke2-coredns-rke2-coredns --timeout=20m
-  # CoreDNS can be slightly slower; wait for it too.
-  kubectl_rke2 -n kube-system rollout status deploy/rke2-coredns-rke2-coredns --timeout=20m || true
 
   log "System pods are settled"
   kubectl_rke2 get pods -A || true