fix(system-upgrade-controller): talos health check

buroa · Nov 28, 2023 · f9f598c · f9f598c
1 parent 44a19bd
commit f9f598c
Show file tree

Hide file tree

Showing 3 changed files with 35 additions and 41 deletions.
diff --git a/kubernetes/apps/kube-system/system-upgrade-controller/plans/kubernetes.yaml b/kubernetes/apps/kube-system/system-upgrade-controller/plans/kubernetes.yaml
@@ -19,14 +19,23 @@ spec:
   nodeSelector:
     matchExpressions:
       - { key: node-role.kubernetes.io/control-plane, operator: Exists }
-  upgrade:
-    image: ghcr.io/siderolabs/talosctl:v1.5.5
-    envs:
+  # TODO: upstream talos health check fix
+  prepare:
+    image: &image ghcr.io/siderolabs/talosctl:v1.5.5
+    envs: &envs
       - name: NODE_IP
         valueFrom:
           fieldRef:
             fieldPath: status.hostIP
     args:
-      - "--nodes=$(NODE_IP)"
-      - "upgrade-k8s"
-      - "--to=$(SYSTEM_UPGRADE_PLAN_LATEST_VERSION)"
+      - --nodes=$(NODE_IP)
+      - health
+      - --control-plane-nodes=10.0.0.10,10.0.0.11,10.0.0.12
+      - --worker-nodes=10.0.0.20,10.0.0.21,10.0.0.22
+  upgrade:
+    image: *image
+    envs: *envs
+    args:
+      - --nodes=$(NODE_IP)
+      - upgrade-k8s
+      - --to=$(SYSTEM_UPGRADE_PLAN_LATEST_VERSION)
diff --git a/kubernetes/apps/kube-system/system-upgrade-controller/plans/talos.yaml b/kubernetes/apps/kube-system/system-upgrade-controller/plans/talos.yaml
@@ -19,18 +19,27 @@ spec:
   nodeSelector:
     matchExpressions:
       - { key: kubernetes.io/os, operator: In, values: [linux] }
-  upgrade:
-    image: ghcr.io/siderolabs/talosctl
-    envs:
+  # TODO: upstream talos health check fix
+  prepare:
+    image: &image ghcr.io/siderolabs/talosctl
+    envs: &envs
       - name: NODE_IP
         valueFrom:
           fieldRef:
             fieldPath: status.hostIP
     args:
-      - "--nodes=$(NODE_IP)"
-      - "upgrade"
-      - "--image=ghcr.io/buroa/installer:$(SYSTEM_UPGRADE_PLAN_LATEST_VERSION)"
-      - "--preserve=true"
-      - "--stage=true"
-      - "--wait=false"
-      - "--reboot-mode=powercycle" # thunderbolt issue with kexec
+      - --nodes=$(NODE_IP)
+      - health
+      - --control-plane-nodes=10.0.0.10,10.0.0.11,10.0.0.12
+      - --worker-nodes=10.0.0.20,10.0.0.21,10.0.0.22
+  upgrade:
+    image: *image
+    envs: *envs
+    args:
+      - --nodes=$(NODE_IP)
+      - upgrade
+      - --image=ghcr.io/buroa/installer:$(SYSTEM_UPGRADE_PLAN_LATEST_VERSION)
+      - --preserve=true
+      - --stage=true
+      - --wait=false
+      - --reboot-mode=powercycle # thunderbolt issue with kexec
diff --git a/talos/talconfig.yaml b/talos/talconfig.yaml
@@ -56,13 +56,9 @@ nodes:
               ip: ${clusterEndpointIP}
           - vlanId: 12
             dhcp: false
-            addresses:
-              - 10.0.2.10/24
             mtu: 1500
           - vlanId: 13
             dhcp: false
-            addresses:
-              - 10.0.3.10/24
             mtu: 1500
     nodeLabels: &masterLabels
       topology.kubernetes.io/region: *cluster
@@ -100,13 +96,9 @@ nodes:
               ip: ${clusterEndpointIP}
           - vlanId: 12
             dhcp: false
-            addresses:
-              - 10.0.2.11/24
             mtu: 1500
           - vlanId: 13
             dhcp: false
-            addresses:
-              - 10.0.3.11/24
             mtu: 1500
     nodeLabels: *masterLabels
 
@@ -142,13 +134,9 @@ nodes:
               ip: ${clusterEndpointIP}
           - vlanId: 12
             dhcp: false
-            addresses:
-              - 10.0.2.12/24
             mtu: 1500
           - vlanId: 13
             dhcp: false
-            addresses:
-              - 10.0.3.12/24
             mtu: 1500
     nodeLabels: *masterLabels
 
@@ -182,13 +170,9 @@ nodes:
                 gateway: 10.0.0.1
           - vlanId: 12
             dhcp: false
-            addresses:
-              - 10.0.2.20/24
             mtu: 1500
           - vlanId: 13
             dhcp: false
-            addresses:
-              - 10.0.3.20/24
             mtu: 1500
       - deviceSelector:
           busPath: 1-1.0 # w1
@@ -241,13 +225,9 @@ nodes:
                 gateway: 10.0.0.1
           - vlanId: 12
             dhcp: false
-            addresses:
-              - 10.0.2.21/24
             mtu: 1500
           - vlanId: 13
             dhcp: false
-            addresses:
-              - 10.0.3.21/24
             mtu: 1500
       - deviceSelector:
           busPath: 1-1.0 # w0
@@ -297,13 +277,9 @@ nodes:
                 gateway: 10.0.0.1
           - vlanId: 12
             dhcp: false
-            addresses:
-              - 10.0.2.22/24
             mtu: 1500
           - vlanId: 13
             dhcp: false
-            addresses:
-              - 10.0.3.22/24
             mtu: 1500
       - deviceSelector:
           busPath: 1-1.0 # w0
@@ -371,7 +347,7 @@ controlPlane:
           port: 7445
       logging:
         destinations:
-          - endpoint: udp://vector.${domainName}:6002/
+          - endpoint: udp://vector.${domainName}:6002
             format: json_lines
       sysfs:
         devices.system.cpu.intel_pstate.no_turbo: 1