upgrade taskfile in preparation for talos v1.9.0

.taskfiles/talos/Taskfile.yaml

@@ -55,94 +55,99 @@ tasks:
 
   generate-clusterconfig:
     desc: Generate clusterconfig for Talos
+    preconditions:
+      - which test talhelper
+      - test -f {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talsecret.sops.yaml
+      - test -f {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talconfig.yaml
+    requires:
+      vars:
+        - K8S_CLUSTER
     cmds:
       - talhelper genconfig
         --env-file {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talenv.sops.yaml
         --secret-file {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talsecret.sops.yaml
         --config-file {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talconfig.yaml
         --out-dir {{.K8S_CLUSTER_DIR}}/bootstrap/talos/clusterconfig
-    requires:
-      vars:
-        - K8S_CLUSTER
-    preconditions:
-      - test -f {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talenv.sops.yaml
-      - test -f {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talsecret.sops.yaml
-      - test -f {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talconfig.yaml
-
-  upgrade:
-    desc: Upgrade Talos version for a node
-    vars:
-      TALOS_VERSION:
-        sh: |
-          yq -r ".talosVersion" {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talconfig.yaml
-      TALOS_IMAGE:
-        sh: |
-          talhelper genurl installer \
-            --env-file {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talenv.sops.yaml \
-            --config-file {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talconfig.yaml \
-          | grep {{.NODE}} \
-          | awk '{split($0,u," "); print u[2]}'
-    cmds:
-      - talosctl upgrade -n {{.NODE}} --image {{.TALOS_IMAGE }}
-    requires:
-      vars:
-        - K8S_CLUSTER
-        - NODE
-    preconditions:
-      - test -f {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talenv.sops.yaml
-      - test -f {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talconfig.yaml
-      - msg: "Talos image could not be determined for node={{.NODE}}"
-        sh: 'test -n "{{.TALOS_IMAGE}}"'
-
-  upgrade-k8s:
-    desc: Upgrade Kubernetes version for a Talos cluster
-    silent: false
-    vars:
-      KUBERNETES_VERSION:
-        sh: |
-          yq -r ".kubernetesVersion" {{.K8S_CLUSTER_DIR}}/bootstrap/talos/talconfig.yaml
-      TALOS_CONTROLLER:
-        sh: talosctl config info --output json | jq --raw-output '.endpoints[]' | shuf -n 1
-    cmds:
-      - until kubectl wait --timeout=5m --for=condition=Complete jobs --all --all-namespaces; do sleep 10; done
-      - talosctl upgrade-k8s -n {{.TALOS_CONTROLLER}} --to {{.KUBERNETES_VERSION}}
-    requires:
-      vars:
-        - K8S_CLUSTER
-    preconditions:
-      - talosctl config info &>/dev/null
-      - talosctl --nodes {{.TALOS_CONTROLLER}} get machineconfig &>/dev/null
 
   apply-clusterconfig:
     desc: Apply clusterconfig for a Talos cluster
     vars:
       CLUSTERCONFIG_FILES:
-        sh: find {{.K8S_CLUSTER_DIR}}/bootstrap/talos/clusterconfig -type f -name '*.yaml' -printf '%f\n'
+        sh: ls {{.K8S_CLUSTER_DIR}}/bootstrap/talos/clusterconfig/*.yaml
+    preconditions:
+      - which ls
+      - test -f "${TALOSCONFIG}"
+    requires:
+      vars:
+        - K8S_CLUSTER
     cmds:
       - for:
           var: CLUSTERCONFIG_FILES
         task: _apply-machineconfig
         vars:
-          filename: "{{.ITEM}}"
+          FILENAME: "{{.ITEM}}"
-          hostname: |-
+          HOSTNAME: |-
-            {{ trimPrefix (printf "%s-" .K8S_CLUSTER) .ITEM | trimSuffix ".yaml" }}
+            {{ trimPrefix (printf "%s-" .K8S_CLUSTER) (base .ITEM) | trimSuffix ".yaml" }}
           DRY_RUN: "{{ .DRY_RUN }}"
+          INSECURE: "{{ .INSECURE }}"
+
+  apply-node:
+    desc: Apply Talos config to a node [NODE=required]
+    preconditions:
+      - which talosctl
+      - test -f "${TALOSCONFIG}"
+      - talosctl --nodes {{.NODE}} get machineconfig
     requires:
       vars:
         - K8S_CLUSTER
+        - NODE
+    vars:
+      FILE:
+        sh: ls {{.K8S_CLUSTER_DIR}}/bootstrap/talos/clusterconfig/{{.K8S_CLUSTER}}-{{.NODE}}*.yaml
+    cmds:
+      - task: _apply-machineconfig
+        vars:
+          FILENAME: "{{.FILE}}"
+          HOSTNAME: "{{.NODE}}"
+      - talosctl --nodes {{.NODE}} health --wait-timeout=10m --server=false
+
+  upgrade-node:
+    desc: Upgrade Talos on a single node [NODE=required]
+    preconditions:
+      - which talosctl
+      - test -f "${TALOSCONFIG}"
+      - talosctl --nodes {{.NODE}} get machineconfig
+    requires:
+      vars:
+        - K8S_CLUSTER
+        - NODE
+    vars:
+      FILE:
+        sh: ls {{.K8S_CLUSTER_DIR}}/bootstrap/talos/clusterconfig/{{.K8S_CLUSTER}}-{{.NODE}}*.yaml
+      TALOS_IMAGE:
+        sh: yq '.machine.install.image' < "{{.FILE}}"
+    cmds:
+      - echo "Upgrading Talos on node {{.NODE}}"
+      - talosctl --nodes {{.NODE}} upgrade --image="{{.TALOS_IMAGE}}" --timeout=10m
+      - talosctl --nodes {{.NODE}} health --wait-timeout=10m --server=false
 
   _apply-machineconfig:
     internal: true
     desc: Apply a single Talos machineConfig to a Talos node
-    cmds:
+    vars:
-      - talosctl apply-config
+      MODE: '{{.MODE | default "auto"}}'
-        --nodes "{{.hostname}}"
+    preconditions:
-        --file "{{.K8S_CLUSTER_DIR}}/bootstrap/talos/clusterconfig/{{.filename}}"
+      - which talosctl
-        {{ if eq "true" .DRY_RUN }}--dry-run{{ end }}
+      - test -f "{{.FILENAME}}"
     requires:
       vars:
         - K8S_CLUSTER
-        - hostname
+        - HOSTNAME
-        - filename
+        - FILENAME
-    preconditions:
+    cmds:
-      - test -f {{.K8S_CLUSTER_DIR}}/bootstrap/talos/clusterconfig/{{.filename}}
+      - talosctl apply-config
+        --nodes "{{.HOSTNAME}}"
+        --file "{{.FILENAME}}"
+        --mode="{{.MODE}}"
+        {{ if eq "true" .INSECURE }}--insecure{{ end }}
+        {{ if eq "true" .DRY_RUN }}--dry-run{{ end }}

kubernetes/bootstrap/talos/talconfig.yaml

@@ -20,7 +20,7 @@ nodes:
     ipAddress: 10.1.1.62
     controlPlane: true
     installDiskSelector:
-      wwid: naa.55cd2e4150654449
+      busPath: /pci0000:00/0000:00:1f.2/ata1/host0/target0:0:0/0:0:0:0/
     networkInterfaces:
       - interface: eno1
         dhcp: true
@@ -28,15 +28,16 @@ nodes:
       customization:
         systemExtensions:
           officialExtensions:
+            - siderolabs/i915
             - siderolabs/intel-ucode
-            - siderolabs/i915-ucode
+            - siderolabs/glibc
 
   - hostname: frodo
     disableSearchDomain: true
     ipAddress: 10.1.1.63
     controlPlane: true
     installDiskSelector:
-      wwid: naa.55cd2e4150650b54
+      busPath: /pci0000:00/0000:00:1f.2/ata1/host0/target0:0:0/0:0:0:0/
     networkInterfaces:
       - interface: eno1
         dhcp: true
@@ -44,15 +45,16 @@ nodes:
       customization:
         systemExtensions:
           officialExtensions:
+            - siderolabs/i915
             - siderolabs/intel-ucode
-            - siderolabs/i915-ucode
+            - siderolabs/glibc
 
   - hostname: sam
     disableSearchDomain: true
     ipAddress: 10.1.1.64
     controlPlane: true
     installDiskSelector:
-      wwid: naa.55cd2e41521a8480
+      busPath: /pci0000:00/0000:00:1f.2/ata1/host0/target0:0:0/0:0:0:0/
     networkInterfaces:
       - interface: eno1
         dhcp: true
@@ -60,15 +62,16 @@ nodes:
       customization:
         systemExtensions:
           officialExtensions:
+            - siderolabs/i915
             - siderolabs/intel-ucode
-            - siderolabs/i915-ucode
+            - siderolabs/glibc
 
   - hostname: pippin
     disableSearchDomain: true
     ipAddress: 10.1.1.65
     controlPlane: false
     installDiskSelector:
-      wwid: naa.55cd2e4150155dbf
+      busPath: /pci0000:00/0000:00:1f.2/ata1/host0/target0:0:0/0:0:0:0/
     networkInterfaces:
       - interface: eno1
         dhcp: true
@@ -76,8 +79,9 @@ nodes:
       customization:
         systemExtensions:
           officialExtensions:
+            - siderolabs/i915
             - siderolabs/intel-ucode
-            - siderolabs/i915-ucode
+            - siderolabs/glibc
 
   - hostname: merry
     disableSearchDomain: true
@@ -103,15 +107,15 @@ nodes:
         systemExtensions:
           officialExtensions:
             - siderolabs/amd-ucode
-            - siderolabs/nonfree-kmod-nvidia-production
             - siderolabs/nvidia-container-toolkit-production
-
+            - siderolabs/nvidia-open-gpu-kernel-modules-production
+            - siderolabs/glibc
   - hostname: rosie
     disableSearchDomain: true
     ipAddress: 10.1.1.67
     controlPlane: false
     installDiskSelector:
-      wwid: naa.55cd2e4150a6183a
+      busPath: /pci0000:00/0000:00:1f.2/ata1/host0/target0:0:0/0:0:0:0/
     networkInterfaces:
       - interface: eno1
         dhcp: true
@@ -119,8 +123,9 @@ nodes:
       customization:
         systemExtensions:
           officialExtensions:
+            - siderolabs/i915
             - siderolabs/intel-ucode
-            - siderolabs/i915-ucode
+            - siderolabs/glibc
 
 worker:
   schematic: