fix: cloud init problems with ubuntu (#1150) (#1158)

* fix: cloud init problems with ubuntu (#1150)

* fix: install guest info if cloud init version is > 23.1.2

* ci: adds basic ubuntu test

* fix: adds handling for redhat system and cloud init

* !fixup: higher parallel

* fix: adds override for ubuntu tests

* do not merge: build basic os for ubuntu

* do not merge: ensure new cloud init for tests

* fix: remove package pining

* fix: remvove cloud-init dropin for custom vmware guest datasource

* fix: magefile configuration to create vsphere infa override file

---------

Co-authored-by: Shalin Patel <shalin.patel@nutanix.com>

* ci: run rocky basic e2e tests

---------

Co-authored-by: Faiq <faiq.raza@nutanix.com>

.github/workflows/release-vsphere-template.yaml

@@ -10,7 +10,7 @@ jobs:
   build-e2e:
     strategy:
       fail-fast: false
-      max-parallel: 3
+      max-parallel: 10
       matrix:
         include:
           - os: "redhat 8.4"
@@ -22,7 +22,7 @@ jobs:
           - os: "redhat 8.4"
             buildConfig: "offline-fips"
           - os: "ubuntu 20.04"
-            buildConfig: "offline"
+            buildConfig: "basic"
           - os: "rocky 9.1"
             buildConfig: "offline"
           - os: "flatcar"

.github/workflows/vsphere-e2e.yaml

@@ -12,7 +12,7 @@ jobs:
   rune2e:
     strategy:
       fail-fast: false
-      max-parallel: 5
+      max-parallel: 8
       matrix:
         include:
         - os: "redhat 7.9"
@@ -33,8 +33,12 @@ jobs:
           buildConfig: "offline-fips"
         - os: "ubuntu 20.04"
           buildConfig: "offline"
+        - os: "ubuntu 20.04"
+          buildConfig: "basic"
         - os: "rocky 9.1"
           buildConfig: "offline"
+        - os: "rocky 9.1"
+          buildConfig: "basic"
         - os: "flatcar"
           buildConfig: "basic"
     runs-on: 

ansible/roles/providers/tasks/main.yml

@@ -30,3 +30,8 @@
 
 - include_tasks: misc.yml
   when: packer_builder_type and packer_builder_type != ""
+
+- include_tasks: vmware-redhat.yaml
+  when: 
+  - packer_builder_type is search('vmware') or packer_builder_type is search('vsphere')
+  - ansible_os_family == "RedHat"

ansible/roles/providers/tasks/vmware-redhat.yaml

@@ -0,0 +1,52 @@
+- name: Get package facts
+  ansible.builtin.package_facts:
+    manager: "auto"
+
+- name: Set cloud-init version
+  ansible.builtin.set_fact:
+    cloud_init_version: "{{ ansible_facts.packages['cloud-init'][0].version }}"
+
+# this program used by ds-identify to determine whether or not the
+# VMwareGuestInfo datasource is useable.
+- name: Directly install GuestInfo
+  when: cloud_init_version is version('21.3', '<')
+  block:
+
+    - name: Copy vmware guestinfo datasource
+      copy:
+        src: "{{ item }}"
+        dest: /tmp/
+        owner: root
+        group: root
+        mode: 0755
+      with_items:
+        - cloud-init-vmware.sh
+        - DataSourceVMwareGuestInfo.py
+
+    - name: Create ds-check program to verify VMwareGuestInfo datasource
+      copy:
+        src: files/dscheck_VMwareGuestInfo.sh
+        dest: /usr/bin/dscheck_VMwareGuestInfo
+        owner: root
+        group: root
+        mode: 0755
+
+    - name: Execute cloud-init-vmware.sh
+      shell: bash -o errexit -o pipefail /tmp/cloud-init-vmware.sh
+      environment:
+        VMWARE_DS_PATH: '/tmp/DataSourceVMwareGuestInfo.py'
+
+    - name: Remove cloud-init-vmware.sh
+      file:
+        path: /tmp/cloud-init-vmware.sh
+        state: absent
+
+    # sets the datasource_list to VMwareGuestInfo for all OS
+    # ensure that VMwareGuestInfo overrides existing datasource drop-ins if it exists.
+    - name: Copy cloud-init config file for vmware
+      copy:
+        src: files/etc/cloud/cloud.cfg.d/99-DataSourceVMwareGuestInfo.cfg
+        dest: /etc/cloud/cloud.cfg.d/99-DataSourceVMwareGuestInfo.cfg
+        owner: root
+        group: root
+        mode: 0644

ansible/roles/providers/tasks/vmware.yml

@@ -25,6 +25,7 @@
       - cloud-initramfs-dyn-netconf
   when: ansible_os_family == "Debian"
 
+
 - name: Install cloud-init packages
   yum:
     name: "{{ packages }}"
@@ -58,36 +59,6 @@
     packages: "cloud-init cloud-utils python3-netifaces"
   when: ansible_os_family == "VMware Photon OS"
 
-# pip on CentOS needs to be upgraded, but since it's still
-# Python 2.7, need < 21.0
-- name: Upgrade pip
-  pip:
-    name: pip<21.0
-    extra_args: "{{ '--no-index --find-links=' + pip_packages_remote_filesystem_repo_path if offline_mode_enabled }}"
-    state: forcereinstall
-  when: ansible_os_family == "RedHat" and ansible_distribution_major_version == '7'
-
-- name: Copy vmware guestinfo datasource
-  copy:
-    src: "{{ item }}"
-    dest: /tmp/
-    owner: root
-    group: root
-    mode: 0755
-  with_items:
-    - cloud-init-vmware.sh
-    - DataSourceVMwareGuestInfo.py
-  when: ansible_os_family != "Flatcar"
-
-- name: Copy cloud-init config file for vmware
-  copy:
-    src: files/etc/cloud/cloud.cfg.d/99-DataSourceVMwareGuestInfo.cfg
-    dest: /etc/cloud/cloud.cfg.d/99-DataSourceVMwareGuestInfo.cfg
-    owner: root
-    group: root
-    mode: 0644
-  when: ansible_os_family != "Flatcar"
-
 - name: Remove subiquity-disable-cloudinit-networking.cfg
   file:
     path: /etc/cloud/cloud.cfg.d/subiquity-disable-cloudinit-networking.cfg
@@ -100,29 +71,6 @@
     state: absent
   when: ansible_os_family != "Flatcar"
 
-# this program used by ds-identify to determine whether or not the
-# VMwareGuestInfo datasource is useable.
-- name: Create ds-check program to verify VMwareGuestInfo datasource
-  copy:
-    src: files/dscheck_VMwareGuestInfo.sh
-    dest: /usr/bin/dscheck_VMwareGuestInfo
-    owner: root
-    group: root
-    mode: 0755
-  when: ansible_os_family != "Flatcar"
-
-- name: Execute cloud-init-vmware.sh
-  shell: bash -o errexit -o pipefail /tmp/cloud-init-vmware.sh
-  environment:
-    VMWARE_DS_PATH: '/tmp/DataSourceVMwareGuestInfo.py'
-  when: ansible_os_family != "Flatcar"
-
-- name: Remove cloud-init-vmware.sh
-  file:
-    path: /tmp/cloud-init-vmware.sh
-    state: absent
-  when: ansible_os_family != "Flatcar"
-
 - name: >-
     Remove cloud-init /etc/cloud/cloud.cfg.d/99-disable-networking-config.cfg
   file:
@@ -166,15 +114,15 @@
   when: ('ufw.service' in ansible_facts.services) and (ansible_os_family == "Debian")
 
 # See https://kb.vmware.com/s/article/82229
-# From systemd docs: 
-# "If a valid D-Bus machine ID is already configured for the system, 
+# From systemd docs:
+# "If a valid D-Bus machine ID is already configured for the system,
 # the D-Bus machine ID is copied and used to initialize the machine ID in /etc/machine-id"
 # This needs to be reset/truncated as well on Ubuntu.
 - name: Truncate D-Bus machine-id
   file:
     path: /var/lib/dbus/machine-id
     state: absent
-  when: ansible_os_family == "Debian" 
+  when: ansible_os_family == "Debian"
 
 - name: Link D-Bus machine-id to /etc/machine-id
   file:

magefile.go

@@ -140,14 +140,14 @@ func RunE2e(buildOS, buildConfig, buildInfra string, dryRun bool) error {
 		}
 	}
 
-	if buildOS == "flatcar" && buildInfra == ova {
-		flatcarOverride := "packer-ova-flatcar-override.yaml"
-		flatcarOverrideFlag := fmt.Sprintf("--overrides=%s", flatcarOverride)
-		overrideFlagForCmd = append(overrideFlagForCmd, flatcarOverrideFlag)
-		fmt.Printf("making flatcar override %s \n", flatcarOverride)
+	if buildConfig == basic && buildInfra == ova {
+		basicOverride := "packer-ova-basic-override.yaml"
+		basicOverrideFlag := fmt.Sprintf("--overrides=%s", basicOverride)
+		overrideFlagForCmd = append(overrideFlagForCmd, basicOverrideFlag)
+		fmt.Printf("making basic override %s \n", basicOverride)
 		// TODO: @faiq - move this to mage
-		if err := sh.RunV("make", flatcarOverride); err != nil {
-			return fmt.Errorf("failed to override for flatcar ova %s %v", flatcarOverride, err)
+		if err := sh.RunV("make", basicOverride); err != nil {
+			return fmt.Errorf("failed to override for basic ova %s %v", basicOverride, err)
 		}
 	}
 
@@ -391,18 +391,19 @@ func downloadAirgappedArtifacts(buildOS, buildConfig string) error {
 	if err := fetchPipPackages(artifactsDir); err != nil {
 		return fmt.Errorf("failed to fetch pip packages %w", err)
 	}
-	if buildConfig == offlineNvidia {
+	isGPU := buildConfig == offlineNvidia
+	if isGPU {
 		if err := fetchNvidiaRunFile(); err != nil {
 			return fmt.Errorf("failed to fetch nvidiaRunFile %w", err)
 		}
 	}
-	if err := createOSBundle(buildOS, kubeVersion, artifactsDir, isFips); err != nil {
+	if err := createOSBundle(buildOS, kubeVersion, artifactsDir, isFips, isGPU); err != nil {
 		return fmt.Errorf("failed to fetch OS bundle %w", err)
 	}
 	return nil
 }
 
-func createOSBundle(osName, kubernetesVersion, downloadDir string, fips bool) error {
+func createOSBundle(osName, kubernetesVersion, downloadDir string, fips, gpu bool) error {
 	osInfo := strings.Replace(osName, " ", "-", 1)
 	args := []string{
 		"create-package-bundle", fmt.Sprintf("--os=%s", osInfo),
@@ -411,6 +412,12 @@ func createOSBundle(osName, kubernetesVersion, downloadDir string, fips bool) er
 	if fips {
 		args = append(args, "--fips=true")
 	}
+	if osName == "redhat 8.8" || osName == "redhat 8.6" {
+		args = append(args, "--enable-eus-repos=true")
+	}
+	if gpu {
+		args = append(args, "--fetch-kernel-headers=true")
+	}
 	return sh.RunV(wrapperCmd, args...)
 }
 

test/infra/Makefile

@@ -77,8 +77,8 @@ packer-aws-offline-override.yaml: infra.aws.create install-envsubst
 packer-ova-offline-override.yaml: infra.vsphere.create install-envsubst
 	@$(INFRA_MODULES_DIR)/vsphere/export.sh $@
 
-packer-ova-flatcar-override.yaml: install-envsubst
-	"${ENVSUBST_ASSETS}"/envsubst < "$(INFRA_MODULES_DIR)"/vsphere/packer-vsphere-flatcar.yaml.tmpl > $@
+packer-ova-basic-override.yaml: install-envsubst
+	"${ENVSUBST_ASSETS}"/envsubst < "$(INFRA_MODULES_DIR)"/vsphere/packer-vsphere-basic.yaml.tmpl > $@
 
 .PHONY: infra.aws.destroy
 infra.aws.destroy: ## Destroy infrastructure