Fix multinode CI (#19)

* Fix multinode CI, fix review comments

.github/workflows/deploy-vagrant.yaml

@@ -13,18 +13,11 @@ jobs:
   deploy:
     runs-on:
       - self-hosted
-      - libvirt-vagrant
+      - vagrant
     steps:
       - name: Checkout x1 repository
         uses: actions/checkout@v4
 
-      - name: Checkout x1-cluster-profiles repository
-        uses: actions/checkout@v4
-        with:
-          repository: intel-sandbox/x1-cluster-profiles
-          path: x1-cluster-profiles
-          token: ${{ secrets.X1_GITHUB_TOKEN }}
-
       - name: Deploy X1 to Vagrant VMs
         run: ./scripts/ci/vagrant.sh
 

icl-cluster-profiles/profiles/ci.yaml

@@ -0,0 +1,42 @@
+# Generated from Kapitan target ci
+
+calico_endpoint_to_host_action: ACCEPT
+calico_ipip_mode: Never
+calico_network_backend: bird
+calico_vxlan_mode: Never
+containerd_use_config_path: true
+dns_etchosts: '{{ ansible_default_ipv4.address }} {{ ingress_domain }}
+
+  {{ ansible_default_ipv4.address }} hub.{{ ingress_domain }}
+
+  {{ ansible_default_ipv4.address }} jupyter.{{ ingress_domain }}
+
+  {{ ansible_default_ipv4.address }} prefect.{{ ingress_domain }}
+
+  {{ ansible_default_ipv4.address }} ray.{{ ingress_domain }}
+
+  {{ ansible_default_ipv4.address }} registry.{{ ingress_domain }}
+
+  {{ ansible_default_ipv4.address }} s3.{{ ingress_domain }}
+
+  '
+download_container: false
+ingress_domain: localtest.me
+kube_pods_subnet: 192.168.128.0/18
+kube_service_addresses: 192.168.64.0/18
+ntp_enabled: true
+ntp_manage_config: true
+vm_mode: true
+
+# Overwrite default Terraform variables. Keeping this for compatibility with the older installation
+# method. New method uses a dedicated terraform.tfvars file instead.
+# For a complete list of variables and their default values see x1/terraform/variables.tf.
+# Note that this variable is deprecated, use a generated "terraform.tfvars.json" instead.
+terraform_tfvars: |
+  ingress_domain = "localtest.me"
+  prefect_api_url = "http://prefect.localtest.me/api"
+  jupyterhub_pre_puller_enabled = false
+  prometheus_enabled = false
+  kubernetes_dashboard_enabled = false
+  use_node_ip_for_user_ports = true
+  use_external_node_ip_for_user_ports = true

scripts/ci/init-multipass.sh

@@ -24,7 +24,6 @@ host_key_checking = False
 EOF
 
     # Fix ERROR: Ansible could not initialize the preferred locale: unsupported locale setting
-    export LC_ALL="C.UTF-8"
     locale
 
     multipass launch --name "vm-$WORKFLOW_PREFIX_ID" -m "$VM_MEMORY"M -c "$VM_CPU" -d "$VM_DISK"G --cloud-init=./cloud-init.yaml rocky

scripts/ci/init-vagrant.sh

@@ -35,6 +35,7 @@ ensure_vagrant_plugins() {
 set_vagrant_env() {
     export X1_LIBVIRT_DEFAULT_PREFIX="$WORKFLOW_PREFIX_ID"
     export VAGRANT_DEFAULT_PROVIDER=libvirt
+    export VAGRANT_HOME=/.vagrant.d
 }
 
 vm_clean_before() {
@@ -62,6 +63,6 @@ vm_cleanup() {
     vm_clean_before
 }
 
-ensure_vagrant_plugins
 set_vagrant_env
+ensure_vagrant_plugins
 

scripts/ci/init.sh

@@ -4,6 +4,7 @@ set -e
 set -vx
 
 set_env() {
+    echo "HOSTNAME: "$(hostname)
     AGENT_ID=$(hostname | sed -e 's/.*[^0-9]//')
     AGENT_ID=${AGENT_ID:-100}
 
@@ -15,7 +16,7 @@ set_env() {
 
     WORKFLOW_PREFIX_ID="$WORKFLOW_PREFIX-$AGENT_ID"
 
-    X1_K8S_EXTRA_SETTINGS_FILE="$WORKSPACE_DIR/x1-cluster-profiles/profiles/ci.yaml"
+    X1_K8S_EXTRA_SETTINGS_FILE="$WORKSPACE_DIR/icl-cluster-profiles/profiles/ci.yaml"
     no_proxy=localtest.me,.localtest.me,$no_proxy
 
     VM_MEMORY=16384
@@ -30,6 +31,9 @@ set_env() {
 
     echo "WORKFLOW_DIR: $WORKFLOW_DIR"
     echo "X1_K8S_EXTRA_SETTINGS_FILE: $X1_K8S_EXTRA_SETTINGS_FILE"
+
+    export LC_ALL="C.UTF-8"
+    locale
 }
 
 run_kind() {

vagrant/Vagrantfile

@@ -21,7 +21,7 @@ Vagrant.configure(2) do |config|
 
   config.vm.provider "libvirt" do |lv|
     lv.driver = "kvm"
-    lv.storage_pool_name = "images"
+    lv.storage_pool_name = "default"
     lv.default_prefix = "#{libvirt_default_prefix}-"
   end