Miscellaneous resilience improvements

ansible/ci/group_vars/all.yml

@@ -7,7 +7,7 @@ gcp_instance_prefix: ci
 gcp_available_zones:
   - us-central1-a
   - us-central1-b
-  - us-central1-c
+  - us-central1-d
   - us-central1-f
 ansible_user: runner
 gcp_network_name: collector-ci

ansible/dev/group_vars/all.yml

@@ -8,7 +8,6 @@ gcp_instance_prefix: "{{ lookup('env', 'USER') }}-dev"
 gcp_available_zones:
   - us-central1-a
   - us-central1-b
-  - us-central1-c
   - us-central1-f
 ansible_user: "{{ lookup('env', 'USER') }}"
 gcp_network_name: default

ansible/roles/create-vm/tasks/create-gcp-instance.yml

@@ -0,0 +1,39 @@
+---
+- name: Safely create GCP compute instance
+  block:
+    - name: Create compute instance
+      google.cloud.gcp_compute_instance:
+        name: "{{ vm_name }}"
+        project: "{{ gcp_project }}"
+        auth_kind: "{{ gcp_auth_kind }}"
+        service_account_file: "{{ gcp_service_account_file }}"
+        zone: "{{ gcp_zone }}"
+        machine_type: "{{ vm_machine_type | default('e2-standard-2') }}"
+        disks:
+          - auto_delete: true
+            boot: true
+            initialize_params:
+              disk_size_gb: "{{ vm_disk_size | default(20) }}"
+              source_image: "{{ gcp_source_image }}"
+        network_interfaces:
+          - network: "{{ network }}"
+            access_configs:
+              - name: External NAT
+                type: ONE_TO_ONE_NAT
+        labels: "{{ gcp_default_labels | combine(gcp_extra_labels) }}"
+        metadata: "{{ gcp_meta_data | default({}) }}"
+      register: instance_result
+
+    - set_fact:
+        instance_created: true
+      when: instance_result is defined and instance_result.changed
+  rescue:
+    - name: Remove failed GCP compute instance
+      google.cloud.gcp_compute_instance:
+        name: "{{ vm_name }}"
+        project: "{{ gcp_project }}"
+        auth_kind: "{{ gcp_auth_kind }}"
+        service_account_file: "{{ gcp_service_account_file }}"
+        zone: "{{ gcp_zone }}"
+        state: absent
+  when: not instance_created

ansible/roles/create-vm/tasks/create-gcp-vm.yml

@@ -50,27 +50,13 @@
     state: present
   register: network
 
+
+- set_fact:
+    instance_created: false
+
 - name: Create GCP VM
-  google.cloud.gcp_compute_instance:
-    name: "{{ vm_name }}"
-    project: "{{ gcp_project }}"
-    auth_kind: "{{ gcp_auth_kind }}"
-    service_account_file: "{{ gcp_service_account_file }}"
-    zone: "{{ gcp_zone }}"
-    machine_type: "{{ vm_machine_type | default('e2-standard-2') }}"
-    disks:
-      - auto_delete: true
-        boot: true
-        initialize_params:
-          disk_size_gb: "{{ vm_disk_size | default(20) }}"
-          source_image: "{{ gcp_source_image }}"
-    network_interfaces:
-      - network: "{{ network }}"
-        access_configs:
-          - name: External NAT
-            type: ONE_TO_ONE_NAT
-    labels: "{{ gcp_default_labels | combine(gcp_extra_labels) }}"
-    metadata: "{{ gcp_meta_data | default({}) }}"
+  ansible.builtin.include_tasks:
+    file: create-gcp-instance.yml
 
   # This looks like a lot, and its mostly because ansible support
   # for "breaking" from a loop is not all that great at the moment.
@@ -80,9 +66,5 @@
   loop: "{{ vm_available_zones }}"
   loop_control:
     loop_var: gcp_zone
-  register: instance_result
   when: >
-    (vm_arch == "amd64" or vm_arch == "arm64") and not condition and excluded == false
-  vars:
-    condition: "{{ (instance_result | default({'changed': false})).changed }}"
-
+    (vm_arch == "amd64" or vm_arch == "arm64") and excluded == false