cloud_init_wait: factor out to file_wait

7 files changed, 44 insertions, 29 deletions

diff --git a/roles/cloud_init_wait/defaults/main.yml b/roles/cloud_init_wait/defaults/main.yml
index e5971c9..b857f1d 100644
--- a/roles/cloud_init_wait/defaults/main.yml
+++ b/roles/cloud_init_wait/defaults/main.yml
@@ -1,3 +1,2 @@
 cloud_init_wait_path: /var/lib/cloud/instance/boot-finished
 cloud_init_wait_seconds: 300
-cloud_init_wait_delay: 10
diff --git a/roles/cloud_init_wait/tasks/check.yml b/roles/cloud_init_wait/tasks/check.yml
deleted file mode 100644
index 2dc20aa..0000000
--- a/roles/cloud_init_wait/tasks/check.yml
+++ /dev/null
@@ -1,23 +0,0 @@
-# Make a playbook usable immediately after creating a VM - even before
-# the SSH server is started.
-- name: Wait for SSH connection
-  ansible.builtin.wait_for_connection:
-    timeout: '{{ cloud_init_wait_seconds }}'
-
-# Again, make a playbook usable immediately after creating a VM - that includes
-# waiting until cloud-init is done.
-- name: Wait until cloud-init is finished
-  ansible.builtin.file:
-    path: '{{ cloud_init_wait_path }}'
-    state: file
-  register: boot_finished
-  until: boot_finished is succeeded
-  # About 5 minutes worth of attempts.
-  retries: '{{ cloud_init_wait_seconds // cloud_init_wait_delay }}'
-  delay: '{{ cloud_init_wait_delay }}'
-  ignore_unreachable: true
-  ignore_errors: true
-
-- ansible.builtin.include_tasks: check.yml
-  when:
-    - boot_finished is unreachable
diff --git a/roles/cloud_init_wait/tasks/main.yml b/roles/cloud_init_wait/tasks/main.yml
index dec4162..22c258d 100644
--- a/roles/cloud_init_wait/tasks/main.yml
+++ b/roles/cloud_init_wait/tasks/main.yml
@@ -1,5 +1,5 @@
-- ansible.builtin.include_tasks: check.yml
-
-- ansible.builtin.fail:
-    msg: Failed to wait until cloud-init is done
-  when: boot_finished is not defined or boot_finished is not succeeded
+- ansible.builtin.include_role:
+    name: file_wait
+  vars:
+    file_wait_path: '{{ cloud_init_wait_path }}'
+    file_wait_seconds: '{{ cloud_init_wait_seconds }}'
diff --git a/roles/file_wait/README.md b/roles/file_wait/README.md
new file mode 100644
index 0000000..0f4c22b
--- /dev/null
+++ b/roles/file_wait/README.md
@@ -0,0 +1,4 @@
+This weird and extremely convoluted way to wait until a file exists
+(disregarding reboots) was borrowed from RedHat themselves:
+
+    https://www.ansible.com/blog/tolerable-ansible
diff --git a/roles/file_wait/defaults/main.yml b/roles/file_wait/defaults/main.yml
new file mode 100644
index 0000000..1e8625c
--- /dev/null
+++ b/roles/file_wait/defaults/main.yml
@@ -0,0 +1,3 @@
+file_wait_seconds: 300
+file_wait_delay: 10
+file_wait_reboots: 3
diff --git a/roles/file_wait/tasks/check.yml b/roles/file_wait/tasks/check.yml
new file mode 100644
index 0000000..1a7c138
--- /dev/null
+++ b/roles/file_wait/tasks/check.yml
@@ -0,0 +1,25 @@
+- name: Wait for connection
+  ansible.builtin.wait_for_connection:
+    timeout: '{{ file_wait_seconds }}'
+
+- name: Check if the file exists
+  ansible.builtin.file:
+    path: '{{ file_wait_path }}'
+    state: file
+  register: file_wait_check
+  until: file_wait_check is succeeded
+  retries: '{{ file_wait_seconds // file_wait_delay }}'
+  delay: '{{ file_wait_delay }}'
+  ignore_unreachable: true
+  ignore_errors: true
+
+- name: If the host restarted, try again
+  block:
+    - name: Note a reboot happened
+      ansible.builtin.set_fact:
+        file_wait_reboots: '{{ (file_wait_reboots | int) - 1 }}'
+
+    - name: Retry if there're more attempts
+      ansible.builtin.include_tasks: check.yml
+      when: (file_wait_reboots | int > 0)
+  when: file_wait_check is unreachable
diff --git a/roles/file_wait/tasks/main.yml b/roles/file_wait/tasks/main.yml
new file mode 100644
index 0000000..76dce68
--- /dev/null
+++ b/roles/file_wait/tasks/main.yml
@@ -0,0 +1,7 @@
+- name: Checking if the file exists
+  ansible.builtin.include_tasks: check.yml
+
+- name: Fail unless the file exists
+  ansible.builtin.fail:
+    msg: 'File is missing: {{ file_wait_path }}'
+  when: file_wait_check is not defined or file_wait_check is not succeeded or file_wait_check is unreachable