ceph_migrate refactor restart mgr handler as task to fix role usage

This is a follow-up to #1089.

Previously, the 'restart mgr' operation was implemented as a handler.
The previous handler approach caused unexpected behaviors, as handlers are only executed
once at the end of the playbook, even if notified multiple times. Additionally, the last
restart must call 'cephadm' directly on the compute node, not via the ceph_cli container
on the controller node. This commit moves the restart logic inline, directly after host
removal in the MON removal block.

- The handler file (handlers/main.yml) is deleted.
- The restart mgr command is now an explicit task, using 'cephadm shell -- ceph mgr fail'
  and running on a ComputeHCI node when available, to ensure manager failover is properly triggered.

Signed-off-by: Roberto Alfieri <[email protected]>

Author: rebtoor

tests/roles/ceph_migrate/handlers/main.yml

@@ -81,4 +81,12 @@
       when: lsh.stdout | from_json | community.general.json_query('[*].hostname') | length > 0
       ansible.builtin.command:
         "{{ ceph_cli }} orch host rm {{ cur_mon }} --force"
-      notify: restart mgr
+
+    - name: restart mgr on compute
+      become: true
+      ansible.builtin.command:
+        "cephadm shell -- ceph mgr fail"
+      delegate_to: "{{ groups['ComputeHCI'][0] | default(inventory_hostname) }}"
+      when:
+        - groups['ComputeHCI'] is defined
+        - groups['ComputeHCI'] | length > 0

tests/roles/ceph_migrate/tasks/drain.yaml

@@ -120,13 +120,17 @@
         CEPH_CONTAINER_IMAGE: "{{ ceph_container }}"
         CEPH_CONTAINER_BINARY: "{{ ceph_container_cli }}"
         CEPH_CONF: "{{ ceph_config_tmp_client_home }}"
-      notify: restart mgr
 
     - name: Print the resulting spec
       when: debug | default(false)
       ansible.builtin.debug:
         msg: "{{ spc }}"
 
+    - name: restart mgr
+      become: true
+      ansible.builtin.command:
+        "{{ ceph_cli }} mgr fail"
+
     # Check if a mon exists in the target node and remove it
     - name: MON - Get tmp mon
       ansible.builtin.command: "{{ ceph_cli }} orch ps --daemon_type mon --daemon_id {{ daemon_id }} -f json"
@@ -143,7 +147,11 @@
       delay: "{{ ceph_retry_delay }}"
       loop_control:
         label: "MON - Get tmp mon"
-      notify: restart mgr
+
+    - name: restart mgr
+      become: true
+      ansible.builtin.command:
+        "{{ ceph_cli }} mgr fail"
 
     - name: MON - Wait for the current mon to be deleted
       ansible.builtin.pause:
@@ -157,7 +165,11 @@
     - name: MON - Redeploy mon on {{ target_node }}
       ansible.builtin.command:
         "{{ ceph_cli }} orch daemon add mon {{ target_node }}:{{ mon_ipaddr }}"
-      notify: restart mgr
+
+    - name: restart mgr
+      become: true
+      ansible.builtin.command:
+        "{{ ceph_cli }} mgr fail"
 
     - name: MON - Wait for the spec to be updated
       ansible.builtin.pause:
@@ -197,7 +209,11 @@
     CEPH_CONTAINER_IMAGE: "{{ ceph_container }}"
     CEPH_CONTAINER_BINARY: "{{ ceph_container_cli }}"
     CEPH_CONF: "{{ ceph_config_tmp_client_home }}"
-  notify: restart mgr
+
+- name: restart mgr
+  become: true
+  ansible.builtin.command:
+    "{{ ceph_cli }} mgr fail"
 
 # Wait for the redeploy to finish before moving to the next stage
 - name: MON - wait daemons