v2.0.0

roles/restore_primary/tasks/main.yml

@@ -1,19 +1,122 @@
-- name: Stop postgres on primary
-  service:
+---
+# Przywrócenie primary z pg2 (replica)
+# Używany w przypadku awarii pg1
+
+# --- Auto-detect PostgreSQL version na replica ---
+- name: Detect PostgreSQL version on replica
+  shell: "ls -d /etc/postgresql/*/ | head -1 | xargs basename"
+  register: pg_version_detect
+  changed_when: false
+  delegate_to: pg2
+  run_once: true
+
+- name: Set PostgreSQL variables from detection
+  set_fact:
+    pg_version: "{{ pg_version_detect.stdout | trim }}"
+    pg_cluster: "main"
+    pg_data: "/var/lib/postgresql/{{ pg_version_detect.stdout | trim }}/main"
+    pg_bin_path: "/usr/lib/postgresql/{{ pg_version_detect.stdout | trim }}/bin"
+
+- name: Wait for replica to be accessible
+  wait_for:
+    host: pg2
+    port: 5432
+    timeout: 300
+
+# --- Promowanie replica ---
+- name: Promote replica to primary
+  shell: "{{ pg_bin_path }}/pg_ctl promote -D {{ pg_data }}"
+  become: true
+  become_user: postgres
+  delegate_to: pg2
+  register: promote_result
+
+- name: Wait for promoted replica to be ready
+  wait_for:
+    host: pg2
+    port: 5432
+    timeout: 60
+
+# --- Przywrócenie na pg1 ---
+- name: Stop PostgreSQL on primary (pg1)
+  systemd:
     name: postgresql
     state: stopped
+  ignore_errors: true
 
-- name: Remove old data on primary
+- name: Remove old cluster data on primary
   file:
     path: "{{ pg_data }}"
     state: absent
+  become: true
 
-- name: Restore base backup from replica
+# --- Base backup z nowego primary (pg2) ---
+- name: Perform base backup from new primary (pg2)
+  shell: |
+    pg_basebackup \
+      -h pg2 \
+      -D {{ pg_data }} \
+      -U replicator \
+      -Fp \
+      -Xs \
+      -P \
+      -R
+  become: true
   become_user: postgres
-  command: >
-    pg_basebackup -h pg2 -D {{ pg_data }} -U replicator -Fp -Xs -P -R
+  register: restore_basebackup
 
-- name: Start postgres on primary
-  service:
+- name: Fix permissions on restored data directory
+  file:
+    path: "{{ pg_data }}"
+    owner: postgres
+    group: postgres
+    mode: "0700"
+    recurse: yes
+  become: true
+
+- name: Configure recovery settings
+  lineinfile:
+    path: "{{ pg_data }}/standby.signal"
+    line: ""
+    create: yes
+    owner: postgres
+    group: postgres
+    mode: "0644"
+
+# --- Uruchomienie pg1 ---
+- name: Start PostgreSQL on primary
+  systemd:
     name: postgresql
     state: started
+    enabled: yes
+
+- name: Wait for PostgreSQL to start
+  wait_for:
+    path: "/var/run/postgresql/.s.PGSQL.5432"
+    timeout: 60
+
+# --- Weryfikacja ---
+- name: Verify primary is accessible
+  postgresql_query:
+    db: postgres
+    query: "SELECT version()"
+    login_unix_socket_directory: /var/run/postgresql
+  become: true
+  become_user: postgres
+  register: version_check
+
+- name: Display restoration result
+  debug:
+    msg: "Primary restored successfully. PostgreSQL: {{ version_check.query_result[0].version }}"
+
+- name: Display manual failover instructions
+  debug:
+    msg: |
+      Restoration complete!
+
+      Next steps:
+      1. Verify pg1 and pg2 connectivity
+      2. Update application connection strings if needed
+      3. Monitor replication status:
+         SELECT * FROM pg_stat_replication;
+      4. Schedule regular test restores