Initial managed-container HA resource agent baseline

2026-09-17 10:54:08 -04:00
commit b6c4fc848f
24 changed files with 4951 additions and 0 deletions
@@ -0,0 +1,110 @@
+# Sourced by managed-container; uses the shared OCF and parameter environment.
+
+managed_promote()
+{
+    local monitor_rc role i
+
+    # Belt-and-suspenders safety: promotion scoring already returns
+    # -INFINITY while this marker exists, but refuse the action itself too.
+    # This protects against an administrative/manual promotion request or a
+    # future scoring regression.
+    if pg_rejoin_pending; then
+        ocf_exit_reason \
+            "Cannot promote PostgreSQL while rejoin-required is set; instance must rejoin the elected primary first"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    managed_monitor
+    monitor_rc=$?
+
+    case "${monitor_rc}" in
+        "${OCF_RUNNING_PROMOTED}")
+            ocf_log info \
+                "PostgreSQL is already primary; reconciling primary invariants"
+
+            if ! ensure_local_ha_slots; then
+                ocf_exit_reason \
+                    "HA slot reconciliation failed on existing primary"
+                return "${OCF_ERR_GENERIC}"
+            fi
+
+            if ! ensure_rewind_privileges; then
+                ocf_exit_reason \
+                    "pg_rewind privilege reconciliation failed on existing primary"
+                return "${OCF_ERR_GENERIC}"
+            fi
+
+            pg_clear_rejoin_pending >/dev/null 2>&1 || true
+            return "${OCF_SUCCESS}"
+            ;;
+
+        "${OCF_SUCCESS}")
+            # Healthy standby: continue with promotion.
+            ;;
+
+        "${OCF_NOT_RUNNING}")
+            ocf_exit_reason \
+                "Cannot promote PostgreSQL because the resource is not running"
+            return "${OCF_ERR_GENERIC}"
+            ;;
+
+        *)
+            ocf_exit_reason \
+                "Cannot promote unhealthy PostgreSQL instance (monitor rc=${monitor_rc})"
+            return "${monitor_rc}"
+            ;;
+    esac
+
+    ocf_log info \
+        "Promoting PostgreSQL standby inside ${CONTAINER_NAME}"
+
+    if ! pg_command '
+        BINDIR="$(pg_config --bindir)" || exit 1
+        exec "$BINDIR/pg_ctl" \
+            -D "$PGDATA" \
+            promote -w
+    '
+    then
+        ocf_exit_reason "pg_ctl promote failed"
+        return "${OCF_ERR_GENERIC}"
+    fi
+
+    for i in $(seq 1 30); do
+        role="$(pg_role 2>/dev/null)" || {
+            sleep 1
+            continue
+        }
+
+        if [ "${role}" = "f" ]; then
+            ocf_log info \
+                "PostgreSQL became primary; reconciling primary invariants"
+
+            if ! ensure_local_ha_slots; then
+                ocf_exit_reason \
+                    "PostgreSQL promoted but HA slot reconciliation failed"
+                return "${OCF_ERR_GENERIC}"
+            fi
+
+            if ! ensure_rewind_privileges; then
+                ocf_exit_reason \
+                    "PostgreSQL promoted but pg_rewind privilege reconciliation failed"
+                return "${OCF_ERR_GENERIC}"
+            fi
+
+            if ! pg_clear_rejoin_pending; then
+                ocf_log warn \
+                    "PostgreSQL promoted successfully but provisional rejoin marker could not be cleared"
+            fi
+
+            ocf_log info \
+                "PostgreSQL promotion completed successfully"
+            return "${OCF_SUCCESS}"
+        fi
+
+        sleep 1
+    done
+
+    ocf_exit_reason \
+        "PostgreSQL did not become primary after promotion"
+    return "${OCF_ERR_GENERIC}"
+}