fix(backend): 修复spider异常无法提单问题 #8709

dbm-ui/backend/db_meta/api/cluster/tendbha/detail.py

@@ -52,19 +52,11 @@ def scan_cluster(cluster: Cluster) -> Graphic:
             dummy_slave_be_node, slave_be_group = graph.add_node(slave_be, to_group=slave_be_group)
             graph.add_line(source=slave_be_group, target=receiver_instance_group, label=LineLabel.Bind)
 
-        for otr in (
-            StorageInstanceTuple.objects.filter(ejector=ejector_instance)
-            .prefetch_related("cluster")
-            .exclude(receiver__cluster=cluster)
-        ):
+        for otr in StorageInstanceTuple.objects.filter(ejector=ejector_instance).exclude(receiver__cluster=cluster):
             foreign_receiver_cluster = otr.receiver.cluster.get()
             graph.add_foreign_cluster(ForeignRelationType.RepTo, foreign_receiver_cluster)
 
-        for otr in (
-            StorageInstanceTuple.objects.filter(receiver=ejector_instance)
-            .prefetch_related("cluster")
-            .exclude(ejector__cluster=cluster)
-        ):
+        for otr in StorageInstanceTuple.objects.filter(receiver=ejector_instance).exclude(ejector__cluster=cluster):
             foreign_ejector_cluster = otr.ejector.cluster.get()
             graph.add_foreign_cluster(ForeignRelationType.RepFrom, foreign_ejector_cluster)
 

dbm-ui/backend/tests/mock_data/components/itsm.py

@@ -79,7 +79,6 @@ def update_service(cls, *args, **kwargs):
     def create_ticket(cls, *args, **kwargs):
         response_data = copy.deepcopy(cls.base_info)
         response_data["data"] = {"sn": ticket_flow.SN}
-
         return response_data["data"]
 
     @classmethod
@@ -99,6 +98,22 @@ def ticket_approval_result(cls, *args, **kwargs):
 
         return response_data["data"]
 
+    @classmethod
+    def get_ticket_logs(cls, *args, **kwargs):
+        response_data = copy.deepcopy(cls.base_info)
+        response_data["data"] = {
+            "sn": "REQ20200831000005",
+            "title": "测试内置审批日志",
+            "create_at": "2020-08-31 20:57:22",
+            "creator": "xxx(xxx)",
+            "logs": [
+                {"operator": "xxx", "message": "流程开始"},
+                {"operator": "xxx", "message": "xxx 处理节点【提单】(提交)"},
+                {"operator": "admin", "message": "admin 处理节点【审批】(通过)"},
+            ],
+        }
+        return response_data["data"]
+
     @classmethod
     def get_ticket_info(cls, *args, **kwargs):
         response_data = copy.deepcopy(cls.base_info)

dbm-ui/backend/tests/mock_data/ticket/sqlserver_flow.py

@@ -140,12 +140,12 @@
         "bk_cloud_id": 0,
         "city_code": "深圳",
         "db_module_id": DB_MODULE_ID + 1,
-        "cluster_count": 1,
+        "cluster_count": 2,
         "inst_num": 1,
         "ip_source": "resource_pool",
         "nodes": {"backend": []},
         "resource_spec": {
-            "sqlserver_ha": {
+            "backend_group": {
                 "spec_id": 102,
                 "spec_name": "2核_4G_10G",
                 "spec_cluster_type": "sqlserver_ha",

dbm-ui/backend/tests/mock_data/ticket/ticket_flow.py

@@ -43,11 +43,9 @@
 MYSQL_FULL_BACKUP_TICKET_DATA = {
     "bk_biz_id": constant.BK_BIZ_ID,
     "details": {
-        "infos": {
-            "backup_type": "logical",
-            "file_tag": "DBFILE1M",
-            "clusters": [{"cluster_id": 1, "backup_local": "master"}],
-        }
+        "backup_type": "logical",
+        "file_tag": "DBFILE1M",
+        "infos": [{"cluster_id": 1, "backup_local": "master"}],
     },
     "remark": "",
     "ticket_type": "MYSQL_HA_FULL_BACKUP",

dbm-ui/backend/tests/ticket/test_ticket_revoke.py

@@ -22,7 +22,7 @@
 from backend.tests.mock_data.components.itsm import ItsmApiMock
 from backend.tests.mock_data.iam_app.permission import PermissionMock
 from backend.tests.mock_data.ticket.ticket_flow import MYSQL_FULL_BACKUP_TICKET_DATA, SN
-from backend.ticket.builders.mysql.mysql_ha_full_backup import MySQLHaFullBackupDetailSerializer
+from backend.ticket.builders.mysql.mysql_full_backup import MySQLFullBackupDetailSerializer
 from backend.ticket.constants import TicketStatus, TodoStatus, TodoType
 from backend.ticket.flow_manager.inner import InnerFlow
 from backend.ticket.handler import TicketHandler
@@ -46,19 +46,16 @@ class TestTicketRevoke:
     """
 
     @patch.object(TicketViewSet, "permission_classes")
-    @patch.object(MySQLHaFullBackupDetailSerializer, "validate")
     @patch.object(InnerFlow, "status", new_callable=PropertyMock)
     @patch.object(TicketViewSet, "get_permissions", lambda x: [])
+    @patch.object(MySQLFullBackupDetailSerializer, "validate", lambda self, attrs: attrs)
     @patch("backend.ticket.flow_manager.itsm.ItsmApi", ItsmApiMock())
     @patch("backend.db_services.cmdb.biz.CCApi", CCApiMock())
     @patch("backend.db_services.cmdb.biz.Permission", PermissionMock)
-    def test_ticket_revoke(
-        self, mocked_status, mocked_validate, mocked_permission_classes, query_fixture, db, init_app
-    ):
+    def test_ticket_revoke(self, mocked_status, mocked_permission_classes, db, init_app):
         # 以全库备份为例，测试流程：start --> itsm --> inner --> end
         mocked_status.return_value = TicketStatus.SUCCEEDED
         mocked_permission_classes.return_value = [AllowAny]
-        mocked_validate.return_value = MYSQL_FULL_BACKUP_TICKET_DATA
 
         client.login(username="admin")
         # 创建单据

dbm-ui/backend/ticket/builders/sqlserver/sqlserver_ha_apply.py

@@ -85,22 +85,30 @@ def format_cluster_domains(self) -> List[Dict[str, str]]:
 
     @classmethod
     def insert_ip_into_apply_infos(cls, ticket_data, infos: List[Dict]):
-        backend_nodes = ticket_data["nodes"]["backend_group"]
+        backend_nodes = ticket_data["nodes"][MachineType.SQLSERVER_HA.value]
         for index, apply_info in enumerate(infos):
-            # # 每组集群需要两个后端 IP 和两个 Proxy IP
-            # start, end = index * 2, (index + 1) * 2
-            apply_info["mssql_master_host"] = backend_nodes[index]["master"]
-            apply_info["mssql_slave_host"] = backend_nodes[index]["slave"]
+            # 每组集群需要两个后端 IP 和两个 Proxy IP
+            start, end = index * 2, (index + 1) * 2
+            apply_info["mssql_master_host"] = backend_nodes[start:end][0]
+            apply_info["mssql_slave_host"] = backend_nodes[start:end][1]
 
 
 class SQLServerHaApplyResourceParamBuilder(SQLServerSingleApplyResourceParamBuilder):
     def format(self):
         super().format()
 
+    @classmethod
+    def insert_ip_into_apply_infos(cls, ticket_data, infos: List[Dict]):
+        backend_nodes = ticket_data["nodes"]["backend_group"]
+        for index, apply_info in enumerate(infos):
+            # 每组集群需要两个后端 IP 和两个 Proxy IP
+            apply_info["mssql_master_host"] = backend_nodes[index]["master"]
+            apply_info["mssql_slave_host"] = backend_nodes[index]["slave"]
+
     def post_callback(self):
         next_flow = self.ticket.next_flow()
         infos = next_flow.details["ticket_data"]["infos"]
-        SQLServerHAApplyFlowParamBuilder.insert_ip_into_apply_infos(self.ticket.details, infos)
+        self.insert_ip_into_apply_infos(self.ticket.details, infos)
         next_flow.details["ticket_data"]["resource_spec"]["sqlserver_ha"] = next_flow.details["ticket_data"][
             "resource_spec"
         ]["master"]

dbm-ui/backend/ticket/builders/tendbcluster/base.py

@@ -51,10 +51,8 @@ class TendbBaseOperateDetailSerializer(MySQLBaseOperateDetailSerializer):
 
     #  实例不可用时，还能正常提单类型的白名单
     # spider 接入层异常, 只允许修复接入层异常的单据 1. 踢出故障 spider 2. 上架 (扩容) 新的 spider
-    # slave或者 spider 不可用不影响分区单据
-    SPIDER_UNAVAILABLE_WHITELIST = [
-        TicketType.TENDBCLUSTER_PARTITION.value,
-    ]
+    # slave或者 spider 不可用不影响分区单据。TODO：暂时放开spider异常限制
+    SPIDER_UNAVAILABLE_WHITELIST = TicketType.get_values()
     # 存储层 master 异常 (dbha 因为某些问题未正常介入),只有切换单据可用
     REMOTE_MASTER_UNAVAILABLE_WHITELIST = [
         TicketType.TENDBCLUSTER_MASTER_SLAVE_SWITCH,

dbm-ui/backend/ticket/flow_manager/itsm.py

@@ -31,20 +31,34 @@ def __init__(self, flow_obj: Flow):
     @property
     def ticket_approval_result(self):
         # 优先读取缓存，避免同一个对象内多次请求 ITSM
-        # TODO: ITSM接口请求比较缓慢
         if getattr(self, "_ticket_approval_result", None):
             return self._ticket_approval_result
 
         # 调用ITSM接口查询审批状态
-        data = ItsmApi.ticket_approval_result({"sn": [self.flow_obj.flow_obj_id]}, use_admin=True)
         try:
+            data = ItsmApi.ticket_approval_result({"sn": [self.flow_obj.flow_obj_id]}, use_admin=True)
             itsm_ticket_result = data[0]
-        except IndexError:
+        except (IndexError, ApiResultError):
             itsm_ticket_result = None
 
         setattr(self, "_ticket_approval_result", itsm_ticket_result)
         return itsm_ticket_result
 
+    @property
+    def ticket_logs(self):
+        # 同ticket_approval_result，优先读取缓存
+        if getattr(self, "_ticket_logs", None):
+            return self._ticket_logs
+
+        try:
+            itsm_logs = ItsmApi.get_ticket_logs({"sn": [self.flow_obj.flow_obj_id]}, use_admin=True)
+            ticket_logs = itsm_logs["logs"]
+        except (KeyError, ApiResultError):
+            ticket_logs = []
+
+        setattr(self, "_ticket_logs", ticket_logs)
+        return ticket_logs
+
     @property
     def _start_time(self) -> str:
         return datetime2str(self.flow_obj.create_at)
@@ -58,23 +72,18 @@ def _end_time(self) -> Union[datetime, Any]:
 
     @property
     def _summary(self) -> dict:
-        try:
-            logs = ItsmApi.get_ticket_logs({"sn": [self.flow_obj.flow_obj_id]})
-        except ApiResultError:
-            return _("未知单据")
-
         # 获取单据审批状态
         current_status = self.ticket_approval_result["current_status"]
         approve_result = self.ticket_approval_result["approve_result"]
         summary = {"status": current_status, "approve_result": approve_result}
 
         # 目前审批流程是固定的，取流程中第三个节点的日志作为概览即可
         try:
-            summary.update(operator=logs["logs"][2]["operator"], message=logs["logs"][2]["message"])
+            summary.update(operator=self.ticket_logs[2]["operator"], message=self.ticket_logs[2]["message"])
         except (IndexError, KeyError):
             # 异常时根据状态取默认的概览
             msg = TicketStatus.get_choice_label(self.status)
-            summary.update(operator=logs["logs"][-1]["operator"], status=self.status, message=msg)
+            summary.update(operator=self.ticket_logs[-1]["operator"], status=self.status, message=msg)
         return summary
 
     @property
@@ -88,8 +97,8 @@ def _status(self) -> str:
             return self.flow_obj.update_status(TicketFlowStatus.RUNNING)
 
         todo = self.flow_obj.todo_of_flow.first()
-        updater = self.ticket_approval_result["updated_by"]
-
+        # 非进行中的单据，肯定已经来到了第三个节点，否则也无法处理todo
+        updater = self.ticket_logs[2]["operator"]
         # 撤单
         if current_status == ItsmTicketStatus.REVOKED:
             todo.set_status(username=updater, status=TodoStatus.DONE_FAILED)