feat(backend): 轮值策略可包含/排除指定业务 #8893

dbm-ui/backend/db_monitor/exceptions.py

@@ -40,3 +40,9 @@ class AutofixException(DBMonitorBaseException):
     ERROR_CODE = "203"
     MESSAGE = _("故障自愈异常")
     MESSAGE_TPL = _("故障自愈异常: {message}")
+
+
+class DutyRuleSaveException(DBMonitorBaseException):
+    ERROR_CODE = "204"
+    MESSAGE = _("轮值策略保存失败")
+    MESSAGE_TPL = _("轮值策略保存失败: {message}")

dbm-ui/backend/db_monitor/migrations/0021_dutyrule_biz_config.py

@@ -0,0 +1,18 @@
+# Generated by Django 3.2.25 on 2025-01-02 11:46
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ("db_monitor", "0020_auto_20240621_1216"),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name="dutyrule",
+            name="biz_config",
+            field=models.JSONField(default=dict, verbose_name="业务设置(包含业务include/排除业务exclude)"),
+        ),
+    ]

dbm-ui/backend/db_monitor/models/alarm.py

@@ -47,8 +47,9 @@
     BkMonitorDeleteAlarmException,
     BkMonitorSaveAlarmException,
     BuiltInNotAllowDeleteException,
+    DutyRuleSaveException,
 )
-from backend.db_monitor.tasks import update_app_policy
+from backend.db_monitor.tasks import delete_monitor_duty_rule, update_app_policy, update_db_notice_group
 from backend.db_monitor.utils import (
     bkm_delete_alarm_strategy,
     bkm_save_alarm_strategy,
@@ -167,12 +168,8 @@ def save_monitor_group(self) -> int:
             resp = BKMonitorV3Api.save_duty_rule(save_duty_rule_params, use_admin=True, raw=True)
             if resp.get("result"):
                 self.monitor_duty_rule_id = resp["data"]["id"]
-                monitor_duty_rule_ids = (
-                    DutyRule.objects.filter(db_type=self.db_type)
-                    .exclude(monitor_duty_rule_id=0)
-                    .order_by("-priority")
-                    .values_list("monitor_duty_rule_id", flat=True)
-                )
+                duty_rules = DutyRule.get_biz_db_duty_rules(self.bk_biz_id, self.db_type)
+                monitor_duty_rule_ids = [rule.monitor_duty_rule_id for rule in duty_rules]
                 save_monitor_group_params["need_duty"] = True
                 save_monitor_group_params["duty_rules"] = list(monitor_duty_rule_ids) + [self.monitor_duty_rule_id]
             else:
@@ -275,11 +272,15 @@ class DutyRule(AuditedModel):
     category = models.CharField(verbose_name=_("轮值类型"), choices=DutyRuleCategory.get_choices(), max_length=LEN_SHORT)
     db_type = models.CharField(_("数据库类型"), choices=DBType.get_choices(), max_length=LEN_SHORT)
     duty_arranges = models.JSONField(_("轮值人员设置"))
+    biz_config = models.JSONField(_("业务设置(包含业务include/排除业务exclude)"), default=dict)
 
     def save(self, *args, **kwargs):
         """
         保存轮值
         """
+        # 0. (前置校验)不允许同时存在包含业务和排除业务两个设置
+        if self.biz_config.get("include") and self.biz_config.get("exclude"):
+            raise DutyRuleSaveException(_("不允许通知存在包含业务和排除业务配置"))
         # 1. 新建监控轮值
         params = {
             "name": f"{self.db_type}_{self.name}",
@@ -343,26 +344,43 @@ def save(self, *args, **kwargs):
         # 3. 判断是否需要变更用户组
         # 3.1 非老规则（即新建的规则）
         need_update_user_group = not is_old_rule
-        # 3.2 调整了优先级的规则
+        # 3.2 调整了优先级的规则，或者调整了业务配置
         if self.pk:
             old_rule = DutyRule.objects.get(pk=self.pk)
-            if old_rule.priority != self.priority:
+            if old_rule.priority != self.priority or old_rule.biz_config != self.biz_config:
                 need_update_user_group = True
         # 4. 保存本地轮值规则
         super().save(*args, **kwargs)
-        # 5. 变更告警组
+        # 5. 变更告警组-异步执行
         if need_update_user_group:
-            for notice_group in NoticeGroup.objects.filter(is_built_in=True, db_type=self.db_type):
-                notice_group.save()
+            update_db_notice_group.delay(self.db_type)
 
     def delete(self, using=None, keep_parents=False):
-        BKMonitorV3Api.delete_duty_rules({"ids": [self.monitor_duty_rule_id], "bk_biz_ids": [env.DBA_APP_BK_BIZ_ID]})
+        """删除轮值"""
         super().delete()
+        delete_monitor_duty_rule.delay(self.db_type, self.monitor_duty_rule_id)
 
     @classmethod
     def priority_distinct(cls) -> list:
         return list(cls.objects.values_list("priority", flat=True).distinct().order_by("-priority"))
 
+    @classmethod
+    def get_biz_db_duty_rules(cls, bk_biz_id: int, db_type: str):
+        """获取指定业务DB组件的轮值策略"""
+        duty_rules = DutyRule.objects.filter(db_type=db_type).exclude(monitor_duty_rule_id=0).order_by("-priority")
+        active_biz_duty_rules: list = []
+
+        for rule in duty_rules:
+            # 如果业务不在包含名单，或者业务在排除名单，则本策略不属于该业务下
+            if rule.biz_config:
+                include, exclude = rule.biz_config.get("include"), rule.biz_config.get("exclude")
+                if (include and bk_biz_id not in include) or (exclude and bk_biz_id in exclude):
+                    continue
+            # 添加该业务下的轮值策略
+            active_biz_duty_rules.append(rule)
+
+        return active_biz_duty_rules
+
     class Meta:
         verbose_name_plural = verbose_name = _("轮值规则(DutyRule)")
 

dbm-ui/backend/db_monitor/serializers.py

@@ -9,6 +9,7 @@
 specific language governing permissions and limitations under the License.
 """
 import urllib.parse
+from collections import defaultdict
 
 from django.utils.translation import gettext_lazy as _
 from rest_framework import serializers
@@ -17,6 +18,7 @@
 from backend.bk_web.serializers import AuditedSerializer
 from backend.configuration.constants import DBType
 from backend.db_meta.enums import ClusterType
+from backend.db_meta.models import AppCache
 from backend.db_monitor import mock_data
 from backend.db_monitor.constants import AlertLevelEnum, DetectAlgEnum, OperatorEnum, TargetLevel
 from backend.db_monitor.exceptions import AutofixException
@@ -64,21 +66,36 @@ class Meta:
 
 
 class DutyRuleSerializer(AuditedSerializer, serializers.ModelSerializer):
+    biz_config_display = serializers.SerializerMethodField(help_text=_("业务配置信息"))
+
+    @property
+    def biz_name_map(self):
+        if not hasattr(self, "_biz_name_map"):
+            setattr(self, "_biz_name_map", AppCache.get_appcache(key="appcache_dict"))
+        return self._biz_name_map
+
     class Meta:
         model = DutyRule
         fields = "__all__"
 
+    def get_biz_config_display(self, obj):
+        biz_config_display = defaultdict(dict)
+        for key, bizs in obj.biz_config.items():
+            infos = [{"bk_biz_id": biz, "bk_biz_name": self.biz_name_map[str(biz)]["bk_biz_name"]} for biz in bizs]
+            biz_config_display[key] = infos
+        return biz_config_display
+
 
 class DutyRuleCreateSerializer(DutyRuleSerializer):
     class Meta:
-        model = NoticeGroup
+        model = DutyRule
         fields = "__all__"
         swagger_schema_fields = {"example": mock_data.CREATE_HANDOFF_DUTY_RULE}
 
 
 class DutyRuleUpdateSerializer(DutyRuleSerializer):
     class Meta:
-        model = NoticeGroup
+        model = DutyRule
         fields = "__all__"
         swagger_schema_fields = {"example": mock_data.CREATE_CUSTOM_DUTY_RULE}
 

dbm-ui/backend/db_monitor/tasks.py

@@ -13,7 +13,10 @@
 
 from celery import shared_task
 
+from backend import env
+from backend.components import BKMonitorV3Api
 from backend.configuration.constants import PLAT_BIZ_ID
+from backend.exceptions import ApiResultError
 
 logger = logging.getLogger("celery")
 
@@ -85,3 +88,27 @@ def update_app_policy(bk_biz_id, notify_group_id, db_type):
                 notify_group_id,
                 db_type,
             )
+
+
+@shared_task
+def update_db_notice_group(db_type: str):
+    """更新DB类型的告警组"""
+    from backend.db_monitor.models import NoticeGroup
+
+    for notice_group in NoticeGroup.objects.filter(is_built_in=True, db_type=db_type):
+        logger.info("[local_notice_group] update notice group: %s", notice_group.name)
+        notice_group.save()
+
+
+@shared_task
+def delete_monitor_duty_rule(db_type: str, monitor_duty_rule_id):
+    """解绑相关告警组，删除轮值策略，调用此函数之前保证轮值已从DBM中删除"""
+    update_db_notice_group(db_type)
+
+    logger.info("[duty_rule] delete duty rule: %s", monitor_duty_rule_id)
+
+    try:
+        BKMonitorV3Api.delete_duty_rules({"ids": [monitor_duty_rule_id], "bk_biz_ids": [env.DBA_APP_BK_BIZ_ID]})
+    except (ApiResultError, Exception) as e:
+        # 轮值删除错误暂可忽略，因为删除之前已经停用不会生效，并且在DBM数据也清理。只是会在监控平台留下一条脏数据
+        logger.error("[duty_rule] error in deleting duty: %s", e)