feat(backend): 监控目标支持自定义key条件 close #1589

TencentBlueKing · Nov 1, 2023 · 110130f · 110130f
1 parent 2f1f12a
commit 110130f
Show file tree

Hide file tree

Showing 93 changed files with 24,403 additions and 8,433 deletions.
diff --git a/dbm-ui/backend/db_monitor/management/commands/extract_alarm.py b/dbm-ui/backend/db_monitor/management/commands/extract_alarm.py
@@ -31,6 +31,7 @@ class Command(BaseCommand):
     def add_arguments(self, parser):
         parser.add_argument("db_type", choices=DBType.get_values(), type=str, help="db类型")
         parser.add_argument("bkmonitor_strategy_list", nargs="+", type=int, help="监控策略ID列表")
+        parser.add_argument("-c", "--custom-conditions", nargs="*", help="自定义过滤条件的key列表", type=str)
         parser.add_argument("-d", "--disabled", dest="is_disabled", action="store_true", help="disable by default")
 
     def to_template(self, db_type: str, instance: dict):
@@ -99,9 +100,10 @@ def clear_id(self, objs, id_name="id"):
 
     def handle(self, *args, **options):
         bkmonitor_strategy_list = options["bkmonitor_strategy_list"]
+        custom_conditions = options["custom_conditions"] or []
         db_type = options["db_type"]
         is_disabled = options["is_disabled"]
-
+        print(custom_conditions)
         res = BKMonitorV3Api.search_alarm_strategy_v3(
             {
                 "page": 1,
@@ -143,6 +145,7 @@ def handle(self, *args, **options):
             strategy_template["labels"] = sorted(set(strategy_template["labels"]))
 
             data_type_label = ""
+            custom_agg_conditions = []
             for item in strategy_template["items"]:
                 # 清空监控目标
                 item["target"] = []
@@ -178,19 +181,26 @@ def handle(self, *args, **options):
                     if "agg_dimension" in query_config and "appid" not in query_config["agg_dimension"]:
                         query_config["agg_dimension"].append("appid")
 
+                    custom_agg_conditions = list(
+                        filter(lambda x: x["key"] in custom_conditions, query_config.get("agg_condition", []))
+                    )
+
             self.clear_id(strategy_template["items"])
 
-            with open(os.path.join(TPLS_ALARM_DIR, f"{template_name}_test.json"), "w") as template_file:
+            with open(os.path.join(TPLS_ALARM_DIR, f"{template_name}.json"), "w") as template_file:
+                is_enabled = not is_disabled
+                strategy_template["is_enabled"] = is_enabled
                 template_dict = OrderedDict(
                     {
                         "bk_biz_id": 0,
-                        "version": 0,
-                        "is_enabled": not is_disabled,
                         "name": template_name,
-                        "details": strategy_template,
                         "db_type": db_type,
-                        "alert_source": data_type_label,
+                        "details": strategy_template,
+                        "is_enabled": is_enabled,
                         "monitor_indicator": strategy_template["items"][0]["name"],
+                        "version": 0,
+                        "alert_source": data_type_label,
+                        "custom_conditions": custom_agg_conditions,
                     }
                 )
 

diff --git a/dbm-ui/backend/db_monitor/management/commands/fix_policy_template.py b/dbm-ui/backend/db_monitor/management/commands/fix_policy_template.py
@@ -56,6 +56,7 @@ def handle(self, *args, **options):
         for alarm_json in alarm_jsons:
             with open(alarm_json, "r+") as f:
                 template_dict = json.loads(f.read())
+                template_dict["custom_conditions"] = []
 
                 old_template_name = template_dict["name"]
                 template_name = template_dict["name"]

diff --git a/dbm-ui/backend/db_monitor/migrations/0016_monitorpolicy_custom_conditions.py b/dbm-ui/backend/db_monitor/migrations/0016_monitorpolicy_custom_conditions.py
@@ -0,0 +1,18 @@
+# Generated by Django 3.2.19 on 2023-10-31 11:35
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ("db_monitor", "0015_auto_20231031_1238"),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name="monitorpolicy",
+            name="custom_conditions",
+            field=models.JSONField(default=list, verbose_name="自定义过滤列表"),
+        ),
+    ]
diff --git a/dbm-ui/backend/db_monitor/models/alarm.py b/dbm-ui/backend/db_monitor/models/alarm.py
@@ -523,6 +523,9 @@ class MonitorPolicy(AuditedModel):
     target_priority = models.PositiveIntegerField(verbose_name=_("监控策略优先级，跟随targets调整"))
     target_keyword = models.TextField(verbose_name=_("监控目标检索冗余字段"), default="")
 
+    # [{"key": "abc", "method": "eq", "value": ["aa", "bb"], "condition": "and", "dimension_name": "abc"}]
+    custom_conditions = models.JSONField(verbose_name=_("自定义过滤列表"), default=list)
+
     # Type 目前仅支持 Threshold
     # level: 1（致命）、2（预警）、3(提醒)
     # item[*].algorithms[*]:
@@ -707,10 +710,13 @@ def patch_priority_and_agg_conditions(self, details):
             for query_config in item["query_configs"]:
                 if "agg_condition" in query_config:
                     # remove same type conditions
+                    exclude_keys = list(map(lambda x: x["key"], self.custom_conditions)) + TargetLevel.get_values()
                     query_config_agg_condition = list(
-                        filter(lambda cond: cond["key"] not in TargetLevel.get_values(), query_config["agg_condition"])
+                        filter(lambda cond: cond["key"] not in exclude_keys, query_config["agg_condition"])
                     )
                     query_config_agg_condition.extend(agg_conditions)
+                    query_config_agg_condition.extend(self.custom_conditions)
+
                     # overwrite agg_condition
                     query_config["agg_condition"] = query_config_agg_condition
                 else:
@@ -807,8 +813,8 @@ def save(self, force_insert=False, force_update=False, using=None, update_fields
         super().save(force_insert, force_update, using, update_fields)
 
     def delete(self, using=None, keep_parents=False):
-        if self.bk_biz_id == PLAT_BIZ_ID:
-            raise BuiltInNotAllowDeleteException
+        # if self.bk_biz_id == PLAT_BIZ_ID:
+        #     raise BuiltInNotAllowDeleteException
 
         if self.monitor_policy_id:
             self.bkm_delete_alarm_strategy()

diff --git a/dbm-ui/backend/db_monitor/models/collect.py b/dbm-ui/backend/db_monitor/models/collect.py
@@ -115,7 +115,7 @@ def init_collect_strategy(bk_biz_id=env.DBA_APP_BK_BIZ_ID):
                     )
                     collect_params["id"] = collect_instance.collect_id
 
-                    if template.version != collect_instance.version:
+                    if template.version <= collect_instance.version:
                         logger.warning("[init_collect_strategy] skip update bkmonitor collector: %s " % template.name)
                         continue
 

diff --git a/dbm-ui/backend/db_monitor/readme.md b/dbm-ui/backend/db_monitor/readme.md
@@ -34,7 +34,7 @@ python manage.py extract_collect influxdb 6
 python manage.py extract_alarm mysql 39901 39900 39763 39759 39783 39760 39785 39779 39723 39738 39739 39787 39767 39751 39733 39762 39729 39782 39734 39788 39730 39768 39731 39769 39737 39721 39773 39786 39771 39742 39755 39745 39789 39918 39913 39917 39916 39915 39914 39912 39911
 python manage.py extract_alarm redis 39764 39743 39722 39744 39727 39776 39748 39774 39758 39718 39754 39750 39728 39925 39923 39922 39921
 python manage.py extract_alarm es 39777 39752 39717 39747 39756 39740 39757 39725
-python manage.py extract_alarm kafka 5676 5677 5678 5679 5685
+python manage.py extract_alarm kafka 39772 -d -c consumergroup topic
 python manage.py extract_alarm pulsar 5683 5682 5681 5680
 python manage.py extract_alarm hdfs 5891 5894 5895 5898 5899
 python manage.py extract_alarm influxdb 5946 5947 5948 5949 5950 5951 5952

diff --git a/dbm-ui/backend/db_monitor/serializers.py b/dbm-ui/backend/db_monitor/serializers.py
@@ -161,10 +161,11 @@ class TestRuleConfigSerializer(serializers.Serializer):
         child=serializers.ChoiceField(choices=NoticeSignalEnum.get_choices()), allow_empty=False
     )
     notify_groups = serializers.ListField(child=serializers.IntegerField(), allow_empty=True)
+    custom_conditions = serializers.ListSerializer(child=serializers.JSONField(), allow_empty=True)
 
     class Meta:
         model = MonitorPolicy
-        fields = ["targets", "test_rules", "notify_rules", "notify_groups"]
+        fields = ["targets", "test_rules", "notify_rules", "notify_groups", "custom_conditions"]
 
 
 class MonitorPolicyCloneSerializer(MonitorPolicyUpdateSerializer):

diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/ES 主机 CPU 使用率.json b/dbm-ui/backend/db_monitor/tpls/alarm/ES 主机 CPU 使用率.json
@@ -237,6 +237,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(CPU\u4f7f\u7528\u7387)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/ES 主机磁盘IO利用率.json b/dbm-ui/backend/db_monitor/tpls/alarm/ES 主机磁盘IO利用率.json
@@ -238,6 +238,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(I/O\u4f7f\u7528\u7387)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/ES 主机磁盘空间使用率.json b/dbm-ui/backend/db_monitor/tpls/alarm/ES 主机磁盘空间使用率.json
@@ -238,6 +238,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(\u78c1\u76d8\u7a7a\u95f4\u4f7f\u7528\u7387)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/ES 主机网络入流量.json b/dbm-ui/backend/db_monitor/tpls/alarm/ES 主机网络入流量.json
@@ -238,6 +238,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(\u7f51\u5361\u5165\u6d41\u91cf)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/ES 写拒绝.json b/dbm-ui/backend/db_monitor/tpls/alarm/ES 写拒绝.json
@@ -251,6 +251,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "MAX(ES\u7ebf\u7a0b\u6c60\u4e2d\u88ab\u62d2\u7edd\u4efb\u52a1\u6570)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/ES 写排队.json b/dbm-ui/backend/db_monitor/tpls/alarm/ES 写排队.json
@@ -245,6 +245,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "MAX(ES\u7ebf\u7a0b\u6c60\u4e2d\u7684\u6392\u961f\u4efb\u52a1\u6570)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/ES 搜索拒绝.json b/dbm-ui/backend/db_monitor/tpls/alarm/ES 搜索拒绝.json
@@ -251,6 +251,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "MAX(ES\u7ebf\u7a0b\u6c60\u4e2d\u88ab\u62d2\u7edd\u7684\u4efb\u52a1\u6570)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/ES 集群状态.json b/dbm-ui/backend/db_monitor/tpls/alarm/ES 集群状态.json
@@ -251,6 +251,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "MAX(ES\u96c6\u7fa4\u5065\u5eb7\u72b6\u6001)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/HDFS CorruptBlocks.json b/dbm-ui/backend/db_monitor/tpls/alarm/HDFS CorruptBlocks.json
@@ -305,6 +305,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(\u574f\u6570\u636e\u5757)/MIN(NameNode\u72b6\u6001)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/HDFS DataNode Last Contact.json b/dbm-ui/backend/db_monitor/tpls/alarm/HDFS DataNode Last Contact.json
@@ -232,6 +232,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "DataNode\u6700\u540e\u4e00\u6b21\u901a\u4fe1\u65f6\u95f4",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/HDFS EditLogRoll.json b/dbm-ui/backend/db_monitor/tpls/alarm/HDFS EditLogRoll.json
@@ -306,6 +306,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "NameNode\u7f16\u8f91\u65e5\u5fd7\u6eda\u52a8\u64cd\u4f5c\u6570",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/HDFS MissingBlocks.json b/dbm-ui/backend/db_monitor/tpls/alarm/HDFS MissingBlocks.json
@@ -305,6 +305,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "\u4e22\u5931\u7684\u6570\u636e\u5757\u6570\u91cf",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/HDFS NameNode Failover.json b/dbm-ui/backend/db_monitor/tpls/alarm/HDFS NameNode Failover.json
@@ -297,6 +297,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "NameNode\u6545\u969c\u8f6c\u79fb\u6b21\u6570",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/HDFS 容量使用率.json b/dbm-ui/backend/db_monitor/tpls/alarm/HDFS 容量使用率.json
@@ -322,6 +322,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "\u5bb9\u91cf\u4f7f\u7528\u7387",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB CPU 使用率告警.json b/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB CPU 使用率告警.json
@@ -277,6 +277,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(CPU\u4f7f\u7528\u7387)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB Series 增长量.json b/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB Series 增长量.json
@@ -251,6 +251,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(\u65f6\u95f4\u5e8f\u5217\u6570\u91cf)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 内存使用告警.json b/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 内存使用告警.json
@@ -277,6 +277,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(\u5e94\u7528\u7a0b\u5e8f\u5185\u5b58\u4f7f\u7528\u5360\u6bd4)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 写入异常告警.json b/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 写入异常告警.json
@@ -251,6 +251,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(\u5199\u5165\u5931\u8d25\u7684\u6570\u636e\u70b9\u6570\u91cf)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 实例端口异常.json b/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 实例端口异常.json
@@ -241,6 +241,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "MAX(http_response_result_code)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 磁盘 IO 利用率.json b/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 磁盘 IO 利用率.json
@@ -278,6 +278,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(I/O\u4f7f\u7528\u7387)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 磁盘空间告警.json b/dbm-ui/backend/db_monitor/tpls/alarm/InfluxDB 磁盘空间告警.json
@@ -278,6 +278,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "MAX(\u78c1\u76d8\u7a7a\u95f4\u4f7f\u7528\u7387)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }
diff --git a/dbm-ui/backend/db_monitor/tpls/alarm/Kafka 主机 CPU 使用率.json b/dbm-ui/backend/db_monitor/tpls/alarm/Kafka 主机 CPU 使用率.json
@@ -233,6 +233,7 @@
   },
   "is_enabled": true,
   "monitor_indicator": "AVG(CPU\u4f7f\u7528\u7387)",
-  "version": 10,
-  "alert_source": "time_series"
+  "version": 11,
+  "alert_source": "time_series",
+  "custom_conditions": []
 }