feat: 监控告警内容注入自定义文档链接

apiserver/paasng/paasng/accessories/smart_advisor/tags.py

@@ -64,6 +64,7 @@ def __init__(self, tag_value):
 AppFeatureTag = type("AppFeatureTag", (BaseFixedTypeTag,), {"tag_type": "app-feature"})
 DeploymentFailureTag = type("DeploymentFailureTag", (BaseFixedTypeTag,), {"tag_type": "deploy-failure"})
 DeployPhaseTag = type("DeployPhaseTag", (BaseFixedTypeTag,), {"tag_type": "deploy_phase"})
+SaasMonitorTag = type("SaasMonitorTag", (BaseFixedTypeTag,), {"tag_type": "saas_monitor"})
 
 
 TAG_TYPES: List[Type[BaseFixedTypeTag]] = [
@@ -73,6 +74,7 @@ def __init__(self, tag_value):
     AppFeatureTag,
     DeploymentFailureTag,
     DeployPhaseTag,
+    SaasMonitorTag,
 ]
 TAG_TYPES_MAP = {t.tag_type: t for t in TAG_TYPES}
 

apiserver/paasng/paasng/misc/monitoring/monitor/alert_rules/ascode/client.py

@@ -25,6 +25,8 @@
 from django.conf import settings
 from django.utils.translation import gettext_lazy as _
 
+from paasng.accessories.smart_advisor.advisor import DocumentaryLinkAdvisor
+from paasng.accessories.smart_advisor.tags import get_dynamic_tag
 from paasng.infras.bkmonitorv3.client import make_bk_monitor_client
 from paasng.infras.bkmonitorv3.shim import get_or_create_bk_monitor_space
 from paasng.misc.monitoring.monitor.alert_rules.config import RuleConfig
@@ -73,6 +75,24 @@ def _validate(self, rule_configs: List[RuleConfig]):
                     f"app_code({self.app_code})"
                 )
 
+    def _update_metric_name_prefixes(self, ctx: dict, alert_code: str):
+        """根据告警代码添加需要的指标前缀到上下文中"""
+        alert_config_mapping = {
+            "rabbitmq": settings.RABBITMQ_MONITOR_CONF,
+            "bkrepo": settings.BKREPO_MONITOR_CONF,
+            "gcs_mysql": settings.GCS_MYSQL_MONITOR_CONF,
+        }
+        for code_snippet, config in alert_config_mapping.items():
+            if code_snippet in alert_code:
+                ctx[f"{code_snippet}_metric_name_prefix"] = config.get("metric_name_prefix", "")
+
+    def _update_docs_url(self, ctx: dict, alert_code: str):
+        """根据告警代码添加需要的文档信息"""
+        tag = get_dynamic_tag(f"saas_monitor:{alert_code}")
+        links = DocumentaryLinkAdvisor().search_by_tags([tag], limit=1)
+        if links:
+            ctx["doc_url"] = f"处理建议: {links[0].location}"
+
     def _render_rule_configs(self, rule_configs: List[RuleConfig]) -> Dict:
         """按照 MonitorAsCode 规则, 渲染出如下示例目录结构:
 
@@ -88,14 +108,10 @@ def _render_rule_configs(self, rule_configs: List[RuleConfig]) -> Dict:
         for conf in rule_configs:
             ctx = conf.to_dict()
             ctx["notice_group_name"] = self.default_notice_group_name
-            # 涉及到 rabbitmq 的告警策略, 指标是通过 bkmonitor 配置的采集器采集, 需要添加指标前缀
-            if "rabbitmq" in conf.alert_code:
-                ctx["rabbitmq_metric_name_prefix"] = settings.RABBITMQ_MONITOR_CONF.get("metric_name_prefix", "")
-            if "bkrepo" in conf.alert_code:
-                ctx["bkrepo_metric_name_prefix"] = settings.BKREPO_MONITOR_CONF.get("metric_name_prefix", "")
-            if "gcs_mysql" in conf.alert_code:
-                ctx["gcs_mysql_metric_name_prefix"] = settings.GCS_MYSQL_MONITOR_CONF.get("metric_name_prefix", "")
-
+            # 涉及到增强服务的告警策略, 指标是通过 bkmonitor 配置的采集器采集, 需要添加指标前缀
+            self._update_metric_name_prefixes(ctx, conf.alert_code)
+            # 为通知添加操作文档链接，若未在'管理后台--智能顾问--文档管理'配置文档则不添加
+            self._update_docs_url(ctx, conf.alert_code)
             configs[f"rule/{conf.alert_rule_name}.yaml"] = j2_env.get_template(f"{conf.alert_code}.yaml.j2").render(
                 **ctx
             )

apiserver/paasng/paasng/misc/monitoring/monitor/alert_rules/ascode/rules_tpl/gcs_mysql_slow_query.yaml.j2

@@ -25,3 +25,33 @@ notice: # 通知配置
   interval: 30 # 通知收敛间隔(分钟)
   user_groups: # 通知组配置
   - '{{ notice_group_name }}'
+  template:
+    abnormal:
+      content: '{% raw %}{{content.level}}
+
+        {{content.begin_time}}
+
+        {{content.time}}
+
+        {{content.duration}}
+
+        {{content.target_type}}
+
+        {{content.data_source}}
+
+        {{content.content}}{% endraw %} {{ doc_url }}
+
+        {% raw %}{{content.current_value}}
+
+        {{content.biz}}
+
+        {{content.target}}
+
+        {{content.dimension}}
+
+        {{content.detail}}
+
+        {{content.assign_detail}}
+
+        {{content.related_info}}'
+      title: '{{business.bk_biz_name}} - {{alarm.name}}{{alarm.display_type}}{% endraw %}'

apiserver/paasng/paasng/misc/monitoring/monitor/alert_rules/ascode/rules_tpl/high_bkrepo_quota_usage.yaml.j2

@@ -25,3 +25,33 @@ notice: # 通知配置
   interval: 30 # 通知收敛间隔(分钟)
   user_groups: # 通知组配置
   - '{{ notice_group_name }}'
+  template:
+    abnormal:
+      content: '{% raw %}{{content.level}}
+
+        {{content.begin_time}}
+
+        {{content.time}}
+
+        {{content.duration}}
+
+        {{content.target_type}}
+
+        {{content.data_source}}
+
+        {{content.content}}{% endraw %} {{ doc_url }}
+
+        {% raw %}{{content.current_value}}
+
+        {{content.biz}}
+
+        {{content.target}}
+
+        {{content.dimension}}
+
+        {{content.detail}}
+
+        {{content.assign_detail}}
+
+        {{content.related_info}}'
+      title: '{{business.bk_biz_name}} - {{alarm.name}}{{alarm.display_type}}{% endraw %}'

apiserver/paasng/paasng/misc/monitoring/monitor/alert_rules/ascode/rules_tpl/high_cpu_usage.yaml.j2

@@ -25,3 +25,33 @@ notice: # 通知配置
   interval: 120 # 通知收敛间隔(分钟)
   user_groups: # 通知组配置
   - '{{ notice_group_name }}'
+  template:
+    abnormal:
+      content: '{% raw %}{{content.level}}
+
+        {{content.begin_time}}
+
+        {{content.time}}
+
+        {{content.duration}}
+
+        {{content.target_type}}
+
+        {{content.data_source}}
+
+        {{content.content}}{% endraw %} {{ doc_url }}
+
+        {% raw %}{{content.current_value}}
+
+        {{content.biz}}
+
+        {{content.target}}
+
+        {{content.dimension}}
+
+        {{content.detail}}
+
+        {{content.assign_detail}}
+
+        {{content.related_info}}'
+      title: '{{business.bk_biz_name}} - {{alarm.name}}{{alarm.display_type}}{% endraw %}'

apiserver/paasng/paasng/misc/monitoring/monitor/alert_rules/ascode/rules_tpl/high_mem_usage.yaml.j2

@@ -25,3 +25,33 @@ notice: # 通知配置
   interval: 120 # 通知收敛间隔(分钟)
   user_groups: # 通知组配置
   - '{{ notice_group_name }}'
+  template:
+    abnormal:
+      content: '{% raw %}{{content.level}}
+
+        {{content.begin_time}}
+
+        {{content.time}}
+
+        {{content.duration}}
+
+        {{content.target_type}}
+
+        {{content.data_source}}
+
+        {{content.content}}{% endraw %} {{ doc_url }}
+
+        {% raw %}{{content.current_value}}
+
+        {{content.biz}}
+
+        {{content.target}}
+
+        {{content.dimension}}
+
+        {{content.detail}}
+
+        {{content.assign_detail}}
+
+        {{content.related_info}}'
+      title: '{{business.bk_biz_name}} - {{alarm.name}}{{alarm.display_type}}{% endraw %}'

apiserver/paasng/paasng/misc/monitoring/monitor/alert_rules/ascode/rules_tpl/high_rabbitmq_queue_messages.yaml.j2

@@ -25,3 +25,33 @@ notice: # 通知配置
   interval: 30 # 通知收敛间隔(分钟)
   user_groups: # 通知组配置
   - '{{ notice_group_name }}'
+  template:
+    abnormal:
+      content: '{% raw %}{{content.level}}
+
+        {{content.begin_time}}
+
+        {{content.time}}
+
+        {{content.duration}}
+
+        {{content.target_type}}
+
+        {{content.data_source}}
+
+        {{content.content}}{% endraw %} {{ doc_url }}
+
+        {% raw %}{{content.current_value}}
+
+        {{content.biz}}
+
+        {{content.target}}
+
+        {{content.dimension}}
+
+        {{content.detail}}
+
+        {{content.assign_detail}}
+
+        {{content.related_info}}'
+      title: '{{business.bk_biz_name}} - {{alarm.name}}{{alarm.display_type}}{% endraw %}'

apiserver/paasng/paasng/misc/monitoring/monitor/alert_rules/ascode/rules_tpl/oom_killed.yaml.j2

@@ -25,3 +25,33 @@ notice: # 通知配置
   interval: 120 # 通知收敛间隔(分钟)
   user_groups: # 通知组配置
   - '{{ notice_group_name }}'
+  template:
+    abnormal:
+      content: '{% raw %}{{content.level}}
+
+        {{content.begin_time}}
+
+        {{content.time}}
+
+        {{content.duration}}
+
+        {{content.target_type}}
+
+        {{content.data_source}}
+
+        {{content.content}}{% endraw %} {{ doc_url }}
+
+        {% raw %}{{content.current_value}}
+
+        {{content.biz}}
+
+        {{content.target}}
+
+        {{content.dimension}}
+
+        {{content.detail}}
+
+        {{content.assign_detail}}
+
+        {{content.related_info}}'
+      title: '{{business.bk_biz_name}} - {{alarm.name}}{{alarm.display_type}}{% endraw %}'

apiserver/paasng/paasng/misc/monitoring/monitor/alert_rules/ascode/rules_tpl/pod_restart.yaml.j2

@@ -25,3 +25,33 @@ notice: # 通知配置
   interval: 30 # 通知收敛间隔(分钟)
   user_groups: # 通知组配置
   - '{{ notice_group_name }}'
+  template:
+    abnormal:
+      content: '{% raw %}{{content.level}}
+
+        {{content.begin_time}}
+
+        {{content.time}}
+
+        {{content.duration}}
+
+        {{content.target_type}}
+
+        {{content.data_source}}
+
+        {{content.content}}{% endraw %} {{ doc_url }}
+
+        {% raw %}{{content.current_value}}
+
+        {{content.biz}}
+
+        {{content.target}}
+
+        {{content.dimension}}
+
+        {{content.detail}}
+
+        {{content.assign_detail}}
+
+        {{content.related_info}}'
+      title: '{{business.bk_biz_name}} - {{alarm.name}}{{alarm.display_type}}{% endraw %}'