feat: Metadata自监控--指标化改造，后台任务耗时统计&接入链路趋势统计指标 --story=120586640 (#3766)

TencentBlueKing · Nov 7, 2024 · 95638c5 · 95638c5
1 parent b52e3b6
commit 95638c5
Show file tree

Hide file tree

Showing 4 changed files with 155 additions and 141 deletions.
diff --git a/bkmonitor/core/prometheus/metrics.py b/bkmonitor/core/prometheus/metrics.py
@@ -1158,12 +1158,13 @@ def from_exc(cls, expr):
     name="bkmonitor_metadata_cron_task_cost_seconds",
     documentation="监控元数据定时任务耗时统计",
     labelnames=("task_name", "process_target"),
+    buckets=(0, 1, 5, 10, 30, 60, 120, 180, 240, 300, 600, 900, 1800, 3000, 6000, INF),
 )
 
-METADATA_DATA_LINK_ACCESS_INFO = Gauge(
-    name="bkmonitor_metadata_data_link_access_info",
+METADATA_DATA_LINK_ACCESS_TOTAL = Counter(
+    name="bkmonitor_metadata_data_link_access_total",
     documentation="监控元数据数据链路接入统计",
-    labelnames=("version", "biz_id", "data_id", 'table_id', 'strategy'),
+    labelnames=("version", "biz_id", 'strategy', 'status'),
 )
 
 

diff --git a/bkmonitor/metadata/models/vm/utils.py b/bkmonitor/metadata/models/vm/utils.py
@@ -122,7 +122,6 @@ def access_bkdata(bk_biz_id: int, table_id: str, data_id: int):
             version=DATA_LINK_V3_VERSION_NAME,
             data_id=data_id,
             biz_id=bk_biz_id,
-            table_id=table_id,
             status=ACCESS_DATA_LINK_SUCCESS_STATUS,
             strategy=DataLink.BK_STANDARD_V2_TIME_SERIES,
         )
@@ -133,7 +132,6 @@ def access_bkdata(bk_biz_id: int, table_id: str, data_id: int):
             version=DATA_LINK_V3_VERSION_NAME,
             data_id=data_id,
             biz_id=bk_biz_id,
-            table_id=table_id,
             status=ACCESS_DATA_LINK_FAILURE_STATUS,
             strategy=DataLink.BK_STANDARD_V2_TIME_SERIES,
         )
@@ -314,7 +312,6 @@ def report_metadata_data_link_access_metric(
     status: int,
     biz_id: int,
     data_id: int,
-    table_id: str,
     strategy: str,
 ) -> None:
     """
@@ -323,18 +320,13 @@ def report_metadata_data_link_access_metric(
     @param status: 接入状态（失败-1/成功1） 以是否成功向bkbase发起请求为准
     @param biz_id: 业务ID
     @param data_id: 数据ID
-    @param table_id: 结果表ID
     @param strategy: 链路策略（套餐类型）
     """
     try:
         logger.info("try to report metadata data link component status metric,data_id->[%s]", data_id)
-        metrics.METADATA_DATA_LINK_ACCESS_INFO.labels(
-            version=version,
-            biz_id=biz_id,
-            data_id=data_id,
-            table_id=table_id,
-            strategy=strategy,
-        ).set(status)
+        metrics.METADATA_DATA_LINK_ACCESS_TOTAL.labels(
+            version=version, biz_id=biz_id, strategy=strategy, status=status
+        ).inc()
         metrics.report_all()
     except Exception as err:  # pylint: disable=broad-except
         logger.error("report metadata data link access metric error->[%s],data_id->[%s]", err, data_id)
@@ -553,7 +545,6 @@ def access_v2_bkdata_vm(bk_biz_id: int, table_id: str, data_id: int):
             version=DATA_LINK_V4_VERSION_NAME,
             data_id=data_id,
             biz_id=bk_biz_id,
-            table_id=table_id,
             status=ACCESS_DATA_LINK_SUCCESS_STATUS,
             strategy=DataLink.BK_STANDARD_V2_TIME_SERIES,
         )
@@ -563,7 +554,6 @@ def access_v2_bkdata_vm(bk_biz_id: int, table_id: str, data_id: int):
             version=DATA_LINK_V4_VERSION_NAME,
             data_id=data_id,
             biz_id=bk_biz_id,
-            table_id=table_id,
             status=ACCESS_DATA_LINK_FAILURE_STATUS,
             strategy=DataLink.BCS_FEDERAL_SUBSET_TIME_SERIES,
         )
@@ -581,7 +571,6 @@ def access_v2_bkdata_vm(bk_biz_id: int, table_id: str, data_id: int):
             version=DATA_LINK_V4_VERSION_NAME,
             data_id=data_id,
             biz_id=bk_biz_id,
-            table_id=table_id,
             status=ACCESS_DATA_LINK_SUCCESS_STATUS,
             strategy=DataLink.BCS_FEDERAL_SUBSET_TIME_SERIES,
         )
@@ -591,7 +580,6 @@ def access_v2_bkdata_vm(bk_biz_id: int, table_id: str, data_id: int):
             version=DATA_LINK_V4_VERSION_NAME,
             data_id=data_id,
             biz_id=bk_biz_id,
-            table_id=table_id,
             status=ACCESS_DATA_LINK_FAILURE_STATUS,
             strategy=DataLink.BCS_FEDERAL_SUBSET_TIME_SERIES,
         )

diff --git a/bkmonitor/metadata/task/refresh_data_link.py b/bkmonitor/metadata/task/refresh_data_link.py
@@ -33,4 +33,4 @@ def refresh_data_link_status():
 
     bkbase_rt_records = models.BkBaseResultTable.objects.filter(monitor_table_id__in=table_id_list)
     logger.info("refresh_data_link_status: now try to bulk_refresh_data_link_status,len->[%s] ", len(bkbase_rt_records))
-    bulk_refresh_data_link_status.delay(bkbase_rt_records)
+    bulk_refresh_data_link_status.delay(bkbase_rt_records)  # task_id