fix potential data race when report resource group metric (#8235)

close #8233

dbms/src/Common/TiFlashMetrics.h

@@ -28,6 +28,7 @@
 
 #include <ext/scope_guard.h>
 #include <mutex>
+#include <shared_mutex>
 
 // to make GCC 11 happy
 #include <cassert>
@@ -863,11 +864,17 @@ struct MetricFamily
     T & get(size_t idx = 0) { return *(metrics[idx]); }
     T & get(size_t idx, const String & resource_group_name)
     {
-        if (metrics_map.find(resource_group_name) == metrics_map.end())
         {
-            addMetricsForResourceGroup(resource_group_name);
+            std::shared_lock lock(resource_group_metrics_mu);
+            if (resource_group_metrics_map.find(resource_group_name) != resource_group_metrics_map.end())
+                return *(resource_group_metrics_map[resource_group_name][idx]);
         }
-        return *(metrics_map[resource_group_name][idx]);
+
+        std::lock_guard lock(resource_group_metrics_mu);
+        if (resource_group_metrics_map.find(resource_group_name) == resource_group_metrics_map.end())
+            addMetricsForResourceGroup(resource_group_name);
+
+        return *(resource_group_metrics_map[resource_group_name][idx]);
     }
 
 private:
@@ -886,14 +893,15 @@ struct MetricFamily
             auto & metric = MetricTrait::add(*store_family, resource_group_name, MetricArgType{});
             metrics_temp.emplace_back(&metric);
         }
-        metrics_map[resource_group_name] = metrics_temp;
+        resource_group_metrics_map[resource_group_name] = metrics_temp;
     }
 
     std::vector<T *> metrics;
     prometheus::Family<T> * store_family;
     std::vector<MetricArgType> store_args;
     // <resource_group_name, metrics>
-    std::unordered_map<String, std::vector<T *>> metrics_map;
+    std::shared_mutex resource_group_metrics_mu;
+    std::unordered_map<String, std::vector<T *>> resource_group_metrics_map;
 };
 
 /// Centralized registry of TiFlash metrics.