add a histogram sensor for node usages KIKIMR-18705

3 files changed, 38 insertions, 10 deletions

diff --git a/ydb/core/mind/hive/hive_impl.cpp b/ydb/core/mind/hive/hive_impl.cpp
index 1fe5feaca1..98f3bada80 100644
--- a/ydb/core/mind/hive/hive_impl.cpp
+++ b/ydb/core/mind/hive/hive_impl.cpp
@@ -2077,23 +2077,17 @@ void THive::ProcessTabletBalancer() {
 }
 
 THive::THiveStats THive::GetStats() const {
-    struct TNodeStat {
-        TNodeId NodeId;
-        double Usage;
-    };
-
     THiveStats stats = {};
-    TVector<TNodeStat> values;
-    values.reserve(Nodes.size());
+    stats.Values.reserve(Nodes.size());
     for (const auto& ni : Nodes) {
         if (ni.second.IsAlive() && !ni.second.Down) {
-            values.push_back({ni.first, ni.second.GetNodeUsage()});
+            stats.Values.push_back({ni.first, ni.second.GetNodeUsage()});
         }
     }
-    if (values.empty()) {
+    if (stats.Values.empty()) {
         return stats;
     }
-    auto it = std::minmax_element(values.begin(), values.end(), [](const TNodeStat& a, const TNodeStat& b) -> bool {
+    auto it = std::minmax_element(stats.Values.begin(), stats.Values.end(), [](const THiveStats::TNodeStat& a, const THiveStats::TNodeStat& b) -> bool {
         return a.Usage < b.Usage;
     });
     stats.MaxUsage = it.second->Usage;
@@ -2105,6 +2099,8 @@ THive::THiveStats THive::GetStats() const {
         double minUsage = std::max(stats.MinUsage, minUsageToBalance);
         double maxUsage = std::max(stats.MaxUsage, minUsageToBalance);
         stats.Scatter = (maxUsage - minUsage) / maxUsage;
+    } else {
+        stats.Scatter = 0;
     }
     return stats;
 }
@@ -2137,6 +2133,12 @@ void THive::Handle(TEvPrivate::TEvProcessTabletBalancer::TPtr&) {
     TabletCounters->Simple()[NHive::COUNTER_BALANCE_USAGE_MIN].Set(stats.MinUsage * 100);
     TabletCounters->Simple()[NHive::COUNTER_BALANCE_USAGE_MAX].Set(stats.MaxUsage * 100);
 
+    auto& nodeUsageHistogram = TabletCounters->Percentile()[NHive::COUNTER_NODE_USAGE];
+    nodeUsageHistogram.Clear();
+    for (const auto& record : stats.Values) {
+        nodeUsageHistogram.IncrementFor(record.Usage * 100);
+    }
+
     if (stats.MaxUsage >= GetMaxNodeUsageToKick()) {
         std::vector<TNodeId> overloadedNodes;
         for (const auto& [nodeId, nodeInfo] : Nodes) {
diff --git a/ydb/core/mind/hive/hive_impl.h b/ydb/core/mind/hive/hive_impl.h
index 41ac9a7c05..a26b4b9422 100644
--- a/ydb/core/mind/hive/hive_impl.h
+++ b/ydb/core/mind/hive/hive_impl.h
@@ -846,11 +846,17 @@ protected:
     double GetUsage() const;
 
     struct THiveStats {
+        struct TNodeStat {
+            TNodeId NodeId;
+            double Usage;
+        };
+
         double MinUsage;
         TNodeId MinUsageNodeId;
         double MaxUsage;
         TNodeId MaxUsageNodeId;
         double Scatter;
+        std::vector<TNodeStat> Values;
     };
 
     THiveStats GetStats() const;
diff --git a/ydb/core/protos/counters_hive.proto b/ydb/core/protos/counters_hive.proto
index e62a1c02e7..8c628efe19 100644
--- a/ydb/core/protos/counters_hive.proto
+++ b/ydb/core/protos/counters_hive.proto
@@ -49,6 +49,26 @@ enum EPercentileCounters {
     };
 
     COUNTER_PERCENTILE_IGNORE = 0;
+
+    COUNTER_NODE_USAGE = 1 [(CounterOpts) = {
+        Name: "NodeUsage",
+        Integral: true,
+        Ranges: { Value: 30    Name: "30%"       },
+        Ranges: { Value: 35    Name: "35%"       },
+        Ranges: { Value: 40    Name: "40%"       },
+        Ranges: { Value: 45    Name: "45%"       },
+        Ranges: { Value: 50    Name: "50%"       },
+        Ranges: { Value: 55    Name: "55%"       },
+        Ranges: { Value: 60    Name: "60%"       },
+        Ranges: { Value: 65    Name: "65%"       },
+        Ranges: { Value: 70    Name: "70%"       },
+        Ranges: { Value: 75    Name: "75%"       },
+        Ranges: { Value: 80    Name: "80%"       },
+        Ranges: { Value: 85    Name: "85%"       },
+        Ranges: { Value: 90    Name: "90%"       },
+        Ranges: { Value: 95    Name: "95%"       },
+        Ranges: { Value: 100   Name: "100%"      },
+    }];
 }
 
 enum ETxTypes {