add network throughput aggregation (#14364)

Author: adameat

ydb/core/protos/node_whiteboard.proto

@@ -345,6 +345,7 @@ message TSystemStateInfo {
     optional double CoresUsed = 39;
     optional uint32 CoresTotal = 40;
     optional float NetworkUtilization = 41;
+    optional uint64 NetworkWriteThroughput = 42;
 }
 
 message TEvSystemStateRequest {

ydb/core/tablet/node_whiteboard.cpp

@@ -74,6 +74,7 @@ class TNodeWhiteboardService : public TActorBootstrapped<TNodeWhiteboardService>
     i64 MaxClockSkewWithPeerUs;
     ui32 MaxClockSkewPeerId;
     float MaxNetworkUtilization = 0.0;
+    ui64 SumNetworkWriteThroughput = 0;
     NKikimrWhiteboard::TSystemStateInfo SystemStateInfo;
     THolder<NTracing::ITraceCollection> TabletIntrospectionData;
 
@@ -568,6 +569,7 @@ class TNodeWhiteboardService : public TActorBootstrapped<TNodeWhiteboardService>
         }
         // TODO: need better way to calculate network utilization
         MaxNetworkUtilization = std::max(MaxNetworkUtilization, ev->Get()->Record.GetUtilization());
+        SumNetworkWriteThroughput += nodeStateInfo.GetWriteThroughput();
         nodeStateInfo.MergeFrom(ev->Get()->Record);
         nodeStateInfo.SetChangeTime(currentChangeTime);
     }
@@ -1095,6 +1097,7 @@ class TNodeWhiteboardService : public TActorBootstrapped<TNodeWhiteboardService>
     }
 
     void Handle(TEvPrivate::TEvUpdateRuntimeStats::TPtr &, const TActorContext &ctx) {
+        static constexpr TDuration UPDATE_PERIOD = TDuration::Seconds(15);
         {
             NKikimrWhiteboard::TSystemStateInfo systemStatsUpdate;
             TVector<double> loadAverage = GetLoadAverage();
@@ -1120,8 +1123,13 @@ class TNodeWhiteboardService : public TActorBootstrapped<TNodeWhiteboardService>
             MaxNetworkUtilization = 0;
         }
 
+        {
+            SystemStateInfo.SetNetworkWriteThroughput(SumNetworkWriteThroughput / UPDATE_PERIOD.Seconds());
+            SumNetworkWriteThroughput = 0;
+        }
+
         UpdateSystemState(ctx);
-        ctx.Schedule(TDuration::Seconds(15), new TEvPrivate::TEvUpdateRuntimeStats());
+        ctx.Schedule(UPDATE_PERIOD, new TEvPrivate::TEvUpdateRuntimeStats());
     }
 
     void Handle(TEvPrivate::TEvCleanupDeadTablets::TPtr &, const TActorContext &ctx) {

ydb/core/viewer/protos/viewer.proto

@@ -344,6 +344,7 @@ message TClusterInfo {
     uint64 Hosts = 60;
     uint64 Tenants = 61;
     double NetworkUtilization = 62;
+    uint64 NetworkWriteThroughput = 63;
 }
 
 enum ETenantType {
@@ -400,6 +401,7 @@ message TTenant {
     uint32 CoresTotal = 50;
     optional NKikimrMemory.TMemoryStats MemoryStats = 51;
     float NetworkUtilization = 52;
+    uint64 NetworkWriteThroughput = 53;
 }
 
 message TTenants {