add network throughput aggregation

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 {

ydb/core/viewer/tests/canondata/result.json

@@ -1435,6 +1435,7 @@
         "MapVersions": "not-empty-object",
         "MemoryTotal": "not-zero-number-text",
         "NetworkUtilization": "number",
+        "NetworkWriteThroughput": "text",
         "NodesAlive": 3,
         "NodesTotal": 3,
         "NumberOfCpus": "not-zero-number",
@@ -4129,6 +4130,7 @@
                 "Metrics": {},
                 "Name": "/Root",
                 "NetworkUtilization": "number",
+                "NetworkWriteThroughput": "text",
                 "NodeIds": [
                     1
                 ],
@@ -4175,6 +4177,7 @@
                 "Metrics": "not-empty-object",
                 "Name": "/Root/dedicated_db",
                 "NetworkUtilization": "number",
+                "NetworkWriteThroughput": "text",
                 "NodeIds": [
                     50000
                 ],
@@ -4261,6 +4264,7 @@
                 "Metrics": "not-empty-object",
                 "Name": "/Root/shared_db",
                 "NetworkUtilization": "number",
+                "NetworkWriteThroughput": "text",
                 "NodeIds": [
                     50001
                 ],
@@ -4327,6 +4331,7 @@
                     "Metrics": {},
                     "Name": "/Root",
                     "NetworkUtilization": "number",
+                    "NetworkWriteThroughput": "text",
                     "NodeIds": [
                         1
                     ],
@@ -4377,6 +4382,7 @@
                     "Metrics": "not-empty-object",
                     "Name": "/Root/dedicated_db",
                     "NetworkUtilization": "number",
+                    "NetworkWriteThroughput": "text",
                     "NodeIds": [
                         50000
                     ],
@@ -4470,6 +4476,7 @@
                     "Metrics": "not-empty-object",
                     "Name": "/Root/shared_db",
                     "NetworkUtilization": "number",
+                    "NetworkWriteThroughput": "text",
                     "NodeIds": [
                         50001
                     ],

ydb/core/viewer/tests/test.py

@@ -326,6 +326,7 @@ def normalize_result_nodes(result):
                                            'NetworkUtilization',
                                            'NetworkUtilizationMin',
                                            'NetworkUtilizationMax',
+                                           'NetworkWriteThroughput',
                                            'PingTimeUs',
                                            'PingTimeMinUs',
                                            'PingTimeMaxUs',

ydb/core/viewer/viewer_cluster.h

@@ -347,6 +347,7 @@ class TJsonCluster : public TViewerPipeClient {
         request->AddFieldsRequired(NKikimrWhiteboard::TSystemStateInfo::kCoresUsedFieldNumber);
         request->AddFieldsRequired(NKikimrWhiteboard::TSystemStateInfo::kCoresTotalFieldNumber);
         request->AddFieldsRequired(NKikimrWhiteboard::TSystemStateInfo::kNetworkUtilizationFieldNumber);
+        request->AddFieldsRequired(NKikimrWhiteboard::TSystemStateInfo::kNetworkWriteThroughputFieldNumber);
     }
 
     void InitTabletWhiteboardRequest(NKikimrWhiteboard::TEvTabletStateRequest* request) {
@@ -544,6 +545,9 @@ class TJsonCluster : public TViewerPipeClient {
                 ClusterInfo.SetNetworkUtilization(ClusterInfo.GetNetworkUtilization() + systemState.GetNetworkUtilization());
                 ++nodesWithNetworkUtilization;
             }
+            if (systemState.HasNetworkWriteThroughput()) {
+                ClusterInfo.SetNetworkWriteThroughput(ClusterInfo.GetNetworkWriteThroughput() + systemState.GetNetworkWriteThroughput());
+            }
         }
 
         if (nodesWithNetworkUtilization != 0) {

ydb/core/viewer/viewer_tenantinfo.h

@@ -280,6 +280,7 @@ class TJsonTenantInfo : public TViewerPipeClient {
             request.AddFieldsRequired(NKikimrWhiteboard::TSystemStateInfo::kMemoryStatsFieldNumber);
         }
         request.AddFieldsRequired(NKikimrWhiteboard::TSystemStateInfo::kNetworkUtilizationFieldNumber);
+        request.AddFieldsRequired(NKikimrWhiteboard::TSystemStateInfo::kNetworkWriteThroughputFieldNumber);
     }
 
     void SendWhiteboardSystemStateRequest(const TNodeId nodeId) {
@@ -850,6 +851,9 @@ class TJsonTenantInfo : public TViewerPipeClient {
                             tenant.SetNetworkUtilization(tenant.GetNetworkUtilization() + nodeInfo.GetNetworkUtilization());
                             ++nodesWithNetworkUtilization;
                         }
+                        if (nodeInfo.HasNetworkWriteThroughput()) {
+                            tenant.SetNetworkWriteThroughput(tenant.GetNetworkWriteThroughput() + nodeInfo.GetNetworkWriteThroughput());
+                        }
                         overall = Max(overall, GetViewerFlag(nodeInfo.GetSystemState()));
                     }
                     tenantNodes.emplace(nodeId);