add network diagnostics columns to viewer/nodes handler (ydb-platform…

…#11344)
GrigoriyPA · Nov 7, 2024 · 804f3d6 · 804f3d6
1 parent 6887352
commit 804f3d6
Show file tree

Hide file tree

Showing 6 changed files with 789 additions and 80 deletions.
diff --git a/ydb/core/protos/node_whiteboard.proto b/ydb/core/protos/node_whiteboard.proto
@@ -105,7 +105,7 @@ message TNodeStateInfo {
     optional uint64 ConnectTime = 7; // changed every time the connection changes it state
     optional uint32 PeerNodeId = 8 [(DefaultField) = true]; // node if of the peer
     optional int64 ClockSkewUs = 9; // a positive value means the peer is ahead in time; a negative value means it's behind
-    optional uint32 PingTimeUs = 10; // RTT for the peer
+    optional uint64 PingTimeUs = 10; // RTT for the peer
     optional float Utilization = 11; // network utilization 0-1
     optional NActorsInterconnect.TScopeId ScopeId = 12; // scope id of the peer
     optional uint32 Count = 13; // filled during group count

diff --git a/ydb/core/viewer/json_handlers_viewer.cpp b/ydb/core/viewer/json_handlers_viewer.cpp
@@ -244,7 +244,7 @@ void InitViewerHealthCheckJsonHandler(TJsonHandlers& handlers) {
 }
 
 void InitViewerNodesJsonHandler(TJsonHandlers& handlers) {
-    handlers.AddHandler("/viewer/nodes", new TJsonHandler<TJsonNodes>(TJsonNodes::GetSwagger()), 12);
+    handlers.AddHandler("/viewer/nodes", new TJsonHandler<TJsonNodes>(TJsonNodes::GetSwagger()), 13);
 }
 
 void InitViewerACLJsonHandler(TJsonHandlers &jsonHandlers) {

diff --git a/ydb/core/viewer/json_pipe_req.cpp b/ydb/core/viewer/json_pipe_req.cpp
@@ -254,6 +254,15 @@ TViewerPipeClient::TRequestResponse<TEvViewer::TEvViewerResponse> TViewerPipeCli
             case NKikimrViewer::TEvViewerRequest::kSystemRequest:
                 response.Span.Attribute("request_type", "SystemRequest");
                 break;
+            case NKikimrViewer::TEvViewerRequest::kPDiskRequest:
+                response.Span.Attribute("request_type", "PDiskRequest");
+                break;
+            case NKikimrViewer::TEvViewerRequest::kVDiskRequest:
+                response.Span.Attribute("request_type", "VDiskRequest");
+                break;
+            case NKikimrViewer::TEvViewerRequest::kNodeRequest:
+                response.Span.Attribute("request_type", "NodeRequest");
+                break;
             case NKikimrViewer::TEvViewerRequest::kQueryRequest:
                 response.Span.Attribute("request_type", "QueryRequest");
                 break;

diff --git a/ydb/core/viewer/protos/viewer.proto b/ydb/core/viewer/protos/viewer.proto
@@ -532,10 +532,27 @@ message TNodeInfo {
     bool Disconnected = 4;
     float CpuUsage = 5;
     float DiskSpaceUsage = 6;
-    NKikimrWhiteboard.TSystemStateInfo SystemState = 10;
-    repeated NKikimrWhiteboard.TPDiskStateInfo PDisks = 20;
-    repeated NKikimrWhiteboard.TVDiskStateInfo VDisks = 30;
-    repeated TTabletStateInfo Tablets = 40;
+    optional uint32 Connections = 7; // total number of live connections
+    EFlag ConnectStatus = 10; // Max
+    optional uint64 ReceiveThroughput = 11;
+    optional uint64 SendThroughput = 12;
+    optional float NetworkUtilization = 20; // Sum
+    optional float NetworkUtilizationMin = 21;
+    optional float NetworkUtilizationMax = 22;
+    optional int64 ClockSkewUs = 30; // Avg
+    optional int64 ClockSkewMinUs = 31;
+    optional int64 ClockSkewMaxUs = 32;
+    optional int64 ReverseClockSkewUs = 33; // Avg
+    optional uint64 PingTimeUs = 40; // Avg
+    optional uint64 PingTimeMinUs = 41;
+    optional uint64 PingTimeMaxUs = 42;
+    optional uint64 ReversePingTimeUs = 43; // Avg
+    NKikimrWhiteboard.TSystemStateInfo SystemState = 50;
+    repeated NKikimrWhiteboard.TPDiskStateInfo PDisks = 51;
+    repeated NKikimrWhiteboard.TVDiskStateInfo VDisks = 52;
+    repeated TTabletStateInfo Tablets = 53;
+    repeated NKikimrWhiteboard.TNodeStateInfo Peers = 54;
+    repeated NKikimrWhiteboard.TNodeStateInfo ReversePeers = 55;
 }
 
 message TNodesInfo {
@@ -668,13 +685,14 @@ message TEvViewerRequest {
         NKikimrWhiteboard.TEvVDiskStateRequest VDiskRequest = 16;
         NKikimrWhiteboard.TEvPDiskStateRequest PDiskRequest = 17;
         NKikimrWhiteboard.TEvBSGroupStateRequest BSGroupRequest = 18;
+        NKikimrWhiteboard.TEvNodeStateRequest NodeRequest = 19;
         THttpProxyRequest QueryRequest = 13;
         THttpProxyRequest RenderRequest = 14;
         TSchemeCacheRequest AutocompleteRequest = 15;
-        bytes Reserved19 = 19;
         bytes Reserved20 = 20;
         bytes Reserved21 = 21;
         bytes Reserved22 = 22;
+        bytes Reserved23 = 23;
     }
 }
 
@@ -686,13 +704,14 @@ message TEvViewerResponse {
         NKikimrWhiteboard.TEvVDiskStateResponse VDiskResponse = 16;
         NKikimrWhiteboard.TEvPDiskStateResponse PDiskResponse = 17;
         NKikimrWhiteboard.TEvBSGroupStateResponse BSGroupResponse = 18;
+        NKikimrWhiteboard.TEvNodeStateResponse NodeResponse = 19;
         NKikimrKqp.TEvQueryResponse QueryResponse = 13;
         NKikimrGraph.TEvMetricsResult RenderResponse = 14;
         TQueryAutocomplete AutocompleteResponse = 15;
-        bytes Reserved19 = 19;
         bytes Reserved20 = 20;
         bytes Reserved21 = 21;
         bytes Reserved22 = 22;
+        bytes Reserved23 = 23;
     }
 }
 
@@ -789,3 +808,4 @@ message TFeatureFlagsConfig {
     uint32 Version = 1;
     repeated TDatabase Databases = 2;
 }
+