Split IC peer name labels (#12401)

Author: alexvru

ydb/library/actors/interconnect/interconnect_counters.cpp

@@ -261,8 +261,10 @@ namespace {
             UpdateUtilization(PrevStarvation, Starvation, starvation);
         }
 
-        void SetPeerInfo(const TString& name, const TString& dataCenterId) override {
-            if (name != std::exchange(HumanFriendlyPeerHostName, name)) {
+        void SetPeerInfo(ui32 nodeId, const TString& name, const TString& dataCenterId) override {
+            if (nodeId != PeerNodeId || name != HumanFriendlyPeerHostName) {
+                PeerNodeId = nodeId;
+                HumanFriendlyPeerHostName = name;
                 PerSessionCounters.Reset();
             }
             VALGRIND_MAKE_READABLE(&DataCenterId, sizeof(DataCenterId));
@@ -278,7 +280,9 @@ namespace {
             const bool updatePerSession = !PerSessionCounters || updatePerDataCenter;
             if (HasSessionCounters && updatePerSession) {
                 auto base = MergePerDataCenterCounters ? PerDataCenterCounters : Counters;
-                PerSessionCounters = base->GetSubgroup("peer", *HumanFriendlyPeerHostName);
+                PerSessionCounters = base
+                    ->GetSubgroup("peer_node_id", ToString(*PeerNodeId))
+                    ->GetSubgroup("peer_name", *HumanFriendlyPeerHostName);
             }
 
             const bool updateGlobal = !Initialized;
@@ -601,8 +605,10 @@ namespace {
             UpdateUtilization(PrevStarvation_, Starvation_, starvation);
         }
 
-        void SetPeerInfo(const TString& name, const TString& dataCenterId) override {
-            if (name != std::exchange(HumanFriendlyPeerHostName, name)) {
+        void SetPeerInfo(ui32 nodeId, const TString& name, const TString& dataCenterId) override {
+            if (nodeId != PeerNodeId || name != HumanFriendlyPeerHostName) {
+                PeerNodeId = nodeId;
+                HumanFriendlyPeerHostName = name;
                 PerSessionMetrics_.reset();
             }
             VALGRIND_MAKE_READABLE(&DataCenterId, sizeof(DataCenterId));
@@ -620,7 +626,10 @@ namespace {
             if (updatePerSession) {
                 auto base = MergePerDataCenterMetrics_ ? PerDataCenterMetrics_ : Metrics_;
                 PerSessionMetrics_ = std::make_shared<NMonitoring::TMetricSubRegistry>(
-                        NMonitoring::TLabels{{"peer", *HumanFriendlyPeerHostName}}, base);
+                    NMonitoring::TLabels{
+                        {"peer_node_id", ToString(*PeerNodeId)},
+                        {"peer_name", *HumanFriendlyPeerHostName},
+                    }, base);
             }
 
             const bool updateGlobal = !Initialized_;

ydb/library/actors/interconnect/interconnect_counters.h

@@ -35,7 +35,7 @@ class IInterconnectMetrics {
     virtual void IncDisconnectByReason(const TString& s) = 0;
     virtual void IncUsefulReadWakeups() = 0;
     virtual void IncSpuriousReadWakeups() = 0;
-    virtual void SetPeerInfo(const TString& name, const TString& dataCenterId) = 0;
+    virtual void SetPeerInfo(ui32 nodeId, const TString& name, const TString& dataCenterId) = 0;
     virtual void AddInputChannelsIncomingTraffic(ui16 channel, ui64 incomingTraffic) = 0;
     virtual void IncInputChannelsIncomingEvents(ui16 channel) = 0;
     virtual void IncRecvSyscalls(ui64 ns) = 0;
@@ -50,6 +50,7 @@ class IInterconnectMetrics {
 
 protected:
     std::optional<TString> DataCenterId;
+    std::optional<ui32> PeerNodeId;
     std::optional<TString> HumanFriendlyPeerHostName;
 };
 

ydb/library/actors/interconnect/interconnect_tcp_proxy.cpp

@@ -9,10 +9,13 @@
 namespace NActors {
     static constexpr TDuration GetNodeRequestTimeout = TDuration::Seconds(5);
 
-    static TString PeerNameForHuman(ui32 nodeNum, const TString& longName, ui16 port) {
+    static TString PeerNameForHuman(const TString& longName, ui16 port) {
         TStringBuf token;
         TStringBuf(longName).NextTok('.', token);
-        return ToString<ui32>(nodeNum) + ":" + (token.size() > 0 ? TString(token) : longName) + ":" + ToString<ui16>(port);
+        return TStringBuilder()
+            << (token.size() > 0 ? TString(token) : longName)
+            << ':'
+            << port;
     }
 
     TInterconnectProxyTCP::TInterconnectProxyTCP(const ui32 node, TInterconnectProxyCommon::TPtr common,
@@ -106,12 +109,12 @@ namespace NActors {
             TransitToErrorState("cannot get node info");
         } else {
             auto& info = *ev->Get()->Node;
-            TString name = PeerNameForHuman(PeerNodeId, info.Host, info.Port);
+            TString name = PeerNameForHuman(info.Host, info.Port);
             TechnicalPeerHostName = info.Host;
             if (!Metrics) {
                 Metrics = Common->Metrics ? CreateInterconnectMetrics(Common) : CreateInterconnectCounters(Common);
             }
-            Metrics->SetPeerInfo(name, info.Location.GetDataCenterId());
+            Metrics->SetPeerInfo(PeerNodeId, name, info.Location.GetDataCenterId());
 
             LOG_DEBUG_IC("ICP02", "configured for host %s", name.data());
 

ydb/library/actors/interconnect/ut/channel_scheduler_ut.cpp

@@ -10,7 +10,7 @@ Y_UNIT_TEST_SUITE(ChannelScheduler) {
         auto common = MakeIntrusive<TInterconnectProxyCommon>();
         common->MonCounters = MakeIntrusive<NMonitoring::TDynamicCounters>();
         std::shared_ptr<IInterconnectMetrics> ctr = CreateInterconnectCounters(common);
-        ctr->SetPeerInfo("peer", "1");
+        ctr->SetPeerInfo(1, "peer", "1");
         auto callback = [](THolder<IEventBase>) {};
         TEventHolderPool pool(common, callback);
         TSessionParams p;