Add epoch version & nodes count sensors in NodeBroker & DynamicNameserver (#20625)

pixcc · web-flow · commit 9adb3a04dfd5 · 2025-07-08T13:22:51.000+03:00
diff --git a/ydb/core/mind/dynamic_nameserver.cpp b/ydb/core/mind/dynamic_nameserver.cpp
@@ -1,6 +1,7 @@
 #include "dynamic_nameserver_impl.h"
 
 #include <ydb/core/base/appdata.h>
+#include <ydb/core/base/counters.h>
 #include <ydb/core/base/nameservice.h>
 #include <ydb/core/mon/mon.h>
 #include <ydb/library/services/services.pb.h>
@@ -91,6 +92,7 @@ class TActorCacheMiss : public TActor<TActorCacheMiss<TCacheMiss>>, public TCach
                 ResetInterconnectProxyConfig(NodeId, ctx);
                 ListNodesCache->Invalidate(); // node was erased
             }
+            Owner->UpdateCounters();
             OnError(rec.GetStatus().GetReason(), ctx);
             return;
         }
@@ -105,6 +107,7 @@ class TActorCacheMiss : public TActor<TActorCacheMiss<TCacheMiss>>, public TCach
             ResetInterconnectProxyConfig(NodeId, ctx);
         Config->DynamicNodes.emplace(NodeId, node);
 
+        Owner->UpdateCounters();
         OnSuccess(ctx);
     }
 
@@ -240,6 +243,13 @@ void TDynamicNameserver::Bootstrap(const TActorContext &ctx)
         OpenPipe(domain->DomainUid);
     }
 
+    auto counters = GetServiceCounters(AppData(ctx)->Counters, "utils")->GetSubgroup("component", "nameserver");
+    ActiveDynamicNodesCounter = counters->GetCounter("ActiveDynamicNodes");
+    ExpiredDynamicNodesCounter = counters->GetCounter("ExpireDynamicNodes");
+    StaticNodesCounter = counters->GetCounter("StaticNodes");
+    EpochVersionCounter = counters->GetCounter("EpochVersion");
+    UpdateCounters();
+
     Send(MakeBlobStorageNodeWardenID(SelfId().NodeId()), new TEvNodeWardenQueryStorageConfig(true));
 
     Become(&TDynamicNameserver::StateFunc);
@@ -282,6 +292,7 @@ void TDynamicNameserver::ReplaceNameserverSetup(TIntrusivePtr<TTableNameserverSe
         for (const auto& subscriber : StaticNodeChangeSubscribers) {
             TActivationContext::Send(new IEventHandle(SelfId(), subscriber, new TEvInterconnect::TEvListNodes));
         }
+        UpdateCounters();
     }
 }
 
@@ -518,6 +529,8 @@ void TDynamicNameserver::UpdateState(const NKikimrNodeBroker::TNodesInfo &rec,
         }
         config->Epoch = rec.GetEpoch();
     }
+
+    UpdateCounters();
 }
 
 void TDynamicNameserver::OnPipeDestroyed(ui32 domain, const TActorContext &ctx)
@@ -542,6 +555,18 @@ void TDynamicNameserver::OnPipeDestroyed(ui32 domain, const TActorContext &ctx)
     OpenPipe(DynamicConfigs[domain]->NodeBrokerPipe);
 }
 
+void TDynamicNameserver::UpdateCounters() {
+    auto info = AppData()->DomainsInfo;
+    if (info->Domain) {
+        const auto &config = DynamicConfigs[info->Domain->DomainUid];
+
+        *EpochVersionCounter = config->Epoch.Version;
+        *ActiveDynamicNodesCounter = config->DynamicNodes.size();
+        *ExpiredDynamicNodesCounter = config->ExpiredNodes.size();
+        *StaticNodesCounter = StaticConfig->StaticNodeTable.size();
+    }
+}
+
 void TDynamicNameserver::Handle(TEvInterconnect::TEvResolveNode::TPtr &ev,
                                 const TActorContext &ctx)
 {
@@ -805,6 +830,8 @@ void TDynamicNameserver::Handle(TEvNodeBroker::TEvUpdateNodes::TPtr &ev, const T
                 break;
         }
     }
+
+    UpdateCounters();
 }
 
 void TDynamicNameserver::Handle(TEvNodeBroker::TEvSyncNodesResponse::TPtr &ev, const TActorContext &ctx)
diff --git a/ydb/core/mind/dynamic_nameserver_impl.h b/ydb/core/mind/dynamic_nameserver_impl.h
@@ -261,6 +261,8 @@ class TDynamicNameserver : public TActorBootstrapped<TDynamicNameserver> {
     void OnPipeDestroyed(ui32 domain,
                          const TActorContext &ctx);
 
+    void UpdateCounters();
+
     void Handle(TEvInterconnect::TEvResolveNode::TPtr &ev, const TActorContext &ctx);
     void Handle(TEvResolveAddress::TPtr &ev, const TActorContext &ctx);
     void Handle(TEvInterconnect::TEvListNodes::TPtr &ev, const TActorContext &ctx);
@@ -305,6 +307,11 @@ class TDynamicNameserver : public TActorBootstrapped<TDynamicNameserver> {
     bool SyncInProgress = false;
     ui64 SyncCookie = 0;
     ui64 SeqNo = 0;
+
+    ::NMonitoring::TDynamicCounters::TCounterPtr StaticNodesCounter;
+    ::NMonitoring::TDynamicCounters::TCounterPtr ActiveDynamicNodesCounter;
+    ::NMonitoring::TDynamicCounters::TCounterPtr ExpiredDynamicNodesCounter;
+    ::NMonitoring::TDynamicCounters::TCounterPtr EpochVersionCounter;
 };
 
 } // NNodeBroker
diff --git a/ydb/core/mind/dynamic_nameserver_mon.cpp b/ydb/core/mind/dynamic_nameserver_mon.cpp
@@ -21,6 +21,19 @@ TString ToString(TInstant t)
     return t.FormatLocalTime("%d %b %Y %H:%M:%S %Z");
 }
 
+TString ToString(EProtocolState s)
+{
+    switch (s) {
+        case EProtocolState::Connecting:
+            return "Connecting";
+        case EProtocolState::UseEpochProtocol:
+            return "EpochProtocol";
+        case EProtocolState::UseDeltaProtocol:
+            return "DeltaProtocol";
+    }
+    return "Unknown";
+}
+
 void OutputStaticContent(IOutputStream &str)
 {
     str << R"__(
@@ -194,6 +207,10 @@ void TDynamicNameserver::Handle(NMon::TEvHttpInfo::TPtr &ev, const TActorContext
                 << "      <td class='right-align'>Max dynamic node ID:</td>" << Endl
                 << "      <td>" << config->MaxDynamicNodeId << "</td>" << Endl
                 << "    </tr>" << Endl
+                << "    <tr>" << Endl
+                << "      <td class='right-align'>Protocol state:</td>" << Endl
+                << "      <td>" << ToString(ProtocolState) << "</td>" << Endl
+                << "    </tr>" << Endl
                 << "  </tbody>" << Endl
                 << "</table></div>" << Endl;
 
diff --git a/ydb/core/mind/node_broker.cpp b/ydb/core/mind/node_broker.cpp
@@ -786,6 +786,13 @@ bool TNodeBroker::HasOutdatedSubscription(TActorId subscriber, ui64 newSeqNo) co
     return false;
 }
 
+void TNodeBroker::UpdateCommittedStateCounters() {
+    TabletCounters->Simple()[COUNTER_ACTIVE_NODES].Set(Committed.Nodes.size());
+    TabletCounters->Simple()[COUNTER_EXPIRED_NODES].Set(Committed.ExpiredNodes.size());
+    TabletCounters->Simple()[COUNTER_REMOVED_NODES].Set(Committed.RemovedNodes.size());
+    TabletCounters->Simple()[COUNTER_EPOCH_VERSION].Set(Committed.Epoch.Version);
+}
+
 void TNodeBroker::TState::LoadConfigFromProto(const NKikimrNodeBroker::TConfig &config)
 {
     Config = config;
diff --git a/ydb/core/mind/node_broker__extend_lease.cpp b/ydb/core/mind/node_broker__extend_lease.cpp
@@ -90,6 +90,8 @@ class TNodeBroker::TTxExtendLease : public TTransactionBase<TNodeBroker> {
             Self->AddNodeToUpdateNodesLog(node);
             Self->ScheduleProcessSubscribersQueue(ctx);
         }
+
+        Self->UpdateCommittedStateCounters();
     }
 
 private:
diff --git a/ydb/core/mind/node_broker__graceful_shutdown.cpp b/ydb/core/mind/node_broker__graceful_shutdown.cpp
@@ -52,6 +52,8 @@ class TNodeBroker::TTxGracefulShutdown : public TTransactionBase<TNodeBroker> {
             Self->Committed.ReleaseSlotIndex(Self->Committed.Nodes.at(Event->Get()->Record.GetNodeId()));
         }
         ctx.Send(Event->Sender, Response.Release());
+
+        Self->UpdateCommittedStateCounters();
     }
 
 private:
diff --git a/ydb/core/mind/node_broker__migrate_state.cpp b/ydb/core/mind/node_broker__migrate_state.cpp
@@ -106,6 +106,8 @@ class TNodeBroker::TTxMigrateState : public TTransactionBase<TNodeBroker> {
             NKikimrNodeBroker::TVersionInfo versionInfo;
             versionInfo.SetSupportDeltaProtocol(true);
             Self->SignalTabletActive(ctx, versionInfo.SerializeAsString());
+
+            Self->UpdateCommittedStateCounters();
         } else {
             Self->Execute(Self->CreateTxMigrateState(std::move(DbChanges)));
         }
diff --git a/ydb/core/mind/node_broker__register_node.cpp b/ydb/core/mind/node_broker__register_node.cpp
@@ -235,6 +235,8 @@ class TNodeBroker::TTxRegisterNode : public TTransactionBase<TNodeBroker> {
         }
 
         Reply(ctx);
+
+        Self->UpdateCommittedStateCounters();
     }
 
 private:
diff --git a/ydb/core/mind/node_broker__update_config.cpp b/ydb/core/mind/node_broker__update_config.cpp
@@ -92,6 +92,8 @@ class TNodeBroker::TTxUpdateConfig : public TTransactionBase<TNodeBroker> {
                         "TTxUpdateConfig reply with: " << Response->ToString());
             ctx.Send(Response);
         }
+
+        Self->UpdateCommittedStateCounters();
     }
 
 private:
diff --git a/ydb/core/mind/node_broker__update_config_subscription.cpp b/ydb/core/mind/node_broker__update_config_subscription.cpp
@@ -41,6 +41,8 @@ class TNodeBroker::TTxUpdateConfigSubscription : public TTransactionBase<TNodeBr
                     "Using new subscription id=" << SubscriptionId);
 
         Self->Committed.ConfigSubscriptionId = SubscriptionId;
+
+        Self->UpdateCommittedStateCounters();
     }
 
 private:
diff --git a/ydb/core/mind/node_broker__update_epoch.cpp b/ydb/core/mind/node_broker__update_epoch.cpp
@@ -36,6 +36,8 @@ class TNodeBroker::TTxUpdateEpoch : public TTransactionBase<TNodeBroker> {
         Self->PrepareUpdateNodesLog();
         Self->ProcessDelayedListNodesRequests();
         Self->ScheduleProcessSubscribersQueue(ctx);
+
+        Self->UpdateCommittedStateCounters();
     }
 
 private:
diff --git a/ydb/core/mind/node_broker_impl.h b/ydb/core/mind/node_broker_impl.h
@@ -305,6 +305,8 @@ class TNodeBroker : public TActor<TNodeBroker>
     void RemoveSubscriber(TActorId subscriber, const TActorContext &ctx);
     bool HasOutdatedSubscription(TActorId subscriber, ui64 newSeqNo) const;
 
+    void UpdateCommittedStateCounters();
+
     void Handle(TEvConsole::TEvConfigNotificationRequest::TPtr &ev,
                 const TActorContext &ctx);
     void Handle(TEvConsole::TEvReplaceConfigSubscriptionsResponse::TPtr &ev,
diff --git a/ydb/core/protos/counters_node_broker.proto b/ydb/core/protos/counters_node_broker.proto
@@ -10,6 +10,10 @@ enum ESimpleCounters {
     COUNTER_EPOCH_SIZE_BYTES = 0            [(CounterOpts) = {Name: "EpochSizeBytes"}];
     COUNTER_EPOCH_DELTAS_SIZE_BYTES = 1     [(CounterOpts) = {Name: "EpochDeltasSizeBytes"}];
     COUNTER_UPDATE_NODES_LOG_SIZE_BYTES = 2 [(CounterOpts) = {Name: "UpdateNodesLogSizeBytes"}];
+    COUNTER_EPOCH_VERSION = 3               [(CounterOpts) = {Name: "EpochVersion"}];
+    COUNTER_ACTIVE_NODES = 4                [(CounterOpts) = {Name: "ActiveNodes"}];
+    COUNTER_EXPIRED_NODES = 5               [(CounterOpts) = {Name: "ExpiredNodes"}];
+    COUNTER_REMOVED_NODES = 6               [(CounterOpts) = {Name: "RemovedNodes"}];
 }
 
 enum ECumulativeCounters {

Original file line number	Diff line number	Diff line change
`@@ -1,6 +1,7 @@`
`1`	`1`	`#include "dynamic_nameserver_impl.h"`
`2`	`2`
`3`	`3`	`#include <ydb/core/base/appdata.h>`
	`4`	`+#include <ydb/core/base/counters.h>`
`4`	`5`	`#include <ydb/core/base/nameservice.h>`
`5`	`6`	`#include <ydb/core/mon/mon.h>`
`6`	`7`	`#include <ydb/library/services/services.pb.h>`
`@@ -91,6 +92,7 @@ class TActorCacheMiss : public TActor<TActorCacheMiss<TCacheMiss>>, public TCach`
`91`	`92`	`ResetInterconnectProxyConfig(NodeId, ctx);`
`92`	`93`	`ListNodesCache->Invalidate(); // node was erased`
`93`	`94`	`}`
	`95`	`+ Owner->UpdateCounters();`
`94`	`96`	`OnError(rec.GetStatus().GetReason(), ctx);`
`95`	`97`	`return;`
`96`	`98`	`}`
`@@ -105,6 +107,7 @@ class TActorCacheMiss : public TActor<TActorCacheMiss<TCacheMiss>>, public TCach`
`105`	`107`	`ResetInterconnectProxyConfig(NodeId, ctx);`
`106`	`108`	`Config->DynamicNodes.emplace(NodeId, node);`
`107`	`109`
	`110`	`+ Owner->UpdateCounters();`
`108`	`111`	`OnSuccess(ctx);`
`109`	`112`	`}`
`110`	`113`
`@@ -240,6 +243,13 @@ void TDynamicNameserver::Bootstrap(const TActorContext &ctx)`
`240`	`243`	`OpenPipe(domain->DomainUid);`
`241`	`244`	`}`
`242`	`245`
	`246`	`+ auto counters = GetServiceCounters(AppData(ctx)->Counters, "utils")->GetSubgroup("component", "nameserver");`
	`247`	`+ ActiveDynamicNodesCounter = counters->GetCounter("ActiveDynamicNodes");`
	`248`	`+ ExpiredDynamicNodesCounter = counters->GetCounter("ExpireDynamicNodes");`
	`249`	`+ StaticNodesCounter = counters->GetCounter("StaticNodes");`
	`250`	`+ EpochVersionCounter = counters->GetCounter("EpochVersion");`
	`251`	`+ UpdateCounters();`
	`252`	`+`
`243`	`253`	`Send(MakeBlobStorageNodeWardenID(SelfId().NodeId()), new TEvNodeWardenQueryStorageConfig(true));`
`244`	`254`
`245`	`255`	`Become(&TDynamicNameserver::StateFunc);`
`@@ -282,6 +292,7 @@ void TDynamicNameserver::ReplaceNameserverSetup(TIntrusivePtr<TTableNameserverSe`
`282`	`292`	`for (const auto& subscriber : StaticNodeChangeSubscribers) {`
`283`	`293`	`TActivationContext::Send(new IEventHandle(SelfId(), subscriber, new TEvInterconnect::TEvListNodes));`
`284`	`294`	`}`
	`295`	`+ UpdateCounters();`
`285`	`296`	`}`
`286`	`297`	`}`
`287`	`298`
`@@ -518,6 +529,8 @@ void TDynamicNameserver::UpdateState(const NKikimrNodeBroker::TNodesInfo &rec,`
`518`	`529`	`}`
`519`	`530`	`config->Epoch = rec.GetEpoch();`
`520`	`531`	`}`
	`532`	`+`
	`533`	`+ UpdateCounters();`
`521`	`534`	`}`
`522`	`535`
`523`	`536`	`void TDynamicNameserver::OnPipeDestroyed(ui32 domain, const TActorContext &ctx)`
`@@ -542,6 +555,18 @@ void TDynamicNameserver::OnPipeDestroyed(ui32 domain, const TActorContext &ctx)`
`542`	`555`	`OpenPipe(DynamicConfigs[domain]->NodeBrokerPipe);`
`543`	`556`	`}`
`544`	`557`
	`558`	`+void TDynamicNameserver::UpdateCounters() {`
	`559`	`+ auto info = AppData()->DomainsInfo;`
	`560`	`+ if (info->Domain) {`
	`561`	`+ const auto &config = DynamicConfigs[info->Domain->DomainUid];`
	`562`	`+`
	`563`	`+ *EpochVersionCounter = config->Epoch.Version;`
	`564`	`+ *ActiveDynamicNodesCounter = config->DynamicNodes.size();`
	`565`	`+ *ExpiredDynamicNodesCounter = config->ExpiredNodes.size();`
	`566`	`+ *StaticNodesCounter = StaticConfig->StaticNodeTable.size();`
	`567`	`+ }`
	`568`	`+}`
	`569`	`+`
`545`	`570`	`void TDynamicNameserver::Handle(TEvInterconnect::TEvResolveNode::TPtr &ev,`
`546`	`571`	`const TActorContext &ctx)`
`547`	`572`	`{`
`@@ -805,6 +830,8 @@ void TDynamicNameserver::Handle(TEvNodeBroker::TEvUpdateNodes::TPtr &ev, const T`
`805`	`830`	`break;`
`806`	`831`	`}`
`807`	`832`	`}`
	`833`	`+`
	`834`	`+ UpdateCounters();`
`808`	`835`	`}`
`809`	`836`
`810`	`837`	`void TDynamicNameserver::Handle(TEvNodeBroker::TEvSyncNodesResponse::TPtr &ev, const TActorContext &ctx)`
Original file line number	Diff line number	Diff line change
`@@ -786,6 +786,13 @@ bool TNodeBroker::HasOutdatedSubscription(TActorId subscriber, ui64 newSeqNo) co`
`786`	`786`	`return false;`
`787`	`787`	`}`
`788`	`788`
	`789`	`+void TNodeBroker::UpdateCommittedStateCounters() {`
	`790`	`+ TabletCounters->Simple()[COUNTER_ACTIVE_NODES].Set(Committed.Nodes.size());`
	`791`	`+ TabletCounters->Simple()[COUNTER_EXPIRED_NODES].Set(Committed.ExpiredNodes.size());`
	`792`	`+ TabletCounters->Simple()[COUNTER_REMOVED_NODES].Set(Committed.RemovedNodes.size());`
	`793`	`+ TabletCounters->Simple()[COUNTER_EPOCH_VERSION].Set(Committed.Epoch.Version);`
	`794`	`+}`
	`795`	`+`
`789`	`796`	`void TNodeBroker::TState::LoadConfigFromProto(const NKikimrNodeBroker::TConfig &config)`
`790`	`797`	`{`
`791`	`798`	`Config = config;`
Original file line number	Diff line number	Diff line change
`@@ -90,6 +90,8 @@ class TNodeBroker::TTxExtendLease : public TTransactionBase<TNodeBroker> {`
`90`	`90`	`Self->AddNodeToUpdateNodesLog(node);`
`91`	`91`	`Self->ScheduleProcessSubscribersQueue(ctx);`
`92`	`92`	`}`
	`93`	`+`
	`94`	`+ Self->UpdateCommittedStateCounters();`
`93`	`95`	`}`
`94`	`96`
`95`	`97`	`private:`
Original file line number	Diff line number	Diff line change
`@@ -52,6 +52,8 @@ class TNodeBroker::TTxGracefulShutdown : public TTransactionBase<TNodeBroker> {`
`52`	`52`	`Self->Committed.ReleaseSlotIndex(Self->Committed.Nodes.at(Event->Get()->Record.GetNodeId()));`
`53`	`53`	`}`
`54`	`54`	`ctx.Send(Event->Sender, Response.Release());`
	`55`	`+`
	`56`	`+ Self->UpdateCommittedStateCounters();`
`55`	`57`	`}`
`56`	`58`
`57`	`59`	`private:`
Original file line number	Diff line number	Diff line change
`@@ -106,6 +106,8 @@ class TNodeBroker::TTxMigrateState : public TTransactionBase<TNodeBroker> {`
`106`	`106`	`NKikimrNodeBroker::TVersionInfo versionInfo;`
`107`	`107`	`versionInfo.SetSupportDeltaProtocol(true);`
`108`	`108`	`Self->SignalTabletActive(ctx, versionInfo.SerializeAsString());`
	`109`	`+`
	`110`	`+ Self->UpdateCommittedStateCounters();`
`109`	`111`	`} else {`
`110`	`112`	`Self->Execute(Self->CreateTxMigrateState(std::move(DbChanges)));`
`111`	`113`	`}`
Original file line number	Diff line number	Diff line change
`@@ -235,6 +235,8 @@ class TNodeBroker::TTxRegisterNode : public TTransactionBase<TNodeBroker> {`
`235`	`235`	`}`
`236`	`236`
`237`	`237`	`Reply(ctx);`
	`238`	`+`
	`239`	`+ Self->UpdateCommittedStateCounters();`
`238`	`240`	`}`
`239`	`241`
`240`	`242`	`private:`
Original file line number	Diff line number	Diff line change
`@@ -92,6 +92,8 @@ class TNodeBroker::TTxUpdateConfig : public TTransactionBase<TNodeBroker> {`
`92`	`92`	`"TTxUpdateConfig reply with: " << Response->ToString());`
`93`	`93`	`ctx.Send(Response);`
`94`	`94`	`}`
	`95`	`+`
	`96`	`+ Self->UpdateCommittedStateCounters();`
`95`	`97`	`}`
`96`	`98`
`97`	`99`	`private:`
Original file line number	Diff line number	Diff line change
`@@ -41,6 +41,8 @@ class TNodeBroker::TTxUpdateConfigSubscription : public TTransactionBase<TNodeBr`
`41`	`41`	`"Using new subscription id=" << SubscriptionId);`
`42`	`42`
`43`	`43`	`Self->Committed.ConfigSubscriptionId = SubscriptionId;`
	`44`	`+`
	`45`	`+ Self->UpdateCommittedStateCounters();`
`44`	`46`	`}`
`45`	`47`
`46`	`48`	`private:`
Original file line number	Diff line number	Diff line change
`@@ -36,6 +36,8 @@ class TNodeBroker::TTxUpdateEpoch : public TTransactionBase<TNodeBroker> {`
`36`	`36`	`Self->PrepareUpdateNodesLog();`
`37`	`37`	`Self->ProcessDelayedListNodesRequests();`
`38`	`38`	`Self->ScheduleProcessSubscribersQueue(ctx);`
	`39`	`+`
	`40`	`+ Self->UpdateCommittedStateCounters();`
`39`	`41`	`}`
`40`	`42`
`41`	`43`	`private:`