return up-to-date node stats KIKIMR-20697 (#910)

Author: vporyadke

ydb/core/mind/hive/hive_impl.cpp

@@ -1782,6 +1782,8 @@ bool THive::IsTabletMoveExpedient(const TTabletInfo& tablet, const TNodeInfo& no
 
 void THive::FillTabletInfo(NKikimrHive::TEvResponseHiveInfo& response, ui64 tabletId, const TLeaderTabletInfo *info, const NKikimrHive::TEvRequestHiveInfo &req) {
     if (info) {
+        TInstant now = TActivationContext::Now();
+        TInstant restartsBarrierTime = now - GetTabletRestartsPeriod();
         auto& tabletInfo = *response.AddTablets();
         tabletInfo.SetTabletID(tabletId);
         tabletInfo.SetTabletType(info->Type);
@@ -1800,7 +1802,7 @@ void THive::FillTabletInfo(NKikimrHive::TEvResponseHiveInfo& response, ui64 tabl
         if (!info->IsRunning()) {
             tabletInfo.SetLastAliveTimestamp(info->Statistics.GetLastAliveTimestamp());
         }
-        tabletInfo.SetRestartsPerPeriod(info->Statistics.RestartTimestampSize());
+        tabletInfo.SetRestartsPerPeriod(info->GetRestartsPerPeriod(restartsBarrierTime));
         if (req.GetReturnMetrics()) {
             tabletInfo.MutableMetrics()->CopyFrom(info->GetResourceValues());
         }
@@ -1831,7 +1833,7 @@ void THive::FillTabletInfo(NKikimrHive::TEvResponseHiveInfo& response, ui64 tabl
                 if (!follower.IsRunning()) {
                     tabletInfo.SetLastAliveTimestamp(follower.Statistics.GetLastAliveTimestamp());
                 }
-                tabletInfo.SetRestartsPerPeriod(follower.Statistics.RestartTimestampSize());
+                tabletInfo.SetRestartsPerPeriod(follower.GetRestartsPerPeriod(restartsBarrierTime));
                 if (req.GetReturnMetrics()) {
                     tabletInfo.MutableMetrics()->CopyFrom(follower.GetResourceValues());
                 }
@@ -1843,16 +1845,14 @@ void THive::FillTabletInfo(NKikimrHive::TEvResponseHiveInfo& response, ui64 tabl
 void THive::Handle(TEvHive::TEvRequestHiveInfo::TPtr& ev) {
     const auto& record = ev->Get()->Record;
     TAutoPtr<TEvHive::TEvResponseHiveInfo> response = new TEvHive::TEvResponseHiveInfo();
-    TInstant now = TlsActivationContext->Now();
     if (record.HasTabletID()) {
         TTabletId tabletId = record.GetTabletID();
         NKikimrHive::TForwardRequest forwardRequest;
         if (CheckForForwardTabletRequest(tabletId, forwardRequest)) {
             response->Record.MutableForwardRequest()->CopyFrom(forwardRequest);
         }
-        TLeaderTabletInfo* tablet = FindTablet(tabletId);
+        const TLeaderTabletInfo* tablet = FindTablet(tabletId);
         if (tablet) {
-            tablet->ActualizeTabletStatistics(now);
             FillTabletInfo(response->Record, record.GetTabletID(), tablet, record);
         } else {
             BLOG_W("Can't find the tablet from RequestHiveInfo(TabletID=" << tabletId << ")");
@@ -1866,7 +1866,6 @@ void THive::Handle(TEvHive::TEvRequestHiveInfo::TPtr& ev) {
             if (it->second.IsDeleting()) {
                 continue;
             }
-            it->second.ActualizeTabletStatistics(now);
             FillTabletInfo(response->Record, it->first, &it->second, record);
         }
         response->Record.set_starttimetimestamp(StartTime().MilliSeconds());
@@ -1955,13 +1954,15 @@ void THive::Handle(TEvHive::TEvRequestHiveDomainStats::TPtr& ev) {
 
 void THive::Handle(TEvHive::TEvRequestHiveNodeStats::TPtr& ev) {
     const auto& request(ev->Get()->Record);
+    TInstant now = TActivationContext::Now();
+    TInstant restartsBarrierTime = now - GetNodeRestartWatchPeriod();
     THolder<TEvHive::TEvResponseHiveNodeStats> response = MakeHolder<TEvHive::TEvResponseHiveNodeStats>();
     auto& record = response->Record;
     if (request.GetReturnExtendedTabletInfo()) {
         record.SetExtendedTabletInfo(true);
     }
     for (auto it = Nodes.begin(); it != Nodes.end(); ++it) {
-        const TNodeInfo& node = it->second;
+        TNodeInfo& node = it->second;
         if (node.IsUnknown()) {
             continue;
         }
@@ -2035,7 +2036,7 @@ void THive::Handle(TEvHive::TEvRequestHiveNodeStats::TPtr& ev) {
         if (!node.IsAlive()) {
             nodeStats.SetLastAliveTimestamp(node.Statistics.GetLastAliveTimestamp());
         }
-        nodeStats.SetRestartsPerPeriod(node.Statistics.RestartTimestampSize());
+        nodeStats.SetRestartsPerPeriod(node.GetRestartsPerPeriod(restartsBarrierTime));
     }
     Send(ev->Sender, response.Release(), 0, ev->Cookie);
 }
@@ -3299,6 +3300,11 @@ void THive::ActualizeRestartStatistics(google::protobuf::RepeatedField<google::p
     array.erase(begin, it);
 }
 
+ui64 THive::GetRestartsPerPeriod(const google::protobuf::RepeatedField<google::protobuf::uint64>& restartTimestamps, ui64 barrier) {
+    auto it = std::lower_bound(restartTimestamps.begin(), restartTimestamps.end(), barrier);
+    return restartTimestamps.end() - it;
+}
+
 bool THive::IsSystemTablet(TTabletTypes::EType type) {
     switch (type) {
         case TTabletTypes::Coordinator:

ydb/core/mind/hive/hive_impl.h

@@ -665,7 +665,7 @@ class THive : public TActor<THive>, public TTabletExecutedFlat, public THiveShar
             const NKikimrTabletBase::TMetrics& after,
             NKikimr::NHive::TResourceRawValues deltaRaw,
             NKikimr::NHive::TResourceNormalizedValues deltaNormalized);
-    static void FillTabletInfo(NKikimrHive::TEvResponseHiveInfo& response, ui64 tabletId, const TLeaderTabletInfo* info, const NKikimrHive::TEvRequestHiveInfo& req);
+    void FillTabletInfo(NKikimrHive::TEvResponseHiveInfo& response, ui64 tabletId, const TLeaderTabletInfo* info, const NKikimrHive::TEvRequestHiveInfo& req);
     void ExecuteStartTablet(TFullTabletId tabletId, const TActorId& local, ui64 cookie, bool external);
     ui32 GetDataCenters();
     ui32 GetRegisteredDataCenters();
@@ -914,6 +914,7 @@ class THive : public TActor<THive>, public TTabletExecutedFlat, public THiveShar
     }
 
     static void ActualizeRestartStatistics(google::protobuf::RepeatedField<google::protobuf::uint64>& restartTimestamps, ui64 barrier);
+    static ui64 GetRestartsPerPeriod(const google::protobuf::RepeatedField<google::protobuf::uint64>& restartTimestamps, ui64 barrier);
     static bool IsSystemTablet(TTabletTypes::EType type);
 
 protected:

ydb/core/mind/hive/node_info.cpp

@@ -471,6 +471,10 @@ void TNodeInfo::ActualizeNodeStatistics(TInstant now) {
     Hive.ActualizeRestartStatistics(*Statistics.MutableRestartTimestamp(), barierTime.MilliSeconds());
 }
 
+ui64 TNodeInfo::GetRestartsPerPeriod(TInstant barrier) const {
+    return Hive.GetRestartsPerPeriod(Statistics.GetRestartTimestamp(), barrier.MilliSeconds());
+}
+
 TString TNodeInfo::GetLogPrefix() const {
     return Hive.GetLogPrefix();
 }

ydb/core/mind/hive/node_info.h

@@ -235,6 +235,7 @@ struct TNodeInfo {
 
     void UpdateResourceTotalUsage(const NKikimrHive::TEvTabletMetrics& metrics);
     void ActualizeNodeStatistics(TInstant now);
+    ui64 GetRestartsPerPeriod(TInstant barrier) const;
 
     TDataCenterId GetDataCenter() const {
         return Location.GetDataCenterId();

ydb/core/mind/hive/tablet_info.cpp

@@ -496,13 +496,8 @@ void TTabletInfo::ActualizeTabletStatistics(TInstant now) {
     Hive.ActualizeRestartStatistics(*Statistics.MutableRestartTimestamp(), barierTime.MilliSeconds());
 }
 
-ui64 TTabletInfo::GetRestartsPerPeriod(TInstant barrier) {
-    const auto& array(Statistics.GetRestartTimestamp());
-    ui64 restarts = 0;
-    for (auto itRestart = array.rbegin(); (itRestart != array.rend()) && (TInstant::MilliSeconds(*itRestart) >= barrier); ++itRestart) {
-        ++restarts;
-    }
-    return restarts;
+ui64 TTabletInfo::GetRestartsPerPeriod(TInstant barrier) const {
+    return Hive.GetRestartsPerPeriod(Statistics.GetRestartTimestamp(), barrier.MilliSeconds());
 }
 
 bool TTabletInfo::RestartsOften() const {

ydb/core/mind/hive/tablet_info.h

@@ -294,7 +294,7 @@ struct TTabletInfo {
     }
 
     void ActualizeTabletStatistics(TInstant now);
-    ui64 GetRestartsPerPeriod(TInstant barrier);
+    ui64 GetRestartsPerPeriod(TInstant barrier) const;
     bool RestartsOften() const;
 
     bool HasCounter() {