tablet metrics - less proto operations + batching

ydb/core/mind/hive/balancer.cpp

@@ -325,9 +325,9 @@ class THiveBalancer : public NActors::TActorBootstrapped<THiveBalancer>, public
                     tablet->ActorsToNotifyOnRestart.emplace_back(SelfId()); // volatile settings, will not persist upon restart
                     ++KickInFlight;
                     ++Movements;
-                    BLOG_D("Balancer moving tablet " << tablet->ToString() << " " << tablet->GetResourceValues()
-                           << " from node " << tablet->Node->Id << " " << tablet->Node->ResourceValues
-                           << " to node " << node->Id << " " << node->ResourceValues);
+                    BLOG_D("Balancer moving tablet " << tablet->ToString()
+                           << " from node " << tablet->Node->Id
+                           << " to node " << node->Id);
                     Hive->RecordTabletMove(THive::TTabletMoveInfo(now, *tablet, tablet->Node->Id, node->Id));
                     Hive->Execute(Hive->CreateRestartTablet(tablet->GetFullTabletId(), node->Id));
                     UpdateProgress();

ydb/core/mind/hive/drain.cpp

@@ -85,17 +85,17 @@ class THiveDrain : public NActors::TActorBootstrapped<THiveDrain>, public ISubAc
                     ++KickInFlight;
                     ++Movements;
                     BLOG_D("Drain " << SelfId() << " moving tablet "
-                                << tablet->ToString() << " " << tablet->GetResourceValues()
-                                << " from node " << tablet->Node->Id << " " << tablet->Node->ResourceValues
-                                << " to node " << node->Id << " " << node->ResourceValues);
+                                << tablet->ToString()
+                                << " from node " << tablet->Node->Id
+                                << " to node " << node->Id);
                     Hive->TabletCounters->Cumulative()[NHive::COUNTER_DRAIN_EXECUTED].Increment(1);
                     Hive->RecordTabletMove(THive::TTabletMoveInfo(TInstant::Now(), *tablet, tablet->Node->Id, node->Id));
                     Hive->Execute(Hive->CreateRestartTablet(tabletId, node->Id));
                 } else {
                     if (std::holds_alternative<THive::TNoNodeFound>(result) || std::holds_alternative<THive::TNotEnoughResources>(result)) {
                         Hive->TabletCounters->Cumulative()[NHive::COUNTER_DRAIN_FAILED].Increment(1);
-                        BLOG_D("Drain " << SelfId() << " could not move tablet " << tablet->ToString() << " " << tablet->GetResourceValues()
-                               << " from node " << tablet->Node->Id << " " << tablet->Node->ResourceValues);
+                        BLOG_D("Drain " << SelfId() << " could not move tablet " << tablet->ToString()
+                               << " from node " << tablet->Node->Id);
                     } else if (std::holds_alternative<THive::TTooManyTabletsStarting>(result)){
                         BLOG_D("Drain " << SelfId() << " could not move tablet " << tablet->ToString() << " and will try again later");
                         Hive->WaitToMoveTablets(SelfId());

ydb/core/mind/hive/fill.cpp

@@ -61,9 +61,9 @@ class THiveFill : public NActors::TActorBootstrapped<THiveFill>, public ISubActo
                         tablet->ActorsToNotifyOnRestart.emplace_back(SelfId()); // volatile settings, will not persist upon restart
                         ++KickInFlight;
                         ++Movements;
-                        BLOG_D("Fill " << SelfId() << " moving tablet " << tablet->ToString() << " " << tablet->GetResourceValues()
-                               << " from node " << tablet->Node->Id << " " << tablet->Node->ResourceValues
-                               << " to node " << node->Id << " " << node->ResourceValues);
+                        BLOG_D("Fill " << SelfId() << " moving tablet " << tablet->ToString()
+                               << " from node " << tablet->Node->Id
+                               << " to node " << node->Id);
                         Hive->TabletCounters->Cumulative()[NHive::COUNTER_FILL_EXECUTED].Increment(1);
                         Hive->RecordTabletMove(THive::TTabletMoveInfo(TInstant::Now(), *tablet, tablet->Node->Id, node->Id));
                         Hive->Execute(Hive->CreateRestartTablet(tablet->GetFullTabletId(), node->Id), ctx);

ydb/core/mind/hive/hive.cpp

@@ -119,6 +119,53 @@ bool TNodeFilter::IsAllowedPile(TBridgePileId pile) const {
     }
 }
 
+TMetrics& TMetrics::operator+=(const TMetrics& other) {
+    CPU += other.CPU;
+    Memory += other.Memory;
+    Network += other.Network;
+    Counter += other.Counter;
+    Storage += other.Storage;
+    ReadThroughput += other.ReadThroughput;
+    WriteThroughput += other.WriteThroughput;
+    ReadIops += other.ReadIops;
+    WriteIops += other.WriteIops;
+    return *this;
+}
+
+void TMetrics::ToProto(NKikimrTabletBase::TMetrics* proto) const {
+    if (CPU) {
+        proto->SetCPU(CPU);
+    }
+    if (Memory) {
+        proto->SetMemory(Memory);
+    }
+    if (Network) {
+        proto->SetNetwork(Network);
+    }
+    if (Counter) {
+        proto->SetCounter(Counter);
+    }
+    if (Storage) {
+        proto->SetStorage(Storage);
+    }
+    if (ReadThroughput) {
+        proto->SetReadThroughput(ReadThroughput);
+    }
+    if (WriteThroughput) {
+        proto->SetWriteThroughput(WriteThroughput);
+    }
+    if (ReadIops) {
+        proto->SetReadIops(ReadIops);
+    }
+    if (WriteIops) {
+        proto->SetWriteIops(WriteIops);
+    }
+    proto->MutableGroupReadThroughput()->Assign(GroupReadThroughput.begin(), GroupReadThroughput.end());
+    proto->MutableGroupWriteThroughput()->Assign(GroupWriteThroughput.begin(), GroupWriteThroughput.end());
+    proto->MutableGroupReadIops()->Assign(GroupReadIops.begin(), GroupReadIops.end());
+    proto->MutableGroupWriteIops()->Assign(GroupWriteIops.begin(), GroupWriteIops.end());
+}
+
 template <typename K, typename V>
 std::unordered_map<V, K> MakeReverseMap(const std::unordered_map<K, V>& map) {
     std::unordered_map<V, K> result;

ydb/core/mind/hive/hive.h

@@ -381,6 +381,27 @@ struct TFollowerUpdates {
     }
 };
 
+// same as NKikimrTabletBase::TMetrics, except not a protobuf - for lighter operations
+struct TMetrics {
+    ui64 CPU = 0;
+    ui64 Memory = 0;
+    ui64 Network = 0;
+    ui64 Counter = 0;
+    ui64 Storage = 0;
+    TVector<NKikimrTabletBase::TThroughputRecord> GroupReadThroughput;
+    TVector<NKikimrTabletBase::TThroughputRecord> GroupWriteThroughput;
+    ui64 ReadThroughput = 0;
+    ui64 WriteThroughput = 0;
+    TVector<NKikimrTabletBase::TIopsRecord> GroupReadIops;
+    TVector<NKikimrTabletBase::TIopsRecord> GroupWriteIops;
+    ui64 ReadIops = 0;
+    ui64 WriteIops = 0;
+
+    TMetrics& operator+=(const TMetrics& other);
+
+    void ToProto(NKikimrTabletBase::TMetrics* proto) const;
+};
+
 
 } // NHive
 } // NKikimr

ydb/core/mind/hive/hive_events.h

@@ -39,6 +39,7 @@ struct TEvPrivate {
         EvRefreshScaleRecommendation,
         EvUpdateFollowers,
         EvUpdateBalanceCounters,
+        EvProcessMetrics,
         EvEnd
     };
 
@@ -145,6 +146,8 @@ struct TEvPrivate {
     };
 
     struct TEvUpdateBalanceCounters : TEventLocal<TEvUpdateBalanceCounters, EvUpdateBalanceCounters> {};
+
+    struct TEvProcessMetrics : TEventLocal<TEvProcessMetrics, EvProcessMetrics> {};
 };
 
 } // NHive