fix

ivanmorozov333 · ivanmorozov333 · commit d47fd48642a4 · 2024-11-13T13:18:02.000+03:00
diff --git a/ydb/core/protos/config.proto b/ydb/core/protos/config.proto
@@ -634,6 +634,7 @@ message TLimiterConfig {
 message TGroupedMemoryLimiterConfig {
     optional bool Enabled = 1 [default = true];
     optional uint64 MemoryLimit = 2;
+    optional uint64 HardMemoryLimit = 3;
 }
 
 message TExternalIndexConfig {
diff --git a/ydb/core/tx/columnshard/engines/reader/abstract/read_context.h b/ydb/core/tx/columnshard/engines/reader/abstract/read_context.h
@@ -64,7 +64,8 @@ class TReadContext {
 
     void AbortWithError(const TString& errorMessage) {
         if (AtomicCas(&AbortFlag, 1, 0)) {
-            NActors::TActivationContext::Send(ScanActorId, std::make_unique<NColumnShard::TEvPrivate::TEvTaskProcessedResult>(errorMessage));
+            NActors::TActivationContext::Send(
+                ScanActorId, std::make_unique<NColumnShard::TEvPrivate::TEvTaskProcessedResult>(TConclusionStatus::Fail(errorMessage)));
         }
     }
 
diff --git a/ydb/core/tx/columnshard/engines/reader/plain_reader/iterator/fetching.cpp b/ydb/core/tx/columnshard/engines/reader/plain_reader/iterator/fetching.cpp
@@ -189,6 +189,14 @@ TAllocateMemoryStep::TFetchingStepAllocation::TFetchingStepAllocation(
     , TasksGuard(source->GetContext()->GetCommonContext()->GetCounters().GetResourcesAllocationTasksGuard()) {
 }
 
+void TAllocateMemoryStep::TFetchingStepAllocation::DoOnAllocationImpossible(const TString& errorMessage) {
+    auto sourcePtr = Source.lock();
+    if (sourcePtr) {
+        sourcePtr->GetContext()->GetCommonContext()->AbortWithError(
+            "cannot allocate memory for step " + Step.GetName() + ": '" + errorMessage + "'");
+    }
+}
+
 TConclusion<bool> TAllocateMemoryStep::DoExecuteInplace(
     const std::shared_ptr<IDataSource>& source, const TFetchingScriptCursor& step) const {
 
diff --git a/ydb/core/tx/columnshard/engines/reader/plain_reader/iterator/fetching.h b/ydb/core/tx/columnshard/engines/reader/plain_reader/iterator/fetching.h
@@ -237,9 +237,7 @@ class TAllocateMemoryStep: public IFetchingStep {
         NColumnShard::TCounterGuard TasksGuard;
         virtual bool DoOnAllocated(std::shared_ptr<NGroupedMemoryManager::TAllocationGuard>&& guard,
             const std::shared_ptr<NGroupedMemoryManager::IAllocation>& allocation) override;
-        virtual void DoOnAllocationImpossible(const TString& errorMessage) override {
-            Source->GetContext()->GetCommonContext()->AbortWithError("cannot allocate memory for step " + Step.GetName() + ": '" + errorMessage + "'");
-        }
+        virtual void DoOnAllocationImpossible(const TString& errorMessage) override;
     public:
         TFetchingStepAllocation(const std::shared_ptr<IDataSource>& source, const ui64 mem, const TFetchingScriptCursor& step);
     };
diff --git a/ydb/core/tx/columnshard/engines/reader/sys_view/chunks/chunks.cpp b/ydb/core/tx/columnshard/engines/reader/sys_view/chunks/chunks.cpp
@@ -199,4 +199,8 @@ TStatsIterator::TFetchingAccessorAllocation::TFetchingAccessorAllocation(
     , Context(context) {
 }
 
+void TStatsIterator::TFetchingAccessorAllocation::DoOnAllocationImpossible(const TString& errorMessage) {
+    Context->AbortWithError("cannot allocate memory for take accessors info: " + errorMessage);
+}
+
 }   // namespace NKikimr::NOlap::NReader::NSysView::NChunks
diff --git a/ydb/core/tx/columnshard/engines/reader/sys_view/chunks/chunks.h b/ydb/core/tx/columnshard/engines/reader/sys_view/chunks/chunks.h
@@ -117,9 +117,7 @@ class TStatsIterator: public NAbstract::TStatsIterator<NKikimr::NSysView::Schema
             AccessorsManager->AskData(std::move(Request));
             return true;
         }
-        virtual void DoOnAllocationImpossible(const TString& errorMessage) override {
-            Context->AbortWithError("cannot allocate memory for take accessors info: " + errorMessage);
-        }
+        virtual void DoOnAllocationImpossible(const TString& errorMessage) override;
 
         virtual void DoOnRequestsFinished(TDataAccessorsResult&& result) override {
             if (result.HasErrors()) {
diff --git a/ydb/core/tx/limiter/grouped_memory/service/allocation.h b/ydb/core/tx/limiter/grouped_memory/service/allocation.h
@@ -48,9 +48,10 @@ class TAllocationInfo: public NColumnShard::TMonitoringObjectsCounter<TAllocatio
         AFL_TRACE(NKikimrServices::GROUPED_MEMORY_LIMITER)("event", "allocated")("allocation_id", Identifier)("stage", Stage->GetName());
         AFL_VERIFY(Allocation)("status", GetAllocationStatus())("volume", AllocatedVolume)("id", Identifier)("stage", Stage->GetName())(
             "allocation_internal_group_id", AllocationInternalGroupId);
-        if (!Stage->Allocate(AllocatedVolume)) {
+        auto allocationResult = Stage->Allocate(AllocatedVolume);
+        if (allocationResult.IsFail()) {
             AllocationFailed = true;
-            Allocation->OnAllocationImpossible();
+            Allocation->OnAllocationImpossible(allocationResult.GetErrorMessage());
             return false;
         }
         const bool result = Allocation->OnAllocated(
diff --git a/ydb/core/tx/limiter/grouped_memory/usage/abstract.h b/ydb/core/tx/limiter/grouped_memory/usage/abstract.h
@@ -5,6 +5,7 @@
 #include <ydb/library/actors/core/actor.h>
 #include <ydb/library/actors/core/actorid.h>
 #include <ydb/library/actors/core/log.h>
+#include <ydb/library/conclusion/status.h>
 
 namespace NKikimr::NOlap::NGroupedMemoryManager {
 
@@ -94,8 +95,8 @@ class TPositiveControlInteger {
 class TStageFeatures {
 private:
     YDB_READONLY_DEF(TString, Name);
-    YDB_READONLY(ui64, HardLimit, 0);
     YDB_READONLY(ui64, Limit, 0);
+    YDB_READONLY(ui64, HardLimit, 0);
     YDB_ACCESSOR_DEF(TPositiveControlInteger, Usage);
     YDB_ACCESSOR_DEF(TPositiveControlInteger, Waiting);
     std::shared_ptr<TStageFeatures> Owner;
@@ -124,9 +125,9 @@ class TStageFeatures {
         , Counters(counters) {
     }
 
-    [[nodiscard]] bool Allocate(const ui64 volume) {
+    [[nodiscard]] TConclusionStatus Allocate(const ui64 volume) {
         if (HardLimit < Usage.Val() + volume) {
-            return false;
+            return TConclusionStatus::Fail(TStringBuilder() << "limit:" << HardLimit << ";val:" << Usage.Val() << ";delta=" << volume << ";");
         }
         Waiting.Sub(volume);
         Usage.Add(volume);
@@ -135,12 +136,13 @@ class TStageFeatures {
             Counters->Sub(volume, false);
         }
         if (Owner) {
-            if (!Owner->Allocate(volume)) {
+            const auto ownerResult = Owner->Allocate(volume);
+            if (ownerResult.IsFail()) {
                 Free(volume, true);
-                return false;
+                return ownerResult;
             }
         }
-        return true;
+        return TConclusionStatus::Success();
     }
 
     void Free(const ui64 volume, const bool allocated) {
@@ -208,7 +210,7 @@ class IAllocation {
     YDB_READONLY(ui64, Identifier, Counter.Inc());
     YDB_READONLY(ui64, Memory, 0);
     bool Allocated = false;
-    virtual void DoOnAllocationImpossible() = 0;
+    virtual void DoOnAllocationImpossible(const TString& errorMessage) = 0;
     virtual bool DoOnAllocated(
         std::shared_ptr<TAllocationGuard>&& guard, const std::shared_ptr<NGroupedMemoryManager::IAllocation>& allocation) = 0;
 
@@ -226,8 +228,8 @@ class IAllocation {
         return Allocated;
     }
 
-    void OnAllocationImpossible() {
-        DoOnAllocationImpossible();
+    void OnAllocationImpossible(const TString& errorMessage) {
+        DoOnAllocationImpossible(errorMessage);
     }
 
     [[nodiscard]] bool OnAllocated(
diff --git a/ydb/core/tx/limiter/grouped_memory/usage/config.cpp b/ydb/core/tx/limiter/grouped_memory/usage/config.cpp
@@ -7,13 +7,16 @@ bool TConfig::DeserializeFromProto(const NKikimrConfig::TGroupedMemoryLimiterCon
     if (config.HasMemoryLimit()) {
         MemoryLimit = config.GetMemoryLimit();
     }
+    if (config.HasHardMemoryLimit()) {
+        HardMemoryLimit = config.GetHardMemoryLimit();
+    }
     Enabled = config.GetEnabled();
     return true;
 }
 
 TString TConfig::DebugString() const {
     TStringBuilder sb;
-    sb << "MemoryLimit=" << MemoryLimit << ";Enabled=" << Enabled << ";";
+    sb << "MemoryLimit=" << MemoryLimit << ";HardMemoryLimit=" << HardMemoryLimit << ";Enabled=" << Enabled << ";";
     return sb;
 }
 
diff --git a/ydb/core/tx/limiter/grouped_memory/usage/config.h b/ydb/core/tx/limiter/grouped_memory/usage/config.h
@@ -8,6 +8,7 @@ class TConfig {
 private:
     YDB_READONLY(bool, Enabled, true);
     YDB_READONLY(ui64, MemoryLimit, ui64(3) << 30);
+    YDB_READONLY(ui64, HardMemoryLimit, ui64(10) << 30);
 
 public:
 
diff --git a/ydb/core/tx/limiter/grouped_memory/usage/service.h b/ydb/core/tx/limiter/grouped_memory/usage/service.h
@@ -21,8 +21,8 @@ class TServiceOperatorImpl {
     static void Register(const TConfig& serviceConfig, TIntrusivePtr<::NMonitoring::TDynamicCounters> counters) {
         Singleton<TSelf>()->Counters = std::make_shared<TCounters>(counters, TMemoryLimiterPolicy::Name);
         Singleton<TSelf>()->ServiceConfig = serviceConfig;
-        Singleton<TSelf>()->DefaultStageFeatures = std::make_shared<TStageFeatures>(
-            "GLOBAL", serviceConfig.GetMemoryLimit(), nullptr, Singleton<TSelf>()->Counters->BuildStageCounters("general"));
+        Singleton<TSelf>()->DefaultStageFeatures = std::make_shared<TStageFeatures>("GLOBAL", serviceConfig.GetMemoryLimit(),
+            serviceConfig.GetHardMemoryLimit(), nullptr, Singleton<TSelf>()->Counters->BuildStageCounters("general"));
     }
     static const TString& GetMemoryLimiterName() {
         Y_ABORT_UNLESS(TMemoryLimiterPolicy::Name.size() == 4);

Original file line number	Diff line number	Diff line change
`@@ -634,6 +634,7 @@ message TLimiterConfig {`
`634`	`634`	`message TGroupedMemoryLimiterConfig {`
`635`	`635`	`optional bool Enabled = 1 [default = true];`
`636`	`636`	`optional uint64 MemoryLimit = 2;`
	`637`	`+ optional uint64 HardMemoryLimit = 3;`
`637`	`638`	`}`
`638`	`639`
`639`	`640`	`message TExternalIndexConfig {`
Original file line number	Diff line number	Diff line change
`@@ -64,7 +64,8 @@ class TReadContext {`
`64`	`64`
`65`	`65`	`void AbortWithError(const TString& errorMessage) {`
`66`	`66`	`if (AtomicCas(&AbortFlag, 1, 0)) {`
`67`		`- NActors::TActivationContext::Send(ScanActorId, std::make_unique<NColumnShard::TEvPrivate::TEvTaskProcessedResult>(errorMessage));`
	`67`	`+ NActors::TActivationContext::Send(`
	`68`	`+ ScanActorId, std::make_unique<NColumnShard::TEvPrivate::TEvTaskProcessedResult>(TConclusionStatus::Fail(errorMessage)));`
`68`	`69`	`}`
`69`	`70`	`}`
`70`	`71`
Original file line number	Diff line number	Diff line change
`@@ -199,4 +199,8 @@ TStatsIterator::TFetchingAccessorAllocation::TFetchingAccessorAllocation(`
`199`	`199`	`, Context(context) {`
`200`	`200`	`}`
`201`	`201`
	`202`	`+void TStatsIterator::TFetchingAccessorAllocation::DoOnAllocationImpossible(const TString& errorMessage) {`
	`203`	`+ Context->AbortWithError("cannot allocate memory for take accessors info: " + errorMessage);`
	`204`	`+}`
	`205`	`+`
`202`	`206`	`} // namespace NKikimr::NOlap::NReader::NSysView::NChunks`
Original file line number	Diff line number	Diff line change
`@@ -7,13 +7,16 @@ bool TConfig::DeserializeFromProto(const NKikimrConfig::TGroupedMemoryLimiterCon`
`7`	`7`	`if (config.HasMemoryLimit()) {`
`8`	`8`	`MemoryLimit = config.GetMemoryLimit();`
`9`	`9`	`}`
	`10`	`+ if (config.HasHardMemoryLimit()) {`
	`11`	`+ HardMemoryLimit = config.GetHardMemoryLimit();`
	`12`	`+ }`
`10`	`13`	`Enabled = config.GetEnabled();`
`11`	`14`	`return true;`
`12`	`15`	`}`
`13`	`16`
`14`	`17`	`TString TConfig::DebugString() const {`
`15`	`18`	`TStringBuilder sb;`
`16`		`- sb << "MemoryLimit=" << MemoryLimit << ";Enabled=" << Enabled << ";";`
	`19`	`+ sb << "MemoryLimit=" << MemoryLimit << ";HardMemoryLimit=" << HardMemoryLimit << ";Enabled=" << Enabled << ";";`
`17`	`20`	`return sb;`
`18`	`21`	`}`
`19`	`22`