dont reallocate memory after merge through incorrect memory consumpti… (#12623)

Author: ivanmorozov333

ydb/core/tx/columnshard/engines/reader/plain_reader/iterator/context.cpp

@@ -13,7 +13,6 @@ std::unique_ptr<NArrow::NMerger::TMergePartialStream> TSpecialReadContext::Build
 ui64 TSpecialReadContext::GetMemoryForSources(const THashMap<ui32, std::shared_ptr<IDataSource>>& sources) {
     ui64 result = 0;
     for (auto&& i : sources) {
-        auto fetchingPlan = GetColumnsFetchingPlan(i.second);
         AFL_VERIFY(i.second->GetIntervalsCount());
         const ui64 sourceMemory = std::max<ui64>(1, i.second->GetResourceGuardsMemory() / i.second->GetIntervalsCount());
         result += sourceMemory;

ydb/core/tx/columnshard/engines/reader/plain_reader/iterator/fetching.cpp

@@ -222,6 +222,7 @@ TConclusion<bool> TAllocateMemoryStep::DoExecuteInplace(const std::shared_ptr<ID
         }
         size += sizeLocal;
     }
+    AFL_DEBUG(NKikimrServices::TX_COLUMNSHARD_SCAN)("event", "TAllocateMemoryStep::DoExecuteInplace")("source", source->GetSourceIdx())("memory", size);
 
     auto allocation = std::make_shared<TFetchingStepAllocation>(source, size, step, StageIndex);
     NGroupedMemoryManager::TScanMemoryLimiterOperator::SendToAllocation(source->GetContext()->GetProcessMemoryControlId(),

ydb/core/tx/columnshard/engines/reader/plain_reader/iterator/interval.cpp

@@ -10,15 +10,12 @@ void TFetchingInterval::ConstructResult() {
     if (ready != WaitSourcesCount) {
         AFL_DEBUG(NKikimrServices::TX_COLUMNSHARD_SCAN)("event", "skip_construct_result")("interval_idx", IntervalIdx)(
             "count", WaitSourcesCount)("ready", ready)("interval_id", GetIntervalId());
-        return;
-    } else {
-        AFL_DEBUG(NKikimrServices::TX_COLUMNSHARD_SCAN)("event", "start_construct_result")("interval_idx", IntervalIdx)(
-            "interval_id", GetIntervalId());
-    }
-    if (AtomicCas(&SourcesFinalized, 1, 0)) {
+    } else if (AtomicCas(&SourcesFinalized, 1, 0)) {
         IntervalStateGuard.SetStatus(NColumnShard::TScanCounters::EIntervalStatus::WaitMergerStart);
 
         MergingContext->SetIntervalChunkMemory(Context->GetMemoryForSources(Sources));
+        AFL_DEBUG(NKikimrServices::TX_COLUMNSHARD_SCAN)("event", "start_construct_result")("interval_idx", IntervalIdx)(
+            "interval_id", GetIntervalId())("memory", MergingContext->GetIntervalChunkMemory())("count", WaitSourcesCount);
 
         auto task = std::make_shared<TStartMergeTask>(MergingContext, Context, std::move(Sources));
         task->SetPriority(NConveyor::ITask::EPriority::High);
@@ -81,6 +78,8 @@ void TFetchingInterval::OnPartSendingComplete() {
     }
     IntervalStateGuard.SetStatus(NColumnShard::TScanCounters::EIntervalStatus::WaitMergerContinue);
 
+    AFL_DEBUG(NKikimrServices::TX_COLUMNSHARD_SCAN)("event", "continue_construct_result")("interval_idx", IntervalIdx)(
+        "interval_id", GetIntervalId())("memory", MergingContext->GetIntervalChunkMemory())("count", WaitSourcesCount);
     auto task = std::make_shared<TContinueMergeTask>(MergingContext, Context, std::move(Merger));
     task->SetPriority(NConveyor::ITask::EPriority::High);
     NGroupedMemoryManager::TScanMemoryLimiterOperator::SendToAllocation(Context->GetProcessMemoryControlId(),

ydb/core/tx/columnshard/engines/reader/plain_reader/iterator/merge.cpp

@@ -33,6 +33,8 @@ void TBaseMergeTask::PrepareResultBatch() {
         LastPK = nullptr;
         return;
     }
+    const ui64 dataSizeBefore = NArrow::GetTableDataSize(ResultBatch);
+    const ui64 memorySizeBefore = NArrow::GetTableMemorySize(ResultBatch);
     {
         ResultBatch = NArrow::TColumnOperator().VerifyIfAbsent().Extract(ResultBatch, Context->GetProgramInputColumns()->GetColumnNamesVector());
         AFL_VERIFY((ui32)ResultBatch->num_columns() == Context->GetProgramInputColumns()->GetColumnNamesVector().size());
@@ -45,7 +47,10 @@ void TBaseMergeTask::PrepareResultBatch() {
         } else {
             ShardedBatch = NArrow::TShardedRecordBatch(ResultBatch);
         }
-        AllocationGuard->Update(NArrow::GetTableMemorySize(ResultBatch));
+        AFL_DEBUG(NKikimrServices::TX_COLUMNSHARD_SCAN)("event", "update_memory_merger")("before_data", dataSizeBefore)(
+            "before_memory", memorySizeBefore)("after_memory", NArrow::GetTableMemorySize(ResultBatch))(
+            "after_data", NArrow::GetTableDataSize(ResultBatch))("guard", AllocationGuard->GetMemory());
+        //        AllocationGuard->Update(NArrow::GetTableMemorySize(ResultBatch));
         AFL_VERIFY(!!LastPK == !!ShardedBatch->GetRecordsCount())("lpk", !!LastPK)("sb", ShardedBatch->GetRecordsCount());
     } else {
         AllocationGuard = nullptr;

ydb/core/tx/limiter/grouped_memory/usage/abstract.h

@@ -129,9 +129,13 @@ class TStageFeatures {
         Waiting.Sub(volume);
         if (HardLimit < Usage.Val() + volume) {
             Counters->OnCannotAllocate();
+            AFL_DEBUG(NKikimrServices::GROUPED_MEMORY_LIMITER)("name", Name)("event", "cannot_allocate")("limit", HardLimit)(
+                "usage", Usage.Val())(
+                "delta", volume);
             return TConclusionStatus::Fail(TStringBuilder() << Name << "::(limit:" << HardLimit << ";val:" << Usage.Val() << ";delta=" << volume << ");");
         }
         Usage.Add(volume);
+        AFL_DEBUG(NKikimrServices::GROUPED_MEMORY_LIMITER)("name", Name)("event", "allocate")("usage", Usage.Val())("delta", volume);
         if (Counters) {
             Counters->Add(volume, true);
             Counters->Sub(volume, false);
@@ -155,6 +159,7 @@ class TStageFeatures {
         } else {
             Waiting.Sub(volume);
         }
+        AFL_DEBUG(NKikimrServices::GROUPED_MEMORY_LIMITER)("name", Name)("event", "free")("usage", Usage.Val())("delta", volume);
 
         if (withOwner && Owner) {
             Owner->Free(volume, allocated);
@@ -166,6 +171,8 @@ class TStageFeatures {
             Counters->Sub(from, allocated);
             Counters->Add(to, allocated);
         }
+        AFL_DEBUG(NKikimrServices::GROUPED_MEMORY_LIMITER)("name", Name)("event", "update")("usage", Usage.Val())("waiting", Waiting.Val())(
+            "allocated", allocated)("from", from)("to", to);
         if (allocated) {
             Usage.Sub(from);
             Usage.Add(to);