fix allocation cleaning race from separated thread after scope cleani… (

#10096)
ydb-platform · Oct 5, 2024 · 66e070c · 66e070c
1 parent 516adee
commit 66e070c
Show file tree

Hide file tree

Showing 3 changed files with 19 additions and 10 deletions.
diff --git a/ydb/core/tx/limiter/grouped_memory/service/allocation.h b/ydb/core/tx/limiter/grouped_memory/service/allocation.h
@@ -1,4 +1,5 @@
 #pragma once
+#include <ydb/core/tx/columnshard/counters/common/object_counter.h>
 #include <ydb/core/tx/limiter/grouped_memory/usage/abstract.h>
 
 namespace NKikimr::NOlap::NGroupedMemoryManager {
@@ -9,7 +10,7 @@ enum class EAllocationStatus {
     Failed
 };
 
-class TAllocationInfo {
+class TAllocationInfo: public NColumnShard::TMonitoringObjectsCounter<TAllocationInfo> {
 private:
     std::shared_ptr<IAllocation> Allocation;
     YDB_READONLY(ui64, AllocationInternalGroupId, 0);
@@ -25,7 +26,7 @@ class TAllocationInfo {
         if (GetAllocationStatus() != EAllocationStatus::Failed) {
             Stage->Free(AllocatedVolume, GetAllocationStatus() == EAllocationStatus::Allocated);
         }
-        
+
         AFL_TRACE(NKikimrServices::GROUPED_MEMORY_LIMITER)("event", "destroy")("allocation_id", Identifier)("stage", Stage->GetName());
     }
 
@@ -69,8 +70,8 @@ class TAllocationInfo {
         }
     }
 
-    TAllocationInfo(const ui64 processId, const ui64 scopeId, const ui64 allocationInternalGroupId, const std::shared_ptr<IAllocation>& allocation,
-        const std::shared_ptr<TStageFeatures>& stage);
+    TAllocationInfo(const ui64 processId, const ui64 scopeId, const ui64 allocationInternalGroupId,
+        const std::shared_ptr<IAllocation>& allocation, const std::shared_ptr<TStageFeatures>& stage);
 };
 
 }   // namespace NKikimr::NOlap::NGroupedMemoryManager
diff --git a/ydb/core/tx/limiter/grouped_memory/service/group.h b/ydb/core/tx/limiter/grouped_memory/service/group.h
@@ -1,11 +1,13 @@
 #pragma once
 #include "allocation.h"
 
+#include <ydb/core/tx/columnshard/counters/common/object_counter.h>
+
 namespace NKikimr::NOlap::NGroupedMemoryManager {
 
 class TProcessMemoryScope;
 
-class TGrouppedAllocations {
+class TGrouppedAllocations: public NColumnShard::TMonitoringObjectsCounter<TGrouppedAllocations> {
 private:
     THashMap<ui64, std::shared_ptr<TAllocationInfo>> Allocations;
 

diff --git a/ydb/core/tx/limiter/grouped_memory/service/process.h b/ydb/core/tx/limiter/grouped_memory/service/process.h
@@ -2,11 +2,13 @@
 #include "group.h"
 #include "ids.h"
 
+#include <ydb/core/tx/columnshard/counters/common/object_counter.h>
+
 #include <ydb/library/accessor/validator.h>
 
 namespace NKikimr::NOlap::NGroupedMemoryManager {
 
-class TProcessMemoryScope {
+class TProcessMemoryScope: public NColumnShard::TMonitoringObjectsCounter<TProcessMemoryScope> {
 private:
     const ui64 ExternalProcessId;
     const ui64 ExternalScopeId;
@@ -63,6 +65,8 @@ class TProcessMemoryScope {
         }
         GroupIds.Clear();
         AllocationInfo.clear();
+        AFL_INFO(NKikimrServices::GROUPED_MEMORY_LIMITER)("event", "scope_cleaned")("process_id", ExternalProcessId)(
+            "external_scope_id", ExternalScopeId);
         return true;
     }
 
@@ -106,7 +110,11 @@ class TProcessMemoryScope {
     bool UnregisterAllocation(const ui64 allocationId) {
         ui64 memoryAllocated = 0;
         auto it = AllocationInfo.find(allocationId);
-        AFL_VERIFY(it != AllocationInfo.end());
+        if (it == AllocationInfo.end()) {
+            AFL_WARN(NKikimrServices::GROUPED_MEMORY_LIMITER)("reason", "allocation_cleaned_in_previous_scope_id_live")(
+                "allocation_id", allocationId)("process_id", ExternalProcessId)("external_scope_id", ExternalScopeId);
+            return true;
+        }
         bool waitFlag = false;
         const ui64 internalGroupId = it->second->GetAllocationInternalGroupId();
         switch (it->second->GetAllocationStatus()) {
@@ -144,7 +152,7 @@ class TProcessMemoryScope {
     }
 };
 
-class TProcessMemory {
+class TProcessMemory: public NColumnShard::TMonitoringObjectsCounter<TProcessMemory> {
 private:
     const ui64 ExternalProcessId;
 
@@ -217,7 +225,6 @@ class TProcessMemory {
         if (it->second->Unregister()) {
             AllocationScopes.erase(it);
         }
-
     }
 
     void RegisterScope(const ui64 externalScopeId) {
@@ -227,7 +234,6 @@ class TProcessMemory {
         } else {
             it->second->Register();
         }
-
     }
 
     void SetPriorityProcess() {