additional info on abort (#3815)

Author: ivanmorozov333

ydb/core/tx/columnshard/blobs_action/transaction/tx_draft.h

@@ -12,7 +12,7 @@ class TTxWriteDraft: public TTransactionBase<TColumnShard> {
 public:
     ~TTxWriteDraft() {
         if (!Completed) {
-            WriteController->Abort();
+            WriteController->Abort("TTxWriteDraft aborted before complete");
         }
     }
 

ydb/core/tx/columnshard/blobs_action/transaction/tx_write_index.cpp

@@ -38,7 +38,7 @@ bool TTxWriteIndex::Execute(TTransactionContext& txc, const TActorContext& ctx)
             auto& portion = changes->GetWritePortionInfo(i)->GetPortionResult();
             LOG_S_WARN(TxPrefix() << "(" << changes->TypeString() << ":" << portion.DebugString() << ") blob cannot apply changes: " << TxSuffix());
         }
-        NOlap::TChangesFinishContext context("cannot write index blobs");
+        NOlap::TChangesFinishContext context("cannot write index blobs: " + ::ToString(Ev->Get()->GetPutStatus()));
         changes->Abort(*Self, context);
         LOG_S_ERROR(TxPrefix() << " (" << changes->TypeString() << ") cannot write index blobs" << TxSuffix());
     }
@@ -71,7 +71,7 @@ TTxWriteIndex::~TTxWriteIndex() {
     if (Ev) {
         auto changes = Ev->Get()->IndexChanges;
         if (!CompleteReady && changes) {
-            changes->AbortEmergency();
+            changes->AbortEmergency("TTxWriteIndex destructor withno CompleteReady flag");
         }
     }
 }

ydb/core/tx/columnshard/engines/changes/abstract/abstract.cpp

@@ -52,8 +52,8 @@ void TColumnEngineChanges::WriteIndexOnComplete(NColumnShard::TColumnShard* self
 }
 
 void TColumnEngineChanges::Compile(TFinalizationContext& context) noexcept {
-    Y_ABORT_UNLESS(Stage != EStage::Aborted);
-    Y_ABORT_UNLESS(Stage == EStage::Constructed);
+    AFL_VERIFY(Stage != EStage::Aborted);
+    AFL_VERIFY(Stage == EStage::Constructed);
 
     DoCompile(context);
     DoOnAfterCompile();
@@ -66,7 +66,7 @@ TColumnEngineChanges::~TColumnEngineChanges() {
 }
 
 void TColumnEngineChanges::Abort(NColumnShard::TColumnShard& self, TChangesFinishContext& context) {
-    Y_ABORT_UNLESS(Stage != EStage::Finished && Stage != EStage::Created && Stage != EStage::Aborted);
+    AFL_VERIFY(Stage != EStage::Finished && Stage != EStage::Created && Stage != EStage::Aborted)("stage", Stage)("reason", context.ErrorMessage);
     Stage = EStage::Aborted;
     OnFinish(self, context);
 }
@@ -91,8 +91,8 @@ void TColumnEngineChanges::StartEmergency() {
     }
 }
 
-void TColumnEngineChanges::AbortEmergency() {
-    AFL_WARN(NKikimrServices::TX_COLUMNSHARD)("event", "AbortEmergency");
+void TColumnEngineChanges::AbortEmergency(const TString& reason) {
+    AFL_WARN(NKikimrServices::TX_COLUMNSHARD)("event", "AbortEmergency")("reason", reason);
     Stage = EStage::Aborted;
     OnAbortEmergency();
 }

ydb/core/tx/columnshard/engines/changes/abstract/abstract.h

@@ -264,7 +264,7 @@ class TColumnEngineChanges {
     }
 
     void StartEmergency();
-    void AbortEmergency();
+    void AbortEmergency(const TString& reason);
 
     void Abort(NColumnShard::TColumnShard& self, TChangesFinishContext& context);
     void Start(NColumnShard::TColumnShard& self);

ydb/core/tx/columnshard/engines/ut/ut_logs_engine.cpp

@@ -311,7 +311,7 @@ bool Insert(TColumnEngineForLogs& engine, TTestDbWrapper& db, TSnapshot snap,
     changes->WriteIndexOnExecute(nullptr, contextExecute);
     NOlap::TWriteIndexCompleteContext contextComplete(NActors::TActivationContext::AsActorContext(), 0, 0, TDuration::Zero(), engine);
     changes->WriteIndexOnComplete(nullptr, contextComplete);
-    changes->AbortEmergency();
+    changes->AbortEmergency("testing");
     return result;
 }
 
@@ -348,7 +348,7 @@ bool Compact(TColumnEngineForLogs& engine, TTestDbWrapper& db, TSnapshot snap, N
             }
         }
     }
-    changes->AbortEmergency();
+    changes->AbortEmergency("testing");
     return result;
 }
 
@@ -368,7 +368,7 @@ bool Cleanup(TColumnEngineForLogs& engine, TTestDbWrapper& db, TSnapshot snap, u
     changes->WriteIndexOnExecute(nullptr, contextExecute);
     NOlap::TWriteIndexCompleteContext contextComplete(NActors::TActivationContext::AsActorContext(), 0, 0, TDuration::Zero(), engine);
     changes->WriteIndexOnComplete(nullptr, contextComplete);
-    changes->AbortEmergency();
+    changes->AbortEmergency("testing");
     return result;
 }
 
@@ -386,7 +386,7 @@ bool Ttl(TColumnEngineForLogs& engine, TTestDbWrapper& db,
     changes->WriteIndexOnExecute(nullptr, contextExecute);
     NOlap::TWriteIndexCompleteContext contextComplete(NActors::TActivationContext::AsActorContext(), 0, 0, TDuration::Zero(), engine);
     changes->WriteIndexOnComplete(nullptr, contextComplete);
-    changes->AbortEmergency();
+    changes->AbortEmergency("testing");
     return result;
 }
 

ydb/core/tx/columnshard/engines/writer/compacted_blob_constructor.cpp

@@ -34,16 +34,18 @@ void TCompactedWriteController::DoOnReadyResult(const NActors::TActorContext& ct
 
 TCompactedWriteController::~TCompactedWriteController() {
     if (WriteIndexEv && WriteIndexEv->IndexChanges) {
-        WriteIndexEv->IndexChanges->AbortEmergency();
+        WriteIndexEv->IndexChanges->AbortEmergency("TCompactedWriteController destructed with WriteIndexEv and WriteIndexEv->IndexChanges");
     }
 }
 
 const NKikimr::NOlap::TBlobsAction& TCompactedWriteController::GetBlobsAction() {
     return WriteIndexEv->IndexChanges->GetBlobsAction();
 }
 
-void TCompactedWriteController::DoAbort() {
-    WriteIndexEv->IndexChanges->AbortEmergency();
+void TCompactedWriteController::DoAbort(const TString& reason) {
+    if (WriteIndexEv && WriteIndexEv->IndexChanges) {
+        WriteIndexEv->IndexChanges->AbortEmergency("TCompactedWriteController aborted: " + reason);
+    }
 }
 
 }

ydb/core/tx/columnshard/engines/writer/compacted_blob_constructor.h

@@ -15,7 +15,7 @@ class TCompactedWriteController : public NColumnShard::IWriteController {
     TActorId DstActor;
 protected:
     void DoOnReadyResult(const NActors::TActorContext& ctx, const NColumnShard::TBlobPutResult::TPtr& putResult) override;
-    virtual void DoAbort() override;
+    virtual void DoAbort(const TString& reason) override;
 public:
     const TBlobsAction& GetBlobsAction();
 

ydb/core/tx/columnshard/engines/writer/write_controller.h

@@ -41,7 +41,7 @@ class IWriteController {
     }
 
     NOlap::TBlobWriteInfo& AddWriteTask(NOlap::TBlobWriteInfo&& task);
-    virtual void DoAbort() {
+    virtual void DoAbort(const TString& /*reason*/) {
     }
 public:
     const NOlap::TWriteActionsCollection& GetBlobActions() const {
@@ -61,11 +61,12 @@ class IWriteController {
         return TStringBuilder() << "size=" << size << ";count=" << WriteTasks.size() << ";actions=" << sb << ";";
     }
 
-    void Abort() {
+    void Abort(const TString& reason) {
+        AFL_WARN(NKikimrServices::TX_COLUMNSHARD)("event", "IWriteController aborted")("reason", reason);
         for (auto&& i : WritingActions) {
             i.second->Abort();
         }
-        DoAbort();
+        DoAbort(reason);
     }
 
     using TPtr = std::shared_ptr<IWriteController>;

ydb/core/tx/columnshard/write_actor.cpp

@@ -36,7 +36,7 @@ class TWriteActor: public TActorBootstrapped<TWriteActor>, public TMonitoringObj
 
         if (status != NKikimrProto::OK) {
             ACFL_ERROR("event", "TEvPutResult")("blob_id", msg->Id.ToString())("status", status)("error", msg->ErrorReason);
-            WriteController->Abort();
+            WriteController->Abort("cannot write blob " + msg->Id.ToString() + ", status: " + ::ToString(status) + ". reason: " + msg->ErrorReason);
             return SendResultAndDie(ctx, status);
         }