additional info on abort (#3815)

ivanmorozov333 · web-flow · commit 79f96db2e7e1 · 2024-04-17T15:31:01.000+03:00
diff --git a/ydb/core/tx/columnshard/blobs_action/transaction/tx_draft.h b/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");
         }
     }
 
diff --git a/ydb/core/tx/columnshard/blobs_action/transaction/tx_write_index.cpp b/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");
         }
     }
 }
diff --git a/ydb/core/tx/columnshard/engines/changes/abstract/abstract.cpp b/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();
 }
diff --git a/ydb/core/tx/columnshard/engines/changes/abstract/abstract.h b/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);
diff --git a/ydb/core/tx/columnshard/engines/ut/ut_logs_engine.cpp b/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;
 }
 
diff --git a/ydb/core/tx/columnshard/engines/writer/compacted_blob_constructor.cpp b/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);
+    }
 }
 
 }
diff --git a/ydb/core/tx/columnshard/engines/writer/compacted_blob_constructor.h b/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();
 
diff --git a/ydb/core/tx/columnshard/engines/writer/write_controller.h b/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>;
diff --git a/ydb/core/tx/columnshard/write_actor.cpp b/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);
         }
 

Original file line number	Diff line number	Diff line change
`@@ -12,7 +12,7 @@ class TTxWriteDraft: public TTransactionBase<TColumnShard> {`
`12`	`12`	`public:`
`13`	`13`	`~TTxWriteDraft() {`
`14`	`14`	`if (!Completed) {`
`15`		`- WriteController->Abort();`
	`15`	`+ WriteController->Abort("TTxWriteDraft aborted before complete");`
`16`	`16`	`}`
`17`	`17`	`}`
`18`	`18`
Original file line number	Diff line number	Diff line change
`@@ -38,7 +38,7 @@ bool TTxWriteIndex::Execute(TTransactionContext& txc, const TActorContext& ctx)`
`38`	`38`	`auto& portion = changes->GetWritePortionInfo(i)->GetPortionResult();`
`39`	`39`	`LOG_S_WARN(TxPrefix() << "(" << changes->TypeString() << ":" << portion.DebugString() << ") blob cannot apply changes: " << TxSuffix());`
`40`	`40`	`}`
`41`		`- NOlap::TChangesFinishContext context("cannot write index blobs");`
	`41`	`+ NOlap::TChangesFinishContext context("cannot write index blobs: " + ::ToString(Ev->Get()->GetPutStatus()));`
`42`	`42`	`changes->Abort(*Self, context);`
`43`	`43`	`LOG_S_ERROR(TxPrefix() << " (" << changes->TypeString() << ") cannot write index blobs" << TxSuffix());`
`44`	`44`	`}`
`@@ -71,7 +71,7 @@ TTxWriteIndex::~TTxWriteIndex() {`
`71`	`71`	`if (Ev) {`
`72`	`72`	`auto changes = Ev->Get()->IndexChanges;`
`73`	`73`	`if (!CompleteReady && changes) {`
`74`		`- changes->AbortEmergency();`
	`74`	`+ changes->AbortEmergency("TTxWriteIndex destructor withno CompleteReady flag");`
`75`	`75`	`}`
`76`	`76`	`}`
`77`	`77`	`}`
Original file line number	Diff line number	Diff line change
`@@ -52,8 +52,8 @@ void TColumnEngineChanges::WriteIndexOnComplete(NColumnShard::TColumnShard* self`
`52`	`52`	`}`
`53`	`53`
`54`	`54`	`void TColumnEngineChanges::Compile(TFinalizationContext& context) noexcept {`
`55`		`- Y_ABORT_UNLESS(Stage != EStage::Aborted);`
`56`		`- Y_ABORT_UNLESS(Stage == EStage::Constructed);`
	`55`	`+ AFL_VERIFY(Stage != EStage::Aborted);`
	`56`	`+ AFL_VERIFY(Stage == EStage::Constructed);`
`57`	`57`
`58`	`58`	`DoCompile(context);`
`59`	`59`	`DoOnAfterCompile();`
`@@ -66,7 +66,7 @@ TColumnEngineChanges::~TColumnEngineChanges() {`
`66`	`66`	`}`
`67`	`67`
`68`	`68`	`void TColumnEngineChanges::Abort(NColumnShard::TColumnShard& self, TChangesFinishContext& context) {`
`69`		`- Y_ABORT_UNLESS(Stage != EStage::Finished && Stage != EStage::Created && Stage != EStage::Aborted);`
	`69`	`+ AFL_VERIFY(Stage != EStage::Finished && Stage != EStage::Created && Stage != EStage::Aborted)("stage", Stage)("reason", context.ErrorMessage);`
`70`	`70`	`Stage = EStage::Aborted;`
`71`	`71`	`OnFinish(self, context);`
`72`	`72`	`}`
`@@ -91,8 +91,8 @@ void TColumnEngineChanges::StartEmergency() {`
`91`	`91`	`}`
`92`	`92`	`}`
`93`	`93`
`94`		`-void TColumnEngineChanges::AbortEmergency() {`
`95`		`- AFL_WARN(NKikimrServices::TX_COLUMNSHARD)("event", "AbortEmergency");`
	`94`	`+void TColumnEngineChanges::AbortEmergency(const TString& reason) {`
	`95`	`+ AFL_WARN(NKikimrServices::TX_COLUMNSHARD)("event", "AbortEmergency")("reason", reason);`
`96`	`96`	`Stage = EStage::Aborted;`
`97`	`97`	`OnAbortEmergency();`
`98`	`98`	`}`
Original file line number	Diff line number	Diff line change
`@@ -264,7 +264,7 @@ class TColumnEngineChanges {`
`264`	`264`	`}`
`265`	`265`
`266`	`266`	`void StartEmergency();`
`267`		`- void AbortEmergency();`
	`267`	`+ void AbortEmergency(const TString& reason);`
`268`	`268`
`269`	`269`	`void Abort(NColumnShard::TColumnShard& self, TChangesFinishContext& context);`
`270`	`270`	`void Start(NColumnShard::TColumnShard& self);`
Original file line number	Diff line number	Diff line change
`@@ -311,7 +311,7 @@ bool Insert(TColumnEngineForLogs& engine, TTestDbWrapper& db, TSnapshot snap,`
`311`	`311`	`changes->WriteIndexOnExecute(nullptr, contextExecute);`
`312`	`312`	`NOlap::TWriteIndexCompleteContext contextComplete(NActors::TActivationContext::AsActorContext(), 0, 0, TDuration::Zero(), engine);`
`313`	`313`	`changes->WriteIndexOnComplete(nullptr, contextComplete);`
`314`		`- changes->AbortEmergency();`
	`314`	`+ changes->AbortEmergency("testing");`
`315`	`315`	`return result;`
`316`	`316`	`}`
`317`	`317`
`@@ -348,7 +348,7 @@ bool Compact(TColumnEngineForLogs& engine, TTestDbWrapper& db, TSnapshot snap, N`
`348`	`348`	`}`
`349`	`349`	`}`
`350`	`350`	`}`
`351`		`- changes->AbortEmergency();`
	`351`	`+ changes->AbortEmergency("testing");`
`352`	`352`	`return result;`
`353`	`353`	`}`
`354`	`354`
`@@ -368,7 +368,7 @@ bool Cleanup(TColumnEngineForLogs& engine, TTestDbWrapper& db, TSnapshot snap, u`
`368`	`368`	`changes->WriteIndexOnExecute(nullptr, contextExecute);`
`369`	`369`	`NOlap::TWriteIndexCompleteContext contextComplete(NActors::TActivationContext::AsActorContext(), 0, 0, TDuration::Zero(), engine);`
`370`	`370`	`changes->WriteIndexOnComplete(nullptr, contextComplete);`
`371`		`- changes->AbortEmergency();`
	`371`	`+ changes->AbortEmergency("testing");`
`372`	`372`	`return result;`
`373`	`373`	`}`
`374`	`374`
`@@ -386,7 +386,7 @@ bool Ttl(TColumnEngineForLogs& engine, TTestDbWrapper& db,`
`386`	`386`	`changes->WriteIndexOnExecute(nullptr, contextExecute);`
`387`	`387`	`NOlap::TWriteIndexCompleteContext contextComplete(NActors::TActivationContext::AsActorContext(), 0, 0, TDuration::Zero(), engine);`
`388`	`388`	`changes->WriteIndexOnComplete(nullptr, contextComplete);`
`389`		`- changes->AbortEmergency();`
	`389`	`+ changes->AbortEmergency("testing");`
`390`	`390`	`return result;`
`391`	`391`	`}`
`392`	`392`
Original file line number	Diff line number	Diff line change
`@@ -34,16 +34,18 @@ void TCompactedWriteController::DoOnReadyResult(const NActors::TActorContext& ct`
`34`	`34`
`35`	`35`	`TCompactedWriteController::~TCompactedWriteController() {`
`36`	`36`	`if (WriteIndexEv && WriteIndexEv->IndexChanges) {`
`37`		`- WriteIndexEv->IndexChanges->AbortEmergency();`
	`37`	`+ WriteIndexEv->IndexChanges->AbortEmergency("TCompactedWriteController destructed with WriteIndexEv and WriteIndexEv->IndexChanges");`
`38`	`38`	`}`
`39`	`39`	`}`
`40`	`40`
`41`	`41`	`const NKikimr::NOlap::TBlobsAction& TCompactedWriteController::GetBlobsAction() {`
`42`	`42`	`return WriteIndexEv->IndexChanges->GetBlobsAction();`
`43`	`43`	`}`
`44`	`44`
`45`		`-void TCompactedWriteController::DoAbort() {`
`46`		`- WriteIndexEv->IndexChanges->AbortEmergency();`
	`45`	`+void TCompactedWriteController::DoAbort(const TString& reason) {`
	`46`	`+ if (WriteIndexEv && WriteIndexEv->IndexChanges) {`
	`47`	`+ WriteIndexEv->IndexChanges->AbortEmergency("TCompactedWriteController aborted: " + reason);`
	`48`	`+ }`
`47`	`49`	`}`
`48`	`50`
`49`	`51`	`}`
Original file line number	Diff line number	Diff line change
`@@ -41,7 +41,7 @@ class IWriteController {`
`41`	`41`	`}`
`42`	`42`
`43`	`43`	`NOlap::TBlobWriteInfo& AddWriteTask(NOlap::TBlobWriteInfo&& task);`
`44`		`- virtual void DoAbort() {`
	`44`	`+ virtual void DoAbort(const TString& /reason/) {`
`45`	`45`	`}`
`46`	`46`	`public:`
`47`	`47`	`const NOlap::TWriteActionsCollection& GetBlobActions() const {`
`@@ -61,11 +61,12 @@ class IWriteController {`
`61`	`61`	`return TStringBuilder() << "size=" << size << ";count=" << WriteTasks.size() << ";actions=" << sb << ";";`
`62`	`62`	`}`
`63`	`63`
`64`		`- void Abort() {`
	`64`	`+ void Abort(const TString& reason) {`
	`65`	`+ AFL_WARN(NKikimrServices::TX_COLUMNSHARD)("event", "IWriteController aborted")("reason", reason);`
`65`	`66`	`for (auto&& i : WritingActions) {`
`66`	`67`	`i.second->Abort();`
`67`	`68`	`}`
`68`		`- DoAbort();`
	`69`	`+ DoAbort(reason);`
`69`	`70`	`}`
`70`	`71`
`71`	`72`	`using TPtr = std::shared_ptr<IWriteController>;`
Original file line number	Diff line number	Diff line change
`@@ -36,7 +36,7 @@ class TWriteActor: public TActorBootstrapped<TWriteActor>, public TMonitoringObj`
`36`	`36`
`37`	`37`	`if (status != NKikimrProto::OK) {`
`38`	`38`	`ACFL_ERROR("event", "TEvPutResult")("blob_id", msg->Id.ToString())("status", status)("error", msg->ErrorReason);`
`39`		`- WriteController->Abort();`
	`39`	`+ WriteController->Abort("cannot write blob " + msg->Id.ToString() + ", status: " + ::ToString(status) + ". reason: " + msg->ErrorReason);`
`40`	`40`	`return SendResultAndDie(ctx, status);`
`41`	`41`	`}`
`42`	`42`