Datashard: DISK_SPACE_EXHAUSTED error (#8318)

ydb-platform · Aug 27, 2024 · ac92ce0 · ac92ce0
1 parent cb032f1
commit ac92ce0
Show file tree

Hide file tree

Showing 9 changed files with 32 additions and 7 deletions.
diff --git a/ydb/core/kqp/executer_actor/kqp_data_executer.cpp b/ydb/core/kqp/executer_actor/kqp_data_executer.cpp
@@ -842,6 +842,7 @@ class TKqpDataExecuter : public TKqpExecuterBase<TKqpDataExecuter, EExecType::Da
             case NKikimrDataEvents::TEvWriteResult::STATUS_ABORTED: {
                 return ReplyErrorAndDie(Ydb::StatusIds::ABORTED, issues);
             }
+            case NKikimrDataEvents::TEvWriteResult::STATUS_DISK_SPACE_EXHAUSTED:
             case NKikimrDataEvents::TEvWriteResult::STATUS_INTERNAL_ERROR: {
                 return ReplyErrorAndDie(Ydb::StatusIds::INTERNAL_ERROR, issues);
             }

diff --git a/ydb/core/kqp/runtime/kqp_write_actor.cpp b/ydb/core/kqp/runtime/kqp_write_actor.cpp
@@ -425,6 +425,20 @@ class TKqpDirectWriteActor : public TActorBootstrapped<TKqpDirectWriteActor>, pu
             }
             return;
         }
+        case NKikimrDataEvents::TEvWriteResult::STATUS_DISK_SPACE_EXHAUSTED: {
+            CA_LOG_E("Got DISK_SPACE_EXHAUSTED for table `"
+                    << SchemeEntry->TableId.PathId.ToString() << "`."
+                    << " ShardID=" << ev->Get()->Record.GetOrigin() << ","
+                    << " Sink=" << this->SelfId() << "."
+                    << getIssues().ToOneLineString());
+
+        RuntimeError(
+            TStringBuilder() << "Got DISK_SPACE_EXHAUSTED for table `"
+                << SchemeEntry->TableId.PathId.ToString() << "`.",
+            NYql::NDqProto::StatusIds::PRECONDITION_FAILED,
+            getIssues());
+            return;
+        }        
         case NKikimrDataEvents::TEvWriteResult::STATUS_OVERLOADED: {
             CA_LOG_W("Got OVERLOADED for table `"
                 << SchemeEntry->TableId.PathId.ToString() << "`."

diff --git a/ydb/core/protos/counters_datashard.proto b/ydb/core/protos/counters_datashard.proto
@@ -147,6 +147,8 @@ enum ECumulativeCounters {
     COUNTER_WRITE_CANCELLED = 109                         [(CounterOpts) = {Name: "WriteCancelled"}];
     COUNTER_WRITE_ROWS = 110                              [(CounterOpts) = {Name: "WriteRows"}];
     COUNTER_WRITE_BYTES = 111                             [(CounterOpts) = {Name: "WriteBytes"}];
+    COUNTER_WRITE_DISK_SPACE_EXHAUSTED = 112              [(CounterOpts) = {Name: "WriteDiskSpaceExhausted"}];
+    COUNTER_PREPARE_DISK_SPACE_EXHAUSTED = 113            [(CounterOpts) = {Name: "PrepareSpaceExhausted"}];
 }
 
 enum EPercentileCounters {

diff --git a/ydb/core/protos/data_events.proto b/ydb/core/protos/data_events.proto
@@ -126,6 +126,7 @@ message TEvWriteResult {
         STATUS_BAD_REQUEST = 7;
         STATUS_SCHEME_CHANGED = 8;
         STATUS_LOCKS_BROKEN = 9;
+        STATUS_DISK_SPACE_EXHAUSTED = 10;
     }
 
     // Status

diff --git a/ydb/core/tx/datashard/check_data_tx_unit.cpp b/ydb/core/tx/datashard/check_data_tx_unit.cpp
@@ -212,9 +212,9 @@ EExecutionStatus TCheckDataTxUnit::Execute(TOperation::TPtr op,
                             // Updates are not allowed when database is out of space
                             TString err = "Cannot perform writes: database is out of disk space";
 
-                            DataShard.IncCounter(COUNTER_PREPARE_OUT_OF_SPACE);
+                            DataShard.IncCounter(COUNTER_PREPARE_DISK_SPACE_EXHAUSTED);
 
-                            BuildResult(op)->AddError(NKikimrTxDataShard::TError::OUT_OF_SPACE, err);
+                            BuildResult(op)->AddError(NKikimrTxDataShard::TError::DISK_SPACE_EXHAUSTED, err);
                             op->Abort(EExecutionUnitKind::FinishPropose);
 
                             LOG_LOG_S_THROTTLE(DataShard.GetLogThrottler(TDataShard::ELogThrottlerType::CheckDataTxUnit_Execute), ctx, NActors::NLog::PRI_ERROR, NKikimrServices::TX_DATASHARD, err);

diff --git a/ydb/core/tx/datashard/check_write_unit.cpp b/ydb/core/tx/datashard/check_write_unit.cpp
@@ -88,9 +88,9 @@ EExecutionStatus TCheckWriteUnit::Execute(TOperation::TPtr op,
                             // Updates are not allowed when database is out of space
                             TString err = "Cannot perform writes: database is out of disk space";
 
-                            DataShard.IncCounter(COUNTER_WRITE_OUT_OF_SPACE);
+                            DataShard.IncCounter(COUNTER_WRITE_DISK_SPACE_EXHAUSTED);
 
-                            writeOp->SetError(NKikimrDataEvents::TEvWriteResult::STATUS_OVERLOADED, err);
+                            writeOp->SetError(NKikimrDataEvents::TEvWriteResult::STATUS_DISK_SPACE_EXHAUSTED, err);
                             op->Abort(EExecutionUnitKind::FinishProposeWrite);
 
                             DataShard.SetOverloadSubscribed(writeOp->GetWriteTx()->GetOverloadSubscribe(), writeOp->GetRecipient(), op->GetTarget(), ERejectReasons::YellowChannels, writeOp->GetWriteResult()->Record);

diff --git a/ydb/core/tx/datashard/datashard__op_rows.cpp b/ydb/core/tx/datashard/datashard__op_rows.cpp
@@ -203,7 +203,7 @@ static bool MaybeReject(TDataShard* self, TEvRequest& ev, const TActorContext& c
             Reject<TEvResponse, TEvRequest>(self, ev, txDesc, rejectReasons, rejectDescription, &OutOfSpace, ctx, logThrottlerType);
             return true;
         } else if (self->IsSubDomainOutOfSpace()) {
-            self->IncCounter(COUNTER_PREPARE_OUT_OF_SPACE);
+            self->IncCounter(COUNTER_PREPARE_DISK_SPACE_EXHAUSTED);
             rejectReasons = ERejectReasons::DiskSpace;
             rejectDescription = "Cannot perform writes: database is out of disk space";
             Reject<TEvResponse, TEvRequest>(self, ev, txDesc, rejectReasons, rejectDescription, &DiskSpaceExhausted, ctx, logThrottlerType);

diff --git a/ydb/core/tx/datashard/datashard__write.cpp b/ydb/core/tx/datashard/datashard__write.cpp
@@ -293,8 +293,9 @@ NKikimrDataEvents::TEvWriteResult::EStatus NEvWrite::TConvertor::ConvertErrCode(
         case NKikimrTxDataShard::TError_EKind_SCHEME_CHANGED:
             return NKikimrDataEvents::TEvWriteResult::STATUS_SCHEME_CHANGED;
         case NKikimrTxDataShard::TError_EKind_OUT_OF_SPACE:
-        case NKikimrTxDataShard::TError_EKind_DISK_SPACE_EXHAUSTED:
             return NKikimrDataEvents::TEvWriteResult::STATUS_OVERLOADED;
+        case NKikimrTxDataShard::TError_EKind_DISK_SPACE_EXHAUSTED:
+            return NKikimrDataEvents::TEvWriteResult::STATUS_DISK_SPACE_EXHAUSTED;
         default:
             return NKikimrDataEvents::TEvWriteResult::STATUS_INTERNAL_ERROR;
     }

diff --git a/ydb/core/tx/datashard/datashard_impl.h b/ydb/core/tx/datashard/datashard_impl.h
@@ -3276,7 +3276,13 @@ class TDataShard
             ev->Record.MutableTableStats()->SetImmediateTxCompleted(TabletCounters->Cumulative()[COUNTER_PREPARE_IMMEDIATE].Get() + TabletCounters->Cumulative()[COUNTER_WRITE_IMMEDIATE].Get());
             ev->Record.MutableTableStats()->SetPlannedTxCompleted(TabletCounters->Cumulative()[COUNTER_PLANNED_TX_COMPLETE].Get());
             ev->Record.MutableTableStats()->SetTxRejectedByOverload(TabletCounters->Cumulative()[COUNTER_PREPARE_OVERLOADED].Get() + TabletCounters->Cumulative()[COUNTER_WRITE_OVERLOADED].Get());
-            ev->Record.MutableTableStats()->SetTxRejectedBySpace(TabletCounters->Cumulative()[COUNTER_PREPARE_OUT_OF_SPACE].Get() + TabletCounters->Cumulative()[COUNTER_WRITE_OUT_OF_SPACE].Get());
+            ev->Record.MutableTableStats()->SetTxRejectedBySpace(
+                TabletCounters->Cumulative()[COUNTER_PREPARE_OUT_OF_SPACE].Get() 
+              + TabletCounters->Cumulative()[COUNTER_PREPARE_DISK_SPACE_EXHAUSTED].Get()
+              + TabletCounters->Cumulative()[COUNTER_WRITE_OUT_OF_SPACE].Get()
+              + TabletCounters->Cumulative()[COUNTER_WRITE_DISK_SPACE_EXHAUSTED].Get()
+            );
+
             ev->Record.MutableTableStats()->SetTxCompleteLagMsec(TabletCounters->Simple()[COUNTER_TX_COMPLETE_LAG].Get());
             ev->Record.MutableTableStats()->SetInFlightTxCount(TabletCounters->Simple()[COUNTER_TX_IN_FLY].Get()
                 + TabletCounters->Simple()[COUNTER_IMMEDIATE_TX_IN_FLY].Get());