Data erasure remove waiting shards (#16313)

Author: molotkov-and

ydb/core/tx/schemeshard/schemeshard__delete_tablet_reply.cpp

@@ -180,7 +180,7 @@ struct TSchemeShard::TTxDeleteTabletReply : public TSchemeShard::TRwTxBase {
                             "Close pipe to deleted shardIdx " << ShardIdx << " tabletId " << TabletId);
                 Self->PipeClientCache->ForceClose(ctx, ui64(TabletId));
             }
-            if (Self->DataErasureManager->GetStatus() == EDataErasureStatus::IN_PROGRESS) {
+            if (Self->EnableDataErasure && Self->DataErasureManager->GetStatus() == EDataErasureStatus::IN_PROGRESS) {
                 Self->Execute(Self->CreateTxCancelDataErasureShards({ShardIdx}));
             }
         }

ydb/core/tx/schemeshard/schemeshard__root_data_erasure_manager.cpp

@@ -313,22 +313,14 @@ void TRootDataErasureManager::OnDone(const TPathId& pathId, NIceDb::TNiceDb& db)
     ActivePipes.erase(pathId);
     auto it = WaitingDataErasureTenants.find(pathId);
     if (it != WaitingDataErasureTenants.end()) {
-        it->second = EDataErasureStatus::COMPLETED;
-        db.Table<Schema::WaitingDataErasureTenants>().Key(pathId.OwnerId, pathId.LocalPathId).Update<Schema::WaitingDataErasureTenants::Status>(it->second);
+        db.Table<Schema::WaitingDataErasureTenants>().Key(pathId.OwnerId, pathId.LocalPathId).Delete();
+        WaitingDataErasureTenants.erase(it);
     }
 
     SchemeShard->TabletCounters->Cumulative()[COUNTER_DATA_ERASURE_OK].Increment(1);
     UpdateMetrics();
 
-    bool isDataErasureCompleted = true;
-    for (const auto& [pathId, status] : WaitingDataErasureTenants) {
-        if (status == EDataErasureStatus::IN_PROGRESS) {
-            isDataErasureCompleted = false;
-            break;
-        }
-    }
-
-    if (isDataErasureCompleted) {
+    if (WaitingDataErasureTenants.empty()) {
         LOG_INFO_S(ctx, NKikimrServices::FLAT_TX_SCHEMESHARD,
             "[RootDataErasureManager] Data erasure in tenants is completed. Send request to BS controller");
         Status = EDataErasureStatus::IN_PROGRESS_BSC;
@@ -465,16 +457,10 @@ bool TRootDataErasureManager::Remove(const TPathId& pathId) {
     if (it != WaitingDataErasureTenants.end()) {
         Queue->Remove(pathId);
         ActivePipes.erase(pathId);
-        WaitingDataErasureTenants[pathId] = EDataErasureStatus::COMPLETED;
-        bool isDataErasureCompleted = true;
-        for (const auto& [pathId, status] : WaitingDataErasureTenants) {
-            if (status == EDataErasureStatus::IN_PROGRESS) {
-                isDataErasureCompleted = false;
-                break;
-            }
-        }
+        WaitingDataErasureTenants.erase(it);
 
-        if (isDataErasureCompleted) {
+        if (WaitingDataErasureTenants.empty()) {
+            Status = EDataErasureStatus::IN_PROGRESS_BSC;
             SendRequestToBSC();
         }
         return true;

ydb/core/tx/schemeshard/schemeshard__tenant_data_erasure_manager.cpp

@@ -114,7 +114,7 @@ void TTenantDataErasureManager::Run(NIceDb::TNiceDb& db) {
     Status = EDataErasureStatus::IN_PROGRESS;
     for (const auto& [shardIdx, shardInfo] : SchemeShard->ShardInfos) {
         if (shardInfo.TabletType == ETabletType::DataShard) {
-            Enqueue(shardIdx); // forward generation
+            Enqueue(shardIdx);
             WaitingDataErasureShards[shardIdx] = EDataErasureStatus::IN_PROGRESS;
             db.Table<Schema::WaitingDataErasureShards>().Key(shardIdx.GetOwnerId(), shardIdx.GetLocalId()).Update<Schema::WaitingDataErasureShards::Status>(WaitingDataErasureShards[shardIdx]);
         }
@@ -291,21 +291,15 @@ void TTenantDataErasureManager::OnDone(const TTabletId& tabletId, NIceDb::TNiceD
     {
         auto it = WaitingDataErasureShards.find(shardIdx);
         if (it != WaitingDataErasureShards.end()) {
-            it->second = EDataErasureStatus::COMPLETED;
+            db.Table<Schema::WaitingDataErasureShards>().Key(shardIdx.GetOwnerId(), shardIdx.GetLocalId()).Delete();
+            WaitingDataErasureShards.erase(it);
         }
-        db.Table<Schema::WaitingDataErasureShards>().Key(shardIdx.GetOwnerId(), shardIdx.GetLocalId()).Update<Schema::WaitingDataErasureShards::Status>(it->second);
     }
 
     SchemeShard->TabletCounters->Cumulative()[COUNTER_TENANT_DATA_ERASURE_OK].Increment(1);
     UpdateMetrics();
 
-    bool isTenantDataErasureCompleted = true;
-    for (const auto& [shardIdx, status] : WaitingDataErasureShards) {
-        if (status == EDataErasureStatus::IN_PROGRESS) {
-            isTenantDataErasureCompleted = false;
-        }
-    }
-    if (isTenantDataErasureCompleted) {
+    if (WaitingDataErasureShards.empty()) {
         LOG_INFO_S(ctx, NKikimrServices::FLAT_TX_SCHEMESHARD,
             "[TenantDataErasureManager] Data erasure in shards is completed. Send response to root schemeshard");
         Complete();
@@ -399,14 +393,8 @@ bool TTenantDataErasureManager::Remove(const TShardIdx& shardIdx) {
     if (it != WaitingDataErasureShards.end()) {
         Queue->Remove(shardIdx);
         ActivePipes.erase(shardIdx);
-        it->second = EDataErasureStatus::COMPLETED;
-        bool isTenantDataErasureCompleted = true;
-        for (const auto& [shardIdx, status] : WaitingDataErasureShards) {
-            if (status == EDataErasureStatus::IN_PROGRESS) {
-                isTenantDataErasureCompleted = false;
-            }
-        }
-        if (isTenantDataErasureCompleted) {
+        WaitingDataErasureShards.erase(it);
+        if (WaitingDataErasureShards.empty()) {
             auto ctx = SchemeShard->ActorContext();
             LOG_INFO_S(ctx, NKikimrServices::FLAT_TX_SCHEMESHARD,
                 "[TenantDataErasureManager] [Remove] Data erasure in shards is completed. Send response to root schemeshard");
@@ -620,7 +608,7 @@ struct TSchemeShard::TTxCancelDataErasureShards : public TSchemeShard::TRwTxBase
         NIceDb::TNiceDb db(txc.DB);
         for (const auto& shard : DataErasureShards) {
             if (Self->DataErasureManager->Remove(shard)) {
-                db.Table<Schema::WaitingDataErasureShards>().Key(shard.GetOwnerId(), shard.GetLocalId()).Update<Schema::WaitingDataErasureShards::Status>(EDataErasureStatus::COMPLETED);
+                db.Table<Schema::WaitingDataErasureShards>().Key(shard.GetOwnerId(), shard.GetLocalId()).Delete();
             }
         }
         if (Self->DataErasureManager->GetStatus() == EDataErasureStatus::COMPLETED) {

ydb/core/tx/schemeshard/schemeshard_impl.cpp

@@ -1362,7 +1362,7 @@ bool TSchemeShard::CheckApplyIf(const NKikimrSchemeOp::TModifyScheme& scheme, TS
 
             if (item.HasLockedTxId()) {
                 const auto lockOwnerTxId = TTxId(item.GetLockedTxId());
-    
+
                 TString lockErr = "fail user constraint in ApplyIf section:";
                 if (!CheckLocks(pathId, lockOwnerTxId, lockErr)) {
                     errStr = lockErr;
@@ -2295,7 +2295,7 @@ void TSchemeShard::PersistRemoveSubDomain(NIceDb::TNiceDb& db, const TPathId& pa
         }
 
         if (DataErasureManager->Remove(pathId)) {
-            db.Table<Schema::WaitingDataErasureTenants>().Key(pathId.OwnerId, pathId.LocalPathId).Update<Schema::WaitingDataErasureTenants::Status>(EDataErasureStatus::COMPLETED);
+            db.Table<Schema::WaitingDataErasureTenants>().Key(pathId.OwnerId, pathId.LocalPathId).Delete();
         }
 
         db.Table<Schema::SubDomains>().Key(pathId.LocalPathId).Delete();
@@ -7150,7 +7150,7 @@ void TSchemeShard::SetPartitioning(TPathId pathId, TTableInfo::TPtr tableInfo, T
         newPartitioningSet.reserve(newPartitioning.size());
         const auto& oldPartitioning = tableInfo->GetPartitions();
 
-        std::vector<TShardIdx> dataErasureShards;
+        std::vector<TShardIdx> newDataErasureShards;
         for (const auto& p: newPartitioning) {
             if (!oldPartitioning.empty())
                 newPartitioningSet.insert(p.ShardIdx);
@@ -7160,11 +7160,11 @@ void TSchemeShard::SetPartitioning(TPathId pathId, TTableInfo::TPtr tableInfo, T
             if (it != partitionStats.end()) {
                 EnqueueBackgroundCompaction(p.ShardIdx, it->second);
                 UpdateShardMetrics(p.ShardIdx, it->second);
-                dataErasureShards.push_back(p.ShardIdx);
+                newDataErasureShards.push_back(p.ShardIdx);
             }
         }
-        if (DataErasureManager->GetStatus() == EDataErasureStatus::IN_PROGRESS) {
-            Execute(CreateTxAddEntryToDataErasure(dataErasureShards), this->ActorContext());
+        if (EnableDataErasure && DataErasureManager->GetStatus() == EDataErasureStatus::IN_PROGRESS) {
+            Execute(CreateTxAddEntryToDataErasure(newDataErasureShards), this->ActorContext());
         }
 
         for (const auto& p: oldPartitioning) {