[25-2] Do not crash on unknown enum values (state, build kind) during index build (#26228)

Author: vitalif

ydb/core/tx/schemeshard/schemeshard_info_types.cpp

@@ -2440,6 +2440,47 @@ void TIndexBuildInfo::AddParent(const TSerializedTableRange& range, TShardIdx sh
     Cluster2Shards.emplace_hint(itFrom, parentTo, TClusterShards{.From = parentFrom, .Shards = {shard}});
 }
 
+bool TIndexBuildInfo::IsValidState(EState value)
+{
+    switch (value) {
+        case EState::Invalid:
+        case EState::AlterMainTable:
+        case EState::Locking:
+        case EState::GatheringStatistics:
+        case EState::Initiating:
+        case EState::Filling:
+        case EState::DropBuild:
+        case EState::CreateBuild:
+        case EState::LockBuild:
+        case EState::Applying:
+        case EState::Unlocking:
+        case EState::Done:
+        case EState::Cancellation_Applying:
+        case EState::Cancellation_Unlocking:
+        case EState::Cancellation_DroppingColumns:
+        case EState::Cancelled:
+        case EState::Rejection_Applying:
+        case EState::Rejection_Unlocking:
+        case EState::Rejection_DroppingColumns:
+        case EState::Rejected:
+            return true;
+    }
+    return false;
+}
+
+bool TIndexBuildInfo::IsValidBuildKind(EBuildKind value)
+{
+    switch (value) {
+        case EBuildKind::BuildKindUnspecified:
+        case EBuildKind::BuildSecondaryIndex:
+        case EBuildKind::BuildVectorIndex:
+        case EBuildKind::BuildPrefixedVectorIndex:
+        case EBuildKind::BuildColumns:
+            return true;
+    }
+    return false;
+}
+
 TColumnFamiliesMerger::TColumnFamiliesMerger(NKikimrSchemeOp::TPartitionConfig &container)
     : Container(container)
     , DeduplicationById(TPartitionConfigMerger::DeduplicateColumnFamiliesById(Container))

ydb/core/tx/schemeshard/schemeshard_info_types.h

@@ -3290,6 +3290,9 @@ struct TIndexBuildInfo: public TSimpleRefCount<TIndexBuildInfo> {
         return result;
     }
 
+    static bool IsValidState(EState value);
+    static bool IsValidBuildKind(EBuildKind value);
+
     struct TClusterShards {
         NTableIndex::TClusterId From = std::numeric_limits<NTableIndex::TClusterId>::max();
         std::vector<TShardIdx> Shards;
@@ -3347,16 +3350,27 @@ struct TIndexBuildInfo: public TSimpleRefCount<TIndexBuildInfo> {
         indexInfo->Id = id;
         indexInfo->Uid = uid;
 
-        indexInfo->State = TIndexBuildInfo::EState(
-            row.template GetValue<Schema::IndexBuild::State>());
         indexInfo->Issue =
             row.template GetValueOrDefault<Schema::IndexBuild::Issue>();
 
+        indexInfo->State = TIndexBuildInfo::EState(
+            row.template GetValue<Schema::IndexBuild::State>());
+        if (!IsValidState(indexInfo->State)) {
+            indexInfo->IsBroken = true;
+            indexInfo->AddIssue(TStringBuilder() << "Unknown build state: " << ui32(indexInfo->State));
+            indexInfo->State = TIndexBuildInfo::EState::Invalid;
+        }
+
         // note: please note that here we specify BuildSecondaryIndex as operation default,
         // because previously this table was dedicated for build secondary index operations only.
         indexInfo->BuildKind = TIndexBuildInfo::EBuildKind(
             row.template GetValueOrDefault<Schema::IndexBuild::BuildKind>(
                 ui32(TIndexBuildInfo::EBuildKind::BuildSecondaryIndex)));
+        if (!IsValidBuildKind(indexInfo->BuildKind)) {
+            indexInfo->IsBroken = true;
+            indexInfo->AddIssue(TStringBuilder() << "Unknown build kind: " << ui32(indexInfo->BuildKind));
+            indexInfo->BuildKind = TIndexBuildInfo::EBuildKind::BuildKindUnspecified;
+        }
 
         indexInfo->DomainPathId =
             TPathId(row.template GetValue<Schema::IndexBuild::DomainOwnerId>(),

ydb/core/tx/schemeshard/ut_index_build/ut_vector_index_build.cpp

@@ -1332,4 +1332,108 @@ Y_UNIT_TEST_SUITE(VectorIndexBuildTest) {
             UNIT_ASSERT_STRING_CONTAINS(buildIndexOperation.DebugString(), "Processed: UploadRows: 0 UploadBytes: 0 ReadRows: 0 ReadBytes: 0");
         }
     }
+
+    Y_UNIT_TEST(UnknownState) {
+        TTestBasicRuntime runtime;
+        TTestEnv env(runtime);
+        ui64 txId = 100;
+
+        runtime.SetLogPriority(NKikimrServices::TX_DATASHARD, NLog::PRI_TRACE);
+        runtime.SetLogPriority(NKikimrServices::BUILD_INDEX, NLog::PRI_TRACE);
+
+        TestCreateTable(runtime, ++txId, "/MyRoot", R"(
+            Name: "vectors"
+            Columns { Name: "id" Type: "Uint64" }
+            Columns { Name: "embedding" Type: "String" }
+            KeyColumnNames: [ "id" ]
+        )");
+        env.TestWaitNotification(runtime, txId);
+
+        NYdb::NTable::TGlobalIndexSettings globalIndexSettings;
+
+        std::unique_ptr<NYdb::NTable::TKMeansTreeSettings> kmeansTreeSettings;
+        {
+            Ydb::Table::KMeansTreeSettings proto;
+            UNIT_ASSERT(google::protobuf::TextFormat::ParseFromString(R"(
+                settings {
+                    metric: DISTANCE_COSINE
+                    vector_type: VECTOR_TYPE_FLOAT
+                    vector_dimension: 1024
+                }
+                levels: 5
+                clusters: 4
+            )", &proto));
+            using T = NYdb::NTable::TKMeansTreeSettings;
+            kmeansTreeSettings = std::make_unique<T>(T::FromProto(proto));
+        }
+
+        TBlockEvents<TEvDataShard::TEvLocalKMeansRequest> kmeansBlocker(runtime, [&](const auto& ) {
+            return true;
+        });
+
+        const ui64 buildIndexTx = ++txId;
+        AsyncBuildVectorIndex(runtime, buildIndexTx, TTestTxConfig::SchemeShard, "/MyRoot", "/MyRoot/vectors", "index1", {"embedding"});
+
+        runtime.WaitFor("LocalKMeansRequest", [&]{ return kmeansBlocker.size(); });
+
+        {
+            // set unknown State value
+            TString writeQuery = Sprintf(R"(
+                (
+                    (let key '( '('Id (Uint64 '%lu)) ) )
+                    (let value '('('State (Uint32 '999999)) ) )
+                    (return (AsList (UpdateRow 'IndexBuild key value) ))
+                )
+            )", buildIndexTx);
+            NKikimrMiniKQL::TResult result;
+            TString err;
+            NKikimrProto::EReplyStatus status = LocalMiniKQL(runtime, TTestTxConfig::SchemeShard, writeQuery, result, err);
+            UNIT_ASSERT_VALUES_EQUAL_C(status, NKikimrProto::EReplyStatus::OK, err);
+        }
+
+        Cerr << "... rebooting scheme shard" << Endl;
+        RebootTablet(runtime, TTestTxConfig::SchemeShard, runtime.AllocateEdgeActor());
+        kmeansBlocker.Stop().Unblock();
+
+        {
+            auto buildIndexOperation = TestGetBuildIndex(runtime, TTestTxConfig::SchemeShard, "/MyRoot", buildIndexTx);
+            auto buildIndexHtml = TestGetBuildIndexHtml(runtime, TTestTxConfig::SchemeShard, buildIndexTx);
+            UNIT_ASSERT_VALUES_EQUAL_C(
+                buildIndexOperation.GetIndexBuild().GetState(), Ydb::Table::IndexBuildState::STATE_UNSPECIFIED,
+                buildIndexOperation.DebugString()
+            );
+            UNIT_ASSERT_STRING_CONTAINS(buildIndexOperation.DebugString(), "Unknown build state");
+            UNIT_ASSERT_STRING_CONTAINS(buildIndexHtml, "IsBroken: 1");
+        }
+
+        {
+            // set a known State but unknown BuildKind
+            TString writeQuery = Sprintf(R"(
+                (
+                    (let key '( '('Id (Uint64 '%lu)) ) )
+                    (let value '('('State (Uint32 '40)) '('BuildKind (Uint32 '999999)) ) )
+                    (return (AsList (UpdateRow 'IndexBuild key value) ))
+                )
+            )", buildIndexTx);
+            NKikimrMiniKQL::TResult result;
+            TString err;
+            NKikimrProto::EReplyStatus status = LocalMiniKQL(runtime, TTestTxConfig::SchemeShard, writeQuery, result, err);
+            UNIT_ASSERT_VALUES_EQUAL_C(status, NKikimrProto::EReplyStatus::OK, err);
+        }
+
+        Cerr << "... rebooting scheme shard" << Endl;
+        RebootTablet(runtime, TTestTxConfig::SchemeShard, runtime.AllocateEdgeActor());
+
+        {
+            auto buildIndexOperation = TestGetBuildIndex(runtime, TTestTxConfig::SchemeShard, "/MyRoot", buildIndexTx);
+            auto buildIndexHtml = TestGetBuildIndexHtml(runtime, TTestTxConfig::SchemeShard, buildIndexTx);
+            UNIT_ASSERT_VALUES_EQUAL_C(
+                buildIndexOperation.GetIndexBuild().GetState(), Ydb::Table::IndexBuildState::STATE_TRANSFERING_DATA,
+                buildIndexOperation.DebugString()
+            );
+            UNIT_ASSERT_STRING_CONTAINS(buildIndexOperation.DebugString(), "Unknown build kind");
+            UNIT_ASSERT_STRING_CONTAINS(buildIndexHtml, "IsBroken: 1");
+        }
+    }
+
 }