24-3: Index build: do not lose the requested partitioning settings of an indexImplTable in case of SchemeShard reboots (#10579)

ydb/core/tx/schemeshard/schemeshard_build_index.cpp

@@ -34,7 +34,8 @@ void TSchemeShard::Handle(TEvPrivate::TEvIndexBuildingMakeABill::TPtr& ev, const
 
 void TSchemeShard::PersistCreateBuildIndex(NIceDb::TNiceDb& db, const TIndexBuildInfo::TPtr info) {
     Y_ABORT_UNLESS(info->BuildKind != TIndexBuildInfo::EBuildKind::BuildKindUnspecified);
-    db.Table<Schema::IndexBuild>().Key(info->Id).Update(
+    auto persistedBuildIndex = db.Table<Schema::IndexBuild>().Key(info->Id);
+    persistedBuildIndex.Update(
         NIceDb::TUpdate<Schema::IndexBuild::Uid>(info->Uid),
         NIceDb::TUpdate<Schema::IndexBuild::DomainOwnerId>(info->DomainPathId.OwnerId),
         NIceDb::TUpdate<Schema::IndexBuild::DomainLocalId>(info->DomainPathId.LocalPathId),
@@ -48,6 +49,17 @@ void TSchemeShard::PersistCreateBuildIndex(NIceDb::TNiceDb& db, const TIndexBuil
         NIceDb::TUpdate<Schema::IndexBuild::MaxRetries>(info->Limits.MaxRetries),
         NIceDb::TUpdate<Schema::IndexBuild::BuildKind>(ui32(info->BuildKind))
     );
+    // Persist details of the index build operation: ImplTableDescription.
+    // We have chosen TIndexCreationConfig's string representation as the serialization format.
+    {
+        NKikimrSchemeOp::TIndexCreationConfig serializableRepresentation;
+
+        *serializableRepresentation.MutableIndexImplTableDescription() = info->ImplTableDescription;
+
+        persistedBuildIndex.Update(
+            NIceDb::TUpdate<Schema::IndexBuild::CreationConfig>(serializableRepresentation.SerializeAsString())
+        );
+    }
 
     ui32 columnNo = 0;
     for (ui32 i = 0; i < info->IndexColumns.size(); ++i, ++columnNo) {

ydb/core/tx/schemeshard/schemeshard_info_types.h

@@ -3092,6 +3092,14 @@ struct TIndexBuildInfo: public TSimpleRefCount<TIndexBuildInfo> {
         indexInfo->IndexName = row.template GetValue<Schema::IndexBuild::IndexName>();
         indexInfo->IndexType = row.template GetValue<Schema::IndexBuild::IndexType>();
 
+        // Restore the operation details: ImplTableDescription.
+        if (row.template HaveValue<Schema::IndexBuild::CreationConfig>()) {
+            NKikimrSchemeOp::TIndexCreationConfig creationConfig;
+            Y_ABORT_UNLESS(creationConfig.ParseFromString(row.template GetValue<Schema::IndexBuild::CreationConfig>()));
+
+            indexInfo->ImplTableDescription = std::move(*creationConfig.MutableIndexImplTableDescription());
+        }
+
         indexInfo->State = TIndexBuildInfo::EState(
             row.template GetValue<Schema::IndexBuild::State>());
         indexInfo->Issue =

ydb/core/tx/schemeshard/schemeshard_schema.h

@@ -1318,6 +1318,9 @@ struct Schema : NIceDb::Schema {
         struct AlterMainTableTxStatus : Column<32, NScheme::NTypeIds::Uint32> { using Type = NKikimrScheme::EStatus; };
         struct AlterMainTableTxDone : Column<33, NScheme::NTypeIds::Bool> {};
 
+        // Serialized as string NKikimrSchemeOp::TIndexCreationConfig protobuf.
+        struct CreationConfig : Column<34, NScheme::NTypeIds::String> { using Type = TString; };
+
         using TKey = TableKey<Id>;
         using TColumns = TableColumns<
             Id,
@@ -1352,7 +1355,8 @@ struct Schema : NIceDb::Schema {
             BuildKind,
             AlterMainTableTxId,
             AlterMainTableTxStatus,
-            AlterMainTableTxDone
+            AlterMainTableTxDone,
+            CreationConfig
         >;
     };
 

ydb/core/tx/schemeshard/ut_helpers/ls_checks.cpp

@@ -1252,11 +1252,63 @@ TCheckFunc PartitionKeys(TVector<TString> lastShardKeys) {
         const auto& pathDescr = record.GetPathDescription();
         UNIT_ASSERT_VALUES_EQUAL(lastShardKeys.size(), pathDescr.TablePartitionsSize());
         for (size_t i = 0; i < lastShardKeys.size(); ++i) {
-            UNIT_ASSERT_STRING_CONTAINS(pathDescr.GetTablePartitions(i).GetEndOfRangeKeyPrefix(), lastShardKeys[i]);
+            const auto& partition = pathDescr.GetTablePartitions(i);
+            UNIT_ASSERT_STRING_CONTAINS_C(
+                partition.GetEndOfRangeKeyPrefix(), lastShardKeys[i],
+                "partition index: " << i << '\n'
+                    << "actual key prefix: " << partition.GetEndOfRangeKeyPrefix().Quote() << '\n'
+                    << "expected key prefix: " << lastShardKeys[i].Quote() << '\n'
+            );
         }
     };
 }
 
+namespace {
+
+// Serializes / deserializes a value of type T to a cell vector string representation.
+template <typename T>
+struct TSplitBoundarySerializer {
+    static TString Serialize(T splitBoundary) {
+        const auto cell = TCell::Make(splitBoundary);
+        TSerializedCellVec cellVec(TArrayRef<const TCell>(&cell, 1));
+        return cellVec.ReleaseBuffer();
+    }
+
+    static TVector<T> Deserialize(const TString& serializedCells) {
+        TSerializedCellVec cells(serializedCells);
+        TVector<T> values;
+        for (const auto& cell : cells.GetCells()) {
+            if (cell.IsNull()) {
+                // the last cell
+                break;
+            }
+            values.emplace_back(cell.AsValue<T>());
+        }
+        return values;
+    }
+};
+
+}
+
+template <typename T>
+TCheckFunc SplitBoundaries(TVector<T>&& expectedBoundaries) {
+    return [expectedBoundaries = std::move(expectedBoundaries)] (const NKikimrScheme::TEvDescribeSchemeResult& record) {
+        const auto& pathDescr = record.GetPathDescription();
+        UNIT_ASSERT_VALUES_EQUAL(pathDescr.TablePartitionsSize(), expectedBoundaries.size() + 1);
+        for (size_t i = 0; i < expectedBoundaries.size(); ++i) {
+            const auto& partition = pathDescr.GetTablePartitions(i);
+            const auto actualBoundary = TSplitBoundarySerializer<T>::Deserialize(partition.GetEndOfRangeKeyPrefix()).at(0);
+            UNIT_ASSERT_VALUES_EQUAL_C(
+                actualBoundary, expectedBoundaries[i],
+                "partition index: " << i << '\n'
+                    << "actual key prefix: " << partition.GetEndOfRangeKeyPrefix().Quote() << '\n'
+            );
+        }
+    };
+}
+
+template TCheckFunc SplitBoundaries<ui32>(TVector<ui32>&&);
+
 TCheckFunc ServerlessComputeResourcesMode(NKikimrSubDomains::EServerlessComputeResourcesMode serverlessComputeResourcesMode) {
     return [=] (const NKikimrScheme::TEvDescribeSchemeResult& record) {
         UNIT_ASSERT_C(IsGoodDomainStatus(record.GetStatus()), "Unexpected status: " << record.GetStatus());

ydb/core/tx/schemeshard/ut_helpers/ls_checks.h

@@ -99,6 +99,10 @@ namespace NLs {
     void CheckBoundaries(const NKikimrScheme::TEvDescribeSchemeResult& record);
     TCheckFunc PartitionCount(ui32 count);
     TCheckFunc PartitionKeys(TVector<TString> lastShardKeys);
+    // Checks if the serialized representation of an expected boundary is a prefix of the actual one.
+    // Similar to PartitionKeys check, but does not require you to pass split boundaries in a serialized form.
+    template <typename T>
+    TCheckFunc SplitBoundaries(TVector<T>&& expectedBoundaries);
     TCheckFunc FollowerCount(ui32 count);
     TCheckFunc CrossDataCenterFollowerCount(ui32 count);
     TCheckFunc AllowFollowerPromotion(bool val);

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

@@ -910,6 +910,78 @@ Y_UNIT_TEST_SUITE(IndexBuildTest) {
         }
     }
 
+    Y_UNIT_TEST(IndexPartitioningIsPersisted) {
+        TTestBasicRuntime runtime;
+        TTestEnv env(runtime);
+        ui64 txId = 100;
+
+        TestCreateTable(runtime, ++txId, "/MyRoot", R"(
+            Name: "Table"
+            Columns { Name: "key" Type: "Uint64" }
+            Columns { Name: "value" Type: "Utf8" }
+            KeyColumnNames: [ "key" ]
+        )");
+        env.TestWaitNotification(runtime, txId);
+
+        Ydb::Table::GlobalIndexSettings settings;
+        UNIT_ASSERT(google::protobuf::TextFormat::ParseFromString(R"(
+            partition_at_keys {
+                split_points {
+                    type { tuple_type { elements { optional_type { item { type_id: UTF8 } } } } }
+                    value { items { text_value: "alice" } }
+                }
+                split_points {
+                    type { tuple_type { elements { optional_type { item { type_id: UTF8 } } } } }
+                    value { items { text_value: "bob" } }
+                }
+            }
+            partitioning_settings {
+                min_partitions_count: 3
+                max_partitions_count: 3
+            }
+        )", &settings));
+
+        TBlockEvents<TEvSchemeShard::TEvModifySchemeTransaction> indexCreationBlocker(runtime, [](const auto& ev) {
+            const auto& modifyScheme = ev->Get()->Record.GetTransaction(0);
+            return modifyScheme.GetOperationType() == NKikimrSchemeOp::ESchemeOpCreateIndexBuild;
+        });
+
+        const ui64 buildIndexTx = ++txId;
+        TestBuildIndex(runtime, buildIndexTx, TTestTxConfig::SchemeShard, "/MyRoot", "/MyRoot/Table", TBuildIndexConfig{
+            "Index", NKikimrSchemeOp::EIndexTypeGlobal, { "value" }, {},
+            { NYdb::NTable::TGlobalIndexSettings::FromProto(settings) }
+        });
+
+        RebootTablet(runtime, TTestTxConfig::SchemeShard, runtime.AllocateEdgeActor());
+
+        indexCreationBlocker.Stop().Unblock();
+        env.TestWaitNotification(runtime, buildIndexTx);
+
+        auto buildIndexOperation = TestGetBuildIndex(runtime, TTestTxConfig::SchemeShard, "/MyRoot", buildIndexTx);
+        UNIT_ASSERT_VALUES_EQUAL_C(
+            (int)buildIndexOperation.GetIndexBuild().GetState(), (int)Ydb::Table::IndexBuildState::STATE_DONE,
+            buildIndexOperation.DebugString()
+        );
+
+        TestDescribeResult(DescribePath(runtime, "/MyRoot/Table"), {
+            NLs::IsTable,
+            NLs::IndexesCount(1)
+        });
+
+        TestDescribeResult(DescribePrivatePath(runtime, "/MyRoot/Table/Index"), {
+            NLs::PathExist,
+            NLs::IndexState(NKikimrSchemeOp::EIndexState::EIndexStateReady)
+        });
+
+        TestDescribeResult(DescribePrivatePath(runtime, "/MyRoot/Table/Index/indexImplTable", true, true), {
+            NLs::IsTable,
+            NLs::PartitionCount(3),
+            NLs::MinPartitionsCountEqual(3),
+            NLs::MaxPartitionsCountEqual(3),
+            NLs::PartitionKeys({"alice", "bob", ""})
+        });
+    }
+
     Y_UNIT_TEST(DropIndex) {
         TTestBasicRuntime runtime;
         TTestEnv env(runtime);

ydb/core/tx/schemeshard/ut_index_build/ya.make

@@ -14,6 +14,7 @@ PEERDIR(
     ydb/core/testlib/default
     ydb/core/tx
     ydb/core/tx/schemeshard/ut_helpers
+    ydb/public/sdk/cpp/client/ydb_table
 )
 
 YQL_LAST_ABI_VERSION()

ydb/core/tx/schemeshard/ut_index_build_reboots/ut_index_build_reboots.cpp

@@ -450,4 +450,75 @@ Y_UNIT_TEST_SUITE(IndexBuildTestReboots) {
 
         });
     }
+
+    Y_UNIT_TEST(IndexPartitioning) {
+        TTestWithReboots t(false);
+        t.Run([&](TTestActorRuntime& runtime, bool& activeZone) {
+            {
+                TInactiveZone inactive(activeZone);
+
+                TestCreateTable(runtime, ++t.TxId, "/MyRoot", R"(
+                    Name: "Table"
+                    Columns { Name: "key" Type: "Uint32" }
+                    Columns { Name: "value" Type: "Utf8" }
+                    KeyColumnNames: [ "key" ]
+                )");
+                t.TestEnv->TestWaitNotification(runtime, t.TxId);
+            }
+
+            Ydb::Table::GlobalIndexSettings settings;
+            UNIT_ASSERT(google::protobuf::TextFormat::ParseFromString(R"(
+                partition_at_keys {
+                    split_points {
+                        type { tuple_type { elements { optional_type { item { type_id: UTF8 } } } } }
+                        value { items { text_value: "alice" } }
+                    }
+                    split_points {
+                        type { tuple_type { elements { optional_type { item { type_id: UTF8 } } } } }
+                        value { items { text_value: "bob" } }
+                    }
+                }
+                partitioning_settings {
+                    min_partitions_count: 3
+                    max_partitions_count: 3
+                }
+            )", &settings));
+
+            const ui64 buildIndexId = ++t.TxId;
+            AsyncBuildIndex(runtime, buildIndexId, TTestTxConfig::SchemeShard, "/MyRoot", "/MyRoot/Table", TBuildIndexConfig{
+                "Index", NKikimrSchemeOp::EIndexTypeGlobal, { "value" }, {},
+                { NYdb::NTable::TGlobalIndexSettings::FromProto(settings) }
+            });
+
+            {
+                auto descr = TestGetBuildIndex(runtime, TTestTxConfig::SchemeShard, "/MyRoot", buildIndexId);
+                UNIT_ASSERT_VALUES_EQUAL((int)descr.GetIndexBuild().GetState(), (int)Ydb::Table::IndexBuildState::STATE_PREPARING);
+            }
+
+            t.TestEnv->TestWaitNotification(runtime, buildIndexId);
+
+            {
+                auto descr = TestGetBuildIndex(runtime, TTestTxConfig::SchemeShard, "/MyRoot", buildIndexId);
+                UNIT_ASSERT_VALUES_EQUAL((int)descr.GetIndexBuild().GetState(), (int)Ydb::Table::IndexBuildState::STATE_DONE);
+            }
+
+            TestDescribeResult(DescribePath(runtime, "/MyRoot/Table"), {
+                NLs::IsTable,
+                NLs::IndexesCount(1)
+            });
+
+            TestDescribeResult(DescribePrivatePath(runtime, "/MyRoot/Table/Index"), {
+                NLs::PathExist,
+                NLs::IndexState(NKikimrSchemeOp::EIndexState::EIndexStateReady)
+            });
+
+            TestDescribeResult(DescribePrivatePath(runtime, "/MyRoot/Table/Index/indexImplTable", true, true), {
+                NLs::IsTable,
+                NLs::PartitionCount(3),
+                NLs::MinPartitionsCountEqual(3),
+                NLs::MaxPartitionsCountEqual(3),
+                NLs::PartitionKeys({"alice", "bob", ""})
+            });
+        });
+    }
 }

ydb/core/tx/schemeshard/ut_index_build_reboots/ya.make

@@ -18,6 +18,7 @@ PEERDIR(
     ydb/core/tx
     ydb/core/tx/schemeshard/ut_helpers
     ydb/library/yql/public/udf/service/exception_policy
+    ydb/public/sdk/cpp/client/ydb_table
 )
 
 SRCS(

ydb/public/sdk/cpp/client/ydb_table/table.cpp

@@ -251,8 +251,7 @@ static void SerializeTo(const TRenameIndex& rename, Ydb::Table::RenameIndexItem&
     proto.set_replace_destination(rename.ReplaceDestination_);
 }
 
-template <typename TProto>
-TExplicitPartitions TExplicitPartitions::FromProto(const TProto& proto) {
+TExplicitPartitions TExplicitPartitions::FromProto(const Ydb::Table::ExplicitPartitions& proto) {
     TExplicitPartitions out;
     for (const auto& splitPoint : proto.split_points()) {
         TValue value(TType(splitPoint.type()), splitPoint.value());
@@ -2307,13 +2306,12 @@ ui64 TIndexDescription::GetSizeBytes() const {
     return SizeBytes;
 }
 
-template <typename TProto>
-TGlobalIndexSettings TGlobalIndexSettings::FromProto(const TProto& proto) {
-    auto partitionsFromProto = [](const auto& proto) -> TUniformOrExplicitPartitions {
+TGlobalIndexSettings TGlobalIndexSettings::FromProto(const Ydb::Table::GlobalIndexSettings& proto) {
+    auto partitionsFromProto = [](const Ydb::Table::GlobalIndexSettings& proto) -> TUniformOrExplicitPartitions {
         switch (proto.partitions_case()) {
-        case TProto::kUniformPartitions:
+        case Ydb::Table::GlobalIndexSettings::kUniformPartitions:
             return proto.uniform_partitions();
-        case TProto::kPartitionAtKeys:
+        case Ydb::Table::GlobalIndexSettings::kPartitionAtKeys:
             return TExplicitPartitions::FromProto(proto.partition_at_keys());
         default:
             return {};

ydb/public/sdk/cpp/client/ydb_table/table.h

@@ -184,9 +184,7 @@ struct TExplicitPartitions {
 
     FLUENT_SETTING_VECTOR(TValue, SplitPoints);
 
-    template <typename TProto>
-    static TExplicitPartitions FromProto(const TProto& proto);
-
+    static TExplicitPartitions FromProto(const Ydb::Table::ExplicitPartitions& proto);
     void SerializeTo(Ydb::Table::ExplicitPartitions& proto) const;
 };
 
@@ -196,9 +194,7 @@ struct TGlobalIndexSettings {
     TPartitioningSettings PartitioningSettings;
     TUniformOrExplicitPartitions Partitions;
 
-    template <typename TProto>
-    static TGlobalIndexSettings FromProto(const TProto& proto);
-
+    static TGlobalIndexSettings FromProto(const Ydb::Table::GlobalIndexSettings& proto);
     void SerializeTo(Ydb::Table::GlobalIndexSettings& proto) const;
 };
 

ydb/tests/functional/scheme_tests/canondata/tablet_scheme_tests.TestTabletSchemes.test_tablet_schemes_flat_schemeshard_/flat_schemeshard.schema

@@ -4788,6 +4788,11 @@
                 "ColumnId": 33,
                 "ColumnName": "AlterMainTableTxDone",
                 "ColumnType": "Bool"
+            },
+            {
+                "ColumnId": 34,
+                "ColumnName": "CreationConfig",
+                "ColumnType": "String"
             }
         ],
         "ColumnsDropped": [],
@@ -4826,7 +4831,8 @@
                     30,
                     31,
                     32,
-                    33
+                    33,
+                    34
                 ],
                 "RoomID": 0,
                 "Codec": 0,