24-4-4-hotfix: Datashard configurable stats report intervals

ydb/core/client/flat_ut.cpp

@@ -2950,7 +2950,8 @@ Y_UNIT_TEST_SUITE(TFlatTest) {
         UNIT_ASSERT_VALUES_EQUAL(partitions.size(), 1);
 
         // Force stats reporting without delays
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
+        cleverServer.GetRuntime()->GetAppData()
+            .DataShardConfig.SetStatsReportIntervalSeconds(0);
 
         cleverServer.GetRuntime()->SetLogPriority(NKikimrServices::FLAT_TX_SCHEMESHARD, NActors::NLog::PRI_DEBUG);
 //        cleverServer.GetRuntime()->SetLogPriority(NKikimrServices::TABLET_EXECUTOR, NActors::NLog::PRI_DEBUG);
@@ -3123,7 +3124,8 @@ Y_UNIT_TEST_SUITE(TFlatTest) {
         UNIT_ASSERT_VALUES_EQUAL(partitions.size(), 4);
 
         // Force stats reporting without delays
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
+        cleverServer.GetRuntime()->GetAppData()
+            .DataShardConfig.SetStatsReportIntervalSeconds(0);
 
         cleverServer.GetRuntime()->SetLogPriority(NKikimrServices::FLAT_TX_SCHEMESHARD, NActors::NLog::PRI_DEBUG);
 //        cleverServer.GetRuntime()->SetLogPriority(NKikimrServices::TABLET_EXECUTOR, NActors::NLog::PRI_DEBUG);
@@ -3229,7 +3231,8 @@ Y_UNIT_TEST_SUITE(TFlatTest) {
         TVector<ui64> initialPartitions = annoyingClient.GetTablePartitions("/dc-1/Dir/T1");
 
         // Force stats reporting without delays
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
+        cleverServer.GetRuntime()->GetAppData()
+            .DataShardConfig.SetStatsReportIntervalSeconds(0);
         NDataShard::gDbStatsDataSizeResolution = 80000;
 
         TString bigValue(100*1024, '0');

ydb/core/protos/datashard_config.proto

@@ -21,4 +21,6 @@ message TDataShardConfig {
     optional string CdcInitialScanTaskName = 17 [default = "cdc_initial_scan"];
     optional uint32 CdcInitialScanTaskPriority = 18 [default = 10];
     optional bool DisabledOnSchemeShard = 19 [default = false];
+    optional uint32 StatsReportIntervalSeconds = 25 [default = 10];
+    optional uint32 BackupTableStatsReportIntervalSeconds = 26 [default = 30];
 }

ydb/core/sys_view/ut_common.cpp

@@ -26,7 +26,10 @@ NKikimrSubDomains::TSubDomainSettings GetSubDomainDefaultSettings(const TString
     return subdomain;
 }
 
-TTestEnv::TTestEnv(ui32 staticNodes, ui32 dynamicNodes, ui32 storagePools, ui32 pqTabletsN, bool enableSVP, bool disableSources) {
+TTestEnv::TTestEnv(
+    ui32 staticNodes, ui32 dynamicNodes, ui32 storagePools,
+    ui32 pqTabletsN, bool enableSVP, bool disableSources,
+    std::optional<ui32> dataShardStatsReportIntervalSeconds) {
     auto mbusPort = PortManager.GetPort();
     auto grpcPort = PortManager.GetPort();
 
@@ -56,6 +59,11 @@ TTestEnv::TTestEnv(ui32 staticNodes, ui32 dynamicNodes, ui32 storagePools, ui32
     Settings->AppConfig->MutableTableServiceConfig()->SetEnableKqpDataQuerySourceRead(!disableSources);
     Settings->AppConfig->MutableHiveConfig()->AddBalancerIgnoreTabletTypes(NKikimrTabletBase::TTabletTypes::SysViewProcessor);
 
+    if (dataShardStatsReportIntervalSeconds) {
+        Settings->AppConfig->MutableDataShardConfig()
+            ->SetStatsReportIntervalSeconds(*dataShardStatsReportIntervalSeconds);
+    }
+
     Server = new Tests::TServer(*Settings);
     Server->EnableGRpc(grpcPort);
 

ydb/core/sys_view/ut_common.h

@@ -23,7 +23,8 @@ class TTestEnv {
 
 public:
     TTestEnv(ui32 staticNodes = 1, ui32 dynamicNodes = 4, ui32 storagePools = 0,
-        ui32 pqTabletsN = 0, bool enableSVP = false, bool disableSources = false);
+        ui32 pqTabletsN = 0, bool enableSVP = false, bool disableSources = false,
+        std::optional<ui32> dataShardStatsReportIntervalSeconds = std::nullopt);
 
     template<typename... Args>
     TTestEnv(TDisableSourcesTag, Args&&... args)

ydb/core/sys_view/ut_kqp.cpp

@@ -717,11 +717,9 @@ Y_UNIT_TEST_SUITE(SystemView) {
     }
 
     Y_UNIT_TEST(PartitionStatsFields) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         auto nowUs = TInstant::Now().MicroSeconds();
 
-        TTestEnv env;
+        TTestEnv env(1, 4, 0, 0, false, false, 0);
         CreateRootTable(env);
 
         TTableClient client(env.GetDriver());
@@ -1210,11 +1208,9 @@ Y_UNIT_TEST_SUITE(SystemView) {
     }
 
     Y_UNIT_TEST(TopPartitionsFields) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         auto nowUs = TInstant::Now().MicroSeconds();
 
-        TTestEnv env(1, 4, 0, 0, true);
+        TTestEnv env(1, 4, 0, 0, true, false, 0);
         CreateTenantsAndTables(env);
 
         TTableClient client(env.GetDriver());
@@ -1262,11 +1258,9 @@ Y_UNIT_TEST_SUITE(SystemView) {
     }
 
     Y_UNIT_TEST(TopPartitionsTables) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         constexpr ui64 partitionCount = 5;
 
-        TTestEnv env(1, 4, 0, 0, true);
+        TTestEnv env(1, 4, 0, 0, true, false, 0);
         CreateTenantsAndTables(env, true, partitionCount);
 
         TTableClient client(env.GetDriver());
@@ -1292,11 +1286,9 @@ Y_UNIT_TEST_SUITE(SystemView) {
     }
 
     Y_UNIT_TEST(TopPartitionsRanges) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         constexpr ui64 partitionCount = 5;
 
-        TTestEnv env(1, 4, 0, 0, true);
+        TTestEnv env(1, 4, 0, 0, true, false, 0);
         CreateTenantsAndTables(env, true, partitionCount);
 
         TTableClient client(env.GetDriver());

ydb/core/testlib/test_client.cpp

@@ -252,6 +252,7 @@ namespace Tests {
             appData.DataStreamsAuthFactory = Settings->DataStreamsAuthFactory.get();
             appData.PersQueueMirrorReaderFactory = Settings->PersQueueMirrorReaderFactory.get();
             appData.HiveConfig.MergeFrom(Settings->AppConfig->GetHiveConfig());
+            appData.DataShardConfig.MergeFrom(Settings->AppConfig->GetDataShardConfig());
             appData.GraphConfig.MergeFrom(Settings->AppConfig->GetGraphConfig());
             appData.SqsConfig.MergeFrom(Settings->AppConfig->GetSqsConfig());
 

ydb/core/tx/datashard/datashard.cpp

@@ -34,7 +34,6 @@ using namespace NTabletFlatExecutor;
 // But in unit tests we want to test both scenarios
 bool gAllowLogBatchingDefaultValue = true;
 
-TDuration gDbStatsReportInterval = TDuration::Seconds(10);
 ui64 gDbStatsDataSizeResolution = 10*1024*1024;
 ui64 gDbStatsRowCountResolution = 100000;
 ui32 gDbStatsHistogramBucketsCount = 10;

ydb/core/tx/datashard/datashard.h

@@ -148,7 +148,6 @@ namespace NDataShard {
 
     // NOTE: this switch should be modified only in tests !!!
     extern bool gAllowLogBatchingDefaultValue;
-    extern TDuration gDbStatsReportInterval;
     extern ui64 gDbStatsDataSizeResolution;
     extern ui64 gDbStatsRowCountResolution;
     extern ui32 gDbStatsHistogramBucketsCount;

ydb/core/tx/datashard/datashard__stats.cpp

@@ -597,13 +597,31 @@ class TDataShard::TTxInitiateStatsUpdate : public NTabletFlatExecutor::TTransact
     }
 };
 
+TDuration TDataShard::GetStatsReportInterval(const TAppData& appData) const {
+    const auto& userTables = GetUserTables();
+    const bool isBackup = !userTables.empty() && std::all_of(userTables.begin(), userTables.end(),
+        [](const auto& kv) { return kv.second->IsBackup; });
+
+    if (isBackup) {
+        // Clamp the interval for backup tables to the value for ordinary tables, as it
+        // makes no sense for the latter to be longer than the former.
+        auto interval = std::max(
+            appData.DataShardConfig.GetBackupTableStatsReportIntervalSeconds(),
+            appData.DataShardConfig.GetStatsReportIntervalSeconds());
+        return TDuration::Seconds(interval);
+    } else {
+        return TDuration::Seconds(appData.DataShardConfig.GetStatsReportIntervalSeconds());
+    }
+}
+
 void TDataShard::UpdateTableStats(const TActorContext &ctx) {
     if (StatisticsDisabled)
         return;
 
-    TInstant now = AppData(ctx)->TimeProvider->Now();
+    auto* appData = AppData(ctx);
+    TInstant now = appData->TimeProvider->Now();
 
-    if (LastDbStatsUpdateTime + gDbStatsReportInterval > now)
+    if (LastDbStatsUpdateTime + GetStatsReportInterval(*appData) > now)
         return;
 
     if (State != TShardState::Ready)

ydb/core/tx/datashard/datashard_impl.h

@@ -3209,13 +3209,16 @@ class TDataShard
         NTabletPipe::SendData(ctx, StateReportPipe, ev.Release());
     }
 
+    TDuration GetStatsReportInterval(const TAppData&) const;
+
     void SendPeriodicTableStats(const TActorContext &ctx) {
         if (StatisticsDisabled)
             return;
 
-        TInstant now = AppData(ctx)->TimeProvider->Now();
+        auto* appData = AppData(ctx);
+        TInstant now = appData->TimeProvider->Now();
 
-        if (LastDbStatsReportTime + gDbStatsReportInterval > now)
+        if (LastDbStatsReportTime + GetStatsReportInterval(*appData) > now)
             return;
 
         auto* resourceMetrics = Executor()->GetResourceMetrics();

ydb/core/tx/datashard/datashard_ut_stats.cpp

@@ -388,6 +388,8 @@ Y_UNIT_TEST_SUITE(DataShardStats) {
         TServerSettings serverSettings(pm.GetPort(2134));
         serverSettings.SetDomainName("Root")
             .SetUseRealThreads(false);
+        serverSettings.AppConfig->MutableDataShardConfig()
+            ->SetStatsReportIntervalSeconds(0);
 
         TServer::TPtr server = new TServer(serverSettings);
         auto& runtime = *server->GetRuntime();
@@ -399,8 +401,6 @@ Y_UNIT_TEST_SUITE(DataShardStats) {
 
         InitRoot(server, sender);
 
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         NLocalDb::TCompactionPolicyPtr policy = NLocalDb::CreateDefaultUserTablePolicy();
         policy->InMemForceStepsToSnapshot = 1;
 
@@ -696,6 +696,47 @@ Y_UNIT_TEST_SUITE(DataShardStats) {
         WaitTableStats(runtime, shard1, DoesNotHaveSchemaChangesCondition());
     }
 
+    Y_UNIT_TEST(BackupTableStatsReportInterval) {
+        TPortManager pm;
+        TServerSettings serverSettings(pm.GetPort(2134));
+        serverSettings.SetDomainName("Root")
+            .SetUseRealThreads(false);
+
+        TServer::TPtr server = new TServer(serverSettings);
+        auto& runtime = *server->GetRuntime();
+        auto sender = runtime.AllocateEdgeActor();
+
+        runtime.SetLogPriority(NKikimrServices::TX_DATASHARD, NLog::PRI_TRACE);
+
+        InitRoot(server, sender);
+
+        auto [shards, tableId1] = CreateShardedTable(server, sender, "/Root", "table-1", 1);
+
+        // Create a backup copy table
+        {
+            auto senderCopy = runtime.AllocateEdgeActor();
+            ui64 txId = AsyncCreateCopyTable(
+                server, senderCopy, "/Root", "table-2", "/Root/table-1", /*isBackup=*/true);
+            WaitTxNotification(server, senderCopy, txId);
+        }
+        auto tableId2 = ResolveTableId(server, sender, "/Root/table-2");
+
+        std::unordered_map<TLocalPathId, size_t> tableToStatsCount;
+        auto observerFunc = [&](auto& ev) {
+            const NKikimrTxDataShard::TEvPeriodicTableStats& record = ev->Get()->Record;
+            ++tableToStatsCount[record.GetTableLocalId()];
+        };
+        auto observer = runtime.AddObserver<TEvDataShard::TEvPeriodicTableStats>(observerFunc);
+
+        runtime.WaitFor("First stats event", [&]{ return !tableToStatsCount.empty(); });
+        runtime.SimulateSleep(TDuration::Seconds(45));
+        // Once every 10 seconds
+        UNIT_ASSERT_GE(tableToStatsCount[tableId1.PathId.LocalPathId], 4);
+        UNIT_ASSERT_LE(tableToStatsCount[tableId1.PathId.LocalPathId], 5);
+        // Once every 30 seconds
+        UNIT_ASSERT_GE(tableToStatsCount[tableId2.PathId.LocalPathId], 1);
+        UNIT_ASSERT_LE(tableToStatsCount[tableId2.PathId.LocalPathId], 2);
+    }
 }
 
 }

ydb/core/tx/datashard/ut_common/datashard_ut_common.cpp

@@ -1254,12 +1254,14 @@ ui64 AsyncCreateCopyTable(
         TActorId sender,
         const TString &root,
         const TString &name,
-        const TString &from)
+        const TString &from,
+        bool isBackup)
 {
     auto request = SchemeTxTemplate(NKikimrSchemeOp::ESchemeOpCreateTable, root);
     auto& desc = *request->Record.MutableTransaction()->MutableModifyScheme()->MutableCreateTable();
     desc.SetName(name);
     desc.SetCopyFromTable(from);
+    desc.SetIsBackup(isBackup);
 
     return RunSchemeTx(*server->GetRuntime(), std::move(request), sender);
 }

ydb/core/tx/datashard/ut_common/datashard_ut_common.h

@@ -527,7 +527,8 @@ ui64 AsyncCreateCopyTable(Tests::TServer::TPtr server,
                           TActorId sender,
                           const TString &root,
                           const TString &name,
-                          const TString &from);
+                          const TString &from,
+                          bool isBackup = false);
 
 NKikimrTxDataShard::TEvCompactTableResult CompactTable(
     TTestActorRuntime& runtime, ui64 shardId, const TTableId& tableId, bool compactBorrowed = false);

ydb/core/tx/schemeshard/ut_base/ut_base.cpp

@@ -10568,12 +10568,12 @@ Y_UNIT_TEST_SUITE(TSchemeShardTest) {
 
         TTestEnvOptions opts;
         opts.EnableBackgroundCompaction(false);
+        opts.DataShardStatsReportIntervalSeconds(1);
 
         TTestEnv env(runtime, opts);
 
         ui64 txId = 100;
 
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(1);
         NDataShard::gDbStatsDataSizeResolution = 10;
         NDataShard::gDbStatsRowCountResolution = 10;
 

ydb/core/tx/schemeshard/ut_compaction/ut_compaction.cpp

@@ -768,12 +768,14 @@ Y_UNIT_TEST_SUITE(TSchemeshardBorrowedCompactionTest) {
         // 1. Copy table is not compacted until we want to split it
         // 2. After borrow compaction both src and dst tables are background compacted
 
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(1);
         NDataShard::gDbStatsDataSizeResolution = 10;
         NDataShard::gDbStatsRowCountResolution = 10;
 
+        TTestEnvOptions opts;
+        opts.DataShardStatsReportIntervalSeconds(1);
+
         TTestBasicRuntime runtime;
-        TTestEnv env(runtime);
+        TTestEnv env(runtime, opts);
 
         runtime.SetLogPriority(NKikimrServices::TX_DATASHARD, NLog::PRI_DEBUG);
         runtime.SetLogPriority(NKikimrServices::FLAT_TX_SCHEMESHARD, NActors::NLog::PRI_TRACE);
@@ -899,12 +901,14 @@ Y_UNIT_TEST_SUITE(TSchemeshardBorrowedCompactionTest) {
         // I.e. we compact borrowed data ASAP except copy table case, when
         // we don't want to compact at all.
 
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(1);
         NDataShard::gDbStatsDataSizeResolution = 10;
         NDataShard::gDbStatsRowCountResolution = 10;
 
+        TTestEnvOptions opts;
+        opts.DataShardStatsReportIntervalSeconds(1);
+
         TTestBasicRuntime runtime;
-        TTestEnv env(runtime);
+        TTestEnv env(runtime, opts);
 
         runtime.SetLogPriority(NKikimrServices::TX_DATASHARD, NLog::PRI_DEBUG);
         runtime.SetLogPriority(NKikimrServices::FLAT_TX_SCHEMESHARD, NActors::NLog::PRI_TRACE);
@@ -973,12 +977,14 @@ Y_UNIT_TEST_SUITE(TSchemeshardBorrowedCompactionTest) {
 
     Y_UNIT_TEST(SchemeshardShouldNotCompactBorrowedAfterSplitMergeWhenDisabled) {
 
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(1);
         NDataShard::gDbStatsDataSizeResolution = 10;
         NDataShard::gDbStatsRowCountResolution = 10;
 
+        TTestEnvOptions opts;
+        opts.DataShardStatsReportIntervalSeconds(1);
+
         TTestBasicRuntime runtime;
-        TTestEnv env(runtime);
+        TTestEnv env(runtime, opts);
 
         runtime.SetLogPriority(NKikimrServices::TX_DATASHARD, NLog::PRI_DEBUG);
         runtime.SetLogPriority(NKikimrServices::FLAT_TX_SCHEMESHARD, NActors::NLog::PRI_TRACE);

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

@@ -580,6 +580,11 @@ NSchemeShardUT_Private::TTestEnv::TTestEnv(TTestActorRuntime& runtime, const TTe
         app.AddSystemBackupSID(sid);
     }
 
+    if (opts.DataShardStatsReportIntervalSeconds_) {
+        app.DataShardConfig.SetStatsReportIntervalSeconds(
+            *opts.DataShardStatsReportIntervalSeconds_);
+    }
+
     AddDomain(runtime, app, TTestTxConfig::DomainUid, hive, schemeRoot);
 
     SetupLogging(runtime);

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

@@ -68,6 +68,7 @@ namespace NSchemeShardUT_Private {
         OPTION(std::optional<bool>, EnableChangefeedsOnIndexTables, std::nullopt);
         OPTION(std::optional<bool>, EnableTieringInColumnShard, std::nullopt);
         OPTION(std::optional<bool>, EnableParameterizedDecimal, std::nullopt);
+        OPTION(std::optional<ui32>, DataShardStatsReportIntervalSeconds, std::nullopt);
 
         #undef OPTION
     };

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

@@ -446,7 +446,7 @@ Y_UNIT_TEST_SUITE(IndexBuildTest) {
                             NLs::PathVersionEqual(3)});
 
         // Force stats reporting without delays
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
+        runtime.GetAppData().DataShardConfig.SetStatsReportIntervalSeconds(0);
         NDataShard::gDbStatsDataSizeResolution = 80000;
 
         auto upgradeEvent = [&](TAutoPtr<IEventHandle>& ev) -> auto {
@@ -768,10 +768,9 @@ Y_UNIT_TEST_SUITE(IndexBuildTest) {
         TTestEnvOptions opts;
         opts.EnableBackgroundCompaction(false);
         opts.DisableStatsBatching(true);
+        opts.DataShardStatsReportIntervalSeconds(0);
         TTestEnv env(runtime, opts);
 
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         ui64 txId = 100;
 
         TestCreateTable(runtime, ++txId, "/MyRoot", R"(