25-3: Datashard configurable stats report intervals

ydb/core/client/flat_ut.cpp

@@ -2627,7 +2627,8 @@ 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);
@@ -2800,7 +2801,8 @@ 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);
@@ -2906,7 +2908,8 @@ 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

@@ -26,4 +26,6 @@ message TDataShardConfig {
     optional uint64 IncrementalRestoreReadAheadLo = 20 [default = 524288];
     optional uint64 IncrementalRestoreReadAheadHi = 21 [default = 1048576];
     optional uint64 InMemoryStateMigrationTimeoutMs = 24 [default = 1000];
+    optional uint32 StatsReportIntervalSeconds = 25 [default = 10];
+    optional uint32 BackupTableStatsReportIntervalSeconds = 26 [default = 30];
 }

ydb/core/sys_view/ut_common.cpp

@@ -77,6 +77,11 @@ TTestEnv::TTestEnv(ui32 staticNodes, ui32 dynamicNodes, const TTestEnvSettings&
 
     Settings->AppConfig->MutableHiveConfig()->AddBalancerIgnoreTabletTypes(NKikimrTabletBase::TTabletTypes::SysViewProcessor);
 
+    if (settings.DataShardStatsReportIntervalSeconds) {
+        Settings->AppConfig->MutableDataShardConfig()
+            ->SetStatsReportIntervalSeconds(*settings.DataShardStatsReportIntervalSeconds);
+    }
+
     Server = new Tests::TServer(*Settings);
     Server->EnableGRpc(grpcPort);
 

ydb/core/sys_view/ut_common.h

@@ -28,6 +28,7 @@ struct TTestEnvSettings {
     bool EnableOlapCompression = false;
     TMaybe<bool> EnableRealSystemViewPaths;
     NKikimrProto::TAuthConfig AuthConfig = {};
+    TMaybe<ui32> DataShardStatsReportIntervalSeconds;
 };
 
 class TTestEnv {

ydb/core/sys_view/ut_kqp.cpp

@@ -2704,9 +2704,7 @@ ALTER OBJECT `/Root/test_show_create` (TYPE TABLE) SET (ACTION = UPSERT_OPTIONS,
     }
 
     Y_UNIT_TEST(PartitionStatsLocksFields) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
-        TTestEnv env;
+        TTestEnv env({.DataShardStatsReportIntervalSeconds = 0});
         CreateRootTable(env, /* partitionCount */ 1, /* fillTable */ true);
 
         TTableClient client(env.GetDriver());
@@ -2734,11 +2732,9 @@ ALTER OBJECT `/Root/test_show_create` (TYPE TABLE) SET (ACTION = UPSERT_OPTIONS,
     }
 
     Y_UNIT_TEST(PartitionStatsFields) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         auto nowUs = TInstant::Now().MicroSeconds();
 
-        TTestEnv env;
+        TTestEnv env({.DataShardStatsReportIntervalSeconds = 0});
         CreateRootTable(env);
         const auto describeResult = env.GetClient().Describe(env.GetServer().GetRuntime(), "Root/Table0");
         const auto tablePathId = describeResult.GetPathId();
@@ -3229,11 +3225,9 @@ ALTER OBJECT `/Root/test_show_create` (TYPE TABLE) SET (ACTION = UPSERT_OPTIONS,
     }
 
     Y_UNIT_TEST(TopPartitionsByCpuFields) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         auto nowUs = TInstant::Now().MicroSeconds();
 
-        TTestEnv env(1, 4, {.EnableSVP = true});
+        TTestEnv env(1, 4, {.EnableSVP = true, .DataShardStatsReportIntervalSeconds = 0});
         CreateTenantsAndTables(env);
 
         TTableClient client(env.GetDriver());
@@ -3281,11 +3275,9 @@ ALTER OBJECT `/Root/test_show_create` (TYPE TABLE) SET (ACTION = UPSERT_OPTIONS,
     }
 
     Y_UNIT_TEST(TopPartitionsByCpuTables) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         constexpr ui64 partitionCount = 5;
 
-        TTestEnv env(1, 4, {.EnableSVP = true});
+        TTestEnv env(1, 4, {.EnableSVP = true, .DataShardStatsReportIntervalSeconds = 0});
         CreateTenantsAndTables(env, true, partitionCount);
 
         TTableClient client(env.GetDriver());
@@ -3311,11 +3303,9 @@ ALTER OBJECT `/Root/test_show_create` (TYPE TABLE) SET (ACTION = UPSERT_OPTIONS,
     }
 
     Y_UNIT_TEST(TopPartitionsByCpuRanges) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         constexpr ui64 partitionCount = 5;
 
-        TTestEnv env(1, 4, {.EnableSVP = true});
+        TTestEnv env(1, 4, {.EnableSVP = true, .DataShardStatsReportIntervalSeconds = 0});
         CreateTenantsAndTables(env, true, partitionCount);
 
         TTableClient client(env.GetDriver());
@@ -3392,11 +3382,13 @@ ALTER OBJECT `/Root/test_show_create` (TYPE TABLE) SET (ACTION = UPSERT_OPTIONS,
     }
 
     Y_UNIT_TEST(TopPartitionsByCpuFollowers) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         auto nowUs = TInstant::Now().MicroSeconds();
 
-        TTestEnv env(1, 4, {.EnableSVP = true, .EnableForceFollowers = true});
+        TTestEnv env(1, 4, {
+            .EnableSVP = true,
+            .EnableForceFollowers = true,
+            .DataShardStatsReportIntervalSeconds = 0,
+        });
 
         auto& runtime = *env.GetServer().GetRuntime();
         runtime.SetLogPriority(NKikimrServices::TX_DATASHARD, NLog::PRI_TRACE);
@@ -3567,9 +3559,7 @@ ALTER OBJECT `/Root/test_show_create` (TYPE TABLE) SET (ACTION = UPSERT_OPTIONS,
     }
 
     Y_UNIT_TEST(TopPartitionsByTliFields) {
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
-        TTestEnv env(1, 4, {.EnableSVP = true});
+        TTestEnv env(1, 4, {.EnableSVP = true, .DataShardStatsReportIntervalSeconds = 0});
         CreateTenantsAndTables(env);
 
         TTableClient client(env.GetDriver());

ydb/core/testlib/test_client.cpp

@@ -554,6 +554,7 @@ 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());
             appData.SharedCacheConfig.MergeFrom(Settings->AppConfig->GetSharedCacheConfig());

ydb/core/tx/datashard/datashard.cpp

@@ -35,7 +35,6 @@ 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

@@ -153,7 +153,6 @@ 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

@@ -596,13 +596,31 @@ 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 && State != TShardState::Readonly)

ydb/core/tx/datashard/datashard_impl.h

@@ -3334,13 +3334,16 @@ 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

@@ -391,6 +391,8 @@ 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();
@@ -402,8 +404,6 @@ Y_UNIT_TEST_SUITE(DataShardStats) {
 
         InitRoot(server, sender);
 
-        NDataShard::gDbStatsReportInterval = TDuration::Seconds(0);
-
         NLocalDb::TCompactionPolicyPtr policy = NLocalDb::CreateDefaultUserTablePolicy();
         policy->InMemForceStepsToSnapshot = 1;
 
@@ -834,6 +834,47 @@ 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

@@ -1319,12 +1319,14 @@ 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

@@ -531,7 +531,8 @@ 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

@@ -10900,12 +10900,12 @@ 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

@@ -771,12 +771,14 @@ 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);
@@ -902,12 +904,14 @@ 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);
@@ -976,12 +980,14 @@ 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

@@ -656,6 +656,11 @@ 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, opts.NStoragePools_);
 
     SetupLogging(runtime);