Move MaxTxInFly to Immediate Control Board

Author: neyrox

ydb/core/base/appdata.cpp

@@ -81,7 +81,6 @@ struct TAppData::TImpl {
     NKikimrConfig::TStatisticsConfig StatisticsConfig;
     TMetricsConfig MetricsConfig;
     NKikimrConfig::TSystemTabletBackupConfig SystemTabletBackupConfig;
-    NKikimrConfig::TSystemTabletConfig SystemTabletConfig;
 };
 
 TAppData::TAppData(
@@ -148,7 +147,6 @@ TAppData::TAppData(
     , StatisticsConfig(Impl->StatisticsConfig)
     , MetricsConfig(Impl->MetricsConfig)
     , SystemTabletBackupConfig(Impl->SystemTabletBackupConfig)
-    , SystemTabletConfig(Impl->SystemTabletConfig)
     , KikimrShouldContinue(kikimrShouldContinue)
     , TracingConfigurator(MakeIntrusive<NJaegerTracing::TSamplingThrottlingConfigurator>(TimeProvider, RandomProvider))
 {}

ydb/core/base/appdata_fwd.h

@@ -81,7 +81,6 @@ namespace NKikimrConfig {
     class TBridgeConfig;
     class TStatisticsConfig;
     class TSystemTabletBackupConfig;
-    class TSystemTabletConfig;
 }
 
 namespace NKikimrReplication {
@@ -271,7 +270,6 @@ struct TAppData {
     NKikimrConfig::TStatisticsConfig& StatisticsConfig;
     TMetricsConfig& MetricsConfig;
     NKikimrConfig::TSystemTabletBackupConfig& SystemTabletBackupConfig;
-    NKikimrConfig::TSystemTabletConfig& SystemTabletConfig;
     bool EnforceUserTokenRequirement = false;
     bool EnforceUserTokenCheckRequirement = false; // check token if it was specified
     bool AllowHugeKeyValueDeletes = true; // delete when all clients limit deletes per request

ydb/core/driver_lib/run/run.cpp

@@ -1466,10 +1466,6 @@ void TKikimrRunner::InitializeAppData(const TKikimrRunConfig& runConfig)
         AppData->DataIntegrityTrailsConfig = runConfig.AppConfig.GetDataIntegrityTrailsConfig();
     }
 
-    if (runConfig.AppConfig.HasSystemTabletConfig()) {
-        AppData->SystemTabletConfig = runConfig.AppConfig.GetSystemTabletConfig();
-    }
-
     // setup resource profiles
     AppData->ResourceProfiles = new TResourceProfiles;
     if (runConfig.AppConfig.GetBootstrapConfig().ResourceProfilesSize())

ydb/core/protos/config.proto

@@ -1747,6 +1747,12 @@ message TImmediateControlsConfig {
             MinValue: 8,
             MaxValue: 4096,
             DefaultValue: 256 }];
+
+        optional uint64 MaxTxInFly = 2 [(ControlOptions) = {
+            Description: "Maximum tx queue size for all tablets",
+            MinValue: 0,
+            MaxValue: 1000000,
+            DefaultValue: 10000 }];
     }
 
     message TDSProxyControls {
@@ -2741,10 +2747,6 @@ message TSystemTabletBackupConfig {
     }
 }
 
-message TSystemTabletConfig {
-    optional uint64 MaxTxInFly = 1 [default = 10000];
-}
-
 message TAppConfig {
     option (NMarkers.Root) = true;
     optional TActorSystemConfig ActorSystemConfig = 1;
@@ -2852,7 +2854,6 @@ message TAppConfig {
     optional TStatisticsConfig StatisticsConfig = 113;
     optional TMetricsConfig MetricsConfig = 114;
     optional TSystemTabletBackupConfig SystemTabletBackupConfig = 115;
-    optional TSystemTabletConfig SystemTabletConfig = 116;
 }
 
 message TYdbVersion {

ydb/core/tablet_flat/flat_executor.cpp

@@ -133,12 +133,12 @@ TExecutor::TExecutor(
     , Time(TAppData::TimeProvider)
     , Owner(owner)
     , OwnerActorId(ownerActorId)
-    , MaxTxInFly(AppData()->SystemTabletConfig.GetMaxTxInFly())
     , Emitter(new TIdEmitter)
     , CounterEventsInFlight(new TEvTabletCounters::TInFlightCookie)
     , Stats(new TExecutorStatsImpl())
     , LogFlushDelayOverrideUsec(-1, -1, 60*1000*1000)
     , MaxCommitRedoMB(256, 1, 4096)
+    , MaxTxInFly(10000, 0, 1000000)
 {}
 
 TExecutor::~TExecutor() {
@@ -183,6 +183,7 @@ void TExecutor::Registered(TActorSystem *sys, const TActorId&)
         TControlBoard::RegisterSharedControl(LogFlushDelayOverrideUsec, icb.LogFlushDelayOverrideUsec[static_cast<size_t>(Owner->TabletType())]);
     }
     TControlBoard::RegisterSharedControl(MaxCommitRedoMB, icb.TabletControls.MaxCommitRedoMB);
+    TControlBoard::RegisterSharedControl(MaxTxInFly, icb.TabletControls.MaxTxInFly);
 
     // instantiate alert counters so even never reported alerts are created
     GetServiceCounters(AppData()->Counters, "tablets")->GetCounter("alerts_pending_nodata", true);
@@ -4076,8 +4077,7 @@ void TExecutor::ForceSendCounters() {
 
 float TExecutor::GetRejectProbability() const {
     // Limit number of in-flight TXs
-    // TODO: make configurable
-    if (Stats->TxInFly > MaxTxInFly) {
+    if (Stats->TxInFly > ui64(MaxTxInFly)) {
         HadRejectProbabilityByTxInFly = true;
         return 1.0;
     }
@@ -4113,7 +4113,7 @@ float TExecutor::GetRejectProbability() const {
 }
 
 void TExecutor::MaybeRelaxRejectProbability() {
-    if (HadRejectProbabilityByTxInFly && Stats->TxInFly <= MaxTxInFly ||
+    if (HadRejectProbabilityByTxInFly && Stats->TxInFly <= ui64(MaxTxInFly) ||
         HadRejectProbabilityByOverload)
     {
         HadRejectProbabilityByTxInFly = false;

ydb/core/tablet_flat/flat_executor.h

@@ -386,8 +386,6 @@ class TExecutor
     // Counts the number of times LeaseDuration was increased
     size_t LeaseDurationIncreases = 0;
 
-    const ui64 MaxTxInFly;
-
     struct TLeaseCommit : public TIntrusiveListItem<TLeaseCommit> {
         using TByEndMap = std::multimap<TMonotonic, TLeaseCommit*>;
 
@@ -502,6 +500,7 @@ class TExecutor
 
     TControlWrapper LogFlushDelayOverrideUsec;
     TControlWrapper MaxCommitRedoMB;
+    TControlWrapper MaxTxInFly;
 
     ui64 Stamp() const noexcept;
     void Registered(TActorSystem*, const TActorId&) override;

ydb/tests/olap/test_overloads.py

@@ -5,6 +5,7 @@
 import yatest.common
 import ydb
 import random
+import requests
 
 from ydb.tests.library.harness.kikimr_config import KikimrConfigGenerator
 from ydb.tests.library.harness.kikimr_runner import KiKiMR
@@ -110,13 +111,13 @@ def _setup_ydb_rp(cls):
         logger.info(yatest.common.execute([ydb_path, "-V"], wait=True).stdout.decode("utf-8"))
         config = KikimrConfigGenerator(
             extra_feature_flags={"enable_olap_reject_probability": True},
-            system_tablet_config={"max_tx_in_fly": 0},
         )
         cls.cluster = KiKiMR(config)
         cls.cluster.start()
         node = cls.cluster.nodes[1]
         cls.ydb_client = YdbClient(endpoint=f"grpc://{node.host}:{node.port}", database=f"/{config.domain_name}")
         cls.ydb_client.wait_connection(timeout=60)
+        cls.mon_url = f"http://{node.host}:{node.mon_port}"
 
     def get_row_count(self) -> int:
         return self.ydb_client.query(f"select count(*) as Rows from `{self.table_name}`")[0].rows[0]["Rows"]
@@ -208,8 +209,16 @@ def test_overloads_workload(self, writing_in_flight_requests_count_limit, writin
         logging.info(f"Count rows after insert {self.get_row_count()}")
         assert self.get_row_count() != 0
 
+    def tune_icb(self):
+        response = requests.post(
+            self.mon_url + "/actors/icb",
+            data = "TabletControls.MaxTxInFly=0"
+        )
+        response.raise_for_status()
+
     def test_overloads_reject_probability(self):
         self._setup_ydb_rp()
+        self.tune_icb()
 
         table_path = f"{self.ydb_client.database}/table_for_test_overloads_reject_probability"
         self.ydb_client.query(