Merge ae027eb into 8f381d4

ydb/core/base/events.h

@@ -178,7 +178,8 @@ struct TKikimrEvents : TEvents {
         ES_REPLICATION_SERVICE,
         ES_BACKUP_SERVICE,
         ES_TX_BACKGROUND,
-        ES_SS_BG_TASKS
+        ES_SS_BG_TASKS,
+        ES_LIMITER
     };
 };
 

ydb/core/driver_lib/run/kikimr_services_initializers.cpp

@@ -176,9 +176,12 @@
 #include <ydb/services/metadata/ds_table/service.h>
 #include <ydb/services/metadata/service.h>
 
-#include <ydb/core/tx/conveyor/usage/config.h>
 #include <ydb/core/tx/conveyor/service/service.h>
+#include <ydb/core/tx/conveyor/usage/config.h>
 #include <ydb/core/tx/conveyor/usage/service.h>
+#include <ydb/core/tx/limiter/service/service.h>
+#include <ydb/core/tx/limiter/usage/config.h>
+#include <ydb/core/tx/limiter/usage/service.h>
 
 #include <ydb/core/backup/controller/tablet.h>
 
@@ -2169,6 +2172,28 @@ void TKqpServiceInitializer::InitializeServices(NActors::TActorSystemSetup* setu
     }
 }
 
+TCompDiskLimiterInitializer::TCompDiskLimiterInitializer(const TKikimrRunConfig& runConfig)
+    : IKikimrServicesInitializer(runConfig) {
+}
+
+void TCompDiskLimiterInitializer::InitializeServices(NActors::TActorSystemSetup* setup, const NKikimr::TAppData* appData) {
+    NLimiter::TConfig serviceConfig;
+    if (Config.HasCompDiskLimiterConfig()) {
+        Y_ABORT_UNLESS(serviceConfig.DeserializeFromProto<NLimiter::TCompDiskLimiterPolicy>(Config.GetCompDiskLimiterConfig()));
+    }
+
+    if (serviceConfig.IsEnabled()) {
+        TIntrusivePtr<::NMonitoring::TDynamicCounters> tabletGroup = GetServiceCounters(appData->Counters, "tablets");
+        TIntrusivePtr<::NMonitoring::TDynamicCounters> countersGroup = tabletGroup->GetSubgroup("type", "TX_COMP_DISK_LIMITER");
+
+        auto service = NLimiter::TCompDiskOperator::CreateService(serviceConfig, countersGroup);
+
+        setup->LocalServices.push_back(std::make_pair(
+            NLimiter::TCompDiskOperator::MakeServiceId(NodeId),
+            TActorSetupCmd(service, TMailboxType::HTSwap, appData->UserPoolId)));
+    }
+}
+
 TCompConveyorInitializer::TCompConveyorInitializer(const TKikimrRunConfig& runConfig)
     : IKikimrServicesInitializer(runConfig) {
 }
@@ -2178,9 +2203,7 @@ void TCompConveyorInitializer::InitializeServices(NActors::TActorSystemSetup* se
     if (Config.HasCompConveyorConfig()) {
         Y_ABORT_UNLESS(serviceConfig.DeserializeFromProto(Config.GetCompConveyorConfig()));
     }
-    if (!serviceConfig.HasDefaultFractionOfThreadsCount()) {
-        serviceConfig.SetDefaultFractionOfThreadsCount(0.33);
-    }
+    serviceConfig.SetWorkersCount(2);
 
     if (serviceConfig.IsEnabled()) {
         TIntrusivePtr<::NMonitoring::TDynamicCounters> tabletGroup = GetServiceCounters(appData->Counters, "tablets");

ydb/core/driver_lib/run/kikimr_services_initializers.h

@@ -391,6 +391,12 @@ class TKqpServiceInitializer : public IKikimrServicesInitializer {
     IGlobalObjectStorage& GlobalObjects;
 };
 
+class TCompDiskLimiterInitializer: public IKikimrServicesInitializer {
+public:
+    TCompDiskLimiterInitializer(const TKikimrRunConfig& runConfig);
+    void InitializeServices(NActors::TActorSystemSetup* setup, const NKikimr::TAppData* appData) override;
+};
+
 class TCompConveyorInitializer: public IKikimrServicesInitializer {
 public:
     TCompConveyorInitializer(const TKikimrRunConfig& runConfig);

ydb/core/driver_lib/run/ya.make

@@ -112,6 +112,7 @@ PEERDIR(
     ydb/core/tx/columnshard
     ydb/core/tx/coordinator
     ydb/core/tx/conveyor/service
+    ydb/core/tx/limiter/service
     ydb/core/tx/datashard
     ydb/core/tx/long_tx_service
     ydb/core/tx/long_tx_service/public

ydb/core/protos/config.proto

@@ -594,6 +594,12 @@ message TConveyorConfig {
     optional double DefaultFractionOfThreadsCount = 4;
 }
 
+message TLimiterConfig {
+    optional bool Enabled = 1 [default = true];
+    optional uint64 Limit = 2;
+    optional uint64 PeriodMilliSeconds = 3 [default = 1000];
+}
+
 message TExternalIndexConfig {
     optional bool Enabled = 1 [default = true];
     optional TInternalRequestConfig RequestConfig = 2;
@@ -1840,6 +1846,7 @@ message TAppConfig {
     optional TS3ProxyResolverConfig S3ProxyResolverConfig = 76;
     optional TBackgroundCleaningConfig BackgroundCleaningConfig = 77;
     optional TBlobCacheConfig BlobCacheConfig = 78;
+    optional TLimiterConfig CompDiskLimiterConfig = 79;
 
     repeated TNamedConfig NamedConfigs = 100;
     optional string ClusterYamlConfig = 101;

ydb/core/tx/columnshard/columnshard__write_index.cpp

@@ -6,10 +6,34 @@
 #include "engines/changes/abstract/abstract.h"
 #include "engines/writer/compacted_blob_constructor.h"
 
+#include <ydb/core/tx/limiter/usage/abstract.h>
+#include <ydb/core/tx/limiter/usage/service.h>
+
 #include <ydb/library/actors/core/log.h>
 
 namespace NKikimr::NColumnShard {
 
+class TDiskResourcesRequest: public NLimiter::IResourceRequest {
+private:
+    using TBase = NLimiter::IResourceRequest;
+    std::shared_ptr<NOlap::TCompactedWriteController> WriteController;
+    const ui64 TabletId;
+
+private:
+    virtual void DoOnResourceAllocated() override {
+        NActors::TActivationContext::AsActorContext().Register(CreateWriteActor(TabletId, WriteController, TInstant::Max()));
+    }
+
+public:
+    TDiskResourcesRequest(const std::shared_ptr<NOlap::TCompactedWriteController>& writeController, const ui64 tabletId)
+        : TBase(writeController->GetWriteVolume())
+        , WriteController(writeController)
+        , TabletId(tabletId)
+    {
+
+    }
+};
+
 void TColumnShard::Handle(TEvPrivate::TEvWriteIndex::TPtr& ev, const TActorContext& ctx) {
     auto putStatus = ev->Get()->GetPutStatus();
 
@@ -32,7 +56,7 @@ void TColumnShard::Handle(TEvPrivate::TEvWriteIndex::TPtr& ev, const TActorConte
             if (*needDraftTransaction) {
                 Execute(new TTxWriteDraft(this, writeController));
             } else {
-                ctx.Register(CreateWriteActor(TabletID(), writeController, TInstant::Max()));
+                NLimiter::TCompDiskOperator::AskResource(std::make_shared<TDiskResourcesRequest>(writeController, TabletID()));
             }
         }
     } else {

ydb/core/tx/columnshard/engines/writer/compacted_blob_constructor.cpp

@@ -23,6 +23,7 @@ TCompactedWriteController::TCompactedWriteController(const TActorId& dstActor, T
         for (auto&& b : portionWithBlobs.GetBlobs()) {
             auto& task = AddWriteTask(TBlobWriteInfo::BuildWriteTask(b.GetBlob(), changes.MutableBlobsAction().GetWriting(b.GetOperator()->GetStorageId())));
             b.RegisterBlobId(portionWithBlobs, task.GetBlobId());
+            WriteVolume += b.GetSize();
         }
     }
 }

ydb/core/tx/columnshard/engines/writer/compacted_blob_constructor.h

@@ -13,11 +13,17 @@ class TCompactedWriteController : public NColumnShard::IWriteController {
 private:
     TAutoPtr<NColumnShard::TEvPrivate::TEvWriteIndex> WriteIndexEv;
     TActorId DstActor;
+    ui64 WriteVolume = 0;
+
 protected:
     void DoOnReadyResult(const NActors::TActorContext& ctx, const NColumnShard::TBlobPutResult::TPtr& putResult) override;
     virtual void DoAbort(const TString& reason) override;
+
 public:
     const TBlobsAction& GetBlobsAction();
+    ui64 GetWriteVolume() const {
+        return WriteVolume;
+    }
 
     TCompactedWriteController(const TActorId& dstActor, TAutoPtr<NColumnShard::TEvPrivate::TEvWriteIndex> writeEv);
     ~TCompactedWriteController();

ydb/core/tx/limiter/service/service.cpp

@@ -0,0 +1,59 @@
+#include "service.h"
+
+namespace NKikimr::NLimiter {
+
+TLimiterActor::TLimiterActor(const TConfig& config, const TString& limiterName, TIntrusivePtr<::NMonitoring::TDynamicCounters> baseCounters)
+    : LimiterName(limiterName)
+    , Config(config)
+    , Counters(LimiterName, baseCounters)
+{
+    Counters.InProgressLimit->Set(Config.GetLimit());
+}
+
+void TLimiterActor::HandleMain(TEvExternal::TEvAskResource::TPtr& ev) {
+    const auto now = TMonotonic::Now();
+    if (VolumeInFlight + ev->Get()->GetRequest()->GetVolume() <= Config.GetLimit()) {
+        VolumeInFlight += ev->Get()->GetRequest()->GetVolume();
+        RequestsInFlight.emplace_back(now, ev->Get()->GetRequest()->GetVolume());
+        if (RequestsInFlight.size() == 1) {
+            Schedule(now + Config.GetPeriod(), new NActors::TEvents::TEvWakeup());
+        }
+        ev->Get()->GetRequest()->OnResourceAllocated();
+        Counters.InProgressStart->Inc();
+    } else {
+        RequestsQueue.emplace_back(now, ev->Get()->GetRequest());
+        VolumeInWaiting += ev->Get()->GetRequest()->GetVolume();
+    }
+    Counters.InProgressCount->Set(RequestsInFlight.size());
+    Counters.InProgressVolume->Set(VolumeInFlight);
+    Counters.WaitingQueueCount->Set(RequestsQueue.size());
+    Counters.WaitingQueueVolume->Set(VolumeInWaiting);
+}
+
+void TLimiterActor::HandleMain(NActors::TEvents::TEvWakeup::TPtr& /*ev*/) {
+    const auto now = TMonotonic::Now();
+    AFL_VERIFY(RequestsInFlight.size());
+    while (RequestsInFlight.size() && now + Config.GetPeriod() <= RequestsInFlight.front().GetInstant()) {
+        AFL_VERIFY(RequestsInFlight.front().GetVolume() <= VolumeInFlight);
+        VolumeInFlight = VolumeInFlight - RequestsInFlight.front().GetVolume();
+        RequestsInFlight.pop_front();
+    }
+    if (RequestsInFlight.size()) {
+        Schedule(RequestsInFlight.front().GetInstant() + Config.GetPeriod(), new NActors::TEvents::TEvWakeup());
+    }
+    while (RequestsQueue.size() && (!VolumeInFlight || VolumeInFlight + RequestsQueue.front().GetRequest()->GetVolume() <= Config.GetLimit())) {
+        Counters.WaitingHistogram->Collect((i64)(now - RequestsQueue.front().GetInstant()).MilliSeconds(), 1);
+        VolumeInFlight += RequestsQueue.front().GetRequest()->GetVolume();
+        RequestsInFlight.emplace_back(now, RequestsQueue.front().GetRequest()->GetVolume());
+        RequestsQueue.front().GetRequest()->OnResourceAllocated();
+        VolumeInWaiting -= RequestsQueue.front().GetRequest()->GetVolume();
+        RequestsQueue.pop_front();
+        Counters.InProgressStart->Inc();
+    }
+    Counters.InProgressCount->Set(RequestsInFlight.size());
+    Counters.InProgressVolume->Set(VolumeInFlight);
+    Counters.WaitingQueueCount->Set(RequestsQueue.size());
+    Counters.WaitingQueueVolume->Set(VolumeInWaiting);
+}
+
+}

ydb/core/tx/limiter/service/service.h

@@ -0,0 +1,101 @@
+#pragma once
+#include <ydb/core/tx/columnshard/counters/common/owner.h>
+#include <ydb/core/tx/limiter/usage/abstract.h>
+#include <ydb/core/tx/limiter/usage/config.h>
+#include <ydb/core/tx/limiter/usage/events.h>
+
+#include <ydb/library/actors/core/actor_bootstrapped.h>
+#include <ydb/library/actors/core/log.h>
+#include <ydb/library/accessor/accessor.h>
+
+#include <library/cpp/monlib/dynamic_counters/counters.h>
+
+#include <queue>
+
+namespace NKikimr::NLimiter {
+
+class TCounters: public NColumnShard::TCommonCountersOwner {
+private:
+    using TBase = NColumnShard::TCommonCountersOwner;
+public:
+    const ::NMonitoring::TDynamicCounters::TCounterPtr WaitingQueueCount;
+    const ::NMonitoring::TDynamicCounters::TCounterPtr WaitingQueueVolume;
+    const ::NMonitoring::TDynamicCounters::TCounterPtr InProgressLimit;
+
+    const ::NMonitoring::TDynamicCounters::TCounterPtr InProgressCount;
+    const ::NMonitoring::TDynamicCounters::TCounterPtr InProgressVolume;
+
+    const ::NMonitoring::TDynamicCounters::TCounterPtr InProgressStart;
+
+    const ::NMonitoring::THistogramPtr WaitingHistogram;
+
+    TCounters(const TString& limiterName, TIntrusivePtr<::NMonitoring::TDynamicCounters> baseSignals)
+        : TBase("Limiter/" + limiterName, baseSignals)
+        , WaitingQueueCount(TBase::GetValue("WaitingQueue/Count"))
+        , WaitingQueueVolume(TBase::GetValue("WaitingQueue/Volume"))
+        , InProgressLimit(TBase::GetValue("InProgress/Limit/Volume"))
+        , InProgressCount(TBase::GetValue("InProgress/Count"))
+        , InProgressVolume(TBase::GetValue("InProgress/Volume"))
+        , InProgressStart(TBase::GetDeriviative("InProgress"))
+        , WaitingHistogram(TBase::GetHistogram("Waiting", NMonitoring::ExponentialHistogram(20, 2))) {
+    }
+};
+
+class TLimiterActor: public NActors::TActorBootstrapped<TLimiterActor> {
+private:
+    const TString LimiterName;
+    const TConfig Config;
+    TCounters Counters;
+    class TResourceRequest {
+    private:
+        YDB_READONLY(TMonotonic, Instant, TMonotonic::Zero());
+        YDB_READONLY_DEF(std::shared_ptr<IResourceRequest>, Request);
+    public:
+        TResourceRequest(const TMonotonic instant, const std::shared_ptr<IResourceRequest>& req)
+            : Instant(instant)
+            , Request(req) {
+
+        }
+    };
+
+    class TResourceRequestInFlight {
+    private:
+        YDB_READONLY(TMonotonic, Instant, TMonotonic::Zero());
+        YDB_READONLY(ui64, Volume, 0);
+    public:
+        TResourceRequestInFlight(const TMonotonic instant, const ui64 volume)
+            : Instant(instant)
+            , Volume(volume) {
+
+        }
+    };
+
+    ui64 VolumeInFlight = 0;
+    ui64 VolumeInWaiting = 0;
+    std::deque<TResourceRequest> RequestsQueue;
+    std::deque<TResourceRequestInFlight> RequestsInFlight;
+
+    void HandleMain(TEvExternal::TEvAskResource::TPtr& ev);
+    void HandleMain(NActors::TEvents::TEvWakeup::TPtr& ev);
+
+public:
+
+    STATEFN(StateMain) {
+        switch (ev->GetTypeRewrite()) {
+            hFunc(TEvExternal::TEvAskResource, HandleMain);
+            hFunc(NActors::TEvents::TEvWakeup, HandleMain);
+        default:
+            AFL_ERROR(NKikimrServices::TX_LIMITER)("limiter", LimiterName)("problem", "unexpected event")("type", ev->GetTypeRewrite());
+            AFL_VERIFY_DEBUG(false)("type", ev->GetTypeRewrite());
+            break;
+        }
+    }
+
+    TLimiterActor(const TConfig& config, const TString& limiterName, TIntrusivePtr<::NMonitoring::TDynamicCounters> baseCounters);
+
+    void Bootstrap() {
+        Become(&TLimiterActor::StateMain);
+    }
+};
+
+}

ydb/core/tx/limiter/service/ya.make

@@ -0,0 +1,12 @@
+LIBRARY()
+
+SRCS(
+    service.cpp
+)
+
+PEERDIR(
+    ydb/core/tx/limiter/usage
+    ydb/core/protos
+)
+
+END()

ydb/core/tx/limiter/usage/abstract.cpp

@@ -0,0 +1,5 @@
+#include "abstract.h"
+
+namespace NKikimr::NLimiter {
+
+}

ydb/core/tx/limiter/usage/abstract.h

@@ -0,0 +1,24 @@
+#pragma once
+#include <memory>
+#include <ydb/core/tx/columnshard/counters/common/owner.h>
+
+namespace NKikimr::NLimiter {
+class IResourceRequest {
+private:
+    YDB_READONLY(ui64, Volume, 0);
+    virtual void DoOnResourceAllocated() = 0;
+public:
+    void OnResourceAllocated() {
+        return DoOnResourceAllocated();
+    }
+
+    virtual ~IResourceRequest() = default;
+
+    IResourceRequest(const ui64 volume)
+        : Volume(volume)
+    {
+
+    }
+};
+
+}

ydb/core/tx/limiter/usage/config.cpp

@@ -0,0 +1,12 @@
+#include "config.h"
+#include <util/string/builder.h>
+
+namespace NKikimr::NLimiter {
+
+TString TConfig::DebugString() const {
+    TStringBuilder sb;
+    sb << "Period=" << Period << ";Limit=" << Limit << ";Enabled=" << EnabledFlag << ";";
+    return sb;
+}
+
+}