Ydb stable 22-4-31

x-stable-origin-commit: 2bc59c7eeae4a8f3d396867de193d1375dd388ce

ydb/core/blobstorage/nodewarden/node_warden_impl.cpp

@@ -401,8 +401,14 @@ void TNodeWarden::Handle(TEvStatusUpdate::TPtr ev) {
     auto *msg = ev->Get();
     const TVSlotId vslotId(msg->NodeId, msg->PDiskId, msg->VSlotId);
     if (const auto it = LocalVDisks.find(vslotId); it != LocalVDisks.end() && it->second.Status != msg->Status) {
-        it->second.Status = msg->Status;
+        auto& vdisk = it->second;
+        vdisk.Status = msg->Status;
         SendDiskMetrics(false);
+
+        if (msg->Status == NKikimrBlobStorage::EVDiskStatus::READY && vdisk.WhiteboardVDiskId) {
+            Send(WhiteboardId, new NNodeWhiteboard::TEvWhiteboard::TEvVDiskDropDonors(*vdisk.WhiteboardVDiskId,
+                vdisk.WhiteboardInstanceGuid, NNodeWhiteboard::TEvWhiteboard::TEvVDiskDropDonors::TDropAllDonors()));
+        }
     }
 }
 

ydb/core/blobstorage/nodewarden/node_warden_impl.h

@@ -224,6 +224,12 @@ namespace NKikimr::NStorage {
                 return MakeBlobStorageVDiskID(NodeId, PDiskId, VDiskSlotId);
             }
 
+            void Serialize(NKikimrBlobStorage::TVSlotId *proto) const {
+                proto->SetNodeId(NodeId);
+                proto->SetPDiskId(PDiskId);
+                proto->SetVSlotId(VDiskSlotId);
+            }
+
             auto AsTuple() const { return std::make_tuple(NodeId, PDiskId, VDiskSlotId); }
             friend bool operator <(const TVSlotId& x, const TVSlotId& y) { return x.AsTuple() < y.AsTuple(); }
             friend bool operator <=(const TVSlotId& x, const TVSlotId& y) { return x.AsTuple() <= y.AsTuple(); }
@@ -253,6 +259,7 @@ namespace NKikimr::NStorage {
 
             // Last VDiskId reported to Node Whiteboard.
             std::optional<TVDiskID> WhiteboardVDiskId;
+            ui64 WhiteboardInstanceGuid;
 
             bool SlayInFlight = false;
 

ydb/core/blobstorage/nodewarden/node_warden_vdisk.cpp

@@ -175,6 +175,7 @@ namespace NKikimr::NStorage {
         Send(WhiteboardId, new NNodeWhiteboard::TEvWhiteboard::TEvVDiskStateUpdate(vdiskId, groupInfo->GetStoragePoolName(),
             vslotId.PDiskId, vslotId.VDiskSlotId, pdiskGuid, kind, donorMode, whiteboardInstanceGuid, std::move(donors)));
         vdisk.WhiteboardVDiskId.emplace(vdiskId);
+        vdisk.WhiteboardInstanceGuid = whiteboardInstanceGuid;
 
         // create an actor
         auto *as = TActivationContext::ActorSystem();
@@ -281,9 +282,19 @@ namespace NKikimr::NStorage {
 
     void TNodeWarden::Handle(TEvBlobStorage::TEvDropDonor::TPtr ev) {
         auto *msg = ev->Get();
-        STLOG(PRI_INFO, BS_NODE, NW34, "TEvDropDonor", (VSlotId, TVSlotId(msg->NodeId, msg->PDiskId, msg->VSlotId)),
-            (VDiskId, msg->VDiskId));
+        const TVSlotId vslotId(msg->NodeId, msg->PDiskId, msg->VSlotId);
+        STLOG(PRI_INFO, BS_NODE, NW34, "TEvDropDonor", (VSlotId, vslotId), (VDiskId, msg->VDiskId));
         SendDropDonorQuery(msg->NodeId, msg->PDiskId, msg->VSlotId, msg->VDiskId);
+
+        if (const auto it = LocalVDisks.find(vslotId); it != LocalVDisks.end()) {
+            const auto& vdisk = it->second;
+            if (vdisk.WhiteboardVDiskId) {
+                NKikimrBlobStorage::TVSlotId id;
+                vslotId.Serialize(&id);
+                Send(WhiteboardId, new NNodeWhiteboard::TEvWhiteboard::TEvVDiskDropDonors(*vdisk.WhiteboardVDiskId,
+                    vdisk.WhiteboardInstanceGuid, {id}));
+            }
+        }
     }
 
     void TNodeWarden::UpdateGroupInfoForDisk(TVDiskRecord& vdisk, const TIntrusivePtr<TBlobStorageGroupInfo>& newInfo) {

ydb/core/grpc_services/CMakeLists.txt

@@ -53,6 +53,7 @@ target_link_libraries(ydb-core-grpc_services PUBLIC
   cpp-client-resources
 )
 target_sources(ydb-core-grpc_services PRIVATE
+  ${CMAKE_SOURCE_DIR}/ydb/core/grpc_services/audit_log.cpp
   ${CMAKE_SOURCE_DIR}/ydb/core/grpc_services/grpc_endpoint_publish_actor.cpp
   ${CMAKE_SOURCE_DIR}/ydb/core/grpc_services/grpc_helper.cpp
   ${CMAKE_SOURCE_DIR}/ydb/core/grpc_services/grpc_mon.cpp

ydb/core/grpc_services/audit_log.cpp

@@ -0,0 +1,21 @@
+#include "defs.h"
+#include "audit_log.h"
+
+#include "base/base.h"
+
+namespace NKikimr {
+namespace NGRpcService {
+
+void AuditLog(const IRequestProxyCtx* reqCtx, const TString& database,
+              const TString& subject, const TActorContext& ctx)
+{
+    LOG_NOTICE_S(ctx, NKikimrServices::GRPC_SERVER, "AUDIT: "
+        << "request name: " << reqCtx->GetRequestName()
+        << ", database: " << database
+        << ", peer: " << reqCtx->GetPeerName()
+        << ", subject: " << subject);
+}
+
+}
+}
+

ydb/core/grpc_services/audit_log.h

@@ -0,0 +1,12 @@
+#pragma once
+
+namespace NKikimr {
+namespace NGRpcService {
+
+class IRequestProxyCtx;
+
+void AuditLog(const IRequestProxyCtx* reqCtx, const TString& database,
+              const TString& subject, const TActorContext& ctx);
+
+}
+}

ydb/core/grpc_services/grpc_request_check_actor.h

@@ -1,5 +1,6 @@
 #pragma once
 #include "defs.h"
+#include "audit_log.h"
 #include "service_ratelimiter_events.h"
 #include "local_rate_limiter.h"
 #include "operation_helpers.h"
@@ -112,6 +113,10 @@ class TGrpcRequestCheckActor
             }
         }
 
+        if (AppData(ctx)->FeatureFlags.GetEnableGrpcAudit()) {
+            AuditLog(GrpcRequestBaseCtx_, CheckedDatabaseName_, GetSubject(), ctx);
+        }
+
         // Simple rps limitation
         static NRpcService::TRlConfig rpsRlConfig(
             "serverless_rt_coordination_node_path",
@@ -205,6 +210,11 @@ class TGrpcRequestCheckActor
     }
 
 private:
+    TString GetSubject() const {
+        const auto sid = TBase::GetUserSID();
+        return sid ? sid : "no subject";
+    }
+
     static NYql::TIssues GetRlIssues(const Ydb::RateLimiter::AcquireResourceResponse& resp) {
         NYql::TIssues opIssues;
         NYql::IssuesFromMessage(resp.operation().issues(), opIssues);

ydb/core/mind/hive/balancer.cpp

@@ -146,6 +146,8 @@ class THiveBalancer : public NActors::TActorBootstrapped<THiveBalancer>, public
         if (RecheckOnFinish && MaxMovements != 0 && Movements >= MaxMovements) {
             BLOG_D("Balancer initiated recheck");
             Hive->ProcessTabletBalancer();
+        } else {
+            Send(Hive->SelfId(), new TEvPrivate::TEvBalancerOut());
         }
         return IActor::PassAway();
     }
@@ -239,6 +241,7 @@ class THiveBalancer : public NActors::TActorBootstrapped<THiveBalancer>, public
                     tablets.emplace_back(tablet);
                 }
             }
+            BLOG_TRACE("Balancer on node " << node->Id <<  ": " << tablets.size() << "/" << nodeTablets.size() << " tablets is suitable for balancing");
             if (!tablets.empty()) {
                 switch (Hive->GetTabletBalanceStrategy()) {
                 case NKikimrConfig::THiveConfig::HIVE_TABLET_BALANCE_STRATEGY_OLD_WEIGHTED_RANDOM:

ydb/core/mind/hive/hive_events.h

@@ -23,6 +23,7 @@ struct TEvPrivate {
         EvUnlockTabletReconnectTimeout,
         EvProcessPendingOperations,
         EvRestartComplete,
+        EvBalancerOut,
         EvEnd
     };
 
@@ -37,7 +38,7 @@ struct TEvPrivate {
     };
 
     struct TEvProcessBootQueue : TEventLocal<TEvProcessBootQueue, EvProcessBootQueue> {};
-    
+
     struct TEvPostponeProcessBootQueue : TEventLocal<TEvPostponeProcessBootQueue, EvPostponeProcessBootQueue> {};
 
     struct TEvProcessDisconnectNode : TEventLocal<TEvProcessDisconnectNode, EvProcessDisconnectNode> {
@@ -78,6 +79,8 @@ struct TEvPrivate {
     };
 
     struct TEvProcessPendingOperations : TEventLocal<TEvProcessPendingOperations, EvProcessPendingOperations> {};
+
+    struct TEvBalancerOut : TEventLocal<TEvBalancerOut, EvBalancerOut> {};
 };
 
 } // NHive

ydb/core/mind/hive/hive_impl.cpp

@@ -308,6 +308,10 @@ void THive::Handle(TEvPrivate::TEvProcessPendingOperations::TPtr&) {
     BLOG_D("Handle ProcessPendingOperations");
 }
 
+void THive::Handle(TEvPrivate::TEvBalancerOut::TPtr&) {
+    BLOG_D("Handle BalancerOut");
+}
+
 void THive::Handle(TEvHive::TEvBootTablet::TPtr& ev) {
     TTabletId tabletId = ev->Get()->Record.GetTabletID();
     TTabletInfo* tablet = FindTablet(tabletId);
@@ -554,6 +558,14 @@ void THive::BuildCurrentConfig() {
     for (const NKikimrConfig::THiveTabletPreference& tabletPreference : CurrentConfig.GetDefaultTabletPreference()) {
         DefaultDataCentersPreference[tabletPreference.GetType()] = tabletPreference.GetDataCentersPreference();
     }
+    BalancerIgnoreTabletTypes.clear();
+    for (auto i : CurrentConfig.GetBalancerIgnoreTabletTypes()) {
+        const auto type = TTabletTypes::EType(i);
+        if (IsValidTabletType(type)) {
+            BalancerIgnoreTabletTypes.emplace_back(type);
+        }
+    }
+    MakeTabletTypeSet(BalancerIgnoreTabletTypes);
 }
 
 void THive::Cleanup() {
@@ -1612,6 +1624,7 @@ void THive::FillTabletInfo(NKikimrHive::TEvResponseHiveInfo& response, ui64 tabl
         auto& tabletInfo = *response.AddTablets();
         tabletInfo.SetTabletID(tabletId);
         tabletInfo.SetTabletType(info->Type);
+        tabletInfo.SetObjectId(info->ObjectId);
         tabletInfo.SetState(static_cast<ui32>(info->State));
         tabletInfo.SetTabletBootMode(info->BootMode);
         tabletInfo.SetVolatileState(info->GetVolatileState());
@@ -1620,6 +1633,9 @@ void THive::FillTabletInfo(NKikimrHive::TEvResponseHiveInfo& response, ui64 tabl
         tabletInfo.MutableTabletOwner()->SetOwnerIdx(info->Owner.second);
         tabletInfo.SetGeneration(info->KnownGeneration);
         tabletInfo.MutableObjectDomain()->CopyFrom(info->ObjectDomain);
+        if (info->BalancerPolicy != NKikimrHive::EBalancerPolicy::POLICY_BALANCE) {
+            tabletInfo.SetBalancerPolicy(info->BalancerPolicy);
+        }
         if (!info->IsRunning()) {
             tabletInfo.SetLastAliveTimestamp(info->Statistics.GetLastAliveTimestamp());
         }
@@ -2437,6 +2453,7 @@ STFUNC(THive::StateWork) {
         hFunc(NSysView::TEvSysView::TEvGetTabletsRequest, Handle);
         hFunc(TEvHive::TEvRequestTabletOwners, Handle);
         hFunc(TEvHive::TEvTabletOwnersReply, Handle);
+        hFunc(TEvPrivate::TEvBalancerOut, Handle);
     default:
         if (!HandleDefaultEvents(ev, ctx)) {
             BLOG_W("THive::StateWork unhandled event type: " << ev->GetTypeRewrite()

ydb/core/mind/hive/hive_impl.h

@@ -140,6 +140,8 @@ TString GetConditionalRedString(const TString& str, bool condition);
 TString GetDataCenterName(ui64 dataCenterId);
 TString LongToShortTabletName(const TString& longTabletName);
 TString GetLocationString(const NActors::TNodeLocation& location);
+void MakeTabletTypeSet(std::vector<TTabletTypes::EType>& list);
+bool IsValidTabletType(TTabletTypes::EType type);
 
 class THive : public TActor<THive>, public TTabletExecutedFlat, public THiveSharedSettings {
 public:
@@ -404,6 +406,9 @@ class THive : public TActor<THive>, public TTabletExecutedFlat, public THiveShar
     std::unordered_map<TDataCenterId, std::unordered_set<TNodeId>> RegisteredDataCenterNodes;
     std::unordered_set<TNodeId> ConnectedNodes;
 
+    // normalized to be sorted list of unique values
+    std::vector<TTabletTypes::EType> BalancerIgnoreTabletTypes; // built from CurrentConfig
+
     // to be removed later
     bool TabletOwnersSynced = false;
     // to be removed later
@@ -482,6 +487,7 @@ class THive : public TActor<THive>, public TTabletExecutedFlat, public THiveShar
     void Handle(TEvPrivate::TEvProcessTabletBalancer::TPtr&);
     void Handle(TEvPrivate::TEvUnlockTabletReconnectTimeout::TPtr&);
     void Handle(TEvPrivate::TEvProcessPendingOperations::TPtr&);
+    void Handle(TEvPrivate::TEvBalancerOut::TPtr&);
     void Handle(TEvTxProxySchemeCache::TEvNavigateKeySetResult::TPtr& ev);
     void Handle(NConsole::TEvConsole::TEvConfigNotificationRequest::TPtr& ev);
     void Handle(NConsole::TEvConfigsDispatcher::TEvSetConfigSubscriptionResponse::TPtr& ev);
@@ -748,6 +754,12 @@ class THive : public TActor<THive>, public TTabletExecutedFlat, public THiveShar
         return initialMaximum;
     }
 
+    bool IsInBalancerIgnoreList(TTabletTypes::EType type) const {
+        const auto& ignoreList = BalancerIgnoreTabletTypes;
+        auto found = std::find(ignoreList.begin(), ignoreList.end(), type);
+        return (found != ignoreList.end());
+    }
+
     static void ActualizeRestartStatistics(google::protobuf::RepeatedField<google::protobuf::uint64>& restartTimestamps, ui64 barrier);
     static bool IsSystemTablet(TTabletTypes::EType type);
 

ydb/core/mind/hive/hive_schema.h

@@ -93,6 +93,8 @@ struct Schema : NIceDb::Schema {
         struct DataCentersPreference : Column<121, NScheme::NTypeIds::String> { using Type = NKikimrHive::TDataCentersPreference; };
         struct AllowedDataCenterIds : Column<122, NScheme::NTypeIds::String> { using Type = TVector<TString>; };
 
+        struct BalancerPolicy : Column<123, NScheme::NTypeIds::Uint64> { using Type = NKikimrHive::EBalancerPolicy; static constexpr NKikimrHive::EBalancerPolicy Default = NKikimrHive::EBalancerPolicy::POLICY_BALANCE; };
+
         using TKey = TableKey<ID>;
         using TColumns = TableColumns<
             ID,
@@ -118,7 +120,8 @@ struct Schema : NIceDb::Schema {
             ReassignReason,
             Statistics,
             DataCentersPreference,
-            AllowedDataCenterIds
+            AllowedDataCenterIds,
+            BalancerPolicy
         >;
     };
 

ydb/core/mind/hive/hive_statics.cpp

@@ -340,5 +340,16 @@ TString GetLocationString(const NActors::TNodeLocation& location) {
     return proto.ShortDebugString();
 }
 
+void MakeTabletTypeSet(std::vector<TTabletTypes::EType>& list) {
+    std::sort(list.begin(), list.end());
+    list.erase(std::unique(list.begin(), list.end()), list.end());
+}
+
+bool IsValidTabletType(TTabletTypes::EType type) {
+    return (type > TTabletTypes::Unknown
+            && type < TTabletTypes::Reserved40
+            );
+}
+
 } // NHive
 } // NKikimr