make TabletOwners synchronization more safe KIKIMR-14545

ref:916aa66629b7d7a4f7430766720b3dee316a432f

8 files changed, 69 insertions, 19 deletions

diff --git a/ydb/core/mind/hive/CMakeLists.txt b/ydb/core/mind/hive/CMakeLists.txt
index 4995dceb2a..d47111b5c9 100644
--- a/ydb/core/mind/hive/CMakeLists.txt
+++ b/ydb/core/mind/hive/CMakeLists.txt
@@ -62,6 +62,7 @@ target_sources(core-mind-hive PRIVATE
   ${CMAKE_SOURCE_DIR}/ydb/core/mind/hive/tx__register_node.cpp
   ${CMAKE_SOURCE_DIR}/ydb/core/mind/hive/tx__release_tablets.cpp
   ${CMAKE_SOURCE_DIR}/ydb/core/mind/hive/tx__release_tablets_reply.cpp
+  ${CMAKE_SOURCE_DIR}/ydb/core/mind/hive/tx__request_tablet_owners.cpp
   ${CMAKE_SOURCE_DIR}/ydb/core/mind/hive/tx__request_tablet_seq.cpp
   ${CMAKE_SOURCE_DIR}/ydb/core/mind/hive/tx__response_tablet_seq.cpp
   ${CMAKE_SOURCE_DIR}/ydb/core/mind/hive/tx__restart_tablet.cpp
diff --git a/ydb/core/mind/hive/hive_impl.cpp b/ydb/core/mind/hive/hive_impl.cpp
index 93c0afcd25..bbe2f60200 100644
--- a/ydb/core/mind/hive/hive_impl.cpp
+++ b/ydb/core/mind/hive/hive_impl.cpp
@@ -452,12 +452,14 @@ void THive::Handle(TEvPrivate::TEvBootTablets::TPtr&) {
     } else {
         BLOG_D("SubDomain Hive is ready");
 
-        // this code should be removed later
-        THolder<TEvHive::TEvRequestTabletOwners> request(new TEvHive::TEvRequestTabletOwners());
-        request->Record.SetOwnerID(TabletID());
-        BLOG_D("Requesting TabletOwners from the Root");
-        SendToRootHivePipe(request.Release());
-        // this code should be removed later
+        if (!TabletOwnersSynced) {
+            // this code should be removed later
+            THolder<TEvHive::TEvRequestTabletOwners> request(new TEvHive::TEvRequestTabletOwners());
+            request->Record.SetOwnerID(TabletID());
+            BLOG_D("Requesting TabletOwners from the Root");
+            SendToRootHivePipe(request.Release());
+            // this code should be removed later
+        }
     }
     if (!tabletsToReleaseFromParent.empty()) {
         THolder<TEvHive::TEvReleaseTablets> request(new TEvHive::TEvReleaseTablets());
@@ -2551,18 +2553,7 @@ void THive::Handle(TEvHive::TEvReleaseTabletsReply::TPtr& ev) {
 
 void THive::Handle(TEvHive::TEvRequestTabletOwners::TPtr& ev) {
     BLOG_D("Handle TEvHive::TEvRequestTabletOwners(" << ev->Get()->Record.ShortDebugString() << ")");
-    auto ownerId = ev->Get()->Record.GetOwnerID();
-    std::vector<TSequencer::TSequence> sequences;
-    Keeper.GetOwnedSequences(ownerId, sequences);
-    BLOG_D("TEvHive::TEvRequestTabletOwners() - replying with " << sequences.size() << " sequences");
-    THolder<TEvHive::TEvTabletOwnersReply> reply(new TEvHive::TEvTabletOwnersReply());
-    for (const auto& seq : sequences) {
-        auto* tabletOwners = reply->Record.AddTabletOwners();
-        tabletOwners->SetOwnerID(ownerId);
-        tabletOwners->SetBegin(seq.Begin);
-        tabletOwners->SetEnd(seq.End);
-    }
-    Send(ev->Sender, reply.Release());
+    Execute(CreateRequestTabletOwners(std::move(ev)));
 }
 
 void THive::Handle(TEvHive::TEvTabletOwnersReply::TPtr& ev) {
diff --git a/ydb/core/mind/hive/hive_impl.h b/ydb/core/mind/hive/hive_impl.h
index 2f0d6fd1b7..cc1b642cad 100644
--- a/ydb/core/mind/hive/hive_impl.h
+++ b/ydb/core/mind/hive/hive_impl.h
@@ -235,6 +235,7 @@ protected:
     friend class TTxSwitchDrainOn;
     friend class TTxSwitchDrainOff;
     friend class TTxTabletOwnersReply;
+    friend class TTxRequestTabletOwners;
 
     friend class TDeleteTabletActor;
 
@@ -290,6 +291,7 @@ protected:
     ITransaction* CreateSwitchDrainOn(TNodeId nodeId, TDrainSettings settings, const TActorId& initiator);
     ITransaction* CreateSwitchDrainOff(TNodeId nodeId, TDrainSettings settings, NKikimrProto::EReplyStatus status, ui32 movements);
     ITransaction* CreateTabletOwnersReply(TEvHive::TEvTabletOwnersReply::TPtr event);
+    ITransaction* CreateRequestTabletOwners(TEvHive::TEvRequestTabletOwners::TPtr event);
 
 public:
     TDomainsView DomainsView;
@@ -423,6 +425,10 @@ protected:
     std::unordered_map<TTabletTypes::EType, NKikimrHive::TDataCentersPreference> DefaultDataCentersPreference;
     std::unordered_set<TDataCenterId> RegisteredDataCenterIds;
 
+    // to be removed later
+    bool TabletOwnersSynced = false;
+    // to be removed later
+
     void OnActivateExecutor(const TActorContext& ctx) override;
     void DefaultSignalTabletActive(const TActorContext& ctx) override;
     void OnDetach(const TActorContext&) override;
diff --git a/ydb/core/mind/hive/hive_schema.h b/ydb/core/mind/hive/hive_schema.h
index 6289499cc3..2b664b6fdc 100644
--- a/ydb/core/mind/hive/hive_schema.h
+++ b/ydb/core/mind/hive/hive_schema.h
@@ -36,6 +36,7 @@ struct TSchemeIds {
         ResourceChangeReactionPeriod,
         TabletKickCooldownPeriod,
         ResourceOvercommitment,
+        TabletOwnersSynced,
     };
 };
 
diff --git a/ydb/core/mind/hive/tx__load_everything.cpp b/ydb/core/mind/hive/tx__load_everything.cpp
index e4c1cf1f51..739304494f 100644
--- a/ydb/core/mind/hive/tx__load_everything.cpp
+++ b/ydb/core/mind/hive/tx__load_everything.cpp
@@ -164,6 +164,9 @@ public:
                     case TSchemeIds::State::ResourceOvercommitment:
                         Self->DatabaseConfig.SetResourceOvercommitment((double)stateRowset.GetValue<Schema::State::Value>() / 100);
                         break;
+                    case TSchemeIds::State::TabletOwnersSynced:
+                        Self->TabletOwnersSynced = (bool)stateRowset.GetValue<Schema::State::Value>();
+                        break;
                     }
                 }
                 stateRowset.Next();
diff --git a/ydb/core/mind/hive/tx__request_tablet_owners.cpp b/ydb/core/mind/hive/tx__request_tablet_owners.cpp
new file mode 100644
index 0000000000..d07e45216a
--- /dev/null
+++ b/ydb/core/mind/hive/tx__request_tablet_owners.cpp
@@ -0,0 +1,46 @@
+#include "hive_impl.h"
+#include "hive_log.h"
+
+namespace NKikimr {
+namespace NHive {
+
+class TTxRequestTabletOwners : public TTransactionBase<THive> {
+    THolder<TEvHive::TEvRequestTabletOwners::THandle> Request;
+    THolder<TEvHive::TEvTabletOwnersReply> Response;
+
+public:
+    TTxRequestTabletOwners(THolder<TEvHive::TEvRequestTabletOwners::THandle> event, THive *hive)
+        : TBase(hive)
+        , Request(std::move(event))
+        , Response(new TEvHive::TEvTabletOwnersReply())
+    {}
+
+    TTxType GetTxType() const override { return NHive::TXTYPE_REQUEST_TABLET_OWNERS; }
+
+    bool Execute(TTransactionContext&, const TActorContext&) override {
+        BLOG_D("THive::TTxRequestTabletOwners::Execute");
+        auto ownerId = Request->Get()->Record.GetOwnerID();
+        std::vector<TSequencer::TSequence> sequences;
+        Self->Keeper.GetOwnedSequences(ownerId, sequences);
+        BLOG_D("THive::TTxRequestTabletOwners - replying with " << sequences.size() << " sequences");
+        for (const auto& seq : sequences) {
+            auto* tabletOwners = Response->Record.AddTabletOwners();
+            tabletOwners->SetOwnerID(ownerId);
+            tabletOwners->SetBegin(seq.Begin);
+            tabletOwners->SetEnd(seq.End);
+        }
+        return true;
+    }
+
+    void Complete(const TActorContext&) override {
+        BLOG_D("THive::TTxRequestTabletOwners::Complete");
+        Self->Send(Request->Sender, Response.Release());
+    }
+};
+
+ITransaction* THive::CreateRequestTabletOwners(TEvHive::TEvRequestTabletOwners::TPtr event) {
+    return new TTxRequestTabletOwners(THolder(std::move(event.Release())), this);
+}
+
+} // NHive
+} // NKikimr
diff --git a/ydb/core/mind/hive/tx__tablet_owners_reply.cpp b/ydb/core/mind/hive/tx__tablet_owners_reply.cpp
index b1e73dde71..9023e034a6 100644
--- a/ydb/core/mind/hive/tx__tablet_owners_reply.cpp
+++ b/ydb/core/mind/hive/tx__tablet_owners_reply.cpp
@@ -6,7 +6,6 @@ namespace NHive {
 
 class TTxTabletOwnersReply : public TTransactionBase<THive> {
     THolder<TEvHive::TEvTabletOwnersReply::THandle> Request;
-    TVector<TTabletId> TabletIds;
 
 public:
     TTxTabletOwnersReply(THolder<TEvHive::TEvTabletOwnersReply::THandle> event, THive *hive)
@@ -29,6 +28,8 @@ public:
                 db.Table<Schema::TabletOwners>().Key(seq.Begin, seq.End).Update<Schema::TabletOwners::OwnerId>(ownerId);
             }
         }
+        db.Table<Schema::State>().Key(TSchemeIds::State::TabletOwnersSynced).Update<Schema::State::Value>(true);
+        Self->TabletOwnersSynced = true;
         return true;
     }
 
diff --git a/ydb/core/protos/counters_hive.proto b/ydb/core/protos/counters_hive.proto
index 0d9a2c4d41..bc7322f46b 100644
--- a/ydb/core/protos/counters_hive.proto
+++ b/ydb/core/protos/counters_hive.proto
@@ -106,4 +106,5 @@ enum ETxTypes {
     TXTYPE_RESTART_TABLET = 56                            [(TxTypeOpts) = {Name: "TxRestartTablet"}];
     TXTYPE_MON_MOVE_TABLET = 57                           [(TxTypeOpts) = {Name: "TxMonMoveTablet"}];
     TXTYPE_TABLET_OWNERS_REPLY = 58                       [(TxTypeOpts) = {Name: "TxTabletOwnersReply"}];
+    TXTYPE_REQUEST_TABLET_OWNERS = 59                     [(TxTypeOpts) = {Name: "TxRequestTabletOwners"}];
 }