Some BlobDepot minor fixes

8 files changed, 130 insertions, 62 deletions

diff --git a/ydb/core/blob_depot/agent/storage_range.cpp b/ydb/core/blob_depot/agent/storage_range.cpp
index 0a4f4e89f4..1c113adf25 100644
--- a/ydb/core/blob_depot/agent/storage_range.cpp
+++ b/ydb/core/blob_depot/agent/storage_range.cpp
@@ -49,12 +49,13 @@ namespace NKikimr::NBlobDepot {
                 ++ResolvesInFlight;
             }
 
-            void IssueResolve(TLogoBlobID id) {
+            void IssueResolve(ui64 tag) {
                 NKikimrBlobDepot::TEvResolve resolve;
                 auto *item = resolve.AddItems();
-                item->SetExactKey(id.AsBinaryString());
+                item->SetExactKey(Reads[tag].AsBinaryString());
                 item->SetTabletId(Request.TabletId);
                 item->SetMustRestoreFirst(Request.MustRestoreFirst);
+                item->SetCookie(tag);
 
                 Agent.Issue(std::move(resolve), this, nullptr);
                 ++ResolvesInFlight;
@@ -79,14 +80,29 @@ namespace NKikimr::NBlobDepot {
                     return EndWithError(msg.GetStatus(), msg.GetErrorReason());
                 }
 
+#ifndef NDEBUG
+                THashSet<TLogoBlobID> ids;
+#endif
+
                 for (const auto& key : msg.GetResolvedKeys()) {
                     const TString& blobId = key.GetKey();
                     auto id = TLogoBlobID::FromBinary(blobId);
 
+#ifndef NDEBUG
+                    const bool inserted = ids.insert(id).second;
+                    Y_VERIFY(inserted);
+#endif
+
                     if (key.HasErrorReason()) {
                         return EndWithError(NKikimrProto::ERROR, TStringBuilder() << "failed to resolve blob# " << id
                             << ": " << key.GetErrorReason());
                     } else if (!Request.IsIndexOnly) {
+                        const ui64 tag = key.HasCookie() ? key.GetCookie() : Reads.size();
+                        if (tag == Reads.size()) {
+                            Reads.push_back(id);
+                        } else {
+                            Y_VERIFY(Reads[tag] == id);
+                        }
                         TReadArg arg{
                             key.GetValueChain(),
                             NKikimrBlobStorage::EGetHandleClass::FastRead,
@@ -94,17 +110,14 @@ namespace NKikimr::NBlobDepot {
                             this,
                             0,
                             0,
-                            Reads.size(),
+                            tag,
                             {}};
-                        Reads.push_back(id);
                         ++ReadsInFlight;
                         TString error;
                         if (!Agent.IssueRead(arg, error)) {
                             return EndWithError(NKikimrProto::ERROR, TStringBuilder() << "failed to read discovered blob: "
                                 << error);
                         }
-                    } else if (Request.MustRestoreFirst) {
-                        Y_FAIL("not implemented yet");
                     } else {
                         FoundBlobs.try_emplace(id);
                     }
@@ -128,7 +141,7 @@ namespace NKikimr::NBlobDepot {
                     }
 
                     case NKikimrProto::NODATA:
-                        IssueResolve(Reads[tag]);
+                        IssueResolve(tag);
                         break;
 
                     default:
diff --git a/ydb/core/blob_depot/data_resolve.cpp b/ydb/core/blob_depot/data_resolve.cpp
index c81d1756cb..449fb7cc1d 100644
--- a/ydb/core/blob_depot/data_resolve.cpp
+++ b/ydb/core/blob_depot/data_resolve.cpp
@@ -166,6 +166,8 @@ namespace NKikimr::NBlobDepot {
             bool progress = false;
             if (!KeysLoaded && !LoadKeys(txc, progress)) {
                 return progress;
+            } else if (SuccessorTx) {
+                return true;
             } else {
                 KeysLoaded = true;
             }
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy.h b/ydb/core/blobstorage/dsproxy/dsproxy.h
index c4bb0d7e39..ab8820c93b 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy.h
+++ b/ydb/core/blobstorage/dsproxy/dsproxy.h
@@ -516,7 +516,7 @@ private:
 protected:
     using TThis = TDerived;
 
-    TIntrusivePtr<TBlobStorageGroupInfo> Info;
+    const TIntrusivePtr<TBlobStorageGroupInfo> Info;
     TIntrusivePtr<TGroupQueues> GroupQueues;
     TIntrusivePtr<TBlobStorageGroupProxyMon> Mon;
     TIntrusivePtr<TStoragePoolCounters> PoolCounters;
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_assimilate.cpp b/ydb/core/blobstorage/dsproxy/dsproxy_assimilate.cpp
index 5b8394ccfb..15138025e5 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_assimilate.cpp
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_assimilate.cpp
@@ -314,7 +314,7 @@ public:
 
         const auto& record = ev->Get()->Record;
         const TVDiskID vdiskId = VDiskIDFromVDiskID(record.GetVDiskID());
-        const ui32 orderNumber = Info->GetOrderNumber(vdiskId);
+        const ui32 orderNumber = Info->GetTopology().GetOrderNumber(vdiskId);
         Y_VERIFY(orderNumber < PerVDiskInfo.size());
 
         if (record.GetStatus() == NKikimrProto::OK) {
diff --git a/ydb/core/blobstorage/groupinfo/blobstorage_groupinfo.cpp b/ydb/core/blobstorage/groupinfo/blobstorage_groupinfo.cpp
index e663567105..6a8fa70cc4 100644
--- a/ydb/core/blobstorage/groupinfo/blobstorage_groupinfo.cpp
+++ b/ydb/core/blobstorage/groupinfo/blobstorage_groupinfo.cpp
@@ -816,7 +816,7 @@ TVDiskID TBlobStorageGroupInfo::GetVDiskInSubgroup(ui32 idxInSubgroup, ui32 hash
 
 ui32 TBlobStorageGroupInfo::GetOrderNumber(const TVDiskID &vdisk) const {
     Y_VERIFY_S(vdisk.GroupID == GroupID, "Expected GroupID# " << GroupID << ", given GroupID# " << vdisk.GroupID);
-    Y_VERIFY_S(vdisk.GroupGeneration == GroupGeneration, "Expected GroupGeeration# " << GroupGeneration
+    Y_VERIFY_S(vdisk.GroupGeneration == GroupGeneration, "Expected GroupGeneration# " << GroupGeneration
             << ", given GroupGeneration# " << vdisk.GroupGeneration);
     return Topology->GetOrderNumber(vdisk);
 }
diff --git a/ydb/core/mind/bscontroller/impl.h b/ydb/core/mind/bscontroller/impl.h
index ce9bf37bb1..65ca1b30a8 100644
--- a/ydb/core/mind/bscontroller/impl.h
+++ b/ydb/core/mind/bscontroller/impl.h
@@ -508,7 +508,6 @@ public:
         TMaybe<Table::ErrorReason::Type> ErrorReason;
         TMaybe<Table::NeedAlter::Type> NeedAlter;
         std::optional<NKikimrBlobStorage::TGroupMetrics> GroupMetrics;
-        bool CommitInProgress = false;
 
         bool Down = false; // is group are down right now (not selectable)
         TVector<TIndirectReferable<TVSlotInfo>::TPtr> VDisksInGroup;
@@ -1646,6 +1645,7 @@ private:
     void RenderFooter(IOutputStream& out);
     void RenderMonPage(IOutputStream& out);
     void RenderInternalTables(IOutputStream& out, const TString& table);
+    void RenderVirtualGroups(IOutputStream& out);
     void RenderGroupDetail(IOutputStream &out, TGroupId groupId);
     void RenderGroupsInStoragePool(IOutputStream &out, const TBoxStoragePoolId& id);
     void RenderVSlotTable(IOutputStream& out, std::function<void()> callback);
diff --git a/ydb/core/mind/bscontroller/monitoring.cpp b/ydb/core/mind/bscontroller/monitoring.cpp
index 60645b190a..5f1bf21ebc 100644
--- a/ydb/core/mind/bscontroller/monitoring.cpp
+++ b/ydb/core/mind/bscontroller/monitoring.cpp
@@ -909,6 +909,8 @@ bool TBlobStorageController::OnRenderAppHtmlPage(NMon::TEvRemoteHttpInfo::TPtr e
         } else if (page == "InternalTables") {
             const TString table = cgi.Has("table") ? cgi.Get("table") : "pdisks";
             RenderInternalTables(str, table);
+        } else if (page == "VirtualGroups") {
+            RenderVirtualGroups(str);
         } else if (page == "StopGivingGroups") {
             StopGivingGroups = true;
             str << "OK";
@@ -960,6 +962,7 @@ void TBlobStorageController::RenderMonPage(IOutputStream& out) {
         (SelfHealEnable ? "enabled" : "disabled") << ")<br>";
     out << "<a href='app?TabletID=" << TabletID() << "&page=HealthEvents'>Health events</a><br>";
     out << "<a href='app?TabletID=" << TabletID() << "&page=Scrub'>Scrub state</a><br>";
+    out << "<a href='app?TabletID=" << TabletID() << "&page=VirtualGroups'>Virtual groups</a><br>";
     out << "<a href='app?TabletID=" << TabletID() << "&page=InternalTables'>Internal tables</a><br>";
 
     HTML(out) {
diff --git a/ydb/core/mind/bscontroller/virtual_group.cpp b/ydb/core/mind/bscontroller/virtual_group.cpp
index 66359ee855..a8abe9645f 100644
--- a/ydb/core/mind/bscontroller/virtual_group.cpp
+++ b/ydb/core/mind/bscontroller/virtual_group.cpp
@@ -185,64 +185,37 @@ namespace NKikimr::NBsController {
         class TTxUpdateGroup : public TTransactionBase<TBlobStorageController> {
             TVirtualGroupSetupMachine *Machine;
             std::optional<TConfigState> State;
+            std::function<void(TGroupInfo&)> Callback;
 
         public:
-            TTxUpdateGroup(TVirtualGroupSetupMachine *machine)
+            TTxUpdateGroup(TVirtualGroupSetupMachine *machine, std::function<void(TGroupInfo&)>&& callback)
                 : TTransactionBase(machine->Self)
                 , Machine(machine)
+                , Callback(std::move(callback))
             {}
 
             bool Execute(TTransactionContext& txc, const TActorContext&) override {
-                TGroupInfo *group = GetGroup(true);
-                NIceDb::TNiceDb db(txc.DB);
-                using T = Schema::Group;
-                auto row = db.Table<T>().Key(group->ID);
-#define PARAM(NAME) \
-                if (const auto& cell = group->NAME) { \
-                    row.Update<T::NAME>(*cell); \
-                }
-                PARAM(VirtualGroupState)
-                PARAM(BlobDepotConfig)
-                PARAM(BlobDepotId)
-                PARAM(ErrorReason)
-                PARAM(NeedAlter)
-#undef PARAM
-                if (group->SeenOperational) {
-                    row.Update<T::SeenOperational>(true);
-                }
-                if (group->VirtualGroupState == NKikimrBlobStorage::EVirtualGroupState::WORKING &&
-                        group->DecommitStatus == NKikimrBlobStorage::TGroupDecommitStatus::PENDING) {
-                    State.emplace(*Self, Self->HostRecords, TActivationContext::Now());
-                    TGroupInfo *group = State->Groups.FindForUpdate(Machine->GroupId);
-                    Y_VERIFY(group);
-                    group->DecommitStatus = NKikimrBlobStorage::TGroupDecommitStatus::IN_PROGRESS;
-                    group->ContentChanged = true;
-                    TString error;
-                    if (State->Changed() && !Self->CommitConfigUpdates(*State, false, false, false, txc, &error)) {
-                        STLOG(PRI_ERROR, BS_CONTROLLER, BSCVG08, "failed to commit update",
-                            (VirtualGroupId, Machine->GroupId), (Error, error));
-                        State->Rollback();
-                        State.reset();
-                    }
+                State.emplace(*Self, Self->HostRecords, TActivationContext::Now());
+                TGroupInfo *group = State->Groups.FindForUpdate(Machine->GroupId);
+                Y_VERIFY(group);
+                Callback(*group);
+                TString error;
+                if (State->Changed() && !Self->CommitConfigUpdates(*State, false, false, false, txc, &error)) {
+                    STLOG(PRI_ERROR, BS_CONTROLLER, BSCVG08, "failed to commit update",
+                        (VirtualGroupId, Machine->GroupId), (Error, error));
+                    State->Rollback();
+                    State.reset();
                 }
                 return true;
             }
 
             void Complete(const TActorContext&) override {
-                TGroupInfo *group = GetGroup(false);
+                TGroupInfo *group = Machine->GetGroup();
                 Y_VERIFY(group->VirtualGroupSetupMachineId == Machine->SelfId());
-                TActivationContext::Send(new IEventHandle(TEvents::TSystem::Bootstrap, 0, Machine->SelfId(), {}, nullptr, 0));
                 if (State) {
                     State->ApplyConfigUpdates();
                 }
-            }
-
-        private:
-            TGroupInfo *GetGroup(bool commitInProgress) {
-                TGroupInfo *group = Machine->GetGroup();
-                Y_VERIFY(group->CommitInProgress != commitInProgress);
-                group->CommitInProgress = commitInProgress;
-                return group;
+                TActivationContext::Send(new IEventHandle(TEvents::TSystem::Bootstrap, 0, Machine->SelfId(), {}, nullptr, 0));
             }
         };
 
@@ -382,15 +355,17 @@ namespace NKikimr::NBsController {
             STLOG(PRI_INFO, BS_CONTROLLER, BSCVG04, "received TEvCreateTabletReply", (TabletId, Self->TabletID()),
                 (Msg, ev->Get()->Record));
 
-            TGroupInfo *group = GetGroup();
             auto& record = ev->Get()->Record;
             if (record.GetStatus() == NKikimrProto::OK || record.GetStatus() == NKikimrProto::ALREADY) {
                 BlobDepotTabletId = record.GetTabletID();
                 Y_VERIFY(BlobDepotTabletId);
             } else {
-                group->VirtualGroupState = NKikimrBlobStorage::EVirtualGroupState::CREATE_FAILED;
-                group->ErrorReason = SingleLineProto(record);
-                Self->Execute(new TTxUpdateGroup(this));
+                Self->Execute(new TTxUpdateGroup(this, [&](TGroupInfo& group) {
+                    group.VirtualGroupState = NKikimrBlobStorage::EVirtualGroupState::CREATE_FAILED;
+                    group.ErrorReason = TStringBuilder() << "failed to create BlobDepot tablet"
+                        << " Reason# " << NKikimrHive::EErrorReason_Name(record.GetErrorReason())
+                        << " Status# " << NKikimrProto::EReplyStatus_Name(record.GetStatus());
+                }));
             }
 
             NTabletPipe::CloseAndForgetClient(SelfId(), HivePipeId);
@@ -416,12 +391,17 @@ namespace NKikimr::NBsController {
         }
 
         void Handle(TEvBlobDepot::TEvApplyConfigResult::TPtr /*ev*/) {
-            TGroupInfo *group = GetGroup();
-            group->VirtualGroupState = NKikimrBlobStorage::EVirtualGroupState::WORKING;
-            group->BlobDepotId = BlobDepotTabletId;
-            group->NeedAlter = false; // race-check
-            Self->Execute(new TTxUpdateGroup(this));
             NTabletPipe::CloseAndForgetClient(SelfId(), BlobDepotPipeId);
+
+            Self->Execute(new TTxUpdateGroup(this, [&](TGroupInfo& group) {
+                group.VirtualGroupState = NKikimrBlobStorage::EVirtualGroupState::WORKING;
+                group.BlobDepotId = BlobDepotTabletId;
+                group.NeedAlter = false;
+                if (group.DecommitStatus == NKikimrBlobStorage::TGroupDecommitStatus::PENDING) {
+                    group.DecommitStatus = NKikimrBlobStorage::TGroupDecommitStatus::IN_PROGRESS;
+                    group.ContentChanged = true;
+                }
+            }));
         }
 
         ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -566,4 +546,74 @@ namespace NKikimr::NBsController {
         Execute(new TTxDecommitGroup(this, ev));
     }
 
+    void TBlobStorageController::RenderVirtualGroups(IOutputStream& out) {
+        HTML(out) {
+            DIV_CLASS("panel panel-info") {
+                DIV_CLASS("panel-heading") {
+                    out << "Virtual groups";
+                }
+                DIV_CLASS("panel-body") {
+                    TABLE_CLASS("table table-condensed") {
+                        TABLEHEAD() {
+                            TABLER() {
+                                TABLEH() { out << "GroupId"; }
+                                TABLEH() { out << "StoragePoolName"; }
+                                TABLEH() { out << "Name"; }
+                                TABLEH() { out << "State"; }
+                                TABLEH() { out << "HiveId"; }
+                                TABLEH() { out << "ErrorReason"; }
+                                TABLEH() { out << "DecommitStatus"; }
+                            }
+                        }
+                        TABLEBODY() {
+                            for (const auto& [groupId, group] : GroupMap) {
+                                if (!group->VirtualGroupState) {
+                                    continue;
+                                }
+
+                                TABLER() {
+                                    TABLED() {
+                                        out << groupId;
+                                    }
+                                    TABLED() {
+                                        const auto it = StoragePools.find(group->StoragePoolId);
+                                        Y_VERIFY(it != StoragePools.end());
+                                        out << it->second.Name;
+                                    }
+                                    TABLED() {
+                                        if (group->VirtualGroupName) {
+                                            out << *group->VirtualGroupName;
+                                        } else {
+                                            out << "<i>null</i>";
+                                        }
+                                    }
+                                    TABLED() {
+                                        out << NKikimrBlobStorage::EVirtualGroupState_Name(*group->VirtualGroupState);
+                                    }
+                                    TABLED() {
+                                        if (group->HiveId) {
+                                            out << *group->HiveId;
+                                        } else {
+                                            out << "<i>null</i>";
+                                        }
+                                    }
+                                    TABLED() {
+                                        if (group->ErrorReason) {
+                                            out << *group->ErrorReason;
+                                        } else {
+                                            out << "<i>null</i>";
+                                        }
+                                    }
+                                    TABLED() {
+                                        out << NKikimrBlobStorage::TGroupDecommitStatus::E_Name(group->DecommitStatus);
+                                    }
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+        }
+    }
+
 } // NKikimr::NBsController