Fix NodeWarden pipe connect race KIKIMR-13568

ref:b7652cfd6c7d3752ae1533df9ec2140bd73dc9e9

4 files changed, 11 insertions, 9 deletions

diff --git a/ydb/core/mind/bscontroller/config.cpp b/ydb/core/mind/bscontroller/config.cpp
index aa7d6efceea..47b522f24dc 100644
--- a/ydb/core/mind/bscontroller/config.cpp
+++ b/ydb/core/mind/bscontroller/config.cpp
@@ -23,7 +23,7 @@ namespace NKikimr::NBsController {
                 for (auto &pair : Services) {
                     const TNodeId &nodeId = pair.first;
 
-                    if (TNodeInfo *node = Self->FindNode(nodeId); node && node->IsRegistered) {
+                    if (TNodeInfo *node = Self->FindNode(nodeId); node && node->ConnectedCount) {
                         auto event = MakeHolder<TEvBlobStorage::TEvControllerNodeServiceSetUpdate>();
                         auto& record = event->Record;
                         pair.second.Swap(&record);
diff --git a/ydb/core/mind/bscontroller/group_reconfigure_wipe.cpp b/ydb/core/mind/bscontroller/group_reconfigure_wipe.cpp
index b3bf230de82..7c274224baf 100644
--- a/ydb/core/mind/bscontroller/group_reconfigure_wipe.cpp
+++ b/ydb/core/mind/bscontroller/group_reconfigure_wipe.cpp
@@ -46,7 +46,7 @@ public:
             db.Table<Schema::VSlot>().Key(id.GetKey()).Update<Schema::VSlot::Mood>(info->Mood);
 
             // Prepare results for nodes
-            if (TNodeInfo *node = Self->FindNode(id.NodeId); node && node->IsRegistered) {
+            if (TNodeInfo *node = Self->FindNode(id.NodeId); node && node->ConnectedCount) {
                 auto& msg = ResultForNode[id.NodeId];
                 msg = MakeHolder<TEvBlobStorage::TEvControllerNodeServiceSetUpdate>(NKikimrProto::OK, id.NodeId);
                 Self->ReadVSlot(*info, msg.Get());
diff --git a/ydb/core/mind/bscontroller/impl.h b/ydb/core/mind/bscontroller/impl.h
index ba311a254c3..2cf3989ca42 100644
--- a/ydb/core/mind/bscontroller/impl.h
+++ b/ydb/core/mind/bscontroller/impl.h
@@ -730,7 +730,7 @@ public:
     public:
         using Table = Schema::Node;
 
-        bool IsRegistered = false;
+        ui32 ConnectedCount = 0;
         Table::NextPDiskID::Type NextPDiskID;
         // in-mem only
         std::map<TString, NPDisk::TDriveData> KnownDrives;
diff --git a/ydb/core/mind/bscontroller/register_node.cpp b/ydb/core/mind/bscontroller/register_node.cpp
index 9e0df640a8e..2a88fec38e2 100644
--- a/ydb/core/mind/bscontroller/register_node.cpp
+++ b/ydb/core/mind/bscontroller/register_node.cpp
@@ -233,8 +233,6 @@ public:
             }
         }
 
-        TNodeInfo& nodeInfo = Self->GetNode(nodeId);
-
         auto res = std::make_unique<TEvBlobStorage::TEvControllerNodeServiceSetUpdate>(NKikimrProto::OK, nodeId);
 
         TSet<ui32> groupIDsToRead;
@@ -285,8 +283,6 @@ public:
 
         Self->ReadGroups(groupsToDiscard, true, res.get());
 
-        nodeInfo.IsRegistered = true;
-
         for (auto it = Self->PDisks.lower_bound(minPDiskId); it != Self->PDisks.end() && it->first.NodeId == nodeId; ++it) {
             Self->ReadPDisk(it->first, *it->second, res.get(), NKikimrBlobStorage::INITIAL);
         }
@@ -426,6 +422,9 @@ void TBlobStorageController::OnRegisterNode(const TActorId& serverId, TNodeId no
 }
 
 void TBlobStorageController::OnWardenConnected(TNodeId nodeId) {
+    TNodeInfo& node = GetNode(nodeId);
+    ++node.ConnectedCount;
+
     for (auto it = PDisks.lower_bound(TPDiskId::MinForNode(nodeId)); it != PDisks.end() && it->first.NodeId == nodeId; ++it) {
         it->second->UpdateOperational(true);
         SysViewChangedPDisks.insert(it->first);
@@ -433,6 +432,11 @@ void TBlobStorageController::OnWardenConnected(TNodeId nodeId) {
 }
 
 void TBlobStorageController::OnWardenDisconnected(TNodeId nodeId) {
+    TNodeInfo& node = GetNode(nodeId);
+    if (--node.ConnectedCount) {
+        return; // there are still some connections from this NW
+    }
+
     const TInstant now = TActivationContext::Now();
     std::vector<std::pair<TVSlotId, TInstant>> lastSeenReadyQ;
     for (auto it = PDisks.lower_bound(TPDiskId::MinForNode(nodeId)); it != PDisks.end() && it->first.NodeId == nodeId; ++it) {
@@ -461,8 +465,6 @@ void TBlobStorageController::OnWardenDisconnected(TNodeId nodeId) {
         Send(SelfHealId, sh.Release());
     }
     ScrubState.OnNodeDisconnected(nodeId);
-    TNodeInfo& node = GetNode(nodeId);
-    node.IsRegistered = false;
     EraseKnownDrivesOnDisconnected(&node);
     if (!lastSeenReadyQ.empty()) {
         Execute(CreateTxUpdateLastSeenReady(std::move(lastSeenReadyQ)));