Add counters for number of unconnected disks to ds proxy KIKIMR-13568

ref:75801ddb5751aef9278677fd0cbd6f70fb61510c

6 files changed, 48 insertions, 0 deletions

diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_impl.h b/ydb/core/blobstorage/dsproxy/dsproxy_impl.h
index 508eb10a05..25eee78a6e 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_impl.h
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_impl.h
@@ -175,6 +175,8 @@ class TBlobStorageGroupProxy : public TActorBootstrapped<TBlobStorageGroupProxy>
    ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
     // Establishing Sessions state
 
+    ui32 NumUnconnectedDisks = Max<ui32>();
+
     void SwitchToWorkWhenGoodToGo();
     void WakeupEstablishingSessions(TEvEstablishingSessionTimeout::TPtr ev);
     void Handle(TEvProxyQueueState::TPtr& ev);
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_nodemon.cpp b/ydb/core/blobstorage/dsproxy/dsproxy_nodemon.cpp
index addca8c533..b12c5fcf89 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_nodemon.cpp
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_nodemon.cpp
@@ -81,6 +81,10 @@ TDsProxyNodeMon::TDsProxyNodeMon(TIntrusivePtr<NMonitoring::TDynamicCounters> &c
         EstablishingSessionsTimeout5min = group->GetCounter("EstablishingSessionsTimeout5min", false);
         UnconfiguredTimeout = group->GetCounter("UnconfiguredTimeout", false);
         UnconfiguredTimeout5min = group->GetCounter("UnconfiguredTimeout5min", false);
+        ConnectedAll = group->GetCounter("ConnectedAll", false);
+        ConnectedMinus1 = group->GetCounter("ConnectedMinus1", false);
+        ConnectedMinus2 = group->GetCounter("ConnectedMinus2", false);
+        ConnectedMinus3more = group->GetCounter("ConnectedMinus3more", false);
     }
 }
 
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_nodemon.h b/ydb/core/blobstorage/dsproxy/dsproxy_nodemon.h
index 150da172fb..956e470e37 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_nodemon.h
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_nodemon.h
@@ -89,6 +89,10 @@ struct TDsProxyNodeMon : public TThrRefBase {
     NMonitoring::TDynamicCounters::TCounterPtr EstablishingSessionsTimeout5min;
     NMonitoring::TDynamicCounters::TCounterPtr UnconfiguredTimeout;
     NMonitoring::TDynamicCounters::TCounterPtr UnconfiguredTimeout5min;
+    NMonitoring::TDynamicCounters::TCounterPtr ConnectedAll;
+    NMonitoring::TDynamicCounters::TCounterPtr ConnectedMinus1;
+    NMonitoring::TDynamicCounters::TCounterPtr ConnectedMinus2;
+    NMonitoring::TDynamicCounters::TCounterPtr ConnectedMinus3more;
 
     TDsProxyNodeMon(TIntrusivePtr<NMonitoring::TDynamicCounters> &counters, bool initForAllDeviceTypes);
     void CountPutPesponseTime(TPDiskCategory::EDeviceType type, NKikimrBlobStorage::EPutHandleClass cls, ui32 size,
@@ -99,6 +103,24 @@ struct TDsProxyNodeMon : public TThrRefBase {
 
     // Called only from NodeWarder
     void CheckNodeMonCountersForDeviceType(TPDiskCategory::EDeviceType type);
+
+    void IncNumUnconnected(ui32 num) {
+        switch (num) {
+            case 0: ++*ConnectedAll; break;
+            case 1: ++*ConnectedMinus1; break;
+            case 2: ++*ConnectedMinus2; break;
+            default: ++*ConnectedMinus3more; break;
+        }
+    }
+
+    void DecNumUnconnected(ui32 num) {
+        switch (num) {
+            case 0: --*ConnectedAll; break;
+            case 1: --*ConnectedMinus1; break;
+            case 2: --*ConnectedMinus2; break;
+            default: --*ConnectedMinus3more; break;
+        }
+    }
 };
 
 } // NKikimr
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_state.cpp b/ydb/core/blobstorage/dsproxy/dsproxy_state.cpp
index af2739699d..9cf4b55e58 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_state.cpp
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_state.cpp
@@ -151,6 +151,8 @@ namespace NKikimr {
             } else { // this is the first time configuration arrives -- no queues are created yet
                 EnsureMonitoring(false);
                 Sessions = MakeIntrusive<TGroupSessions>(Info, BSProxyCtx, MonActor, SelfId());
+                NumUnconnectedDisks = Sessions->GetNumUnconnectedDisks();
+                NodeMon->IncNumUnconnected(NumUnconnectedDisks);
             }
             SetStateEstablishingSessions();
         } else { // configuration has been reset, we have to request it via NodeWarden
@@ -208,6 +210,11 @@ namespace NKikimr {
                 CurrentStateFunc() == &TThis::StateEstablishingSessionsTimeout)) {
             SwitchToWorkWhenGoodToGo();
         }
+
+        if (const ui32 prev = std::exchange(NumUnconnectedDisks, Sessions->GetNumUnconnectedDisks()); prev != NumUnconnectedDisks) {
+            NodeMon->IncNumUnconnected(NumUnconnectedDisks);
+            NodeMon->DecNumUnconnected(prev);
+        }
     }
 
     void TBlobStorageGroupProxy::Bootstrap() {
@@ -256,6 +263,10 @@ namespace NKikimr {
         ErrorDescription = "DsProxy got a Poison Pill";
         SetStateEstablishingSessionsTimeout();
 
+        if (NumUnconnectedDisks != Max<ui32>()) {
+            NodeMon->DecNumUnconnected(NumUnconnectedDisks);
+        }
+
         TActorBootstrapped::PassAway();
         // TODO: Unsubscribe
     }
diff --git a/ydb/core/blobstorage/dsproxy/group_sessions.cpp b/ydb/core/blobstorage/dsproxy/group_sessions.cpp
index 016952e1e2..9923c6b784 100644
--- a/ydb/core/blobstorage/dsproxy/group_sessions.cpp
+++ b/ydb/core/blobstorage/dsproxy/group_sessions.cpp
@@ -122,4 +122,12 @@ void TGroupSessions::QueueConnectUpdate(ui32 orderNumber, NKikimrBlobStorage::EV
     }
 }
 
+ui32 TGroupSessions::GetNumUnconnectedDisks() {
+    ui32 n = 0;
+    for (const ui8 mask : ConnectedQueuesMask) {
+        n += mask != AllQueuesMask;
+    }
+    return n;
+}
+
 } // NKikimr
diff --git a/ydb/core/blobstorage/dsproxy/group_sessions.h b/ydb/core/blobstorage/dsproxy/group_sessions.h
index b4b9db6284..9284051408 100644
--- a/ydb/core/blobstorage/dsproxy/group_sessions.h
+++ b/ydb/core/blobstorage/dsproxy/group_sessions.h
@@ -221,6 +221,7 @@ namespace NKikimr {
         void Poison();
         bool GoodToGo(const TBlobStorageGroupInfo::TTopology& topology, bool waitForAllVDisks);
         void QueueConnectUpdate(ui32 orderNumber, NKikimrBlobStorage::EVDiskQueueId queueId, bool connected);
+        ui32 GetNumUnconnectedDisks();
     };
 
     struct TEvRequestProxySessionsState : TEventLocal<TEvRequestProxySessionsState, TEvBlobStorage::EvRequestProxySessionsState>