storage health check fixes (#8291)

Author: vporyadke

ydb/core/health_check/health_check.cpp

@@ -251,6 +251,7 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
     struct TGroupState {
         TString ErasureSpecies;
         std::vector<const NKikimrSysView::TVSlotEntry*> VSlots;
+        ui32 Generation;
     };
 
     struct TSelfCheckResult {
@@ -1276,12 +1277,17 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
         for (const auto& group : Groups->GetEntries()) {
             auto groupId = group.GetKey().GetGroupId();
             auto poolId = group.GetInfo().GetStoragePoolId();
-            GroupState[groupId].ErasureSpecies = group.GetInfo().GetErasureSpeciesV2();
+            auto& groupState = GroupState[groupId];
+            groupState.ErasureSpecies = group.GetInfo().GetErasureSpeciesV2();
+            groupState.Generation = group.GetInfo().GetGeneration();
             StoragePoolState[poolId].Groups.emplace(groupId);
         }
         for (const auto& vSlot : VSlots->GetEntries()) {
             auto vSlotId = GetVSlotId(vSlot.GetKey());
-            GroupState[vSlot.GetInfo().GetGroupId()].VSlots.push_back(&vSlot);
+            auto groupStateIt = GroupState.find(vSlot.GetInfo().GetGroupId());
+            if (groupStateIt != GroupState.end() && vSlot.GetInfo().GetGroupGeneration() == groupStateIt->second.Generation) {
+                groupStateIt->second.VSlots.push_back(&vSlot);
+            }
         }
         for (const auto& pool : StoragePools->GetEntries()) { // there is no specific pool for static group here
             ui64 poolId = pool.GetKey().GetStoragePoolId();
@@ -1782,6 +1788,13 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
 
         storageVDiskStatus.set_id(GetVSlotId(vSlot->GetKey()));
 
+        if (!vSlot->GetInfo().HasStatusV2()) {
+            // this should mean that BSC recently restarted and does not have accurate data yet - we should not report to avoid false positives
+            context.ReportStatus(Ydb::Monitoring::StatusFlag::GREEN);
+            storageVDiskStatus.set_overall(context.GetOverallStatus());
+            return;
+        }
+
         const auto *descriptor = NKikimrBlobStorage::EVDiskStatus_descriptor();
         auto status = descriptor->FindValueByName(vSlot->GetInfo().GetStatusV2());
         if (!status) { // this case is not expected because becouse bsc assignes status according EVDiskStatus enum
@@ -1801,16 +1814,12 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
                 storageVDiskStatus.set_overall(context.GetOverallStatus());
                 return;
             }
-            case NKikimrBlobStorage::INIT_PENDING: { // initialization in process
-                context.ReportStatus(Ydb::Monitoring::StatusFlag::YELLOW, TStringBuilder() << "VDisk is being initialized", ETags::VDiskState);
-                storageVDiskStatus.set_overall(context.GetOverallStatus());
-                return;
-            }
             case NKikimrBlobStorage::REPLICATING: { // the disk accepts queries, but not all the data was replicated
                 context.ReportStatus(Ydb::Monitoring::StatusFlag::BLUE, TStringBuilder() << "Replication in progress", ETags::VDiskState);
                 storageVDiskStatus.set_overall(context.GetOverallStatus());
                 return;
             }
+            case NKikimrBlobStorage::INIT_PENDING:
             case NKikimrBlobStorage::READY: { // the disk is fully operational and does not affect group fault tolerance
                 context.ReportStatus(Ydb::Monitoring::StatusFlag::GREEN);
             }