do not let faulty pdisks make group status dead in healthcheck (ydb-p…

…latform#9744)
vporyadke · Oct 21, 2024 · d70273a · d70273a
1 parent 483ce39
commit d70273a
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 5 deletions.
diff --git a/ydb/core/health_check/health_check.cpp b/ydb/core/health_check/health_check.cpp
@@ -1996,6 +1996,8 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
             }
         }
 
+        // do not propagate RED status to vdisk - so that vdisk is not considered down when computing group status
+        context.OverallStatus = MinStatus(context.OverallStatus, Ydb::Monitoring::StatusFlag::ORANGE);
         storagePDiskStatus.set_overall(context.GetOverallStatus());
     }
 

diff --git a/ydb/core/health_check/health_check_ut.cpp b/ydb/core/health_check/health_check_ut.cpp
@@ -69,7 +69,8 @@ Y_UNIT_TEST_SUITE(THealthCheckTest) {
 
     struct TTestVSlotInfo {
         std::optional<NKikimrBlobStorage::EVDiskStatus> Status;
-        ui32 Generation;
+        ui32 Generation = DEFAULT_GROUP_GENERATION;
+        NKikimrBlobStorage::EDriveStatus PDiskStatus = NKikimrBlobStorage::ACTIVE;
 
         TTestVSlotInfo(std::optional<NKikimrBlobStorage::EVDiskStatus> status = NKikimrBlobStorage::READY,
                        ui32 generation = DEFAULT_GROUP_GENERATION)
@@ -78,7 +79,11 @@ Y_UNIT_TEST_SUITE(THealthCheckTest) {
         {
         }
 
-        TTestVSlotInfo(NKikimrBlobStorage::EVDiskStatus status) : Status(status), Generation(DEFAULT_GROUP_GENERATION) {}
+        TTestVSlotInfo(NKikimrBlobStorage::EVDiskStatus status, NKikimrBlobStorage::EDriveStatus pDiskStatus = NKikimrBlobStorage::ACTIVE)
+            : Status(status)
+            , PDiskStatus(pDiskStatus)
+        {
+        }
     };
 
     using TVDisks = TVector<TTestVSlotInfo>;
@@ -223,18 +228,20 @@ Y_UNIT_TEST_SUITE(THealthCheckTest) {
         entry->mutable_info()->set_name(STORAGE_POOL_NAME);
     }
 
-    void AddPDisksToSysViewResponse(NSysView::TEvSysView::TEvGetPDisksResponse::TPtr* ev, size_t count, double occupancy) {
+    void AddPDisksToSysViewResponse(NSysView::TEvSysView::TEvGetPDisksResponse::TPtr* ev, const TVDisks& vslots, double occupancy) {
         auto& record = (*ev)->Get()->Record;
         auto entrySample = record.entries(0);
         record.clear_entries();
         auto pdiskId = PDISK_START_ID;
         const size_t totalSize = 3'200'000'000'000ull;
-        for (size_t i = 0; i < count; ++i) {
+        const auto *descriptor = NKikimrBlobStorage::EDriveStatus_descriptor();
+        for (const auto& vslot : vslots) {
             auto* entry = record.add_entries();
             entry->CopyFrom(entrySample);
             entry->mutable_key()->set_pdiskid(pdiskId);
             entry->mutable_info()->set_totalsize(totalSize);
             entry->mutable_info()->set_availablesize((1 - occupancy) * totalSize);
+            entry->mutable_info()->set_statusv2(descriptor->FindValueByNumber(vslot.PDiskStatus)->name());
             ++pdiskId;
         }
     }
@@ -483,7 +490,7 @@ Y_UNIT_TEST_SUITE(THealthCheckTest) {
                 }
                 case NSysView::TEvSysView::EvGetPDisksResponse: {
                     auto* x = reinterpret_cast<NSysView::TEvSysView::TEvGetPDisksResponse::TPtr*>(&ev);
-                    AddPDisksToSysViewResponse(x, vdisks.size(), occupancy);
+                    AddPDisksToSysViewResponse(x, vdisks, occupancy);
                     break;
                 }
                 case NSysView::TEvSysView::EvGetGroupsResponse: {
@@ -711,6 +718,14 @@ Y_UNIT_TEST_SUITE(THealthCheckTest) {
         CheckHcResultHasIssuesWithStatus(result, "STORAGE_GROUP", Ydb::Monitoring::StatusFlag::RED, 1);
     }
 
+    Y_UNIT_TEST(YellowIssueReadyVDisksOnFaultyPDisks) {
+        auto result = RequestHcWithVdisks(NKikimrBlobStorage::TGroupStatus::PARTIAL, TVDisks{3, {NKikimrBlobStorage::READY, NKikimrBlobStorage::FAULTY}});
+        Cerr << result.ShortDebugString() << Endl;
+        CheckHcResultHasIssuesWithStatus(result, "STORAGE_GROUP", Ydb::Monitoring::StatusFlag::YELLOW, 1);
+        CheckHcResultHasIssuesWithStatus(result, "STORAGE_GROUP", Ydb::Monitoring::StatusFlag::ORANGE, 0);
+        CheckHcResultHasIssuesWithStatus(result, "STORAGE_GROUP", Ydb::Monitoring::StatusFlag::RED, 0);
+    }
+
     /* HC currently infers group status on its own, so it's never unknown
     Y_UNIT_TEST(RedGroupIssueWhenUnknownGroupStatus) {
         auto result = RequestHcWithVdisks(NKikimrBlobStorage::TGroupStatus::UNKNOWN, {});
-Original file line number
+Diff line change
@@ Expand Up @@
                 }
             }
+            // do not propagate RED status to vdisk - so that vdisk is not considered down when computing group status
+            context.OverallStatus = MinStatus(context.OverallStatus, Ydb::Monitoring::StatusFlag::ORANGE);
             storagePDiskStatus.set_overall(context.GetOverallStatus());
         }
@@ Expand Down @@