FIx restoration shadow disk from backup (#2934)

* FIx restoration shadow disk from backup * Fix nits
ydb-platform · Jan 28, 2025 · ca6a2ee · ca6a2ee
1 parent ef9294e
commit ca6a2ee
Show file tree

Hide file tree

Showing 4 changed files with 153 additions and 8 deletions.
diff --git a/cloud/blockstore/libs/storage/disk_registry/actors/restore_validator_actor.cpp b/cloud/blockstore/libs/storage/disk_registry/actors/restore_validator_actor.cpp
@@ -233,7 +233,7 @@ TStringBuf NormalizeMirrorId(TStringBuf diskId) {
 
 bool CheckMirrorDiskId(
     const TSet<TString>& disksInSS,
-    const TVector<NProto::TDiskConfig> disksInBackup,
+    const TVector<NProto::TDiskConfig>& disksInBackup,
     const NProto::TDiskConfig& disk)
 {
     const TString& diskId = disk.GetDiskId();
@@ -532,12 +532,21 @@ void TRestoreValidationActor::HandleListVolumesResponse(
                 ValidSnapshot.DisksToCleanup,
                 NormalizeMirrorId(itr->GetDiskId())))
         {
-            LOG_WARN_S(
-                ctx,
-                Component,
-                RESTORE_PREFIX
-                    << " DiskID " << itr->GetDiskId().Quote()
-                    << " is found in backup but not in SS");
+            const bool isShadowDisk =
+                !itr->GetCheckpointReplica().GetCheckpointId().empty();
+            if (isShadowDisk) {
+                LOG_WARN_S(
+                    ctx,
+                    Component,
+                    RESTORE_PREFIX << " ShadowDisk " << itr->GetDiskId().Quote()
+                                   << " is found in backup");
+            } else {
+                LOG_WARN_S(
+                    ctx,
+                    Component,
+                    RESTORE_PREFIX << " DiskID " << itr->GetDiskId().Quote()
+                                   << " is found in backup, but not in SS");
+            }
             SetErrorDevicesInBackup(itr->GetDeviceUUIDs(), ctx.Now());
             DisksInBackup.erase(itr->GetDiskId());
             itr = ValidSnapshot.Disks.erase(itr);

diff --git a/cloud/blockstore/libs/storage/disk_registry/actors/restore_validator_actor_ut.cpp b/cloud/blockstore/libs/storage/disk_registry/actors/restore_validator_actor_ut.cpp
@@ -790,6 +790,86 @@ Y_UNIT_TEST_SUITE(TRestoreValidatorActorTest)
         UNIT_ASSERT_EQUAL(
             state.Disks[4].GetDiskId(), "Disk 4/1");
     }
+
+    Y_UNIT_TEST_F(CheckSkipRestoreShadowDisk, TSetupEnvironment)
+    {
+        TDiskRegistryStateSnapshot backup;
+        {   // Source disk
+            auto& diskConfig = backup.Disks.emplace_back();
+            diskConfig.SetDiskId("Disk 1");
+            diskConfig.SetFolderId("Folder 1");
+            diskConfig.SetCloudId("Cloud 1");
+            diskConfig.SetBlockSize(41);
+        }
+        {   // checkpoint disk
+            auto& diskConfig = backup.Disks.emplace_back();
+            diskConfig.SetDiskId("Disk 1-cp1");
+            diskConfig.SetFolderId("Folder 1");
+            diskConfig.SetCloudId("Cloud 1");
+            diskConfig.SetBlockSize(41);
+            auto* checkpoint = diskConfig.MutableCheckpointReplica();
+            checkpoint->SetSourceDiskId("Disk 1");
+            checkpoint->SetCheckpointId("cp1");
+        }
+
+        auto validatorId = ActorSystem.Register(
+            new TRestoreValidationActor(EdgeActor, {}, 0, backup));
+
+        ActorSystem.GrabEdgeEvent<TEvService::TEvListVolumesRequest>();
+
+        auto volumeListResponse =
+            std::make_unique<TEvService::TEvListVolumesResponse>();
+        volumeListResponse->Record.AddVolumes("Disk 1");
+
+        ActorSystem.Send(new NActors::IEventHandle(
+            validatorId,
+            EdgeActor,
+            volumeListResponse.release()));
+
+        ActorSystem.GrabEdgeEvent<TEvSSProxy::TEvDescribeVolumeRequest>();
+        {
+            NKikimrSchemeOp::TPathDescription description;
+            auto* mutableVolumeConfig =
+                description.MutableBlockStoreVolumeDescription()
+                    ->MutableVolumeConfig();
+            mutableVolumeConfig->SetDiskId("Disk 1");
+            mutableVolumeConfig->SetBlockSize(41);
+            mutableVolumeConfig->SetFolderId("Folder 1");
+            mutableVolumeConfig->SetCloudId("Cloud 1");
+            auto describeVolumeResponse =
+                std::make_unique<TEvSSProxy::TEvDescribeVolumeResponse>(
+                    "",
+                    std::move(description));
+
+            ActorSystem.Send(new NActors::IEventHandle(
+                validatorId,
+                EdgeActor,
+                describeVolumeResponse.release()));
+        }
+
+        {
+            UNIT_ASSERT_EQUAL(
+                ActorSystem.GrabEdgeEvent<TEvVolume::TEvGetVolumeInfoRequest>()
+                    ->Record.GetDiskId(),
+                "Disk 1");
+
+            auto volumeInfoResponse =
+                std::make_unique<TEvVolume::TEvGetVolumeInfoResponse>();
+            auto& volume = *volumeInfoResponse->Record.MutableVolume();
+            volume.SetDiskId("Disk 1");
+
+            ActorSystem.Send(new NActors::IEventHandle(
+                validatorId,
+                EdgeActor,
+                volumeInfoResponse.release()));
+        }
+
+        auto response = ActorSystem.GrabEdgeEvent<
+            TEvDiskRegistryPrivate::TEvRestoreDiskRegistryValidationResponse>();
+        auto& state = response->LoadDBState;
+        UNIT_ASSERT_EQUAL(state.Disks.size(), 1);
+        UNIT_ASSERT_EQUAL(state.Disks[0].GetDiskId(), "Disk 1");
+    }
 }
 
 }   // namespace NDiskRegistry

diff --git a/cloud/blockstore/libs/storage/volume/actors/shadow_disk_actor.cpp b/cloud/blockstore/libs/storage/volume/actors/shadow_disk_actor.cpp
@@ -775,7 +775,9 @@ void TShadowDiskActor::HandleShadowDiskAcquired(
     }
 
     if (HasError(msg->Error)) {
-        if (acquireReason != EAcquireReason::PeriodicalReAcquire) {
+        if (msg->Error.GetCode() == E_NOT_FOUND ||
+            acquireReason != EAcquireReason::PeriodicalReAcquire)
+        {
             SetErrorState(ctx);
         }
         return;

diff --git a/cloud/blockstore/libs/storage/volume/volume_ut_checkpoint.cpp b/cloud/blockstore/libs/storage/volume/volume_ut_checkpoint.cpp
@@ -3725,6 +3725,60 @@ Y_UNIT_TEST_SUITE(TVolumeCheckpointTest)
         }
     }
 
+    Y_UNIT_TEST(ShouldStopAcquiringAfterENotFound)
+    {
+        NProto::TStorageServiceConfig config;
+        config.SetUseShadowDisksForNonreplDiskCheckpoints(true);
+        config.SetMaxAcquireShadowDiskTotalTimeoutWhenNonBlocked(2000);
+
+        auto runtime = PrepareTestActorRuntime(config);
+
+        auto describeDiskRequestsFilter = [&](TAutoPtr<IEventHandle>& event)
+        {
+            if (event->GetTypeRewrite() ==
+                TEvDiskRegistry::EvDescribeDiskResponse)
+            {   // Simulate response with E_NOT_FOUND error from DiskRegistry.
+                auto* msg =
+                    event->Get<TEvDiskRegistry::TEvDescribeDiskResponse>();
+                msg->Record.MutableError()->SetCode(E_NOT_FOUND);
+            }
+            return TTestActorRuntime::DefaultObserverFunc(event);
+        };
+        runtime->SetObserverFunc(describeDiskRequestsFilter);
+
+        // Create volume.
+        TVolumeClient volume(*runtime);
+        volume.UpdateVolumeConfig(
+            0,
+            0,
+            0,
+            0,
+            false,
+            1,
+            NCloud::NProto::STORAGE_MEDIA_SSD_NONREPLICATED,
+            32768);
+
+        volume.WaitReady();
+
+        auto clientInfo = CreateVolumeClientInfo(
+            NProto::VOLUME_ACCESS_READ_WRITE,
+            NProto::VOLUME_MOUNT_LOCAL,
+            0);
+        volume.AddClient(clientInfo);
+
+        // Create checkpoint.
+        volume.CreateCheckpoint("c1");
+
+        // Reconnect pipe since partition has restarted.
+        volume.ReconnectPipe();
+
+        // Shadow disk entered the error state.
+        auto status =
+            volume.GetCheckpointStatus("c1")->Record.GetCheckpointStatus();
+
+        UNIT_ASSERT_EQUAL(NProto::ECheckpointStatus::ERROR, status);
+    }
+
     Y_UNIT_TEST(ShouldBlockWritesWhenReAcquire)
     {
         NProto::TStorageServiceConfig config;