Add PDisk Move command (#21496)

(cherry picked from commit 63183ec)

Author: SammyVimes

ydb/apps/dstool/lib/commands.py

@@ -7,6 +7,7 @@
 import ydb.apps.dstool.lib.dstool_cmd_pdisk_stop as pdisk_stop
 import ydb.apps.dstool.lib.dstool_cmd_pdisk_restart as pdisk_restart
 import ydb.apps.dstool.lib.dstool_cmd_pdisk_readonly as pdisk_readonly
+import ydb.apps.dstool.lib.dstool_cmd_pdisk_move as pdisk_move
 
 import ydb.apps.dstool.lib.dstool_cmd_vdisk_evict as vdisk_evict
 import ydb.apps.dstool.lib.dstool_cmd_vdisk_list as vdisk_list
@@ -51,14 +52,14 @@
     pool_list, pool_create_virtual,
     group_check, group_decommit, group_show_blob_info, group_show_storage_efficiency, group_show_usage_by_tablets,
     group_state, group_take_snapshot, group_add, group_list, group_virtual_create, group_virtual_cancel,
-    pdisk_add_by_serial, pdisk_remove_by_serial, pdisk_set, pdisk_list, pdisk_stop, pdisk_restart, pdisk_readonly,
+    pdisk_add_by_serial, pdisk_remove_by_serial, pdisk_set, pdisk_list, pdisk_stop, pdisk_restart, pdisk_readonly, pdisk_move,
     vdisk_evict, vdisk_list, vdisk_set_read_only, vdisk_remove_donor, vdisk_wipe,
     device_list,
 ]
 
 default_structure = [
     ('device', ['list']),
-    ('pdisk', ['add-by-serial', 'remove-by-serial', 'set', 'list', 'stop', 'restart', 'readonly']),
+    ('pdisk', ['add-by-serial', 'remove-by-serial', 'set', 'list', 'stop', 'restart', 'readonly', 'move']),
     ('vdisk', ['evict', 'list', 'set-read-only', 'remove-donor', 'wipe']),
     ('group', ['add', 'check', 'decommit', ('show', ['blob-info', 'storage-efficiency', 'usage-by-tablets']), 'state', 'take-snapshot', 'list', ('virtual', ['create', 'cancel'])]),
     ('pool', ['list', ('create', ['virtual'])]),

ydb/apps/dstool/lib/dstool_cmd_pdisk_move.py

@@ -0,0 +1,64 @@
+import ydb.apps.dstool.lib.common as common
+import sys
+
+description = 'Move PDisk'
+
+
+def add_options(p):
+    common.add_pdisk_select_options(p)
+    common.add_ignore_degraded_group_check_option(p)
+    common.add_ignore_failure_model_group_check_option(p)
+    common.add_ignore_vslot_quotas_option(p)
+    common.add_basic_format_options(p)
+    p.add_argument('--source-pdisk', '-s', type=str, required=True,
+                   help='Source PDisk in the format nodeId:pdiskId (e.g. "1:1000")')
+    p.add_argument('--destination-pdisk', '-d', type=str, required=True,
+                   help='Destination PDisk in the format nodeId:pdiskId (e.g. "2:1000")')
+
+
+def create_request(args, source_pdisk, target_pdisk):
+    request = common.create_bsc_request(args)
+    cmd = request.Command.add().MovePDisk
+
+    cmd.SourcePDisk.TargetPDiskId.NodeId = int(source_pdisk.split(';')[0])
+    cmd.SourcePDisk.TargetPDiskId.PDiskId = int(source_pdisk.split(';')[1])
+
+    cmd.DestinationPDisk.TargetPDiskId.NodeId = int(target_pdisk.split(';')[0])
+    cmd.DestinationPDisk.TargetPDiskId.PDiskId = int(target_pdisk.split(';')[1])
+
+    return request
+
+
+def perform_request(request):
+    return common.invoke_bsc_request(request)
+
+
+def is_successful_response(response):
+    return common.is_successful_bsc_response(response)
+
+
+def do(args):
+    assert not args.dry_run, '--dry-run is not supported for this command'
+
+    src = args.source_pdisk
+    if src is None:
+        common.print_status(args, success=False, error_reason='Source PDisk is not specified')
+        sys.exit(1)
+
+    dst = args.destination_pdisk
+    if dst is None:
+        common.print_status(args, success=False, error_reason='Destination PDisk is not specified')
+        sys.exit(1)
+
+    success = True
+    error_reason = ''
+
+    request = create_request(args, src, dst)
+    response = perform_request(request)
+    if not is_successful_response(response):
+        success = False
+        error_reason += 'Request has failed: \n{0}\n{1}\n'.format(request, response)
+
+    common.print_status(args, success, error_reason)
+    if not success:
+        sys.exit(1)

ydb/apps/dstool/lib/ya.make

@@ -18,6 +18,7 @@ PY_SRCS(
     dstool_cmd_pdisk_restart.py
     dstool_cmd_pdisk_set.py
     dstool_cmd_pdisk_stop.py
+    dstool_cmd_pdisk_move.py
 
     dstool_cmd_vdisk_evict.py
     dstool_cmd_vdisk_list.py

ydb/core/blobstorage/nodewarden/node_warden_vdisk.cpp

@@ -25,7 +25,10 @@ namespace NKikimr::NStorage {
         STLOG(PRI_INFO, BS_NODE, NW00, "PoisonLocalVDisk", (VDiskId, vdisk.GetVDiskId()), (VSlotId, vdisk.GetVSlotId()),
             (RuntimeData, vdisk.RuntimeData.has_value()));
 
+        bool vdiskRunning = false;
+
         if (vdisk.RuntimeData) {
+            vdiskRunning = true;
             vdisk.TIntrusiveListItem<TVDiskRecord, TGroupRelationTag>::Unlink();
             TActivationContext::Send(new IEventHandle(TEvents::TSystem::Poison, 0, vdisk.GetVDiskServiceId(), {}, nullptr, 0));
             vdisk.RuntimeData.reset();
@@ -50,7 +53,7 @@ namespace NKikimr::NStorage {
         vdisk.ScrubCookie = 0; // disable reception of Scrub messages from this disk
         vdisk.ScrubCookieForController = 0; // and from controller too
         vdisk.Status = NKikimrBlobStorage::EVDiskStatus::ERROR;
-        vdisk.ShutdownPending = true;
+        vdisk.ShutdownPending = vdiskRunning; // Shutdown pending only if VDisk was running before poison
         VDiskStatusChanged = true;
     }
 

ydb/core/blobstorage/pdisk/mock/pdisk_mock.cpp

@@ -34,7 +34,7 @@ struct TPDiskMockState::TImpl {
 
     const ui32 NodeId;
     const ui32 PDiskId;
-    const ui64 PDiskGuid;
+    ui64 PDiskGuid;
     const ui64 Size;
     const ui32 ChunkSize;
     const ui32 TotalChunks;
@@ -904,6 +904,8 @@ class TPDiskMockActor : public TActorBootstrapped<TPDiskMockActor> {
         bool restartAllowed = ev->Get()->RestartAllowed;
 
         if (restartAllowed) {
+            Become(&TThis::StateNormal);
+            Impl.StateErrorReason = "";
             Impl.IsDiskReadOnly = ev->Get()->Config->ReadOnly;
             Send(ev->Sender, new TEvBlobStorage::TEvNotifyWardenPDiskRestarted(Impl.PDiskId));
         }
@@ -1033,6 +1035,13 @@ class TPDiskMockActor : public TActorBootstrapped<TPDiskMockActor> {
         Send(ev->Sender, new NPDisk::TEvWriteMetadataResult(NPDisk::EPDiskMetadataOutcome::OK, Impl.PDiskGuid), 0, ev->Cookie);
     }
 
+    void Handle(TEvMoveDrive::TPtr& ev) {
+        State = ev->Get()->State;
+        Impl.Blocks = State->Impl->Blocks;
+        Impl.Owners = State->Impl->Owners;
+        Impl.PDiskGuid = State->Impl->PDiskGuid;
+    }
+
     void HandleMoveToErrorState() {
         Impl.StateErrorReason = "Some error reason";
         Become(&TThis::StateError);
@@ -1064,6 +1073,7 @@ class TPDiskMockActor : public TActorBootstrapped<TPDiskMockActor> {
         cFunc(TEvents::TSystem::Wakeup, ReportMetrics);
         hFunc(NPDisk::TEvReadMetadata, Handle);
         hFunc(NPDisk::TEvWriteMetadata, Handle);
+        hFunc(TEvMoveDrive, Handle);
 
         cFunc(EvBecomeError, HandleMoveToErrorState);
 
@@ -1085,6 +1095,8 @@ class TPDiskMockActor : public TActorBootstrapped<TPDiskMockActor> {
         hFunc(NPDisk::TEvReadMetadata, ErrorHandle);
         hFunc(NPDisk::TEvWriteMetadata, ErrorHandle);
         hFunc(TEvBlobStorage::TEvAskWardenRestartPDiskResult, Handle);
+        hFunc(NPDisk::TEvYardControl, Handle);
+        hFunc(TEvMoveDrive, Handle);
 
         cFunc(TEvents::TSystem::Wakeup, ReportMetrics);
         cFunc(EvBecomeNormal, HandleMoveToNormalState);

ydb/core/blobstorage/pdisk/mock/pdisk_mock.h

@@ -9,7 +9,8 @@ namespace NKikimr {
 
     enum EPDiskMockEvents {
         EvBecomeError = TEvBlobStorage::EvEnd + 1,
-        EvBecomeNormal
+        EvBecomeNormal,
+        EvReplaceData,
     };
 
     class TPDiskMockState : public TThrRefBase {
@@ -51,6 +52,14 @@ namespace NKikimr {
         bool IsDiskReadOnly() const;
     };
 
+    struct TEvMoveDrive : TEventLocal<TEvMoveDrive, EvReplaceData> {
+        TIntrusivePtr<TPDiskMockState>& State;
+
+        TEvMoveDrive(TIntrusivePtr<TPDiskMockState>& state)
+            : State(state)
+        {}
+    };
+
     IActor *CreatePDiskMockActor(TPDiskMockState::TPtr state);
 
 } // NKikimr

ydb/core/blobstorage/ut_blobstorage/lib/env.h

@@ -542,7 +542,7 @@ struct TEnvironmentSetup {
     }
 
     void CreateBoxAndPool(ui32 numDrivesPerNode = 0, ui32 numGroups = 0, ui32 numStorageNodes = 0,
-            NKikimrBlobStorage::EPDiskType pdiskType = NKikimrBlobStorage::EPDiskType::ROT) {
+            NKikimrBlobStorage::EPDiskType pdiskType = NKikimrBlobStorage::EPDiskType::ROT, const std::optional<NKikimrBlobStorage::TGroupGeometry>& geometry = std::nullopt) {
         NKikimrBlobStorage::TConfigRequest request;
 
         auto *cmd = request.AddCommand()->MutableDefineHostConfig();
@@ -579,6 +579,10 @@ struct TEnvironmentSetup {
             cmd2->SetEncryptionMode(TBlobStorageGroupInfo::EEncryptionMode::EEM_ENC_V1);
         }
 
+        if (geometry) {
+            cmd2->MutableGeometry()->CopyFrom(*geometry);
+        }
+
         auto response = Invoke(request);
         UNIT_ASSERT_C(response.GetSuccess(), response.GetErrorDescription());
     }