PDisk restart and readonly commands with FQDN and disk's path (#13196)

ydb/apps/dstool/lib/commands.py

@@ -5,6 +5,8 @@
 import ydb.apps.dstool.lib.dstool_cmd_pdisk_set as pdisk_set
 import ydb.apps.dstool.lib.dstool_cmd_pdisk_list as pdisk_list
 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_vdisk_evict as vdisk_evict
 import ydb.apps.dstool.lib.dstool_cmd_vdisk_list as vdisk_list
@@ -49,14 +51,14 @@
     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_add_by_serial, pdisk_remove_by_serial, pdisk_set, pdisk_list, pdisk_stop, pdisk_restart, pdisk_readonly,
     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']),
+    ('pdisk', ['add-by-serial', 'remove-by-serial', 'set', 'list', 'stop', 'restart', 'readonly']),
     ('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_readonly.py

@@ -0,0 +1,58 @@
+import ydb.apps.dstool.lib.common as common
+import sys
+
+description = 'Change PDisk read-only status'
+
+
+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('--enabled', type=str, choices=['true', 'false'], help='Enable read-only mode')
+
+
+def create_request(args, pdisk):
+    request = common.create_bsc_request(args)
+    cmd = request.Command.add().SetPDiskReadOnly
+
+    cmd.TargetPDiskId.NodeId = pdisk[0]
+    cmd.TargetPDiskId.PDiskId = pdisk[1]
+
+    cmd.Value = args.enabled == 'true'
+
+    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):
+    base_config = common.fetch_base_config()
+
+    assert not args.dry_run, '--dry-run is not supported for this command'
+
+    pdisks = common.get_selected_pdisks(args, base_config)
+
+    if len(pdisks) != 1:
+        common.print_status(args, success=False, error_reason='Only change one PDisk read-only status at a time')
+        sys.exit(1)
+
+    success = True
+    error_reason = ''
+
+    request = create_request(args, list(pdisks)[0])
+    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/dstool_cmd_pdisk_restart.py

@@ -0,0 +1,55 @@
+import ydb.apps.dstool.lib.common as common
+import sys
+
+description = 'Restart 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)
+
+
+def create_request(args, pdisk):
+    request = common.create_bsc_request(args)
+    cmd = request.Command.add().RestartPDisk
+
+    cmd.TargetPDiskId.NodeId = pdisk[0]
+    cmd.TargetPDiskId.PDiskId = pdisk[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):
+    base_config = common.fetch_base_config()
+
+    assert not args.dry_run, '--dry-run is not supported for this command'
+
+    pdisks = common.get_selected_pdisks(args, base_config)
+
+    if len(pdisks) != 1:
+        common.print_status(args, success=False, error_reason='Only restart one PDisk at a time')
+        sys.exit(1)
+
+    success = True
+    error_reason = ''
+
+    request = create_request(args, list(pdisks)[0])
+    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/dstool_cmd_pdisk_stop.py

@@ -16,8 +16,8 @@ def create_request(args, pdisk):
     request = common.create_bsc_request(args)
     cmd = request.Command.add().StopPDisk
 
-    cmd.HostKey.NodeId = pdisk[0]
-    cmd.PDiskId = pdisk[1]
+    cmd.TargetPDiskId.NodeId = pdisk[0]
+    cmd.TargetPDiskId.PDiskId = pdisk[1]
 
     return request
 

ydb/apps/dstool/lib/ya.make

@@ -13,7 +13,9 @@ PY_SRCS(
 
     dstool_cmd_pdisk_add_by_serial.py
     dstool_cmd_pdisk_list.py
+    dstool_cmd_pdisk_readonly.py
     dstool_cmd_pdisk_remove_by_serial.py
+    dstool_cmd_pdisk_restart.py
     dstool_cmd_pdisk_set.py
     dstool_cmd_pdisk_stop.py
 

ydb/core/blobstorage/ut_blobstorage/stop_pdisk.cpp

@@ -23,9 +23,9 @@ Y_UNIT_TEST_SUITE(BSCStopPDisk) {
         NKikimrBlobStorage::TConfigRequest request;
 
         NKikimrBlobStorage::TStopPDisk* cmd = request.AddCommand()->MutableStopPDisk();
-        auto* hostKey = cmd->MutableHostKey();
-        hostKey->SetNodeId(targetNodeId);
-        cmd->SetPDiskId(targetPDiskId);
+        auto* pdiskId = cmd->MutableTargetPDiskId();
+        pdiskId->SetNodeId(targetNodeId);
+        pdiskId->SetPDiskId(targetPDiskId);
 
         auto response = env.Invoke(request);
 

ydb/core/mind/bscontroller/cmds_box.cpp

@@ -226,10 +226,45 @@ namespace NKikimr::NBsController {
         boxes.erase(cmd.GetOriginBoxId());
     }
 
-    void TBlobStorageController::TConfigState::ExecuteStep(const NKikimrBlobStorage::TRestartPDisk& cmd, TStatus& /*status*/) {
-        auto targetPDiskId = cmd.GetTargetPDiskId();
+    template <class T>
+    TPDiskId GetPDiskId(const TBlobStorageController::TConfigState& state, const T& command) {
+        if (command.HasTargetPDiskId() && command.HasTargetPDiskLocation()) {
+            throw TExError() << "Only one of TargetPDiskId or PDiskLocation can be specified";
+        }
 
-        TPDiskId pdiskId(targetPDiskId.GetNodeId(), targetPDiskId.GetPDiskId());
+        if (command.HasTargetPDiskId()) {
+            const NKikimrBlobStorage::TPDiskId& pdiskId = command.GetTargetPDiskId();
+            ui32 targetNodeId = pdiskId.GetNodeId();
+            ui32 targetPDiskId = pdiskId.GetPDiskId();
+            if (const auto& hostId = state.HostRecords->GetHostId(targetNodeId)) {
+                TPDiskId target(targetNodeId, targetPDiskId);
+                if (state.PDisks.Find(target) && !state.PDisksToRemove.count(target)) {
+                    return target;
+                }
+                throw TExPDiskNotFound(targetNodeId, targetPDiskId);
+            }
+            throw TExHostNotFound(targetNodeId);
+        } else if (command.HasTargetPDiskLocation()) {
+            const NKikimrBlobStorage::TPDiskLocation& pdiskLocation = command.GetTargetPDiskLocation();
+            const TString& targetFqdn = pdiskLocation.GetFqdn();
+            const TString& targetDiskPath = pdiskLocation.GetPath();
+
+            auto range = state.HostRecords->ResolveNodeId(targetFqdn);
+
+            for (auto it = range.first; it != range.second; ++it) {
+                const TNodeId nodeId = it->second;
+                if (const auto& pdiskId = state.FindPDiskByLocation(nodeId, targetDiskPath)) {
+                    return *pdiskId;
+                }
+            }
+
+            throw TExPDiskNotFound(targetFqdn, targetDiskPath);
+        }
+        throw TExError() << "Either TargetPDiskId or PDiskLocation must be specified";
+    }
+
+    void TBlobStorageController::TConfigState::ExecuteStep(const NKikimrBlobStorage::TRestartPDisk& cmd, TStatus& /*status*/) {
+        TPDiskId pdiskId = GetPDiskId(*this, cmd);
 
         TPDiskInfo *pdisk = PDisks.FindForUpdate(pdiskId);
 
@@ -252,9 +287,7 @@ namespace NKikimr::NBsController {
     }
 
     void TBlobStorageController::TConfigState::ExecuteStep(const NKikimrBlobStorage::TSetPDiskReadOnly& cmd, TStatus& /*status*/) {
-        auto targetPDiskId = cmd.GetTargetPDiskId();
-
-        TPDiskId pdiskId(targetPDiskId.GetNodeId(), targetPDiskId.GetPDiskId());
+        TPDiskId pdiskId = GetPDiskId(*this, cmd);
 
         TPDiskInfo *pdisk = PDisks.FindForUpdate(pdiskId);
 
@@ -281,25 +314,7 @@ namespace NKikimr::NBsController {
     }
 
     void TBlobStorageController::TConfigState::ExecuteStep(const NKikimrBlobStorage::TStopPDisk& cmd, TStatus& /*status*/) {
-        const auto& host = NormalizeHostKey(cmd.GetHostKey());
-
-        TPDiskId pdiskId;
-        if (cmd.GetPDiskId()) {
-            if (cmd.GetPath()) {
-                throw TExError() << "TUpdateDriveStatus.Path and PDiskId are mutually exclusive";
-            }
-            pdiskId = TPDiskId(host.GetNodeId(), cmd.GetPDiskId());
-            if (!PDisks.Find(pdiskId) || PDisksToRemove.count(pdiskId)) {
-                throw TExPDiskNotFound(host, cmd.GetPDiskId(), TString());
-            }
-        } else {
-            const std::optional<TPDiskId> found = FindPDiskByLocation(host.GetNodeId(), cmd.GetPath());
-            if (found && !PDisksToRemove.count(*found)) {
-                pdiskId = *found;
-            } else {
-                throw TExPDiskNotFound(host, 0, cmd.GetPath());
-            }
-        }
+        TPDiskId pdiskId = GetPDiskId(*this, cmd);
 
         TPDiskInfo *pdisk = PDisks.FindForUpdate(pdiskId);
 

ydb/core/mind/bscontroller/error.h

@@ -105,6 +105,11 @@ namespace NKikimr::NBsController {
                 << TErrorParams::IcPort(std::get<1>(hostKey));
         }
 
+        TExHostNotFound(ui32 nodeId) {
+            *this << "Host not found"
+                << TErrorParams::NodeId(nodeId);
+        }
+
         NKikimrBlobStorage::TConfigResponse::TStatus::EFailReason GetFailReason() const override {
             return NKikimrBlobStorage::TConfigResponse::TStatus::kHostNotFound;
         }
@@ -126,6 +131,12 @@ namespace NKikimr::NBsController {
                 << TErrorParams::PDiskId(pdiskId);
         }
 
+        TExPDiskNotFound(const TString& fqdn, TString path) {
+            *this << "PDisk not found"
+                << TErrorParams::Fqdn(fqdn)
+                << TErrorParams::Path(path);
+        }
+
         NKikimrBlobStorage::TConfigResponse::TStatus::EFailReason GetFailReason() const override {
             return NKikimrBlobStorage::TConfigResponse::TStatus::kPDiskNotFound;
         }

ydb/core/mind/bscontroller/impl.h

@@ -10,6 +10,8 @@
 #include "self_heal.h"
 #include "storage_pool_stat.h"
 
+#include <util/generic/hash_multi_map.h>
+
 inline IOutputStream& operator <<(IOutputStream& o, NKikimr::TErasureType::EErasureSpecies p) {
     return o << NKikimr::TErasureType::ErasureSpeciesName(p);
 }
@@ -1432,6 +1434,7 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
     class THostRecordMapImpl {
         THashMap<THostId, THostRecord> HostIdToRecord;
         THashMap<TNodeId, THostId> NodeIdToHostId;
+        THashMultiMap<TString, TNodeId> FqdnToNodeId;
 
     public:
         THostRecordMapImpl() = default;
@@ -1441,6 +1444,7 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
                 const THostId hostId(nodeInfo.Host, nodeInfo.Port);
                 NodeIdToHostId.emplace(nodeInfo.NodeId, hostId);
                 HostIdToRecord.emplace(hostId, nodeInfo);
+                FqdnToNodeId.emplace(nodeInfo.Host, nodeInfo.NodeId);
             }
         }
 
@@ -1450,6 +1454,7 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
                 const TNodeId nodeId = item.GetNodeId();
                 NodeIdToHostId.emplace(nodeId, hostId);
                 HostIdToRecord.emplace(hostId, item);
+                FqdnToNodeId.emplace(item.GetHost(), nodeId);
             }
         }
 
@@ -1482,6 +1487,10 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
             }
         }
 
+        auto ResolveNodeId(const TString& fqdn) const {
+            return FqdnToNodeId.equal_range(fqdn);
+        }
+
         auto begin() const {
             return HostIdToRecord.begin();
         }

ydb/core/protos/blobstorage_config.proto

@@ -400,19 +400,31 @@ message TWipeVDisk {
     NKikimrBlobStorage.TVDiskID VDiskId = 2;
 }
 
+message TPDiskLocation {
+    string Fqdn = 1; // fully qualified domain name of the host
+    string Path = 2; // absolute path to the device as enlisted in PDisk configuration
+}
+
 message TRestartPDisk {
-    NKikimrBlobStorage.TPDiskId TargetPDiskId = 1;
+    oneof PDisk {
+        NKikimrBlobStorage.TPDiskId TargetPDiskId = 1;
+        TPDiskLocation TargetPDiskLocation = 2;   
+    }
 }
 
 message TSetPDiskReadOnly {
-    NKikimrBlobStorage.TPDiskId TargetPDiskId = 1;
-    bool Value = 2;
+    oneof PDisk {
+        NKikimrBlobStorage.TPDiskId TargetPDiskId = 1;
+        TPDiskLocation TargetPDiskLocation = 2;
+    }
+    bool Value = 3;
 }
 
 message TStopPDisk {
-    THostKey HostKey = 1; // host on which we are looking for the drive
-    string Path = 2; // absolute path to the device as enlisted in PDisk configuration
-    uint32 PDiskId = 3; // may be set instead of path to identify PDisk
+    oneof PDisk {
+        NKikimrBlobStorage.TPDiskId TargetPDiskId = 1;
+        TPDiskLocation TargetPDiskLocation = 2;
+    }
 }
 
 message TSetScrubPeriodicity {