Add option to restart PDisk without any checks (#13155)

(cherry picked from commit 2490271)
ydb-platform · Jan 3, 2025 · 7bd4196 · 7bd4196
1 parent 518ba6d
commit 7bd4196
Show file tree

Hide file tree

Showing 8 changed files with 78 additions and 19 deletions.
diff --git a/ydb/core/blobstorage/base/blobstorage_events.h b/ydb/core/blobstorage/base/blobstorage_events.h
@@ -374,9 +374,11 @@ namespace NKikimr {
 
     struct TEvBlobStorage::TEvAskWardenRestartPDisk : TEventLocal<TEvAskWardenRestartPDisk, EvAskWardenRestartPDisk> {
         const ui32 PDiskId;
+        const bool IgnoreChecks;
 
-        TEvAskWardenRestartPDisk(const ui32& pdiskId)
+        TEvAskWardenRestartPDisk(const ui32 pdiskId, const bool ignoreChecks)
             : PDiskId(pdiskId)
+            , IgnoreChecks(ignoreChecks)
         {}
     };
 

diff --git a/ydb/core/blobstorage/nodewarden/blobstorage_node_warden_ut.cpp b/ydb/core/blobstorage/nodewarden/blobstorage_node_warden_ut.cpp
@@ -910,7 +910,7 @@ Y_UNIT_TEST_SUITE(TBlobStorageWardenTest) {
 
         runtime.RegisterService(pdiskServiceId, pdiskActorId);
 
-        runtime.Send(new IEventHandle(nodeWarden, pdiskActorId, new TEvBlobStorage::TEvAskWardenRestartPDisk(pdiskId), 0, cookie), nodeId);
+        runtime.Send(new IEventHandle(nodeWarden, pdiskActorId, new TEvBlobStorage::TEvAskWardenRestartPDisk(pdiskId, false), 0, cookie), nodeId);
 
         auto responseEvent = new TEvBlobStorage::TEvControllerConfigResponse();
 

diff --git a/ydb/core/blobstorage/nodewarden/node_warden_impl.cpp b/ydb/core/blobstorage/nodewarden/node_warden_impl.cpp
@@ -541,7 +541,7 @@ void TNodeWarden::Handle(TEvBlobStorage::TEvAskWardenRestartPDisk::TPtr ev) {
 
     PDiskRestartRequests[requestCookie] = pdiskId;
 
-    AskBSCToRestartPDisk(pdiskId, requestCookie);
+    AskBSCToRestartPDisk(pdiskId, ev->Get()->IgnoreChecks, requestCookie);
 }
 
 void TNodeWarden::Handle(TEvBlobStorage::TEvNotifyWardenPDiskRestarted::TPtr ev) {

diff --git a/ydb/core/blobstorage/nodewarden/node_warden_impl.h b/ydb/core/blobstorage/nodewarden/node_warden_impl.h
@@ -216,7 +216,7 @@ namespace NKikimr::NStorage {
 
         TIntrusivePtr<TPDiskConfig> CreatePDiskConfig(const NKikimrBlobStorage::TNodeWardenServiceSet::TPDisk& pdisk);
         void StartLocalPDisk(const NKikimrBlobStorage::TNodeWardenServiceSet::TPDisk& pdisk);
-        void AskBSCToRestartPDisk(ui32 pdiskId, ui64 requestCookie);
+        void AskBSCToRestartPDisk(ui32 pdiskId, bool ignoreDegradedGroups, ui64 requestCookie);
         void OnPDiskRestartFinished(ui32 pdiskId, NKikimrProto::EReplyStatus status);
         void DestroyLocalPDisk(ui32 pdiskId);
 

diff --git a/ydb/core/blobstorage/nodewarden/node_warden_pdisk.cpp b/ydb/core/blobstorage/nodewarden/node_warden_pdisk.cpp
@@ -203,10 +203,19 @@ namespace NKikimr::NStorage {
         SendToController(std::move(report));
     }
 
-    void TNodeWarden::AskBSCToRestartPDisk(ui32 pdiskId, ui64 requestCookie) {
+    void TNodeWarden::AskBSCToRestartPDisk(ui32 pdiskId, bool ignoreDegradedGroups, ui64 requestCookie) {
         auto ev = std::make_unique<TEvBlobStorage::TEvControllerConfigRequest>();
 
-        NKikimrBlobStorage::TRestartPDisk* cmd = ev->Record.MutableRequest()->AddCommand()->MutableRestartPDisk();
+        NKikimrBlobStorage::TConfigRequest* request = ev->Record.MutableRequest();
+
+        NKikimrBlobStorage::TRestartPDisk* cmd = request->AddCommand()->MutableRestartPDisk();
+
+        if (ignoreDegradedGroups) {
+            request->SetIgnoreGroupFailModelChecks(true);
+            request->SetIgnoreDegradedGroupsChecks(true);
+            request->SetIgnoreDisintegratedGroupsChecks(true);
+            request->SetIgnoreGroupSanityChecks(true);
+        }
 
         auto targetPDiskId = cmd->MutableTargetPDiskId();
         targetPDiskId->SetNodeId(LocalNodeId);

diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_actor.cpp b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_actor.cpp
@@ -1004,7 +1004,7 @@ class TPDiskActor : public TActorBootstrapped<TPDiskActor> {
             }
 
             Send(ev->Sender, new TEvBlobStorage::TEvNotifyWardenPDiskRestarted(PDisk->PDiskId, NKikimrProto::EReplyStatus::NOTREADY));
-            
+
             return;
         }
 
@@ -1017,7 +1017,7 @@ class TPDiskActor : public TActorBootstrapped<TPDiskActor> {
             NPDisk::TMainKey newMainKey = ev->Get()->MainKey;
 
             SecureWipeBuffer((ui8*)ev->Get()->MainKey.Keys.data(), sizeof(NPDisk::TKey) * ev->Get()->MainKey.Keys.size());
-            
+
             LOG_NOTICE_S(*TlsActivationContext, NKikimrServices::BS_PDISK, "PDiskId# " << PDisk->PDiskId
                     << " Going to restart PDisk since received TEvAskWardenRestartPDiskResult");
 
@@ -1031,7 +1031,7 @@ class TPDiskActor : public TActorBootstrapped<TPDiskActor> {
             TIntrusivePtr<TPDiskConfig> actorCfg = std::move(Cfg);
 
             auto& newCfg = ev->Get()->Config;
-            
+
             if (newCfg) {
                 Y_VERIFY_S(newCfg->PDiskId == pdiskId,
                         "New config's PDiskId# " << newCfg->PDiskId << " is not equal to real PDiskId# " << pdiskId);
@@ -1046,7 +1046,7 @@ class TPDiskActor : public TActorBootstrapped<TPDiskActor> {
             TGenericExecutorThread& executorThread = actorCtx.ExecutorThread;
 
             PassAway();
-            
+
             CreatePDiskActor(executorThread, counters, actorCfg, newMainKey, pdiskId, poolId, nodeId);
 
             Send(ev->Sender, new TEvBlobStorage::TEvNotifyWardenPDiskRestarted(pdiskId));
@@ -1125,10 +1125,12 @@ class TPDiskActor : public TActorBootstrapped<TPDiskActor> {
             }
         }
         if (cgi.Has("restartPDisk")) {
+            bool ignoreChecks = "true" == cgi.Get("ignoreChecks");
+
             ui32 cookieIdxPart = NextRestartRequestCookie++;
-            ui64 fullCookie = (((ui64) PDisk->PDiskId) << 32) | cookieIdxPart; // This way cookie will be unique no matter the disk.
+            ui64 fullCookie = (((ui64) PDisk->PDiskId) << 32) | cookieIdxPart; // This way cookie will be unique regardless of the disk.
 
-            Send(NodeWardenServiceId, new TEvBlobStorage::TEvAskWardenRestartPDisk(PDisk->PDiskId), fullCookie);
+            Send(NodeWardenServiceId, new TEvBlobStorage::TEvAskWardenRestartPDisk(PDisk->PDiskId, ignoreChecks), fullCookie);
             // Send responce later when restart command will be received.
             PendingRestartResponse = [this, actor = ev->Sender] (bool restartAllowed, TString& details) {
                 TStringStream jsonBuilder;

diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl_http.cpp b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl_http.cpp
@@ -108,13 +108,32 @@ void TPDisk::RenderState(IOutputStream &str, THttpInfo &httpInfo) {
                         }
                     }
 
+                    function toggleButtonColor() {
+                        var checkbox = document.getElementById("ignoreChecks");
+                        var okButton = document.getElementById("restartOkButton");
+                        
+                        if (checkbox.checked) {
+                            okButton.classList.remove("btn-primary");
+                            okButton.classList.add("btn-danger");
+                        } else {
+                            okButton.classList.remove("btn-danger");
+                            okButton.classList.add("btn-primary");
+                        }
+                    }
+
                     function sendRestartRequest() {
+                        let ignoreChecks = document.getElementById("ignoreChecks").checked;
+                        let requestData = "restartPDisk=";
+                        if (ignoreChecks) {
+                            requestData += "&ignoreChecks=true";
+                        }
                         $.ajax({
                             url: "",
-                            data: "restartPDisk=",
+                            data: requestData,
                             method: "POST",
                             success: reloadPage
                         });
+                        $('#restartModal').modal('hide');
                     }
 
                     function sendStopRequest() {
@@ -127,11 +146,38 @@ void TPDisk::RenderState(IOutputStream &str, THttpInfo &httpInfo) {
                     }
                 </script>
             )___";
-            str << "<button onclick='sendRestartRequest()' name='restartPDisk' class='btn btn-default' ";
-            str << "style='background:LightGray; margin:5px' ";
-            str << ">";
-            str << "Restart";
-            str << "</button>";
+
+            str << R"___(
+                <button type="button" class="btn btn-default" style="background: LightGray; margin: 5px" 
+                        data-toggle="modal" data-target="#restartModal">
+                    Restart
+                </button>
+
+                <div id="restartModal" class="modal fade" tabindex="-1">
+                    <div class="modal-dialog">
+                        <div class="modal-content">
+                            <div class="modal-header">
+                                <button type="button" class="close" data-dismiss="modal">
+                                    <span aria-hidden="true">&times;</span>
+                                </button>
+                                <h4 class="modal-title">Confirm Restart</h4>
+                            </div>
+                            <div class="modal-body">
+                                <p>Are you sure you want to restart?</p>
+                                <div class="checkbox">
+                                    <label>
+                                        <input type="checkbox" id="ignoreChecks" onchange="toggleButtonColor()"> Ignore all checks
+                                    </label>
+                                </div>
+                            </div>
+                            <div class="modal-footer">
+                                <button type="button" class="btn btn-default" data-dismiss="modal">Cancel</button>
+                                <button type="button" class="btn btn-primary" id="restartOkButton" onclick="sendRestartRequest()">Restart</button>
+                            </div>
+                        </div>
+                    </div>
+                </div>
+            )___";
 
             if (Cfg->SectorMap) {
                 str << "<button onclick='sendStopRequest()' name='stopPDisk' class='btn btn-default' ";

diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_ut_actions.cpp b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_ut_actions.cpp
@@ -1136,7 +1136,7 @@ void TTestChunkUnlockRestart::TestFSM(const TActorContext &ctx) {
             SignalDoneEvent();
             break;
         }
-        ctx.Send(NodeWardenId, new TEvBlobStorage::TEvAskWardenRestartPDisk(LastResponse.whiteboardPDiskResult->Record.GetPDiskId()));
+        ctx.Send(NodeWardenId, new TEvBlobStorage::TEvAskWardenRestartPDisk(LastResponse.whiteboardPDiskResult->Record.GetPDiskId(), false));
         break;
     case 30:
         TEST_RESPONSE(EvHarakiri, OK);