Merge b359a87 into bd6396c

Author: adameat

.github/config/muted_ya.txt

@@ -89,7 +89,6 @@ ydb/core/tx/schemeshard/ut_pq_reboots TPqGroupTestReboots.AlterWithReboots-PQCon
 ydb/core/tx/schemeshard/ut_pq_reboots TPqGroupTestReboots.CreateAlter-PQConfigTransactionsAtSchemeShard-true
 ydb/core/tx/schemeshard/ut_pq_reboots TPqGroupTestReboots.CreateAlterDropPqGroupWithReboots-PQConfigTransactionsAtSchemeShard-true
 ydb/core/tx/tiering/ut ColumnShardTiers.TTLUsage
-ydb/core/viewer/tests test.py.test_viewer_healthcheck
 ydb/core/viewer/ut Viewer.TabletMerging
 ydb/core/viewer/ut Viewer.TabletMergingPacked
 ydb/library/actors/http/ut sole chunk chunk

ydb/core/protos/node_whiteboard.proto

@@ -69,6 +69,7 @@ message TEvTabletStateRequest {
     optional string Format = 5; // it could be "packed5"
     optional string GroupBy = 20; // it's either empty or "Type,State" for now
     repeated fixed64 FilterTabletId = 22;
+    optional NKikimrSubDomains.TDomainKey FilterTenantId = 23;
 }
 
 message TEvTabletStateResponse {

ydb/core/tablet/node_whiteboard.cpp

@@ -839,24 +839,32 @@ class TNodeWhiteboardService : public TActorBootstrapped<TNodeWhiteboardService>
     void Handle(TEvWhiteboard::TEvTabletStateRequest::TPtr &ev, const TActorContext &ctx) {
         auto now = TMonotonic::Now();
         const auto& request = ev->Get()->Record;
+        auto matchesFilter = [
+            changedSince = request.has_changedsince() ? request.changedsince() : 0,
+            filterTenantId = request.has_filtertenantid() ? NKikimr::TSubDomainKey(request.filtertenantid()) : NKikimr::TSubDomainKey()
+        ](const NKikimrWhiteboard::TTabletStateInfo& tabletStateInfo) {
+            return tabletStateInfo.changetime() >= changedSince
+                && (!filterTenantId || filterTenantId == NKikimr::TSubDomainKey(tabletStateInfo.tenantid()));
+        };
         std::unique_ptr<TEvWhiteboard::TEvTabletStateResponse> response = std::make_unique<TEvWhiteboard::TEvTabletStateResponse>();
         auto& record = response->Record;
         if (request.format() == "packed5") {
             TEvWhiteboard::TEvTabletStateResponsePacked5* ptr = response->AllocatePackedResponse(TabletStateInfo.size());
             for (const auto& [tabletId, tabletInfo] : TabletStateInfo) {
-                ptr->TabletId = tabletInfo.tabletid();
-                ptr->FollowerId = tabletInfo.followerid();
-                ptr->Generation = tabletInfo.generation();
-                ptr->Type = tabletInfo.type();
-                ptr->State = tabletInfo.state();
-                ++ptr;
+                if (matchesFilter(tabletInfo)) {
+                    ptr->TabletId = tabletInfo.tabletid();
+                    ptr->FollowerId = tabletInfo.followerid();
+                    ptr->Generation = tabletInfo.generation();
+                    ptr->Type = tabletInfo.type();
+                    ptr->State = tabletInfo.state();
+                    ++ptr;
+                }
             }
         } else {
             if (request.groupby().empty()) {
-                ui64 changedSince = request.has_changedsince() ? request.changedsince() : 0;
                 if (request.filtertabletid_size() == 0) {
                     for (const auto& pr : TabletStateInfo) {
-                        if (pr.second.changetime() >= changedSince) {
+                        if (matchesFilter(pr.second)) {
                             NKikimrWhiteboard::TTabletStateInfo& tabletStateInfo = *record.add_tabletstateinfo();
                             Copy(tabletStateInfo, pr.second, request);
                         }
@@ -865,7 +873,7 @@ class TNodeWhiteboardService : public TActorBootstrapped<TNodeWhiteboardService>
                     for (auto tabletId : request.filtertabletid()) {
                         auto it = TabletStateInfo.find({tabletId, 0});
                         if (it != TabletStateInfo.end()) {
-                            if (it->second.changetime() >= changedSince) {
+                            if (matchesFilter(it->second)) {
                                 NKikimrWhiteboard::TTabletStateInfo& tabletStateInfo = *record.add_tabletstateinfo();
                                 Copy(tabletStateInfo, it->second, request);
                             }
@@ -876,6 +884,9 @@ class TNodeWhiteboardService : public TActorBootstrapped<TNodeWhiteboardService>
                 std::unordered_map<std::pair<NKikimrTabletBase::TTabletTypes::EType,
                     NKikimrWhiteboard::TTabletStateInfo::ETabletState>, NKikimrWhiteboard::TTabletStateInfo> stateGroupBy;
                 for (const auto& [id, stateInfo] : TabletStateInfo) {
+                    if (!matchesFilter(stateInfo)) {
+                        continue;
+                    }
                     NKikimrWhiteboard::TTabletStateInfo& state = stateGroupBy[{stateInfo.type(), stateInfo.state()}];
                     auto count = state.count();
                     if (count == 0) {

ydb/core/viewer/json_pipe_req.cpp

@@ -825,13 +825,9 @@ void TViewerPipeClient::HandleResolveResource(TEvTxProxySchemeCache::TEvNavigate
         if (ResourceNavigateResponse->IsOk()) {
             TSchemeCacheNavigate::TEntry& entry(ResourceNavigateResponse->Get()->Request->ResultSet.front());
             SharedDatabase = CanonizePath(entry.Path);
-            if (SharedDatabase == AppData()->TenantName) {
-                Direct = true;
-                Bootstrap(); // retry bootstrap without redirect this time
-            } else {
-                DatabaseBoardInfoResponse = MakeRequestStateStorageEndpointsLookup(SharedDatabase);
-                --Requests; // don't count this request
-            }
+            Direct |= (SharedDatabase == AppData()->TenantName);
+            DatabaseBoardInfoResponse = MakeRequestStateStorageEndpointsLookup(SharedDatabase);
+            --Requests; // don't count this request
         } else {
             AddEvent("Failed to resolve database - shared database not found");
             Direct = true;
@@ -850,7 +846,8 @@ void TViewerPipeClient::HandleResolveDatabase(TEvTxProxySchemeCache::TEvNavigate
                 --Requests; // don't count this request
                 Become(&TViewerPipeClient::StateResolveResource);
             } else {
-                DatabaseBoardInfoResponse = MakeRequestStateStorageEndpointsLookup(CanonizePath(entry.Path));
+                Database = CanonizePath(entry.Path);
+                DatabaseBoardInfoResponse = MakeRequestStateStorageEndpointsLookup(Database);
                 --Requests; // don't count this request
             }
         } else {
@@ -865,7 +862,11 @@ void TViewerPipeClient::HandleResolve(TEvStateStorage::TEvBoardInfo::TPtr& ev) {
     if (DatabaseBoardInfoResponse) {
         DatabaseBoardInfoResponse->Set(std::move(ev));
         if (DatabaseBoardInfoResponse->IsOk()) {
-            return ReplyAndPassAway(MakeForward(GetNodesFromBoardReply(DatabaseBoardInfoResponse->GetRef())));
+            if (Direct) {
+                Bootstrap(); // retry bootstrap without redirect this time
+            } else {
+                ReplyAndPassAway(MakeForward(GetNodesFromBoardReply(DatabaseBoardInfoResponse->GetRef())));
+            }
         } else {
             AddEvent("Failed to resolve database nodes");
             Direct = true;
@@ -902,10 +903,11 @@ void TViewerPipeClient::RedirectToDatabase(const TString& database) {
 
 bool TViewerPipeClient::NeedToRedirect() {
     auto request = GetRequest();
-    if (request) {
+    if (NeedRedirect && request) {
+        NeedRedirect = false;
         Direct |= !request.GetHeader("X-Forwarded-From-Node").empty(); // we're already forwarding
         Direct |= (Database == AppData()->TenantName) || Database.empty(); // we're already on the right node or don't use database filter
-        if (Database && !Direct) {
+        if (Database) {
             RedirectToDatabase(Database); // to find some dynamic node and redirect query there
             return true;
         }

ydb/core/viewer/json_pipe_req.h

@@ -42,6 +42,7 @@ class TViewerPipeClient : public TActorBootstrapped<TViewerPipeClient> {
     TString Database;
     TString SharedDatabase;
     bool Direct = false;
+    bool NeedRedirect = true;
     ui32 Requests = 0;
     bool PassedAway = false;
     ui32 MaxRequestsInFlight = 200;