Add stable node names (ydb-platform#5759)

dorooleg · Jun 20, 2024 · 9242688 · 9242688
1 parent 49e8858
commit 9242688
Show file tree

Hide file tree

Showing 30 changed files with 631 additions and 30 deletions.
diff --git a/ydb/core/base/appdata_fwd.h b/ydb/core/base/appdata_fwd.h
@@ -240,6 +240,7 @@ struct TAppData {
     TVector<TString> DefaultUserSIDs;
     TString AllAuthenticatedUsers = "all-users@well-known";
     TString TenantName;
+    TString NodeName;
 
     TIntrusivePtr<TResourceProfiles> ResourceProfiles;
 

diff --git a/ydb/core/driver_lib/cli_utils/cli_cmds_server.cpp b/ydb/core/driver_lib/cli_utils/cli_cmds_server.cpp
@@ -722,8 +722,23 @@ class TClientCommandServerBase : public TClientCommand {
             TRACE_CONFIG_CHANGE_INPLACE_T(TenantPoolConfig, UpdateExplicitly);
         }
 
-        if (config.ParseResult->Has("tenant") && InterconnectPort != DefaultInterconnectPort) {
-            AppConfig.MutableMonitoringConfig()->SetHostLabelOverride(HostAndICPort());
+        if (config.ParseResult->Has("tenant")) {
+            if (AppConfig.GetDynamicNodeConfig().GetNodeInfo().HasName()) {
+                const TString& nodeName = AppConfig.GetDynamicNodeConfig().GetNodeInfo().GetName();
+                AppConfig.MutableMonitoringConfig()->SetHostLabelOverride(nodeName);
+                TRACE_CONFIG_CHANGE_INPLACE_T(MonitoringConfig, UpdateExplicitly);
+            } else if (InterconnectPort != DefaultInterconnectPort) {
+                AppConfig.MutableMonitoringConfig()->SetHostLabelOverride(HostAndICPort());
+                TRACE_CONFIG_CHANGE_INPLACE_T(MonitoringConfig, UpdateExplicitly);
+            }
+        }
+
+        if (config.ParseResult->Has("tenant")) {
+            if (InterconnectPort == DefaultInterconnectPort) {
+                AppConfig.MutableMonitoringConfig()->SetProcessLocation(Host());
+            } else {
+                AppConfig.MutableMonitoringConfig()->SetProcessLocation(HostAndICPort());
+            }
             TRACE_CONFIG_CHANGE_INPLACE_T(MonitoringConfig, UpdateExplicitly);
         }
 
@@ -989,11 +1004,18 @@ class TClientCommandServerBase : public TClientCommand {
         ShuffleRange(addrs);
     }
 
-    TString HostAndICPort() {
+    TString HostAndICPort() const {
+        auto hostname = Host();
+        if (!hostname) {
+            return "";
+        }
+        return TStringBuilder() << hostname << ":" << InterconnectPort;
+    }
+
+    TString Host() const {
         try {
             auto hostname = to_lower(HostName());
-            hostname = hostname.substr(0, hostname.find('.'));
-            return TStringBuilder() << hostname << ":" << InterconnectPort;
+            return hostname.substr(0, hostname.find('.'));
         } catch (TSystemError& error) {
             return "";
         }
@@ -1015,6 +1037,11 @@ class TClientCommandServerBase : public TClientCommand {
                 result = TryToRegisterDynamicNodeViaDiscoveryService(addr, domainName, NodeHost, NodeAddress, NodeResolveHost, GetSchemePath());
                 if (result.IsSuccess()) {
                     Cout << "Success. Registered via discovery service as " << result.GetNodeId() << Endl;
+                    Cout << "Node name: ";
+                    if (result.HasNodeName()) {
+                        Cout << result.GetNodeName();
+                    }
+                    Cout << Endl;
                     break;
                 }
                 Cerr << "Registration error: " << static_cast<NYdb::TStatus>(result) << Endl;
@@ -1052,6 +1079,9 @@ class TClientCommandServerBase : public TClientCommand {
                 nodeInfo->SetAddress(node.Address);
                 nodeInfo->SetExpire(node.Expire);
                 CopyNodeLocation(nodeInfo->MutableLocation(), node.Location);
+                if (result.HasNodeName()) {
+                    nodeInfo->SetName(result.GetNodeName());
+                }
             } else {
                 auto &info = *nsConfig.AddNode();
                 info.SetNodeId(node.NodeId);

diff --git a/ydb/core/driver_lib/run/kikimr_services_initializers.cpp b/ydb/core/driver_lib/run/kikimr_services_initializers.cpp
@@ -2116,7 +2116,7 @@ void TKqpServiceInitializer::InitializeServices(NActors::TActorSystemSetup* setu
 
         auto kqpProxySharedResources = std::make_shared<NKqp::TKqpProxySharedResources>();
 
-        // Crate resource manager
+        // Create resource manager
         auto rm = NKqp::CreateKqpResourceManagerActor(Config.GetTableServiceConfig().GetResourceManager(), nullptr,
             {}, kqpProxySharedResources);
         setup->LocalServices.push_back(std::make_pair(

diff --git a/ydb/core/driver_lib/run/run.cpp b/ydb/core/driver_lib/run/run.cpp
@@ -1075,6 +1075,10 @@ void TKikimrRunner::InitializeAppData(const TKikimrRunConfig& runConfig)
 
     AppData->TenantName = runConfig.TenantName;
 
+    if (runConfig.AppConfig.GetDynamicNodeConfig().GetNodeInfo().HasName()) {
+        AppData->NodeName = runConfig.AppConfig.GetDynamicNodeConfig().GetNodeInfo().GetName();
+    }
+
     if (runConfig.AppConfig.HasBootstrapConfig()) {
         AppData->BootstrapConfig = runConfig.AppConfig.GetBootstrapConfig();
     }

diff --git a/ydb/core/grpc_services/rpc_node_registration.cpp b/ydb/core/grpc_services/rpc_node_registration.cpp
@@ -106,6 +106,9 @@ class TNodeRegistrationRPC : public TActorBootstrapped<TNodeRegistrationRPC> {
 
         auto request = TEvNodeRegistrationRequest::GetProtoRequest(Request);
         Result.set_node_id(rec.GetNode().GetNodeId());
+        if (rec.GetNode().HasName()) {
+            Result.set_node_name(rec.GetNode().GetName());
+        }
         Result.set_expire(rec.GetNode().GetExpire());
         Result.set_domain_path(request->domain_path());
         CopyNodeInfo(Result.add_nodes(), rec.GetNode());

diff --git a/ydb/core/mind/hive/hive_impl.cpp b/ydb/core/mind/hive/hive_impl.cpp
@@ -1974,6 +1974,9 @@ void THive::Handle(TEvHive::TEvRequestHiveNodeStats::TPtr& ev) {
         if (!node.ServicedDomains.empty()) {
             nodeStats.MutableNodeDomain()->CopyFrom(node.ServicedDomains.front());
         }
+        if (!node.Name.empty()) {
+            nodeStats.SetNodeName(node.Name);
+        }
         if (request.GetReturnExtendedTabletInfo()) {
             if (request.HasFilterTabletsByPathId()) {
                 auto itTabletsOfObject = node.TabletsOfObject.find({request.GetFilterTabletsBySchemeShardId(), request.GetFilterTabletsByPathId()});

diff --git a/ydb/core/mind/hive/hive_schema.h b/ydb/core/mind/hive/hive_schema.h
@@ -192,9 +192,10 @@ struct Schema : NIceDb::Schema {
         struct Drain : Column<6, NScheme::NTypeIds::Bool> { static constexpr bool Default = false; };
         struct DrainInitiators : Column<8, NScheme::NTypeIds::String> { using Type = TVector<TActorId>; };
         struct Location : Column<9, NScheme::NTypeIds::String> { using Type = NActorsInterconnect::TNodeLocation; };
+        struct Name : Column<10, NScheme::NTypeIds::String> {};
 
         using TKey = TableKey<ID>;
-        using TColumns = TableColumns<ID, Local, Down, Freeze, ServicedDomains, Statistics, Drain, DrainInitiators, Location>;
+        using TColumns = TableColumns<ID, Local, Down, Freeze, ServicedDomains, Statistics, Drain, DrainInitiators, Location, Name>;
     };
 
     struct TabletCategory : Table<6> {

diff --git a/ydb/core/mind/hive/node_info.h b/ydb/core/mind/hive/node_info.h
@@ -60,6 +60,7 @@ struct TNodeInfo {
     THashSet<TLeaderTabletInfo*> LockedTablets;
     mutable TInstant LastResourceChangeReaction;
     NKikimrHive::TNodeStatistics Statistics;
+    TString Name;
 
     TNodeInfo(TNodeId nodeId, THive& hive);
     TNodeInfo(const TNodeInfo&) = delete;

diff --git a/ydb/core/mind/hive/tx__load_everything.cpp b/ydb/core/mind/hive/tx__load_everything.cpp
@@ -314,6 +314,7 @@ class TTxLoadEverything : public TTransactionBase<THive> {
                 node.DrainInitiators = nodeRowset.GetValueOrDefault<Schema::Node::DrainInitiators>();
                 node.ServicedDomains = nodeRowset.GetValueOrDefault<Schema::Node::ServicedDomains>();
                 node.Statistics = nodeRowset.GetValueOrDefault<Schema::Node::Statistics>();
+                node.Name = nodeRowset.GetValueOrDefault<Schema::Node::Name>();
                 if (nodeRowset.HaveValue<Schema::Node::Location>()) {
                     auto location = nodeRowset.GetValue<Schema::Node::Location>();
                     if (location.HasDataCenter()) {

diff --git a/ydb/core/mind/hive/tx__register_node.cpp b/ydb/core/mind/hive/tx__register_node.cpp
@@ -41,7 +41,14 @@ class TTxRegisterNode : public TTransactionBase<THive> {
             } else {
                 Sort(servicedDomains);
             }
-            db.Table<Schema::Node>().Key(nodeId).Update<Schema::Node::Local, Schema::Node::ServicedDomains, Schema::Node::Statistics>(Local, servicedDomains, node.Statistics);
+            const TString& name = Record.GetName();
+            db.Table<Schema::Node>().Key(nodeId).Update(
+                NIceDb::TUpdate<Schema::Node::Local>(Local),
+                NIceDb::TUpdate<Schema::Node::ServicedDomains>(servicedDomains),
+                NIceDb::TUpdate<Schema::Node::Statistics>(node.Statistics),
+                NIceDb::TUpdate<Schema::Node::Name>(name)
+            );
+
             node.BecomeDisconnected();
             if (node.LastSeenServicedDomains != servicedDomains) {
                 // new tenant - new rules
@@ -52,6 +59,7 @@ class TTxRegisterNode : public TTransactionBase<THive> {
             node.Local = Local;
             node.ServicedDomains.swap(servicedDomains);
             node.LastSeenServicedDomains = node.ServicedDomains;
+            node.Name = name;
         }
         if (Record.HasSystemLocation() && Record.GetSystemLocation().HasDataCenter()) {
             node.Location = TNodeLocation(Record.GetSystemLocation());

diff --git a/ydb/core/mind/local.cpp b/ydb/core/mind/local.cpp
@@ -199,6 +199,9 @@ class TLocalNodeRegistrar : public TActorBootstrapped<TLocalNodeRegistrar> {
             }
             tabletAvailability->SetPriority(tabletInfo.Priority);
         }
+        if (const TString& nodeName = AppData(ctx)->NodeName; !nodeName.Empty()) {
+            request->Record.SetName(nodeName);
+        }
 
         NTabletPipe::SendData(ctx, HivePipeClient, request.Release());