GH-20889 - put conditionals are hcp initialization for consul server (#20926)

* put conditionals are hcp initialization for consul server

* put more things behind configuration flags

* add changelog

* TestServer_hcpManager

* fix TestAgent_scadaProvider

.changelog/20926.txt

@@ -0,0 +1,3 @@
+```release-note:bug
+error running consul server in 1.18.0: failed to configure SCADA provider user's home directory path: $HOME is not defined
+```

agent/agent.go

@@ -1320,7 +1320,7 @@ func (a *Agent) listenHTTP() ([]apiServer, error) {
 	}
 
 	httpAddrs := a.config.HTTPAddrs
-	if a.scadaProvider != nil {
+	if a.config.IsCloudEnabled() && a.scadaProvider != nil {
 		httpAddrs = append(httpAddrs, scada.CAPCoreAPI)
 	}
 

agent/agent_test.go

@@ -6341,6 +6341,7 @@ func TestAgent_scadaProvider(t *testing.T) {
 	pvd.EXPECT().Listen(scada.CAPCoreAPI.Capability()).Return(l, nil).Once()
 	pvd.EXPECT().Stop().Return(nil).Once()
 	a := TestAgent{
+		HCL: `cloud = { resource_id = "test-resource-id" client_id = "test-client-id" client_secret = "test-client-secret" }`,
 		OverrideDeps: func(deps *BaseDeps) {
 			deps.HCP.Provider = pvd
 		},

agent/config/builder.go

@@ -1114,8 +1114,8 @@ func (b *builder) build() (rt RuntimeConfig, err error) {
 		LocalProxyConfigResyncInterval:    30 * time.Second,
 	}
 
-	// host metrics are enabled by default to support HashiCorp Cloud Platform integration
-	rt.Telemetry.EnableHostMetrics = boolValWithDefault(c.Telemetry.EnableHostMetrics, true)
+	// host metrics are enabled if consul is configured with HashiCorp Cloud Platform integration
+	rt.Telemetry.EnableHostMetrics = boolValWithDefault(c.Telemetry.EnableHostMetrics, rt.IsCloudEnabled())
 
 	rt.TLS, err = b.buildTLSConfig(rt, c.TLS)
 	if err != nil {

agent/consul/server.go

@@ -595,32 +595,34 @@ func NewServer(config *Config, flat Deps, externalGRPCServer *grpc.Server,
 		StorageBackend: s.raftStorageBackend,
 	})
 
-	s.hcpManager = hcp.NewManager(hcp.ManagerConfig{
-		CloudConfig:       flat.HCP.Config,
-		StatusFn:          s.hcpServerStatus(flat),
-		Logger:            logger.Named("hcp_manager"),
-		SCADAProvider:     flat.HCP.Provider,
-		TelemetryProvider: flat.HCP.TelemetryProvider,
-		ManagementTokenUpserterFn: func(name, secretId string) error {
-			// Check the state of the server before attempting to upsert the token. Otherwise,
-			// the upsert will fail and log errors that do not require action from the user.
-			if s.config.ACLsEnabled && s.IsLeader() && s.InPrimaryDatacenter() {
-				// Idea for improvement: Upsert a token with a well-known accessorId here instead
-				// of a randomly generated one. This would prevent any possible insertion collision between
-				// this and the insertion that happens during the ACL initialization process (initializeACLs function)
-				return s.upsertManagementToken(name, secretId)
-			}
-			return nil
-		},
-		ManagementTokenDeleterFn: func(secretId string) error {
-			// Check the state of the server before attempting to delete the token.Otherwise,
-			// the delete will fail and log errors that do not require action from the user.
-			if s.config.ACLsEnabled && s.IsLeader() && s.InPrimaryDatacenter() {
-				return s.deleteManagementToken(secretId)
-			}
-			return nil
-		},
-	})
+	if s.config.Cloud.IsConfigured() {
+		s.hcpManager = hcp.NewManager(hcp.ManagerConfig{
+			CloudConfig:       flat.HCP.Config,
+			StatusFn:          s.hcpServerStatus(flat),
+			Logger:            logger.Named("hcp_manager"),
+			SCADAProvider:     flat.HCP.Provider,
+			TelemetryProvider: flat.HCP.TelemetryProvider,
+			ManagementTokenUpserterFn: func(name, secretId string) error {
+				// Check the state of the server before attempting to upsert the token. Otherwise,
+				// the upsert will fail and log errors that do not require action from the user.
+				if s.config.ACLsEnabled && s.IsLeader() && s.InPrimaryDatacenter() {
+					// Idea for improvement: Upsert a token with a well-known accessorId here instead
+					// of a randomly generated one. This would prevent any possible insertion collision between
+					// this and the insertion that happens during the ACL initialization process (initializeACLs function)
+					return s.upsertManagementToken(name, secretId)
+				}
+				return nil
+			},
+			ManagementTokenDeleterFn: func(secretId string) error {
+				// Check the state of the server before attempting to delete the token.Otherwise,
+				// the delete will fail and log errors that do not require action from the user.
+				if s.config.ACLsEnabled && s.IsLeader() && s.InPrimaryDatacenter() {
+					return s.deleteManagementToken(secretId)
+				}
+				return nil
+			},
+		})
+	}
 
 	var recorder *middleware.RequestRecorder
 	if flat.NewRequestRecorderFunc != nil {
@@ -890,22 +892,24 @@ func NewServer(config *Config, flat Deps, externalGRPCServer *grpc.Server,
 	// to enable RPC forwarding.
 	s.grpcLeaderForwarder = flat.LeaderForwarder
 
-	// Start watching HCP Link resource. This needs to be created after
-	// the GRPC services are set up in order for the resource service client to
-	// function. This uses the insecure grpc channel so that it doesn't need to
-	// present a valid ACL token.
-	go hcp.RunHCPLinkWatcher(
-		&lib.StopChannelContext{StopCh: shutdownCh},
-		logger.Named("hcp-link-watcher"),
-		pbresource.NewResourceServiceClient(s.insecureSafeGRPCChan),
-		hcp.HCPManagerLifecycleFn(
-			s.hcpManager,
-			hcpclient.NewClient,
-			bootstrap.LoadManagementToken,
-			flat.HCP.Config,
-			flat.HCP.DataDir,
-		),
-	)
+	if s.config.Cloud.IsConfigured() {
+		// Start watching HCP Link resource. This needs to be created after
+		// the GRPC services are set up in order for the resource service client to
+		// function. This uses the insecure grpc channel so that it doesn't need to
+		// present a valid ACL token.
+		go hcp.RunHCPLinkWatcher(
+			&lib.StopChannelContext{StopCh: shutdownCh},
+			logger.Named("hcp-link-watcher"),
+			pbresource.NewResourceServiceClient(s.insecureSafeGRPCChan),
+			hcp.HCPManagerLifecycleFn(
+				s.hcpManager,
+				hcpclient.NewClient,
+				bootstrap.LoadManagementToken,
+				flat.HCP.Config,
+				flat.HCP.DataDir,
+			),
+		)
+	}
 
 	s.controllerManager = controller.NewManager(
 		// Usage of the insecure + unsafe grpc chan is required for the controller
@@ -1008,13 +1012,15 @@ func isV1CatalogRequest(rpcName string) bool {
 }
 
 func (s *Server) registerControllers(deps Deps, proxyUpdater ProxyUpdater) error {
-	hcpctl.RegisterControllers(
-		s.controllerManager, hcpctl.ControllerDependencies{
-			ResourceApisEnabled:    s.useV2Resources,
-			HCPAllowV2ResourceApis: s.hcpAllowV2Resources,
-			CloudConfig:            deps.HCP.Config,
-		},
-	)
+	if s.config.Cloud.IsConfigured() {
+		hcpctl.RegisterControllers(
+			s.controllerManager, hcpctl.ControllerDependencies{
+				ResourceApisEnabled:    s.useV2Resources,
+				HCPAllowV2ResourceApis: s.hcpAllowV2Resources,
+				CloudConfig:            deps.HCP.Config,
+			},
+		)
+	}
 
 	// When not enabled, the v1 tenancy bridge is used by default.
 	if s.useV2Tenancy {
@@ -2075,8 +2081,10 @@ func (s *Server) trackLeaderChanges() {
 			s.raftStorageBackend.LeaderChanged()
 			s.controllerManager.SetRaftLeader(s.IsLeader())
 
-			// Trigger sending an update to HCP status
-			s.hcpManager.SendUpdate()
+			if s.config.Cloud.IsConfigured() {
+				// Trigger sending an update to HCP status
+				s.hcpManager.SendUpdate()
+			}
 		case <-s.shutdownCh:
 			s.raft.DeregisterObserver(observer)
 			return

agent/consul/server_test.go

@@ -2100,6 +2100,9 @@ func TestServer_hcpManager(t *testing.T) {
 	// Configure the server for the StatusFn
 	conf1.BootstrapExpect = 1
 	conf1.RPCAdvertise = &net.TCPAddr{IP: []byte{127, 0, 0, 2}, Port: conf1.RPCAddr.Port}
+	conf1.Cloud.ClientID = "test-client-id"
+	conf1.Cloud.ResourceID = "test-resource-id"
+	conf1.Cloud.ClientSecret = "test-client-secret"
 	hcp1 := hcpclient.NewMockClient(t)
 	hcp1.EXPECT().PushServerStatus(mock.Anything, mock.MatchedBy(func(status *hcpclient.ServerStatus) bool {
 		return status.ID == string(conf1.NodeID)

agent/setup.go

@@ -138,15 +138,17 @@ func NewBaseDeps(configLoader ConfigLoader, logOut io.Writer, providedLogger hcl
 	cfg.Telemetry.PrometheusOpts.SummaryDefinitions = summaries
 
 	var extraSinks []metrics.MetricSink
-	// This values is set late within newNodeIDFromConfig above
-	cfg.Cloud.NodeID = cfg.NodeID
+	if cfg.IsCloudEnabled() {
+		// This values is set late within newNodeIDFromConfig above
+		cfg.Cloud.NodeID = cfg.NodeID
 
-	d.HCP, err = hcp.NewDeps(cfg.Cloud, d.Logger.Named("hcp"), cfg.DataDir)
-	if err != nil {
-		return d, err
-	}
-	if d.HCP.Sink != nil {
-		extraSinks = append(extraSinks, d.HCP.Sink)
+		d.HCP, err = hcp.NewDeps(cfg.Cloud, d.Logger.Named("hcp"), cfg.DataDir)
+		if err != nil {
+			return d, err
+		}
+		if d.HCP.Sink != nil {
+			extraSinks = append(extraSinks, d.HCP.Sink)
+		}
 	}
 
 	d.MetricsConfig, err = lib.InitTelemetry(cfg.Telemetry, d.Logger, extraSinks...)

command/agent/agent.go

@@ -181,12 +181,9 @@ func (c *cmd) run(args []string) int {
 			ui.Error(err.Error())
 			return 1
 		}
-	}
 
-	// We unconditionally add an Access Control header to our config in order to allow the HCP UI to work.
-	// We do this unconditionally because the cluster can be linked to HCP at any time (not just at startup) and this
-	// is simpler than selectively reloading parts of config at runtime.
-	loader = hcpbootstrap.AddAclPolicyAccessControlHeader(loader)
+		loader = hcpbootstrap.AddAclPolicyAccessControlHeader(loader)
+	}
 
 	bd, err := agent.NewBaseDeps(loader, logGate, nil)
 	if err != nil {