From a9455bdf826d40c3493885f690865495d21598e4 Mon Sep 17 00:00:00 2001
From: rosstimothy <39066650+rosstimothy@users.noreply.github.com>
Date: Thu, 12 Dec 2024 16:23:57 -0500
Subject: [PATCH] Attempt to reduce flakiness of TestIntegrations/DiscoveryNode
 (#50137)

This applies the same measures as https://github.com/gravitational/teleport/pull/49850
applied to various other flaky integration tests to ensure that
nodes are routable before attempting user connections to them.
Additionally, the debug service is disabled by default for all of
the various ways a TeleInstance can be created. This prevents test
failures on macOS due to unix socket name length.

Closes https://github.com/gravitational/teleport/issues/16310.
---
 integration/helpers/instance.go | 9 +++++++++
 integration/integration_test.go | 6 ++++++
 2 files changed, 15 insertions(+)

diff --git a/integration/helpers/instance.go b/integration/helpers/instance.go
index fb4fdc3947f09..a552193cad8e4 100644
--- a/integration/helpers/instance.go
+++ b/integration/helpers/instance.go
@@ -444,6 +444,7 @@ func (i *TeleInstance) Create(t *testing.T, trustedSecrets []*InstanceSecrets, e
 	tconf.Proxy.DisableWebInterface = true
 	tconf.CircuitBreakerConfig = breaker.NoopBreakerConfig()
 	tconf.InstanceMetadataClient = imds.NewDisabledIMDSClient()
+	tconf.DebugService.Enabled = false
 	return i.CreateEx(t, trustedSecrets, tconf)
 }
 
@@ -469,6 +470,7 @@ func (i *TeleInstance) GenerateConfig(t *testing.T, trustedSecrets []*InstanceSe
 	tconf.Auth.ClusterName, err = services.NewClusterNameWithRandomID(types.ClusterNameSpecV2{
 		ClusterName: i.Secrets.SiteName,
 	})
+	tconf.DebugService.Enabled = false
 	if err != nil {
 		return nil, trace.Wrap(err)
 	}
@@ -737,6 +739,7 @@ func (i *TeleInstance) StartNodeWithTargetPort(tconf *servicecfg.Config, authPor
 	}
 	tconf.Auth.Enabled = false
 	tconf.Proxy.Enabled = false
+	tconf.DebugService.Enabled = false
 
 	// Create a new Teleport process and add it to the list of nodes that
 	// compose this "cluster".
@@ -779,6 +782,7 @@ func (i *TeleInstance) StartApp(conf *servicecfg.Config) (*service.TeleportProce
 	conf.Testing.UploadEventsC = i.UploadEventsC
 	conf.Auth.Enabled = false
 	conf.Proxy.Enabled = false
+	conf.DebugService.Enabled = false
 
 	// Create a new Teleport process and add it to the list of nodes that
 	// compose this "cluster".
@@ -829,6 +833,7 @@ func (i *TeleInstance) StartApps(configs []*servicecfg.Config) ([]*service.Telep
 			cfg.Testing.UploadEventsC = i.UploadEventsC
 			cfg.Auth.Enabled = false
 			cfg.Proxy.Enabled = false
+			cfg.DebugService.Enabled = false
 
 			// Create a new Teleport process and add it to the list of nodes that
 			// compose this "cluster".
@@ -894,6 +899,7 @@ func (i *TeleInstance) StartDatabase(conf *servicecfg.Config) (*service.Teleport
 	conf.Proxy.Enabled = false
 	conf.Apps.Enabled = false
 	conf.SSH.Enabled = false
+	conf.DebugService.Enabled = false
 
 	// Create a new Teleport process and add it to the list of nodes that
 	// compose this "cluster".
@@ -956,6 +962,7 @@ func (i *TeleInstance) StartKube(t *testing.T, conf *servicecfg.Config, clusterN
 	conf.Apps.Enabled = false
 	conf.SSH.Enabled = false
 	conf.Databases.Enabled = false
+	conf.DebugService.Enabled = false
 
 	conf.Kube.KubeconfigPath = filepath.Join(dataDir, "kube_config")
 	if err := EnableKube(t, conf, clusterName); err != nil {
@@ -1031,6 +1038,7 @@ func (i *TeleInstance) StartNodeAndProxy(t *testing.T, name string) (sshPort, we
 	}
 	tconf.CircuitBreakerConfig = breaker.NoopBreakerConfig()
 	tconf.InstanceMetadataClient = imds.NewDisabledIMDSClient()
+	tconf.DebugService.Enabled = false
 
 	// Create a new Teleport process and add it to the list of nodes that
 	// compose this "cluster".
@@ -1124,6 +1132,7 @@ func (i *TeleInstance) StartProxy(cfg ProxyConfig, opts ...Option) (reversetunne
 	tconf.Proxy.DisableALPNSNIListener = cfg.DisableALPNSNIListener
 	tconf.CircuitBreakerConfig = breaker.NoopBreakerConfig()
 	tconf.InstanceMetadataClient = imds.NewDisabledIMDSClient()
+	tconf.DebugService.Enabled = false
 	tconf.FileDescriptors = cfg.FileDescriptors
 	// apply options
 	for _, o := range opts {
diff --git a/integration/integration_test.go b/integration/integration_test.go
index f99ea4c01eafc..96e10fe53c448 100644
--- a/integration/integration_test.go
+++ b/integration/integration_test.go
@@ -4420,12 +4420,18 @@ func testDiscoveryNode(t *testing.T, suite *integrationTestSuite) {
 	helpers.WaitForActiveTunnelConnections(t, main.Tunnel, helpers.Site, 1)
 	helpers.WaitForActiveTunnelConnections(t, proxyTunnel, helpers.Site, 1)
 
+	// Wait for the nodes to be visible to both Proxy instances.
+	require.NoError(t, main.WaitForNodeCount(ctx, helpers.Site, 1))
+	instance := helpers.TeleInstance{Tunnel: proxyTunnel}
+	require.NoError(t, instance.WaitForNodeCount(ctx, helpers.Site, 1))
+
 	// Execute the connection via first proxy.
 	cfg := helpers.ClientConfig{
 		Login:   suite.Me.Username,
 		Cluster: helpers.Site,
 		Host:    "cluster-main-node",
 	}
+
 	output, err := runCommand(t, main, []string{"echo", "hello world"}, cfg, 1)
 	require.NoError(t, err)
 	require.Equal(t, "hello world\n", output)