Reapply "remove leftoever impossible to reach kubelet logs due to getHostAliases"

This reverts commit 86af906.

Author: mrdoggopat

pkg/util/cloudproviders/BUILD.bazel

@@ -10,6 +10,7 @@ go_library(
     importpath = "github.com/DataDog/datadog-agent/pkg/util/cloudproviders",
     visibility = ["//visibility:public"],
     deps = [
+        "//pkg/config/helper",
         "//pkg/config/setup",
         "//pkg/util/cloudproviders/alibaba",
         "//pkg/util/cloudproviders/azure",

pkg/util/cloudproviders/cloudproviders.go

@@ -12,6 +12,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/DataDog/datadog-agent/pkg/config/helper"
 	configsetup "github.com/DataDog/datadog-agent/pkg/config/setup"
 	"github.com/DataDog/datadog-agent/pkg/util/hostname/validate"
 	"github.com/DataDog/datadog-agent/pkg/util/kubelet"
@@ -134,16 +135,34 @@ var (
 	hostAliasLogOnce = true
 )
 
+// kubeletDependentHostAliasDetectors are the host-alias detectors that route through
+// the shared kubelet client singleton. Cluster Checks Runners are Deployment
+// replicas, not DaemonSets, so they are never colocated with a node's kubelet and can
+// never reach it.
+var kubeletDependentHostAliasDetectors = map[string]bool{
+	"kubelet":                    true,
+	kubernetes.CloudProviderName: true,
+}
+
 // GetHostAliases returns the hostname aliases and the name of the possible cloud providers
 func GetHostAliases(ctx context.Context) ([]string, string) {
 	aliases := []string{}
 	cloudprovider := ""
+	isCLCRunner := helper.IsCLCRunner(configsetup.Datadog())
 
 	// cloud providers endpoints can take a few seconds to answer. We're using a WaitGroup to call all of them
 	// concurrently since GetHostAliases is called during the agent startup and is blocking.
 	var wg sync.WaitGroup
 
 	for _, hostAliasesDetector := range hostAliasesDetectors {
+		if isCLCRunner && kubeletDependentHostAliasDetectors[hostAliasesDetector.name] {
+			// Skip probing the kubelet client singleton: it can never succeed on a
+			// CCR and would otherwise trigger its exponential-backoff retrier and
+			// its "Impossible to reach Kubelet" warning for no benefit.
+			log.Debugf("Skipping %s Host Alias: Agent is a Cluster Checks Runner and has no reachable local Kubelet", hostAliasesDetector.name)
+			continue
+		}
+
 		wg.Add(1)
 		go func(hostAliasesDetector cloudProviderAliasesDetector) {
 			defer wg.Done()

pkg/util/cloudproviders/cloudproviders_test.go

@@ -64,6 +64,55 @@ func TestCloudProviderAliases(t *testing.T) {
 	assert.Contains(t, []string{"detector1", "detector3"}, cloudprovider)
 }
 
+// TestCloudProviderAliasesSkipsKubeletDependentDetectorsOnCLCRunner ensures
+// that GetHostAliases never invokes the "kubelet" or "kubernetes" detectors
+// on a Cluster Checks Runner, since CCRs are Deployment replicas (not
+// DaemonSets) and can never reach a local kubelet.
+func TestCloudProviderAliasesSkipsKubeletDependentDetectorsOnCLCRunner(t *testing.T) {
+	origDetectors := hostAliasesDetectors
+	defer func() { hostAliasesDetectors = origDetectors }()
+
+	config := configmock.New(t)
+	config.SetInTest("clc_runner_enabled", true)
+	config.SetInTest("config_providers", []map[string]interface{}{{"name": "clusterchecks"}})
+
+	kubeletCalled := false
+	kubernetesDetectorCalled := false
+	otherDetectorCalled := false
+
+	hostAliasesDetectors = []cloudProviderAliasesDetector{
+		{
+			name: "kubelet",
+			callback: func(_ context.Context) ([]string, error) {
+				kubeletCalled = true
+				return []string{"kubelet-alias"}, nil
+			},
+		},
+		{
+			name: "kubernetes",
+			callback: func(_ context.Context) ([]string, error) {
+				kubernetesDetectorCalled = true
+				return []string{"kubernetes-alias"}, nil
+			},
+		},
+		{
+			name:       "other",
+			isCloudEnv: true,
+			callback: func(_ context.Context) ([]string, error) {
+				otherDetectorCalled = true
+				return []string{"other-alias"}, nil
+			},
+		},
+	}
+
+	aliases, cloudprovider := GetHostAliases(context.TODO())
+	assert.False(t, kubeletCalled, "kubelet host alias detector should be skipped on a Cluster Checks Runner")
+	assert.False(t, kubernetesDetectorCalled, "kubernetes host alias detector should be skipped on a Cluster Checks Runner")
+	assert.True(t, otherDetectorCalled, "non-kubelet-dependent host alias detectors should still run on a Cluster Checks Runner")
+	assert.Equal(t, []string{"other-alias"}, aliases)
+	assert.Equal(t, "other", cloudprovider)
+}
+
 func TestCloudProviderHostCCRID(t *testing.T) {
 	origDetectors := hostCCRIDDetectors
 	defer func() { hostCCRIDDetectors = origDetectors }()