Add health check extension for liveness & readiness probes (#1779)

HussainRaza28 · web-flow · commit 14ef2e81218b · 2025-08-06T15:22:39.000-04:00
diff --git a/translator/tocwconfig/sampleConfig/appsignals_and_eks_config.yaml b/translator/tocwconfig/sampleConfig/appsignals_and_eks_config.yaml
@@ -313,6 +313,9 @@ extensions:
         objects:
             - endpointslices
             - services
+    health_check/health_check:
+        endpoint: 0.0.0.0:13133
+        path: /
     server:
         listen_addr: :4311
         tls_ca_path: /etc/amazon-cloudwatch-observability-agent-client-cert/tls-ca.crt
@@ -1538,6 +1541,7 @@ service:
         - k8smetadata
         - entitystore
         - server
+        - health_check/health_check
     pipelines:
         metrics/application_signals:
             exporters:
diff --git a/translator/tocwconfig/sampleConfig/appsignals_and_k8s_config.yaml b/translator/tocwconfig/sampleConfig/appsignals_and_k8s_config.yaml
@@ -314,6 +314,9 @@ extensions:
         objects:
             - endpointslices
             - services
+    health_check/health_check:
+        endpoint: 0.0.0.0:13133
+        path: /
     server:
         listen_addr: :4311
         tls_ca_path: /etc/amazon-cloudwatch-observability-agent-client-cert/tls-ca.crt
@@ -1533,6 +1536,7 @@ service:
         - k8smetadata
         - entitystore
         - server
+        - health_check/health_check
     pipelines:
         metrics/application_signals:
             exporters:
diff --git a/translator/tocwconfig/sampleConfig/appsignals_fallback_and_eks_config.yaml b/translator/tocwconfig/sampleConfig/appsignals_fallback_and_eks_config.yaml
@@ -313,6 +313,9 @@ extensions:
         objects:
             - endpointslices
             - services
+    health_check/health_check:
+        endpoint: 0.0.0.0:13133
+        path: /
     server:
         listen_addr: :4311
         tls_ca_path: /etc/amazon-cloudwatch-observability-agent-client-cert/tls-ca.crt
@@ -1538,6 +1541,7 @@ service:
         - k8smetadata
         - entitystore
         - server
+        - health_check/health_check
     pipelines:
         metrics/application_signals:
             exporters:
diff --git a/translator/tocwconfig/sampleConfig/appsignals_over_fallback_config.yaml b/translator/tocwconfig/sampleConfig/appsignals_over_fallback_config.yaml
@@ -318,6 +318,9 @@ extensions:
         tls_ca_path: /etc/amazon-cloudwatch-observability-agent-client-cert/tls-ca.crt
         tls_cert_path: /etc/amazon-cloudwatch-observability-agent-server-cert/server.crt
         tls_key_path: /etc/amazon-cloudwatch-observability-agent-server-cert/server.key
+    health_check/health_check:
+        endpoint: 0.0.0.0:13133
+        path: /
 processors:
     awsapplicationsignals:
         limiter:
@@ -1538,6 +1541,7 @@ service:
         - k8smetadata
         - entitystore
         - server
+        - health_check/health_check
     pipelines:
         metrics/application_signals:
             exporters:
diff --git a/translator/tocwconfig/sampleConfig/otlp_metrics_cloudwatchlogs_eks_config.yaml b/translator/tocwconfig/sampleConfig/otlp_metrics_cloudwatchlogs_eks_config.yaml
@@ -53,6 +53,9 @@ extensions:
     k8smetadata:
         objects:
             - endpointslices
+    health_check/health_check:
+        endpoint: 0.0.0.0:13133
+        path: /
     server:
         listen_addr: :4311
         tls_ca_path: /etc/amazon-cloudwatch-observability-agent-client-cert/tls-ca.crt
@@ -107,6 +110,7 @@ service:
         - agenthealth/statuscode
         - entitystore
         - server
+        - health_check/health_check
     pipelines:
         metrics/hostOtlpMetrics/cloudwatchlogs:
             exporters:
diff --git a/translator/tocwconfig/sampleConfig/otlp_metrics_eks_config.yaml b/translator/tocwconfig/sampleConfig/otlp_metrics_eks_config.yaml
@@ -31,6 +31,9 @@ extensions:
     k8smetadata:
         objects:
             - endpointslices
+    health_check/health_check:
+        endpoint: 0.0.0.0:13133
+        path: /
     server:
         listen_addr: :4311
         tls_ca_path: /etc/amazon-cloudwatch-observability-agent-client-cert/tls-ca.crt
@@ -91,6 +94,7 @@ service:
         - agenthealth/statuscode
         - entitystore
         - server
+        - health_check/health_check
     pipelines:
         metrics/hostOtlpMetrics:
             exporters:
diff --git a/translator/tocwconfig/sampleConfig/statsd_eks_config.yaml b/translator/tocwconfig/sampleConfig/statsd_eks_config.yaml
@@ -28,6 +28,9 @@ extensions:
         kubernetes_mode: EKS
         mode: ec2
         region: us-west-2
+    health_check/health_check:
+        endpoint: 0.0.0.0:13133
+        path: /
     server:
         listen_addr: :4311
         tls_ca_path: /etc/amazon-cloudwatch-observability-agent-client-cert/tls-ca.crt
@@ -44,6 +47,7 @@ service:
         - agenthealth/statuscode
         - entitystore
         - server
+        - health_check/health_check
     pipelines:
         metrics/hostCustomMetrics:
             exporters:
diff --git a/translator/translate/otel/extension/healthcheck/translator.go b/translator/translate/otel/extension/healthcheck/translator.go
@@ -0,0 +1,37 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: MIT
+
+package healthcheck
+
+import (
+	"go.opentelemetry.io/collector/component"
+	"go.opentelemetry.io/collector/confmap"
+
+	"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/common"
+)
+
+type healthCheckTranslator struct {
+	name string
+}
+
+var _ common.Translator[component.Config, component.ID] = (*healthCheckTranslator)(nil)
+
+func NewHealthCheckTranslator() common.Translator[component.Config, component.ID] {
+	return &healthCheckTranslator{name: "health_check"}
+}
+
+func (t *healthCheckTranslator) ID() component.ID {
+	return component.NewIDWithName(component.MustNewType("health_check"), t.name)
+}
+
+func (t *healthCheckTranslator) Translate(_ *confmap.Conf) (component.Config, error) {
+	cfg := &struct {
+		Endpoint string `mapstructure:"endpoint"`
+		Path     string `mapstructure:"path"`
+	}{
+		Endpoint: "0.0.0.0:13133",
+		Path:     "/",
+	}
+
+	return cfg, nil
+}
diff --git a/translator/translate/otel/extension/healthcheck/translator_test.go b/translator/translate/otel/extension/healthcheck/translator_test.go
@@ -0,0 +1,29 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: MIT
+
+package healthcheck
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"go.opentelemetry.io/collector/confmap"
+)
+
+func TestHealthCheckTranslator(t *testing.T) {
+	translator := NewHealthCheckTranslator()
+	assert.Equal(t, "health_check", translator.ID().Type().String())
+
+	conf := confmap.New()
+	cfg, err := translator.Translate(conf)
+	assert.NoError(t, err)
+
+	// Assert the config has the expected fields
+	healthCheckCfg, ok := cfg.(*struct {
+		Endpoint string `mapstructure:"endpoint"`
+		Path     string `mapstructure:"path"`
+	})
+	assert.True(t, ok)
+	assert.Equal(t, "0.0.0.0:13133", healthCheckCfg.Endpoint)
+	assert.Equal(t, "/", healthCheckCfg.Path)
+}
diff --git a/translator/translate/otel/translate_otel.go b/translator/translate/otel/translate_otel.go
@@ -23,6 +23,7 @@ import (
 	"github.com/aws/amazon-cloudwatch-agent/translator/context"
 	"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/common"
 	"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/extension/entitystore"
+	healthcheckextension "github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/extension/healthcheck"
 	"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/extension/server"
 	pipelinetranslator "github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/pipeline"
 	"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/pipeline/applicationsignals"
@@ -92,6 +93,7 @@ func Translate(jsonConfig interface{}, os string) (*otelcol.Config, error) {
 	}
 	if context.CurrentContext().KubernetesMode() != "" {
 		pipelines.Translators.Extensions.Set(server.NewTranslator())
+		pipelines.Translators.Extensions.Set(healthcheckextension.NewHealthCheckTranslator())
 	}
 
 	cfg := &otelcol.Config{
diff --git a/translator/translate/otel/translate_otel_test.go b/translator/translate/otel/translate_otel_test.go
@@ -13,13 +13,71 @@ import (
 
 	"github.com/aws/amazon-cloudwatch-agent/tool/testutil"
 	"github.com/aws/amazon-cloudwatch-agent/translator"
+	"github.com/aws/amazon-cloudwatch-agent/translator/context"
 	_ "github.com/aws/amazon-cloudwatch-agent/translator/registerrules"
 	"github.com/aws/amazon-cloudwatch-agent/translator/translate/agent"
 	"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/common"
 	"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/pipeline/prometheus"
 	"github.com/aws/amazon-cloudwatch-agent/translator/util/eksdetector"
 )
 
+func TestHealthCheckExtension(t *testing.T) {
+	agent.Global_Config.Region = "us-east-1"
+
+	// Test case 1: Non-Kubernetes environment should NOT have health check extension
+	input := map[string]interface{}{
+		"metrics": map[string]interface{}{
+			"metrics_collected": map[string]interface{}{
+				"cpu": map[string]interface{}{},
+			},
+		},
+	}
+
+	cfg, err := Translate(input, "linux")
+	require.NoError(t, err)
+	require.NotNil(t, cfg)
+
+	// Verify that the health check extension is NOT registered for non-Kubernetes
+	extensionFound := false
+	for _, ext := range cfg.Service.Extensions {
+		if ext.Type().String() == "health_check" {
+			extensionFound = true
+			break
+		}
+	}
+	assert.False(t, extensionFound, "Health check extension should NOT be registered for non-Kubernetes environments")
+
+	// Test case 2: Kubernetes environment should have health check extension
+	inputK8s := map[string]interface{}{
+		"logs": map[string]interface{}{
+			"metrics_collected": map[string]interface{}{
+				"kubernetes": map[string]interface{}{
+					"cluster_name": "TestCluster",
+				},
+			},
+		},
+	}
+
+	// Set Kubernetes mode in context for this test
+	ctx := context.CurrentContext()
+	ctx.SetKubernetesMode("EKS")
+	defer ctx.SetKubernetesMode("") // Reset after test
+
+	cfgK8s, err := Translate(inputK8s, "linux")
+	require.NoError(t, err)
+	require.NotNil(t, cfgK8s)
+
+	// Verify that the health check extension IS registered for Kubernetes
+	extensionFoundK8s := false
+	for _, ext := range cfgK8s.Service.Extensions {
+		if ext.Type().String() == "health_check" {
+			extensionFoundK8s = true
+			break
+		}
+	}
+	assert.True(t, extensionFoundK8s, "Health check extension should be registered for Kubernetes environments")
+}
+
 func TestTranslator(t *testing.T) {
 	agent.Global_Config.Region = "us-east-1"
 	testutil.SetPrometheusRemoteWriteTestingEnv(t)

Original file line number	Diff line number	Diff line change
`@@ -23,6 +23,7 @@ import (`
`23`	`23`	`"github.com/aws/amazon-cloudwatch-agent/translator/context"`
`24`	`24`	`"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/common"`
`25`	`25`	`"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/extension/entitystore"`
	`26`	`+ healthcheckextension "github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/extension/healthcheck"`
`26`	`27`	`"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/extension/server"`
`27`	`28`	`pipelinetranslator "github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/pipeline"`
`28`	`29`	`"github.com/aws/amazon-cloudwatch-agent/translator/translate/otel/pipeline/applicationsignals"`
`@@ -92,6 +93,7 @@ func Translate(jsonConfig interface{}, os string) (*otelcol.Config, error) {`
`92`	`93`	`}`
`93`	`94`	`if context.CurrentContext().KubernetesMode() != "" {`
`94`	`95`	`pipelines.Translators.Extensions.Set(server.NewTranslator())`
	`96`	`+ pipelines.Translators.Extensions.Set(healthcheckextension.NewHealthCheckTranslator())`
`95`	`97`	`}`
`96`	`98`
`97`	`99`	`cfg := &otelcol.Config{`