feat(probe): add startup probes (#1234)

Signed-off-by: Wassim DHIF <[email protected]>
DataDog · Jul 9, 2024 · b6bd845 · b6bd845
1 parent eb49fb8
commit b6bd845
Show file tree

Hide file tree

Showing 8 changed files with 70 additions and 1 deletion.
diff --git a/apis/datadoghq/common/common.go b/apis/datadoghq/common/common.go
@@ -48,6 +48,24 @@ func GetDefaultReadinessProbe() *corev1.Probe {
 	return readinessProbe
 }
 
+// GetDefaultStartupProbe creates a defaulted StartupProbe
+func GetDefaultStartupProbe() *corev1.Probe {
+	startupProbe := &corev1.Probe{
+		InitialDelaySeconds: DefaultStartupProbeInitialDelaySeconds,
+		PeriodSeconds:       DefaultStartupProbePeriodSeconds,
+		TimeoutSeconds:      DefaultStartupProbeTimeoutSeconds,
+		SuccessThreshold:    DefaultStartupProbeSuccessThreshold,
+		FailureThreshold:    DefaultStartupProbeFailureThreshold,
+	}
+	startupProbe.HTTPGet = &corev1.HTTPGetAction{
+		Path: DefaultStartupProbeHTTPPath,
+		Port: intstr.IntOrString{
+			IntVal: DefaultAgentHealthPort,
+		},
+	}
+	return startupProbe
+}
+
 // GetDefaultTraceAgentProbe creates a defaulted liveness/readiness probe for the Trace Agent
 func GetDefaultTraceAgentProbe() *corev1.Probe {
 	probe := &corev1.Probe{

diff --git a/apis/datadoghq/common/const.go b/apis/datadoghq/common/const.go
@@ -81,13 +81,15 @@ const (
 	// DefaultHelmCheckConf default Helm Check ConfigMap name
 	DefaultHelmCheckConf string = "helm-check-config"
 
+	// DefaultAgentHealthPort default agent health port
+	DefaultAgentHealthPort int32 = 5555
+
 	// Liveness probe default config
 	DefaultLivenessProbeInitialDelaySeconds int32 = 15
 	DefaultLivenessProbePeriodSeconds       int32 = 15
 	DefaultLivenessProbeTimeoutSeconds      int32 = 5
 	DefaultLivenessProbeSuccessThreshold    int32 = 1
 	DefaultLivenessProbeFailureThreshold    int32 = 6
-	DefaultAgentHealthPort                  int32 = 5555
 	DefaultLivenessProbeHTTPPath                  = "/live"
 
 	// Readiness probe default config
@@ -98,6 +100,14 @@ const (
 	DefaultReadinessProbeFailureThreshold    int32 = 6
 	DefaultReadinessProbeHTTPPath                  = "/ready"
 
+	// Startup probe default config
+	DefaultStartupProbeInitialDelaySeconds int32 = 15
+	DefaultStartupProbePeriodSeconds       int32 = 15
+	DefaultStartupProbeTimeoutSeconds      int32 = 5
+	DefaultStartupProbeSuccessThreshold    int32 = 1
+	DefaultStartupProbeFailureThreshold    int32 = 6
+	DefaultStartupProbeHTTPPath                  = "/startup"
+
 	// Default Image name
 	DefaultAgentImageName        string = "agent"
 	DefaultClusterAgentImageName string = "cluster-agent"

diff --git a/controllers/datadogagent/agent_test.go b/controllers/datadogagent/agent_test.go
@@ -126,6 +126,24 @@ func defaultReadinessProbe() *corev1.Probe {
 	}
 }
 
+func defaultStartupProbe() *corev1.Probe {
+	return &corev1.Probe{
+		InitialDelaySeconds: 15,
+		PeriodSeconds:       15,
+		TimeoutSeconds:      5,
+		SuccessThreshold:    1,
+		FailureThreshold:    6,
+		ProbeHandler: corev1.ProbeHandler{
+			HTTPGet: &corev1.HTTPGetAction{
+				Path: "/startup",
+				Port: intstr.IntOrString{
+					IntVal: 5555,
+				},
+			},
+		},
+	}
+}
+
 func defaultVolumes() []corev1.Volume {
 	return []corev1.Volume{
 		{

diff --git a/controllers/datadogagent/clusteragent_test.go b/controllers/datadogagent/clusteragent_test.go
@@ -61,6 +61,7 @@ func clusterAgentDefaultPodSpec() corev1.PodSpec {
 				},
 				LivenessProbe:  defaultLivenessProbe(),
 				ReadinessProbe: defaultReadinessProbe(),
+				StartupProbe:   defaultStartupProbe(),
 				SecurityContext: &corev1.SecurityContext{
 					ReadOnlyRootFilesystem:   apiutils.NewBoolPointer(true),
 					AllowPrivilegeEscalation: apiutils.NewBoolPointer(false),

diff --git a/controllers/datadogagent/component/agent/default.go b/controllers/datadogagent/component/agent/default.go
@@ -155,6 +155,7 @@ func coreAgentContainer(dda metav1.Object) corev1.Container {
 		VolumeMounts:   volumeMountsForCoreAgent(),
 		LivenessProbe:  apicommon.GetDefaultLivenessProbe(),
 		ReadinessProbe: apicommon.GetDefaultReadinessProbe(),
+		StartupProbe:   apicommon.GetDefaultStartupProbe(),
 	}
 }
 

diff --git a/controllers/datadogagent/component/clusteragent/default.go b/controllers/datadogagent/component/clusteragent/default.go
@@ -102,6 +102,7 @@ func defaultPodSpec(dda metav1.Object, volumes []corev1.Volume, volumeMounts []c
 				VolumeMounts:   volumeMounts,
 				LivenessProbe:  apicommon.GetDefaultLivenessProbe(),
 				ReadinessProbe: apicommon.GetDefaultReadinessProbe(),
+				StartupProbe:   apicommon.GetDefaultStartupProbe(),
 				Command:        nil,
 				Args:           nil,
 				SecurityContext: &corev1.SecurityContext{

diff --git a/controllers/datadogagent/component/clusteragent/default_test.go b/controllers/datadogagent/component/clusteragent/default_test.go
@@ -80,6 +80,7 @@ func clusterAgentDefaultPodSpec() corev1.PodSpec {
 				},
 				LivenessProbe:  defaultLivenessProbe(),
 				ReadinessProbe: defaultReadinessProbe(),
+				StartupProbe:   defaultStartupProbe(),
 				SecurityContext: &corev1.SecurityContext{
 					ReadOnlyRootFilesystem:   apiutils.NewBoolPointer(true),
 					AllowPrivilegeEscalation: apiutils.NewBoolPointer(false),
@@ -191,6 +192,24 @@ func defaultReadinessProbe() *corev1.Probe {
 	}
 }
 
+func defaultStartupProbe() *corev1.Probe {
+	return &corev1.Probe{
+		InitialDelaySeconds: 15,
+		PeriodSeconds:       15,
+		TimeoutSeconds:      5,
+		SuccessThreshold:    1,
+		FailureThreshold:    6,
+		ProbeHandler: corev1.ProbeHandler{
+			HTTPGet: &corev1.HTTPGetAction{
+				Path: "/startup",
+				Port: intstr.IntOrString{
+					IntVal: 5555,
+				},
+			},
+		},
+	}
+}
+
 func clusterAgentDefaultEnvVars() []corev1.EnvVar {
 	return []corev1.EnvVar{
 		{

diff --git a/controllers/datadogagent/component/clusterchecksrunner/default.go b/controllers/datadogagent/component/clusterchecksrunner/default.go
@@ -233,6 +233,7 @@ func defaultPodSpec(dda metav1.Object, volumes []corev1.Volume, volumeMounts []c
 				},
 				LivenessProbe:  apicommon.GetDefaultLivenessProbe(),
 				ReadinessProbe: apicommon.GetDefaultReadinessProbe(),
+				StartupProbe:   apicommon.GetDefaultStartupProbe(),
 				SecurityContext: &corev1.SecurityContext{
 					ReadOnlyRootFilesystem:   apiutils.NewBoolPointer(true),
 					AllowPrivilegeEscalation: apiutils.NewBoolPointer(false),