Merge pull request #361 from arangodb/feature/disable-probes

Allow to disable probes

Author: neunhoef

docs/Manual/Deployment/Kubernetes/DeploymentResource.md

@@ -401,6 +401,14 @@ for each server of this group.
 This setting is not available for group `coordinators`, `syncmasters` & `syncworkers`
 because servers in these groups do not need persistent storage.
 
+### `spec.<group>.probes.livenessProbeDisabled: bool`
+
+If set to true, the operator does not generate a liveness probe for new pods belonging to this group.
+
+### `spec.<group>.probes.readinessProbeDisabled: bool`
+
+If set to true, the operator does not generate a readiness probe for new pods belonging to this group.
+
 ### `spec.<group>.tolerations: []Toleration`
 
 This setting specifies the `tolerations` for the `Pod`s created

pkg/apis/deployment/v1alpha/server_group_spec.go

@@ -56,6 +56,40 @@ type ServerGroupSpec struct {
 	ServiceAccountName *string `json:"serviceAccountName,omitempty"`
 	// NodeSelector speficies a set of selectors for nodes
 	NodeSelector map[string]string `json:"nodeSelector,omitempty"`
+	// Probes specifies additional behaviour for probes
+	Probes *ServerGroupProbesSpec `json:"probes,omitempty"`
+}
+
+// ServerGroupProbesSpec contains specification for probes for pods of the server group
+type ServerGroupProbesSpec struct {
+	// LivenessProbeDisabled if true livenessProbes are disabled
+	LivenessProbeDisabled *bool `json:"livenessProbeDisabled,omitempty"`
+	// LivenessProbeDisabled if specified the given probes is used as liveness probe
+	//LivenessProbeOverride *v1.Probe `json:"LivenessProbeOverride,omitempty"`
+	// LivenessProbeDisabled if true readinessProbes are disabled
+	ReadinessProbeDisabled *bool `json:"ReadinessProbeDisabled,omitempty"`
+	// ReadinessProbeOverride if specified the given probes is used as readiness probe
+	//ReadinessProbeOverride *v1.Probe `json:"ReadinessProbeOverride,omitempty"`
+}
+
+// // HasLivenessProbeOverride returns true if a livenessprobe override is set
+// func (s ServerGroupProbesSpec) HasLivenessProbeOverride() bool {
+// 	return s.LivenessProbeOverride != nil
+// }
+
+// // HasReadinessProbeOverride returns true if a readinessprobe override is set
+// func (s ServerGroupProbesSpec) HasReadinessProbeOverride() bool {
+// 	return s.ReadinessProbeOverride != nil
+// }
+
+// IsLivenessProbeDisabled returns true if liveness probes are disabled
+func (s ServerGroupProbesSpec) IsLivenessProbeDisabled() bool {
+	return util.BoolOrDefault(s.LivenessProbeDisabled)
+}
+
+// IsReadinessProbeDisabled returns true if readiness probes are disabled
+func (s ServerGroupProbesSpec) IsReadinessProbeDisabled() bool {
+	return util.BoolOrDefault(s.ReadinessProbeDisabled)
 }
 
 // GetCount returns the value of count.
@@ -98,6 +132,19 @@ func (s ServerGroupSpec) GetServiceAccountName() string {
 	return util.StringOrDefault(s.ServiceAccountName)
 }
 
+// HasProbesSpec returns true if Probes is non nil
+func (s ServerGroupSpec) HasProbesSpec() bool {
+	return s.Probes != nil
+}
+
+// GetProbesSpec returns the Probes spec or the nil value if not set
+func (s ServerGroupSpec) GetProbesSpec() ServerGroupProbesSpec {
+	if s.HasProbesSpec() {
+		return *s.Probes
+	}
+	return ServerGroupProbesSpec{}
+}
+
 // Validate the given group spec
 func (s ServerGroupSpec) Validate(group ServerGroup, used bool, mode DeploymentMode, env Environment) error {
 	if used {

pkg/apis/deployment/v1alpha/zz_generated.deepcopy.go

@@ -331,6 +331,15 @@ func createArangoSyncArgs(apiObject metav1.Object, spec api.DeploymentSpec, grou
 
 // createLivenessProbe creates configuration for a liveness probe of a server in the given group.
 func (r *Resources) createLivenessProbe(spec api.DeploymentSpec, group api.ServerGroup) (*k8sutil.HTTPProbeConfig, error) {
+	groupspec := spec.GetServerGroupSpec(group)
+
+	if groupspec.HasProbesSpec() {
+		probesspec := groupspec.GetProbesSpec()
+		if probesspec.IsLivenessProbeDisabled() {
+			return nil, nil
+		}
+	}
+
 	switch group {
 	case api.ServerGroupSingle, api.ServerGroupAgents, api.ServerGroupDBServers:
 		authorization := ""
@@ -394,9 +403,19 @@ func (r *Resources) createLivenessProbe(spec api.DeploymentSpec, group api.Serve
 
 // createReadinessProbe creates configuration for a readiness probe of a server in the given group.
 func (r *Resources) createReadinessProbe(spec api.DeploymentSpec, group api.ServerGroup, version driver.Version) (*k8sutil.HTTPProbeConfig, error) {
+	groupspec := spec.GetServerGroupSpec(group)
+
+	if groupspec.HasProbesSpec() {
+		probesspec := groupspec.GetProbesSpec()
+		if probesspec.IsReadinessProbeDisabled() {
+			return nil, nil
+		}
+	}
+
 	if group != api.ServerGroupSingle && group != api.ServerGroupCoordinators {
 		return nil, nil
 	}
+
 	authorization := ""
 	if spec.IsAuthenticated() {
 		secretData, err := r.getJWTSecret(spec)

pkg/deployment/resources/pod_creator.go

@@ -77,10 +77,10 @@ func (config HTTPProbeConfig) Create() *v1.Probe {
 				HTTPHeaders: headers,
 			},
 		},
-		InitialDelaySeconds: def(config.InitialDelaySeconds, 30), // Wait 30s before first probe
-		TimeoutSeconds:      def(config.TimeoutSeconds, 2),       // Timeout of each probe is 2s
-		PeriodSeconds:       def(config.PeriodSeconds, 10),       // Interval between probes is 10s
-		SuccessThreshold:    def(config.SuccessThreshold, 1),     // Single probe is enough to indicate success
-		FailureThreshold:    def(config.FailureThreshold, 3),     // Need 3 failed probes to consider a failed state
+		InitialDelaySeconds: def(config.InitialDelaySeconds, 15*60), // Wait 15min before first probe
+		TimeoutSeconds:      def(config.TimeoutSeconds, 2),          // Timeout of each probe is 2s
+		PeriodSeconds:       def(config.PeriodSeconds, 60),          // Interval between probes is 10s
+		SuccessThreshold:    def(config.SuccessThreshold, 1),        // Single probe is enough to indicate success
+		FailureThreshold:    def(config.FailureThreshold, 10),       // Need 10 failed probes to consider a failed state
 	}
 }

pkg/util/k8sutil/probes.go

@@ -37,11 +37,11 @@ func TestCreate(t *testing.T) {
 	config := HTTPProbeConfig{path, false, secret, 0, 0, 0, 0, 0, 0}
 	probe := config.Create()
 
-	assert.Equal(t, probe.InitialDelaySeconds, int32(30))
+	assert.Equal(t, probe.InitialDelaySeconds, int32(15*60))
 	assert.Equal(t, probe.TimeoutSeconds, int32(2))
-	assert.Equal(t, probe.PeriodSeconds, int32(10))
+	assert.Equal(t, probe.PeriodSeconds, int32(60))
 	assert.Equal(t, probe.SuccessThreshold, int32(1))
-	assert.Equal(t, probe.FailureThreshold, int32(3))
+	assert.Equal(t, probe.FailureThreshold, int32(10))
 
 	assert.Equal(t, probe.Handler.HTTPGet.Path, path)
 	assert.Equal(t, probe.Handler.HTTPGet.HTTPHeaders[0].Name, "Authorization")