Support hostNetwork (pingcap#774)

* Support hostNetwork

* Use POD_NAME instead of HOSTNAME

* Update chart

* e2e test

* remove PodSpec

charts/tidb-cluster/templates/scripts/_start_pd.sh.tpl

@@ -30,7 +30,7 @@ fi
 
 # the general form of variable PEER_SERVICE_NAME is: "<clusterName>-pd-peer"
 cluster_name=`echo ${PEER_SERVICE_NAME} | sed 's/-pd-peer//'`
-domain="${HOSTNAME}.${PEER_SERVICE_NAME}.${NAMESPACE}.svc"
+domain="${POD_NAME}.${PEER_SERVICE_NAME}.${NAMESPACE}.svc"
 discovery_url="${cluster_name}-discovery.${NAMESPACE}.svc:10261"
 encoded_domain_url=`echo ${domain}:2380 | base64 | tr "\n" " " | sed "s/ //g"`
 
@@ -57,7 +57,7 @@ while true; do
 done
 
 ARGS="--data-dir=/var/lib/pd \
---name=${HOSTNAME} \
+--name=${POD_NAME} \
 --peer-urls=http://0.0.0.0:2380 \
 --advertise-peer-urls=http://${domain}:2380 \
 --client-urls=http://0.0.0.0:2379 \

charts/tidb-cluster/templates/scripts/_start_tikv.sh.tpl

@@ -29,7 +29,7 @@ then
 fi
 
 ARGS="--pd=${CLUSTER_NAME}-pd:2379 \
---advertise-addr=${HOSTNAME}.${HEADLESS_SERVICE_NAME}.${NAMESPACE}.svc:20160 \
+--advertise-addr=${POD_NAME}.${HEADLESS_SERVICE_NAME}.${NAMESPACE}.svc:20160 \
 --addr=0.0.0.0:20160 \
 --status-addr=0.0.0.0:20180 \
 --data-dir=/var/lib/tikv \

charts/tidb-cluster/templates/tidb-cluster.yaml

@@ -49,6 +49,7 @@ spec:
     podSecurityContext:
 {{ toYaml .Values.pd.podSecurityContext | indent 6}}
   {{- end }}
+    hostNetwork: {{ .Values.pd.hostNetwork }}
   tikv:
     replicas: {{ .Values.tikv.replicas }}
     image: {{ .Values.tikv.image }}
@@ -76,6 +77,7 @@ spec:
 {{ toYaml .Values.tikv.podSecurityContext | indent 6}}
   {{- end }}
     maxFailoverCount: {{ .Values.tikv.maxFailoverCount | default 3 }}
+    hostNetwork: {{ .Values.tikv.hostNetwork }}
   tidb:
     replicas: {{ .Values.tidb.replicas }}
     image: {{ .Values.tidb.image }}
@@ -99,6 +101,7 @@ spec:
     podSecurityContext:
 {{ toYaml .Values.tidb.podSecurityContext | indent 6}}
   {{- end }}
+    hostNetwork: {{ .Values.tidb.hostNetwork }}
     binlogEnabled: {{ .Values.binlog.pump.create | default false }}
     maxFailoverCount: {{ .Values.tidb.maxFailoverCount | default 3 }}
     separateSlowLog: {{ .Values.tidb.separateSlowLog | default false }}

charts/tidb-cluster/values.yaml

@@ -158,6 +158,10 @@ pd:
   # refer to https://kubernetes.io/docs/tasks/configure-pod-container/security-context/#set-the-security-context-for-a-pod
   podSecurityContext: {}
 
+  # Use the host's network namespace if enabled.
+  # Default to false.
+  hostNetwork: false
+
 tikv:
   # Please refer to https://github.com/tikv/tikv/blob/master/etc/config-template.toml for the default
   # tikv configurations (change to the tags of your tikv version),
@@ -242,6 +246,10 @@ tikv:
   # refer to https://kubernetes.io/docs/tasks/configure-pod-container/security-context/#set-the-security-context-for-a-pod
   podSecurityContext: {}
 
+  # Use the host's network namespace if enabled.
+  # Default to false.
+  hostNetwork: false
+
 tidb:
   # Please refer to https://github.com/pingcap/tidb/blob/master/config/config.toml.example for the default
   # tidb configurations(change to the tags of your tidb version),
@@ -303,6 +311,10 @@ tidb:
   # refer to https://kubernetes.io/docs/tasks/configure-pod-container/security-context/#set-the-security-context-for-a-pod
   podSecurityContext: {}
 
+  # Use the host's network namespace if enabled.
+  # Default to false.
+  hostNetwork: false
+
   maxFailoverCount: 3
   service:
     type: NodePort

pkg/apis/pingcap.com/v1alpha1/types.go

@@ -158,6 +158,7 @@ type PodAttributesSpec struct {
 	Tolerations        []corev1.Toleration        `json:"tolerations,omitempty"`
 	Annotations        map[string]string          `json:"annotations,omitempty"`
 	PodSecurityContext *corev1.PodSecurityContext `json:"podSecurityContext,omitempty"`
+	HostNetwork        bool                       `json:"hostNetwork,omitempty"`
 }
 
 // Service represent service type used in TidbCluster

pkg/manager/member/pd_member_manager.go

@@ -484,6 +484,11 @@ func (pmm *pdMemberManager) getNewPDSetForTidbCluster(tc *v1alpha1.TidbCluster)
 		}
 	}
 
+	dnsPolicy := corev1.DNSClusterFirst // same as k8s defaults
+	if tc.Spec.PD.HostNetwork {
+		dnsPolicy = corev1.DNSClusterFirstWithHostNet
+	}
+
 	pdSet := &apps.StatefulSet{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:            setName,
@@ -503,6 +508,8 @@ func (pmm *pdMemberManager) getNewPDSetForTidbCluster(tc *v1alpha1.TidbCluster)
 					SchedulerName: tc.Spec.SchedulerName,
 					Affinity:      tc.Spec.PD.Affinity,
 					NodeSelector:  tc.Spec.PD.NodeSelector,
+					HostNetwork:   tc.Spec.PD.HostNetwork,
+					DNSPolicy:     dnsPolicy,
 					Containers: []corev1.Container{
 						{
 							Name:            v1alpha1.PDMemberType.String(),
@@ -532,6 +539,14 @@ func (pmm *pdMemberManager) getNewPDSetForTidbCluster(tc *v1alpha1.TidbCluster)
 										},
 									},
 								},
+								{
+									Name: "POD_NAME",
+									ValueFrom: &corev1.EnvVarSource{
+										FieldRef: &corev1.ObjectFieldSelector{
+											FieldPath: "metadata.name",
+										},
+									},
+								},
 								{
 									Name:  "PEER_SERVICE_NAME",
 									Value: controller.PDPeerMemberName(tcName),

pkg/manager/member/tidb_member_manager.go

@@ -329,6 +329,11 @@ func (tmm *tidbMemberManager) getNewTiDBSetForTidbCluster(tc *v1alpha1.TidbClust
 		},
 	})
 
+	dnsPolicy := corev1.DNSClusterFirst // same as k8s defaults
+	if tc.Spec.PD.HostNetwork {
+		dnsPolicy = corev1.DNSClusterFirstWithHostNet
+	}
+
 	tidbLabel := label.New().Instance(instanceName).TiDB()
 	podAnnotations := CombineAnnotations(controller.AnnProm(10080), tc.Spec.TiDB.Annotations)
 	tidbSet := &apps.StatefulSet{
@@ -355,6 +360,7 @@ func (tmm *tidbMemberManager) getNewTiDBSetForTidbCluster(tc *v1alpha1.TidbClust
 					Tolerations:     tc.Spec.TiDB.Tolerations,
 					Volumes:         vols,
 					SecurityContext: tc.Spec.TiDB.PodSecurityContext,
+					HostNetwork:   tc.Spec.PD.HostNetwork,
 				},
 			},
 			ServiceName:         controller.TiDBPeerMemberName(tcName),

pkg/manager/member/tikv_member_manager.go

@@ -320,6 +320,11 @@ func (tkmm *tikvMemberManager) getNewSetForTidbCluster(tc *v1alpha1.TidbCluster)
 		storageClassName = controller.DefaultStorageClassName
 	}
 
+	dnsPolicy := corev1.DNSClusterFirst // same as k8s defaults
+	if tc.Spec.PD.HostNetwork {
+		dnsPolicy = corev1.DNSClusterFirstWithHostNet
+	}
+
 	tikvset := &apps.StatefulSet{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:            setName,
@@ -339,6 +344,8 @@ func (tkmm *tikvMemberManager) getNewSetForTidbCluster(tc *v1alpha1.TidbCluster)
 					SchedulerName: tc.Spec.SchedulerName,
 					Affinity:      tc.Spec.TiKV.Affinity,
 					NodeSelector:  tc.Spec.TiKV.NodeSelector,
+					HostNetwork:   tc.Spec.PD.HostNetwork,
+					DNSPolicy:     dnsPolicy,
 					Containers: []corev1.Container{
 						{
 							Name:            v1alpha1.TiKVMemberType.String(),
@@ -366,6 +373,14 @@ func (tkmm *tikvMemberManager) getNewSetForTidbCluster(tc *v1alpha1.TidbCluster)
 										},
 									},
 								},
+								{
+									Name: "POD_NAME",
+									ValueFrom: &corev1.EnvVarSource{
+										FieldRef: &corev1.ObjectFieldSelector{
+											FieldPath: "metadata.name",
+										},
+									},
+								},
 								{
 									Name:  "CLUSTER_NAME",
 									Value: tcName,

tests/cluster_info.go

@@ -33,6 +33,17 @@ func (tc *TidbClusterConfig) ScaleTiDB(replicas uint) *TidbClusterConfig {
 	return tc
 }
 
+func (tc *TidbClusterConfig) RunInHost(flag bool) *TidbClusterConfig {
+	val := "false"
+	if flag {
+		val = "true"
+	}
+	tc.set("pd.hostNetwork", val)
+	tc.set("tikv.hostNetwork", val)
+	tc.set("tidb.hostNetwork", val)
+	return tc
+}
+
 func (tc *TidbClusterConfig) UpgradePD(image string) *TidbClusterConfig {
 	tc.PDImage = image
 	return tc

tests/cmd/e2e/main.go

@@ -115,6 +115,16 @@ func main() {
 			UpdateTiDBTokenLimit(cfg.TiDBTokenLimit)
 		oa.UpgradeTidbClusterOrDie(cluster1)
 		oa.CheckTidbClusterStatusOrDie(cluster1)
+
+		// switch to host network
+		cluster1.RunInHost(true)
+		oa.UpgradeTidbClusterOrDie(cluster1)
+		oa.CheckTidbClusterStatusOrDie(cluster1)
+
+		// switch to pod network
+		cluster1.RunInHost(false)
+		oa.UpgradeTidbClusterOrDie(cluster1)
+		oa.CheckTidbClusterStatusOrDie(cluster1)
 	}
 	fn2 := func(wg *sync.WaitGroup) {
 		defer wg.Done()