support the affinity feature of k8s which define the rule of assigning pods to nodes

charts/tidb-cluster/templates/config/_pd-config.tpl

@@ -82,7 +82,7 @@ max-replicas = {{ .Values.pd.maxReplicas }}
 # The placement priorities is implied by the order of label keys.
 # For example, ["zone", "rack"] means that we should place replicas to
 # different zones first, then to different racks if we don't have enough zones.
-location-labels = ["zone", "rack", "host"]
+location-labels = ["region", "zone", "rack", "host"]
 
 [label-property]
 # Do not assign region leaders to stores that have these tags.

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

@@ -27,11 +27,8 @@ spec:
   {{- if .Values.pd.resources }}
 {{ toYaml .Values.pd.resources | indent 4 }}
   {{- end }}
-  {{- if .Values.pd.nodeSelector }}
-    nodeSelector:
-{{ toYaml .Values.pd.nodeSelector | indent 6 }}
-  {{- end }}
-    nodeSelectorRequired: {{ .Values.nodeSelectorRequired | default true }}
+    affinity:
+{{ toYaml .Values.pd.affinity | indent 6 }}
   {{- if .Values.pd.tolerations }}
     tolerations:
 {{ toYaml .Values.pd.tolerations | indent 4 }}
@@ -46,11 +43,8 @@ spec:
   {{- if .Values.tikv.resources }}
 {{ toYaml .Values.tikv.resources | indent 4 }}
   {{- end }}
-  {{- if .Values.tikv.nodeSelector }}
-    nodeSelector:
-{{ toYaml .Values.tikv.nodeSelector | indent 6 }}
-  {{- end }}
-    nodeSelectorRequired: {{ .Values.nodeSelectorRequired | default true }}
+    affinity:
+{{ toYaml .Values.tikv.affinity | indent 6 }}
   {{- if .Values.tikv.tolerations }}
     tolerations:
 {{ toYaml .Values.tikv.tolerations | indent 4 }}
@@ -62,11 +56,8 @@ spec:
   {{- if .Values.tidb.resources }}
 {{ toYaml .Values.tidb.resources | indent 4 }}
   {{- end }}
-  {{- if .Values.tidb.nodeSelector }}
-    nodeSelector:
-{{ toYaml .Values.tidb.nodeSelector | indent 6 }}
-  {{- end }}
-    nodeSelectorRequired: {{ .Values.nodeSelectorRequired | default true }}
+    affinity:
+{{ toYaml .Values.tidb.affinity | indent 6 }}
   {{- if .Values.tidb.tolerations }}
     tolerations:
 {{ toYaml .Values.tidb.tolerations | indent 4 }}

charts/tidb-cluster/values.yaml

@@ -66,16 +66,68 @@ pd:
       # cpu: 4000m
       # memory: 4Gi
       storage: 1Gi
-  # nodeSelector is used for scheduling pod,
-  # if nodeSelectorRequired is true, all the following labels must be matched
-  nodeSelector: {}
-    # kind: pd
-    # # zone is comma separated availability zone list
-    # zone: cn-bj1-01,cn-bj1-02
-    # # region is comma separated region list
-    # region: cn-bj1
-  # Tolerations are applied to pods, and allow pods to schedule onto nodes with matching taints.
-  # refer to https://kubernetes.io/docs/concepts/configuration/taint-and-toleration
+
+  ## affinity defines pd scheduling rules,it's default settings is empty.
+  ## please read the affinity document before set your scheduling rule:
+  ## ref: https://kubernetes.io/docs/concepts/configuration/assign-pod-node/#affinity-and-anti-affinity
+  affinity: {}
+  ## The following is typical example of affinity settings:
+  ## The PodAntiAffinity setting of the example keeps PD pods does not co-locate on a topology node as far as possible to improve the disaster tolerance of PD on Kubernetes.
+  ## The NodeAffinity setting of the example ensure that the PD pods can only be scheduled to nodes with label:[type="pd"],
+  # affinity:
+  #   podAntiAffinity:
+  #     preferredDuringSchedulingIgnoredDuringExecution:
+  #     # this term work when the nodes have the label named region
+  #     - weight: 10
+  #       podAffinityTerm:
+  #         labelSelector:
+  #           matchLabels:
+  #             app.kubernetes.io/instance: <release name>
+  #             app.kubernetes.io/component: "pd"
+  #         topologyKey: "region"
+  #         namespaces:
+  #         - <helm namespace>
+  #     # this term work when the nodes have the label named zone
+  #     - weight: 20
+  #       podAffinityTerm:
+  #         labelSelector:
+  #           matchLabels:
+  #             app.kubernetes.io/instance: <release name>
+  #             app.kubernetes.io/component: "pd"
+  #         topologyKey: "zone"
+  #         namespaces:
+  #         - <helm namespace>
+  #     # this term work when the nodes have the label named rack
+  #     - weight: 40
+  #       podAffinityTerm:
+  #         labelSelector:
+  #           matchLabels:
+  #             app.kubernetes.io/instance: <release name>
+  #             app.kubernetes.io/component: "pd"
+  #         topologyKey: "rack"
+  #         namespaces:
+  #         - <helm namespace>
+  #     # this term work when the nodes have the label named kubernetes.io/hostname
+  #     - weight: 80
+  #       podAffinityTerm:
+  #         labelSelector:
+  #           matchLabels:
+  #             app.kubernetes.io/instance: <release name>
+  #             app.kubernetes.io/component: "pd"
+  #         topologyKey: "kubernetes.io/hostname"
+  #         namespaces:
+  #         - <helm namespace>
+  #   nodeAffinity:
+  #     requiredDuringSchedulingIgnoredDuringExecution:
+  #       nodeSelectorTerms:
+  #       - matchExpressions:
+  #         - key: "kind"
+  #           operator: In
+  #           values:
+  #           - "pd"
+
+  ## Tolerations are applied to pods, and allow pods to schedule onto nodes with matching taints.
+  ## refer to https://kubernetes.io/docs/concepts/configuration/taint-and-toleration
   tolerations: []
   # - key: node-role
   #   operator: Equal
@@ -109,10 +161,14 @@ tikv:
       # cpu: 12000m
       # memory: 24Gi
       storage: 10Gi
-  nodeSelector: {}
-    # kind: tikv
-    # zone: cn-bj1-01,cn-bj1-02
-    # region: cn-bj1
+
+  ## affinity defines tikv scheduling rules,affinity default settings is empty.
+  ## please read the affinity document before set your scheduling rule:
+  ## ref: https://kubernetes.io/docs/concepts/configuration/assign-pod-node/#affinity-and-anti-affinity
+  affinity: {}
+
+  ## Tolerations are applied to pods, and allow pods to schedule onto nodes with matching taints.
+  ## refer to https://kubernetes.io/docs/concepts/configuration/taint-and-toleration
   tolerations: []
   # - key: node-role
   #   operator: Equal
@@ -187,10 +243,15 @@ tidb:
     requests: {}
     #   cpu: 12000m
     #   memory: 12Gi
-  nodeSelector: {}
-    # kind: tidb
-    # zone: cn-bj1-01,cn-bj1-02
-    # region: cn-bj1
+
+
+  ## affinity defines tikv scheduling rules,affinity default settings is empty.
+  ## please read the affinity document before set your scheduling rule:
+  ## ref: https://kubernetes.io/docs/concepts/configuration/assign-pod-node/#affinity-and-anti-affinity
+  affinity: {}
+
+  ## Tolerations are applied to pods, and allow pods to schedule onto nodes with matching taints.
+  ## refer to https://kubernetes.io/docs/concepts/configuration/taint-and-toleration
   tolerations: []
   # - key: node-role
   #   operator: Equal

docs/operation-guide.md

@@ -11,21 +11,68 @@ $ namespace="tidb"
 
 > **Note:** The rest of the document will use `values.yaml` to reference `charts/tidb-cluster/values.yaml`
 
+## Configuration
+
+TiDB Operator use `values.yaml` as TiDB cluster configuration file. It provides the default basic configuration which you can use directly for quick deployment, but if you have specific configuration requirements or for production deployment, you need to manually modify the variables in the `value.yaml` file.
+
+* Resource setting
+
+    * CPU & Memory
+
+        The default deployment doesn't set CPU and memory requests or limits for any of the pods, these settings can make TiDB cluster run on a small Kubernetes cluster like DinD or the default GKE cluster for testing. But for production deployment, you would likely to adjust the cpu, memory and storage resources according to the [recommendations](https://github.com/pingcap/docs/blob/master/op-guide/recommendation.md).
+        The resource limits should be equal or bigger than the resource requests, it is suggested to set limit and request equal to get [`Guaranteed` QoS]( https://kubernetes.io/docs/tasks/configure-pod-container/quality-service-pod/#create-a-pod-that-gets-assigned-a-qos-class-of-guaranteed).
+
+    * Storage
+
+        The variables `pd.storageClassName` and `tikv.storageClassName` in `values.yaml` are used to set `StorageClass` of pd and tikv,their default setting are `local-storage` with minimal size.
+        If you don't want to use the default `StorageClass` or your Kubernetes cluster does not support `local-storage` class, please execute the following command to find an available `StorageClass` and select the ones you want to provide to TiDB cluster.
+
+        ```shell
+        $ kubectl get sc
+        ```
+
+* Disaster Tolerance setting
+
+    TiDB cluster is a distributed database. Its Disaster Tolerance means that when any physical node failed, not only to ensure TiDB server is available, but also ensure the data is complete and available.
+
+    How to guarantee Disaster Tolerance of TiDB cluster on Kubernetes?
+
+    We mainly solve the problem from the scheduling of services and data.
+
+    * Disaster Tolerance of TiDB server
+
+        TiDB Operator provides a external scheduler to guarantee PD/TiKV/TiDB server disaster tolerance on host level. TiDB Cluster have set the external scheduler as default scheduler, you will find the setting in the variable `schedulerName` of `values.yaml`.
+
+        In the other hand use `PodAntiAffinity` term of `affinity` to ensure disaster tolerance on the other topology levels (e.g. rack, zone, region). 
+        refer to the doc: [pod affnity & anti affinity](https://kubernetes.io/docs/concepts/configuration/assign-pod-node/#inter-pod-affinity-and-anti-affinity-beta-feature), moreover `values.yaml` also provides a typical disaster tolerance setting example in the comments of `pd.affinity`.
+
+    * Disaster Tolerance of data
+
+        Disaster tolerance of data is guaranteed by TiDB Cluster itself. The only work Operator needs to do is that collects topology info from specific labels of Kubernetes nodes where TiKV Pod runs on and then PD will schedule data replicas auto according to the topology info.
+        Cause currently TiDB Operator can only recognize some specific labels, so you can only set nodes topology info with the following particular labels
+
+        * `region`: region where node is located
+        * `zone`: zone where node is located
+        * `rack`: rack where node is located
+        * `kubernetes.io/hostname`: hostname of the node
+
+        you need label topology info to nodes of Kubernetes cluster use the following command
+        ```shell
+        # The labels are optional
+        $ kubectl label node <nodeName> region=<regionName> zone=<zoneName> rack=<rackName> kubernetes.io/hostname=<hostName>
+        ```
+
+For other settings, the variables in `values.yaml` are self-explanatory with comments. You can modify them according to your need before installing the charts.
+
 ## Deploy TiDB cluster
 
-After TiDB Operator and Helm are deployed correctly, TiDB cluster can be deployed using following command:
+After TiDB Operator and Helm are deployed correctly and configuration completed, TiDB cluster can be deployed using following command:
 
 ```shell
 $ helm install charts/tidb-cluster --name=${releaseName} --namespace=${namespace}
 $ kubectl get po -n ${namespace} -l app.kubernetes.io/instance=${releaseName}
 ```
 
-The default deployment doesn't set CPU and memory requests or limits for any of the pods, and the storage used is `local-storage` with minimal size. These settings can make TiDB cluster run on a small Kubernetes cluster like DinD or the default GKE cluster for testing. But for production deployment, you would likely to adjust the cpu, memory and storage resources according to the [recommendations](https://github.com/pingcap/docs/blob/master/op-guide/recommendation.md).
-
-The resource limits should be equal or bigger than the resource requests, it is suggested to set limit and request equal to get [`Guaranteed` QoS]( https://kubernetes.io/docs/tasks/configure-pod-container/quality-service-pod/#create-a-pod-that-gets-assigned-a-qos-class-of-guaranteed).
-
-For other settings, the variables in `values.yaml` are self-explanatory with comments. You can modify them according to your need before installing the charts.
-
 ## Access TiDB cluster
 
 By default TiDB service is exposed using [`NodePort`](https://kubernetes.io/docs/concepts/services-networking/service/#nodeport). You can modify it to `ClusterIP` which will disable access from outside of the cluster. Or modify it to [`LoadBalancer`](https://kubernetes.io/docs/concepts/services-networking/service/#loadbalancer) if the underlining Kubernetes supports this kind of service.

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

@@ -15,10 +15,9 @@ package v1alpha1
 
 import (
 	apps "k8s.io/api/apps/v1beta1"
+	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
-
-	corev1 "k8s.io/api/core/v1"
 )
 
 const (
@@ -107,25 +106,23 @@ type TidbClusterStatus struct {
 // PDSpec contains details of PD member
 type PDSpec struct {
 	ContainerSpec
-	Replicas             int32               `json:"replicas"`
-	NodeSelector         map[string]string   `json:"nodeSelector,omitempty"`
-	NodeSelectorRequired bool                `json:"nodeSelectorRequired,omitempty"`
-	StorageClassName     string              `json:"storageClassName,omitempty"`
-	Tolerations          []corev1.Toleration `json:"tolerations,omitempty"`
+	Replicas         int32               `json:"replicas"`
+	Affinity         *corev1.Affinity    `json:"affinity,omitempty"`
+	StorageClassName string              `json:"storageClassName,omitempty"`
+	Tolerations      []corev1.Toleration `json:"tolerations,omitempty"`
 }
 
 // TiDBSpec contains details of PD member
 type TiDBSpec struct {
 	ContainerSpec
-	Replicas             int32                 `json:"replicas"`
-	NodeSelector         map[string]string     `json:"nodeSelector,omitempty"`
-	NodeSelectorRequired bool                  `json:"nodeSelectorRequired,omitempty"`
-	StorageClassName     string                `json:"storageClassName,omitempty"`
-	Tolerations          []corev1.Toleration   `json:"tolerations,omitempty"`
-	BinlogEnabled        bool                  `json:"binlogEnabled,omitempty"`
-	MaxFailoverCount     int32                 `json:"maxFailoverCount,omitempty"`
-	SeparateSlowLog      bool                  `json:"separateSlowLog,omitempty"`
-	SlowLogTailer        TiDBSlowLogTailerSpec `json:"slowLogTailer,omitempty"`
+	Replicas         int32                 `json:"replicas"`
+	Affinity         *corev1.Affinity      `json:"affinity,omitempty"`
+	StorageClassName string                `json:"storageClassName,omitempty"`
+	Tolerations      []corev1.Toleration   `json:"tolerations,omitempty"`
+	BinlogEnabled    bool                  `json:"binlogEnabled,omitempty"`
+	MaxFailoverCount int32                 `json:"maxFailoverCount,omitempty"`
+	SeparateSlowLog  bool                  `json:"separateSlowLog,omitempty"`
+	SlowLogTailer    TiDBSlowLogTailerSpec `json:"slowLogTailer,omitempty"`
 }
 
 // TiDBSlowLogTailerSpec represents an optional log tailer sidecar with TiDB
@@ -136,11 +133,10 @@ type TiDBSlowLogTailerSpec struct {
 // TiKVSpec contains details of PD member
 type TiKVSpec struct {
 	ContainerSpec
-	Replicas             int32               `json:"replicas"`
-	NodeSelector         map[string]string   `json:"nodeSelector,omitempty"`
-	NodeSelectorRequired bool                `json:"nodeSelectorRequired,omitempty"`
-	StorageClassName     string              `json:"storageClassName,omitempty"`
-	Tolerations          []corev1.Toleration `json:"tolerations,omitempty"`
+	Replicas         int32               `json:"replicas"`
+	Affinity         *corev1.Affinity    `json:"affinity,omitempty"`
+	StorageClassName string              `json:"storageClassName,omitempty"`
+	Tolerations      []corev1.Toleration `json:"tolerations,omitempty"`
 }
 
 // TiKVPromGatewaySpec runs as a sidecar with TiKVSpec

pkg/manager/member/pd_member_manager.go

@@ -476,12 +476,7 @@ func (pmm *pdMemberManager) getNewPDSetForTidbCluster(tc *v1alpha1.TidbCluster)
 				},
 				Spec: corev1.PodSpec{
 					SchedulerName: tc.Spec.SchedulerName,
-					Affinity: util.AffinityForNodeSelector(
-						ns,
-						tc.Spec.PD.NodeSelectorRequired,
-						label.New().Instance(instanceName).PD(),
-						tc.Spec.PD.NodeSelector,
-					),
+					Affinity:      tc.Spec.PD.Affinity,
 					Containers: []corev1.Container{
 						{
 							Name:            v1alpha1.PDMemberType.String(),

pkg/manager/member/tidb_member_manager.go

@@ -346,12 +346,7 @@ func (tmm *tidbMemberManager) getNewTiDBSetForTidbCluster(tc *v1alpha1.TidbClust
 				},
 				Spec: corev1.PodSpec{
 					SchedulerName: tc.Spec.SchedulerName,
-					Affinity: util.AffinityForNodeSelector(
-						ns,
-						tc.Spec.TiDB.NodeSelectorRequired,
-						label.New().Instance(instanceName).TiDB(),
-						tc.Spec.TiDB.NodeSelector,
-					),
+					Affinity:      tc.Spec.TiDB.Affinity,
 					Containers:    containers,
 					RestartPolicy: corev1.RestartPolicyAlways,
 					Tolerations:   tc.Spec.TiDB.Tolerations,