TiDB benchmark on GKE

disk-benchmark-statefulset.yaml

apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: disk-benchmark
spec:
  replicas: 1
  selector:
    matchLabels:
      app: disk-benchmark
  serviceName: disk-benchmark
  template:
    metadata:
      labels:
        app: disk-benchmark
    spec:
      affinity:
        nodeAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
            nodeSelectorTerms:
            - matchExpressions:
              - key: kind
                operator: In
                values:
                - monitor

      containers:
      - name: fio
        image: pingcap/tidb-bench:latest
        command: ["tail", "-f", "/dev/null"]
        securityContext:
          privileged: true
        volumeMounts:
        - name: local
          mountPath: /local-disk
        - name: persistent
          mountPath: /persistent-disk
  volumeClaimTemplates:
  - metadata:
      name: local
    spec:
      accessModes: ["ReadWriteOnce"]
      storageClassName: local-scsi # standard | local-scsi
      resources:
        requests:
          storage: 300Gi
  - metadata:
      name: persistent
    spec:
      accessModes: ["ReadWriteOnce"]
      storageClassName: standard # standard | local-scsi
      resources:
        requests:
          storage: 300Gi

tidb-cluster-values.yaml

# Default values for tidb-cluster.
# This is a YAML-formatted file.
# Declare variables to be passed into your templates.

rbac:
  create: true

# clusterName is the TiDB cluster name, it is required and should be unique
# if multiple clusters are deployed in the same namespace.
clusterName: local-disk-cluster

# localtime determines whether containers timezone should be the same as k8s node
# if localtime is false, container timezone would be UTC+0
# localtime should be set to false if the K8s environment doesn't have /etc/localtime
# on host, e.g. Docker for Mac, Minikube, DinD
localtime: false

# nodeSelectorRequired: true

# default reclaim policy of a PV
pvReclaimPolicy: Retain

# services is the service list to expose, default is ClusterIP
# can be ClusterIP | NodePort | LoadBalancer
services:
  - name: pd
    type: ClusterIP
  - name: tidb
    type: NodePort

pd:
  replicas: 3
  image: pingcap/tidb:v2.0.4
  logLevel: info
  # storageClassName is a StorageClass provides a way for administrators to describe the "classes" of storage they offer.
  # different classes might map to quality-of-service levels, or to backup policies,
  # or to arbitrary policies determined by the cluster administrators.
  # refer to https://kubernetes.io/docs/concepts/storage/storage-classes
  storageClassName: local-scsi

  # maxStoreDownTime is how long a store will be considered `down` when disconnected
  # if a store is considered `down`, the regions will be migrated to other stores
  maxStoreDownTime: 1h
  # maxReplicas is the number of replicas for each region
  maxReplicas: 3
  resources:
    limits:
      cpu: 8000m
      memory: 16Gi
    requests:
      cpu: 8000m
      memory: 16Gi
      storage: 10Gi
  # 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

tikv:
  replicas: 3
  image: pingcap/tidb:v2.0.4
  logLevel: info
  # storageClassName is a StorageClass provides a way for administrators to describe the "classes" of storage they offer.
  # different classes might map to quality-of-service levels, or to backup policies,
  # or to arbitrary policies determined by the cluster administrators.
  # refer to https://kubernetes.io/docs/concepts/storage/storage-classes
  storageClassName: local-scsi

  # syncLog is a bool value to enable or disable syc-log for raftstore, default is true
  # enable this can prevent data loss when power failure
  syncLog: true
  resources:
    limits:
      cpu: 16000m
      memory: 32Gi
      storage: 300Gi
    requests:
      cpu: 16000m
      memory: 32Gi
      storage: 300Gi
  nodeSelector:
    kind: tikv
    # zone: cn-bj1-01,cn-bj1-02
    # region: cn-bj1
tikvPromGateway:
  image: prom/pushgateway:v0.3.1
  # limits:
  #   cpu: 100m
  #   memory: 100Mi
  # requests:
  #   cpu: 50m
  #   memory: 50Mi

tidb:
  replicas: 2
  image: pingcap/tidb:v2.0.4
  logLevel: info
  resources:
    limits:
      cpu: 16000m
      memory: 32Gi
    requests:
      cpu: 16000m
      memory: 32Gi
  nodeSelector:
    kind: tidb
    # zone: cn-bj1-01,cn-bj1-02
    # region: cn-bj1

monitor:
  create: true
  dashboardInstaller:
    image: pingcap/tidb-dashboard-installer:v2.0.0
  grafana:
    image: grafana/grafana:4.6.3
    resources: {}
      # limits:
      #   cpu: 8000m
      #   memory: 8Gi
      # requests:
      #   cpu: 4000m
      #   memory: 4Gi
    username: admin
    password: admin
    service:
      type: LoadBalancer
    grafanaUrl: http://localhost:3000
  prometheus:
    image: prom/prometheus:v2.2.1
    resources: {}
      # limits:
      #   cpu: 8000m
      #   memory: 8Gi
      # requests:
      #   cpu: 4000m
      #   memory: 4Gi
    service:
      type: LoadBalancer
    reserveDays: 12
    # alertmanagerURL: ""
  nodeSelector:
    kind: monitor
    # zone: cn-bj1-01,cn-bj1-02
    # region: cn-bj1

privilegedTidb:
  # Whether to deploy privileged tidb
  create: false
  replicas: 1
  image: pingcap/tidb:v2.0.4
  resources: {}
  #  limits:
  #    cpu: 16000m
  #    memory: 16Gi
  #  requests:
  #    cpu: 12000m
  #    memory: 12Gi
  nodeSelector: {}
  #  kind: privileged-tidb
  #  zone: cn-bj1-01,cn-bj1-02
  #  region: cn-bj1
  service:
    # type is privileged tidb service type, can be ClusterIP, NodePort, LoadBalancer
    type: ClusterIP
    # nodePort is optional and only used when type is NodePort
    # nodePort: 33333

metaInstance: "{{ $labels.instance }}"
metaType: "{{ $labels.type }}"
metaValue: "{{ $value }}"