Skip to content
This repository has been archived by the owner on Nov 10, 2023. It is now read-only.

Latest commit

 

History

History
492 lines (413 loc) · 17.4 KB

milvus.md

File metadata and controls

492 lines (413 loc) · 17.4 KB
Raw

Custom Resource Definition

This document guides user to learn the related fields defined in the Milvus CRD and then customize their Milvus cluster deployment stack.

Before you start, you should have a basic understanding of the Custom Resource (CR) in Kubernetes. If not, please refer to Kubernetes CRD doc.

CRD version: v1beta1

CRD spec

Describe the spec fields with YAML code snippets and comments. All the parts here share the head YAML code snippet shown below.

apiVersion: milvus.io/v1beta1
kind: Milvus
metadata:
  name: milvus-sample
  namespace: sample-ns
spec:
  mode: standalone # Optional ("standalone", "cluster") default="standalone"
  components: {} # Optional
  dependencies: {} # Optional
  config: {} # Optional

Components

Top field spec components(optional) includes components' global spec for all 9 types of components (including 8 components in cluster mode and the standalone component in standalone mode), and private spec for each component.

spec:
  components:
    # Components specifications

    # Components global specifications
    # ... Skipped fields

    # Components private specifications
    # ... Skipped fields

Components Global Spec

The global configuration's for all 9 types of components which can be override by their private spec. It contains fields:

  • image proceeding fields sets default configurations about the image milvus cluster should use and how to pull it.
  • env includes custom environment variables.
  • nodeSelector & tolerations controll which k8s nodes the milvus work loads should be scheduled to.
  • resources: compute resources required by each component

Components global configurations example:

spec:
  components:
    # Components specifications

    # Components global specifications

    # Enable rolling update, supported in milvus v2.2.3.
    # For compatity reason defaults to false, but we suggest you to enable it if you are using milvus v2.2.3 or above.
    enableRollingUpdate: true # Optional default=false

    # imageUpdateMode is the mode when update components' image.
    # rollingUpgrade: to update the components' image in the order of coords -> nodes -> proxy
    # rollingDowngrade: to update the components' image in the order of proxy -> nodes -> coords
    # all: to update all the components' image rightaway.
    # one of rollingUpgrade / rollingDowngrade / all
    imageUpdateMode: rollingUpgrade # Optional default=rollingUpgrade

    # Paused is used to pause all components' deployment rollout
    paused: false # Optional

    # Global pod labels.
    podLabels: # Optional
      key: value
    
    # Global pod annotations.
    podAnnotations: # Optional
      key: value

    # Global image name for milvus components. It will override the default one. Default is determined by operator version
    image: milvusdb/milvus:latest # Optional

    # Global image pull policy. It will override the the default one.
    imagePullPolicy: IfNotPresent # Optional, default = IfNotPresent

    # Global image pull secrets.
    imagePullSecrets: # Optional
    - name: mySecret
    # Global environment variables
    env: # Optional
    - name: key
      value: value

    # Global nodeSelector.
    # NodeSelector is a selector which must be true for the component to fit on a node.
    # Selector which must match a node's labels for the pod to be scheduled on that node.
    # More info: https://kubernetes.io/docs/concepts/configuration/assign-pod-node/
    nodeSelector: # Optional
      key: value

    # Global tolerations.
    # If specified, the pod's tolerations.
    # More info: https://kubernetes.io/docs/concepts/scheduling-eviction/taint-and-toleration/
    tolerations: {} # Optional
    
    # Global compute resources required.
    # Compute Resources required by this component.
    # Cannot be updated.
    # More info: https://kubernetes.io/docs/concepts/configuration/manage-compute-resources-container/
    resources: # Optional
      # Limits describes the maximum amount of compute resources allowed.
      # More info: https://kubernetes.io/docs/concepts/configuration/manage-compute-resources-container/
      limits: {} # Optional
      # Requests describes the minimum amount of compute resources required.
      # If Requests is omitted for a container, it defaults to Limits if that is explicitly specified,
      # otherwise to an implementation-defined value.
      # More info: https://kubernetes.io/docs/concepts/configuration/manage-compute-resources-container/
      requests: {} # Optional

    # Global volumes for all components 
    # More info: https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.21/#volume-v1-core
    volumes: [] # Optional
    
    # Global volumeMounts.
    # More info: https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.21/#volumemount-v1-core
    VolumeMounts: [] # Optional

    # Global serviceAccountName.
    serviceAccountName: "" # Optional

    # Disable metrics collection for all components
    disableMetrics: false # Optional

    # The interval of podmonitor metric scraping in string
    metricInterval : "30s" # Optional

    # ToolImage specify tool image to merge milvus config to original one in image, default uses same image as milvus-operator
    toolImage: "" # Optional

    # UpdateToolImage specifies when milvus-operator upgraded, whether milvus should restart to update the tool image, too
    updateToolImage: false # Optional

    # Components private specifications
    # ... Skipped fields

Components Private Spec

Configurations for each component. There are 9 types of components as in the code below:

spec:
  # ... Skipped fields
  
  components:
    # ... Skipped fields

    # cluster components:
    proxy: {} # Optional
    rootCoord: {} # Optional
    indexCoord: {} # Optional
    dataCoord: {} # Optional
    queryCoord: {} # Optional
    indexNode: {} # Optional
    dataNode: {} # Optional
    queryNode: {} # Optional

    # MixCoord is a mixture of all coordinators(rootCoord, indexCoord, dataCoord and queryCoord), running within a single pod & single process. Since the coordinators won't cost much resources, it's recommended to use mixCoord instead of the 4 coordinators.
    mixCoord: {} # Optional

    # standalone component
    standalone: {} # Optional

Each component has its own basic specifications that can overrides global ones:

  • replica: number of replicas
  • port: the port number that server will listen
  • fields same as section Components Global Spec has stated above (including image fields, env, nodeSelector,tolerations, resources)

Take rootCoord as example:

spec:
  components:
    # Global Component Spec fields
    # ... Skipped fields

    rootCoord: # Optional
      # Supply number of replicas.
      replicas: 1 # Optional, default=1

      # Port number the conponent's server will listen
      port: 8080 # Optional

      # Private Component Spec fields overrides the global ones
      image: milvusdb/milvus:latest # Optional
      imagePullPolicy: IfNotPresent # Optional
      imagePullSecrets: # Optional
      - name: mySecret
      env: # Optional
      - name: key
        value: value
      nodeSelector: # Optional
      - key: value
      tolerations: {} # Optional
      resources: {} # Optional
        requests: {} # Optional
        limits: {} # Optional

    # ... Skipped fields
  # ... Skipped fields

The proxy & standalone component is special. These 2 spec have not only all basic fields as other components, but also specifications about its serviceType

proxy component:

spec:
  components:
    # Global Component Spec fields
    # ... Skipped fields
      
    proxy: # Optional
      serviceType: ClusterIP # Optional ("ClusterIP", "NodePort", "LoadBalancer")
      # ... Skipped fields

    # ... Skipped fields
  # ... Skipped fields

standalone component:

spec:
  components:
    # Global Component Spec fields
    # ... Skipped fields
      
    standalone: # Optional
      serviceType: ClusterIP # Optional ("ClusterIP", "NodePort", "LoadBalancer")
      # ... Skipped fields

    # ... Skipped fields
  # ... Skipped fields

Config

Config overrides the fields of Milvus Cluster's config file template.

For example, if you want to change etcd's rootPath and minIO's bucketname:

spec:
  dependencies: {}
  components: {}
  config: # Optional
    etcd:
      rootPath: my-release
    minio:
      bucketName: my-bucket

A complete set of config fields can be found at https://github.com/milvus-io/milvus/blob/master/configs/milvus.yaml

NOTE! The fields of dependencies' address and port cannot be set in the Milvus Cluster CR.

Dependencies

specifications for milvus's dependencies:

spec:
  # ... Skipped fields
  dependencies: # Optional
    etcd: {} # Optional
    storage: {} # Optional
    
    # Optional. msgStreamType determines which message stream to use. It should be one of "pulsar", "kafka", "rocksmq"
    # "rocksmq" is only available for standalone mode
    # by default, the operator will choose "pulsar" for cluster mode and "rocksmq" for standalone mode
    msgStreamType: "kafka" 
    pulsar: {} # Optional
    kafka: {} # Optional

Dependency ETCD

The dependency etcd may be specified as external or in-cluster:

spec:
  # ... Skipped fields
  dependencies: # Optional
    etcd: # Optional
      # Whether (=true) to use an existed external etcd as specified in the field endpoints or 
      # (=false) create a new etcd inside the same kubernetes cluster for milvus.
      external: false # Optional default=false
      # The external etcd endpoints if external=true
      endpoints:
      - 192.168.1.1:2379
      # in-Cluster etcd configuration if external=false
      inCluster: 
        # deletionPolicy of etcd when the milvus cluster is deleted
        deletionPolicy: Retain # Optional ("Delete", "Retain") default="Retain"
        # When deletionPolicy="Delete" whether the PersistantVolumeClaim shoud be deleted when the etcd is deleted
        pvcDeletion: false # Optional default=false
        # ... Skipped fields
    # ... Skipped fields

The inCluster.values field contains etcd's configurable helm values. For example if you want to deploy etcd in its minimun mode:

spec:
  # ... Skipped fields
  dependencies: # Optional
    etcd: # Optional
      # ... Skipped fields
      inCluster:
        # ... Skipped fields
        values: # Optional
          replicaCount: 1

A complete fields doc can be found at https://artifacthub.io/packages/helm/bitnami/etcd/6.3.3.

Dependency Storage

The dependency storage may be specified as external or in-cluster. When use in-cluster storage, only MinIO storage type is supported.

spec:
  # ... Skipped fields
  dependencies: # Optional
    storage: # Optional
      # Whether (=true) to use an existed external storage as specified in the field endpoints or 
      # (=false) create a new storage inside the same kubernetes cluster for milvus.
      external: false # Optional default=false
      type: "MinIO" # Optional ("MinIO", "S3") default:="MinIO"
      # Secret reference of the storage if it has
      secretRef: mySecret # Optional
      # The external storage endpoint if external=true
      endpoint: "storageEndpoint"
      # in-Cluster storage configuration if external=false
      inCluster: 
        # deletionPolicy of storage when the milvus cluster is deleted
        deletionPolicy: Retain # Optional ("Delete", "Retain") default="Retain"
        # When deletionPolicy="Delete" whether the PersistantVolumeClaim shoud be deleted when the storage is deleted
        pvcDeletion: false # Optional default=false
        # ... Skipped fields
    # ... Skipped fields

The inCluster.values field contains minIO's configurable helm values. For example if you want to deploy minIO in its minimun mode:

spec:
  # ... Skipped fields
  dependencies: # Optional
    storage: # Optional
      # ... Skipped fields
      inCluster:
        # ... Skipped fields
        values: # Optional
          mode: standalone

A complete fields doc can be found at https://github.com/milvus-io/milvus-helm/blob/master/charts/minio/values.yaml.

Dependency Pulsar

The dependency pulsar may be specified as external or in-cluster:

spec:
  # ... Skipped fields
  dependencies: # Optional
    pulsar: # Optional
      # Whether (=true) to use an existed external pulsar as specified in the field endpoints or 
      # (=false) create a new pulsar inside the same kubernetes cluster for milvus.
      external: false # Optional default=false
      # The external pulsar endpoints if external=true
      endpoints:
      - 192.168.1.1:6650
      # in-Cluster pulsar configuration if external=false
      inCluster: 
        # deletionPolicy of pulsar when the milvus cluster is deleted
        deletionPolicy: Retain # Optional ("Delete", "Retain") default="Retain"
        # When deletionPolicy="Delete" whether the PersistantVolumeClaim shoud be deleted when the pulsar is deleted
        pvcDeletion: false # Optional default=false
        # ... Skipped fields
    # ... Skipped fields

The inCluster.values field contains pulsar's configurable helm values. For example if you want to deploy pulsar in its minimun mode:

spec:
  # ... Skipped fields
  dependencies: # Optional
    pulsar: # Optional
      # ... Skipped fields
      inCluster:
        # ... Skipped fields
        values:
          components:
            autorecovery: false
          zookeeper:
            replicaCount: 1
          bookkeeper:
            replicaCount: 1
          broker:
            replicaCount: 1
            configData:
              ## Enable `autoSkipNonRecoverableData` since bookkeeper is running
              ## without persistence
              autoSkipNonRecoverableData: "true"
              managedLedgerDefaultEnsembleSize: "1"
              managedLedgerDefaultWriteQuorum: "1"
              managedLedgerDefaultAckQuorum: "1"
          proxy:
            replicaCount: 1

A complete fields doc can be found at https://github.com/kafkaesque-io/pulsar-helm-chart/blob/pulsar-1.0.31/helm-chart-sources/pulsar/values.yaml. And some of its default values are overrided by fields under pulsar: in https://github.com/milvus-io/milvus-helm/blob/master/charts/milvus/values.yaml

Dependency kafka

The dependency kafka may be specified as external or in-cluster:

spec:
  # ... Skipped fields
  dependencies: # Optional
    kafka: # Optional
      # Whether (=true) to use an existed external kafka as specified in the field endpoints or 
      # (=false) create a new pulsar inside the same kubernetes cluster for milvus.
      external: false # Optional default=false
      # The external kafka brokers if external=true
      brokers:
      - 192.168.1.1:9092
      # in-Cluster pulsar configuration if external=false
      inCluster: 
        # deletionPolicy of pulsar when the milvus cluster is deleted
        deletionPolicy: Retain # Optional ("Delete", "Retain") default="Retain"
        # When deletionPolicy="Delete" whether the PersistantVolumeClaim shoud be deleted when the kafka is deleted
        pvcDeletion: false # Optional default=false
        # ... Skipped fields
    # ... Skipped fields

The inCluster.values field contains kafka's configurable helm values. For example if you want to deploy pulsar in its minimun mode:

spec:
  # ... Skipped fields
  dependencies: # Optional
    kafka: # Optional
      # ... Skipped fields
      inCluster:
        # ... Skipped fields
        values:
          defaultReplicationFactor: 1
          offsetsTopicReplicationFactor: 1
          replicaCount: 1
          zookeeper:
            replicaCount: 1

A complete fields doc can be found at https://github.com/bitnami/charts/blob/1fdd2283f0e5a8772e4a763b455733c77e01b119/bitnami/kafka/values.yaml. And some of its default values are overrided by fields under kafka: in https://github.com/milvus-io/milvus-helm/blob/master/charts/milvus/values.yaml

Status

The status of the CR Milvus is described as below:

status:
  # Show the generous status of the Milvus
  # It can be "Pending", "Healthy", "Unhealthy", "Stopped"
  # Healthy means all milvus components are ready
  # Unhealthy means at least one milvus component is not ready
  # Stopped means all milvus components are stopped
  # Pending means it's being created or resumed from stopped status.
  status: "Healthy"
  # Contains details for the current condition of Milvus and its dependency
  conditions: 
    # Condition type
    # It can be "EtcdReady", "StorageReady", "MsgStream", "MilvusReady", "MilvusUpdated"
  - type: "MilvusReady" 
    # Status is the status of the condition.
    # Can be True, False, Unknown.
    status: True
    # Last time the condition transitioned from one status to another.
    lastTransitionTime: <time> # Optional
     # Unique, one-word, CamelCase reason for the condition's last transition.
    reason: "reason" # Optional
    # Human-readable message indicating details about last transition.
    message: "message" # Optional
  # The Milvus's endpoint of service
  endpoint: "milvus:19530"
  # ComponentsDeployStatus contains the map of component's name to the status of each component deployment
  componentsDeployStatus: {}
  # When observedGeneration is smaller than spec.generation, all the above fields are out of date, the operator should update them later.
  observedGeneration: 1