Helm Subchart order of execution in an umbrella chart

前端 未结 3 951
误落风尘
误落风尘 2020-12-31 03:13

I have an umbrella chart with multiple sub-charts, I simply want to make sure that subchart1 runs before subchart2, etc. How can we define the order of subchart execution?

相关标签:
3条回答
  • 2020-12-31 03:42

    While you cannot make an order, you can use Init Containers to validate your Pods have everything they need to run before actually running them: https://kubernetes.io/docs/concepts/workloads/pods/init-containers/

    0 讨论(0)
  • 2020-12-31 03:45

    Here is a part of the Helm documentation related to execution order of charts:

    The above sections explain how to specify chart dependencies, but how does this affect chart installation using helm install and helm upgrade?

    Suppose that a chart named "A" creates the following Kubernetes objects

    namespace "A-Namespace"
    statefulset "A-StatefulSet"
    service "A-Service"

    Furthermore, A is dependent on chart B that creates objects

    namespace "B-Namespace"
    replicaset "B-ReplicaSet"
    service "B-Service"

    After installation/upgrade of chart A a single Helm release is created/modified. The release will create/update all of the above Kubernetes objects in the following order:

    A-Namespace
    B-Namespace
    A-StatefulSet
    B-ReplicaSet
    A-Service
    B-Service

    This is because when Helm installs/upgrades charts, the Kubernetes objects from the charts and all its dependencies are aggregrated into a single set; then sorted by type followed by name; and then created/updated in that order.
    Hence a single release is created with all the objects for the chart and its dependencies.

    The install order of Kubernetes types is given by the enumeration InstallOrder in kind_sorter.go (Helm v2 source file).

    Part of kind_sorter.go (Helm v3 source) is related to install charts:

    var InstallOrder KindSortOrder = []string{
        "Namespace",
        "NetworkPolicy",
        "ResourceQuota",
        "LimitRange",
        "PodSecurityPolicy",
        "PodDisruptionBudget",
        "Secret",
        "ConfigMap",
        "StorageClass",
        "PersistentVolume",
        "PersistentVolumeClaim",
        "ServiceAccount",
        "CustomResourceDefinition",
        "ClusterRole",
        "ClusterRoleList",
        "ClusterRoleBinding",
        "ClusterRoleBindingList",
        "Role",
        "RoleList",
        "RoleBinding",
        "RoleBindingList",
        "Service",
        "DaemonSet",
        "Pod",
        "ReplicationController",
        "ReplicaSet",
        "Deployment",
        "HorizontalPodAutoscaler",
        "StatefulSet",
        "Job",
        "CronJob",
        "Ingress",
        "APIService",
    }
    

    There is a workaround, that can change the default behaviour, shared by elementalvoid in this issue:

    I've been setting my services, secrets, and configmaps as pre-install hooks to achieve this behavior.

    Example:

    apiVersion: v1
    kind: Service
    metadata:
      name: foo
      annotations:
        "helm.sh/hook": "pre-install"
    

    --

    It is possible to define a weight for a hook which will help build a deterministic executing order. Weights are defined using the following annotation:

      annotations:
        "helm.sh/hook-weight": "5"
    

    Hook weights can be positive or negative numbers but must be represented as strings. When Tiller starts the execution cycle of hooks of a particular Kind it will sort those hooks in ascending order.

    More detailed information about hooks can be found here (v2 doc, v3 doc) and in the source file (helm v2 source, helm v3 source)

    0 讨论(0)
  • 2020-12-31 03:56

    You cannot make an order, and I don't think you will need it. You can rely on health checks, so if one chart depends on another, it will probably fail and restart until health check passes.

    0 讨论(0)
提交回复
热议问题