Skip to content
This repository has been archived by the owner. It is now read-only.
Switch branches/tags
Go to file
Cannot retrieve contributors at this time

CRI Plugin Config Guide

This document provides the description of the CRI plugin configuration. The CRI plugin config is part of the containerd config (default path: /etc/containerd/config.toml).

See here for more information about containerd config.

The explanation and default value of each configuration item are as follows:

# Use config version 2 to enable new configuration fields.
# Config file is parsed as version 1 by default.
# Version 2 uses long plugin names, i.e. "io.containerd.grpc.v1.cri" vs "cri".
version = 2

# The 'plugins."io.containerd.grpc.v1.cri"' table contains all of the server options.

  # disable_tcp_service disables serving CRI on the TCP server.
  # Note that a TCP server is enabled for containerd if TCPAddress is set in section [grpc].
  disable_tcp_service = true

  # stream_server_address is the ip address streaming server is listening on.
  stream_server_address = ""

  # stream_server_port is the port streaming server is listening on.
  stream_server_port = "0"

  # stream_idle_timeout is the maximum time a streaming connection can be
  # idle before the connection is automatically closed.
  # The string is in the golang duration format, see:
  stream_idle_timeout = "4h"

  # enable_selinux indicates to enable the selinux support.
  enable_selinux = false

  # selinux_category_range allows the upper bound on the category range to be set.
  # if not specified or set to 0, defaults to 1024 from the selinux package.
  selinux_category_range = 1024

  # sandbox_image is the image used by sandbox container.
  sandbox_image = ""

  # stats_collect_period is the period (in seconds) of snapshots stats collection.
  stats_collect_period = 10

  # enable_tls_streaming enables the TLS streaming support.
  # It generates a self-sign certificate unless the following x509_key_pair_streaming are both set.
  enable_tls_streaming = false

  # tolerate_missing_hugetlb_controller if set to false will error out on create/update
  # container requests with huge page limits if the cgroup controller for hugepages is not present.
  # This helps with supporting Kubernetes <=1.18 out of the box. (default is `true`)
  tolerate_missing_hugetlb_controller = true

  # ignore_image_defined_volumes ignores volumes defined by the image. Useful for better resource
	# isolation, security and early detection of issues in the mount configuration when using
	# ReadOnlyRootFilesystem since containers won't silently mount a temporary volume.
  ignore_image_defined_volumes = false

  # 'plugins."io.containerd.grpc.v1.cri".x509_key_pair_streaming' contains a x509 valid key pair to stream with tls.
    # tls_cert_file is the filepath to the certificate paired with the "tls_key_file"
    tls_cert_file = ""

    # tls_key_file is the filepath to the private key paired with the "tls_cert_file"
    tls_key_file = ""

  # max_container_log_line_size is the maximum log line size in bytes for a container.
  # Log line longer than the limit will be split into multiple lines. -1 means no
  # limit.
  max_container_log_line_size = 16384

  # disable_cgroup indicates to disable the cgroup support.
  # This is useful when the daemon does not have permission to access cgroup.
  disable_cgroup = false

  # disable_apparmor indicates to disable the apparmor support.
  # This is useful when the daemon does not have permission to access apparmor.
  disable_apparmor = false

  # restrict_oom_score_adj indicates to limit the lower bound of OOMScoreAdj to
  # the containerd's current OOMScoreAdj.
  # This is useful when the containerd does not have permission to decrease OOMScoreAdj.
  restrict_oom_score_adj = false

  # max_concurrent_downloads restricts the number of concurrent downloads for each image.
  max_concurrent_downloads = 3

  # disable_proc_mount disables Kubernetes ProcMount support. This MUST be set to `true`
  # when using containerd with Kubernetes <=1.11.
  disable_proc_mount = false

  # unsetSeccompProfile is the profile containerd/cri will use if the provided seccomp profile is
  # unset (`""`) for a container (default is `unconfined`)
  unset_seccomp_profile = ""

  # 'plugins."io.containerd.grpc.v1.cri".containerd' contains config related to containerd

    # snapshotter is the snapshotter used by containerd.
    snapshotter = "overlayfs"

    # no_pivot disables pivot-root (linux only), required when running a container in a RamDisk with runc.
    # This only works for runtime type "io.containerd.runtime.v1.linux".
    no_pivot = false

    # disable_snapshot_annotations disables to pass additional annotations (image
    # related information) to snapshotters. These annotations are required by
    # stargz snapshotter (
    disable_snapshot_annotations = false

    # discard_unpacked_layers allows GC to remove layers from the content store after
    # successfully unpacking these layers to the snapshotter.
    discard_unpacked_layers = false

    # default_runtime_name is the default runtime name to use.
    default_runtime_name = "runc"

    # 'plugins."io.containerd.grpc.v1.cri".containerd.default_runtime' is the runtime to use in containerd.
    # DEPRECATED: use `default_runtime_name` and `plugins."io.containerd.grpc.v1.cri".runtimes` instead.
    # Remove in containerd 1.4.

    # 'plugins."io.containerd.grpc.v1.cri".containerd.untrusted_workload_runtime' is a runtime to run untrusted workloads on it.
    # DEPRECATED: use `untrusted` runtime in `plugins."io.containerd.grpc.v1.cri".runtimes` instead.
    # Remove in containerd 1.4.

    # 'plugins."io.containerd.grpc.v1.cri".containerd.runtimes' is a map from CRI RuntimeHandler strings, which specify types
    # of runtime configurations, to the matching configurations.
    # In this example, 'runc' is the RuntimeHandler string to match.
      # runtime_type is the runtime type to use in containerd.
      # The default value is "io.containerd.runc.v2" since containerd 1.4.
      # The default value was "io.containerd.runc.v1" in containerd 1.3, "io.containerd.runtime.v1.linux" in prior releases.
      runtime_type = "io.containerd.runc.v2"

      # pod_annotations is a list of pod annotations passed to both pod
      # sandbox as well as container OCI annotations. Pod_annotations also
      # supports golang path match pattern -
      # e.g. ["*"], ["*"], ["*"].
      # For the naming convention of annotation keys, please reference:
      # * Kubernetes:
      # * OCI:
      pod_annotations = []

      # container_annotations is a list of container annotations passed through to the OCI config of the containers.
      # Container annotations in CRI are usually generated by other Kubernetes node components (i.e., not users).
      # Currently, only device plugins populate the annotations.
      container_annotations = []

      # privileged_without_host_devices allows overloading the default behaviour of passing host
      # devices through to privileged containers. This is useful when using a runtime where it does
      # not make sense to pass host devices to the container when privileged. Defaults to false -
      # i.e pass host devices through to privileged containers.
      privileged_without_host_devices = false

      # base_runtime_spec is a file path to a JSON file with the OCI spec that will be used as the base spec that all
      # container's are created from.
      # Use containerd's `ctr oci spec > /etc/containerd/cri-base.json` to output initial spec file.
      # Spec files are loaded at launch, so containerd daemon must be restared on any changes to refresh default specs.
      # Still running containers and restarted containers will still be using the original spec from which that container was created.
      base_runtime_spec = ""

      # 'plugins."io.containerd.grpc.v1.cri".containerd.runtimes.runc.options' is options specific to
      # "io.containerd.runc.v1" and "io.containerd.runc.v2". Its corresponding options type is:
      # .
        # NoPivotRoot disables pivot root when creating a container.
        NoPivotRoot = false

        # NoNewKeyring disables new keyring for the container.
        NoNewKeyring = false

        # ShimCgroup places the shim in a cgroup.
        ShimCgroup = ""

        # IoUid sets the I/O's pipes uid.
        IoUid = 0

        # IoGid sets the I/O's pipes gid.
        IoGid = 0

        # BinaryName is the binary name of the runc binary.
        BinaryName = ""

        # Root is the runc root directory.
        Root = ""

        # CriuPath is the criu binary path.
        CriuPath = ""

        # SystemdCgroup enables systemd cgroups.
        SystemdCgroup = false

        # CriuImagePath is the criu image path
        CriuImagePath = ""

        # CriuWorkPath is the criu work path.
        CriuWorkPath = ""

  # 'plugins."io.containerd.grpc.v1.cri".cni' contains config related to cni
    # bin_dir is the directory in which the binaries for the plugin is kept.
    bin_dir = "/opt/cni/bin"

    # conf_dir is the directory in which the admin places a CNI conf.
    conf_dir = "/etc/cni/net.d"

    # max_conf_num specifies the maximum number of CNI plugin config files to
    # load from the CNI config directory. By default, only 1 CNI plugin config
    # file will be loaded. If you want to load multiple CNI plugin config files
    # set max_conf_num to the number desired. Setting max_config_num to 0 is
    # interpreted as no limit is desired and will result in all CNI plugin
    # config files being loaded from the CNI config directory.
    max_conf_num = 1

    # conf_template is the file path of golang template used to generate
    # cni config.
    # If this is set, containerd will generate a cni config file from the
    # template. Otherwise, containerd will wait for the system admin or cni
    # daemon to drop the config file into the conf_dir.
    # This is a temporary backward-compatible solution for kubenet users
    # who don't have a cni daemonset in production yet.
    # This will be deprecated when kubenet is deprecated.
    # See the "CNI Config Template" section for more details.
    conf_template = ""

  # 'plugins."io.containerd.grpc.v1.cri".registry' contains config related to the registry

    # 'plugins."io.containerd.grpc.v1.cri.registry.headers sets the http request headers to send for all registry requests
        Foo = ["bar"]

    # 'plugins."io.containerd.grpc.v1.cri".registry.mirrors' are namespace to mirror mapping for all namespaces.
        endpoint = ["", ]

  # 'plugins."io.containerd.grpc.v1.cri".image_decryption' contains config related
  # to handling decryption of encrypted container images.
    # key_model defines the name of the key model used for how the cri obtains
    # keys used for decryption of encrypted container images.
    # The [decryption document](
    # contains additional information about the key models available. 
    # Set of available string options: {"", "node"}
    # Omission of this field defaults to the empty string "", which indicates no key model, 
    # disabling image decryption.
    # In order to use the decryption feature, additional configurations must be made.
    # The [decryption document](
    # provides information of how to set up stream processors and the containerd imgcrypt decoder
    # with the appropriate key models.
    # Additional information:
    # * Stream processors:
    # * Containerd imgcrypt:
    key_model = "node"

Untrusted Workload

The recommended way to run untrusted workload is to use RuntimeClass api introduced in Kubernetes 1.12 to select RuntimeHandlers configured to run untrusted workload in plugins."io.containerd.grpc.v1.cri".containerd.runtimes.

However, if you are using the legacy io.kubernetes.cri.untrusted-workloadpod annotation to request a pod be run using a runtime for untrusted workloads, the RuntimeHandler plugins."io.containerd.grpc.v1.cri"cri.containerd.runtimes.untrusted must be defined first. When the annotation io.kubernetes.cri.untrusted-workload is set to true the untrusted runtime will be used. For example, see Create an untrusted pod using Kata Containers.

CNI Config Template

Ideally the cni config should be placed by system admin or cni daemon like calico, weaveworks etc. However, there are still users using kubenet today, who don't have a cni daemonset in production. The cni config template is a temporary backward-compatible solution for them. This is expected to be deprecated when kubenet is deprecated.

The cni config template uses the golang template format. Currently supported values are:

  • .PodCIDR is a string of the first CIDR assigned to the node.
  • .PodCIDRRanges is a string array of all CIDRs assigned to the node. It is usually used for dualstack support.
  • .Routes is a string array of all routes needed. It is usually used for dualstack support or single stack but IPv4 or IPv6 is decided at runtime.

The golang template actions can be used to render the cni config. For example, you can use the following template to add CIDRs and routes for dualstack in the CNI config:

"ipam": {
  "type": "host-local",
  "ranges": [{{range $i, $range := .PodCIDRRanges}}{{if $i}}, {{end}}[{"subnet": "{{$range}}"}]{{end}}],
  "routes": [{{range $i, $route := .Routes}}{{if $i}}, {{end}}{"dst": "{{$route}}"}{{end}}]


The config options of the CRI plugin follow the Kubernetes deprecation policy of "admin-facing CLI components".

In summary, when a config option is announced to be deprecated:

  • It is kept functional for 6 months or 1 release (whichever is longer);
  • A warning is emitted when it is used.