c12s-kubespray/docs/cri-o.md
Nicolas MASSE f01f7c54aa
Add support for CRI-O user namespaces (#8268)
* add support for cri-o user namespaces

* comply with yamllint rules
2021-12-20 06:37:25 -08:00

2.2 KiB

CRI-O

CRI-O is a lightweight container runtime for Kubernetes. Kubespray supports basic functionality for using CRI-O as the default container runtime in a cluster.

  • Kubernetes supports CRI-O on v1.11.1 or later.
  • etcd: configure either kubeadm managed etcd or host deployment

To use the CRI-O container runtime set the following variables:

all/all.yml

download_container: false
skip_downloads: false
etcd_kubeadm_enabled: true

k8s_cluster/k8s_cluster.yml

container_manager: crio

etcd.yml

etcd_deployment_type: host # optionally and mutually exclusive with etcd_kubeadm_enabled

all/crio.yml

Enable docker hub registry mirrors

crio_registries_mirrors:
  - prefix: docker.io
    insecure: false
    blocked: false
    location: registry-1.docker.io
    mirrors:
      - location: 192.168.100.100:5000
        insecure: true
      - location: mirror.gcr.io
        insecure: false

Note about pids_limit

For heavily mult-threaded workloads like databases, the default of 1024 for pids-limit is too low. This parameter controls not just the number of processes but also the amount of threads (since a thread is technically a process with shared memory). See cri-o#1921

In order to increase the default pids_limit for cri-o based deployments you need to set the crio_pids_limit for your k8s_cluster ansible group or per node depending on the use case.

crio_pids_limit: 4096

Note about user namespaces

CRI-O has support for user namespaces. This feature is optional and can be enabled by setting the following two variables.

crio_runtimes:
  - name: runc
    path: /usr/bin/runc
    type: oci
    root: /run/runc
    allowed_annotations:
    - "io.kubernetes.cri-o.userns-mode"

crio_remap_enable: true

The allowed_annotations configures crio.conf accordingly.

The crio_remap_enable configures the /etc/subuid and /etc/subgid files to add an entry for the containers user. By default, 16M uids and gids are reserved for user namespaces (256 pods * 65536 uids/gids) at the end of the uid/gid space.