Configuring registries will be done by specifying (optionally) a hosts.toml file for
each desired registry host in a configuration directory. Note: Updates under this directory
do not require restarting the containerd daemon.
All configured registry hosts are expected to comply with the OCI Distribution Specification. Registries which are non-compliant or implement non-standard behavior are not guaranteed to be supported and may break unexpectedly between releases.
Currently supported OCI Distribution version: v1.0.0
When pulling via ctr use the --hosts-dir option:
ctr images pull --hosts-dir "/etc/containerd/certs.d"
The old CRI config pattern for specifying registry.mirrors and registry.configs has
been DEPRECATED. You should now point your registry config_path to the path where your
hosts.toml files are located.
Modify your config.toml (default location: /etc/containerd/config.toml) as follows:
[plugins."io.containerd.grpc.v1.cri".registry]
config_path = "/etc/containerd/certs.d"If no hosts.toml configuration exists in the host directory, it will fallback to check certificate files based on Docker's certificate file pattern (".crt" files for CA certificates and ".cert"/".key" files for client certificates).
A registry host is the location where container images and artifacts are sourced. These registry hosts may be local or remote and are typically accessed via http/https using the OCI distribution specification. A registry mirror is not a registry host but these mirrors can also be used to pull content. Registry hosts are typically referred to by their internet domain names, aka. registry host names. For example, docker.io, quay.io, gcr.io, and ghcr.io.
A registry host namespace is, for the purpose of containerd registry configuration, a
path to the hosts.toml file specified by the registry host name, or ip address, and an
optional port identifier. When making a pull request for an image the format is
typically as follows:
pull [registry_host_name|IP address][:port][/v2][/org_path]<image_name>[:tag|@DIGEST]
The registry host namespace portion is [registry_host_name|IP address][:port]. Example
tree for docker.io:
$ tree /etc/containerd/certs.d
/etc/containerd/certs.d
└── docker.io
└── hosts.toml
The /v2 portion of the pull request format shown above refers to the version of the
distribution api. If not included in the pull request, /v2 is added by default for all
clients compliant to the distribution specification linked above.
For example when pulling image_name:tag from a private registry named myregistry.io over port 5000:
pull myregistry.io:5000/image_name:tag
The pull will resolve to https://myregistry.io:5000/v2/image_name:tag
When performing image operations via ctr use the --help option to get a list of options you can set for specifying credentials:
ctr i pull --help
...
OPTIONS:
--skip-verify, -k skip SSL certificate validation
--plain-http allow connections using plain HTTP
--user value, -u value user[:password] Registry user and password
--refresh value refresh token for authorization server
--hosts-dir value Custom hosts configuration directory
--tlscacert value path to TLS root CA
--tlscert value path to TLS client certificate
--tlskey value path to TLS client key
--http-dump dump all HTTP request/responses when interacting with container registry
--http-trace enable HTTP tracing for registry interactions
--snapshotter value snapshotter name. Empty value stands for the default value. [$CONTAINERD_SNAPSHOTTER]
--label value labels to attach to the image
--platform value Pull content from a specific platform
--all-platforms pull content and metadata from all platforms
--all-metadata Pull metadata for all platforms
--print-chainid Print the resulting image's chain ID
--max-concurrent-downloads value Set the max concurrent downloads for each pull (default: 0)
Although we have deprecated the old CRI config pattern for specifying registry.mirrors and registry.configs you can still specify your credentials via CRI config.
Additionally, the containerd CRI plugin implements/supports the authentication parameters passed in through CRI pull image service requests.
For example, when containerd is the container runtime implementation for Kubernetes, the containerd CRI plugin receives
authentication credentials from kubelet as retrieved from
Kubernetes Image Pull Secrets
Here is a simple example for a default registry hosts configuration. Set
config_path = "/etc/containerd/certs.d" in your config.toml for containerd.
Make a directory tree at the config path that includes docker.io as a directory
representing the host namespace to be configured. Then add a hosts.toml file
in the docker.io to configure the host namespace. It should look like this:
$ tree /etc/containerd/certs.d
/etc/containerd/certs.d
└── docker.io
└── hosts.toml
$ cat /etc/containerd/certs.d/docker.io/hosts.toml
server = "https://docker.io"
[host."https://registry-1.docker.io"]
capabilities = ["pull", "resolve"]
server = "https://registry-1.docker.io" # Exclude this to not use upstream
[host."https://public-mirror.example.com"]
capabilities = ["pull"] # Requires less trust, won't resolve tag to digest from this host
[host."https://docker-mirror.internal"]
capabilities = ["pull", "resolve"]
ca = "docker-mirror.crt" # Or absolute path /etc/containerd/certs.d/docker.io/docker-mirror.crt
To bypass the TLS verification for a private registry at 192.168.31.250:5000
Create a path and hosts.toml text at the path "/etc/containerd/certs.d/docker.io/hosts.toml" with following or similar contents:
server = "https://registry-1.docker.io"
[host."http://192.168.31.250:5000"]
capabilities = ["pull", "resolve", "push"]
skip_verify = trueFor each registry host namespace directory in your registry config_path you may
include a hosts.toml configuration file. The following root level toml fields
apply to the registry host namespace:
Note: All paths specified in the hosts.toml file may be absolute or relative
to the hosts.toml file.
server specifies the default server for this registry host namespace. When
host(s) are specified, the hosts are tried first in the order listed.
server = "https://docker.io"
capabilities is an optional setting for specifying what operations a host is
capable of performing. Include only the values that apply.
capabilities = ["pull", "resolve", "push"]
capabilities (or Host capabilities) represent the capabilities of the registry host. This also represents the set of operations for which the registry host may be trusted to perform.
For example, pushing is a capability which should only be performed on an upstream source, not a mirror.
Resolving (the process of converting a name into a digest) must be considered a trusted operation and only done by a host which is trusted (or more preferably by secure process which can prove the provenance of the mapping).
A public mirror should never be trusted to do a resolve action.
| Registry Type | Pull | Resolve | Push |
|---|---|---|---|
| Public Registry | yes | yes | yes |
| Private Registry | yes | yes | yes |
| Public Mirror | yes | no | no |
| Private Mirror | yes | yes | no |
ca (Certificate Authority Certification) can be set to a path or an array of
paths each pointing to a ca file for use in authenticating with the registry
namespace.
ca = "/etc/certs/mirror.pem"
or
ca = ["/etc/certs/test-1-ca.pem", "/etc/certs/special.pem"]
client certificates are configured as follows
a path:
client = "/etc/certs/client.pem"
an array of paths:
client = ["/etc/certs/client-1.pem", "/etc/certs/client-2.pem"]
an array of pairs of paths:
client = [["/etc/certs/client.cert", "/etc/certs/client.key"],["/etc/certs/client.pem", ""]]
skip_verify skips verifications of the registry's certificate chain and
host name when set to true. This should only be used for testing or in
combination with other method of verifying connections. (Defaults to false)
skip_verify = false
[header] contains some number of keys where each key is to one of a string or
an array of strings as follows:
[header]
x-custom-1 = "custom header"
or
[header]
x-custom-1 = ["custom header part a","part b"]
or
[header]
x-custom-1 = "custom header",
x-custom-1-2 = "another custom header"
override_path is used to indicate the host's API root endpoint is defined
in the URL path rather than by the API specification. This may be used with
non-compliant OCI registries which are missing the /v2 prefix.
(Defaults to false)
override_path = true
[host]."https://namespace" and [host].http://namespace entries in the
hosts.toml configuration are registry namespaces used in lieu of the default
registry host namespace. These hosts are sometimes called mirrors because they
may contain a copy of the container images and artifacts you are attempting to
retrieve from the default registry. Each host/mirror namespace is also
configured in much the same way as the default registry namespace. Notably the
server is not specified in the host description because it is specified in
the namespace. Here are a few rough examples configuring host mirror namespaces
for this registry host namespace:
[host."https://mirror.registry"]
capabilities = ["pull"]
ca = "/etc/certs/mirror.pem"
skip_verify = false
[host."https://mirror.registry".header]
x-custom-2 = ["value1", "value2"]
[host."https://mirror-bak.registry/us"]
capabilities = ["pull"]
skip_verify = true
[host."http://mirror.registry"]
capabilities = ["pull"]
[host."https://test-1.registry"]
capabilities = ["pull", "resolve", "push"]
ca = ["/etc/certs/test-1-ca.pem", "/etc/certs/special.pem"]
client = [["/etc/certs/client.cert", "/etc/certs/client.key"],["/etc/certs/client.pem", ""]]
[host."https://test-2.registry"]
client = "/etc/certs/client.pem"
[host."https://test-3.registry"]
client = ["/etc/certs/client-1.pem", "/etc/certs/client-2.pem"]
[host."https://non-compliant-mirror.registry/v2/upstream"]
capabilities = ["pull"]
override_path = true
Note: Recursion is not supported in the specification of host mirror namespaces in the hosts.toml file. Thus the following is not allowed/supported:
[host."http://mirror.registry"]
capabilities = ["pull"]
[host."http://double-mirror.registry"]
capabilities = ["pull"]