Skip to content

Latest commit

 

History

History
736 lines (617 loc) · 29.8 KB

README.md

File metadata and controls

736 lines (617 loc) · 29.8 KB
Raw

podman

ansible-lint.yml ansible-test.yml markdownlint.yml woke.yml

This role manages podman configuration, containers, and systemd services which run podman containers.

Requirements

The role requires podman version 4.2 or later. The role requires podman version 4.4 or later for quadlet support and secret support. The role requires podman version 4.5 or later for support for using healthchecks (only supported when using quadlet Container types).

Collection requirements

The role requires the following collections:

  • containers.podman
  • fedora.linux_system_roles

Use this to install the collections:

ansible-galaxy collection install -vv -r meta/collection-requirements.yml

Users, groups, subuid, subgid

Users and groups specified in podman_run_as_user, podman_run_as_group, and specified in a kube spec as run_as_user and run_as_group have the following restrictions:

  • They must be already present on the system - the role will not create the users or groups - the role will exit with an error if a non-existent user or group is specified
  • They must already exist in /etc/subuid and /etc/subgid, or are otherwise provided by your identity management system - the role will exit with an error if a specified user is not present in /etc/subuid, or if a specified group is not in /etc/subgid. The role uses getsubids to check the user and group if available, or checks the files directly if getsubids is not available.

Role Variables

podman_kube_specs

This is a list. Each element of the list is a dict describing a podman pod and corresponding systemd unit to manage. The format of the dict is mostly like the podman_play module except for the following:

  • state - default is created. This takes 3 values:
    • started - Create the pods and systemd services, and start them running
    • created - Create the pods and systemd services, but do not start them
    • absent - Remove the pods and systemd services
  • run_as_user - Use this to specify a per-pod user. If you do not specify this, then the global default podman_run_as_user value will be used. Otherwise, root will be used. NOTE: The user must already exist - the role will not create one. The user must be present in /etc/subuid.
  • run_as_group - Use this to specify a per-pod group. If you do not specify this, then the global default podman_run_as_group value will be used. Otherwise, root will be used. NOTE: The group must already exist - the role will not create one. The group must be present in /etc/subgid.
  • systemd_unit_scope - The scope to use for the systemd unit. If you do not specify this, then the global default podman_systemd_unit_scope will be used. Otherwise, the scope will be system for root containers, and user for user containers.
  • activate_systemd_unit - Whether or not to activate the systemd unit when it is created. If you do not specify this, then the global default podman_activate_systemd_unit will be used, which is true by default.
  • pull_image - Ensure the image is pulled before use. If you do not specify this, then the global default podman_pull_image will be used, which is true by default.
  • continue_if_pull_fails - If pulling the image, and the pull fails, do not treat this as a fatal error, and continue with the role. If you do not specify this, then the global default podman_continue_if_pull_fails will be used, which is false by default.
  • kube_file_src - This is the name of a file on the controller node which will be copied to kube_file on the managed node. This is a file in Kubernetes YAML format. Do not specify this if you specify kube_file_content. kube_file_content takes precedence over kube_file_src.
  • kube_file_content - This is either a string in Kubernetes YAML format, or a dict in Kubernetes YAML format. It will be used as the contents of kube_file on the managed node. Do not specify this if you specify kube_file_src. kube_file_content takes precedence over kube_file_src.
  • kube_file - This is the name of a file on the managed node that contains the Kubernetes specification of the container/pod. You typically do not have to specify this unless you need to somehow copy this file to the managed node outside of the role. If you specify either kube_file_src or kube_file_content, you do not have to specify this. It is highly recommended to omit kube_file and instead specify either kube_file_src or kube_file_content and let the role manage the file path and name.
    • The file basename will be the metadata.name value from the K8s yaml, with a .yml suffix appended to it.
    • The directory will be /etc/containers/ansible-kubernetes.d for system services.
    • The directory will be $HOME/.config/containers/ansible-kubernetes.d for user services.

For example, if you have

    podman_kube_specs:
      - state: started
        kube_file_content:
          apiVersion: v1
          kind: Pod
          metadata:
            name: myappname

This will be copied to the file /etc/containers/ansible-kubernetes.d/myappname.yml on the managed node.

podman_quadlet_specs

List of [Quadlet specifications] (https://docs.podman.io/en/latest/markdown/podman-systemd.unit.5.html) A quadlet spec is uniquely identified by a name and a type, where type is one of the types of units like container, kube, network, volume, etc. You can either pass in name and type explicitly, or the name and type will be derived from the file name given in file, file_src, or template_src.

By default, the files will be copied to or created in /etc/containers/systemd/$name.$type for root containers, and $HOME/.config/containers/systemd/$name.$type for rootless containers, on the managed node. You can provide a different location by using file, but then you will likely need to change the systemd configuration to find the file, which is not supported by this role.

When a quadlet spec depends on some other file e.g. a quadlet.kube that depends on the Yaml file or a ConfigMap, then that file must be specified in the podman_quadlet_specs list before the file that uses it. For example, if you have a file my-app.kube:

[Kube]
ConfigMap=my-app-config.yml
Yaml=my-app.yml
...

Then you must specify my-app-config.yml and my-app.yml before my-app.kube:

podman_quadlet_specs:
  - file_src: my-app-config.yml
  - file_src: my-app.yml
  - file_src: my-app.kube

Most of the parameters for each quadlet spec are the same as for podman_kube_spec above except that the *kube* parameters are not supported, and the following are:

  • name - The name of the unit. If not given, it will be derived from file, file_src, or template_src. For example, if you specify file_src: /path/to/my-container.container then the name will be my-container.
  • type - The type of unit (container, kube, volume, etc.). If not given, it will be derived from file, file_src, or template_src. For example, if you specify file_src: /path/to/my-container.container then the type will be container. If the derived type is not recognized as a valid quadlet type, for example, if you specify file_src: my-kube.yml, then it will just be copied and not processed as a quadlet spec.
  • file_src - The name of the file on the control node to copy to the managed node to use as the source of the quadlet unit. If this file is in the quadlet unit format and has a valid quadlet unit suffix, it will be used as a quadlet unit, otherwise, it will just be copied.
  • file - The name of the file on the managed node to use as the source of the quadlet unit. If this file is in the quadlet unit format and has a valid quadlet unit suffix, it will be used as a quadlet unit, otherwise, it will be treated as a regular file.
  • file_content - The contents of a file to copy to the managed node, in string format. This is useful to pass in short files that can easily be specified inline. You must also specify name and type.
  • template_src - The name of the file on the control node which will be processed as a Jinja template file then copied to the managed node to use as the source of the quadlet unit. If this file is in the quadlet unit format and has a valid quadlet unit suffix, it will be used as a quadlet unit, otherwise, it will just be copied. If the file has a .j2 suffix, that suffix will be stripped to determine the quadlet file type.

For example, if you specify:

podman_quadlet_specs:
  - template_src: my-app.container.j2

Then the local file templates/my-app.container.j2 will be processed as a Jinja template file, then copied to /etc/containers/systemd/my-app.container as a quadlet container unit spec on the managed node.

NOTE: When removing quadlets, you must remove networks last. You cannot remove a network that is in use by a container.

podman_secrets

This is a list of secret specs in almost the same format as used by podman_secret There is an additional field:

  • run_as_user - Use this to specify a secret for a specific user. If you do not specify this, then the global default podman_run_as_user value will be used. Otherwise, root will be used. NOTE: The user must already exist - the role will not create one.

You are strongly encouraged to use Ansible Vault to encrypt the value of the data field.

podman_create_host_directories

This is a boolean, default value is false. If true, the role will ensure host directories specified in host mounts in volumes.hostPath specifications in the Kubernetes YAML given in podman_kube_specs, and from Volume configuration in quadlet Container specification where a host path is specified. NOTE: Directories must be specified as absolute paths (for root containers), or paths relative to the home directory (for non-root containers), in order for the role to manage them. Anything else will be assumed to be some other sort of volume and will be ignored. The role will apply its default ownership/permissions to the directories. If you need to set ownership/permissions, see podman_host_directories.

podman_host_directories

This is a dict. When using podman_create_host_directories, this tells the role what permissions/ownership to apply to automatically created host directories. Each key is the absolute path of host directory to manage. The value is in the format of the parameters to the file module. If you do not specify a value, the role will use its built-in default values. If you want to specify a value to be used for all host directories, use the special key DEFAULT.

podman_host_directories:
  "/var/lib/data":
    owner: dbuser
    group: dbgroup
    mode: "0600"
  DEFAULT:
    owner: root
    group: root
    mode: "0644"

The role will use dbuser:dbgroup 0600 for /var/lib/data, and root:root 0644 for all other host directories created by the role.

podman_firewall

This is a list of dict in the same format as used by the fedora.linux_system_roles.firewall role. Use this to specify ports that you want the role to manage in the firewall.

podman_firewall:
  - port: 8080/tcp

podman_selinux_ports

This is a list of dict in the same format as used by the fedora.linux_system_roles.selinux role. Use this if you want the role to manage the SELinux policy for ports used by the role.

podman_selinux_ports:
  - ports: 8080
    protocol: tcp
    setype: http_port_t

podman_run_as_user

This is the name of the user to use for all rootless containers. You can also specify per-container username with run_as_user in podman_kube_specs. NOTE: The user must already exist - the role will not create one. The user must be present in /etc/subuid.

podman_run_as_group

This is the name of the group to use for all rootless containers. You can also specify per-container group name with run_as_group in podman_kube_specs. NOTE: The group must already exist - the role will not create one. The group must be present in /etc/subgid.

podman_systemd_unit_scope

This is systemd scope to use by default for all systemd units. You can also specify per-container scope with systemd_unit_scope in podman_kube_specs. By default, rootless containers will use user and root containers will use system.

podman_activate_systemd_units

Activate each systemd unit as soon as it is created. The default is true. You can also do this on a per-unit basis by using activate_systemd_units in the spec for each unit. For example, if you are deploying several specs, and you only want the last one in the list to activate which will trigger the others to activate via dependencies, then set activate_systemd_unit: false for each one except the last one uses activate_systemd_unit: true. NOTE: quadlet units are implicitly enabled when created - you cannot currently use activate_systemd_unit to disable those units - you can use activate_systemd_unit to create units stopped or started.

podman_pull_image

Ensure that each image mentioned in a kube or quadlet spec is present by pulling the image before it is used. The default is true. Use false if the managed node already has the correct version, or is not able to pull images. You can also specify this on a per-unit basis with pull_image.

podman_continue_if_pull_fails

If the image pull attempt fails, do not treat this as a fatal error, and continue with the role run. The default is false - a pull attempt failure is a fatal error. You can set this on a per-unit basis with continue_if_pull_fails.

podman_containers_conf

These are the containers.conf(5) settings, provided as a dict. These settings will be provided in a drop-in file in the containers.conf.d directory. If running as root (see podman_run_as_user), the system settings will be managed, otherwise, the user settings will be managed. See the man page for the directory locations.

podman_containers_conf:
  containers:
    default_sysctls:
      - net.ipv4.ping_group_range=0 1000
      - user.max_ipc_namespaces=125052

podman_registries_conf

These are the containers-registries.conf(5) settings, provided as a dict. These settings will be provided in a drop-in file in the registries.conf.d directory. If running as root (see podman_run_as_user), the system settings will be managed, otherwise, the user settings will be managed. See the man page for the directory locations.

podman_registries_conf:
  aliases:
    myregistry: registry.example.com

podman_storage_conf

These are the containers-storage.conf(5) settings, provided as a dict. If running as root (see podman_run_as_user), the system settings will be managed, otherwise, the user settings will be managed. See the man page for the file locations.

podman_storage_conf:
  storage:
    runroot: /var/big/partition/container/storage

podman_policy_json

These are the containers-policy.json(5) settings, provided as a dict. If running as root (see podman_run_as_user), the system settings will be managed, otherwise, the user settings will be managed. See the man page for the file locations.

podman_policy_json:
  default:
    type: insecureAcceptAnything

podman_use_copr (EXPERIMENTAL)

Boolean - default is unset - if you want to enable the copr repo to use the latest development version of podman, use podman_use_copr: true

podman_fail_if_too_old (EXPERIMENTAL)

Boolean - default is unset - by default, the role will fail with an error if you are using an older version of podman and try to use a feature only supported by a newer version. For example, if you attempt to manage quadlet or secrets with podman 4.3 or earlier, the role will fail with an error. If you want the role to be skipped instead, use podman_fail_if_too_old: false.

podman_registry_username

String - default is unset - username to use to authenticate to the registry. You must also set podman_registry_password. You can override this on a per-spec basis with registry_username. The use of container_image_user was unsupported and is deprecated.

podman_registry_password

String - default is unset - password to use to authenticate to the registry. You must also set podman_registry_username. You can override this on a per-spec basis with registry_password. The use of container_image_password was unsupported and is deprecated.

podman_credential_files

This is a list. Each element of the list is a dict describing a podman credential file used to authenticate to registries. See man containers-auth.json and man containers-registries.conf:credential-helpers for more information about the format of these files, and the default directory search path. NOTE: These files contain authentication credentials. Please be careful with them. You are strongly encouraged to use Ansible Vault to encrypt them. The keys of each dict are as follows:

  • state - default is present. Use absent to remove files.
  • file_src - This is the name of a file on the controller node which will be copied to file on the managed node. Do not specify this if you specify file_content or template_src, which will take precedence over file_src.
  • template_src - This is the name of a file on the controller node which will be templated using the template module and copied to file on the managed node. Do not specify this if you specify file_content or file_src.
  • file_content - This is a string in containers-auth.json format. It will be used as the contents of file on the managed node. Do not specify this if you specify file_src or template_src.
  • file - This is the name of a file on the managed node that will contain the auth.json contents. The default value will be $HOME/.config/containers/auth.json. If you specify a relative path, it will be relative to $HOME/.config/containers. If you specify something other than the defaults mentioned in man containers-auth.json, you will also need to configure credential-helpers in containers-registries.conf using podman_registries_conf. Any missing parent directories will be created.
  • run_as_user - Use this to specify a per-credential file owner. If you do not specify this, then the global default podman_run_as_user value will be used. Otherwise, root will be used. NOTE: The user must already exist - the role will not create one. The user must be present in /etc/subuid. NOTE: This is used as the user for the $HOME directory if file is not specified, and as the owner of the file. If you want the owner of the file to be different than the user used for $HOME, specify file as an absolute path.
  • run_as_group - Use this to specify a per-credential file group. If you do not specify this, then the global default podman_run_as_group value will be used. Otherwise, root will be used. NOTE: The group must already exist - the role will not create one. The group must be present in /etc/subgid.
  • mode - The mode of the file - default is "0600".

For example, if you have

    podman_credential_files:
      - file_src: auth.json
        run_as_user: my_user

The local file auth.json will be looked up in the usual Ansible file search paths and will be copied to the file /home/my_user/.config/containers/auth.json on the managed node. The file owner will be my_user and the mode will be "0600". The directories /home/my_user/.config and /home/my_user/.config/containers will be created if they do not exist.

podman_registry_certificates

This variable is a list of dict elements that allows you to manage TLS certificates and keys used to connect to registries. The directories, formats, and files are as described in man containers-certs.d. The names of the keys used for TLS certificates and keys follow the system roles TLS naming conventions. NOTE: the client_ prefix has been dropped here for cert and private_key because there are only clients in this context.

NOTE: You are strongly encouraged to use Ansible Vault to encrypt private keys and any other sensitive values.

The keys of each dict are as follows:

  • state - default is present. Use absent to remove files.
  • run_as_user - This is the user that will be the owner of the files, and is used to find the $HOME directory for the files. If you do not specify this, then the global default podman_run_as_user value will be used. Otherwise, root will be used.
  • run_as_group - This is the group that will be the owner of the files. If you do not specify this, then the global default podman_run_as_group value will be used. Otherwise, root will be used.
  • registry_host - Required - the hostname or hostname:port of the registry. This will be used as the name of the directory under $HOME/.config/containers/certs.d (for rootless containers) or /etc/containers/certs.d (for system containers) which will hold the certificates and keys. If using state: absent and all of the files are removed, the directory will be removed.
  • cert - name of the file in the certs.d directory holding the TLS client certificate. If not specified, use the basename of cert_src. If that isn't specified, use client.cert.
  • private_key - name of the file in the certs.d directory holding the TLS client private key. If not specified, use the basename of private_key_src. If that isn't specified, use client.key
  • ca_cert - name of the file in the certs.d directory holding the TLS CA certificate. If not specified, use the basename of ca_cert_src. If that isn't specified, use ca.crt
  • cert_src - name of the file on the control node to be copied to cert.
  • private_key_src - name of the file on the control node to be copied to private_key.
  • ca_cert_src - name of the file on the control node to be copied to ca_cert.
  • cert_content - contents of the certificate to be copied to cert.
  • private_key_content - contents of the private key to be copied to private_key.
podman_run_as_user: root
podman_registry_certificates:
  - registry_host: quay.io:5000
    cert_src: client.cert
    private_key_content: !vault |
      $ANSIBLE_VAULT.....
    ca_cert_src: ca.crt

This will create the directory /etc/containers/certs.d/quay.io:5000/, will copy the local file client.cert looked up from the usual Ansible file lookup path to /etc/containers/certs.d/quay.io:5000/client.cert, will copy the contents of the Ansible Vault encrypted private_key_content to /etc/containers/certs.d/quay.io:5000/client.key, and will copy the local file ca.crt looked up from the usual Ansible file lookup path to /etc/containers/certs.d/quay.io:5000/ca.crt.

podman_validate_certs

Boolean - default is null - use this to control if pulling images from registries will validate TLS certs or not. The default null means to use whatever is the default used by the containers.podman.podman_image module. You can override this on a per-spec basis using validate_certs.

Variables Exported by the Role

podman_version

This is the version string of the version used by podman. You can generally use this in your templates. For example, if you want to specify a quadlet template_src for a container, and have it use healthchecks and secrets if using podman 4.5 or later:

podman_quadlet_specs:
  - template_src: my-app.container.j2
podman_secrets:
  - name: my-app-pwd
    data: .....

my-app.container.j2:

[Container]
{% if podman_version is version("4.5", ">=") %}
Secret=my-app-pwd,type=env,target=MYAPP_PASSWORD
HealthCmd=/usr/bin/test -f /path/to/my-app.file
HealthOnFailure=kill
{% else %}
PodmanArgs=--secret=my-app-pwd,type=env,target=MYAPP_PASSWORD
{% endif %}

podman_subuid_info, podman_subgid_info

The role needs to ensure any users and groups are present in the subuid and subgid information. Once it extracts this data, it will be available in podman_subuid_info and podman_subgid_info. These are dicts. The key is the user or group name, and the value is a dict with two fields:

  • start - the start of the id range for that user or group, as an int
  • range - the id range for that user or group, as an int
podman_host_directories:
  "/var/lib/db":
    mode: "0777"
    owner: "{{ 1001 + podman_subuid_info['dbuser']['start'] - 1 }}"
    group: "{{ 1001 + podman_subgid_info['dbgroup']['start'] - 1 }}"

Where 1001 is the uid for user dbuser, and 1001 is the gid for group dbgroup.

NOTE: depending on the namespace used by your containers, you might not be able to use the subuid and subgid information, which comes from getsubids if available, or directly from the files /etc/subuid and /etc/subgid on the host. See podman user namespace modes for more information.

Example Playbooks

Create rootless container with volume mount:

- name: Manage podman containers and services
  hosts: all
  vars:
    podman_create_host_directories: true
    podman_firewall:
      - port: 8080-8081/tcp
        state: enabled
      - port: 12340/tcp
        state: enabled
    podman_selinux_ports:
      - ports: 8080-8081
        setype: http_port_t
    podman_kube_specs:
      - state: started
        run_as_user: dbuser
        run_as_group: dbgroup
        kube_file_content:
          apiVersion: v1
          kind: Pod
          metadata:
            name: db
          spec:
            containers:
              - name: db
                image: quay.io/db/db:stable
                ports:
                  - containerPort: 1234
                    hostPort: 12340
                volumeMounts:
                  - mountPath: /var/lib/db:Z
                    name: db
            volumes:
              - name: db
                hostPath:
                  path: /var/lib/db
      - state: started
        run_as_user: webapp
        run_as_group: webapp
        kube_file_src: /path/to/webapp.yml
  roles:
    - linux-system-roles.podman

Create container running as root with Podman volume:

- name: Manage podman root containers and services
  hosts: all
  vars:
    podman_firewall:
      - port: 8080/tcp
        state: enabled
    podman_kube_specs:
      - state: started
        kube_file_content:
          apiVersion: v1
          kind: Pod
          metadata:
            name: ubi8-httpd
          spec:
            containers:
              - name: ubi8-httpd
                image: registry.access.redhat.com/ubi8/httpd-24
                ports:
                  - containerPort: 8080
                    hostPort: 8080
                volumeMounts:
                  - mountPath: /var/www/html:Z
                    name: ubi8-html
            volumes:
              - name: ubi8-html
                persistentVolumeClaim:
                  claimName: ubi8-html-volume
  roles:
    - linux-system-roles.podman

Create quadlet application with secrets. Defer starting the application until all of the units have been created. Note the order of the files in podman_quadlet_specs are in dependency order. Using podman_create_host_directories: true will create any host mounted directories specified by a Volume= directive in the container spec.

podman_create_host_directories: true
podman_activate_systemd_unit: false
podman_quadlet_specs:
  - name: quadlet-demo
    type: network
    file_content: |
      [Network]
      Subnet=192.168.30.0/24
      Gateway=192.168.30.1
      Label=app=wordpress
  - file_src: quadlet-demo-mysql.volume
  - template_src: quadlet-demo-mysql.container.j2
  - file_src: envoy-proxy-configmap.yml
  - file_src: quadlet-demo.yml
  - file_src: quadlet-demo.kube
    activate_systemd_unit: true
podman_firewall:
  - port: 8000/tcp
    state: enabled
  - port: 9000/tcp
    state: enabled
podman_secrets:
  - name: mysql-root-password-container
    state: present
    skip_existing: true
    data: "{{ root_password_from_vault }}"
  - name: mysql-root-password-kube
    state: present
    skip_existing: true
    data: |
      apiVersion: v1
      data:
        password: "{{ root_password_from_vault | b64encode }}"
      kind: Secret
      metadata:
        name: mysql-root-password-kube
  - name: envoy-certificates
    state: present
    skip_existing: true
    data: |
      apiVersion: v1
      data:
        certificate.key: {{ key_from_vault | b64encode }}
        certificate.pem: {{ cert_from_vault | b64encode }}
      kind: Secret
      metadata:
        name: envoy-certificates

License

MIT.

Author Information

Based on podman-container-systemd by Ilkka Tengvall [email protected].

Authors: Thom Carlin, Rich Megginson, Adam Miller, Valentin Rothberg