Skip to content

Experimental work on Ansible GitOps Framework

License

Notifications You must be signed in to change notification settings

validatedpatterns/agof

Repository files navigation

License

Ansible GitOps Framework (AGOF)

This repository is part of Red Hat's Validated Patterns effort, which in general aims to provide Reference Architectures that can be run through CI/CD systems.

The goal of this project in particular is to provide an extensible framework to do GitOps with Ansible Automation Platform, and as such to provide useful facilities for developing Patterns (community and validated) that function with Ansible Automation Platform as the GitOps engine.

The thinking behind this effort is documented in the Ansible Pattern Theory repository.

1. How to Use It

The default installation will provide an AAP 2.4 installation deployed via the Containerized Installer, with services deployed this way:

URL Pattern Service
https://aapnode.fqdn:8443/ Controller API
https://aapnode.fqdn:8444/ Private Automation Hub
https://aapnode.fqdn:8445/ EDA Automation Controller

By default, the framework will apply license content specified by the manifest_content variable, but will not further configure Controller or Automation Hub beyond the defaults.

From there, a minimal example pattern is available to download and run here. To use this example, set the following variables in your agof_vault.yml. agof_statedir is where the config repo will be checked out by the process. Any repo that can be used with the controller_configuration collection can be used as the agof_iac_repo.

agof_statedir: "{{ '~/agof' | expanduser }}"
agof_iac_repo: "https://github.com/validatedpatterns-demos/agof_minimal_config.git"

1.1. Installation

./pattern.sh make install

This builds the default pattern configuration on AWS, which (by default) includes a containerized install of AAP 2.4 on a single AWS VM. Various add-ons can be included by adding variables to the ~/agof_vault.yml file as described below - these options will all be honored as the pattern installs itself.

1.2. Uninstallation

./pattern.sh make aws_uninstall

This destroys all of the AWS infrastructure built for the pattern - starting with the VPC it creates for the pattern and all of the resources attached to it.

2. Entry Points

This is a framework for building Validated Patterns that use Ansible Automation Platform (AAP) as their underlying GitOps engine. To that end, the framework has these deployment models (in increasing level of complexity):

3. Using the Containerized Installer vs. The traditional one

As of AAP 2.4, the containerized installer feature is in Tech Preview, but because of certain advantages it has over the traditional installer, AGOF uses it by default.

3.1. "API" Install (aka "Bare")

./pattern.sh make api_install

In this model, you provide an (already provisioned) AAP endpoint. It does not need to be entitled, it just needs to be running the AAP Controller. You supply the manifest contents, endpoint hostname, admin username (defaults to "admin"), and admin password, and then the installation hands off to a agof_controller_config_dir you define. This is provided for users who have their own AAP installations on bare metal or on-prem or do not want to run on AWS. It is also useful in situations where the AAP deployment topology is more complex than what we provide in the pattern provisioner.

3.2. Legacy "From OS" Install

./pattern.sh make legacy_from_os_install INVENTORY=(your_inventory_file)

In this model, you provide an inventory file with up to two fresh RHEL installations. The model is tested with one AAP and one Hub instance. (Many other topologies are possible with the AAP installation framework; see the Ansible Automation Platform Planning and Installation guides for details.) If you need to install a pattern on a cluster with a different topology than this, use the API install mechanism. This mechanism will run a (slightly) opinionated install of the AAP and Hub components, and will add some conveniences like default execution environments and credentials. Like the "API" install, the install will then be handed over to a agof_controller_config_dir you define.

The reason for making this a separate option is to make it easy for those who are not used to installing AAP to get up and running with it given a couple of VMs (or baremetal instances). Requirements for this mode are as follows:

  • Must be running a version of RHEL that AAP supports
  • Must be properly entitled with a subscription that makes the appropriate AAP repository available

It is not possible to test all possible scenarios in this mode, and we do not try.

Note that INVENTORY defaults to '~/inventory_agof' if you do not specify one.

Your inventory must define an aap_controllers group (which will be configured as the AAP node) and an automation_hub group which will be configured as the automation hub, if you want one. You must also specify username if you want it to be something besides the default 'ec2-user' (which it does not create or otherwise manage). Similarly, you should set controller_hostname (including the 'https://' schema) because that will default to something AWS-specific otherwise. As with all Ansible inventory files, you can set other variables here and the plays will use them. (Other variables of interest might be aap_version, for example.)

Example ~/agof_inventory (for just AAP, which is the default):

[build_control]
localhost

[aap_controllers]
192.168.5.207

[automation_hub]

[eda_controllers]

[aap_controllers:vars]

[automation_hub:vars]

[all:vars]
ansible_user=myuser
ansible_ssh_pass=mypass
ansible_become_pass=mypass
ansible_remote_tmp=/tmp/.ansible
username=myuser
controller_hostname=192.168.5.207

Example ~/agof_inventory (including both AAP and Hub):

[build_control]
localhost

[aap_controllers]
192.168.5.207

[automation_hub]
192.168.5.209

[eda_controllers]

[aap_controllers:vars]

[automation_hub:vars]

[all:vars]
ansible_user=myuser
ansible_ssh_pass=mypass
ansible_become_pass=mypass
ansible_remote_tmp=/tmp/.ansible
username=myuser
controller_hostname=192.168.5.207

3.3. Default Install

./pattern.sh make install

In this model, you provide AWS credentials in addition to the other components needed in the "bare" install. The framework will build an AWS image using Red Hat's ImageBuilder, deploy that image onto a new AWS VPC and subnet, and deploy AAP on that image using the command line installer. It will then hand over the configuration of the AAP installation to the specified agof_controller_config_dir.

3.4. Convenience Features Install (defined by options in the "default" install)

./pattern.sh make install

This model builds on the Default installation by adding the options for building extra VMs besides AAP - the configured VM options include Private Automation Hub (hub), Identity Management (idm), and Satellite (satellite). The framework will configure Automation Hub in this mode, and configure AAP to use it; it will not configure idm or satellite beyond what they need to be managed by AAP. (The configuration of idm and satellite is the subject of the first pattern to use this framework.)

The variables to include builds for the extra components are all Ansible booleans that can be included in your configuration:

automation_hub eda build_idm build_sat

4. agof_vault.yml Configuration

4.1. Common Configuration

Name Description Required Default Notes
admin_user Admin User (for AAP and/or Hub) false 'admin'
admin_password Admin Password (for AAP and/or Hub) true
aap_verison AAP Version to Use true '2.4' Can also be '2.3' currently
containerized_install Flag to direct containerized install true true TP in AAP 2.4; Expected to become AAP's default in AAP 2.5
redhat_username Red Hat Subscriber Username true
redhat_password Red Hat Subscriber Password true
redhat_registry_username_vault Red Hat Subscriber Username true
redhat_registry_password_vault Red Hat Subscriber Password true
manifest_content Base64 encoded Manifest to Entitle true Can be loaded directly from a file using a construct like this: `"{{ lookup('file', '~/Downloads/manifest.zip')
automation_hub_certified_url URL for Certified Content false https://console.redhat.com/api/automation-hub/content/published/ This refers to the automation hub section on https://console.redhat.com. It is the endpoint that is used to download Validated Content in addition to any public Galaxy content needed
automation_hub_validated_url URL for Validated Content false https://console.redhat.com/api/automation-hub/content/validated/ This refers to the automation hub section on https://console.redhat.com. It is the endpoint that is used to download Certified Content in addition to any public Galaxy content needed
automation_hub_token_vault Subscriber-specific token for Content true
automation_hub Flag to build and enable Automation Hub false false
eda Flag to build and enable Event Driven Automation controller false true
agof_controller_config_dir Directory to pass to controller_configuration false This directory is the key one to load all other AAP Controller configuration. The framework will populate it by checking out the agof_iac_repo version agof_iac_repo_version (default: main)
controller_launch_jobs List of jobs to run after controller_configuration has run false Use this to start a job (or jobs) that do not have aggressive schedules, and that are ready to run as soon as the controller is configured. The fewer jobs listed here the better.

4.2. Initialization Environment Configuration

Name Description Required Default Notes
init_env_collection_install Whether to install collections required by the framework false true
init_env_collection_install_force Whether to use the force argument when installing collections false false Forces the installation of declared dependencies if true
special_collection_installs "Bundled" collection installations (references files in repodir) false [] A mechanism to allow the installation of collections bundled into the pattern, if the ones published in galaxy and/or Automation Hub are not sufficient
offline_token Red Hat Offline Token false Used to build the imagebuilder image
offline_token Red Hat Offline Token false Used to build the imagebuilder image

4.3. Automation Hub Specific Configuration

Name Description Required Default Notes
aap_admin_username Admin User (for AAP) false '{{ admin_user }}'
aap_admin_password Admin Password (for AAP) true '{{ admin_password }} '
private_hub_username Admin User (for Automation Hub) true '{{ admin_user }}'
private_hub_password Admin Password (for Automation Hub) true '{{ admin_password }} '
custom_execution_environments Array of Execution Environments to build on Hub true [] The execution environments will be built on the hub node, and then pushed into the hub registry from there.

4.4. AWS-Specific Configuration

Name Description Required Default Notes
aws_account_nbr_vault AWS Account Number false The AWS Account Number is used by ImageBuilder to share the resulting image to as an AMI
aws_access_key_vault AWS Access Key String false The AWS Access Key - can be included if other pattern elements cannot read AWS credentials directly for subsequent pattern use
aws_secret_key_vault AWS Secret Key String false The AWS Secret Key - like the access key, can be added if other pattern elements need it
ec2_region EC2 region to use for builds false us-east-1 is a reasonable value to use because this is where imagebuilder puts images by default
ec2_name_prefix Text to add for EC2 false This is a name to disambiguate your pattern from anything else that might be running in your AWS account. A VPC, subnet, and security group is built from this, and the prefix is added to the pattern_dns_zone by default. Additionally, the SSH private key is stored locally in ~/{{ ec2_name_prefix }}.
pattern_dns_zone Zone to use for route53 updates false Definitely set this if doing DNS updates
build_idm Flag to build an idm VM on AWS false false This is a flag to indicate whether to build a VM called idm on AWS, for later installation of the idm software. Note that the framework just instantiates the VM, it does not install the IDM software.
build_sat Flag to build a Satellite VM on AWS false false This is a flag to indicate whether to install a Satellite VM on AWS, for later installation of the Satellite software. The framework does not include the Ansible code to install Satellite itself.
ec2_instances_xtra Dictionary of additional ec2_instances to build false Build additional VMs as part of the pattern

4.5. ImageBuilder-Specific Configuration

Note: If you are providing an AMI via the imagebuilder_ami variable as opposed to building one from console.redhat.com for the pattern, the pattern still assumes that the RHEL instance will be entitled and will be running a suitable version of RHEL.

Name Description Required Default Notes
org_number_vault Red Hat Subscriber Organization Number true This is the organization number associated with the RHEL instances you need to entitle
activation_key_vault Activation Key Name to embed in image true This is an activation key for the Red Hat CDN. It is expected to be able to enable both the base RHEL repos and the AAP repos.
skip_imagebuilder_build Flag to skip imagebuilder build (also set imagebuilder_ami if true) false false
imagebuilder_ami AMI to use for VM creation in AWS false It is very possible to re-use another imagebuilder build from a previous installation of the pattern framework, and saves ~15 minutes on a new pattern install to re-use such an image.
ami_source_region Source Region to copy AMI from if not present in target region false us-east-1 Imagebuilder puts its images in us-east-1 by default. If you specify a non-imagebuilder ami and it "starts out" in a different region, you can specify that here.

5. What the Framework Does, Step-by-Step

5.1. Pre-GitOps Steps

5.1.1. Pre-init (mandatory)

Pre-initialization, for our purposes, refers to things that have to be installed on the installation workstation/provisioner node to set up the Ansible environment to deploy the rest of the pattern.

Build ansible.cfg

Because ansible.cfg needs to be configured with the automation hub endpooint, we generate it from a template. The ansible.cfg includes endpoints for both public and Automation Hub galaxy servers and a vault password path (for ansible-vault). The template used is here. The automation hub token is read from the vault and injected into the environment for the collection installation phase.

Collection dependency install

The next thing the pre-init play does is install dependency collections based on what's contained in requirements.yml. If there are any "special" collections contained in the root of the framework they can also be installed at this time. When the framework was under development this was necessary with redhat.satellite but the necessary changes have been upstreamed as of 3.11; 3.12 has been released since then, so this mechanism is no longer needed but is left in as it might be necessary in the future.

5.1.2. Environment Initialization (optional; enabled by default; make install entry point)

Environment initialization includes the steps necessary to build the environment (VMs) to install AAP and related tooling; currently the framework can do this on AWS. This step is optional if you wish to provide either an AAP controller API endpoint OR else bring your own inventory/VMs to install AAP Controller and (optionally) Automation Hub on.

This play builds an image using the Red Hat Console's imagebuilder service; the end result of this process is an image that will serve as an AMI in AWS. (ImageBuilder can build other types of images as well.) The key aspects of this image are that they include the cloud-init package (which helps with certain aspects of initialization), but more importantly, they include the organization number and activation key so that images that are instantiated with this AMI are automatically registered and enabled to install content via the Red Hat CDN.

These images tend to be fairly static, so it is not necessary to build a brand new image every time you run the pattern. You can save the AMI from a previous pattern run, and re-use it as long as you like. Building and uploading the image to AWS represents about 15 minutes of the runtime of the pattern installation.

This play handles all the AWS-specific setup necessary to run AAP, (optionally) Automation Hub, (optionally) IdM, (optionally) Satellite, and also offers you the ability to install VMs of your own for a pattern via overrides. The reason for this is that installing the VMs at this stage allows the customer/user to not have to involve AWS credentials in the pattern itself if they do not wish to do so. Of course they are free to include workflows that interact with AWS or other infrastructure if they want - this was done with the intention of simplifying that process if AWS was an implementation detail of the pattern as opposed to the intention of it.

The two key roles that are invoked here are manage_ec2_infra and manage_ec2_instances. Both of these have been adapted from ansible-workshops. The manage_ec2_infra role is responsible for setting up the VPC, subnet, and security group for the patten. manage_ec2_instances is responsible for actually building the VM instances. It will also manage route53 DNS entries. It is safe to run these roles repeatedly; they will not "double allocate" VMs as long as the VMs are running.

These roles also include code for making hostnames durable across reboots, as well as maintaining /etc/hosts on all AWS nodes in the bootstrap set that include all of the other servers in the bootstrap set.

The standard public IPs that are offered by AWS (which are used by this framework) are not permanently associated with the VMs. When the VMs cold start, in particular, we can expect IP change events. The VMs themselves are not directly aware of their external IPs. In order to handle this situation, this play can be run from the provisioner node/workstation to update the route53 DNS mappings based on the AWS API.

Teardown AWS Environment) (make aws_uninstall)

The teardown play will terminate all VMs associated with a VPC and subnet, and remove the DNS entries managed by the "bootstrap set" (which are just the VMs the framework has been told it has to build).

5.1.3. Containerized Install (make install default)

Name Description Required Default Notes
containerized_install Whether to use the containerized installer true true
containerized_installer_user Unprivileged user to create to run AAP true aap
containerized_installer_user_home Directory to install containerized AAP into true /var/lib/{{ containerized_installer_user }}
containerized_installer_shasum SHA256 sum of the installer image to download true 714de904596f2f3312b8804eee0d2fdce6842ac2108dbf718f2107ef61d1598d This currently points to the image for AAP 2.4
automation_hub Boolean to indicate whether to install automation_hub true true
postgresql_admin_username Name of postgres admin for services true postgres
postgresql_admin_password Name of postgres admin for services true {{ db_password }}
ee_extra_images Specifications for extra execution environments to load false []
de_extra_images Specifications for extra decision environments to load (for EDA) false []
controller_postinstall Whether or not to use controller_postinstall feature false false
controller_postinstall_repo_url Repository URL for additional controller configuration false
controller_postinstall_repo_ref Tag, branch or SHA in the repo to use for configuration false main
controller_postinstall_dir Directory to use to load additional controller config false Prefer using the controller_postinstall_repo_url as it is more flexible
hub_admin_password Admin user password for automation_hub service false '{{ admin_password }}'
hub_pg_password Postgres password for automation_hub service false '{{ db_password }}'
hub_postinstall Whether to use the hub_postinstall feature false false
hub_postinstall_dir Directory to use for hub_postinstall if enabled false Prefer hub_postinstall_repo options as they are more flexible
hub_postinstall_repo_url Repository to use for hub_postinstall if enabled false
hub_postinstall_repo_ref Branch, tag, or sha to use in repo for hub postinstall false main
hub_workers How many worker containers to run false 2
hub_collection_signing Whether to sign collections automatically false false
hub_collection_signing_key GPG key to sign collections with, if desired false
hub_container_signing Whether to sign containers automatically false false
hub_container_signing_key GPG key to sign containers (EE's and DE's) with, if desired false
eda_admin_password EDA Service admin password false {{ admin_password }}
eda_pg_password EDA Service postgres password false {{ db_password }}
eda_workers Number of EDA workers false 2
eda_activation_workers Number of EDA activation workers false 2

The containerized installer is a relatively new feature for AAP, and because it makes different assumptions and has very different options from the traditional installer, it has its own path in AGOF, which bypasses the traditional installer steps.

First, we run the Installer Prereqs role. This sets up a user (default: aap) to run AAP as, sets up password-less sudo for that user, and installs the pattern user's ~/.ssh/id_rsa.pub as an authorized_key for the user. The sudo rule simplifies the process of running the containerized installer, and by default is removed in the cleanup stage. This role also sets linger on the AAP user so that services will start at boot time under the AAP user.

Next, we run the Containerized Installer role. This runs the containerized installer as the designated user. This user cannot be root (because of how it uses and configures containerized services). This role downloads and executes the installer, and also places a manifest file (designated by the manifest_content variable) as manifest.zip to entitle the controller. It runs the containerized installer.

Finally, we run the cleanup role if it is enabled (which it is by default). This removed the sudo rule from the AAP user, and removes the manifest.zip file. The installation is now ready for you to use.

5.1.4. Host Configuration (make legacy_from_os_install entry point)

This is normally run inline with the make install entry point, when the pattern frame is not using the containerized installer.. You may, at your option, provide a suitable inventory file that specifies a node to install AAP on, and optionally a node to install Automation Hub on, and pass that inventory file in, as desribed here.

The common role here prepares AAP (on both the controller node, as well as the automation_hub node, if one is configured to be installed).

AAP Installation (mandatory for non-containerized install)

The control node configuration role templates an inventory file for the AAP installer and runs it. Additionally, this role will entitle the AAP controller, as well as loading execution environments from the Red Hat registry. The bulk of controller configuration, however, is designed to happen later, when control is handed to the controller_configuration dispatch role.

This is one of the few elements in the pattern that is mandatory. (Technically, if you use the "API only" installation mode, this part is also optional.)

This role configures an instance of the AAP Private Automation Hub, if the automation_hub variable is set to true. Private Automation Hub serves as a content repository for Ansible content; in particular, it serves as a container registry for Ansible Execution Environments; since one of the main reasons to use Execution Environments is to encapsulate Validated Content, this provides a mechanism for the pattern to facilitate the use of Validated Content in a safe and subscription agreement-compliant manner.

You will need a Private Automation Hub if your pattern builds an Execution Environment with Validated Content. You may also want one for other reasons. The automation_hub variable defaults to false.

If you specify custom execution environments to be built, they will be built on the Hub node and pushed to the registry on the Hub node.

AAP Configuration (mandatory for non-containerized install; make api_install entry point)

This play is really the heart and focus of this framework. The rest of the framework exists to facilitate running this play, and providing additional capabilities to the environment in which the play runs. The logic for the play is contained in the configure_aap role. The make api_install entry point uses just the variables related to controller installation as desribed here. Otherwise, it is called inline from the make install entry point. It is safe to run multiple times.

Entitle Controller

The play uses the same technique and variables to entitle the AAP controller as the role above does; but since we want to preserve the option to call it outside that role, it is duplicated. The key thing is to populate manifest_contents with a suitable manifest file.

Configure Controller

Using the credentials you supply, the play will now proceed to invoke the controller_configuration dispatch role on the controller endpoint based on the configuration found in controller_configuration_dir and in your agof_vault.yml file. This controls all of the controller configuration. Based on the variables defined in the controller configuration, dispatch will call the necessary roles and modules in the right order to configure AAP to run your pattern.

Run "immediate" jobs

The final step in the play allows you to run job(s) immediately. This mechanism is provided in case you have a potentially long-running play for configuration that is scheduled to run repeatedly but at a long interval. This ensures the play (job) runs immediately.

5.2. GitOps Step

The framework is only truly in GitOps mode once the configuration has been fully applied to the AAP controller and the AAP controller has taken responsibility for managing the environment. Prior to that point, there are many opportunities (by design) to inject non-declarative elements into the environment. Beyond that point, changes should be made to the environment or the workflows that configure it by pushing git commits to the repositories the pattern uses.

6. Acknowledgements

This repository represents an interpretation of GitOps principles, as developed in the Hybrid Cloud Patterns GitOps framework for Kubernetes, and an adaptation and fusion of two previous ongoing efforts at Red Hat: Ansible-Workshops and LabBuilder2/RHISbuilder.

The AWS interactions come from Ansible Workshops (with some changes); the IDM and Satellite build components come from LabBuilder2/RHISbuilder. I would also like to specifically thank Mike Savage and Paul Armstrong, co-workers of mine at Red Hat, who have provided encouragement and advice on how to proceed with various techniques in Ansible in particular.

About

Experimental work on Ansible GitOps Framework

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published