16 KiB

Raw Blame History

Ansible Bootstrap Deployment

Storyboard: https://storyboard.openstack.org/#!/story/2004695.

This spec describes the initial phase of StarlingX deployment improvement effort.

Problem description

The primary controller is currently configured using the config_controller Python script which can only be executed on the controller console. The script requires input for many networking aspects upfront in order to run both bootstrap operations and host configuration to completion. Over time, the script logic has grown overly complex to accommodate a plethora of host configuration scenarios and so has increased the configuration time.

Furthermore, once all required input configuration parameters have been successfully validated, the script will run all its steps. If the script fails due to a software issue or a configuration mistake, a re-install will be required. It is not possible for the user to apply a software patch and/or rerun the script to apply updated configurations.

Use Cases

As a developer/tester/operator, I need the ability to configure the controller remotely.
As a developer/tester/operator, I need to the ability to modify and reapply configurations during initial host config.
As a developer/tester/operator, I need the ability to automate the initial host deployment and build out my system from there.
As a developer of StarlingX community, I would like to streamline the initial host config using an industry adopted tool to enable automation and to promote process/code visibility and customization.

Proposed change

Existing workflow with config_controller (high level)

Config_controller:

Create bootstrap hiera config
Apply bootstrap puppet manifest
Persist local configuration
Populate initial system inventory
Create system hiera config
Apply controller puppet manifest
Finalize controller configuration
Activate all services

Host-configuration:

Manual or scripted configurations required for unlock.

Host-unlock:

Apply controller puppet manifest (and worker, storage puppet manifests for All-in-one)
Activate all services

Proposed workflow with Ansible Playbook (high level)

The bootstrap and configuration of the initial host will be orchestrated by an Ansible Playbook¹.

Playbook:

Apply bootstrap puppet manifest
Populate system configuration (with defaults and user-supplied config)
Bring up Kubernetes master node and essential services

Host-configuration:

Manual or scripted configurations required for unlock.

Host-unlock

Apply controller puppet manifest (and worker, storage puppet manifests for All-in-one)
Activate all services

After phase #2 of the Playbook, the host configuration will resemble All-in-one simplex (i.e. defaulting to the loopback interface) until it is unlocked for the first time. Interface configuration is being deferred to ensure the network connection is not interrupted while the playbook is being played. Interface reconfiguration will only take effect on unlock operations. Previously, this would occur as part of the controller manifest apply which has been eliminated.

Scope of the new workflow

The new workflow will cover the initial config for all supported system configurations in a containerized platform.

Bootstrap playbook roles and tasks (high level)

Below is a list of major roles and tasks. The names are deliberately long to make them self-explanatory for review purpose. They can be renamed to be more terse as role variables should be prefixed with role names. During implementation, some roles and tasks will likely be decomposed or combined.

Role: validate-config-input

Task: validate-config

Role: prepare-environment-for-execution

Task: validate-environment
Task: set-environment-variables

Role: cleanup-environment-after-execution

Task: unset-environment-variables
Task: remove-temp-files

Role: store-admin-password

Task: validate-password
Task: store-password

Role: apply-bootstrap-manifest

Task: generate-bootstrap-data
Task: apply-manifest

Role: populate-initial-config

Task: persist-keyring
Task: set-permanent-puppet-workdir
Task: set-permanent-pxe-configdir
Task: set-postgres-config-for-mate
Task: process-branding-and-banner
Task: populate-system-config
Task: populate-load-config
Task: populate-network-config
Task: populate-controller-config
Task: create-loopback-interface
Task: update-local-dns
Task: update-platform-config-file
Task: add-dns-server

Role: bring-up-kubernetes-master-and-dependent-services

Task: bring-up-kubernetes-master
Task: bring-up-tiller
Task: bring-up-fault-management
Task: bring-up-maintenance
Task: bring-up-vim

Playbook directory layout

The directory layout of the playbook initially could be as follows:

bootstrap.yml

roles/

validate-config-input/

tasks/: main.yml
handlers/: main.yml
files/: <scripts, files>
vars/: main.yml
defaults/: main.yml
meta/: main.yml

prepare-environment-for-execution/

cleanup-environment-after-execution/

store-admin-password/

apply-bootstrap-manifest/

popupate-initial-config/

bring-up-Kubernetes-master-and-dependent-services/

Playbook pre_tasks and post_tasks

The pre_tasks and post_tasks can be as simple as marking the start and end of the playbook execution.

Running `bootstrap playbook`

ansible-playbook bootstrap.yml -u <named-account-with-sudo-privileges> [-K -i <config-input-file> -e <list-of-variable-value-pairs-to-overwrite> --ask-vault-password]

The playbook should be run using wrsroot account. However, it can be run using another account with sudo privileges if desired provided that the account has already been setup beforehand. Many playbook tasks must be run as root. The option -K will prompt for privilege escalation password.

Overwriting playbook defaults

The bootstrap playbook will come with default variables and Ansible hosts file /etc/ansible/hosts.yml. These defaults and content of the hosts file are meant for running the playbook locally and bootstrapping the initial controller for All-in-one simplex in virtual box. In practice, some of these defaults will need to be overwritten with user supplied values.

Variables that usually require overwriting are:

host IP (for running the playbook remotely)
system properties
Management, OAM, PXE, cluster subnets
Default DNS server

There are various ways to overwrite variables in Ansible Playbook.

Overwrite with configuration input file

One simple and clean option is to overwrite with -i command line parameter. The content of the provided configuration input file must be in YAML format.

The default hosts (Ansible inventory) file will have the following entries:

bootstrap:

hosts:

local:: ansible_connection: local

vars:

ansible_user: wrsroot ansible_become: true

To overwrite the bootstrap host for remote execution and/or user in the custom configuration input file:

bootstrap:

hosts:

remote:: ansible_host: '128.224.150.83' ansible_connection: ssh

vars:

ansible_user: wrsroot ansible_become: true

To overwrite the role default variables, one option is to add the list of of overwritten variables under vars section of the configuration input file:

vars:

system_mode: duplex-direct dns_server: 8.8.8.8

Overwrite with role vars

Another option to overwrite role defaults is to replace main.yml file under vars directory of the corresponding role(s) with custom one(s) before running the playbook. This takes precedence over the overwriting method above.

Overwrite with extra vars

Command line -e option which has the highest precedence can also be used to overwrite defaults. However, this method can be cumbersome if many defaults need overwriting and the playbook is run manually.

The list of role defaults as well as the preferred method to overwrite these defaults will be documented after the playbook has been developed.

Overwriting sensitive variables

The admin password is a sensitive variable that usually needs to be overwritten. To ensure sensitive information is encrypted, sensitive variables and values are copied to a vault file and secure using ansible-vault encrypt command. The corresponding defaults will need to be mapped to the variables in vaulted file using jinja2 syntax.

The command line argument --ask-vault-pass or --vault-password-file will need to be supplied when running the playbook with encrypted vault file.

For development/test purposes, these variables can simply be overwritten using the command line -e option.

Validating configuration parameters

The config_controller script has extensive logic to validate config parameters in user input file which could be leveraged in validate-config-input role of the bootstrap playbook.

Config_controller script changes

Currently this complex script has multiple uses: a) perform initial configuration required mainly to bring up the controller services, b) backup system configuration, c) restore system configuration from backup file, d) clone the image, and e) restore the system from a clone.

The proposed Ansible bootstrap deployment will replace the initial system configuration aspect of the script. The script will continue to be used for other operations. Relevant code will be removed from the script once the implementation of the playbook is complete.

Puppet changes

The initial bootstrap playbook will leverage the existing Puppet bootstrap.pp manifest to bring up the following services that will be used by the playbook for the remaining tasks:

Required services to bring up Kubernetes master:

docker
etcd

Required services for host unlock:

fm
mtcAgent
nfv-vim

The puppet .pp and in some cases .py files related to these services and Kubernetes will require update.

Sysinv changes

Traditionally, the config_controller script is provided with all required parameters either interactively or via a config file to perform both bootstrap operations and host configuration. Networking and storage provisioning using system commands beyond this point have certain restrictions as the controller manifest has been applied.

With Ansible bootstrap deployment method, some system commands will require changes to support manual configuration adjustments and replays of the bootstrap playbook. The cgtsclient will also need minor modification to avoid requesting for smapi endpoint which is not yet available in this early stage.

Maintenance changes

Some minor tweaks to maintenance code will be required for maintenance Client and Agent to operate properly during the bootstrap phase.

Packaging of `bootstrap playbook` in the ISO and SDK

The playbook will be packaged in the ISO as well as SDK to allow both local and remote execution.

Alternatives

Additional host configuration roles to support the initial host-unlock were considered. However, this would add much of the complex modeling of input configuration (i.e. more upfront planning) to the intial deployment step.

Data model impact

No impact to existing system inventory data model.

REST API impact

At this time, no REST API impact is anticipated.

Security impact

The proposal is to make use of Ansible Playbook which is a well adopted multi-node configuration and deployment orchestration tool partly due to Ansible secure architecture and design.

The scope of the proposed bootstrap playbook is limited to bringing the initial controller to the state where it can be unlocked and allow other Kubernetes nodes on an internal cluster network if configured to join.

The Playbook can only be executed remotely over SSH using a named account with sudo privileges. Ansible vault will be used to store secrets/private information where applicable. As such, no additional security impact is introduced.

Other end user impact

The user will be expected to interact with the feature using ansible-playbook² and ansible-vault³ commands. The bootstrap deployment method will give the user more flexibility to customize and automate the deployment.

Once the initial controller is ready to accept system commands and Kubernetes master is up, the user can: * perform minimum host configurations and unlock the host * join other Kubernetes nodes and perform more extensive custom configurations before the unlock

The playbook can be replayed to update system properties and general networking information. It will not be playable after the host is unlocked.

Performance Impact

Ansible execution overhead is unknown at this time. However, as the controller manifest application and services activation steps are deferred till host-unlock, the time to bring the controller to unlock-ready state should be significantly faster than with the traditional method.

Other deployer impact

None

Developer impact

See end user impact.

The developers can extend the bootstrap playbook with custom host configuration role(s) or another playbook to suit their specific needs.

Upgrade impact

None as this is the initial release of Bootstrap Deployment using Ansible Playbook.

Implementation

Assignee(s)

Primary assignee:

Tee Ngo (teewrs)

Other contributors:

Eric McDonald (emacdona)

Repos Impacted

stx-config
stx-metal
stx-root
stx-docs

Work Items

Modify maintenance to enable maintenance operations during bootstrap phase.
Modify sysinv and cgtsclient to be more flexible with configuration updates during bootstrap deployment using either system commands or APIs.
Modify puppet classes and python scripts to allow launching a limited number of services required for bootstrap operations and initial host unlock.
Create a bootstrap Playbook to bring up Kubernetes master node and configure the primary controller based on default and user-supplied config parameters.
Package the Playbook as part of the ISO & SDK to allow both on premise and remote execution.
Make other necessary changes to support primary controller configuration using either the playbook or traditional config_controller until the transition is complete. This includes lab setup tool changes.

Dependencies

config_controller script
Ansible⁴
Containerized OpenStack based deployment

Testing

This story changes the way StarlingX system is deployed, specifically how the primary controller is configured, which will require changes in existing automated installation and lab setup tools.

The system deployment tests will be limited to All-in-one simplex, All-in-one duplex, and Standard configurations. Deployment tests for Region and Distributed Cloud configurations are deferred until the support for these configurations in a containerized OpenStack based platform is available. At which point, either the bootstrap playbook will be extended with additional roles or with new playbook(s) to process steps in config_region and config_subcloud. This will be documented either in a later version of this spec or in a separate spec.

Documentation Impact

This story affects the StarlingX installation and configuration documentation. Specific details of the documentation changes will be addressed once the implementation is complete.

References

History

Revisions
Release Name	Description
TBD	Introduced

16 KiB Raw Blame History