diff options
Diffstat (limited to 'docs')
| -rw-r--r-- | docs/proposals/role_decomposition.md | 353 |
1 files changed, 353 insertions, 0 deletions
diff --git a/docs/proposals/role_decomposition.md b/docs/proposals/role_decomposition.md new file mode 100644 index 000000000..b6c1d8c5b --- /dev/null +++ b/docs/proposals/role_decomposition.md @@ -0,0 +1,353 @@ +# Scaffolding for decomposing large roles + +## Why? + +Currently we have roles that are very large and encompass a lot of different +components. This makes for a lot of logic required within the role, can +create complex conditionals, and increases the learning curve for the role. + +## How? + +Creating a guide on how to approach breaking up a large role into smaller, +component based, roles. Also describe how to develop new roles, to avoid creating +large roles. + +## Proposal + +Create a new guide or append to the current contributing guide a process for +identifying large roles that can be split up, and how to compose smaller roles +going forward. + +### Large roles + +A role should be considered for decomposition if it: + +1) Configures/installs more than one product. +1) Can configure multiple variations of the same product that can live +side by side. +1) Has different entry points for upgrading and installing a product + +Large roles<sup>1</sup> should be responsible for: +> 1 or composing playbooks + +1) Composing smaller roles to provide a full solution such as an Openshift Master +1) Ensuring that smaller roles are called in the correct order if necessary +1) Calling smaller roles with their required variables +1) Performing prerequisite tasks that small roles may depend on being in place +(openshift_logging certificate generation for example) + +### Small roles + +A small role should be able to: + +1) Be deployed independently of other products (this is different than requiring +being installed after other base components such as OCP) +1) Be self contained and able to determine facts that it requires to complete +1) Fail fast when facts it requires are not available or are invalid +1) "Make it so" based on provided variables and anything that may be required +as part of doing such (this should include data migrations) +1) Have a minimal set of dependencies in meta/main.yml, just enough to do its job + +### Example using decomposition of openshift_logging + +The `openshift_logging` role was created as a port from the deployer image for +the `3.5` deliverable. It was a large role that created the service accounts, +configmaps, secrets, routes, and deployment configs/daemonset required for each +of its different components (Fluentd, Kibana, Curator, Elasticsearch). + +It was possible to configure any of the components independently of one another, +up to a point. However, it was an all of nothing installation and there was a +need from customers to be able to do things like just deploy Fluentd. + +Also being able to support multiple versions of configuration files would become +increasingly messy with a large role. Especially if the components had changes +at different intervals. + +#### Folding of responsibility + +There was a duplicate of work within the installation of three of the four logging +components where there was a possibility to deploy both an 'operations' and +'non-operations' cluster side-by-side. The first step was to collapse that +duplicate work into a single path and allow a variable to be provided to +configure such that either possibility could be created. + +#### Consolidation of responsibility + +The generation of OCP objects required for each component were being created in +the same task file, all Service Accounts were created at the same time, all secrets, +configmaps, etc. The only components that were not generated at the same time were +the deployment configs and the daemonset. The second step was to make the small +roles self contained and generate their own required objects. + +#### Consideration for prerequisites + +Currently the Aggregated Logging stack generates its own certificates as it has +some requirements that prevent it from utilizing the OCP cert generation service. +In order to make sure that all components were able to trust one another as they +did previously, until the cert generation service can be used, the certificate +generation is being handled within the top level `openshift_logging` role and +providing the location of the generated certificates to the individual roles. + +#### Snippets + +[openshift_logging/tasks/install_logging.yaml](https://github.com/ewolinetz/openshift-ansible/blob/logging_component_subroles/roles/openshift_logging/tasks/install_logging.yaml) +```yaml +- name: Gather OpenShift Logging Facts + openshift_logging_facts: + oc_bin: "{{openshift.common.client_binary}}" + openshift_logging_namespace: "{{openshift_logging_namespace}}" + +- name: Set logging project + oc_project: + state: present + name: "{{ openshift_logging_namespace }}" + +- name: Create logging cert directory + file: + path: "{{ openshift.common.config_base }}/logging" + state: directory + mode: 0755 + changed_when: False + check_mode: no + +- include: generate_certs.yaml + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + +## Elasticsearch +- include_role: + name: openshift_logging_elasticsearch + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + +- include_role: + name: openshift_logging_elasticsearch + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + openshift_logging_es_ops_deployment: true + when: + - openshift_logging_use_ops | bool + + +## Kibana +- include_role: + name: openshift_logging_kibana + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + openshift_logging_kibana_namespace: "{{ openshift_logging_namespace }}" + openshift_logging_kibana_master_url: "{{ openshift_logging_master_url }}" + openshift_logging_kibana_master_public_url: "{{ openshift_logging_master_public_url }}" + openshift_logging_kibana_image_prefix: "{{ openshift_logging_image_prefix }}" + openshift_logging_kibana_image_version: "{{ openshift_logging_image_version }}" + openshift_logging_kibana_replicas: "{{ openshift_logging_kibana_replica_count }}" + openshift_logging_kibana_es_host: "{{ openshift_logging_es_host }}" + openshift_logging_kibana_es_port: "{{ openshift_logging_es_port }}" + openshift_logging_kibana_image_pull_secret: "{{ openshift_logging_image_pull_secret }}" + +- include_role: + name: openshift_logging_kibana + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + openshift_logging_kibana_ops_deployment: true + openshift_logging_kibana_namespace: "{{ openshift_logging_namespace }}" + openshift_logging_kibana_master_url: "{{ openshift_logging_master_url }}" + openshift_logging_kibana_master_public_url: "{{ openshift_logging_master_public_url }}" + openshift_logging_kibana_image_prefix: "{{ openshift_logging_image_prefix }}" + openshift_logging_kibana_image_version: "{{ openshift_logging_image_version }}" + openshift_logging_kibana_image_pull_secret: "{{ openshift_logging_image_pull_secret }}" + openshift_logging_kibana_es_host: "{{ openshift_logging_es_ops_host }}" + openshift_logging_kibana_es_port: "{{ openshift_logging_es_ops_port }}" + openshift_logging_kibana_nodeselector: "{{ openshift_logging_kibana_ops_nodeselector }}" + openshift_logging_kibana_cpu_limit: "{{ openshift_logging_kibana_ops_cpu_limit }}" + openshift_logging_kibana_memory_limit: "{{ openshift_logging_kibana_ops_memory_limit }}" + openshift_logging_kibana_hostname: "{{ openshift_logging_kibana_ops_hostname }}" + openshift_logging_kibana_replicas: "{{ openshift_logging_kibana_ops_replica_count }}" + openshift_logging_kibana_proxy_debug: "{{ openshift_logging_kibana_ops_proxy_debug }}" + openshift_logging_kibana_proxy_cpu_limit: "{{ openshift_logging_kibana_ops_proxy_cpu_limit }}" + openshift_logging_kibana_proxy_memory_limit: "{{ openshift_logging_kibana_ops_proxy_memory_limit }}" + openshift_logging_kibana_cert: "{{ openshift_logging_kibana_ops_cert }}" + openshift_logging_kibana_key: "{{ openshift_logging_kibana_ops_key }}" + openshift_logging_kibana_ca: "{{ openshift_logging_kibana_ops_ca}}" + when: + - openshift_logging_use_ops | bool + + +## Curator +- include_role: + name: openshift_logging_curator + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + openshift_logging_curator_namespace: "{{ openshift_logging_namespace }}" + openshift_logging_curator_master_url: "{{ openshift_logging_master_url }}" + openshift_logging_curator_image_prefix: "{{ openshift_logging_image_prefix }}" + openshift_logging_curator_image_version: "{{ openshift_logging_image_version }}" + openshift_logging_curator_image_pull_secret: "{{ openshift_logging_image_pull_secret }}" + +- include_role: + name: openshift_logging_curator + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + openshift_logging_curator_ops_deployment: true + openshift_logging_curator_namespace: "{{ openshift_logging_namespace }}" + openshift_logging_curator_master_url: "{{ openshift_logging_master_url }}" + openshift_logging_curator_image_prefix: "{{ openshift_logging_image_prefix }}" + openshift_logging_curator_image_version: "{{ openshift_logging_image_version }}" + openshift_logging_curator_image_pull_secret: "{{ openshift_logging_image_pull_secret }}" + openshift_logging_curator_cpu_limit: "{{ openshift_logging_curator_ops_cpu_limit }}" + openshift_logging_curator_memory_limit: "{{ openshift_logging_curator_ops_memory_limit }}" + openshift_logging_curator_nodeselector: "{{ openshift_logging_curator_ops_nodeselector }}" + when: + - openshift_logging_use_ops | bool + + +## Fluentd +- include_role: + name: openshift_logging_fluentd + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + +- include: update_master_config.yaml +``` + +[openshift_logging_elasticsearch/meta/main.yaml](https://github.com/ewolinetz/openshift-ansible/blob/logging_component_subroles/roles/openshift_logging_elasticsearch/meta/main.yaml) +```yaml +--- +galaxy_info: + author: OpenShift Red Hat + description: OpenShift Aggregated Logging Elasticsearch Component + company: Red Hat, Inc. + license: Apache License, Version 2.0 + min_ansible_version: 2.2 + platforms: + - name: EL + versions: + - 7 + categories: + - cloud +dependencies: +- role: lib_openshift +``` + +[openshift_logging/meta/main.yaml](https://github.com/ewolinetz/openshift-ansible/blob/logging_component_subroles/roles/openshift_logging/meta/main.yaml) +```yaml +--- +galaxy_info: + author: OpenShift Red Hat + description: OpenShift Aggregated Logging + company: Red Hat, Inc. + license: Apache License, Version 2.0 + min_ansible_version: 2.2 + platforms: + - name: EL + versions: + - 7 + categories: + - cloud +dependencies: +- role: lib_openshift +- role: openshift_facts +``` + +[openshift_logging/tasks/install_support.yaml - old](https://github.com/openshift/openshift-ansible/blob/master/roles/openshift_logging/tasks/install_support.yaml) +```yaml +--- +# This is the base configuration for installing the other components +- name: Check for logging project already exists + command: > + {{ openshift.common.client_binary }} --config={{ mktemp.stdout }}/admin.kubeconfig get project {{openshift_logging_namespace}} --no-headers + register: logging_project_result + ignore_errors: yes + when: not ansible_check_mode + changed_when: no + +- name: "Create logging project" + command: > + {{ openshift.common.admin_binary }} --config={{ mktemp.stdout }}/admin.kubeconfig new-project {{openshift_logging_namespace}} + when: not ansible_check_mode and "not found" in logging_project_result.stderr + +- name: Create logging cert directory + file: path={{openshift.common.config_base}}/logging state=directory mode=0755 + changed_when: False + check_mode: no + +- include: generate_certs.yaml + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + +- name: Create temp directory for all our templates + file: path={{mktemp.stdout}}/templates state=directory mode=0755 + changed_when: False + check_mode: no + +- include: generate_secrets.yaml + vars: + generated_certs_dir: "{{openshift.common.config_base}}/logging" + +- include: generate_configmaps.yaml + +- include: generate_services.yaml + +- name: Generate kibana-proxy oauth client + template: src=oauth-client.j2 dest={{mktemp.stdout}}/templates/oauth-client.yaml + vars: + secret: "{{oauth_secret}}" + when: oauth_secret is defined + check_mode: no + changed_when: no + +- include: generate_clusterroles.yaml + +- include: generate_rolebindings.yaml + +- include: generate_clusterrolebindings.yaml + +- include: generate_serviceaccounts.yaml + +- include: generate_routes.yaml +``` + +# Limitations + +There will always be exceptions for some of these rules, however the majority of +roles should be able to fall within these guidelines. + +# Additional considerations + +## Playbooks including playbooks +In some circumstances it does not make sense to have a composing role but instead +a playbook would be best for orchestrating the role flow. Decisions made regarding +playbooks including playbooks will need to be taken into consideration as part of +defining this process. +Ref: (link to rteague's presentation?) + +## Role dependencies +We want to make sure that our roles do not have any extra or unnecessary dependencies +in meta/main.yml without: + +1. Proposing the inclusion in a team meeting or as part of the PR review and getting agreement +1. Documenting in meta/main.yml why it is there and when it was agreed to (date) + +## Avoiding overly verbose roles +When we are splitting our roles up into smaller components we want to ensure we +avoid creating roles that are, for a lack of a better term, overly verbose. What +do we mean by that? If we have `openshift_master` as an example, and we were to +split it up, we would have a component for `etcd`, `docker`, and possibly for +its rpms/configs. We would want to avoid creating a role that would just create +certificates as those would make sense to be contained with the rpms and configs. +Likewise, when it comes to being able to restart the master, we wouldn't have a +role where that was its sole purpose. + +The same would apply for the `etcd` and `docker` roles. Anything that is required +as part of installing `etcd` such as generating certificates, installing rpms, +and upgrading data between versions should all be contained within the single +`etcd` role. + +## Enforcing standards +Certain naming standards like variable names could be verified as part of a Travis +test. If we were going to also enforce that a role either has tasks or includes +(for example) then we could create tests for that as well. + +## CI tests for individual roles +If we are able to correctly split up roles, it should be possible to test role +installations/upgrades like unit tests (assuming they would be able to be installed +independently of other components). |