Merge pull request #24 from jmrodri/sdk-repo-split

Split SDK into multiple projects

Author: jmrodri

enhancements/split-sdk-into-multiple-projects-deps.png

@@ -0,0 +1,227 @@
+---
+title: split-sdk-into-multiple-projects
+authors:
+  - "@jesusr"
+reviewers:
+  - "@estroz"
+  - "@hasbro17"
+  - "@joelanford"
+  - "@fabianvf"
+  - "@shawn-hurley"
+approvers:
+  - "@estroz"
+  - "@hasbro17"
+  - "@joelanford"
+  - "@fabianvf"
+  - "@shawn-hurley"
+creation-date: 2020-05-18
+last-updated: 2020-05-27
+status: implementable
+see-also:
+replaces:
+superseded-by:
+---
+
+# Split SDK into multiple projects
+
+## Release Signoff Checklist
+
+- [X] Enhancement is `implementable`
+- [X] Design details are appropriately documented from clear requirements
+- [X] Test plan is defined
+- [X] Graduation criteria for dev preview, tech preview, GA
+
+## Open Questions [optional]
+
+1. What do we do with other shared tooling, CI, test helpers, etc. (e.g.
+   changelog generator)? Do we need a separate repo for that?
+   * Answer: We should visit these on a case by case basis as we see what is needed
+     during the splitting.
+
+## Summary
+
+This is the overarching enhancement that will plan how to split functionality
+out of the current operator-sdk project into separate projects. The initial
+thought on component breakdown mirrors the kubebuilder and controller-runtime
+projects:
+
+1. Common libraries and operator building blocks
+1. Go scaffolding plugin and the main SDK CLI
+1. Helm operator library and scaffolding plugin
+1. Ansible operator library and scaffolding plugin
+
+## Motivation
+
+The motivation is to avoid a breaking change in one component causing a major
+version bump for all other unrelated components. Another motivation is to make
+the projects consumable by other projects as libraries. A benefit would be
+better testing and easier to contribute to the smaller projects.
+
+While there are a few complexities with having separate repos, like PR
+coordination between repos, or just the working complexity of having to point
+at different repos to pick up your dep changes, these are obstacles we can
+overcome. They also aren’t that much different from other dependencies we
+work with.
+
+### Goals
+
+Split the operator-sdk repo into 4 separate repos:
+
+1. CLI and Go Operator
+1. SDK library
+1. Helm operator
+1. Ansible operator
+
+This SDK library split would be a requirement for a 1.0 release. The Ansible
+and Helm operators could go afterwards if need be.
+
+### Non-Goals
+
+* We will not solve the downstream SDK in this enhancement.
+* We will not solve all of the items required for a 1.0 release.
+* Defining when version bumps occur in each repo.
+
+## Proposal
+
+Split the operator-sdk repo into 4 separate repos as outlined in the Goals
+section above.
+
+1. CLI and Go Operator
+   * contains the Go operator code
+   * contains the CLI code
+   * related plugin and scaffolding
+   * lives in operator-framework/operator-sdk
+   * the split of the CLI/Go Operator repo is dependent on Kubebuilder Phase 2
+     to avoid creating a moving target.
+1. SDK library
+   * contains utility code like package status, status condition, annotated
+     watcher, etc.
+   * this is something that operator projects (including Helm and Ansible
+     operators) would import
+   * depends on controller-runtime
+   * lives in operator-framework/operator-lib (or some other name TBD)
+   * this library should probably *not* be 1.0; it's most likely to change often
+   * use Go apidiff tool to determine major version bumps
+1. Helm operator
+   * contains Helm operator code
+   * related plugin and scaffolding
+   * lives in operator-framework/helm-operator
+   * the Helm plugin is dependent on Kubebuilder Phase 1.
+1. Ansible operator
+   * contains Ansible operator code
+   * related plugin and scaffolding
+   * lives in operator-framework/ansible-operator
+   * the Ansible plugin is dependent on Kubebuilder Phase 1.
+
+![Image of project dependencies](split-sdk-into-multiple-projects-deps.png)
+
+### Tasks
+
+1. Triage pkg directory to determine what goes to which repo
+1. Use git commands to extract directories or files with history into these
+   new repos. Exact commands will be TBD.
+1. Evaluate if making a separate repo for the hack directory is necessary or
+   if we just copy it to each of the 4 repos.
+1. CLI and Go Operator
+   * keep operator-framework/operator-sdk
+   * cmd, internal directories remain here
+   * prior to 1.0 release, review entire CLI; axe all legacy support for
+     existing project layout
+   * consider making `generate csv` separate like `controller-gen` to support
+     OLM's plan for CSVless bundles (could be done later)
+1. SDK library
+   * create new repo operator-framework/operator-lib
+   * export the directories / files based on task 1 triage.
+1. Helm operator and plugin
+   * create new repo operator-framework/helm-operator
+   * export the pkg/helm tree to the new repo at the top level maintaining
+     history
+     * There is a plan to deprecate pkg/helm and replace with
+       [Joe's helm repo](https://github.com/joelanford/helm-operator) this
+       will be covered in a separate proposal to explicitly explain why and
+       how this will occur.
+1. Ansible operator and plugin
+   * create new repo operator-framework/ansible-operator
+   * export the pkg/ansible tree to the new repo at the top level
+     maintaining history
+
+### Affect on docs website
+
+Today the docs are stored in a website directory in the operator-sdk repo. The
+following illustrates how the docs can be split to follow the repos:
+
+1. CLI and Go Operator
+   * maintains a website directory that contains the SDK docs, hosted at
+     sdk.operatorframework.io
+1. SDK library
+   * will contain Godocs for the library
+1. Helm operator
+   * will contain Godocs for the library portion
+   * would contain a website directory that contains the Helm docs, hosted at
+     helm.operatorframework.io
+     * having its own domain means we can version the docs with the releases of
+       the Helm operator
+1. Ansible operator
+   * will contain Godocs for the library portion
+   * would contain a website directory that contains the Helm docs, hosted at
+     ansible.operatorframework.io
+     * having its own domain means we can version the docs with the releases of
+       the Ansible operator
+
+
+### Implementation Details/Notes/Constraints [optional]
+
+See above
+
+### Risks and Mitigations
+
+The biggest risk is the instability that will occur while the projects are
+split apart. Once all the projects are in place things will come back together.
+
+## Design Details
+
+### Test Plan
+
+The ultimate test is that e2e tests and manual testing of the SDK binary will
+be used to determine it remains functional. Each repo will contain their own
+unit test suites and where appropriate e2e tests.
+
+### Graduation Criteria
+
+N/A
+
+### Upgrade / Downgrade Strategy
+
+N/A
+
+### Version Skew Strategy
+
+N/A
+
+## Implementation History
+
+20200603 - Answer the helm-operator question.
+20200603 - Address questions from the review.
+20200602 - Answered the docs question.
+20200601 - Added versioning as a non-goal; Added KB phase 2 note to CLI repo.
+20200527 - This document is created.
+20200518 - Initial Google Document created to do initial set of review.
+
+## Drawbacks
+
+N/A
+
+## Alternatives
+
+### Docs
+
+An alternative to splitting up the docs would be to build a mechanism to add
+docs from Ansible/Helm repos into the main SDK site at build time. This would
+keep the top level site as it is now. A drawback would be that the documentation
+versioning would be for the entire SDK which negates the individual versioning
+we are attempting with splitting the repos.
+
+There are disussions about whether to version the Operator Framework as a whole,
+which could affect docs in a major way. It could also affect the underlying
+repos as well. This is not something we can solve in this proposal nor should
+we.