Implement ADR 0002: trunk-based automated release process

[screendriver]

Context

ADR 0002 defines the target release process for wire-webapp:

• main becomes the trunk branch.
• Every merge to main deploys to Edge.
• Release branches use release/YYYY-MM-DD.N.
• Release branch updates deploy to Beta.
• Successful Beta deployments create YYYY-MM-DD.N-beta.M tags.
• Production waits for GitHub Environment approval.
• Production promotes the Beta-tested artifact and does not rebuild from source.
• Successful Production deployments create YYYY-MM-DD.N-production tags.
• Production tags are immutable release history.
• Rollback is handled by a dedicated GitHub Actions workflow.
• On-premises maintenance releases are handled separately from the normal cloud release flow.

ADR:

docs/decision-records/0002-trunk-based-automated-release-process.md

Goal

Implement ADR 0002 in small, reviewable, low-risk steps.

The final release process should be GitHub-driven, auditable, and safe:

• build once
• deploy to Beta
• validate on Beta
• approve Production through GitHub Environment approval
• promote the same artifact to Production
• create the Production tag only after successful Production deployment
• keep rollback explicit and auditable

Non-goals for the first phase

The first phase should not try to do everything at once.

Out of scope initially:

• Deleting dev.
• Deleting or renaming master.
• Removing all old release triggers immediately.
• Full on-premises maintenance release automation.
• Perfect durable artifact storage for rollback.
• Full feature flag governance automation.
• Large workflow rewrites without intermediate compatibility.

Progress checklist

• Inspect the current release implementation.
• Add release metadata helpers.
• Prepare reusable E2E / Testiny support.
• Add Edge deployment from trunk.
• Add release branch to Beta workflow.
• Add Production approval and artifact promotion.
• Add release branch creation workflow.
• Add first-class Production rollback workflow.
• Cut over branch model from dev / master to main.
• Retire old tag-driven release path.
• Add maintenance release automation separately.

Phase 1: Inspect the current release implementation

Create an implementation inventory before changing code.

The inventory should cover:

• existing workflows for build, deploy, release, E2E, rollback, Docker, Helm, and notifications
• existing scripts for release tagging and changelog generation
• current branch assumptions
• current tag assumptions
• current deployment targets and environment names
• Testiny integration
• Wire / Deployoholics notification integration
• conflicts between the current implementation and ADR 0002
• a proposed PR sequence with small, reviewable steps

Acceptance criteria:

• No files are changed.
• No branch is created.
• No PR is opened.
• The current implementation is documented well enough to plan the first implementation PR.

Phase 2: Add release metadata helpers

Add small deterministic helpers for release metadata.

Required behavior:

• validate release branch names matching release/YYYY-MM-DD.N
• extract the release identifier YYYY-MM-DD.N
• compute the next Beta tag from existing tags matching YYYY-MM-DD.N-beta.M
• compute the Production tag YYYY-MM-DD.N-production
• detect whether the Production tag already exists
• detect whether an existing Production tag points to the current release commit
• never move, delete, or force-push tags
• add tests for the pure release metadata logic

Acceptance criteria:

• Release naming rules are test-covered.
• Tag selection behavior is deterministic.
• No workflow behavior changes yet.

Phase 3: Prepare reusable E2E / Testiny support

Prepare .github/workflows/e2e-tests.yml so it can be called by the future release workflow.

Required behavior:

• support testinyRunName through workflow_call
• support TESTINY_API_KEY through workflow_call secrets
• keep existing manual dispatch behavior unchanged
• keep Playwright sharding and report generation unchanged

Acceptance criteria:

• A caller workflow can run E2E against a provided webapp URL.
• A caller workflow can upload the result to a named Testiny run.
• Existing manual E2E usage still works.

Phase 4: Add Edge deployment from trunk

Add a dedicated Edge deployment workflow.

Required behavior:

• trigger on pushes to main
• during transition, support the current branch setup only if needed
• build/package the webapp
• deploy to Edge
• do not require GitHub Environment approval
• use concurrency so superseded Edge runs can be cancelled

Acceptance criteria:

• Edge follows trunk.
• Edge deployment is separate from the old release/tag-driven Production flow.
• Existing Production release behavior is not changed in this phase.

Phase 5: Add release branch to Beta workflow

Add the first version of the cloud release workflow.

Required behavior:

• trigger on pushes to release/**
• validate the branch name as release/YYYY-MM-DD.N
• build the release artifact once
• deploy the artifact to Beta
• create YYYY-MM-DD.N-beta.M only after successful Beta deployment
• fetch tags before computing the next Beta tag
• fail if a concurrent workflow created the same Beta tag
• never force-push tags
• run E2E against Beta
• report E2E results to Testiny
• add a GitHub step summary with release identifier, commit SHA, artifact checksum, Beta tag, Beta URL, E2E result, and Testiny run name
• notify the release channel / Deployoholics on failure

Acceptance criteria:

• Pushing to a valid release branch automatically deploys Beta.
• Beta tags are created only after successful Beta deployment.
• E2E and Testiny are part of the release branch workflow.
• No Production deployment is added yet.

Phase 6: Add Production approval and promotion

Extend the release workflow with the Production part.

Required behavior:

• Production waits for GitHub Environment approval.
• Production deploys the same artifact that was deployed to Beta.
• Production does not rebuild from source.
• Production deployment is serialized across the repository.
• Production deployments are not cancelled automatically.
• Before deploying, check whether YYYY-MM-DD.N-production already exists.
• If the Production tag already points to the same commit, skip redeployment.
• If the Production tag exists and points to a different commit, fail.
• Create YYYY-MM-DD.N-production only after successful Production deployment.
• Never move, delete, or force-push Production tags.
• Notify the release channel / Deployoholics on failure.

Acceptance criteria:

• Beta-tested artifact is promoted to Production.
• Production tag represents a successful Production deployment.
• Existing Production tags remain immutable.
• Re-running the workflow does not redeploy a commit that already has the matching Production tag.

Phase 7: Add release branch creation workflow

Add a manual workflow for creating release branches from GitHub Actions.

Required behavior:

• use workflow_dispatch
• input: release_identifier in format YYYY-MM-DD.N
• input: source_ref, defaulting to main
• input: confirmation boolean
• create release/YYYY-MM-DD.N
• fail if the branch already exists
• do not create tags
• do not deploy directly; the pushed branch should trigger the release workflow

Acceptance criteria:

• Release branches can be created without local machine release operations.
• Branch creation is auditable in GitHub Actions.

Phase 8: Add first-class rollback workflow

Add or adapt a dedicated Production rollback workflow.

Required behavior:

• use workflow_dispatch
• input: previous known-good Production tag
• input: rollback reason
• input: optional incident reference
• input: confirmation boolean
• validate the tag exists
• deploy to Production through GitHub Actions
• use the Production GitHub Environment
• serialize against other Production deployments
• notify the release channel / Deployoholics
• do not create, move, delete, or rewrite release tags

Acceptance criteria:

• Rollback is explicit and auditable.
• Rollback is owned by engineering release owners, not QA.
• Runtime state is visible through GitHub deployment metadata, logs, and notifications.

Phase 9: Cut over branch model

After the new workflows are proven:

• rename master to main
• retire dev as a long-lived release branch
• update branch protection rules
• update GitHub workflow branch filters
• update documentation
• update external integrations
• update any wire-builds mappings if still needed

Acceptance criteria:

• main is the single trunk branch.
• Edge follows main.
• Release branches are cut from main.
• dev and master no longer carry release meaning.

Phase 10: Retire old tag-driven release path

After the new release workflow is proven:

• stop Production deployment from pre-created Production tags
• deprecate or remove local release tagging commands
• remove outdated staging / production release tag assumptions where no longer needed
• keep compatibility only where explicitly required

Acceptance criteria:

• Production tags are created after successful Production deployment.
• Local machines are no longer required for normal release operations.
• Old release paths cannot accidentally bypass the new process.

Phase 11: Add maintenance release automation separately

Add maintenance release automation only after the cloud release path is stable.

Required behavior:

• create maintenance branches only when needed
• use branch names like maintenance/<maintenance-line-key>
• do not deploy maintenance branches to normal Beta or Production
• validate maintenance artifacts through a dedicated maintenance path
• create maintenance tags like <maintenance-line-key>-maintenance.X

Acceptance criteria:

• On-premises maintenance releases are separate from cloud releases.
• Normal cloud cadence is not constrained by maintenance lines.

Open questions

• Which existing environment maps to ADR “Beta” during the transition?
• Should wire-webapp-staging be renamed conceptually or physically?
• Where should immutable release artifacts live long term?
• Is GitHub Actions artifact retention sufficient for the first iteration?
• Should Elastic Beanstalk application versions become the durable promotion unit?
• What exact Testiny run naming convention should release workflows use?
• Which GitHub Environment reviewers should approve Production?
• Which Wire channel should receive release workflow failure and stalled approval notifications?
• What is the planned date for master to main rename?
• Which external integrations still depend on dev or master?
• Which parts of wire-builds still need updates from the webapp release workflow?

Risks

• Accidentally keeping both old and new Production paths active.
• Creating Production tags before deployment instead of after deployment.
• Rebuilding for Production and invalidating Beta validation.
• Cancelling an in-progress Production deployment.
• Losing rollback ability because the artifact is not durable enough.
• Blocking release migration on branch renaming too early.
• Mixing maintenance release automation into the cloud release workflow.

Suggested first PR

Start with release metadata helpers and tests.

This should not change any workflow behavior.

Suggested scope:

• validate release/YYYY-MM-DD.N
• extract release identifier
• compute next Beta tag
• compute Production tag
• detect existing Production tag behavior
• add tests
• no GitHub Actions changes

[screendriver]

Implementation inventory completed.

Important findings:

• Current release deployment is still partly driven by pre-existing tags.
• Current Production tag format is YYYY-MM-DD-production.N, but ADR 0002 requires YYYY-MM-DD.N-production.
• Current staging tag format is YYYY-MM-DD-staging.N, but ADR 0002 requires Beta tags like YYYY-MM-DD.N-beta.M.
• Release branches currently trigger the publish workflow, but no Beta deployment target is mapped for them.
• Edge currently follows dev, not main.
• Testiny support exists, but workflow_call does not yet expose testinyRunName / TESTINY_API_KEY.
• The current redeploy workflow rebuilds from a tag and is not yet a first-class rollback workflow.
• dev and master still carry release meaning.

Proposed first PR:

Add pure release metadata helpers and tests.

Scope:

• validate release/YYYY-MM-DD.N
• extract YYYY-MM-DD.N
• compute next YYYY-MM-DD.N-beta.M
• compute YYYY-MM-DD.N-production
• detect whether a Production tag already exists
• detect whether an existing Production tag points to the current release commit
• no GitHub Actions behavior changes

[screendriver]

Phase 3 is done.

Pull request merged:

• Expose Testiny reporting for reusable E2E workflow [WPB-26469] #21634

This prepared the reusable E2E workflow for future release workflows by exposing Testiny reporting through workflow_call.

Next phase:

Add a dedicated Edge deployment workflow.

Important transition note:

The new workflow should not automatically run on dev, because the existing publish workflow still deploys dev to wire-webapp-edge. The new workflow should auto-trigger on main only and optionally support manual dispatch for safe testing.

[screendriver]

Phase 4 is done.

PR merged:

• Add dedicated Edge deployment workflow [WPB-26469] #21641

This adds the future main → Edge deployment workflow without changing the existing dev → Edge path or the current release process.

Next phase:

Add the release branch → Beta workflow.

[screendriver]

Phase 5 is done.

Pull request merged:

• Add release branch Beta deployment workflow [WPB-26469] #21648

This adds the manual transition version of the release-cloud Beta workflow:

• explicit release_branch input
• explicit confirmation before deployment
• validation before checkout
• build once
• deploy to the temporary Beta mapping (wire-webapp-staging)
• create YYYY-MM-DD.N-beta.M after successful Beta deployment
• run E2E against Beta
• report to Testiny

Automatic release/** triggering stays disabled during the transition to avoid disrupting the current release process.