[ShadowCat567]

CloudFormation recently released a new flag on ContinueUpdateRollback, ForceRollback, which specifies whether CloudFormation should force a rollback to continue by skipping resources that are currently in the state UpdateFailed. This is effectively the same behavior we have with cdk rollback --force. The initial implementation of cdk rollback --force would use a poller to look for all the resources that failed to update while a deployment was happening and manually add them to the orphaned resources list. This flag on ContinueUpdateRollback allows us to offload this work to CloudFormation.
Added some new monitoring behavior to the stack-event-poller so we can still print out which resources are going to be skipped for the user to see.
Had to update @aws-sdk version in yarn.lock to get a more recent version of the CloudFormation SDK for the new feature.

---

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache-2.0 license

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.10%. Comparing base (068007f) to head (804b79d).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2078   +/-   ##
=======================================
  Coverage   92.10%   92.10%           
=======================================
  Files          82       82           
  Lines       13189    13189           
  Branches     1880     1880           
=======================================
  Hits        12148    12148           
  Misses       1003     1003           
  Partials       38       38           

Flag	Coverage Δ
suite.unit	92.10% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[maintainer-for-aws]

See the review summary comment for the overview; the notes below are inline.

[maintainer-for-aws]

🟡 Recommended — The Rolled back stack <name> (orphaning: …) warning is emitted inside the rollback loop right after monitor.stop(), before the success/failure outcome is known and before the possible continue (loop again) or throw. On the multi-iteration --force path an iteration that observed skipped resources but has not finished rolling back will announce a completed rollback and then loop again, possibly more than once; and if the rollback ultimately fails and the code throws, the user has already been told the stack rolled back. The past-tense wording asserts a completed, successful rollback that may not have happened — unlike the pre-call Continuing rollback of stack … message, which stays honestly present-progressive.

Suggested change: Emit the orphan-list message only once the rollback has actually succeeded — move it into the if (finalStackState.stackStatus.isRollbackSuccess || !stackErrorMessage) success branch so it prints with the real outcome and not on the continue/throw paths, or reword it to present-progressive to match the pre-call message.

[maintainer-for-aws]

🟡 Recommended — The continueUpdateRollback request now sends ForceRollback: options.orphanFailedResources (true on the --force path) together with ResourcesToSkip: resourcesToSkip, which is an empty array on that path. CloudFormation documents ForceRollback and ResourcesToSkip as mutually exclusive — specifying both in the same request returns a validation error (ContinueUpdateRollback). Since --force is now the only path that sets ForceRollback, if CloudFormation treats a present-but-empty ResourcesToSkip as specifying both, the entire cdk rollback --force flow breaks. In practice Query-protocol serialization of an empty list likely emits no members on the wire, so it most likely does not trip the error — but the unit test asserts ResourcesToSkip: [] against a client mock that does not enforce the service-side rule, so a green test proves nothing here.

Suggested change: Omit ResourcesToSkip entirely when forcing so the two inputs stay on provably separate paths, e.g.:

await cfn.continueUpdateRollback({
  StackName: stackArn,
  ClientRequestToken: randomUUID(),
  RoleARN: executionRoleArn,
  ...(options.orphanFailedResources
    ? { ForceRollback: true }
    : { ResourcesToSkip: resourcesToSkip }),
});

[maintainer-for-aws]

See the review summary comment for the overview; the notes below are inline.

[maintainer-for-aws]

🟡 Recommended — The removed code derived the skip list from the event poller and filtered it to top-level resources only (parentStackLogicalIds.length === 0). Sending ForceRollback: true instead delegates selection to CloudFormation, which skips every resource in UPDATE_FAILED, including those in nested stacks, and sets them to UPDATE_COMPLETE so they no longer match the stack template (ContinueUpdateRollback). So cdk rollback --force can now orphan nested-stack resources it never touched before — an observable behavior change under an existing command, landing under a chore: prefix.

Suggested change: Confirm the widened orphaning scope is intentional and call it out in the PR description and the user-facing rollback --force docs so the semantic change is not silent. --force is already an explicit destructive opt-in, so a feature flag is likely unnecessary, but the broadened behavior should be documented.

[maintainer-for-aws]

⚪ Optional — skippedResourceIds() joins parentStackLogicalIds with the logical ID to produce Parent.Child qualification — the one piece of reporting logic specific to the nested-stack orphaning this PR enables. Both new tests exercise only top-level resources with parentStackLogicalIds: [], so the qualifier join is never asserted and a regression in the Parent.Child format would go uncaught.

Suggested change: Add one case with a non-empty parentStackLogicalIds to lock in the Parent.Child output format of skippedResourceIds().