Progressive delivery
Give on-call staff a clear release stop control
The operator needs to know which action limits exposure, what it leaves running and how to verify the result. Hand over those details before the first automated rollout.
In this article
Name the available actions
Explain feature disable, rollout pause and deployment abort separately. They can have different effects on new requests, existing sessions and background jobs.
Link the supported control and required authority. Avoid a procedure that depends on a release engineer's personal access or an undocumented console location.
State which action is appropriate for the service's common failure classes and when the operator should escalate.
Make release state discoverable
Show the candidate build, current exposure, analysis status and stable recovery version in one maintained view where possible.
Include the data compatibility boundary. An available older image may not be safe to run after an incompatible write or schema change.
Keep temporary flags and cohort rules associated with the release owner so support can reproduce what a user experienced.
Rehearse a stopped rollout
Ask the receiving operator to stop a safe candidate with a controlled fault. Verify the normal request path and inspect work already accepted by the candidate.
Have them locate the operation records needed for reconciliation and the owner of any external effect. This tests more than knowledge of the abort button.
Record gaps in permissions, observability and instructions, then correct the maintained procedure.
Close the rollout's temporary state
Assign ownership for removing flags, duplicate code paths and unused deployment resources after the recovery window. Include any analysis rule that was temporarily bypassed.
Keep the release decision and incident findings available for future changes. A failed canary is useful evidence when the team records what it caught and what its signals missed.
The handover is complete when another operator can limit exposure, explain what remains active and identify the next safe action. That capability matters more than a walkthrough of every controller setting.
Primary sources
Argo Rollouts: canary deploymentsArgo Rollouts: analysisReferences checked 11 September 2026.