A phone number is an entry point, while the agent release is one destination behind it. A rollback can preserve the business number when the routing and deployment model supports that separation, but the entire path still needs verification.
LiveKit's deployment-management guide documents rollback behavior and plan-dependent availability. Its dispatch-rule guide describes selecting agents for inbound rooms. Check the actual project and installed tooling before planning an incident command.
Identify what the release changed
Inventory agent code, shared policy, client configuration, action schema, secrets, dispatch, and remote mappings. Reverting a container may not restore an independently changed fee table or calendar identifier.
Keep a known approved package and a compatibility policy for durable records. The old code must be able to interpret pending actions created by the newer release, or those actions need a separate reviewed recovery path.
Rehearse a fictional booking regression
A new release mislabels uncertain scheduling writes as confirmed. The team suspends affected consequential actions, selects the approved rollback path, and preserves the same business phone entry point.
Controlled calls verify the intended agent, correct client configuration, and restored pending language. Existing uncertain actions are reconciled before retry. The rollback should not cause both old and new implementations to create appointments for one delayed event.
| Rollback fixture | Required evidence |
|---|---|
| Existing phone entry point | Correct active route retained |
| New calls after rollback | Approved behavior package |
| Active calls during transition | Explicit supported treatment |
| Pending newer-version action | Compatible recovery or review |
| Delayed event | One authorized action owner |
Verify recovery beyond deployment status
Inspect real controlled call outcomes and staff queues. A deployment reporting healthy does not prove the right prompt, resource mapping, or remote action semantics are active. Keep a staffed window and an owner for remaining customer follow-up.
Fast rollback reduces exposure to a regression but can reintroduce older limitations. A forward fix may preserve newer compatibility while taking longer to validate. Choose using incident consequence and concrete evidence, not a habit of always preferring one direction.
Review launch checks for the complete behavior package and operating responsibilities.
Questions about stable-number rollback
Does rollback automatically restore every business setting?
No. Identify independently managed configuration and verify its accepted version.
Can active calls be moved transparently to old code?
Do not assume live migration. Verify supported lifecycle behavior and define the customer recovery policy for interrupted sessions.
What closes the rollback incident?
Verified routing and behavior, reconciled consequential actions, owned unresolved requests, current client communication, and a recorded plan for the defective release.