Moving a payment callback endpoint: old orders, certificates and cutover checks

A working homepage does not prove that payment notifications reach your application after a domain or proxy change. First confirm whether an address change affects new orders or existing ones. This is a merchant migration design, not a claim that the platform offers multiple simultaneous callback addresses.

Continue by topic

Inventory outstanding orders before changing configuration

Record the old hostname, complete callback path, HTTP method, certificate expiry and receiving application, then list orders without a final verified result. Exclude secrets and full signatures. Ask the technical team whether the endpoint is captured when an order is created or read when a notification is sent; confirm the retry horizon and authorized reconciliation route.

An order bound to an old endpoint may still need that endpoint after configuration changes. Keep the old domain and certificate until the routing of outstanding orders is understood.

Check the receiving path without depending on redirects

Trace DNS resolution, TLS, reverse-proxy routing and application logs. With a CDN, check client-to-edge and edge-to-origin certificates separately. Do not disable certificate verification. Connectivity probes should use an owned health endpoint without payment side effects.

A redirect response does not prove that the payment sender will follow it. HTTP 307 preserves method and body when followed, but the actual notification client must support that behavior. Agree on a directly reachable final endpoint; avoid login pages, browser challenges and homepage redirects.

Make both receiving paths resolve to the same order

During an agreed overlap, use the same persistent order and deduplication records. Verify each notification using its real protocol and check order identity and amount before fulfillment. Migration is no reason to bypass signatures or keep deduplication only in one server’s memory.

Example: O-OLD arrives through the old endpoint; O-NEW arrives through the new one. A delayed repeat for O-OLD must resolve to its existing fulfillment even if an internal route changes. These are illustrative order names, not 稳付pay fields. Acknowledgements and retries follow the applicable integration agreement.

Roll back from evidence and test the retirement decision

No access log suggests checking DNS, TLS and the proxy. An access log without an application record suggests routing or body-reading problems. Signature failures require checking original fields, encoding and environment configuration, not manually marking the payment successful. Verify unknown orders through the authorized query process before asking a customer to pay again.

Keep cutover time, configuration version, redacted order mappings and an owner. Test delayed old orders, new notifications, duplicates, certificate failures and application errors. Confirm what a rollback changes for already-created orders. Retire the old endpoint only after outstanding orders and the agreed retry window are resolved.

Frequently asked questions

Is a permanent redirect from the old hostname enough?

No. Confirm whether the actual sender follows it and preserves the request. Keep the old endpoint according to outstanding orders and retry conditions.

Can CNY collection acknowledgement rules be reused for USDT?

No. Independently confirm USDT notification fields, verification and acknowledgement with the dedicated technical team; the existing CNY documentation is not its protocol.

References