Freeze one number, call and expected destination
Record one authorized non-emergency inbound call with timestamp and timezone, called number, expected and actual destination IP, SIP Call-ID where available, caller result and whether other nominated addresses received traffic. Export or obtain the current Magrathea destination list and preserve edge and PBX evidence before changing a firewall, listener or pool entry.
The symptom “this server received no calls” does not prove that the address was removed. It may not be nominated, may be healthy but receive no selected call under the current distribution policy, or may accept an INVITE that fails later in PBX routing.
Confirm the nominated pool and service boundary
Match the affected number and SIP Resilience service to the exact ordered IP list, activation state, expected listener and change timestamp. Confirm that each address still belongs to the intended PBX or SBC and that the route back to Magrathea is correct. Do not infer the provider pool from DNS, a local cluster or an old implementation note.
Magrathea describes nominated IP addresses rather than a generic endpoint-registration set. A current SIP registration, if another service uses one, does not prove this inbound resilience pool contains the same address.
Observe the provider health-probe path
Magrathea says it probes each nominated IP several times a minute and removes an address when failure is detected. The public page does not name the protocol, source, destination port, payload, response criteria or failure threshold. Obtain current service-specific probe details, then correlate provider case or pool state with timestamped edge and host evidence.
Do not open all internet traffic, disable the firewall or fabricate a response. Apply only provider-confirmed least-privilege rules and prove that the intended application, not merely the operating system, returns the expected health result.
Separate probe health from SIP-call acceptance
A successful probe can show that the documented health check passed without proving the SIP listener, source authorization, Request-URI, endpoint identification or inbound dialplan will accept a real call. Conversely, a working SIP call observed outside the incident window does not prove the address stayed in the provider pool throughout the failure.
For one selected call, preserve the provider destination, network arrival, INVITE, response, route match and final application result. Correlate the same transaction by timestamp, Call-ID, branch and tags while protecting credentials and personal call data.
Prove removal and restoration as separate transitions
If the address fails the agreed health condition, capture when Magrathea marks or reports it out of pool and where new calls go. After restoring the confirmed probe and SIP path, capture when the address returns and whether a later selected call reaches it. The phrase “the moment a failure is detected” does not publish an exact timer or guarantee a call to every healthy address.
Do not represent this as active-call preservation. Provider pool changes normally concern later inbound attempts; preserve existing-dialog behavior and alternate capacity as separate service questions.
Retest new calls and preserve rollback
In an approved window, exercise a reversible failure and recovery with non-emergency calls. Verify the expected health transition, new-call treatment on remaining addresses, restoration, a later delivered INVITE, PBX route, identity, audio and clean release. Stop if the observed probe or distribution behavior differs from the agreed plan.
Restore the prior destination and firewall state if acceptance checks fail. Record expected and observed results without claiming a universal threshold, load-balancing algorithm, active-call handoff or laboratory validation. Never use 999/112 to test this workflow.
Sources & applicability
Primary references for the technical details above. Operational examples and planning checklists are VoIP.info editorial guidance.
Examples require adaptation to your topology. No live PBX or hardware testing is claimed.