On 21 July 2026 between 14:35 and 15:47 BST, the voice platform experienced a major degradation in call handling. Two processing servers entered a state in which they accepted new calls but could not progress them: callers experienced failed calls, unanswered ringing, or silent calls that could not be hung up. Because the affected servers continued to respond to routine health checks, the platform's load balancer continued to route calls to them and as their reported load figures froze, an increasing share of traffic was directed at the most-affected server as it was reporting the lowest load. By 15:00 the majority of new calls were failing. Calls handled by the remaining healthy servers continued to work normally throughout.
Service was restored by restarting the two affected servers at 15:45. The faulty component has been disabled until the underlying defect and the integration defect are both fixed. We are additionally: adding alerting that detects this failure signature within minutes; hardening the call-processing servers to automatically recover if call handling stalls; changing the routing layer to stop sending calls to servers whose health data has gone stale.