Intermittent call-setup failures
Between 31 August and 8 September 2026, calls routed through our voice platform intermittently failed to connect. The cause has been identified and a fix is deployed. This report explains what happened, what it meant for your traffic, and what we have changed.
Summary
On a small number of occasions during the incident window, one of our regional voice switches stopped starting new calls for a period of between two and eight minutes. During an episode, calls sent to that switch did not connect: the caller heard silence and typically hung up. Between episodes, the platform performed normally.
The cause was a defect in how completed call recordings were uploaded to storage. The upload was performed by the same software that sets up calls, and while a slow upload was in progress it blocked new calls from being routed. Because uploads on our production platform were taking three to four minutes due to issues at our S3 provider, enough slow uploads stalled call setup on that switch for the same length of time.
We have moved recording uploads to a separate background process, so call handling never waits on storage. That change is deployed and we are monitoring production to confirm the fault is gone.
What you would have seen
The failure was visible only at the moment a call was being set up. Specifically:
- New calls did not connect. The call was accepted by the platform and then produced no ringing and no answer. Most callers abandoned after 20–40 seconds.
- A minority received an explicit rejection. Calls that were not abandoned were eventually released with a "no route to destination" failure, after roughly five minutes.
- Calls already in progress kept their audio. Conversations already established were not cut off and audio quality was unaffected.
- Hang-ups were delayed. Calls that ended during an episode were not torn down until the episode cleared, so they may appear in your records as longer than the conversation actually was. If you believe you have been billed for such a call, contact us and we will review it.
- Not all regions at once. An episode affected the switch handling your traffic at that moment, not the platform as a whole.
Call quality, recording integrity and call detail records were not affected. No recordings were lost, and no call content was exposed. The fault delayed and failed call setup; it did not corrupt or disclose data.
We are sorry for the disruption. A fault that fails calls at setup is the most visible kind we can have, and it took longer to identify than we would want because the symptoms pointed convincingly at the network for several days.
Thank you for your understanding.
