Until 2026-02-24 19:15:00 UTC, a misconfigured TTL caused a subset of calls to silently crash. This occurred only when the system was under high load, which is why it affected some calls but not all.

As a result, those calls could not be placed properly and would crash silently. By “silent”, we mean that the calls did crash, but were not marked as such. Consequently, they were effectively “lost” instead of being sent back to our clients’ systems as crashed calls.

As soon as the issue and its root cause were identified, all affected calls were manually resent to our clients’ systems with the correct “crashed” status. Therefore, no data loss occurred. However, there was a longer-than-expected delay between the time a call was triggered and when it was returned to the client’s system.

Once the issue was identified, we immediately released a patch that increased the faulty TTL value, resolving the issue even under high load. Additionally, we plan to revise the way audio streams are initialized after a call is created, to ensure that even a very short TTL cannot block the system as it did during this incident.