my guess would be that the clients are using NDK, which introduces two client side issues, the failure of keeping the service worker alive, plus the flaky reconnect code of NDK, plus the aggressive connection management policies in strfry. i'm quite sure that although the relay is not in the clear, ...