it's probably still significant. if you can find out which relay and versions of the relays it is, and which clients show the worst behaviour you will probably find it's a combination
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, that the client is making it worse
1
