Open LPardue opened 2 years ago
We had client-side idle timeouts in an earlier version, but later realized we probably shouldn't. Gave the reasoning in 4.2 ("Client Response") (and hit it again lightly under "Circuit Breakers" in the "Graceful Degradation" section). Maybe we need more?
And are you suggesting a protocol change, like that we need to refresh the channel state on the connection periodically if it's been joined without traffic for long enough? Or more like text on how to handle it right if no data is flowing?
I probably need to think about it some more :-)
Has any consideration been given to channel idle timeouts? That is, if the channel-related comms go quiet, the channel is cleaned away without a flood of chatter. This might help avoid thundering herds of MC_STATE messages coming back from clients.