libp2p connections can stick around in zombi mode

[dvc94ch]

this happens reliably when hot reloading our flutter application during testing. there is no inject_disconnected or inject_connected emitted, just a new inject_connection_established. this causes all sorts of weird behaviour as the other application when using NotifyHandler::Any will send the on the old connection that refuses to die.

[canewsin]

Oh My, this is the reason My Flutter App hangs after hot reload, have to close and restart the App.

[mxinden]

Does this happen on v0.41.0 or master? In other words, does this happen with #2350 included?

[dvc94ch]

happens with #2350 included. but banning doesn't have anything to do with it. I'm thinking one solution would be to generate a random number at startup and send it with multistream-select. then if there was a state loss the other peer would detect it by receiving a new connection with a new session id or whatever you want to call it. that would then cause the swarm to close all the expired sessions and emit the proper events.

[mxinden]

I am not sure I follow on the bug report itself. Would you mind describing the steps involved? How do connections get into the "zombie" state? Does one endpoint disconnect?

[dvc94ch]

a connects to b
swarm a restarts
a connects to b
b has two connections from a
b sends messages on first connection which never arrive (which can be checked by using NotifyHandler::One(connection_id) instead of NotifyHandler::Any.

would add a test to reproduce it but it requires tons of boilerplate. wasn't there an effort to add an add_address to the swarm?

[dvc94ch]

actually never mind. just thought that the other explanation is that the old swarm isn't shut down and somehow still around. seems like that's more likely