TransportService: Improve connection stability by downgrading connections on substream inactivity

[lexnv]

Context

The transport service handler intent is to downgrade connections (leading to possibly closing them), if substreams where not open in a given timeframe:

litep2p/src/protocol/transport_service.rs

Lines 126 to 130 in d50ec10

	/// Close the connection if no substreams are open within this time frame.
	keep_alive_timeout: Duration,
	/// Pending keep-alive timeouts.
	pending_keep_alive_timeouts: FuturesUnordered<BoxFuture<'static, (PeerId, ConnectionId)>>,

When a connection is established, the timeout is properly tracked:

litep2p/src/protocol/transport_service.rs

Lines 213 to 216 in d50ec10

	self.pending_keep_alive_timeouts.push(Box::pin(async move {
	tokio::time::sleep(keep_alive_timeout).await;
	(peer, connection_id)
	}));

When the timeout expires, the connection is downgraded:

litep2p/src/protocol/transport_service.rs

Lines 414 to 427 in d50ec10

	while let Poll::Ready(Some((peer, connection_id))) =
	self.pending_keep_alive_timeouts.poll_next_unpin(cx)
	{
	if let Some(context) = self.connections.get_mut(&peer) {
	tracing::trace!(
	target: LOG_TARGET,
	?peer,
	?connection_id,
	"keep-alive timeout over, downgrade connection",
	);
	context.downgrade(&connection_id);
	}
	}

Issue

Opening of substreams are not taken into account for downgrading connections:

litep2p/src/protocol/transport_service.rs

Line 410 in d50ec10

Some(event) => return Poll::Ready(Some(event.into())),

Here we should match for Some(InnerTransportEvent::SubstreamOpened and move forward the timeout future.

With the current implementation, we downgrade connections at 5 seconds intervals.

Implications

We need to double-check if the protocol name matches the substream open protocol.
Further, InnerTransportEvent::SubstreamOpened needs to be extended with a connectionID because the TransportService holds up to 2 connections (primary and secondary). We should properly advance the timeout of the connection ID.