swarm/src/behaviour: Add KeepAlive Behaviour

[mxinden]

Description

When getting started with rust-libp2p, it is confusing that connections over time through the KeepAlive mechanism.

See e.g. #2768 #2679

Today, users can import the Ping NetworkBehaviour and set with_keep_alive.

rust-libp2p/protocols/ping/src/handler.rs

Lines 103 to 116 in ce963df

	/// Sets whether the ping protocol itself should keep the connection alive,
	/// apart from the maximum allowed failures.
	///
	/// By default, the ping protocol itself allows the connection to be closed
	/// at any time, i.e. in the absence of ping failures the connection lifetime
	/// is determined by other protocol handlers.
	///
	/// If the maximum number of allowed ping failures is reached, the
	/// connection is always terminated as a result of [`ConnectionHandler::poll`]
	/// returning an error, regardless of the keep-alive setting.
	pub fn with_keep_alive(mut self, b: bool) -> Self {
	self.keep_alive = b;
	self
	}

I think there is value in a simple KeepAlive NetworkBehaviour implementation, keeping the connection alive, without the need to import Ping. The ConnectionHandler of the KeepAlive NetworkBehaviour could be exposed as well, thus allowing it to be used without its parent NetworkBehaviour.

Are you planning to do it yourself in a pull request?

No

[K0UR05H]

It seems like it's possible to keep a connection open using DummyBehaviour::with_keep_alive but I don't know if it's good enough.

[mxinden]

To be honest I haven't been aware of DummyBehaviour::with_keep_alive. Thanks @k0ur0x.

I am closing here for now. I would still consider a pure KeepAlive NetworkBehaviour useful. That said, I doubt it is worth the additional work.

[thomaseizinger]

I am reopening this because:

• the orthogonality of the static keep alive mechanism with the remaining ping functionality is highlighted once again in swarm: Add FromFn ConnectionHandler #2852
• I find DummyBehaviour to be a rather strange component too and thus I think it isn't a particularly good solution to the problem discussed here. Also see swarm: Split off "keep alive" functionality from DummyConnectionHandler #2859.