rust - 如何从派生的NetworkBehaviour发出SwarmEvent::Behaviour？

Question

我正在寻找一种利用mDNS，floodsub和kademlia DHT的网络行为。到目前为止，我已经使所有这些服务都可以使用，但是还不能将这些服务的响应用于有意义的事情。

理想情况下，我将能够将来自行为过程事件的数据(例如将为Kad DHT实现的事件)输送到主Swarm poll循环中。例如，在我的情况下，我对表示通过sled数据库保留在磁盘上的图形的结构的引用。该结构的所有权存在于轮询群的方法中。收到KademliaEvent后，我将如何更新此图表(例如添加条目)？

我尝试过的解决方案:

将数据结构的所有权移到我想更新为ClientBehavior结构的位置，这样我就可以通过self.my_data_structure.add(result); KademliaEvent方法中的inject_event

#[derive(NetworkBehaviour)]根本不喜欢这个

Graph只是一个结构，不会发出任何事件/实现NetworkBehaviour

创建一个Context结构，该结构由derive的NetworkBehaviour组成，可用于在轮询方法和相应的inject_event方法之间来回传递响应

#[derive(NetworkBehaviour)]不能与Arc s/Mutex es

一起使用

这是我的NetworkBehavior的样子:

/// A network behavior describing a client connected to a pub-sub compatible,
/// optionally mDNS-compatible network. Such a "behavior" may be implemented for
/// any libp2p transport, but any transport used with this behavior must implement
/// asynchronous reading & writing capabilities.
#[derive(NetworkBehaviour)]
pub struct ClientBehavior<TSubstream: AsyncRead + AsyncWrite + Send + Unpin + 'static> {
    /// Some pubsub mechanism bound to the above transport
    pub floodsub: Floodsub<TSubstream>,

    /// Some mDNS service bound to the above transport
    pub mdns: Mdns<TSubstream>,

    /// Allow for the client to do some external discovery on the global network through a KAD DHT
    pub kad_dht: Kademlia<TSubstream, MemoryStore>,
}

我的Kademlia NetworkBehaviourEventProceess实现如下所示:

impl<TSubstream: AsyncRead + AsyncWrite + Send + Unpin + 'static>
    NetworkBehaviourEventProcess<KademliaEvent> for ClientBehavior<TSubstream>
{
    fn inject_event(&mut self, event: KademliaEvent) {
        // Some behavior logic, nothing special, really, just a bunch of matches
        // NOTE: this is when I'd want to update the `Graph`, since KademliaEvent will contain data that I need to put in the `Graph` instance
    }
}

以及如何生成 Swarm和 ClientBehavior:

// Initialize a new behavior for a client that we will generate in the not-so-distant future with the given peerId, alongside
// an mDNS service handler as well as a floodsub instance targeted at the given peer
let mut behavior = ClientBehavior {
    floodsub: Floodsub::new(self.peer_id.clone()),
    mdns: Mdns::new().await?,
    kad_dht: Kademlia::new(self.peer_id.clone(), store),
};

// Iterate through bootstrap addresses
for bootstrap_peer in bootstrap_addresses {
    // NOTE: add_address is a method that isn't provided by #[derive(NetworkBehaviour)].
    // It's just a helper method I wrote that adds the peer to the Floodsub & DHT views.
    behavior.add_address(&bootstrap_peer.0, bootstrap_peer.1); // Add the bootstrap peer to the DHT
}

// Bootstrap the behavior's DHT
behavior.kad_dht.bootstrap();

// Note: here, `self` is a reference to a configuration struct holding a peer ID, and a keypair
let mut swarm = Swarm::new(
    libp2p::build_tcp_ws_secio_mplex_yamux(self.keypair.clone())?,
    behavior,
    self.peer_id.clone(),
); // Initialize a swarm

以及我如何轮询群:

// NOTE: Here, `self` has ownership of the aforementioned `Graph` instance. I'd like to update this
// instance from this block, or from the ClientBehavior itself--as long as I'm able to update it once a `KademliaEvent` is received.

// Try to get the address we'll listen on
if let Ok(addr) = format!("/ip4/0.0.0.0/tcp/{}", port).parse::<Multiaddr>() {
    // Try to tell the swarm to listen on this address, return an error if this doesn't work
    if let Err(e) = Swarm::listen_on(&mut swarm, addr.clone()) {
        // Convert the addr err into an io error
        let e: std::io::Error = io::ErrorKind::AddrNotAvailable.into();

        // Return an error
        return Err(e.into());
    };

    // Fetch the hash of the first transaction in the DAG from the network
    swarm
        .kad_dht
        .get_record(&Key::new(&sync::ROOT_TRANSACTION_KEY), Quorum::Majority);

    loop {
        // Poll the swarm
        match swarm.next_event().await {
            // NOTE: Initially, I was under the impression that I would be able to intercept
            // events from the ClientBehavior here. Yet, the below info! call is never reached.
            // This remains the case in libp2p example code that I have experimented with.
            SwarmEvent::Behaviour(e) => info!("idk: {:?}", e),

            _ => debug!("Some other event; this is handled specifically in the actual codebase, but for this question, all I really care about is the above behavior event."),
        };
    }
}

我是否只需要自己实现 NetworkBehaviour？

Answer 1

您的问题有点晚了，但这也许会有所帮助:

Moving ownership of the data structure I want to update to the ClientBehavior struct so I can just self.my_data_structure.add(result); from the KademliaEvent inject_event method

• #[derive(NetworkBehaviour)] does NOT like this at all
• Graph is just a struct, and doesn't emit any events / implement NetworkBehaviour

documentation对此有以下说法:

#[behaviour(ignore)] can be added on a struct field to disable generation of delegation to the fields which do not implement NetworkBehaviour.

因为否则需要所有字段来实现NetworkBehaviour特性。
那应该使它成为可行的解决方案。
但是，无论哪种方式，如果您尝试修改超出Mdns范围的结构，我都相信¹，针对所有这些问题的最惯用的解决方案将是通过poll函数(返回 Poll::Ready(GenerateEvent(TOut)))发出自定义事件，然后您可以监听该事件通过另一个NetworkBehaviourEventProcess impl获取对应的结构，然后应修改该结构。
类似于 libp2p-ping implementation使用VecDeque事件队列，然后在 poll()中传播该事件队列的方法，您可以在ClientBehaviour中有一个类似的队列(标记为 behaviour(ignore)])，追加到队列中，以 poll()发出事件，并在调用代码/等待中捕获该特定事件循环以某种方式？
编辑: should also be able to extend the poll method允许以这种方式返回其他事件。
¹:我对rust-libp2p的能力还差强人意，并且到目前为止还没有足够的经验来说明事实，您的问题对我来说实际上是一个教程