reth_downloaders/bodies/

request.rs

use crate::metrics::{BodyDownloaderMetrics, ResponseMetrics};
use alloy_consensus::BlockHeader;
use alloy_primitives::B256;
use futures::{Future, FutureExt};
use reth_consensus::Consensus;
use reth_network_p2p::{
    bodies::{client::BodiesClient, response::BlockResponse},
    error::{DownloadError, DownloadResult},
    priority::Priority,
};
use reth_network_peers::{PeerId, WithPeerId};
use reth_primitives::{BlockBody, GotExpected, SealedBlock, SealedHeader};
use reth_primitives_traits::InMemorySize;
use std::{
    collections::VecDeque,
    mem,
    pin::Pin,
    sync::Arc,
    task::{ready, Context, Poll},
};

/// Body request implemented as a [Future].
///
/// The future will poll the underlying request until fulfilled.
/// If the response arrived with insufficient number of bodies, the future
/// will issue another request until all bodies are collected.
///
/// It then proceeds to verify the downloaded bodies. In case of an validation error,
/// the future will start over.
///
/// The future will filter out any empty headers (see [`alloy_consensus::Header::is_empty`]) from
/// the request. If [`BodiesRequestFuture`] was initialized with all empty headers, no request will
/// be dispatched and they will be immediately returned upon polling.
///
/// NB: This assumes that peers respond with bodies in the order that they were requested.
/// This is a reasonable assumption to make as that's [what Geth
/// does](https://github.com/ethereum/go-ethereum/blob/f53ff0ff4a68ffc56004ab1d5cc244bcb64d3277/les/server_requests.go#L245).
/// All errors regarding the response cause the peer to get penalized, meaning that adversaries
/// that try to give us bodies that do not match the requested order are going to be penalized
/// and eventually disconnected.
pub(crate) struct BodiesRequestFuture<H, B: BodiesClient> {
    client: Arc<B>,
    consensus: Arc<dyn Consensus<H, B::Body>>,
    metrics: BodyDownloaderMetrics,
    /// Metrics for individual responses. This can be used to observe how the size (in bytes) of
    /// responses change while bodies are being downloaded.
    response_metrics: ResponseMetrics,
    // Headers to download. The collection is shrunk as responses are buffered.
    pending_headers: VecDeque<SealedHeader<H>>,
    /// Internal buffer for all blocks
    buffer: Vec<BlockResponse<H, B::Body>>,
    fut: Option<B::Output>,
    /// Tracks how many bodies we requested in the last request.
    last_request_len: Option<usize>,
}

impl<H, B> BodiesRequestFuture<H, B>
where
    H: BlockHeader,
    B: BodiesClient + 'static,
{
    /// Returns an empty future. Use [`BodiesRequestFuture::with_headers`] to set the request.
    pub(crate) fn new(
        client: Arc<B>,
        consensus: Arc<dyn Consensus<H, B::Body>>,
        metrics: BodyDownloaderMetrics,
    ) -> Self {
        Self {
            client,
            consensus,
            metrics,
            response_metrics: Default::default(),
            pending_headers: Default::default(),
            buffer: Default::default(),
            last_request_len: None,
            fut: None,
        }
    }

    pub(crate) fn with_headers(mut self, headers: Vec<SealedHeader<H>>) -> Self {
        self.buffer.reserve_exact(headers.len());
        self.pending_headers = VecDeque::from(headers);
        // Submit the request only if there are any headers to download.
        // Otherwise, the future will immediately be resolved.
        if let Some(req) = self.next_request() {
            self.submit_request(req, Priority::Normal);
        }
        self
    }

    fn on_error(&mut self, error: DownloadError, peer_id: Option<PeerId>) {
        self.metrics.increment_errors(&error);
        tracing::debug!(target: "downloaders::bodies", ?peer_id, %error, "Error requesting bodies");
        if let Some(peer_id) = peer_id {
            self.client.report_bad_message(peer_id);
        }
        self.submit_request(
            self.next_request().expect("existing hashes to resubmit"),
            Priority::High,
        );
    }

    /// Retrieve header hashes for the next request.
    fn next_request(&self) -> Option<Vec<B256>> {
        let mut hashes =
            self.pending_headers.iter().filter(|h| !h.is_empty()).map(|h| h.hash()).peekable();
        hashes.peek().is_some().then(|| hashes.collect())
    }

    /// Submit the request with the given priority.
    fn submit_request(&mut self, req: Vec<B256>, priority: Priority) {
        tracing::trace!(target: "downloaders::bodies", request_len = req.len(), "Requesting bodies");
        let client = Arc::clone(&self.client);
        self.last_request_len = Some(req.len());
        self.fut = Some(client.get_block_bodies_with_priority(req, priority));
    }

    /// Process block response.
    /// Returns an error if the response is invalid.
    fn on_block_response(&mut self, response: WithPeerId<Vec<B::Body>>) -> DownloadResult<()>
    where
        B::Body: InMemorySize,
    {
        let (peer_id, bodies) = response.split();
        let request_len = self.last_request_len.unwrap_or_default();
        let response_len = bodies.len();

        tracing::trace!(target: "downloaders::bodies", request_len, response_len, ?peer_id, "Received bodies");

        // Increment total downloaded metric
        self.metrics.total_downloaded.increment(response_len as u64);

        // TODO: Malicious peers often return a single block even if it does not exceed the soft
        // response limit (2MB). This could be penalized by checking if this block and the
        // next one exceed the soft response limit, if not then peer either does not have the next
        // block or deliberately sent a single block.
        if bodies.is_empty() {
            return Err(DownloadError::EmptyResponse)
        }

        if response_len > request_len {
            return Err(DownloadError::TooManyBodies(GotExpected {
                got: response_len,
                expected: request_len,
            }))
        }

        // Buffer block responses
        self.try_buffer_blocks(bodies)?;

        // Submit next request if any
        if let Some(req) = self.next_request() {
            self.submit_request(req, Priority::High);
        } else {
            self.fut = None;
        }

        Ok(())
    }

    /// Attempt to buffer body responses. Returns an error if body response fails validation.
    /// Every body preceding the failed one will be buffered.
    ///
    /// This method removes headers from the internal collection.
    /// If the response fails validation, then the header will be put back.
    fn try_buffer_blocks(&mut self, bodies: Vec<B::Body>) -> DownloadResult<()>
    where
        B::Body: InMemorySize,
    {
        let bodies_capacity = bodies.capacity();
        let bodies_len = bodies.len();
        let mut bodies = bodies.into_iter().peekable();

        let mut total_size = bodies_capacity * mem::size_of::<BlockBody>();
        while bodies.peek().is_some() {
            let next_header = match self.pending_headers.pop_front() {
                Some(header) => header,
                None => return Ok(()), // no more headers
            };

            if next_header.is_empty() {
                // increment empty block body metric
                total_size += mem::size_of::<BlockBody>();
                self.buffer.push(BlockResponse::Empty(next_header));
            } else {
                let next_body = bodies.next().unwrap();

                // increment full block body metric
                total_size += next_body.size();

                let block = SealedBlock::new(next_header, next_body);

                if let Err(error) = self.consensus.validate_block_pre_execution(&block) {
                    // Body is invalid, put the header back and return an error
                    let hash = block.hash();
                    let number = block.number();
                    self.pending_headers.push_front(block.header);
                    return Err(DownloadError::BodyValidation {
                        hash,
                        number,
                        error: Box::new(error),
                    })
                }

                self.buffer.push(BlockResponse::Full(block));
            }
        }

        // Increment per-response metric
        self.response_metrics.response_size_bytes.set(total_size as f64);
        self.response_metrics.response_length.set(bodies_len as f64);

        Ok(())
    }
}

impl<H, B> Future for BodiesRequestFuture<H, B>
where
    H: BlockHeader + Unpin + Send + Sync + 'static,
    B: BodiesClient<Body: InMemorySize> + 'static,
{
    type Output = DownloadResult<Vec<BlockResponse<H, B::Body>>>;

    fn poll(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
        let this = self.get_mut();

        loop {
            if this.pending_headers.is_empty() {
                return Poll::Ready(Ok(std::mem::take(&mut this.buffer)))
            }

            // Check if there is a pending requests. It might not exist if all
            // headers are empty and there is nothing to download.
            if let Some(fut) = this.fut.as_mut() {
                match ready!(fut.poll_unpin(cx)) {
                    Ok(response) => {
                        let peer_id = response.peer_id();
                        if let Err(error) = this.on_block_response(response) {
                            this.on_error(error, Some(peer_id));
                        }
                    }
                    Err(error) => {
                        if error.is_channel_closed() {
                            return Poll::Ready(Err(error.into()))
                        }

                        this.on_error(error.into(), None);
                    }
                }
            }

            // Buffer any empty headers
            while this.pending_headers.front().is_some_and(|h| h.is_empty()) {
                let header = this.pending_headers.pop_front().unwrap();
                this.buffer.push(BlockResponse::Empty(header));
            }
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::{
        bodies::test_utils::zip_blocks,
        test_utils::{generate_bodies, TestBodiesClient},
    };
    use reth_consensus::test_utils::TestConsensus;
    use reth_testing_utils::{generators, generators::random_header_range};

    /// Check if future returns empty bodies without dispatching any requests.
    #[tokio::test]
    async fn request_returns_empty_bodies() {
        let mut rng = generators::rng();
        let headers = random_header_range(&mut rng, 0..20, B256::ZERO);

        let client = Arc::new(TestBodiesClient::default());
        let fut = BodiesRequestFuture::new(
            client.clone(),
            Arc::new(TestConsensus::default()),
            BodyDownloaderMetrics::default(),
        )
        .with_headers(headers.clone());

        assert_eq!(
            fut.await.unwrap(),
            headers.into_iter().map(BlockResponse::Empty).collect::<Vec<_>>()
        );
        assert_eq!(client.times_requested(), 0);
    }

    /// Check that the request future
    #[tokio::test]
    async fn request_submits_until_fulfilled() {
        // Generate some random blocks
        let (headers, mut bodies) = generate_bodies(0..=19);

        let batch_size = 2;
        let client = Arc::new(
            TestBodiesClient::default().with_bodies(bodies.clone()).with_max_batch_size(batch_size),
        );
        let fut = BodiesRequestFuture::new(
            client.clone(),
            Arc::new(TestConsensus::default()),
            BodyDownloaderMetrics::default(),
        )
        .with_headers(headers.clone());

        assert_eq!(fut.await.unwrap(), zip_blocks(headers.iter(), &mut bodies));
        assert_eq!(
            client.times_requested(),
            // div_ceild
            (headers.into_iter().filter(|h| !h.is_empty()).count() as u64 + 1) / 2
        );
    }
}