Skip to content

Commit

Permalink
Remove all batches related to a peer on disconnect (sigp#5969)
Browse files Browse the repository at this point in the history 
* Remove all batches related to a peer on disconnect

* Cleanup map entries after disconnect

* Allow lookups to continue in case of disconnections

* Pretty response types

* fmt

* Fix lints

* Remove lookup if it cannot progress

* Fix tests

* Remove poll_close on rpc behaviour

* Remove redundant test

* Fix issue raised by lion

* Revert pretty response types

* Cleanup

* Fix test

* Merge remote-tracking branch 'origin/release-v5.2.1' into rpc-error-on-disconnect-revert

* Apply suggestions from joao

Co-authored-by: João Oliveira <hello@jxs.pt>

* Fix log

* update request status on no peers found

* Do not remove lookup after peer disconnection

* Add comments about expected event api

* Update single_block_lookup.rs

* Update mod.rs

* Merge branch 'rpc-error-on-disconnect-revert' into 5969-review

* Merge pull request #10 from dapplion/5969-review

Add comments about expected event api
  • Loading branch information
@pawanjay176
pawanjay176 authored Jun 26, 2024
1 parent 758b58c commit bf4cbd3
Show file tree
Hide file tree
Showing 12 changed files with 270 additions and 182 deletions.
31 changes: 0 additions & 31 deletions beacon_node/lighthouse_network/src/rpc/handler.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -352,37 +352,6 @@ where
!matches!(self.state, HandlerState::Deactivated)
}

// NOTE: This function gets polled to completion upon a connection close.
fn poll_close(&mut self, _: &mut Context<'_>) -> Poll<Option<Self::ToBehaviour>> {
// Inform the network behaviour of any failed requests

while let Some(substream_id) = self.outbound_substreams.keys().next().cloned() {
let outbound_info = self
.outbound_substreams
.remove(&substream_id)
.expect("The value must exist for a key");
// If the state of the connection is closing, we do not need to report this case to
// the behaviour, as the connection has just closed non-gracefully
if matches!(outbound_info.state, OutboundSubstreamState::Closing(_)) {
continue;
}

// Register this request as an RPC Error
return Poll::Ready(Some(HandlerEvent::Err(HandlerErr::Outbound {
error: RPCError::Disconnected,
proto: outbound_info.proto,
id: outbound_info.req_id,
})));
}

// Also handle any events that are awaiting to be sent to the behaviour
if !self.events_out.is_empty() {
return Poll::Ready(Some(self.events_out.remove(0)));
}

Poll::Ready(None)
}

fn poll(
&mut self,
cx: &mut Context<'_>,
Expand Down
94 changes: 1 addition & 93 deletions beacon_node/lighthouse_network/tests/rpc_tests.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
mod common;

use common::Protocol;
use lighthouse_network::rpc::{methods::*, RPCError};
use lighthouse_network::rpc::methods::*;
use lighthouse_network::{rpc::max_rpc_size, NetworkEvent, ReportSource, Request, Response};
use slog::{debug, warn, Level};
use ssz::Encode;
Expand Down Expand Up @@ -1012,98 +1012,6 @@ fn test_tcp_blocks_by_root_chunked_rpc_terminates_correctly() {
})
}

#[test]
fn test_disconnect_triggers_rpc_error() {
// set up the logging. The level and enabled logging or not
let log_level = Level::Debug;
let enable_logging = false;

let log = common::build_log(log_level, enable_logging);
let spec = E::default_spec();

let rt = Arc::new(Runtime::new().unwrap());
// get sender/receiver
rt.block_on(async {
let (mut sender, mut receiver) = common::build_node_pair(
Arc::downgrade(&rt),
&log,
ForkName::Base,
&spec,
Protocol::Tcp,
)
.await;

// BlocksByRoot Request
let rpc_request = Request::BlocksByRoot(BlocksByRootRequest::new(
// Must have at least one root for the request to create a stream
vec![Hash256::from_low_u64_be(0)],
&spec,
));

// build the sender future
let sender_future = async {
loop {
match sender.next_event().await {
NetworkEvent::PeerConnectedOutgoing(peer_id) => {
// Send a STATUS message
debug!(log, "Sending RPC");
sender
.send_request(peer_id, 42, rpc_request.clone())
.unwrap();
}
NetworkEvent::RPCFailed { error, id: 42, .. } => match error {
RPCError::Disconnected => return,
other => panic!("received unexpected error {:?}", other),
},
other => {
warn!(log, "Ignoring other event {:?}", other);
}
}
}
};

// determine messages to send (PeerId, RequestId). If some, indicates we still need to send
// messages
let mut sending_peer = None;
let receiver_future = async {
loop {
// this future either drives the sending/receiving or times out allowing messages to be
// sent in the timeout
match futures::future::select(
Box::pin(receiver.next_event()),
Box::pin(tokio::time::sleep(Duration::from_secs(1))),
)
.await
{
futures::future::Either::Left((ev, _)) => match ev {
NetworkEvent::RequestReceived { peer_id, .. } => {
sending_peer = Some(peer_id);
}
other => {
warn!(log, "Ignoring other event {:?}", other);
}
},
futures::future::Either::Right((_, _)) => {} // The timeout hit, send messages if required
}

// if we need to send messages send them here. This will happen after a delay
if let Some(peer_id) = sending_peer.take() {
warn!(log, "Receiver got request, disconnecting peer");
receiver.__hard_disconnect_testing_only(peer_id);
}
}
};

tokio::select! {
_ = sender_future => {}
_ = receiver_future => {}
_ = sleep(Duration::from_secs(30)) => {
panic!("Future timed out");
}
}
})
}

/// Establishes a pair of nodes and disconnects the pair based on the selected protocol via an RPC
/// Goodbye message.
fn goodbye_test(log_level: Level, enable_logging: bool, protocol: Protocol) {
Expand Down
38 changes: 36 additions & 2 deletions beacon_node/network/src/sync/backfill_sync/mod.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -307,15 +307,49 @@ impl<T: BeaconChainTypes> BackFillSync<T> {
/// A peer has disconnected.
/// If the peer has active batches, those are considered failed and re-requested.
#[must_use = "A failure here indicates the backfill sync has failed and the global sync state should be updated"]
pub fn peer_disconnected(&mut self, peer_id: &PeerId) -> Result<(), BackFillError> {
pub fn peer_disconnected(
&mut self,
peer_id: &PeerId,
network: &mut SyncNetworkContext<T>,
) -> Result<(), BackFillError> {
if matches!(
self.state(),
BackFillState::Failed | BackFillState::NotRequired
) {
return Ok(());
}

self.active_requests.remove(peer_id);
if let Some(batch_ids) = self.active_requests.remove(peer_id) {
// fail the batches.
for id in batch_ids {
if let Some(batch) = self.batches.get_mut(&id) {
match batch.download_failed(false) {
Ok(BatchOperationOutcome::Failed { blacklist: _ }) => {
self.fail_sync(BackFillError::BatchDownloadFailed(id))?;
}
Ok(BatchOperationOutcome::Continue) => {}
Err(e) => {
self.fail_sync(BackFillError::BatchInvalidState(id, e.0))?;
}
}
// If we have run out of peers in which to retry this batch, the backfill state
// transitions to a paused state.
// We still need to reset the state for all the affected batches, so we should not
// short circuit early.
if self.retry_batch_download(network, id).is_err() {
debug!(
self.log,
"Batch could not be retried";
"batch_id" => id,
"error" => "no synced peers"
);
}
} else {
debug!(self.log, "Batch not found while removing peer";
"peer" => %peer_id, "batch" => id)
}
}
}

// Remove the peer from the participation list
self.participating_peers.remove(peer_id);
Expand Down
42 changes: 26 additions & 16 deletions beacon_node/network/src/sync/block_lookups/mod.rs
View file
Original file line number Diff line number Diff line change
@@ -1,3 +1,25 @@
//! Implements block lookup sync.
//!
//! Block lookup sync is triggered when a peer claims to have imported a block we don't know about.
//! For example, a peer attesting to a head block root that is not in our fork-choice. Lookup sync
//! is recursive in nature, as we may discover that this attested head block root has a parent that
//! is also unknown to us.
//!
//! Block lookup is implemented as an event-driven state machine. It sends events to the network and
//! beacon processor, and expects some set of events back. A discrepancy in the expected event API
//! will result in lookups getting "stuck". A lookup becomes stuck when there is no future event
//! that will trigger the lookup to make progress. There's a fallback mechanism that drops lookups
//! that live for too long, logging the line "Notify the devs a sync lookup is stuck".
//!
//! The expected event API is documented in the code paths that are making assumptions with the
//! comment prefix "Lookup sync event safety:"
//!
//! Block lookup sync attempts to not re-download or re-process data that we already have. Block
//! components are cached temporarily in multiple places before they are imported into fork-choice.
//! Therefore, block lookup sync must peek these caches correctly to decide when to skip a download
//! or consider a lookup complete. These caches are read from the `SyncNetworkContext` and its state
//! returned to this module as `LookupRequestResult` variants.

use self::parent_chain::{compute_parent_chains, NodeChain};
pub use self::single_block_lookup::DownloadResult;
use self::single_block_lookup::{LookupRequestError, LookupResult, SingleBlockLookup};
Expand Down Expand Up @@ -410,21 +432,9 @@ impl<T: BeaconChainTypes> BlockLookups<T> {
/* Error responses */

pub fn peer_disconnected(&mut self, peer_id: &PeerId) {
self.single_block_lookups.retain(|_, lookup| {
for (_, lookup) in self.single_block_lookups.iter_mut() {
lookup.remove_peer(peer_id);

// Note: this condition should be removed in the future. It's not strictly necessary to drop a
// lookup if there are no peers left. Lookup should only be dropped if it can not make progress
if lookup.has_no_peers() {
debug!(self.log,
"Dropping single lookup after peer disconnection";
"block_root" => ?lookup.block_root()
);
false
} else {
true
}
});
}
}

/* Processing responses */
Expand Down Expand Up @@ -787,12 +797,12 @@ impl<T: BeaconChainTypes> BlockLookups<T> {
};

if stuck_lookup.id == ancestor_stuck_lookup.id {
warn!(self.log, "Notify the devs, a sync lookup is stuck";
warn!(self.log, "Notify the devs a sync lookup is stuck";
"block_root" => ?stuck_lookup.block_root(),
"lookup" => ?stuck_lookup,
);
} else {
warn!(self.log, "Notify the devs, a sync lookup is stuck";
warn!(self.log, "Notify the devs a sync lookup is stuck";
"block_root" => ?stuck_lookup.block_root(),
"lookup" => ?stuck_lookup,
"ancestor_block_root" => ?ancestor_stuck_lookup.block_root(),
Expand Down
38 changes: 32 additions & 6 deletions beacon_node/network/src/sync/block_lookups/single_block_lookup.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -197,21 +197,36 @@ impl<T: BeaconChainTypes> SingleBlockLookup<T> {
}

let Some(peer_id) = self.use_rand_available_peer() else {
// Allow lookup to not have any peers. In that case do nothing. If the lookup does
// not have peers for some time, it will be dropped.
// Allow lookup to not have any peers and do nothing. This is an optimization to not
// lose progress of lookups created from a block with unknown parent before we receive
// attestations for said block.
// Lookup sync event safety: If a lookup requires peers to make progress, and does
// not receive any new peers for some time it will be dropped. If it receives a new
// peer it must attempt to make progress.
R::request_state_mut(self)
.get_state_mut()
.update_awaiting_download_status("no peers");
return Ok(());
};

let request = R::request_state_mut(self);
match request.make_request(id, peer_id, downloaded_block_expected_blobs, cx)? {
LookupRequestResult::RequestSent(req_id) => {
// Lookup sync event safety: If make_request returns `RequestSent`, we are
// guaranteed that `BlockLookups::on_download_response` will be called exactly
// with this `req_id`.
request.get_state_mut().on_download_start(req_id)?
}
LookupRequestResult::NoRequestNeeded => {
// Lookup sync event safety: Advances this request to the terminal `Processed`
// state. If all requests reach this state, the request is marked as completed
// in `Self::continue_requests`.
request.get_state_mut().on_completed_request()?
}
// Sync will receive a future event to make progress on the request, do nothing now
LookupRequestResult::Pending(reason) => {
// Lookup sync event safety: Refer to the code paths constructing
// `LookupRequestResult::Pending`
request
.get_state_mut()
.update_awaiting_download_status(reason);
Expand All @@ -222,16 +237,28 @@ impl<T: BeaconChainTypes> SingleBlockLookup<T> {
// Otherwise, attempt to progress awaiting processing
// If this request is awaiting a parent lookup to be processed, do not send for processing.
// The request will be rejected with unknown parent error.
//
// TODO: The condition `block_is_processed || Block` can be dropped after checking for
// unknown parent root when import RPC blobs
} else if !awaiting_parent
&& (block_is_processed || matches!(R::response_type(), ResponseType::Block))
{
// maybe_start_processing returns Some if state == AwaitingProcess. This pattern is
// useful to conditionally access the result data.
if let Some(result) = request.get_state_mut().maybe_start_processing() {
// Lookup sync event safety: If `send_for_processing` returns Ok() we are guaranteed
// that `BlockLookups::on_processing_result` will be called exactly once with this
// lookup_id
return R::send_for_processing(id, result, cx);
}
// Lookup sync event safety: If the request is not in `AwaitingDownload` or
// `AwaitingProcessing` state it is guaranteed to receive some event to make progress.
}

// Lookup sync event safety: If a lookup is awaiting a parent we are guaranteed to either:
// (1) attempt to make progress with `BlockLookups::continue_child_lookups` if the parent
// lookup completes, or (2) get dropped if the parent fails and is dropped.

Ok(())
}

Expand All @@ -246,10 +273,9 @@ impl<T: BeaconChainTypes> SingleBlockLookup<T> {
self.peers.insert(peer_id)
}

/// Remove peer from available peers. Return true if there are no more available peers and all
/// requests are not expecting any future event (AwaitingDownload).
pub fn remove_peer(&mut self, peer_id: &PeerId) -> bool {
self.peers.remove(peer_id)
/// Remove peer from available peers.
pub fn remove_peer(&mut self, peer_id: &PeerId) {
self.peers.remove(peer_id);
}

/// Returns true if this lookup has zero peers
Expand Down
Loading

0 comments on commit bf4cbd3

Please sign in to comment.