Batch same-channel MPP claims into a single commitment

When an MMP payment has multiple parts that arrive over the same final-hop
channel, we previously processed each part sequentially through the full
commitment dance: claim one part, build a ChannelMonitorUpdate, ship a
commitment_signed, wait for the peer's RAA + next commitment_signed, then
move to the next part. This added significant claim latency and wasted
round trips on unnecessary monitor updates.

Instead, queue all MPP part claims into the channel's holding cell first,
then flush them together in a single pass. The holding-cell flush produces
one combined ChannelMonitorUpdate (preimage + commitment steps) and one
commitment_signed carrying all update_fulfill_htlc messages at once,
eliminating the intermediate round trips.

To enable this, the old get_update_fulfill_htlc_and_commit is split into
queue_fulfill_htlc (pushes into the holding cell) and the existing holding-
cell flush path. ClaimedMPPPayment RAA blockers are deliberately ignored
during both the queue and flush phases so the first MPP part on a channel
does not force subsequent parts onto the standalone preimage path.

When the channel cannot immediately flush its holding cell (e.g. peer is
disconnected or another monitor update is in flight), a standalone preimage-
only ChannelMonitorUpdate is persisted for HTLC-preimage safety, matching
the prior behavior.

Author: Abeeujah

lightning/src/ln/chanmon_update_fail_tests.rs

@@ -4654,18 +4654,13 @@ fn test_claim_to_closed_channel_blocks_claimed_event() {
 }
 
 #[test]
-#[cfg(all(feature = "std", not(target_os = "windows")))]
 fn test_single_channel_multiple_mpp() {
 	use crate::util::config::UserConfig;
-	use std::sync::atomic::{AtomicBool, Ordering};
 
-	// Test what happens when we attempt to claim an MPP with many parts that came to us through
-	// the same channel with a synchronous persistence interface which has very high latency.
-	//
-	// Previously, if a `revoke_and_ack` came in while we were still running in
-	// `ChannelManager::claim_payment` we'd end up hanging waiting to apply a
-	// `ChannelMonitorUpdate` until after it completed. See the commit which introduced this test
-	// for more info.
+	// Test that claiming an MPP whose parts all arrived over the same final-hop channel batches
+	// the per-part fulfills into a single `ChannelMonitorUpdate` and a single
+	// `commitment_signed` (with one `update_fulfill_htlc` per part), rather than processing each
+	// part serially with its own monitor update / commitment round trip.
 	let chanmon_cfgs = create_chanmon_cfgs(9);
 	let node_cfgs = create_node_cfgs(9, &chanmon_cfgs);
 	let mut config = test_default_channel_config();
@@ -4674,45 +4669,17 @@ fn test_single_channel_multiple_mpp() {
 		10;
 	let configs: [Option<UserConfig>; 9] = core::array::from_fn(|_| Some(config.clone()));
 	let node_chanmgrs = create_node_chanmgrs(9, &node_cfgs, &configs);
-	let mut nodes = create_network(9, &node_cfgs, &node_chanmgrs);
+	let nodes = create_network(9, &node_cfgs, &node_chanmgrs);
 
-	let node_b_id = nodes[1].node.get_our_node_id();
-	let node_c_id = nodes[2].node.get_our_node_id();
-	let node_d_id = nodes[3].node.get_our_node_id();
-	let node_e_id = nodes[4].node.get_our_node_id();
-	let node_f_id = nodes[5].node.get_our_node_id();
-	let node_g_id = nodes[6].node.get_our_node_id();
 	let node_h_id = nodes[7].node.get_our_node_id();
-	let node_i_id = nodes[8].node.get_our_node_id();
 
 	// Send an MPP payment in six parts along the path shown from top to bottom
 	//        0
 	//   1 2 3 4 5 6
 	//        7
 	//        8
 	//
-	// We can in theory reproduce this issue with fewer channels/HTLCs, but getting this test
-	// robust is rather challenging. We rely on having the main test thread wait on locks held in
-	// the background `claim_funds` thread and unlocking when the `claim_funds` thread completes a
-	// single `ChannelMonitorUpdate`.
-	// This thread calls `get_and_clear_pending_msg_events()` and `handle_revoke_and_ack()`, both
-	// of which require `ChannelManager` locks, but we have to make sure this thread gets a chance
-	// to be blocked on the mutexes before we let the background thread wake `claim_funds` so that
-	// the mutex can switch to this main thread.
-	// This relies on our locks being fair, but also on our threads getting runtime during the test
-	// run, which can be pretty competitive. Thus we do a dumb dance to be as conservative as
-	// possible - we have a background thread which completes a `ChannelMonitorUpdate` (by sending
-	// into the `write_blocker` mpsc) but it doesn't run until a mpsc channel sends from this main
-	// thread to the background thread, and then we let it sleep a while before we send the
-	// `ChannelMonitorUpdate` unblocker.
-	// Further, we give ourselves two chances each time, needing 4 HTLCs just to unlock our two
-	// `ChannelManager` calls. We then need a few remaining HTLCs to actually trigger the bug, so
-	// we use 6 HTLCs.
-	// Finaly, we do not run this test on Winblowz because it, somehow, in 2025, does not implement
-	// actual preemptive multitasking and thinks that cooperative multitasking somehow is
-	// acceptable in the 21st century, let alone a quarter of the way into it.
-	const MAX_THREAD_INIT_TIME: std::time::Duration = std::time::Duration::from_secs(1);
-
+	// All six parts arrive at node 8 over the same node 7 -> node 8 channel.
 	create_announced_chan_between_nodes_with_value(&nodes, 0, 1, 100_000, 0);
 	create_announced_chan_between_nodes_with_value(&nodes, 0, 2, 100_000, 0);
 	create_announced_chan_between_nodes_with_value(&nodes, 0, 3, 100_000, 0);
@@ -4728,7 +4695,7 @@ fn test_single_channel_multiple_mpp() {
 	create_announced_chan_between_nodes_with_value(&nodes, 6, 7, 100_000, 0);
 	create_announced_chan_between_nodes_with_value(&nodes, 7, 8, 1_000_000, 0);
 
-	let (mut route, payment_hash, payment_preimage, payment_secret) =
+	let (route, payment_hash, payment_preimage, payment_secret) =
 		get_route_and_payment_hash!(&nodes[0], nodes[8], 50_000_000);
 
 	send_along_route_with_secret(
@@ -4747,181 +4714,25 @@ fn test_single_channel_multiple_mpp() {
 		payment_secret,
 	);
 
-	let (do_a_write, blocker) = std::sync::mpsc::sync_channel(0);
-	*nodes[8].chain_monitor.write_blocker.lock().unwrap() = Some(blocker);
-
-	// Until we have std::thread::scoped we have to unsafe { turn off the borrow checker }.
-	// We do this by casting a pointer to a `TestChannelManager` to a pointer to a
-	// `TestChannelManager` with different (in this case 'static) lifetime.
-	// This is even suggested in the second example at
-	// https://doc.rust-lang.org/std/mem/fn.transmute.html#examples
-	let claim_node: &'static TestChannelManager<'static, 'static> =
-		unsafe { std::mem::transmute(nodes[8].node as &TestChannelManager) };
-	let thrd = std::thread::spawn(move || {
-		// Initiate the claim in a background thread as it will immediately block waiting on the
-		// `write_blocker` we set above.
-		claim_node.claim_funds(payment_preimage);
-	});
-
-	// First unlock one monitor so that we have a pending
-	// `update_fulfill_htlc`/`commitment_signed` pair to pass to our counterparty.
-	do_a_write.send(()).unwrap();
-
-	let event_node: &'static TestChannelManager<'static, 'static> =
-		unsafe { std::mem::transmute(nodes[8].node as &TestChannelManager) };
-	let thrd_event = std::thread::spawn(move || {
-		let mut have_event = false;
-		while !have_event {
-			let mut events = event_node.get_and_clear_pending_events();
-			assert!(events.len() == 1 || events.len() == 0);
-			if events.len() == 1 {
-				if let Event::PaymentClaimed { .. } = events[0] {
-				} else {
-					panic!("Unexpected event {events:?}");
-				}
-				have_event = true;
-			}
-			if !have_event {
-				std::thread::yield_now();
-			}
-		}
-	});
-
-	// Then fetch the `update_fulfill_htlc`/`commitment_signed`. Note that the
-	// `get_and_clear_pending_msg_events` will immediately hang trying to take a peer lock which
-	// `claim_funds` is holding. Thus, we release a second write after a small sleep in the
-	// background to give `claim_funds` a chance to step forward, unblocking
-	// `get_and_clear_pending_msg_events`.
-	let do_a_write_background = do_a_write.clone();
-	let block_thrd2 = AtomicBool::new(true);
-	let block_thrd2_read: &'static AtomicBool = unsafe { std::mem::transmute(&block_thrd2) };
-	let thrd2 = std::thread::spawn(move || {
-		while block_thrd2_read.load(Ordering::Acquire) {
-			std::thread::yield_now();
-		}
-		std::thread::sleep(MAX_THREAD_INIT_TIME);
-		do_a_write_background.send(()).unwrap();
-		std::thread::sleep(MAX_THREAD_INIT_TIME);
-		do_a_write_background.send(()).unwrap();
-	});
-	block_thrd2.store(false, Ordering::Release);
-	let mut first_updates = get_htlc_update_msgs(&nodes[8], &node_h_id);
-
-	// Thread 2 could unblock first, or it could get blocked waiting on us to process a
-	// `PaymentClaimed` event. Either way, wait until both have finished.
-	thrd2.join().unwrap();
-	thrd_event.join().unwrap();
-
-	// Disconnect node 6 from all its peers so it doesn't bother to fail the HTLCs back
-	nodes[7].node.peer_disconnected(node_b_id);
-	nodes[7].node.peer_disconnected(node_c_id);
-	nodes[7].node.peer_disconnected(node_d_id);
-	nodes[7].node.peer_disconnected(node_e_id);
-	nodes[7].node.peer_disconnected(node_f_id);
-	nodes[7].node.peer_disconnected(node_g_id);
-
-	let first_update_fulfill = first_updates.update_fulfill_htlcs.remove(0);
-	nodes[7].node.handle_update_fulfill_htlc(node_i_id, first_update_fulfill);
-	check_added_monitors(&nodes[7], 1);
-	expect_payment_forwarded!(nodes[7], nodes[1], nodes[8], Some(1000), false, false);
-	nodes[7].node.handle_commitment_signed_batch_test(node_i_id, &first_updates.commitment_signed);
-	check_added_monitors(&nodes[7], 1);
-	let (raa, cs) = get_revoke_commit_msgs(&nodes[7], &node_i_id);
-
-	// Now, handle the `revoke_and_ack` from node 5. Note that `claim_funds` is still blocked on
-	// our peer lock, so we have to release a write to let it process.
-	// After this call completes, the channel previously would be locked up and should not be able
-	// to make further progress.
-	let do_a_write_background = do_a_write.clone();
-	let block_thrd3 = AtomicBool::new(true);
-	let block_thrd3_read: &'static AtomicBool = unsafe { std::mem::transmute(&block_thrd3) };
-	let thrd3 = std::thread::spawn(move || {
-		while block_thrd3_read.load(Ordering::Acquire) {
-			std::thread::yield_now();
-		}
-		std::thread::sleep(MAX_THREAD_INIT_TIME);
-		do_a_write_background.send(()).unwrap();
-		std::thread::sleep(MAX_THREAD_INIT_TIME);
-		do_a_write_background.send(()).unwrap();
-	});
-	block_thrd3.store(false, Ordering::Release);
-	nodes[8].node.handle_revoke_and_ack(node_h_id, &raa);
-	thrd3.join().unwrap();
-	assert!(!thrd.is_finished());
-
-	let thrd4 = std::thread::spawn(move || {
-		do_a_write.send(()).unwrap();
-		do_a_write.send(()).unwrap();
-	});
-
-	thrd4.join().unwrap();
-	thrd.join().unwrap();
-
-	// At the end, we should have 7 ChannelMonitorUpdates - 6 for HTLC claims, and one for the
-	// above `revoke_and_ack`.
-	check_added_monitors(&nodes[8], 7);
-
-	// Now drive everything to the end, at least as far as node 7 is concerned...
-	*nodes[8].chain_monitor.write_blocker.lock().unwrap() = None;
-	nodes[8].node.handle_commitment_signed_batch_test(node_h_id, &cs);
-	check_added_monitors(&nodes[8], 1);
-
-	let (mut updates, raa) = get_updates_and_revoke(&nodes[8], &node_h_id);
-
-	nodes[7].node.handle_update_fulfill_htlc(node_i_id, updates.update_fulfill_htlcs.remove(0));
-	expect_payment_forwarded!(nodes[7], nodes[2], nodes[8], Some(1000), false, false);
-	nodes[7].node.handle_update_fulfill_htlc(node_i_id, updates.update_fulfill_htlcs.remove(0));
-	expect_payment_forwarded!(nodes[7], nodes[3], nodes[8], Some(1000), false, false);
-	let mut next_source = 4;
-	if let Some(update) = updates.update_fulfill_htlcs.get(0) {
-		nodes[7].node.handle_update_fulfill_htlc(node_i_id, update.clone());
-		expect_payment_forwarded!(nodes[7], nodes[4], nodes[8], Some(1000), false, false);
-		next_source += 1;
-	}
+	// Claim the MPP payment. All six parts arrived over the same node 7 -> node 8 channel, so
+	// LDK should queue all six fulfills into that channel's holding cell and release them
+	// together as a single `ChannelMonitorUpdate` and a single `commitment_signed` carrying
+	// all six `update_fulfill_htlc`s, instead of one round trip per part.
+	nodes[8].node.claim_funds(payment_preimage);
+	expect_payment_claimed!(nodes[8], payment_hash, 50_000_000);
 
-	nodes[7].node.handle_commitment_signed_batch_test(node_i_id, &updates.commitment_signed);
-	nodes[7].node.handle_revoke_and_ack(node_i_id, &raa);
-	if updates.update_fulfill_htlcs.get(0).is_some() {
-		check_added_monitors(&nodes[7], 5);
-	} else {
-		check_added_monitors(&nodes[7], 4);
-	}
-
-	let (raa, cs) = get_revoke_commit_msgs(&nodes[7], &node_i_id);
-
-	nodes[8].node.handle_revoke_and_ack(node_h_id, &raa);
-	nodes[8].node.handle_commitment_signed_batch_test(node_h_id, &cs);
-	check_added_monitors(&nodes[8], 2);
-
-	let (mut updates, raa) = get_updates_and_revoke(&nodes[8], &node_h_id);
-
-	nodes[7].node.handle_update_fulfill_htlc(node_i_id, updates.update_fulfill_htlcs.remove(0));
-	expect_payment_forwarded!(nodes[7], nodes[next_source], nodes[8], Some(1000), false, false);
-	next_source += 1;
-	nodes[7].node.handle_update_fulfill_htlc(node_i_id, updates.update_fulfill_htlcs.remove(0));
-	expect_payment_forwarded!(nodes[7], nodes[next_source], nodes[8], Some(1000), false, false);
-	next_source += 1;
-	if let Some(update) = updates.update_fulfill_htlcs.get(0) {
-		nodes[7].node.handle_update_fulfill_htlc(node_i_id, update.clone());
-		expect_payment_forwarded!(nodes[7], nodes[next_source], nodes[8], Some(1000), false, false);
-	}
-
-	nodes[7].node.handle_commitment_signed_batch_test(node_i_id, &updates.commitment_signed);
-	nodes[7].node.handle_revoke_and_ack(node_i_id, &raa);
-	if updates.update_fulfill_htlcs.get(0).is_some() {
-		check_added_monitors(&nodes[7], 5);
-	} else {
-		check_added_monitors(&nodes[7], 4);
-	}
-
-	let (raa, cs) = get_revoke_commit_msgs(&nodes[7], &node_i_id);
-	nodes[8].node.handle_revoke_and_ack(node_h_id, &raa);
-	nodes[8].node.handle_commitment_signed_batch_test(node_h_id, &cs);
-	check_added_monitors(&nodes[8], 2);
+	// Exactly one `ChannelMonitorUpdate` should have been added for all six MPP parts.
+	check_added_monitors(&nodes[8], 1);
 
-	let raa = get_event_msg!(nodes[8], MessageSendEvent::SendRevokeAndACK, node_h_id);
-	nodes[7].node.handle_revoke_and_ack(node_i_id, &raa);
-	check_added_monitors(&nodes[7], 1);
+	// And there should be exactly one batched commitment update to node 7 carrying all six
+	// fulfills.
+	let updates = get_htlc_update_msgs(&nodes[8], &node_h_id);
+	assert_eq!(updates.update_fulfill_htlcs.len(), 6);
+	assert_eq!(updates.commitment_signed.len(), 1);
+	assert!(updates.update_add_htlcs.is_empty());
+	assert!(updates.update_fail_htlcs.is_empty());
+	assert!(updates.update_fail_malformed_htlcs.is_empty());
+	assert!(updates.update_fee.is_none());
 }
 
 #[test]