Peer storage feature #2943

adi2011 · 2024-03-17T19:33:19Z

This would enable nodes to distribute small blobs among their peers, which could be retrieved to resume or force-close the channel operation upon data corruption.

Elaborated in detail here

It's far from complete right now, but please let me know if I am making any mistakes because I am very new to both Rust and LDK :)

lightning/src/ln/channelmanager.rs

TheBlueMatt · 2024-03-26T15:24:31Z

lightning/src/ln/channelmanager.rs

+			if let Some(funding_txo) = chan.funding_txo {
+				found_funded_chan = true;
+				let peer_storage_update = ChannelMonitorUpdate {
+					update_id: CLOSED_CHANNEL_UPDATE_ID,


Lets use a real update_id fetched (and incremented) from the channel.

Can't see any helper in ChannelContext to fetch_and_increment_update_id(), should I create one?

TheBlueMatt · 2024-03-26T15:25:03Z

lightning/src/ln/channelmanager.rs

+
+		let mut found_funded_chan = false;
+		for chan in &sorted_chan_info {
+			if let Some(funding_txo) = chan.funding_txo {


Lets also require the channel have confirmations (or be 0conf).

I am filtering ChannelPhase::Funded, that'd cover this I think?

lightning/src/ln/channelmanager.rs

adi2011 · 2024-04-01T17:37:08Z

Thanks for the review @TheBlueMatt, I was afk last week. I will work on these changes in this week.

lightning/src/ln/channelmanager.rs

lightning/src/chain/channelmonitor.rs

TheBlueMatt

Hey, sorry I missed your review request. I generally don't see them as Github doesn't generate email notifications for them.

lightning/src/ln/channel.rs

TheBlueMatt · 2024-05-14T14:13:41Z

lightning/src/ln/channelmanager.rs

+		let mut peer_storage = VecWriter(Vec::new());
+		self.our_peer_storage.read().unwrap().write(&mut peer_storage).unwrap();
+		let mut encrypted_blob = vec![0;peer_storage.0.len() + 16];
+		self.inbound_payment_key.encrypt_our_peer_storage(&mut encrypted_blob, 0u64, b"", &peer_storage.0[..]);


We should define a new key for this, the inbound_payment_key may be shared across multiple nodes.

Okay, I defined a new method inside NodeManager, get_peer_storage_key. It derives a key from node_secret through the KeysManager.

TheBlueMatt · 2024-05-14T14:14:04Z

lightning/src/ln/channelmanager.rs

+	pub fn get_encrypted_our_peer_storage(&self) -> Vec<u8> {
+		let mut peer_storage = VecWriter(Vec::new());
+		self.our_peer_storage.read().unwrap().write(&mut peer_storage).unwrap();
+		let mut encrypted_blob = vec![0;peer_storage.0.len() + 16];


nit: We shouldn't need a new vec here, we should be able to encrypt in-place.

TheBlueMatt · 2024-05-14T14:17:25Z

lightning/src/ln/channelmanager.rs

@@ -2095,6 +2160,7 @@ where
 	entropy_source: ES,
 	node_signer: NS,
 	signer_provider: SP,
+	our_peer_storage: FairRwLock<OurPeerStorage>,


I dont see where this is updated at runtime? Presumably we need to be able to update each channel when it receives a new RAA

Now this gets updated everytime we send a LatestCounterpartyCommitmentTXInfo.

TheBlueMatt · 2024-05-14T14:22:07Z

lightning/src/chain/channelmonitor.rs

 			best_block,
 			counterparty_node_id: Some(counterparty_node_id),
 			initial_counterparty_commitment_info: None,
 			balances_empty_height: None,
 		})
 	}

+	pub(crate) fn new_stub(secp_ctx: Secp256k1<secp256k1::All>, stub_channel: &StubChannel, best_block: BestBlock, keys: Signer, funding_info_scriptbuf: ScriptBuf) -> ChannelMonitor<Signer> {


I kinda wonder if we shouldn't move StubChannel to the chain module and refactor ChannelMonitor methods to operate on either a full ChannelMonitor or a StubChannel? Its at least something to explore as we go to test this logic and see if its simpler.

But we'd need the chain module to detect the funding spent and to sweep funds, maybe create new methods inside chain module to handle stubchannel?
Watch_stub?

TheBlueMatt

I find myself wondering what the right UX here is - we generally assume that users will not lose ChannelMonitor data, no matter what, and this seems mostly targeted at addressing the case where a user somehow loses a single (or a few) ChannelMonitors for closed channels (if the channels are open, we'll still fail to load).

I think really this feature is useful in one of two cases -

(a) we can use it to detect that we're actually running with stale state by trusting any one of our peers to tell us that, rather than trusting the peer with whom we have a stale channel to tell us our channel is state. This requires much less data (just the latest commitment numbers for each channel) and is a quick and cheap check in ChannelManager which we can respond to by simply panic'ing.

(b) Handling the case where we've lost all our (latest) data and we want to recover from just our cryptographic keys (and maybe some stale ChannelMonitors). In this case, we really don't want to start with a ChannelManager at all, we want to have some new RecoveryPeerConnector struct that handles peer connections and processes the backup data we're adding here.

I think both are relatively easy changes to this PR, though.

TheBlueMatt · 2024-07-23T13:56:44Z

lightning/src/chain/chainmonitor.rs

+		let entry = match stub_monitors.entry(funding_outpoint) {
+			hash_map::Entry::Occupied(_) => {
+				log_error!(logger, "Failed to add new channel data: channel monitor for given outpoint is already present");
+				return Err(());


We should see if this stub monitor has any new information that the previous one did not and update, no?

Thanks for pointing this out. Fixed it.

TheBlueMatt · 2024-07-23T13:59:47Z

lightning/src/chain/channelmonitor.rs

+		Self::from_impl(ChannelMonitorImpl {
+			latest_update_id: STUB_CHANNEL_UPDATE_IDENTIFIER,
+			commitment_transaction_number_obscure_factor: stub_channel.obscure_factor,
+			destination_script: ScriptBuf::new(),


We absolutely need a destination_script to know where to send money.

TheBlueMatt · 2024-07-23T14:02:07Z

lightning/src/chain/channelmonitor.rs

+			funding_redeemscript: ScriptBuf::new(),
+			channel_value_satoshis: stub_channel.channel_value_stoshis,
+			their_cur_per_commitment_points: stub_channel.their_cur_per_commitment_points,
+			on_holder_tx_csv: 1,


We need this to know when we can claim funds (and I think also build witnesses)?

TheBlueMatt · 2024-07-23T14:53:30Z

lightning/src/ln/channelmanager.rs

@@ -2232,7 +2237,8 @@ where
 	entropy_source: ES,
 	node_signer: NS,
 	signer_provider: SP,
-
+	our_peer_storage: FairRwLock<OurPeerStorage>,
+	peer_storage: Mutex<HashMap<PublicKey, Vec<u8>>>,


Shouldn't this be in a PeerHolder?

We can shift it to PeerState, but i think handling it here is simpler?

Putting it here is another thing for us to have to clean up when we no longer have channels. Putting it with the rest of the peer state consolidates logic and simplifies things.

TheBlueMatt · 2024-07-23T14:54:05Z

lightning/src/ln/channelmanager.rs

@@ -7432,6 +7449,77 @@ where
 		}
 	}

+	fn internal_peer_storage(&self, counterparty_node_id: &PublicKey, msg: &msgs::PeerStorageMessage) {
+		let per_peer_state = self.per_peer_state.write().unwrap();


Why does this need a write lock?

My bad, thanks for pointing this out :)

TheBlueMatt · 2024-07-23T15:06:02Z

lightning/src/ln/channelmanager.rs

+
+ 		let mut res = vec![0; msg.data.len() - 16];
+
+		match our_peer_storage.decrypt_our_peer_storage(&mut res, msg.data.as_slice(), self.our_peerstorage_encryption_key) {


Hmm, there's gotta be a more efficient way to handle this message. We're expected to receive it every time we connect, and peers may well send it more often. The ChainMonitor can be smarter here, which may be the best way to go about it - pass the whole blob over the fence and let it decode them/check if it has corresponding ChannelMonitors for everything before deriving the new keys, etc.

Yes, that's better I think. So I'll write a function which processes the blob and only create the keys if we need to panic and persist the channel.

Starting to think instead of doing this we should just compare the decrypted contents with the state of our live channels. Sure, its not entirely robust, but it should be pretty good (and, notably, means at least our current channels are safe to update), but it also avoids changing the interface with ChainMonitor, which would be really nice.

TheBlueMatt · 2024-07-23T15:12:46Z

lightning/src/ln/channelmanager.rs

@@ -9978,6 +10108,16 @@ where
 		let _ = handle_error!(self, self.internal_funding_signed(counterparty_node_id, msg), *counterparty_node_id);
 	}

+	fn handle_peer_storage(&self, counterparty_node_id: &PublicKey, msg: &msgs::PeerStorageMessage) {
+		let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);


Suggested change

let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);

// There's no reason to cause a full persistence just because a peer updated their storage, and it will

// never create new events, so just skip all notification.

let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || NotifyOption::SkipPersistNoEvents);

TheBlueMatt · 2024-07-23T15:13:25Z

lightning/src/ln/channelmanager.rs

 				let pending_msg_events = &mut peer_state.pending_msg_events;
+				let peer_storage = self.peer_storage.lock().unwrap().get(counterparty_node_id).unwrap_or(&Vec::<u8>::new()).clone();
+
+				if peer_storage.len() > 0 {


I dont' see why we should only send if the length is > 0, shouldn't we send if there's any value in the map at all?

fixed this. Thanks :)

TheBlueMatt · 2024-07-23T15:15:04Z

lightning/src/ln/channelmanager.rs

+		(peer_storage.len() as u64).write(writer)?;
+		for (node_id, peer_data) in peer_storage.iter() {
+			node_id.write(writer)?;
+			peer_data.write(writer)?;


This changes the encoding of ChannelManager such that we cannot upgrade/downgrade. Instead, new data have to be written as TLVs.

TheBlueMatt · 2024-07-23T15:17:13Z

lightning/src/ln/channelmanager.rs

+				channel.context.get_commitment_txn_number_obscure_factor(),
+				new_hash_map(),
+				None,
+				channel.context.channel_transaction_parameters.channel_type_features.clone(),


Hum, why are we creating a stub channel for all of our existing channels on each startup? They should already exist as full ChannelMonitors so this is redundant, right?

This is used to build our own latest backup which would be distributed to our peers.

TheBlueMatt · 2024-09-10T19:59:29Z

lightning/src/ln/channelmanager.rs

@@ -2232,7 +2237,8 @@ where
 	entropy_source: ES,
 	node_signer: NS,
 	signer_provider: SP,
-
+	our_peer_storage: FairRwLock<OurPeerStorage>,
+	peer_storage: Mutex<HashMap<PublicKey, Vec<u8>>>,


Putting it here is another thing for us to have to clean up when we no longer have channels. Putting it with the rest of the peer state consolidates logic and simplifies things.

TheBlueMatt · 2024-09-10T19:59:37Z

lightning/src/ln/channelmanager.rs

+/// - [`ChannelMessageHandler`] to handle off-chain channel activity from peers
+/// - [`MessageSendEventsProvider`] to similarly send such messages to peers
+///
+pub struct FundRecoverer<SP: Deref, L:Deref, M: Deref> 


Lets put this in a new module/file.

TheBlueMatt · 2024-09-10T20:00:32Z

lightning/src/chain/mod.rs

@@ -305,6 +305,55 @@ pub trait Watch<ChannelSigner: EcdsaChannelSigner> {
 	/// For details on asynchronous [`ChannelMonitor`] updating and returning
 	/// [`MonitorEvent::Completed`] here, see [`ChannelMonitorUpdateStatus::InProgress`].
 	fn release_pending_monitor_events(&self) -> Vec<(OutPoint, ChannelId, Vec<MonitorEvent>, Option<PublicKey>)>;
+
+	/// Watches a dummy channel identified by `funding_txo` using `monitor`.


Now that we're using a different struct from ChannelManager, IMO we should use a different persistence trait.

TheBlueMatt · 2024-09-10T20:01:27Z

lightning/src/ln/channelmanager.rs

+	SP::Target: SignerProvider,
+	L::Target: Logger
+{
+	fn handle_open_channel(&self, _their_node_id: &PublicKey, _msg: &msgs::OpenChannel) {}


We should reject open channels (see how IgnoringMessageHandler deals with different messages and ~copy that).

TheBlueMatt · 2024-09-10T20:15:47Z

lightning/src/ln/channelmanager.rs

+
+ 		let mut res = vec![0; msg.data.len() - 16];
+
+		match our_peer_storage.decrypt_our_peer_storage(&mut res, msg.data.as_slice(), self.our_peerstorage_encryption_key) {


Starting to think instead of doing this we should just compare the decrypted contents with the state of our live channels. Sure, its not entirely robust, but it should be pretty good (and, notably, means at least our current channels are safe to update), but it also avoids changing the interface with ChainMonitor, which would be really nice.

TheBlueMatt · 2024-09-10T20:24:01Z

lightning/src/ln/channelmanager.rs

@@ -8272,9 +8565,15 @@ where
 						} else { false };
 						let (htlcs_to_fail, monitor_update_opt) = try_chan_phase_entry!(self,
 							chan.revoke_and_ack(&msg, &self.fee_estimator, &&logger, mon_update_blocked), chan_phase_entry);
+
+						let mut our_peer_storage = self.our_peer_storage.write().unwrap();
+						let _ = our_peer_storage.provide_secret(chan.context.channel_id(), chan.get_cur_counterparty_commitment_transaction_number() + 1, msg.per_commitment_secret);


This (and all the other update methods) is actually too early. We allow ChannelMonitorUpdates to be persisted asynchronously, which means we could (a) get a new ChannelMonitorUpdate, updating our peer state, (b) send that new peer state to our peers, (c) restart without finishing persisting the ChannelMonitorUpdate (or ChannelManager), then (d) on restart we're in a perfectly fine state (we haven't sent the new channel messages to our peer, only the new state) but we'll still think we're stale and panic.

Instead, we need to process changes to the channel(s) when we complete ChannelMonitorUpdates (removing them from the pending set in Channel via monitor_updating_restored in channelmanager.rs's handle_monitor_update_completion).

TheBlueMatt · 2024-10-01T15:46:13Z

lightning/src/ln/channelmanager.rs

+		}
+
+		log_trace!(logger, "Received Peer Storage from {}", log_pubkey!(counterparty_node_id));
+		peer_state.peer_storage = msg.data.clone();


I don't think we actually should be willing to store a full 64KiB per peer. Sadly, LDK resilvers ChannelManager very often (something we're working on reducing, but we're a ways from being able to change it), and currently store only ~ 1KiB per channel. If we allow all our peers to store 64KiB, we may ~50x our ChannelManager size, causing a very substantial increase in I/O. For now, we can just limit the max size we accept (maybe 1KiB?) and ignore larger messages, but in a followup we can look at some other I/O interface for storing these in a separate K-V entry.

TheBlueMatt · 2024-10-01T15:47:30Z

lightning/src/chain/channelmonitor.rs

+/// This update ID is used inside [`ChannelMonitorImpl`] to recognise
+/// that we're dealing with a [`StubChannelMonitor`]. Since we require some
+/// exceptions while dealing with it.
+pub const STUB_CHANNEL_UPDATE_IDENTIFIER: u64 = CLOSED_CHANNEL_UPDATE_ID - 1;


Lets set a flag separately. I'm working on removing CLOSED_CHANNEL_UPDATE_ID entirely.

TheBlueMatt · 2024-10-01T15:50:22Z

lightning/src/chain/channelmonitor.rs

+/// This includes timestamp to compare between two given 
+/// [`OurPeerStorage`] and version defines the structure.
+#[derive(Clone, PartialEq, Eq)]
+pub struct OurPeerStorage {


Lets move this into its own file.

TheBlueMatt · 2024-10-01T15:50:38Z

lightning/src/chain/channelmonitor.rs

+	/// [`ChannelMonitorUpdateStep::LatestCounterpartyCommitmentTXInfo`]
+	pub(crate) fn update_state_from_monitor_update(&mut self, cid: ChannelId, monitor_update: ChannelMonitorUpdate) -> Result<(),()> {
+		for update in monitor_update.updates.iter() {
+			match update {


Should this logic move into StubChannel instead?

TheBlueMatt · 2024-10-01T15:50:55Z

lightning/src/chain/channelmonitor.rs

+	pub(crate) features: ChannelTypeFeatures,
+}
+
+impl StubChannel {


To better match LDK terminology, should this be a StubChannelMonitor rather than a StubChannel?

TheBlueMatt · 2024-10-01T16:22:15Z

lightning/src/ln/fundrecoverer.rs

+
+		let persist_res = self
+			.persister
+			.persist_new_channel(stub_channel_monitor.get_funding_txo().0, &stub_channel_monitor);


This is gonna fail if we already have a monitor for this channel, but we should be trying to update that monitor. We may have a somewhat random mix of real (possibly stale) monitors and peer-storage (possibly stale) stubs, and we ideally need to merge all the data we have to be able to recover in the best way possible.

TheBlueMatt · 2024-10-01T16:22:50Z

lightning/src/ln/fundrecoverer.rs

+			funding_txid_u32.wrapping_add(best_height.unwrap_or_default())
+		};
+
+		let partition_factor = if channel_count < 15 {


Don't think we need to bother with this optimization in the recovery stuff.

TheBlueMatt · 2024-10-01T16:23:29Z

lightning/src/ln/fundrecoverer.rs

+		let funding_outpoints = hash_set_from_iter(self.monitors.read().unwrap().keys().cloned());
+		let channel_count = funding_outpoints.len();
+		for funding_outpoint in funding_outpoints.iter() {
+			let monitor_lock = self.monitors.read().unwrap();


It'd be nice if a lot of the code in this file were DRYd with the code elsewhere.

TheBlueMatt · 2024-10-01T16:25:26Z

lightning/src/ln/fundrecoverer.rs

+		}
+
+		let mut monitors = self.monitors.write().unwrap();
+		let entry = match monitors.entry(stub_channel_monitor.get_funding_txo().0) {


Somehow we need to replay blockchain data after we add a new monitor here. The flow needs to be (a) load monitor(s) from peer state, (b) replay any chain parts we need on the new monitor (c) persist it.

TheBlueMatt · 2024-10-01T16:25:38Z

lightning/src/ln/fundrecoverer.rs

+		};
+
+		let persist_res = self
+			.persister


Once we persist a stub monitor here, the ChannelManager should fail to load. This should be pretty doable - just have the ChannelManager check for the stub flag on any ChannelMonitors in read and refuse to load if any are set.

…text.

…ages.

…to the respective peer upon reconnection.

…erytime a LatestCounterpartyCommitmentTxn update is sent. It would be encrypted and sent to our peers.

… or decrpt the peerstorage and send PeerStorage on every RAA and upon reconnection.

…bbed to recover from PeerStorage.

…riate helpers to reload it.

… so that we can just send a BogusChannelReestablish and close all the StubChannelMonitors and sweep the funds from the events.

… peer storage and sweep funds correctly.

…rom a revoked state.

TheBlueMatt

Okay, only one major question and a handful of smaller comments. I think this is structurally correct aside from the one major question, so it may be worth starting to clean this up a bit - new files need to be rustfmt'd (eg by running contrib/run-rustfmt.sh), all files should be tab-indented, git commits should follow a logical progression and not change things introduced in earlier commits, they should have details about why we're doing what we're doing (see https://cbea.ms/git-commit/).

TheBlueMatt · 2024-10-18T18:05:07Z

lightning/src/chain/channelmonitor.rs

+/// This update ID is used inside [`ChannelMonitorImpl`] to recognise
+/// that we're dealing with a [`StubChannelMonitor`]. Since we require some
+/// exceptions while dealing with it.
+pub const STUB_CHANNEL_UPDATE_IDENTIFIER: u64 = core::u64::MAX - 1;


Why not just use CLOSED_CHANNEL_UPDATE_ID? We already have handling for treating that as "final" which it makes sense to reuse wholly.

TheBlueMatt · 2024-10-18T18:07:40Z

lightning/src/chain/channelmonitor.rs

@@ -4253,6 +4393,9 @@ impl<Signer: EcdsaChannelSigner> ChannelMonitorImpl<Signer> {
 					if *idx == input.previous_output.vout {
 						#[cfg(test)]
 						{
+							if self.latest_update_id == STUB_CHANNEL_UPDATE_IDENTIFIER {


This looks like its working around a bug but only in tests? Probably should delete it and fix the bug in prod too :)

TheBlueMatt · 2024-10-18T18:07:54Z

lightning/src/events/mod.rs

@@ -2298,6 +2298,13 @@ impl MaybeReadable for Event {
 	}
 }

+#[derive(Clone, Debug, PartialEq, Eq)]
+pub enum RecoveryEvent {


This needs docs :)

TheBlueMatt · 2024-10-18T18:08:55Z

lightning/src/events/mod.rs

@@ -2298,6 +2298,13 @@ impl MaybeReadable for Event {
 	}
 }

+#[derive(Clone, Debug, PartialEq, Eq)]


This can probably just move to fundrecoverer.rs - its only used there there's no reason for it to be in the global events module.

TheBlueMatt · 2024-10-18T18:09:13Z

lightning/src/events/mod.rs

@@ -2630,6 +2654,10 @@ pub trait EventsProvider {
 	fn process_pending_events<H: Deref>(&self, handler: H) where H::Target: EventHandler;
 }

+pub trait RecoverEventsProvider: EventsProvider {


This is only (ever going to be) implemented by one struct so I'm not sure its worth having a trait. We could have a trait if we use it to force the user to implement both RecoveryEvent and Event handling in one struct (to ensure they don't forget to do both), but then we'd have to un-implement EventsProvider on FundRecoverer. Not really sure its worth it, users have to not forget to fetch events...

TheBlueMatt · 2024-10-18T20:46:44Z

lightning/src/ln/channelmanager.rs

+				Some(mutex) => mutex,
+				None => {
+					log_debug!(logger, "Not able to find peer_state for the counterparty {}, channelId {}", log_pubkey!(ps_channel.counterparty_node_id), ps_channel.channel_id);
+					panic!("Found a missing channel {} through peer storage", ps_channel.channel_id);


Its not a major error worth panicking over, it may just indicate that the peer gave us an old state of ours. Same goes for the below copy.

TheBlueMatt · 2024-10-18T20:56:36Z

lightning/src/ln/fundrecoverer.rs

+use core::cell::RefCell;
+use core::ops::Deref;
+
+// Re-export this for use in the public API.


TheBlueMatt · 2024-10-18T21:03:33Z

lightning/src/ln/fundrecoverer.rs

+
+	fn handle_channel_reestablish(
+		&self, _their_node_id: PublicKey, _msg: &msgs::ChannelReestablish,
+	) {


This needs to respond with a ChannelReestablish with the dummy values from above to ask the peer to FC.

TheBlueMatt · 2024-10-18T21:03:47Z

lightning/src/ln/fundrecoverer.rs

+				);
+
+				let per_peer_state = self.per_peer_state.read().unwrap();
+				if let Some(peer_state_mutex) = per_peer_state.get(&counterparty_node_id) {


Shouldn't this go in the peer_connected handler?

TheBlueMatt · 2024-10-18T21:06:09Z

lightning/src/ln/fundrecoverer.rs

+					if holder.monitor.get_min_seen_secret()
+						> stub_channel_monitor.get_min_seen_secret()
+					{
+						holder.monitor.merge_commitment_secret(stub_channel_monitor);


Just call this on all monitors. It may be that the ChannelMonitor we do have is stale (but is not a stub) and we have more revocation secrets here than we used to.

adi2011 force-pushed the PeerStorageFeature branch 3 times, most recently from b798c8e to c783fd3 Compare March 17, 2024 20:10

TheBlueMatt reviewed Mar 26, 2024

View reviewed changes

adi2011 force-pushed the PeerStorageFeature branch 2 times, most recently from b7f4bbc to 6ecf1b8 Compare April 14, 2024 22:16

coderabbitai bot previously approved these changes Apr 14, 2024

View reviewed changes

adi2011 commented Apr 14, 2024

View reviewed changes

lightning/src/ln/channelmanager.rs Show resolved Hide resolved

adi2011 commented Apr 14, 2024

View reviewed changes

lightning/src/chain/channelmonitor.rs Show resolved Hide resolved

jkczyz assigned TheBlueMatt Apr 29, 2024

adi2011 dismissed coderabbitai[bot]’s stale review via 4b570be May 6, 2024 12:47

adi2011 force-pushed the PeerStorageFeature branch 4 times, most recently from a23367f to 8933816 Compare May 6, 2024 14:38

adi2011 requested a review from TheBlueMatt May 6, 2024 14:38

TheBlueMatt reviewed May 14, 2024

View reviewed changes

adi2011 force-pushed the PeerStorageFeature branch 3 times, most recently from 1317384 to d609779 Compare July 14, 2024 17:46

adi2011 marked this pull request as ready for review July 14, 2024 17:48

adi2011 force-pushed the PeerStorageFeature branch 3 times, most recently from 4a428cb to 31b4912 Compare July 17, 2024 09:15

adi2011 requested a review from TheBlueMatt July 20, 2024 22:08

adi2011 force-pushed the PeerStorageFeature branch 2 times, most recently from a7a12b0 to 6c893d6 Compare July 20, 2024 22:20

TheBlueMatt reviewed Jul 23, 2024

View reviewed changes

adi2011 force-pushed the PeerStorageFeature branch from 6c893d6 to c5c8f06 Compare September 6, 2024 19:34

adi2011 force-pushed the PeerStorageFeature branch from c5c8f06 to 383e41a Compare September 6, 2024 20:22

adi2011 requested a review from TheBlueMatt September 6, 2024 20:23

adi2011 force-pushed the PeerStorageFeature branch 4 times, most recently from 0ea098a to 1574e6d Compare September 7, 2024 11:04

TheBlueMatt reviewed Sep 10, 2024

View reviewed changes

adi2011 force-pushed the PeerStorageFeature branch 3 times, most recently from 5bb20de to 9acd198 Compare September 25, 2024 08:19

TheBlueMatt reviewed Oct 1, 2024

View reviewed changes

adi2011 force-pushed the PeerStorageFeature branch from 9acd198 to 324f57c Compare October 13, 2024 21:59

Aditya Sharma added 11 commits October 14, 2024 03:31

lightning: Add provide_peer_backup_storage to InitContext and NodeCon…

f958657

…text.

lightning: Add message types and their handlers for peer storage mess…

ad13398

…ages.

lightning: Handle peer storage message, it's persistance and send it …

6189a7f

…to the respective peer upon reconnection.

lightning: Add struct for building OurPeerStorage, it gets updated ev…

bab7fcd

…erytime a LatestCounterpartyCommitmentTxn update is sent. It would be encrypted and sent to our peers.

lightning: Add a key inside NodeSigner which would be used to encrypt…

1923863

… or decrpt the peerstorage and send PeerStorage on every RAA and upon reconnection.

channelmonitor: Create StubChannelMonitor so that channels can be stu…

02737f7

…bbed to recover from PeerStorage.

chainmonitor: Add persistence logic for StubChannelMonitor and approp…

e3b2ac9

…riate helpers to reload it.

lightning: Handle your_peer_storage from our peers.

44483f8

channelmanager: Create FundRecoverer to take our node in offline mode…

1ee4c6e

… so that we can just send a BogusChannelReestablish and close all the StubChannelMonitors and sweep the funds from the events.

functional_tests: Add test_peer_storage to confirm if we recover from…

da9085d

… peer storage and sweep funds correctly.

functional_tests: Add test to check if the node is recovering funds f…

fe6e83c

…rom a revoked state.

adi2011 force-pushed the PeerStorageFeature branch from 324f57c to fe6e83c Compare October 13, 2024 22:01

adi2011 requested a review from TheBlueMatt October 13, 2024 22:06

TheBlueMatt reviewed Oct 18, 2024

View reviewed changes

thomash-acinq mentioned this pull request Nov 26, 2024

Peer storage ACINQ/eclair#2888

Open


		let mut res = vec![0; msg.data.len() - 16];

		match our_peer_storage.decrypt_our_peer_storage(&mut res, msg.data.as_slice(), self.our_peerstorage_encryption_key) {

-		let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
+		// There's no reason to cause a full persistence just because a peer updated their storage, and it will
+		// never create new events, so just skip all notification.
+		let _persistence_guard = PersistenceNotifierGuard::optionally_notify(self, || NotifyOption::SkipPersistNoEvents);

Peer storage feature #2943

Are you sure you want to change the base?

Peer storage feature #2943

Conversation

adi2011 commented Mar 17, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

adi2011 commented Apr 1, 2024

TheBlueMatt left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TheBlueMatt left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TheBlueMatt left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

adi2011 commented Mar 17, 2024 •

edited

Loading