Stop cleaning monitor updates on new block connect
authorGursharan Singh <3442979+G8XSU@users.noreply.github.com>
Fri, 8 Dec 2023 19:28:19 +0000 (11:28 -0800)
committerGursharan Singh <3442979+G8XSU@users.noreply.github.com>
Thu, 14 Dec 2023 18:29:12 +0000 (10:29 -0800)
Previously, we used to cleanup monitor updates at both consolidation
threshold and new block connects. With this change we will only
cleanup when our consolidation criteria is met. Also, we remove
monitor read from cleanup logic, in case of update consolidation.
Note: In case of channel-closing monitor update, we still need to
read the old monitor before persisting the new one in order to
determine the cleanup range.

lightning/src/util/persist.rs

index a9f534ee4d3413df6489ea1988c13e75b13272b7..e63290620516ee0b6f62b7bb57c0c6d2b9369d49 100644 (file)
@@ -346,9 +346,10 @@ where
 ///
 /// # Pruning stale channel updates
 ///
-/// Stale updates are pruned when a full monitor is written. The old monitor is first read, and if
-/// that succeeds, updates in the range between the old and new monitors are deleted. The `lazy`
-/// flag is used on the [`KVStore::remove`] method, so there are no guarantees that the deletions
+/// Stale updates are pruned when the consolidation threshold is reached according to `maximum_pending_updates`.
+/// Monitor updates in the range between the latest `update_id` and `update_id - maximum_pending_updates`
+/// are deleted.
+/// The `lazy` flag is used on the [`KVStore::remove`] method, so there are no guarantees that the deletions
 /// will complete. However, stale updates are not a problem for data integrity, since updates are
 /// only read that are higher than the stored [`ChannelMonitor`]'s `update_id`.
 ///
@@ -610,24 +611,6 @@ where
        ) -> chain::ChannelMonitorUpdateStatus {
                // Determine the proper key for this monitor
                let monitor_name = MonitorName::from(funding_txo);
-               let maybe_old_monitor = self.read_monitor(&monitor_name);
-               match maybe_old_monitor {
-                       Ok((_, ref old_monitor)) => {
-                               // Check that this key isn't already storing a monitor with a higher update_id
-                               // (collision)
-                               if old_monitor.get_latest_update_id() > monitor.get_latest_update_id() {
-                                       log_error!(
-                                               self.logger,
-                                               "Tried to write a monitor at the same outpoint {} with a higher update_id!",
-                                               monitor_name.as_str()
-                                       );
-                                       return chain::ChannelMonitorUpdateStatus::UnrecoverableError;
-                               }
-                       }
-                       // This means the channel monitor is new.
-                       Err(ref e) if e.kind() == io::ErrorKind::NotFound => {}
-                       _ => return chain::ChannelMonitorUpdateStatus::UnrecoverableError,
-               }
                // Serialize and write the new monitor
                let mut monitor_bytes = Vec::with_capacity(
                        MONITOR_UPDATING_PERSISTER_PREPEND_SENTINEL.len() + monitor.serialized_length(),
@@ -641,65 +624,12 @@ where
                        &monitor_bytes,
                ) {
                        Ok(_) => {
-                               // Assess cleanup. Typically, we'll clean up only between the last two known full
-                               // monitors.
-                               if let Ok((_, old_monitor)) = maybe_old_monitor {
-                                       let start = old_monitor.get_latest_update_id();
-                                       let end = if monitor.get_latest_update_id() == CLOSED_CHANNEL_UPDATE_ID {
-                                               // We don't want to clean the rest of u64, so just do possible pending
-                                               // updates. Note that we never write updates at
-                                               // `CLOSED_CHANNEL_UPDATE_ID`.
-                                               cmp::min(
-                                                       start.saturating_add(self.maximum_pending_updates),
-                                                       CLOSED_CHANNEL_UPDATE_ID - 1,
-                                               )
-                                       } else {
-                                               monitor.get_latest_update_id().saturating_sub(1)
-                                       };
-                                       // We should bother cleaning up only if there's at least one update
-                                       // expected.
-                                       for update_id in start..=end {
-                                               let update_name = UpdateName::from(update_id);
-                                               #[cfg(debug_assertions)]
-                                               {
-                                                       if let Ok(update) =
-                                                               self.read_monitor_update(&monitor_name, &update_name)
-                                                       {
-                                                               // Assert that we are reading what we think we are.
-                                                               debug_assert_eq!(update.update_id, update_name.0);
-                                                       } else if update_id != start && monitor.get_latest_update_id() != CLOSED_CHANNEL_UPDATE_ID
-                                                       {
-                                                               // We're deleting something we should know doesn't exist.
-                                                               panic!(
-                                                                       "failed to read monitor update {}",
-                                                                       update_name.as_str()
-                                                               );
-                                                       }
-                                                       // On closed channels, we will unavoidably try to read
-                                                       // non-existent updates since we have to guess at the range of
-                                                       // stale updates, so do nothing.
-                                               }
-                                               if let Err(e) = self.kv_store.remove(
-                                                       CHANNEL_MONITOR_UPDATE_PERSISTENCE_PRIMARY_NAMESPACE,
-                                                       monitor_name.as_str(),
-                                                       update_name.as_str(),
-                                                       true,
-                                               ) {
-                                                       log_error!(
-                                                               self.logger,
-                                                               "error cleaning up channel monitor updates for monitor {}, reason: {}",
-                                                               monitor_name.as_str(),
-                                                               e
-                                                       );
-                                               };
-                                       }
-                               };
                                chain::ChannelMonitorUpdateStatus::Completed
                        }
                        Err(e) => {
                                log_error!(
                                        self.logger,
-                                       "error writing channel monitor {}/{}/{} reason: {}",
+                                       "Failed to write ChannelMonitor {}/{}/{} reason: {}",
                                        CHANNEL_MONITOR_PERSISTENCE_PRIMARY_NAMESPACE,
                                        CHANNEL_MONITOR_PERSISTENCE_SECONDARY_NAMESPACE,
                                        monitor_name.as_str(),
@@ -741,7 +671,7 @@ where
                                        Err(e) => {
                                                log_error!(
                                                        self.logger,
-                                                       "error writing channel monitor update {}/{}/{} reason: {}",
+                                                       "Failed to write ChannelMonitorUpdate {}/{}/{} reason: {}",
                                                        CHANNEL_MONITOR_UPDATE_PERSISTENCE_PRIMARY_NAMESPACE,
                                                        monitor_name.as_str(),
                                                        update_name.as_str(),
@@ -751,8 +681,41 @@ where
                                        }
                                }
                        } else {
-                               // We could write this update, but it meets criteria of our design that call for a full monitor write.
-                               self.persist_new_channel(funding_txo, monitor, monitor_update_call_id)
+                               let monitor_name = MonitorName::from(funding_txo);
+                               // In case of channel-close monitor update, we need to read old monitor before persisting
+                               // the new one in order to determine the cleanup range.
+                               let maybe_old_monitor = match monitor.get_latest_update_id() {
+                                       CLOSED_CHANNEL_UPDATE_ID => self.read_monitor(&monitor_name).ok(),
+                                       _ => None
+                               };
+
+                               // We could write this update, but it meets criteria of our design that calls for a full monitor write.
+                               let monitor_update_status = self.persist_new_channel(funding_txo, monitor, monitor_update_call_id);
+
+                               if let chain::ChannelMonitorUpdateStatus::Completed = monitor_update_status {
+                                       let cleanup_range = if monitor.get_latest_update_id() == CLOSED_CHANNEL_UPDATE_ID {
+                                               // If there is an error while reading old monitor, we skip clean up.
+                                               maybe_old_monitor.map(|(_, ref old_monitor)| {
+                                                       let start = old_monitor.get_latest_update_id();
+                                                       // We never persist an update with update_id = CLOSED_CHANNEL_UPDATE_ID
+                                                       let end = cmp::min(
+                                                               start.saturating_add(self.maximum_pending_updates),
+                                                               CLOSED_CHANNEL_UPDATE_ID - 1,
+                                                       );
+                                                       (start, end)
+                                               })
+                                       } else {
+                                               let end = monitor.get_latest_update_id();
+                                               let start = end.saturating_sub(self.maximum_pending_updates);
+                                               Some((start, end))
+                                       };
+
+                                       if let Some((start, end)) = cleanup_range {
+                                               self.cleanup_in_range(monitor_name, start, end);
+                                       }
+                               }
+
+                               monitor_update_status
                        }
                } else {
                        // There is no update given, so we must persist a new monitor.
@@ -761,6 +724,34 @@ where
        }
 }
 
+impl<K: Deref, L: Deref, ES: Deref, SP: Deref> MonitorUpdatingPersister<K, L, ES, SP>
+where
+       ES::Target: EntropySource + Sized,
+       K::Target: KVStore,
+       L::Target: Logger,
+       SP::Target: SignerProvider + Sized
+{
+       // Cleans up monitor updates for given monitor in range `start..=end`.
+       fn cleanup_in_range(&self, monitor_name: MonitorName, start: u64, end: u64) {
+               for update_id in start..=end {
+                       let update_name = UpdateName::from(update_id);
+                       if let Err(e) = self.kv_store.remove(
+                               CHANNEL_MONITOR_UPDATE_PERSISTENCE_PRIMARY_NAMESPACE,
+                               monitor_name.as_str(),
+                               update_name.as_str(),
+                               true,
+                       ) {
+                               log_error!(
+                                       self.logger,
+                                       "Failed to clean up channel monitor updates for monitor {}, reason: {}",
+                                       monitor_name.as_str(),
+                                       e
+                               );
+                       };
+               }
+       }
+}
+
 /// A struct representing a name for a monitor.
 #[derive(Debug)]
 struct MonitorName(String);
@@ -896,20 +887,21 @@ mod tests {
        #[test]
        fn persister_with_real_monitors() {
                // This value is used later to limit how many iterations we perform.
-               let test_max_pending_updates = 7;
+               let persister_0_max_pending_updates = 7;
+               // Intentionally set this to a smaller value to test a different alignment.
+               let persister_1_max_pending_updates = 3;
                let chanmon_cfgs = create_chanmon_cfgs(4);
                let persister_0 = MonitorUpdatingPersister {
                        kv_store: &TestStore::new(false),
                        logger: &TestLogger::new(),
-                       maximum_pending_updates: test_max_pending_updates,
+                       maximum_pending_updates: persister_0_max_pending_updates,
                        entropy_source: &chanmon_cfgs[0].keys_manager,
                        signer_provider: &chanmon_cfgs[0].keys_manager,
                };
                let persister_1 = MonitorUpdatingPersister {
                        kv_store: &TestStore::new(false),
                        logger: &TestLogger::new(),
-                       // Intentionally set this to a smaller value to test a different alignment.
-                       maximum_pending_updates: 3,
+                       maximum_pending_updates: persister_1_max_pending_updates,
                        entropy_source: &chanmon_cfgs[1].keys_manager,
                        signer_provider: &chanmon_cfgs[1].keys_manager,
                };
@@ -934,7 +926,6 @@ mod tests {
                node_cfgs[1].chain_monitor = chain_mon_1;
                let node_chanmgrs = create_node_chanmgrs(2, &node_cfgs, &[None, None]);
                let nodes = create_network(2, &node_cfgs, &node_chanmgrs);
-
                let broadcaster_0 = &chanmon_cfgs[2].tx_broadcaster;
                let broadcaster_1 = &chanmon_cfgs[3].tx_broadcaster;
 
@@ -957,10 +948,11 @@ mod tests {
                                for (_, mon) in persisted_chan_data_0.iter() {
                                        // check that when we read it, we got the right update id
                                        assert_eq!(mon.get_latest_update_id(), $expected_update_id);
-                                       // if the CM is at the correct update id without updates, ensure no updates are stored
+
+                                       // if the CM is at consolidation threshold, ensure no updates are stored.
                                        let monitor_name = MonitorName::from(mon.get_funding_txo().0);
-                                       let (_, cm_0) = persister_0.read_monitor(&monitor_name).unwrap();
-                                       if cm_0.get_latest_update_id() == $expected_update_id {
+                                       if mon.get_latest_update_id() % persister_0_max_pending_updates == 0
+                                                       || mon.get_latest_update_id() == CLOSED_CHANNEL_UPDATE_ID {
                                                assert_eq!(
                                                        persister_0.kv_store.list(CHANNEL_MONITOR_UPDATE_PERSISTENCE_PRIMARY_NAMESPACE,
                                                                monitor_name.as_str()).unwrap().len(),
@@ -975,8 +967,9 @@ mod tests {
                                for (_, mon) in persisted_chan_data_1.iter() {
                                        assert_eq!(mon.get_latest_update_id(), $expected_update_id);
                                        let monitor_name = MonitorName::from(mon.get_funding_txo().0);
-                                       let (_, cm_1) = persister_1.read_monitor(&monitor_name).unwrap();
-                                       if cm_1.get_latest_update_id() == $expected_update_id {
+                                       // if the CM is at consolidation threshold, ensure no updates are stored.
+                                       if mon.get_latest_update_id() % persister_1_max_pending_updates == 0
+                                                       || mon.get_latest_update_id() == CLOSED_CHANNEL_UPDATE_ID {
                                                assert_eq!(
                                                        persister_1.kv_store.list(CHANNEL_MONITOR_UPDATE_PERSISTENCE_PRIMARY_NAMESPACE,
                                                                monitor_name.as_str()).unwrap().len(),
@@ -1001,7 +994,7 @@ mod tests {
                // Send a few more payments to try all the alignments of max pending updates with
                // updates for a payment sent and received.
                let mut sender = 0;
-               for i in 3..=test_max_pending_updates * 2 {
+               for i in 3..=persister_0_max_pending_updates * 2 {
                        let receiver;
                        if sender == 0 {
                                sender = 1;