This commit passes the peer's quit signal to the
gossipSyncer when attempt to hand off gossip query
messages. This allows a rate-limited peer's read
handler to break out immediately, which would
otherwise remain stuck until the rate-limited
gossip syncer pulled the message.
This commit restructures the delivery of gossip
query related messages, such that they are delivered
directly to the gossip syncers. Gossip query rate
limiting was introduced in #1824 on a per-peer basis.
However, since all gossip query messages were being
delivered in the main event loop, the end result is
that one rate-limited peer could stall all other
peers.
In addition, since no other peers would be able to
submit gossip-related messages through the blocked
event loop, the back pressure would eventually rate
limit the read handlers of all peers as well.
The end result would be lengthy delays in reading
messages related to htlc forwarding.
The fix is to lift the delivery of gossip query
messages outside of the main event loop. With
this change, the rate limiting backpressure is
delivered only to the intended peer.
In this commit, we modify the gossiper to no longer broadcast
NodeAnnouncements of nodes who intend to remain private. We do this to
prevent leaking their information to the greater network.
Previously, gossiper was the only object that validated channel
updates. Because updates can also be received as part of a
failed payment session in the routing package, validation logic
needs to be available there too. Gossiper already depends on
routing and having routing call the validation logic inside
gossiper would be a circular dependency. Therefore the validation
was moved to routing.
This commit removes the fallback in fetchGossipSyncer
that creates a gossip syncer if one is not registered
w/in the gossiper. Now that we register gossip syncers
explicitly before reading any gossip query messages,
this should not longer be required. The fallback also
did not honor the cfg.NoChanUpdates flag, which may
have led to inconsistencies between configuration and
actual behavior.
restransmitStaleChannels
In this commit, we add an additional error check for
ErrNoGraphEdgesFound when restransmitting stale channels during the
gossiper's startup. We do this to prevent benign log messages as we'll
log that we were unable to retransmit stale channels when we didn't have
any channels in our graph to begin with.
In this commit, we aim to resolve an issue with nodes requesting for
channel announcements when receiving a channel update for a channel
they're not aware of. This can happen if a node is not caught up with
the chain or if they receive updates for zombie channels. This would
lead to a spam issue, as if a node is not caught up with the chain,
every new update they receive is premature, causing them to manually
request the backing channel announcement. Ideally, we should be able to
detect this as a potential DoS vector and ban the node responsible, but
for now we'll simply remove this functionality.
In this commit, we select on the peer's QuitSignal to allow the caller
to unblock if the peer itself is disconnecting. With this change, we now
ensure that it isn't possible for a peer to block on this method and
prevent a graceful exit.
Previosuly we would immediately return nil on the error channel for
premature ChannelUpdates, which would break the expection that a a
returned non-error meant the update was successfully added to the
database. This meant that the caller would believe the update was added
to the database, while it is actually still in volatile memory and can
be lost during restarts.
This change makes us handle premature ChannelUpdates as we handle other
premature announcements within the gossiper, by deferring sending on the
error channel until we have reprocessed the update.
Previously we wouldn't return anything in the case where the
announcement were meant for a chain we didn't recognize. After this
change we should return an error on the error channel in all flows
within the gossiper.
Corrects an instance that holds a reference to a boltdb
byte slice after returning from the transaction. This
can cause panics under certain conditions, which is
avoided by creating a copy of the key.
In this commit, we allow the gossiper syncer to store the chunk size for
its respective encoding type. We do this to prevent a race condition
that would arise within the unit tests by modifying the values of the
encodingTypeToChunkSize map to allow for easier testing.
In this commit, we fix the logging when adding new gossip syncers. The
old log would log the byte array, rather than the byte slice. We fix
this by slicing before logging.
This commit changes the gossiper to direct messages to
peer objects, instead of sending them through the
server every time. The primary motivation is to reduce
contention on the server's mutex and, more importantly,
avoid deadlocks in the Triangle of Death.
In this commit, we fix an existing deadlock in the
gossiper->server->peer pipeline by ensuring that we're not holding the
syncer mutex while we attempt to have a syncer filter out the rest of
gossip messages.
In this commit we fix an existing bug caused by a scheduling race
condition. We'll now ensure that if we get a gossip message from a peer
before we create an instance for it, then we create one on the spot so
we can service the message. Before this commit, we would drop the first
message, and therefore never sync up with the peer at all, causing them
to miss channel announcements.
In this commit, we extend the AuthenticatedGossiper to take advantage of
the new query features in the case that it gets a channel update w/o
first receiving the full channel announcement. If this happens, we'll
attempt to find a syncer that's fully synced, and request the channel
announcement from it.
In this commit, we create a new concrete implementation for the new
discovery.ChannelGraphTimeSeries interface. We also export the
createChannelAnnouncement method to allow the chanSeries struct to
re-use the existing code for creating wire messages from the database
structs.
In this commit, we update the logic in the AuthenticatedGossiper to
ensure that can properly create, manage, and dispatch messages to any
gossipSyncer instances created by the server.
With this set of changes, the gossip now has complete knowledge of the
current set of peers we're conneted to that support the new range
queries. Upon initial connect, InitSyncState will be called by the
server if the new peer understands the set of gossip queries. This will
then create a new spot in the peerSyncers map for the new syncer. For
each new gossip query message, we'll then attempt to dispatch the
message directly to the gossip syncer. When the peer has disconnected,
we then expect the server to call the PruneSyncState method which will
allow us to free up the resources.
Finally, when we go to broadcast messages, we'll send the messages
directly to the peers that have gossipSyncer instances active, so they
can properly be filtered out. For those that don't we'll broadcast
directly, ensuring we skip *all* peers that have an active gossip
syncer.
In this commit, introduce a new struct, the gossipSyncer. The role of
this struct is to encapsulate the state machine required to implement
the new gossip query range feature recently added to the spec. With this
change, each peer that knows of this new feature will have a new
goroutine that will be managed by the gossiper.
Once created and started, the gossipSyncer will start to progress
through each possible state, finally ending at the chansSynced stage. In
this stage, it has synchronized state with the remote peer, and is
simply awaiting any new messages from the gossiper to send directly to
the peer. Each message will only be sent if the remote peer actually has
a set update horizon, and the message isn't before or after that
horizon.
A set of unit tests has been added to ensure that two state machines
properly terminate and synchronize channel state.
In this commit, we update the testUpdateChannelPolicy to exercise the
recent set of changes within the switch. If one applies this test to a
fresh branch (without those new changes) it should fail. This is due to
the fact that before, Bob would attempt to apply the constraints of the
incoming link (which we updated) instead of the outgoing link. With the
recent set of changes, the test now properly passes.
In this commit, we fix an existing deadlock in the
processChanPolicyUpdate method. Before this commit, within
processChanPolicyUpdate, we would directly call updateChannel *within*
the ForEachChannel closure. This would at times result in a deadlock, as
updateChannel will itself attempt to create a write transaction in order
to persist the newly updated channel.
We fix this deadlock by simply performing another loop once we know the
set of channels that we wish to update. This second loop will actually
update the channels on disk.
In this commit, we reduce the amount of unnecessary work that the
gossiper can carry out. When CPU profiling some nodes, I noticed that
we’d spend a lot of time validating the signatures for an announcement,
only to realize that the router already had it.
To remedy this, we’ll use the new methods added to the channel router
in order to avoid unnecessarily validating an announcement that is
actually stale. This should reduce memory usage (since it uses big
int’s under the scenes), and also idle CPU usage.
