Circuit Breaker

circuitbreaker is to Lightning what firewalls are to the internet.

It allows nodes to protect themselves from being flooded with htlcs. With circuitbreaker a maximum to the number of in-flight htlcs can be set on a per-peer basis. Known and trusted peers for example can be assigned a higher maximum, while a new channel from a previously unseen node may be limited to only a few pending htlcs.

Furthermore it is possible to apply rate limits to the number of forwarded htlcs. This offers protection against DoS/spam attacks that rely on large numbers of fast-resolving htlcs. Rate limiting is implemented with a Token bucket. In configuration the minimum interval between htlcs and a burst size can be specified.

Large numbers of htlcs are also required for probing channel balances. Reducing the information leakage through probing could be another reason to put in place a rate limit for untrusted peers.

Why are limits needed?

In today's Lighting Network payments are routed via a series of hops. Each of those hops will incur a cost for forwarding that payment. While the htlc of an hop is in-flight, the associated amount is locked in the hop's outgoing channel. Those funds cannot be used for another purpose. This can be considered to be an opportunity cost.

Furthermore each channel has a limited number of htlc 'slots'. The current maximum is 483 slots. This means that regardless of channel capacity, there can never be more than 483 htlcs pending. With large channels in particular, it can happen that all slots are occupied while only a fraction of the channel capacity is used. In that case the whole channel is considered to be locked. The duration of the lock can vary from a few seconds to as long as 2 weeks or even more.

When the payment is completed successfully, each hop will collect a routing fee. But depending on the length of the lock and the htlc amounts, this may be far from sufficient to cover the costs.

This is where circuitbreaker comes in. It puts up a defense around that valuable channel liquidity and helps to keep the locked coins at work to maximize routing revenue.

How to use

Requirements

• go 1.13
• lnd version 0.11.0-beta or above.

Configuration

circuitbreaker by default reads its configuration from ~/.circuitbreaker/circuitbreaker.yaml. An example configuration can be found here

Run

• Clone this repository
• cd circuitbreaker/
• go install
• Execute circuitbreaker with the correct command line flags to connect to lnd. See circuitbreaker --help for details.

Operating modes

There are multiple modes in which circuitbreaker can operate. Mode can be configured globally and per peer in the configuration file.

• fail: Fail back htlcs when limits are exceeded. This minimizes the lock-up of liquidity on the incoming side, but does affect your reputation as a routing node.

• queue: Queue htlcs when limits are exceeded. Items are popped from the queue when the number of pending htlcs is below the maximum and the rate limit allows another forward. This mode penalizes upstream nodes for the bad traffic that the deliver by locking up liquidity along the route. This may push upstream nodes to install a firewall too and constrain the mishaving node.

  Even in case of circuit breaker malfunction, queue mode should never cause channel force closes because of lnd's built-in protection that auto-fails htlcs that aren't resolved.

  WARNING 1: Auto-fail is not yet released and scheduled for lnd 0.16. With earlier lnd versions, you risk force-closes!

  WARNING 2: Edge case issues have been reported in queue mode. Usage of this mode isn't recommended until those are resolved.

• queue_peer_initiated: This mode is also queuing htlcs, but only those that come in through channels for which we aren't the channel open initiator. Not being the initiator means that the remote node is carrying the cost of a potential force-closure with stuck htlcs. For channels that we initiated, the safer fail mode is used.

  WARNINGS: See queue mode warnings.

Limitations

• This software is alpha quality. Use at your own risk and be careful in particular on mainnet.
• The interfaces on lnd aren't optimized for this purpose. Therefore the use of a combination of different endpoints is required. This may lead to certain corner cases.