::::::::: ::: ::::::::::: ::::::::::: ::: :::
:+: :+: :+: :+: :+: :+: :+: :+:
+:+ +:+ +:+ +:+ +:+ +:+ +:+ +:+
+#++:++#: +#++:++#++: +#+ +#+ +#+ +:+
+#+ +#+ +#+ +#+ +#+ +#+ +#+ +#+
#+# #+# #+# #+# #+# #+# #+# #+# #+#
### ### ### ### ########### ##### ########
Your friendly bitcoin lightning network helper.
raiju
is a CLI app which sits on top of a lightning node and brings some smarts (perhaps that is debateable) to the channel life-cycle: open, manage, and close. raiju
only supports the lnd node implementation at the moment.
dashboard
raiju
's root command, raiju
, fires up an interactive TUI dashboard which uses the subcommands under the hood.
subcommands
All of raiju
's subcommands can be listed with the global help flag, raiju -h
, and each command has its own help (e.g. raiju candidates -h
).
candidates
Open the most efficient channels
List the best nodes to open a channel to from the current node. candidates
does not automatically open any channels, that needs to be done out-of-band with a different tool such as lncli
. candidates
just lists suggestions and is not intended to be automated (for now...).
The current node has distance 0
to itself and distance 1
to the nodes it has channels with. A node with distance 2
has a channel with a node the current node is connected too, but no channel with the current node, and so on. "Distant Neighbors" are distant (greater than 2
) from the current node, but have a channel with the candidate. By default, only nodes with clearnet addresses are listed. TOR-only nodes tend to be unreliable due to the nature of TOR.
$ raiju candidates
Pubkey Alias Distance Distant Neighbors Capacity Channels Updated
029ef8a775117ba63662a1d1d92b8a184bb1758ed1e12b0cdbb5e92672ef695b73 Carnivore 4 8 14932925 8 2021-04-21 23:17:36 -0700 PDT
0390b5d4492dc2f5318e5233ab2cebf6d48914881a33ef6a9c6bcdbb433ad986d0 LNBIG.com [lnd-01] 3 547 2568240344 547 2021-04-22 12:20:14 -0700 PDT
02c91d6aa51aa940608b497b6beebcb1aec05be3c47704b682b3889424679ca490 LNBIG.com [lnd-21] 3 372 2134427027 372 2021-04-22 08:32:48 -0700 PDT
02a04446caa81636d60d63b066f2814cbd3a6b5c258e3172cbdded7a16e2cfff4c ln.bitstamp.net [Bitstamp] 3 366 1621569578 366 2021-04-22 12:43:28 -0700 PDT
...
The assume
flag allows you to see the remaining candidates and updated stats assuming channels were opened to the given nodes. This can be used to find a set of nodes to open channels too in single batch transaction in order to minimize on onchain fees.
From a "make money routing" perspective, theoretically, these most distant nodes with the most distant neighbor connections are good to open a channel to for some off the beaten path efficient routing vs. just connecting to the biggest node in the network. Your node could offer cheaper, better routing between two "clusters" of nodes than the biggest nodes. From a "make the network stronger in general" perspective, the hope is that this strategy creates a more decentralized network vs. everything being dependent on a handful of large hub nodes.
fees
Passively manage channel liquidity
Set channel fees based on the channel's current liquidity. The idea here is to encourage passive channel re-balancing through fees. If a channel has a too much local liquidity (high), fees are lowered in order to encourage relatively more outbound transactions. Visa versa for a channel with too little local liquidity (low).
The global -liquidity-thresholds
flag determines how channels are grouped into liquidity buckets, while the -liquidity-fees
flag determines the fee settings applied to those groups. For example, if thresholds are set to 80,20
and fees set to 5,50,500
, then channels with over 80% local liquidity will have a 5 PPM fee, channels between 80% and 20% local liquidity will have a 50 PPM fee, and channels with less than 20% liquidity will have a 500 PPM fee.
The -liquidity-stickiness
attempts to avoid extra gossip by waiting for channels to return to a healthier liquidity state before changing fees. If using the same settings as before, plus a stickiness setting of 5%, if a channel moves from 19% liquidity to 23% liquidity it will still have a 500 PPM fee. It needs to move to something better than 25% (20% + 5%) before the fee will change. The stickiness setting only applies to liquidity moving in a healthy (towards center) direction. If you are drastically changing your fee settings, you probably want to set stickiness to 0 temporarily to ensure fees are updated.
The -liquidity-thresholds
, -liquidity-fees
, and -liquidity-stickiness
are global (not fees
specific) because they are also used in the rebalance
command to help coordinate the right amount of fees to pay in active rebalancing.
The -daemon
flag keeps keeps the process alive listening for channel updates that trigger fee updates (e.g. a channel's liquidity sinks below the low level and needs its fees updated). This is helpful when used with the rebalance
command which actively balances channel liquidity. Without the daemon, there is a worst case scenario of: 1. pay a lot of fees to actively rebalance
channel's liquidity from low to standard 2. update the channel's fees to standard 3. have a large payment immediately cancel out the rebalance and it only pays standard fees (instead of higher ones which would have canceled out the cost of the rebalance).
fees
follows the zero-base-fee movement. I am honestly not sure if this is financially sound, but I appreciate the simpler mental model of only thinking in ppm.
systemd automation
Example fees.service
:
[Unit]
Description=Monitor channel fees of LND node
Wants=lnd.service
After=lnd.service
[Service]
User=lightning
Group=lightning
Restart=always
Environment=RAIJU_HOST=localhost:10009
Environment=RAIJU_MAC_PATH=/home/lightning/.lnd/data/chain/bitcoin/mainnet/admin.macaroon
Environment=RAIJU_TLS_PATH=/home/lightning/.lnd/tls.cert
Environment=RAIJU_LIQUIDITY_FEES=5,50,500
Environment=RAIJU_LIQUIDITY_STICKINESS=5
ExecStart=/usr/local/bin/raiju fees -daemon
[Install]
WantedBy=multi-user.target
rebalance
Actively manage channel liquidity
Circular rebalance channels that aren't doing so hot liquidity-wise or force rebalance a single channel.
Where the fees
command attempts to balance channels passively, this is an active approach where liquidity is manually pushed. The cost of active rebalancing are the lightning payment fees. While this command could be used to push large amounts of liquidity, the default settings are intended to just prod things in the right direction.
The maximum fee ppm setting defaults to the low liquidity fee setting used by the fees
command. Theoretically, this means that even if a rebalance is instantly canceled out by a large payment at least fees are re-coup'd.
The command takes two arguments:
- A percentage of the channel capacity to attempt to rebalance per circular payment (the "step").
- The maximum percentage of the channel capacity to attempt to rebalance.
A smaller step percentage will increase the likely hood of a successful payment, but might also increase fees a bit if the payment collects a lot of base_fees
on its route.
If no out channel and last hop pubkey are given, the command will roll through channels with high liquidity and attempt to push it through channels of low liquidity. High and low are defined by the defined by the global -liquidity-thresholds
flag. For example, if liquidity thresholds is set to 80,20
, channels with local liquidity over 80% are considered "high" and channels with local liquidity under 20% are considered "low".
$ raiju rebalance 1 1
If output channel and last hop node flags are specified, than just those channels will be rebalanced. The following example is pushing 1% of the channel 754031881261074944
's capacity to the channel with the 03963169ddfcc5cc6afaff7764fa20dc2e21e9ed8ef0ff0ccd18137d62ae2e01f4
node. A max fee of 2000
ppm will be paid. This is kind of a "force" version of the command since not as many safe guards are in place to ensure fees are re-coup'd.
$ raiju rebalance -last-hop-pubkey 03963169ddfcc5cc6afaff7764fa20dc2e21e9ed8ef0ff0ccd18137d62ae2e01f4 -out-channel-id 754031881261074944 1 1
Why is the out channel a channel ID while the last hop (a.k.a. in channel) a pubkey? This is due to the lightning Network's protocol allowing for non-strict forwarding. There might be some ways to specify an in channel, but I haven't put too much thought into it yet.
systemd automation
Example rebalance.service
:
[Unit]
Description=Rebalance channels of LND node
[Service]
User=lightning
Group=lightning
Environment=RAIJU_HOST=localhost:10009
Environment=RAIJU_MAC_PATH=/home/lightning/.lnd/data/chain/bitcoin/mainnet/admin.macaroon
Environment=RAIJU_TLS_PATH=/home/lightning/.lnd/tls.cert
Environment=RAIJU_LIQUIDITY_FEES=5,50,500
Environment=RAIJU_LIQUIDITY_STICKINESS=5
ExecStart=/usr/local/bin/raiju rebalance 1 5
Example rebalance.timer
:
[Unit]
Description=Rebalance channels daily with a wiggle so not run at the same time every day
[Timer]
OnCalendar=*-*-* 00:00:00
RandomizedDelaySec=1h
[Install]
WantedBy=timers.target
reaper
Close the least efficient channels
Lists channels which should be closed and re-allocated. Similar to the candidates
command, these are just suggestions and no channels are automatically closed. They must be closed out-of-band with another tool like lncli
. This might be made automated in the future.
$ raiju reaper
Channel ID Pubkey Capacity (BTC)
859008852420919297 fewsats 0.1
826864630068281345 Pinky 0.02
installation
To install from source, raiju
requires go
on the system. go install
creates a raiju
executable.
$ go install github.com/nyonson/raiju/cmd/raiju@latest
If a container is preferred, raiju
images are published at ghcr.io/nyonson/raiju
.
docker pull ghcr.io/nyonson/raiju:v0.7.1
A little more configuration is required to pass along settings to the container.
docker run -it \
-v /admin.macaroon:/admin.macaroon:ro -v /tls.cert:/tls.cert:ro \
ghcr.io/nyonson/raiju:v0.7.1 \
-host 192.168.1.187:10009 -mac-path admin.macaroon -tls-path tls.cert
candidates
- Ensure the tls certificate and macaroon are mounted in the container, in the above example they are both mounted to the root of the container's filesystem and then their paths are passed in as cli flags.
- The container may need to be attached to a network depending on your network.
configuration
Global flags (can be found with raiju -h
) can be set through environment variables or with a configuration file. CLI flags overwrite environment variables which overwrite the configuration file values.
Environment variables have a RAIJU_
prefix on the flag name. For example, the global flag host
can be set with the RAIJU_HOST
environment variable.
A configuration file can be provided with the -config
flag or the default location (for Linux it's ~/.config/raiju/config
) can be used. The configuration file format is a flag per line, whitespace delimited.
host localhost:10009
node
Are you here looking for a node to open a channel too? Well, may I offer raiju
's node! Could always use the inbound: 02b6867b56ca1b6a4548b97b009152683fa366bfa1b14119c8f9992e1acacbe1c8