This change modifies create_trampoline_onion to only include as many
available r_tags as there is space left in the trampoline onion payload.
Previously we tried to include all passed invoice r_tags of legacy
trampoline payments into the payload which caused an user facing
exception and payment failure as the onion can only store a max of 400
bytes.
A single, single hop r_tag is around 52 bytes and the payload
without r_tags is already at ~280 bytes. So usually there is enough
space for 2 r_tags.
The implementation shuffles the r_tags on each call
so the payment will try different route hints on the attempts (fee level
increase or user retry).
I have logged the following byte sizes of the trampoline onion with a 2
trampoline onion hop and changing amounts of r_tags:
3 rtags:
payload size [0]: 113 (hop size: 81)
payload size [1]: 440 (hop size: 295) ( 52 bytes/rtag )
payload size [2]: 550 (hop size: 78)
2 rtags:
payload size [0]: 113 (hop size: 81)
payload size [1]: 386 (hop size: 241) ( 52 bytes/rtag )
payload size [2]: 496 (hop size: 78)
1 rtag:
payload size [0]: 113 (hop size: 81)
payload size [1]: 334 (hop size: 189) ( 52 bytes/rtag )
payload size [2]: 444 (hop size: 78)
0 rtags:
payload size [0]: 113 (hop size: 81)
payload size [1]: 282 (hop size: 137)
payload size [2]: 392 (hop size: 78)
As can be seen in the data, using 2 trampoline hops there is not enough
space for even a single r_tag which is why this option is being removed
too.
Make path calculation check if channel is not in our sending channels but still uses our nodeID as starting node of the path.
I noticed an assertion error when trying to pay an invoice from a seed i have opened channels with in different wallet instances (same seed, different wallet).
Because the channel seemed suitable for sending the payment path finding included the channel for sending in the first position of the route but then
in pay_to_route the channel for route[0] could not be found as it is not included in our channel list, causing the assert and payment to fail.
- introduce PaymentFeeBudget, which contains limits for fee budget and cltv budget
- when splitting a payment,
- the fee budget is linearly distributed between the parts
- this resolves a FIXME in lnrouter ("FIXME in case of MPP")
- the cltv budget is simply copied
- we could also add other kinds of budgets later, e.g. for the num in-flight htlcs
- resolves TODO in lnworker ("todo: compare to the fee of the actual route we found")
The convention is that edges (start_node -> edge_node) store
the policy/fees for the *start_node*.
This is what the non-trampoline edges were already using (for a long time),
but the trampoline ones were off-by-one (policy was for end_node),
which was then worked around in multiple places, to correct for...
i.e. I think because of all the workarounds, there was no actual bug,
but it was just very confusing.
Also note that the prior usage of trampoline edges would not work if
we (sender) were not directly connected to a TF (trampoline-forwarder)
but had extra edges in the route to even get to the first TF.
Having the policy corresponding to the start_node of the edge would work
even in that case.
eclair sends CHANNEL_DISABLED if its peer is offline. E.g. we might be
trying to pay a mobile phone with the app closed. In that case we
should not cache the CHANNEL_DISABLED for too long.
LnFeatures.supports() is became part of the hot path of
LNPathFinder.find_path_for_payment()
in 6b43eac6fd,
which made find_path_for_payment considerably slower than before.
It used to take around 0.8 sec originally, then after linked commit went to ~9 sec,
and now this takes it down to ~1.1 sec (on my laptop).
If the short_channel_id of a channel update received
with update_fail_htlc does not match the channel in
our route, blacklist the channel in our route.
Adds liquidity hints for the sending capabilities of routing channels in the
graph. The channel blacklist is incorporated into liquidity hints.
Liquidity hints are updated when a payment fails with a temporary
channel failure or when it succeeds. Liquidity hints are used to give a
penalty in the _edge_cost heuristics used by the pathfinding algorithm.
The base penalty in (_edge_cost) is removed because it is now part of the
liquidity penalty. We don't return early from get_distances, as we want
to explore all channels.
We pass the private edges to lnrouter, and let it find routes end-to-end.
Previously the edge_cost heuristics didn't apply to the private edges
and we were just randomly picking one of the route hints and use that.
So e.g. cheaper private edges were not preferred, but they are now.
PathEdge now stores both start_node and end_node; not just end_node.
Note that for a required feature, BOLT-09 allows setting either:
- only the REQ bit
- both the REQ bit and the OPT bit
Hence, when checking if a feature is supported by e.g. an invoice, both
bits should be checked.
Note that in lnpeer.py, in self.features specifically, REQ implies OPT,
as it is set by ln_compare_features.
- trampoline is enabled by default in config, to prevent download of `gossip_db`.
(if disabled, `gossip_db` will be downloaded, regardless of the existence of channels)
- if trampoline is enabled:
- the wallet can only open channels with trampoline nodes
- already-existing channels with non-trampoline nodes are frozen for sending.
- there are two types of trampoline payments: legacy and end-to-end (e2e).
- we decide to perform legacy or e2e based on the invoice:
- we use trampoline_routing_opt in features to detect Eclair and Phoenix invoices
- we use trampoline_routing_hints to detect Electrum invoices
- when trying a legacy payment, we add a second trampoline to the path to preserve privacy.
(we fall back to a single trampoline if the payment fails for all trampolines)
- the trampoline list is hardcoded, it will remain so until `trampoline_routing_opt` feature flag is in INIT.
- there are currently only two nodes in the hardcoded list, it would be nice to have more.
- similar to Phoenix, we find the fee/cltv by trial-and-error.
- if there is a second trampoline in the path, we use the same fee for both.
- the final spec should add fee info in error messages, so we will be able to fine-tune fees
Old estimate was heavily biased towards simply minimising CLTV sum.
(fees had a too low weight; typically they were ~noise)
Now also take payment_amount into account.
Previously we would put fake chan announcement and fake outgoing chan upd
for own channels into db (to make path finding work). See Peer.add_own_channel().
Now, instead of above, we pass a "my_channels" param to the relevant ChannelDB methods.
