When we force-close a channel, occasionally its due to feerate
disagreements or other non-HTLC-related issues. In those cases,
there's no reason to use a very urgent feerate estimate - we don't
have any timers expiring soon.
Instead, we should give users the information they need to be more
economical on fees in this case, which we do here by splitting
`OnChainSweep` into `UrgentOnChainSweep` and
`NonUrgentOnChainSweep` `ConfirmationTarget`s.
When we broke `ConfirmationTarget` out into task-specific names, we
left `MaxDustHTLCExposure::FeeRateMultiplier` as using the "when we
broadcast feerate" as we were mostly concerned about the dust
thresholds on outbound channels where we pick the fee and drive our
own funds to dust.
In 51bf78d604, that changed to
include transaction fees on both inbound and outbound channels in
our dust exposure amount, but we continued to use
`ConfirmationTarget::OnChainSweep` for the fee estimator threshold.
While the `MaxDustHTLCExposure::FeeRateMultiplier` value is quite
conservative and shouldn't lead to force-closures unless feerate
estimates disagree by something like 500 sat/vB (with only one HTLC
active in a channel), this happened on Aug 22 when feerates spiked
from 4 sat/vB to over 1000 sat/vB in one block.
To avoid simple feerate estimate horizons causing this in the
future, here we add a new
`ConfirmationTarget::MaximumFeeEstimate` which is used for dust
calculations. This allows users to split out the estimates they use
for checking counterparty feerates from the estimates used for
actual broadcasting.
We currently have two structs with identical names in our public
API - `blinded_path::message::ForwardNode` and
`blinded_path::payment::ForwardNode`. This makes the API somewhat
awkward to use - users have to try (and fail) to import
`ForwardNode` twice only to then have to change their imports.
More importantly, however, this makes the API very hard to use in
some bindings languages where rename-imports or module imports
aren't available.
Thus, here, we rename both to give them context.
We never actually build with `#![no_std]` in tests as Rust does
not support it. Thus, many tests have spurious `std` feature gates
when they run just fine without them. Here we remove those gates,
though note that tests that depend on behavior elsewhere in the
codebase which is `std`-gated obviously need to retain their
feature gates.
Not sure why we ever really had this, no one really ever bounds
anything on `std::Error` and its kinda a dead type, so there's no
need for us to `impl` it for our types.
To handle `std` and `no-std` concepts of time in scoring, we'd
originally written a generic `Time` trait which we could use to
fetch the current time, observe real (not wall-clock) elapsed time,
and serialize the time.
Eventually, scoring stopped using this `Time` trait but outbound
payment retry expiry started using it instead to mock time in
tests.
Since scoring no longer uses the full features which required the
`Time` trait, we can substantially simplify by just having the
mocking option.
In 81389dee30 we removed a note about
mixing the `std` and `no-std` feature when de/serializing
`ProbabilisticScorer`s but forgot to note that there was a second
copy of that note in the module documentation.
This removes that note.
The ChannelMonitor::get_claimable_balances method provides a more
straightforward approach to the balance of a channel, which satisfies
most use cases. The computation of AvailableBalances::balance_msat is
complex and originally had a different purpose that is not applicable
anymore. We deprecate AvailableBalances::balance_msat now and will remove
it in a following release.
Co-authored-by: Willem Van Lint <noreply@wvanlint.dev>
In 26c1639ab6 (and later in
50c55dcf32 and
fb693ecbf8) we switched to using
`Default::default()` to initialize `()` for scoring parameters in
tests. One `()`s slipped back in recently, which we replace here.
This test was added some time ago in
0c034e9a82, but never made any sense.
`PeerManager::process_events` will go around its loop as many times
is required to ensure we've always processed all events which were
pending prior to a `process_events` call, so having a test that
checks that we never go around more than twice is obviously broken.
And, indeed, in CI this tests fails with some regularity.
Instead, the test here is changed to ensure that we detectably go
around the loop again at least once.
Fixes#2385
This allow us to forward blinded payments where the blinded path that we are
forwarding within was concatenated to another blinded path that starts at the
next hop.
Also allows constructing blinded paths using this override.
Now that we don't have to have everything in our entire ecosystem
use the same `std`/`no-std` feature combinations we should start by
untangling our own features a bit.
This takes one last step, removing the implications of the `std`
feature from the `lightning` crate.
Now that we don't have to have everything in our entire ecosystem
use the same `std`/`no-std` feature combinations we should start by
untangling our own features a bit.
This takes another step by removing the `no-std` feature entirely
from the `lightning-rapid-gossip-sync` crate and removing all
feature implications on dependencies from the remaining `std`
feature.
Now that we don't have to have everything in our entire ecosystem
use the same `std`/`no-std` feature combinations we should start by
untangling our own features a bit.
This takes another step by removing the `no-std` feature entirely
from the `lightning-invoice` crate and removing all feature
implications on dependencies from the remaining `std` feature.
Now that we don't have to have everything in our entire ecosystem
use the same `std`/`no-std` feature combinations we should start by
untangling our own features a bit.
This takes the first step by removing the `no-std` feature entirely
from the `lightning-background-processor` crate and removing most
feature implications on dependencies from the remaining `std`
feature.
It also addresses a CI oversight where we were not testing
`lightning-background-processor` without the `std` feature in CI at
all.
In order to ensure our crates depend on the workspace copies of
each other in test builds we need to override the crates.io
dependency with a local `path`.
We can do this in one of two ways - either specify the `path` in
the dependency listing in each crate's `Cargo.toml` or use the
workspace `Cargo.toml` to `patch` all dependencies. The first is
tedious while the second lets us have it all in one place. However,
the second option does break `cargo *` in individual crate
directories (forcing the use of `cargo -p crate *` instead) and
makes it rather difficult to depend on local versions of workspace
crates.
Thus, here we drop the `patch.crates-io` from our top-level
`Cargo.toml` entirely.
Still, we do update the `ci/ci-tests.sh` script here to use
`cargo -p crate` instead of switching to each crate's directory as
it allows `cargo` to use a shared `target` and may speed up tests.
`rust-bitcoin` doesn't ever actually *use* its `BufRead`
requirement when deserializing objects, and forcing it is somewhat
inefficient, so we optimize the only (actual) case here by passing
reads straight through to the backing stream.
Since we now have many types in one place, it makes sense to export
them in that place. Further, doing so finally somewhat starts to
reduce our `lightning::ln` module size, which historically is the
dumping ground for everything when most things really should be
top-level modules in `lightning`.
Here we take a step in the right direction by exporting
`lightning_types` as `lightning::types` and encouraging users to
use those paths directly rather than the ones in `lightning::ln`.
During routing, the majority of our time is spent in the scorer.
Given the scorer isn't actually doing all that much computation,
this means we're quite sensitive to memory latency. Thus, the cache
lines our data sits on is incredibly important.
Here, we manually lay out the `ChannelLiquidity` and
`HistoricalLiquidityTracker` structs to ensure that we can do the
non-historical scoring and skip historical scoring for channels
with insufficient data by just looking at the same cache line the
channel's SCID is on.
Sadly, to do the full historical scoring we need to load a second
128-byte cache line pair, but we have some time to get there. We
might consider issuing a preload instruction in the future.
This improves performance a few percent.
When we go to score a channel using the historical liquidity data,
the first thing we do is step through all the valid bucket
combinations, multiply the min and max bucket, and then add them
together to calculate the total number of points tracked. This
isn't a free operation, and for scorers without much data it
represents a large part of the total time spent scoring during
routefinding.
Thus, here we cache this value, updating it every time the buckets
are updated.
Rather than storing the two direction's buckets in
`HistoricalMinMaxBuckets` (renamed
`DirectedHistoricalLiquidityTracker`), we store a single reference
to the `HistoricalLiquidityTracker` as well as the direction bool.
This will allow us in the next commit to reference fields in the
`HistoricalLiquidityTracker` aside from the two directions.
In a comming commit we'll cache some additional data in the
historical bucket tracker. In order to do so, here we isolate the
buckets themselves into the `bucketed_history` module, reducing
the possibility of accidentally updating them directly without
updating caches.
In the coming commits we'll isolate historical bucket logic slightly
further, allowing us to cache some state. This is the first step
towards that, storing the historical liquidity information in a new
`HistoricalLiquidityTracker` rather than in the general
`ChannelLiquidity`.
