Skip to content

Commit

Permalink
Merge bitcoindevkit#1838: Make full-scan/sync flow easier to reason a…
Browse files Browse the repository at this point in the history 
…bout.

800f358 feat!: Improve spk-based syncing flow (志宇)
ee52745 feat(core)!: Make `TxUpdate` non-exhaustive (志宇)

Pull request description:

  ### Description

  bitcoindevkit#1811 introduces the `evicted-at` concept. While adding this to `TxUpdate`, I realized there were some shortcomings in our full-scan & sync flow:

  * Chain sources that use `TxUpdate` to convey updates cannot choose to add transactions without a `seen_at` timestamp as the `TxGraph::apply_update_at` logic adds this timestamp for all unanchored txs if the `seen_at` parameter is specified. Purposefully adding uncanonical txs is useful for wallets that want to know about replaced txs.
  * The `TxGraph::apply_update_at` logic is hard to reason about. `TxUpdate::seen_ats` already has the `seen_at` timestamp per tx, but `apply_update_at()` also takes in a `seen_at` timestamp`.
  * `TxUpdate::seen_ats` currently forces us to only have one `seen-at` per tx as it's a map of `txid -> seen_at`. However, in the future we want to add a `first-seen` timestamp to `TxGraph` which is basically the earliest `seen-at` timestamp introduced so we may want to have multiple `seen_at`s per tx. The other issue is if we merge `TxUpdate`s, we will loose data. I.e. we can only keep the `last_seen` or the `first_seen`.

  These problems were exacerbated when introducing `evicted-at`. In the old design, the chain-source has no concept of sync time (as sync time was introduced after-the-fact when applying the `TxUpdate`). Therefore the introduced `TxUpdate::evicted` field was a `HashSet<Txid>` (no timestamps) and relied on the `TxGraph::apply_upate_at` logic to introduce the `evicted-at` timestamp. All this makes the sync logic harder to understand. What happens if the `seen_at` input is `None`? What happens if updates were applied out of order? What happens when we merge `TxUpdates` before applying?

  The following changes are made in this PR to simplify the sync/full-scan logic to be easier to understand and robust:

  * The `sync_time` is added to the `SyncRequest` and `FullScanRequest`. `sync_time` is mandatory and is added as an input of `builder()`. If the `std` feature is enabled, the `builder_now()` is available which uses the current timestamp. The chain source is now responsible to add this `sync_time` timestamp as `seen_at` for mempool txs. Non-canonical and non-anchored txs can be added without the `seen_at` timestamp.
  * `TxUpdate::seen_ats` is now a `HashSet` of `(Txid, u64)`. This allows multiple `seen_at`s per tx. This is also a much easier to use API as the chain source can just insert into this `HashSet` without checking previous data.
  * `TxGraph::apply_update_at` is removed as we no longer needs to introduce a fallback `seen_at` timestamp after-the-fact. `TxGraph::apply_update` is no longer `std`-only and the logic of applying updates is greatly simplified.

  Additionally, `TxUpdate` is updated to be `#[non_exhaustive]` for backwards-compatibility protection.

  ### Notes to the reviewers

  This is based on bitcoindevkit#1837. Merge that first.

  These are breaking changes to `bdk_core`. It needs to be breaking to properly fix all the issues.

  As mentioned by @notmandatory, `bdk_wallet` changes will be added to a new PR once the new `bdk_wallet` repo is ready. But the PR won't be merged until we're ready for a `bdk_wallet` 2.0.

  ### Changelog notice

  * Change `FullScanRequest::builder` and `SyncRequest::builder` methods to depend on `feature = "std"`. This is because requests now have a `start_time`, instead of specifying a `seen_at` when applying the update.
  * Add `FullScanRequest::builder_at` and `SyncRequest::builder_at` methods which are the non-std version of the `..Request::builder` methods.
  * Change `TxUpdate::seen_ats` field to be a `HashSet` of `(Txid, u64)`. This allows a single update to have multiple `seen_at`s per tx.
  * Change `TxUpdate` to be `non-exhaustive`.
  * Remove `apply_update_at` as we no longer need to apply with a timestamp after-the-fact.

  ### Checklists

  #### All Submissions:

  * [x] I've signed all my commits
  * [x] I followed the [contribution guidelines](https://github.com/bitcoindevkit/bdk/blob/master/CONTRIBUTING.md)
  * [x] I ran `cargo fmt` and `cargo clippy` before committing

  #### New Features:

  ~* [ ] I've added tests for the new feature~ No tests needed as it's more of a refactor.

  * [x] I've added docs for the new feature

ACKs for top commit:
  notmandatory:
    utACK 800f358

Tree-SHA512: 85af8452ac60c4a8087962403fd10c5c67592d3711f7665ae09a2d9c48a868583a41e54f28d639e47bd264ccf95d9254efc8d0d6248c8bcc9343c8290502e061
  • Loading branch information
@evanlinjin
evanlinjin committed Feb 28, 2025
2 parents 209570d + 800f358 commit b26ff89
Show file tree
Hide file tree
Showing 11 changed files with 298 additions and 228 deletions.
18 changes: 8 additions & 10 deletions crates/chain/benches/canonicalization.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -132,11 +132,10 @@ pub fn many_conflicting_unconfirmed(c: &mut Criterion) {
}],
..new_tx(i)
};
let update = TxUpdate {
txs: vec![Arc::new(tx)],
..Default::default()
};
let _ = tx_graph.apply_update_at(update, Some(i as u64));
let mut update = TxUpdate::default();
update.seen_ats = [(tx.compute_txid(), i as u64)].into();
update.txs = vec![Arc::new(tx)];
let _ = tx_graph.apply_update(update);
}
}));
c.bench_function("many_conflicting_unconfirmed::list_canonical_txs", {
Expand Down Expand Up @@ -169,11 +168,10 @@ pub fn many_chained_unconfirmed(c: &mut Criterion) {
..new_tx(i)
};
let txid = tx.compute_txid();
let update = TxUpdate {
txs: vec![Arc::new(tx)],
..Default::default()
};
let _ = tx_graph.apply_update_at(update, Some(i as u64));
let mut update = TxUpdate::default();
update.seen_ats = [(txid, i as u64)].into();
update.txs = vec![Arc::new(tx)];
let _ = tx_graph.apply_update(update);
// Store the next prevout.
previous_output = OutPoint::new(txid, 0);
}
Expand Down
25 changes: 0 additions & 25 deletions crates/chain/src/indexed_tx_graph.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -91,37 +91,12 @@ where
/// Apply an `update` directly.
///
/// `update` is a [`tx_graph::TxUpdate<A>`] and the resultant changes is returned as [`ChangeSet`].
#[cfg(feature = "std")]
#[cfg_attr(docsrs, doc(cfg(feature = "std")))]
pub fn apply_update(&mut self, update: tx_graph::TxUpdate<A>) -> ChangeSet<A, I::ChangeSet> {
let tx_graph = self.graph.apply_update(update);
let indexer = self.index_tx_graph_changeset(&tx_graph);
ChangeSet { tx_graph, indexer }
}

/// Apply the given `update` with an optional `seen_at` timestamp.
///
/// `seen_at` represents when the update is seen (in unix seconds). It is used to determine the
/// `last_seen`s for all transactions in the update which have no corresponding anchor(s). The
/// `last_seen` value is used internally to determine precedence of conflicting unconfirmed
/// transactions (where the transaction with the lower `last_seen` value is omitted from the
/// canonical history).
///
/// Not setting a `seen_at` value means unconfirmed transactions introduced by this update will
/// not be part of the canonical history of transactions.
///
/// Use [`apply_update`](IndexedTxGraph::apply_update) to have the `seen_at` value automatically
/// set to the current time.
pub fn apply_update_at(
&mut self,
update: tx_graph::TxUpdate<A>,
seen_at: Option<u64>,
) -> ChangeSet<A, I::ChangeSet> {
let tx_graph = self.graph.apply_update_at(update, seen_at);
let indexer = self.index_tx_graph_changeset(&tx_graph);
ChangeSet { tx_graph, indexer }
}

/// Insert a floating `txout` of given `outpoint`.
pub fn insert_txout(&mut self, outpoint: OutPoint, txout: TxOut) -> ChangeSet<A, I::ChangeSet> {
let graph = self.graph.insert_txout(outpoint, txout);
Expand Down
62 changes: 15 additions & 47 deletions crates/chain/src/tx_graph.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -110,26 +110,26 @@ use core::{

impl<A: Ord> From<TxGraph<A>> for TxUpdate<A> {
fn from(graph: TxGraph<A>) -> Self {
Self {
txs: graph.full_txs().map(|tx_node| tx_node.tx).collect(),
txouts: graph
.floating_txouts()
.map(|(op, txo)| (op, txo.clone()))
.collect(),
anchors: graph
.anchors
.into_iter()
.flat_map(|(txid, anchors)| anchors.into_iter().map(move |a| (a, txid)))
.collect(),
seen_ats: graph.last_seen.into_iter().collect(),
}
let mut tx_update = TxUpdate::default();
tx_update.txs = graph.full_txs().map(|tx_node| tx_node.tx).collect();
tx_update.txouts = graph
.floating_txouts()
.map(|(op, txo)| (op, txo.clone()))
.collect();
tx_update.anchors = graph
.anchors
.into_iter()
.flat_map(|(txid, anchors)| anchors.into_iter().map(move |a| (a, txid)))
.collect();
tx_update.seen_ats = graph.last_seen.into_iter().collect();
tx_update
}
}

impl<A: Anchor> From<TxUpdate<A>> for TxGraph<A> {
fn from(update: TxUpdate<A>) -> Self {
let mut graph = TxGraph::<A>::default();
let _ = graph.apply_update_at(update, None);
let _ = graph.apply_update(update);
graph
}
}
Expand Down Expand Up @@ -719,52 +719,20 @@ impl<A: Anchor> TxGraph<A> {
///
/// The returned [`ChangeSet`] is the set difference between `update` and `self` (transactions that
/// exist in `update` but not in `self`).
#[cfg(feature = "std")]
#[cfg_attr(docsrs, doc(cfg(feature = "std")))]
pub fn apply_update(&mut self, update: TxUpdate<A>) -> ChangeSet<A> {
use std::time::*;
let now = SystemTime::now()
.duration_since(UNIX_EPOCH)
.expect("current time must be greater than epoch anchor");
self.apply_update_at(update, Some(now.as_secs()))
}

/// Extends this graph with the given `update` alongside an optional `seen_at` timestamp.
///
/// `seen_at` represents when the update is seen (in unix seconds). It is used to determine the
/// `last_seen`s for all transactions in the update which have no corresponding anchor(s). The
/// `last_seen` value is used internally to determine precedence of conflicting unconfirmed
/// transactions (where the transaction with the lower `last_seen` value is omitted from the
/// canonical history).
///
/// Not setting a `seen_at` value means unconfirmed transactions introduced by this update will
/// not be part of the canonical history of transactions.
///
/// Use [`apply_update`](TxGraph::apply_update) to have the `seen_at` value automatically set
/// to the current time.
pub fn apply_update_at(&mut self, update: TxUpdate<A>, seen_at: Option<u64>) -> ChangeSet<A> {
let mut changeset = ChangeSet::<A>::default();
let mut unanchored_txs = HashSet::<Txid>::new();
for tx in update.txs {
if unanchored_txs.insert(tx.compute_txid()) {
changeset.merge(self.insert_tx(tx));
}
changeset.merge(self.insert_tx(tx));
}
for (outpoint, txout) in update.txouts {
changeset.merge(self.insert_txout(outpoint, txout));
}
for (anchor, txid) in update.anchors {
unanchored_txs.remove(&txid);
changeset.merge(self.insert_anchor(txid, anchor));
}
for (txid, seen_at) in update.seen_ats {
changeset.merge(self.insert_seen_at(txid, seen_at));
}
if let Some(seen_at) = seen_at {
for txid in unanchored_txs {
changeset.merge(self.insert_seen_at(txid, seen_at));
}
}
changeset
}

Expand Down
121 changes: 56 additions & 65 deletions crates/chain/tests/test_tx_graph.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -94,7 +94,7 @@ fn insert_txouts() {
// Insert partials transactions.
update.txouts.insert(*outpoint, txout.clone());
// Mark them unconfirmed.
update.seen_ats.insert(outpoint.txid, unconf_seen_at);
update.seen_ats.insert((outpoint.txid, unconf_seen_at));
}

// Insert the full transaction.
Expand Down Expand Up @@ -1231,74 +1231,65 @@ fn tx_graph_update_conversion() {

let test_cases: &[TestCase] = &[
("empty_update", TxUpdate::default()),
(
"single_tx",
TxUpdate {
txs: vec![make_tx(0).into()],
..Default::default()
},
),
(
"two_txs",
TxUpdate {
txs: vec![make_tx(0).into(), make_tx(1).into()],
..Default::default()
},
),
(
"with_floating_txouts",
TxUpdate {
txs: vec![make_tx(0).into(), make_tx(1).into()],
txouts: [
(OutPoint::new(hash!("a"), 0), make_txout(0)),
(OutPoint::new(hash!("a"), 1), make_txout(1)),
(OutPoint::new(hash!("b"), 0), make_txout(2)),
]
.into(),
..Default::default()
},
),
(
"with_anchors",
TxUpdate {
txs: vec![make_tx(0).into(), make_tx(1).into()],
txouts: [
(OutPoint::new(hash!("a"), 0), make_txout(0)),
(OutPoint::new(hash!("a"), 1), make_txout(1)),
(OutPoint::new(hash!("b"), 0), make_txout(2)),
]
.into(),
anchors: [
(ConfirmationBlockTime::default(), hash!("a")),
(ConfirmationBlockTime::default(), hash!("b")),
]
.into(),
..Default::default()
},
),
(
"with_seen_ats",
TxUpdate {
txs: vec![make_tx(0).into(), make_tx(1).into()],
txouts: [
(OutPoint::new(hash!("a"), 0), make_txout(0)),
(OutPoint::new(hash!("a"), 1), make_txout(1)),
(OutPoint::new(hash!("d"), 0), make_txout(2)),
]
.into(),
anchors: [
(ConfirmationBlockTime::default(), hash!("a")),
(ConfirmationBlockTime::default(), hash!("b")),
]
.into(),
seen_ats: [(hash!("c"), 12346)].into_iter().collect(),
},
),
("single_tx", {
let mut tx_update = TxUpdate::default();
tx_update.txs = vec![make_tx(0).into()];
tx_update
}),
("two_txs", {
let mut tx_update = TxUpdate::default();
tx_update.txs = vec![make_tx(0).into(), make_tx(1).into()];
tx_update
}),
("with_floating_txouts", {
let mut tx_update = TxUpdate::default();
tx_update.txs = vec![make_tx(0).into(), make_tx(1).into()];
tx_update.txouts = [
(OutPoint::new(hash!("a"), 0), make_txout(0)),
(OutPoint::new(hash!("a"), 1), make_txout(1)),
(OutPoint::new(hash!("b"), 0), make_txout(2)),
]
.into();
tx_update
}),
("with_anchors", {
let mut tx_update = TxUpdate::default();
tx_update.txs = vec![make_tx(0).into(), make_tx(1).into()];
tx_update.txouts = [
(OutPoint::new(hash!("a"), 0), make_txout(0)),
(OutPoint::new(hash!("a"), 1), make_txout(1)),
(OutPoint::new(hash!("b"), 0), make_txout(2)),
]
.into();
tx_update.anchors = [
(ConfirmationBlockTime::default(), hash!("a")),
(ConfirmationBlockTime::default(), hash!("b")),
]
.into();
tx_update
}),
("with_seen_ats", {
let mut tx_update = TxUpdate::default();
tx_update.txs = vec![make_tx(0).into(), make_tx(1).into()];
tx_update.txouts = [
(OutPoint::new(hash!("a"), 0), make_txout(0)),
(OutPoint::new(hash!("a"), 1), make_txout(1)),
(OutPoint::new(hash!("d"), 0), make_txout(2)),
]
.into();
tx_update.anchors = [
(ConfirmationBlockTime::default(), hash!("a")),
(ConfirmationBlockTime::default(), hash!("b")),
]
.into();
tx_update.seen_ats = [(hash!("c"), 12346)].into_iter().collect();
tx_update
}),
];

for (test_name, update) in test_cases {
let mut tx_graph = TxGraph::<ConfirmationBlockTime>::default();
let _ = tx_graph.apply_update_at(update.clone(), None);
let _ = tx_graph.apply_update(update.clone());
let update_from_tx_graph: TxUpdate<ConfirmationBlockTime> = tx_graph.into();

assert_eq!(
Expand Down
Loading

0 comments on commit b26ff89

Please sign in to comment.