babe: allow skipping over empty epochs (#11727)

* babe: allow skipping epochs in pallet

* babe: detect and skip epochs on client

* babe: cleaner epoch util functions

* babe: add test for runtime handling of skipped epochs

* babe: simpler implementation of client handling of skipped epochs

* babe: test client-side handling of skipped epochs

* babe: add comments on client-side skipped epochs

* babe: remove emptyline

* babe: make it resilient to forks

* babe: typo

* babe: overflow-safe math

* babe: add test for skipping epochs across different forks

* Fix tests

* FMT

Co-authored-by: Bastian Köcher <info@kchr.de>
This commit is contained in:
André Silva
2022-12-24 00:03:31 +00:00
committed by GitHub
parent e053d34cc5
commit 017cf70378
5 changed files with 386 additions and 23 deletions
+46 -3
View File
@@ -543,7 +543,7 @@ fn aux_storage_cleanup<C: HeaderMetadata<Block> + HeaderBackend<Block>, Block: B
let stale_forks = match client.expand_forks(&notification.stale_heads) {
Ok(stale_forks) => stale_forks,
Err((stale_forks, e)) => {
warn!(target: "babe", "{:?}", e,);
warn!(target: LOG_TARGET, "{:?}", e);
stale_forks
},
};
@@ -1511,11 +1511,12 @@ where
if let Some(next_epoch_descriptor) = next_epoch_digest {
old_epoch_changes = Some((*epoch_changes).clone());
let viable_epoch = epoch_changes
let mut viable_epoch = epoch_changes
.viable_epoch(&epoch_descriptor, |slot| Epoch::genesis(&self.config, slot))
.ok_or_else(|| {
ConsensusError::ClientImport(Error::<Block>::FetchEpoch(parent_hash).into())
})?;
})?
.into_cloned();
let epoch_config = next_config_digest
.map(Into::into)
@@ -1528,6 +1529,48 @@ where
log::Level::Info
};
if viable_epoch.as_ref().end_slot() <= slot {
// some epochs must have been skipped as our current slot
// fits outside the current epoch. we will figure out
// which epoch it belongs to and we will re-use the same
// data for that epoch
let mut epoch_data = viable_epoch.as_mut();
let skipped_epochs =
*slot.saturating_sub(epoch_data.start_slot) / epoch_data.duration;
// NOTE: notice that we are only updating a local copy of the `Epoch`, this
// makes it so that when we insert the next epoch into `EpochChanges` below
// (after incrementing it), it will use the correct epoch index and start slot.
// we do not update the original epoch that will be re-used because there might
// be other forks (that we haven't imported) where the epoch isn't skipped, and
// to import those forks we want to keep the original epoch data. not updating
// the original epoch works because when we search the tree for which epoch to
// use for a given slot, we will search in-depth with the predicate
// `epoch.start_slot <= slot` which will still match correctly without updating
// `start_slot` to the correct value as below.
let epoch_index = epoch_data.epoch_index.checked_add(skipped_epochs).expect(
"epoch number is u64; it should be strictly smaller than number of slots; \
slots relate in some way to wall clock time; \
if u64 is not enough we should crash for safety; qed.",
);
let start_slot = skipped_epochs
.checked_mul(epoch_data.duration)
.and_then(|skipped_slots| epoch_data.start_slot.checked_add(skipped_slots))
.expect(
"slot number is u64; it should relate in some way to wall clock time; \
if u64 is not enough we should crash for safety; qed.",
);
warn!(
target: LOG_TARGET,
"👶 Epoch(s) skipped: from {} to {}", epoch_data.epoch_index, epoch_index,
);
epoch_data.epoch_index = epoch_index;
epoch_data.start_slot = Slot::from(start_slot);
}
log!(
target: LOG_TARGET,
log_level,
@@ -28,6 +28,7 @@ use rand_chacha::{
use sc_block_builder::{BlockBuilder, BlockBuilderProvider};
use sc_client_api::{backend::TransactionFor, BlockchainEvents, Finalizer};
use sc_consensus::{BoxBlockImport, BoxJustificationImport};
use sc_consensus_epochs::{EpochIdentifier, EpochIdentifierPosition};
use sc_consensus_slots::BackoffAuthoringOnFinalizedHeadLagging;
use sc_network_test::{Block as TestBlock, *};
use sp_application_crypto::key_types::BABE;
@@ -1109,3 +1110,263 @@ async fn obsolete_blocks_aux_data_cleanup() {
// Present C4, C5
assert!(aux_data_check(&fork3_hashes, true));
}
#[tokio::test]
async fn allows_skipping_epochs() {
let mut net = BabeTestNet::new(1);
let peer = net.peer(0);
let data = peer.data.as_ref().expect("babe link set up during initialization");
let client = peer.client().as_client();
let mut block_import = data.block_import.lock().take().expect("import set up during init");
let mut proposer_factory = DummyFactory {
client: client.clone(),
config: data.link.config.clone(),
epoch_changes: data.link.epoch_changes.clone(),
mutator: Arc::new(|_, _| ()),
};
let epoch_changes = data.link.epoch_changes.clone();
let epoch_length = data.link.config.epoch_length;
// we create all of the blocks in epoch 0 as well as a block in epoch 1
let blocks = propose_and_import_blocks(
&client,
&mut proposer_factory,
&mut block_import,
client.chain_info().genesis_hash,
epoch_length as usize + 1,
)
.await;
// the first block in epoch 0 (#1) announces both epoch 0 and 1 (this is a
// special genesis epoch)
let epoch0 = epoch_changes
.shared_data()
.epoch(&EpochIdentifier {
position: EpochIdentifierPosition::Genesis0,
hash: blocks[0],
number: 1,
})
.unwrap()
.clone();
assert_eq!(epoch0.epoch_index, 0);
assert_eq!(epoch0.start_slot, Slot::from(1));
let epoch1 = epoch_changes
.shared_data()
.epoch(&EpochIdentifier {
position: EpochIdentifierPosition::Genesis1,
hash: blocks[0],
number: 1,
})
.unwrap()
.clone();
assert_eq!(epoch1.epoch_index, 1);
assert_eq!(epoch1.start_slot, Slot::from(epoch_length + 1));
// the first block in epoch 1 (#7) announces epoch 2. we will be skipping
// this epoch and therefore re-using its data for epoch 3
let epoch2 = epoch_changes
.shared_data()
.epoch(&EpochIdentifier {
position: EpochIdentifierPosition::Regular,
hash: blocks[epoch_length as usize],
number: epoch_length + 1,
})
.unwrap()
.clone();
assert_eq!(epoch2.epoch_index, 2);
assert_eq!(epoch2.start_slot, Slot::from(epoch_length * 2 + 1));
// we now author a block that belongs to epoch 3, thereby skipping epoch 2
let last_block = client.expect_header(*blocks.last().unwrap()).unwrap();
let block = propose_and_import_block(
&last_block,
Some((epoch_length * 3 + 1).into()),
&mut proposer_factory,
&mut block_import,
)
.await;
// and the first block in epoch 3 (#8) announces epoch 4
let epoch4 = epoch_changes
.shared_data()
.epoch(&EpochIdentifier {
position: EpochIdentifierPosition::Regular,
hash: block,
number: epoch_length + 2,
})
.unwrap()
.clone();
assert_eq!(epoch4.epoch_index, 4);
assert_eq!(epoch4.start_slot, Slot::from(epoch_length * 4 + 1));
// if we try to get the epoch data for a slot in epoch 3
let epoch3 = epoch_changes
.shared_data()
.epoch_data_for_child_of(
descendent_query(&*client),
&block,
epoch_length + 2,
(epoch_length * 3 + 2).into(),
|slot| Epoch::genesis(&data.link.config, slot),
)
.unwrap()
.unwrap();
// we get back the data for epoch 2
assert_eq!(epoch3, epoch2);
// but if we try to get the epoch data for a slot in epoch 4
let epoch4_ = epoch_changes
.shared_data()
.epoch_data_for_child_of(
descendent_query(&*client),
&block,
epoch_length + 2,
(epoch_length * 4 + 1).into(),
|slot| Epoch::genesis(&data.link.config, slot),
)
.unwrap()
.unwrap();
// we get epoch 4 as expected
assert_eq!(epoch4, epoch4_);
}
#[tokio::test]
async fn allows_skipping_epochs_on_some_forks() {
let mut net = BabeTestNet::new(1);
let peer = net.peer(0);
let data = peer.data.as_ref().expect("babe link set up during initialization");
let client = peer.client().as_client();
let mut block_import = data.block_import.lock().take().expect("import set up during init");
let mut proposer_factory = DummyFactory {
client: client.clone(),
config: data.link.config.clone(),
epoch_changes: data.link.epoch_changes.clone(),
mutator: Arc::new(|_, _| ()),
};
let epoch_changes = data.link.epoch_changes.clone();
let epoch_length = data.link.config.epoch_length;
// we create all of the blocks in epoch 0 as well as two blocks in epoch 1
let blocks = propose_and_import_blocks(
&client,
&mut proposer_factory,
&mut block_import,
client.chain_info().genesis_hash,
epoch_length as usize + 1,
)
.await;
// we now author a block that belongs to epoch 2, built on top of the last
// authored block in epoch 1.
let last_block = client.expect_header(*blocks.last().unwrap()).unwrap();
let epoch2_block = propose_and_import_block(
&last_block,
Some((epoch_length * 2 + 1).into()),
&mut proposer_factory,
&mut block_import,
)
.await;
// if we try to get the epoch data for a slot in epoch 2, we get the data that
// was previously announced when epoch 1 started
let epoch2 = epoch_changes
.shared_data()
.epoch_data_for_child_of(
descendent_query(&*client),
&epoch2_block,
epoch_length + 2,
(epoch_length * 2 + 2).into(),
|slot| Epoch::genesis(&data.link.config, slot),
)
.unwrap()
.unwrap();
// we now author a block that belongs to epoch 3, built on top of the last
// authored block in epoch 1. authoring this block means we're skipping epoch 2
// entirely on this fork
let epoch3_block = propose_and_import_block(
&last_block,
Some((epoch_length * 3 + 1).into()),
&mut proposer_factory,
&mut block_import,
)
.await;
// if we try to get the epoch data for a slot in epoch 3
let epoch3_ = epoch_changes
.shared_data()
.epoch_data_for_child_of(
descendent_query(&*client),
&epoch3_block,
epoch_length + 2,
(epoch_length * 3 + 2).into(),
|slot| Epoch::genesis(&data.link.config, slot),
)
.unwrap()
.unwrap();
// we get back the data for epoch 2
assert_eq!(epoch3_, epoch2);
// if we try to get the epoch data for a slot in epoch 4 in the fork
// where we skipped epoch 2, we should get the epoch data for epoch 4
// that was announced at the beginning of epoch 3
let epoch_data = epoch_changes
.shared_data()
.epoch_data_for_child_of(
descendent_query(&*client),
&epoch3_block,
epoch_length + 2,
(epoch_length * 4 + 1).into(),
|slot| Epoch::genesis(&data.link.config, slot),
)
.unwrap()
.unwrap();
assert!(epoch_data != epoch3_);
// if we try to get the epoch data for a slot in epoch 4 in the fork
// where we didn't skip epoch 2, we should get back the data for epoch 3,
// that was announced when epoch 2 started in that fork
let epoch_data = epoch_changes
.shared_data()
.epoch_data_for_child_of(
descendent_query(&*client),
&epoch2_block,
epoch_length + 2,
(epoch_length * 4 + 1).into(),
|slot| Epoch::genesis(&data.link.config, slot),
)
.unwrap()
.unwrap();
assert!(epoch_data != epoch3_);
let epoch3 = epoch_changes
.shared_data()
.epoch(&EpochIdentifier {
position: EpochIdentifierPosition::Regular,
hash: epoch2_block,
number: epoch_length + 2,
})
.unwrap()
.clone();
assert_eq!(epoch_data, epoch3);
}
+29 -20
View File
@@ -588,10 +588,19 @@ impl<T: Config> Pallet<T> {
return
}
// Update epoch index
let epoch_index = EpochIndex::<T>::get()
.checked_add(1)
.expect("epoch indices will never reach 2^64 before the death of the universe; qed");
// Update epoch index.
//
// NOTE: we figure out the epoch index from the slot, which may not
// necessarily be contiguous if the chain was offline for more than
// `T::EpochDuration` slots. When skipping from epoch N to e.g. N+4, we
// will be using the randomness and authorities for that epoch that had
// been previously announced for epoch N+1, and the randomness collected
// during the current epoch (N) will be used for epoch N+5.
let epoch_index = sp_consensus_babe::epoch_index(
CurrentSlot::<T>::get(),
GenesisSlot::<T>::get(),
T::EpochDuration::get(),
);
EpochIndex::<T>::put(epoch_index);
Authorities::<T>::put(authorities);
@@ -638,11 +647,16 @@ impl<T: Config> Pallet<T> {
}
}
/// Finds the start slot of the current epoch. only guaranteed to
/// give correct results after `initialize` of the first block
/// in the chain (as its result is based off of `GenesisSlot`).
/// Finds the start slot of the current epoch.
///
/// Only guaranteed to give correct results after `initialize` of the first
/// block in the chain (as its result is based off of `GenesisSlot`).
pub fn current_epoch_start() -> Slot {
Self::epoch_start(EpochIndex::<T>::get())
sp_consensus_babe::epoch_start_slot(
EpochIndex::<T>::get(),
GenesisSlot::<T>::get(),
T::EpochDuration::get(),
)
}
/// Produces information about the current epoch.
@@ -666,9 +680,15 @@ impl<T: Config> Pallet<T> {
if u64 is not enough we should crash for safety; qed.",
);
let start_slot = sp_consensus_babe::epoch_start_slot(
next_epoch_index,
GenesisSlot::<T>::get(),
T::EpochDuration::get(),
);
Epoch {
epoch_index: next_epoch_index,
start_slot: Self::epoch_start(next_epoch_index),
start_slot,
duration: T::EpochDuration::get(),
authorities: NextAuthorities::<T>::get().to_vec(),
randomness: NextRandomness::<T>::get(),
@@ -680,17 +700,6 @@ impl<T: Config> Pallet<T> {
}
}
fn epoch_start(epoch_index: u64) -> Slot {
// (epoch_index * epoch_duration) + genesis_slot
const PROOF: &str = "slot number is u64; it should relate in some way to wall clock time; \
if u64 is not enough we should crash for safety; qed.";
let epoch_start = epoch_index.checked_mul(T::EpochDuration::get()).expect(PROOF);
epoch_start.checked_add(*GenesisSlot::<T>::get()).expect(PROOF).into()
}
fn deposit_consensus<U: Encode>(new: U) {
let log = DigestItem::Consensus(BABE_ENGINE_ID, new.encode());
<frame_system::Pallet<T>>::deposit_log(log)
+31
View File
@@ -948,3 +948,34 @@ fn generate_equivocation_report_blob() {
println!("equivocation_proof.encode(): {:?}", equivocation_proof.encode());
});
}
#[test]
fn skipping_over_epochs_works() {
let mut ext = new_test_ext(3);
ext.execute_with(|| {
let epoch_duration: u64 = <Test as Config>::EpochDuration::get();
// this sets the genesis slot to 100;
let genesis_slot = 100;
go_to_block(1, genesis_slot);
// we will author all blocks from epoch #0 and arrive at a point where
// we are in epoch #1. we should already have the randomness ready that
// will be used in epoch #2
progress_to_block(epoch_duration + 1);
assert_eq!(EpochIndex::<Test>::get(), 1);
// genesis randomness is an array of zeros
let randomness_for_epoch_2 = NextRandomness::<Test>::get();
assert!(randomness_for_epoch_2 != [0; 32]);
// we will now create a block for a slot that is part of epoch #4.
// we should appropriately increment the epoch index as well as re-use
// the randomness from epoch #2 on epoch #4
go_to_block(System::block_number() + 1, genesis_slot + epoch_duration * 4);
assert_eq!(EpochIndex::<Test>::get(), 4);
assert_eq!(Randomness::<Test>::get(), randomness_for_epoch_2);
});
}
@@ -360,6 +360,25 @@ pub struct Epoch {
pub config: BabeEpochConfiguration,
}
/// Returns the epoch index the given slot belongs to.
pub fn epoch_index(slot: Slot, genesis_slot: Slot, epoch_duration: u64) -> u64 {
*slot.saturating_sub(genesis_slot) / epoch_duration
}
/// Returns the first slot at the given epoch index.
pub fn epoch_start_slot(epoch_index: u64, genesis_slot: Slot, epoch_duration: u64) -> Slot {
// (epoch_index * epoch_duration) + genesis_slot
const PROOF: &str = "slot number is u64; it should relate in some way to wall clock time; \
if u64 is not enough we should crash for safety; qed.";
epoch_index
.checked_mul(epoch_duration)
.and_then(|slot| slot.checked_add(*genesis_slot))
.expect(PROOF)
.into()
}
sp_api::decl_runtime_apis! {
/// API necessary for block authorship with BABE.
#[api_version(2)]