2020-05-06 11:42:56 +00:00
|
|
|
#![cfg(not(debug_assertions))]
|
|
|
|
|
2023-05-15 02:10:41 +00:00
|
|
|
use beacon_chain::attestation_verification::{
|
|
|
|
batch_verify_aggregated_attestations, batch_verify_unaggregated_attestations, Error,
|
|
|
|
};
|
|
|
|
use beacon_chain::test_utils::{MakeAttestationOptions, HARNESS_GENESIS_TIME};
|
2020-05-06 11:42:56 +00:00
|
|
|
use beacon_chain::{
|
|
|
|
attestation_verification::Error as AttnError,
|
2021-08-30 06:41:31 +00:00
|
|
|
test_utils::{
|
|
|
|
test_spec, AttestationStrategy, BeaconChainHarness, BlockStrategy, EphemeralHarnessType,
|
|
|
|
},
|
2023-08-21 05:02:32 +00:00
|
|
|
BeaconChain, BeaconChainError, BeaconChainTypes, ChainConfig, WhenSlotSkipped,
|
2020-05-06 11:42:56 +00:00
|
|
|
};
|
2023-05-09 10:48:15 +00:00
|
|
|
use genesis::{interop_genesis_state, DEFAULT_ETH1_BLOCK_HASH};
|
2020-07-25 02:03:18 +00:00
|
|
|
use int_to_bytes::int_to_bytes32;
|
2021-10-01 19:57:50 +00:00
|
|
|
use lazy_static::lazy_static;
|
2020-09-27 20:59:40 +00:00
|
|
|
use state_processing::{
|
|
|
|
per_block_processing::errors::AttestationValidationError, per_slot_processing,
|
|
|
|
};
|
2020-05-06 11:42:56 +00:00
|
|
|
use tree_hash::TreeHash;
|
|
|
|
use types::{
|
2023-05-09 10:48:15 +00:00
|
|
|
test_utils::generate_deterministic_keypair, Address, AggregateSignature, Attestation,
|
2023-05-15 02:10:41 +00:00
|
|
|
BeaconStateError, BitList, ChainSpec, Epoch, EthSpec, ForkName, Hash256, Keypair,
|
|
|
|
MainnetEthSpec, SecretKey, SelectionProof, SignedAggregateAndProof, Slot, SubnetId, Unsigned,
|
2020-05-06 11:42:56 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
pub type E = MainnetEthSpec;
|
|
|
|
|
|
|
|
/// The validator count needs to be relatively high compared to other tests to ensure that we can
|
|
|
|
/// have committees where _some_ validators are aggregators but not _all_.
|
|
|
|
pub const VALIDATOR_COUNT: usize = 256;
|
|
|
|
|
2023-05-15 02:10:41 +00:00
|
|
|
pub const CAPELLA_FORK_EPOCH: usize = 1;
|
|
|
|
|
2020-05-06 11:42:56 +00:00
|
|
|
lazy_static! {
|
|
|
|
/// A cached set of keys.
|
|
|
|
static ref KEYPAIRS: Vec<Keypair> = types::test_utils::generate_deterministic_keypairs(VALIDATOR_COUNT);
|
|
|
|
}
|
|
|
|
|
|
|
|
/// Returns a beacon chain harness.
|
2020-10-19 05:58:39 +00:00
|
|
|
fn get_harness(validator_count: usize) -> BeaconChainHarness<EphemeralHarnessType<E>> {
|
2021-08-30 06:41:31 +00:00
|
|
|
let mut spec = test_spec::<E>();
|
|
|
|
|
|
|
|
// A kind-of arbitrary number that ensures that _some_ validators are aggregators, but
|
|
|
|
// not all.
|
|
|
|
spec.target_aggregators_per_committee = 4;
|
|
|
|
|
2021-10-14 02:58:10 +00:00
|
|
|
let harness = BeaconChainHarness::builder(MainnetEthSpec)
|
|
|
|
.spec(spec)
|
2023-08-21 05:02:32 +00:00
|
|
|
.chain_config(ChainConfig {
|
|
|
|
reconstruct_historic_states: true,
|
|
|
|
..ChainConfig::default()
|
|
|
|
})
|
2021-10-14 02:58:10 +00:00
|
|
|
.keypairs(KEYPAIRS[0..validator_count].to_vec())
|
|
|
|
.fresh_ephemeral_store()
|
2022-02-28 22:07:48 +00:00
|
|
|
.mock_execution_layer()
|
2021-10-14 02:58:10 +00:00
|
|
|
.build();
|
2020-05-06 11:42:56 +00:00
|
|
|
|
|
|
|
harness.advance_slot();
|
|
|
|
|
|
|
|
harness
|
|
|
|
}
|
|
|
|
|
2023-05-09 10:48:15 +00:00
|
|
|
/// Returns a beacon chain harness with Capella fork enabled at epoch 1, and
|
|
|
|
/// all genesis validators start with BLS withdrawal credentials.
|
2023-05-15 02:10:41 +00:00
|
|
|
fn get_harness_capella_spec(
|
|
|
|
validator_count: usize,
|
|
|
|
) -> (BeaconChainHarness<EphemeralHarnessType<E>>, ChainSpec) {
|
2023-05-09 10:48:15 +00:00
|
|
|
let mut spec = E::default_spec();
|
|
|
|
spec.altair_fork_epoch = Some(Epoch::new(0));
|
|
|
|
spec.bellatrix_fork_epoch = Some(Epoch::new(0));
|
2023-05-15 02:10:41 +00:00
|
|
|
spec.capella_fork_epoch = Some(Epoch::new(CAPELLA_FORK_EPOCH as u64));
|
2023-05-09 10:48:15 +00:00
|
|
|
|
|
|
|
let validator_keypairs = KEYPAIRS[0..validator_count].to_vec();
|
|
|
|
let genesis_state = interop_genesis_state(
|
|
|
|
&validator_keypairs,
|
|
|
|
HARNESS_GENESIS_TIME,
|
|
|
|
Hash256::from_slice(DEFAULT_ETH1_BLOCK_HASH),
|
|
|
|
None,
|
|
|
|
&spec,
|
|
|
|
)
|
|
|
|
.unwrap();
|
|
|
|
|
|
|
|
let harness = BeaconChainHarness::builder(MainnetEthSpec)
|
2023-05-15 02:10:41 +00:00
|
|
|
.spec(spec.clone())
|
2023-08-21 05:02:32 +00:00
|
|
|
.chain_config(ChainConfig {
|
|
|
|
reconstruct_historic_states: true,
|
|
|
|
..ChainConfig::default()
|
|
|
|
})
|
2023-05-09 10:48:15 +00:00
|
|
|
.keypairs(validator_keypairs)
|
|
|
|
.withdrawal_keypairs(
|
|
|
|
KEYPAIRS[0..validator_count]
|
|
|
|
.iter()
|
|
|
|
.cloned()
|
|
|
|
.map(Some)
|
|
|
|
.collect(),
|
|
|
|
)
|
|
|
|
.genesis_state_ephemeral_store(genesis_state)
|
|
|
|
.mock_execution_layer()
|
|
|
|
.build();
|
|
|
|
|
|
|
|
harness
|
|
|
|
.execution_block_generator()
|
|
|
|
.move_to_terminal_block()
|
|
|
|
.unwrap();
|
|
|
|
|
|
|
|
harness.advance_slot();
|
|
|
|
|
2023-05-15 02:10:41 +00:00
|
|
|
(harness, spec)
|
2023-05-09 10:48:15 +00:00
|
|
|
}
|
|
|
|
|
2020-05-06 11:42:56 +00:00
|
|
|
/// Returns an attestation that is valid for some slot in the given `chain`.
|
|
|
|
///
|
|
|
|
/// Also returns some info about who created it.
|
|
|
|
fn get_valid_unaggregated_attestation<T: BeaconChainTypes>(
|
|
|
|
chain: &BeaconChain<T>,
|
2020-06-18 09:11:03 +00:00
|
|
|
) -> (Attestation<T::EthSpec>, usize, usize, SecretKey, SubnetId) {
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
let head = chain.head_snapshot();
|
2020-05-06 11:42:56 +00:00
|
|
|
let current_slot = chain.slot().expect("should get slot");
|
|
|
|
|
|
|
|
let mut valid_attestation = chain
|
|
|
|
.produce_unaggregated_attestation(current_slot, 0)
|
|
|
|
.expect("should not error while producing attestation");
|
|
|
|
|
|
|
|
let validator_committee_index = 0;
|
|
|
|
let validator_index = *head
|
|
|
|
.beacon_state
|
|
|
|
.get_beacon_committee(current_slot, valid_attestation.data.index)
|
|
|
|
.expect("should get committees")
|
|
|
|
.committee
|
|
|
|
.get(validator_committee_index)
|
|
|
|
.expect("there should be an attesting validator");
|
|
|
|
|
|
|
|
let validator_sk = generate_deterministic_keypair(validator_index).sk;
|
|
|
|
|
|
|
|
valid_attestation
|
|
|
|
.sign(
|
|
|
|
&validator_sk,
|
|
|
|
validator_committee_index,
|
2021-07-09 06:15:32 +00:00
|
|
|
&head.beacon_state.fork(),
|
2020-05-06 11:42:56 +00:00
|
|
|
chain.genesis_validators_root,
|
|
|
|
&chain.spec,
|
|
|
|
)
|
|
|
|
.expect("should sign attestation");
|
|
|
|
|
2020-06-18 09:11:03 +00:00
|
|
|
let subnet_id = SubnetId::compute_subnet_for_attestation_data::<E>(
|
|
|
|
&valid_attestation.data,
|
|
|
|
head.beacon_state
|
|
|
|
.get_committee_count_at_slot(current_slot)
|
|
|
|
.expect("should get committee count"),
|
|
|
|
&chain.spec,
|
|
|
|
)
|
|
|
|
.expect("should get subnet_id");
|
|
|
|
|
2020-05-06 11:42:56 +00:00
|
|
|
(
|
|
|
|
valid_attestation,
|
|
|
|
validator_index,
|
|
|
|
validator_committee_index,
|
|
|
|
validator_sk,
|
2020-06-18 09:11:03 +00:00
|
|
|
subnet_id,
|
2020-05-06 11:42:56 +00:00
|
|
|
)
|
|
|
|
}
|
|
|
|
|
|
|
|
fn get_valid_aggregated_attestation<T: BeaconChainTypes>(
|
|
|
|
chain: &BeaconChain<T>,
|
2020-06-05 01:32:46 +00:00
|
|
|
aggregate: Attestation<T::EthSpec>,
|
2020-05-06 11:42:56 +00:00
|
|
|
) -> (SignedAggregateAndProof<T::EthSpec>, usize, SecretKey) {
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
let head = chain.head_snapshot();
|
|
|
|
let state = &head.beacon_state;
|
2020-05-06 11:42:56 +00:00
|
|
|
let current_slot = chain.slot().expect("should get slot");
|
|
|
|
|
|
|
|
let committee = state
|
|
|
|
.get_beacon_committee(current_slot, aggregate.data.index)
|
|
|
|
.expect("should get committees");
|
|
|
|
let committee_len = committee.committee.len();
|
|
|
|
|
2020-06-05 01:32:46 +00:00
|
|
|
let (aggregator_index, aggregator_sk) = committee
|
2020-05-06 11:42:56 +00:00
|
|
|
.committee
|
|
|
|
.iter()
|
2020-06-05 01:32:46 +00:00
|
|
|
.find_map(|&val_index| {
|
2020-05-06 11:42:56 +00:00
|
|
|
let aggregator_sk = generate_deterministic_keypair(val_index).sk;
|
|
|
|
|
|
|
|
let proof = SelectionProof::new::<T::EthSpec>(
|
|
|
|
aggregate.data.slot,
|
|
|
|
&aggregator_sk,
|
2021-07-09 06:15:32 +00:00
|
|
|
&state.fork(),
|
2020-05-06 11:42:56 +00:00
|
|
|
chain.genesis_validators_root,
|
|
|
|
&chain.spec,
|
|
|
|
);
|
|
|
|
|
|
|
|
if proof.is_aggregator(committee_len, &chain.spec).unwrap() {
|
2020-06-05 01:32:46 +00:00
|
|
|
Some((val_index, aggregator_sk))
|
2020-05-06 11:42:56 +00:00
|
|
|
} else {
|
|
|
|
None
|
|
|
|
}
|
|
|
|
})
|
|
|
|
.expect("should find aggregator for committee");
|
|
|
|
|
|
|
|
let signed_aggregate = SignedAggregateAndProof::from_aggregate(
|
|
|
|
aggregator_index as u64,
|
|
|
|
aggregate,
|
|
|
|
None,
|
|
|
|
&aggregator_sk,
|
2021-07-09 06:15:32 +00:00
|
|
|
&state.fork(),
|
2020-05-06 11:42:56 +00:00
|
|
|
chain.genesis_validators_root,
|
|
|
|
&chain.spec,
|
|
|
|
);
|
|
|
|
|
|
|
|
(signed_aggregate, aggregator_index, aggregator_sk)
|
|
|
|
}
|
|
|
|
|
|
|
|
/// Returns a proof and index for a validator that is **not** an aggregator for the given
|
|
|
|
/// attestation.
|
|
|
|
fn get_non_aggregator<T: BeaconChainTypes>(
|
|
|
|
chain: &BeaconChain<T>,
|
|
|
|
aggregate: &Attestation<T::EthSpec>,
|
|
|
|
) -> (usize, SecretKey) {
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
let head = chain.head_snapshot();
|
|
|
|
let state = &head.beacon_state;
|
2020-05-06 11:42:56 +00:00
|
|
|
let current_slot = chain.slot().expect("should get slot");
|
|
|
|
|
|
|
|
let committee = state
|
|
|
|
.get_beacon_committee(current_slot, aggregate.data.index)
|
|
|
|
.expect("should get committees");
|
|
|
|
let committee_len = committee.committee.len();
|
|
|
|
|
|
|
|
committee
|
|
|
|
.committee
|
|
|
|
.iter()
|
|
|
|
.find_map(|&val_index| {
|
|
|
|
let aggregator_sk = generate_deterministic_keypair(val_index).sk;
|
|
|
|
|
|
|
|
let proof = SelectionProof::new::<T::EthSpec>(
|
|
|
|
aggregate.data.slot,
|
|
|
|
&aggregator_sk,
|
2021-07-09 06:15:32 +00:00
|
|
|
&state.fork(),
|
2020-05-06 11:42:56 +00:00
|
|
|
chain.genesis_validators_root,
|
|
|
|
&chain.spec,
|
|
|
|
);
|
|
|
|
|
|
|
|
if proof.is_aggregator(committee_len, &chain.spec).unwrap() {
|
|
|
|
None
|
|
|
|
} else {
|
|
|
|
Some((val_index, aggregator_sk))
|
|
|
|
}
|
|
|
|
})
|
|
|
|
.expect("should find non-aggregator for committee")
|
|
|
|
}
|
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
struct GossipTester {
|
|
|
|
harness: BeaconChainHarness<EphemeralHarnessType<E>>,
|
2020-05-06 11:42:56 +00:00
|
|
|
/*
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
* Valid unaggregated attestation
|
2020-05-06 11:42:56 +00:00
|
|
|
*/
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
valid_attestation: Attestation<E>,
|
|
|
|
attester_validator_index: usize,
|
|
|
|
attester_committee_index: usize,
|
|
|
|
attester_sk: SecretKey,
|
|
|
|
attestation_subnet_id: SubnetId,
|
2021-04-13 00:24:39 +00:00
|
|
|
/*
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
* Valid unaggregated attestation for batch testing
|
2021-04-13 00:24:39 +00:00
|
|
|
*/
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
invalid_attestation: Attestation<E>,
|
2020-10-02 10:46:37 +00:00
|
|
|
/*
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
* Valid aggregate
|
2020-10-02 10:46:37 +00:00
|
|
|
*/
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
valid_aggregate: SignedAggregateAndProof<E>,
|
|
|
|
aggregator_validator_index: usize,
|
|
|
|
aggregator_sk: SecretKey,
|
2020-05-06 11:42:56 +00:00
|
|
|
/*
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
* Another valid aggregate for batch testing
|
2020-05-06 11:42:56 +00:00
|
|
|
*/
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
invalid_aggregate: SignedAggregateAndProof<E>,
|
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
impl GossipTester {
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
pub async fn new() -> Self {
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
let harness = get_harness(VALIDATOR_COUNT);
|
|
|
|
|
|
|
|
// Extend the chain out a few epochs so we have some chain depth to play with.
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize * 3 - 1,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::AllValidators,
|
|
|
|
)
|
|
|
|
.await;
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
|
|
|
|
// Advance into a slot where there have not been blocks or attestations produced.
|
|
|
|
harness.advance_slot();
|
|
|
|
|
|
|
|
let (
|
|
|
|
valid_attestation,
|
|
|
|
attester_validator_index,
|
|
|
|
attester_committee_index,
|
|
|
|
attester_sk,
|
|
|
|
attestation_subnet_id,
|
|
|
|
) = get_valid_unaggregated_attestation(&harness.chain);
|
|
|
|
|
|
|
|
let (valid_aggregate, aggregator_validator_index, aggregator_sk) =
|
|
|
|
get_valid_aggregated_attestation(&harness.chain, valid_attestation.clone());
|
|
|
|
|
|
|
|
let mut invalid_attestation = valid_attestation.clone();
|
|
|
|
invalid_attestation.data.beacon_block_root = Hash256::repeat_byte(13);
|
|
|
|
|
|
|
|
let (mut invalid_aggregate, _, _) =
|
|
|
|
get_valid_aggregated_attestation(&harness.chain, invalid_attestation.clone());
|
|
|
|
invalid_aggregate.message.aggregator_index = invalid_aggregate
|
|
|
|
.message
|
|
|
|
.aggregator_index
|
|
|
|
.checked_sub(1)
|
|
|
|
.unwrap();
|
|
|
|
|
|
|
|
Self {
|
|
|
|
harness,
|
|
|
|
valid_attestation,
|
|
|
|
attester_validator_index,
|
|
|
|
attester_committee_index,
|
|
|
|
attester_sk,
|
|
|
|
attestation_subnet_id,
|
|
|
|
invalid_attestation,
|
|
|
|
valid_aggregate,
|
|
|
|
aggregator_validator_index,
|
|
|
|
aggregator_sk,
|
|
|
|
invalid_aggregate,
|
2020-05-06 11:42:56 +00:00
|
|
|
}
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
pub fn slot(&self) -> Slot {
|
|
|
|
self.harness.chain.slot().unwrap()
|
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
pub fn epoch(&self) -> Epoch {
|
|
|
|
self.harness.chain.epoch().unwrap()
|
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
2023-08-09 19:44:47 +00:00
|
|
|
pub fn earliest_valid_attestation_slot(&self) -> Slot {
|
|
|
|
let offset = match self.harness.spec.fork_name_at_epoch(self.epoch()) {
|
|
|
|
ForkName::Base | ForkName::Altair | ForkName::Merge | ForkName::Capella => {
|
|
|
|
// Subtract an additional slot since the harness will be exactly on the start of the
|
|
|
|
// slot and the propagation tolerance will allow an extra slot.
|
|
|
|
E::slots_per_epoch() + 1
|
|
|
|
}
|
|
|
|
// EIP-7045
|
|
|
|
ForkName::Deneb => {
|
|
|
|
let epoch_slot_offset = (self.slot() % E::slots_per_epoch()).as_u64();
|
|
|
|
if epoch_slot_offset != 0 {
|
|
|
|
E::slots_per_epoch() + epoch_slot_offset
|
|
|
|
} else {
|
|
|
|
// Here the propagation tolerance will cause the cutoff to be an entire epoch earlier
|
|
|
|
2 * E::slots_per_epoch()
|
|
|
|
}
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
self.slot()
|
|
|
|
.as_u64()
|
2023-08-09 19:44:47 +00:00
|
|
|
.checked_sub(offset)
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
.expect("chain is not sufficiently deep for test")
|
|
|
|
.into()
|
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
pub fn non_aggregator(&self) -> (usize, SecretKey) {
|
|
|
|
get_non_aggregator(&self.harness.chain, &self.valid_aggregate.message.aggregate)
|
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
pub fn import_valid_aggregate(self) -> Self {
|
|
|
|
assert!(
|
|
|
|
self.harness
|
|
|
|
.chain
|
|
|
|
.verify_aggregated_attestation_for_gossip(&self.valid_aggregate)
|
|
|
|
.is_ok(),
|
|
|
|
"valid aggregate should be verified"
|
|
|
|
);
|
|
|
|
self
|
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
pub fn import_valid_unaggregate(self) -> Self {
|
|
|
|
self.harness
|
|
|
|
.chain
|
|
|
|
.verify_unaggregated_attestation_for_gossip(
|
|
|
|
&self.valid_attestation,
|
|
|
|
Some(self.attestation_subnet_id),
|
2020-05-06 11:42:56 +00:00
|
|
|
)
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
.expect("valid attestation should be verified");
|
|
|
|
self
|
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
pub fn inspect_aggregate_err<G, I>(self, desc: &str, get_attn: G, inspect_err: I) -> Self
|
|
|
|
where
|
|
|
|
G: Fn(&Self, &mut SignedAggregateAndProof<E>),
|
|
|
|
I: Fn(&Self, AttnError),
|
|
|
|
{
|
|
|
|
let mut aggregate = self.valid_aggregate.clone();
|
|
|
|
get_attn(&self, &mut aggregate);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Individual verification
|
|
|
|
*/
|
|
|
|
let err = self
|
|
|
|
.harness
|
2020-05-06 11:42:56 +00:00
|
|
|
.chain
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
.verify_aggregated_attestation_for_gossip(&aggregate)
|
|
|
|
.err()
|
|
|
|
.expect(&format!(
|
|
|
|
"{} should error during verify_aggregated_attestation_for_gossip",
|
|
|
|
desc
|
|
|
|
));
|
|
|
|
inspect_err(&self, err);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Batch verification
|
|
|
|
*/
|
|
|
|
let mut results = self
|
|
|
|
.harness
|
|
|
|
.chain
|
|
|
|
.batch_verify_aggregated_attestations_for_gossip(
|
|
|
|
vec![&self.invalid_aggregate, &aggregate].into_iter(),
|
|
|
|
)
|
|
|
|
.unwrap();
|
|
|
|
assert_eq!(results.len(), 2);
|
|
|
|
let batch_err = results.pop().unwrap().err().expect(&format!(
|
|
|
|
"{} should error during batch_verify_aggregated_attestations_for_gossip",
|
|
|
|
desc
|
|
|
|
));
|
|
|
|
inspect_err(&self, batch_err);
|
|
|
|
|
|
|
|
self
|
|
|
|
}
|
2020-05-06 11:42:56 +00:00
|
|
|
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
pub fn inspect_unaggregate_err<G, I>(self, desc: &str, get_attn: G, inspect_err: I) -> Self
|
|
|
|
where
|
|
|
|
G: Fn(&Self, &mut Attestation<E>, &mut SubnetId),
|
|
|
|
I: Fn(&Self, AttnError),
|
|
|
|
{
|
|
|
|
let mut attn = self.valid_attestation.clone();
|
|
|
|
let mut subnet_id = self.attestation_subnet_id;
|
|
|
|
get_attn(&self, &mut attn, &mut subnet_id);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Individual verification
|
|
|
|
*/
|
|
|
|
let err = self
|
|
|
|
.harness
|
|
|
|
.chain
|
|
|
|
.verify_unaggregated_attestation_for_gossip(&attn, Some(subnet_id))
|
|
|
|
.err()
|
|
|
|
.expect(&format!(
|
|
|
|
"{} should error during verify_unaggregated_attestation_for_gossip",
|
|
|
|
desc
|
|
|
|
));
|
|
|
|
inspect_err(&self, err);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Batch verification
|
|
|
|
*/
|
|
|
|
let mut results = self
|
|
|
|
.harness
|
|
|
|
.chain
|
|
|
|
.batch_verify_unaggregated_attestations_for_gossip(
|
|
|
|
vec![
|
|
|
|
(&self.invalid_attestation, Some(subnet_id)),
|
|
|
|
(&attn, Some(subnet_id)),
|
|
|
|
]
|
|
|
|
.into_iter(),
|
|
|
|
)
|
|
|
|
.unwrap();
|
|
|
|
assert_eq!(results.len(), 2);
|
|
|
|
let batch_err = results.pop().unwrap().err().expect(&format!(
|
|
|
|
"{} should error during batch_verify_unaggregated_attestations_for_gossip",
|
|
|
|
desc
|
|
|
|
));
|
|
|
|
inspect_err(&self, batch_err);
|
|
|
|
|
|
|
|
self
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/// Tests verification of `SignedAggregateAndProof` from the gossip network.
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
#[tokio::test]
|
|
|
|
async fn aggregated_gossip_verification() {
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
GossipTester::new()
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
.await
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
/*
|
|
|
|
* The following two tests ensure:
|
|
|
|
*
|
|
|
|
* aggregate.data.slot is within the last ATTESTATION_PROPAGATION_SLOT_RANGE slots (with a
|
|
|
|
* MAXIMUM_GOSSIP_CLOCK_DISPARITY allowance) -- i.e. aggregate.data.slot +
|
|
|
|
* ATTESTATION_PROPAGATION_SLOT_RANGE >= current_slot >= aggregate.data.slot (a client MAY
|
|
|
|
* queue future aggregates for processing at the appropriate slot).
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate from future slot",
|
|
|
|
|tester, a| a.message.aggregate.data.slot = tester.slot() + 1,
|
|
|
|
|tester, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::FutureSlot { attestation_slot, latest_permissible_slot }
|
|
|
|
if attestation_slot == tester.slot() + 1
|
|
|
|
&& latest_permissible_slot == tester.slot()
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate from past slot",
|
2023-08-09 19:44:47 +00:00
|
|
|
|tester, a| {
|
|
|
|
let too_early_slot = tester.earliest_valid_attestation_slot() - 1;
|
|
|
|
a.message.aggregate.data.slot = too_early_slot;
|
|
|
|
a.message.aggregate.data.target.epoch = too_early_slot.epoch(E::slots_per_epoch());
|
|
|
|
},
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
|tester, err| {
|
2023-08-09 19:44:47 +00:00
|
|
|
let valid_early_slot = tester.earliest_valid_attestation_slot();
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::PastSlot {
|
|
|
|
attestation_slot,
|
|
|
|
earliest_permissible_slot
|
|
|
|
}
|
2023-08-09 19:44:47 +00:00
|
|
|
if attestation_slot == valid_early_slot - 1
|
|
|
|
&& earliest_permissible_slot == valid_early_slot
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The aggregate attestation's epoch matches its target -- i.e. `aggregate.data.target.epoch ==
|
|
|
|
* compute_epoch_at_slot(attestation.data.slot)`
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"attestation with invalid target epoch",
|
|
|
|
|_, a| a.message.aggregate.data.target.epoch += 1,
|
|
|
|
|_, err| assert!(matches!(err, AttnError::InvalidTargetEpoch { .. })),
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* This is not in the specification for aggregate attestations (only unaggregates), but we
|
|
|
|
* check it anyway to avoid weird edge cases.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"attestation with invalid target root",
|
|
|
|
|_, a| a.message.aggregate.data.target.root = Hash256::repeat_byte(42),
|
|
|
|
|_, err| assert!(matches!(err, AttnError::InvalidTargetRoot { .. })),
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The block being voted for (aggregate.data.beacon_block_root) passes validation.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate with unknown head block",
|
|
|
|
|_, a| a.message.aggregate.data.beacon_block_root = Hash256::repeat_byte(42),
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::UnknownHeadBlock {
|
|
|
|
beacon_block_root
|
|
|
|
}
|
|
|
|
if beacon_block_root == Hash256::repeat_byte(42)
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The attestation has participants.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate with no participants",
|
|
|
|
|_, a| {
|
|
|
|
let aggregation_bits = &mut a.message.aggregate.aggregation_bits;
|
|
|
|
aggregation_bits.difference_inplace(&aggregation_bits.clone());
|
|
|
|
assert!(aggregation_bits.is_zero());
|
|
|
|
a.message.aggregate.signature = AggregateSignature::infinity();
|
|
|
|
},
|
|
|
|
|_, err| assert!(matches!(err, AttnError::EmptyAggregationBitfield)),
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* This test ensures:
|
|
|
|
*
|
|
|
|
* The aggregator signature, signed_aggregate_and_proof.signature, is valid.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate with bad signature",
|
|
|
|
|tester, a| a.signature = tester.aggregator_sk.sign(Hash256::repeat_byte(42)),
|
|
|
|
|_, err| assert!(matches!(err, AttnError::InvalidSignature)),
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The aggregate_and_proof.selection_proof is a valid signature of the aggregate.data.slot by
|
|
|
|
* the validator with index aggregate_and_proof.aggregator_index.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate with bad signature",
|
|
|
|
|tester, a| {
|
|
|
|
let committee_len = tester
|
|
|
|
.harness
|
|
|
|
.chain
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
.head_snapshot()
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
.beacon_state
|
|
|
|
.get_beacon_committee(tester.slot(), a.message.aggregate.data.index)
|
|
|
|
.expect("should get committees")
|
|
|
|
.committee
|
|
|
|
.len();
|
|
|
|
|
|
|
|
// Generate some random signature until happens to be a valid selection proof. We need
|
|
|
|
// this in order to reach the signature verification code.
|
|
|
|
//
|
|
|
|
// Could run for ever, but that seems _really_ improbable.
|
|
|
|
let mut i: u64 = 0;
|
|
|
|
a.message.selection_proof = loop {
|
|
|
|
i += 1;
|
|
|
|
let proof: SelectionProof = tester
|
|
|
|
.aggregator_sk
|
|
|
|
.sign(Hash256::from_slice(&int_to_bytes32(i)))
|
|
|
|
.into();
|
|
|
|
if proof
|
|
|
|
.is_aggregator(committee_len, &tester.harness.chain.spec)
|
|
|
|
.unwrap()
|
|
|
|
{
|
|
|
|
break proof.into();
|
|
|
|
}
|
|
|
|
};
|
|
|
|
},
|
|
|
|
|_, err| assert!(matches!(err, AttnError::InvalidSignature)),
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The signature of aggregate is valid.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate with bad aggregate signature",
|
|
|
|
|tester, a| {
|
|
|
|
let mut agg_sig = AggregateSignature::infinity();
|
|
|
|
agg_sig.add_assign(&tester.aggregator_sk.sign(Hash256::repeat_byte(42)));
|
|
|
|
a.message.aggregate.signature = agg_sig;
|
|
|
|
},
|
|
|
|
|_, err| assert!(matches!(err, AttnError::InvalidSignature)),
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* Not directly in the specification, but a sanity check.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate with too-high aggregator index",
|
|
|
|
|_, a| {
|
|
|
|
a.message.aggregator_index = <E as EthSpec>::ValidatorRegistryLimit::to_u64() + 1
|
|
|
|
},
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::ValidatorIndexTooHigh(index)
|
|
|
|
if index == (<E as EthSpec>::ValidatorRegistryLimit::to_u64() + 1) as usize
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The aggregator's validator index is within the committee -- i.e.
|
|
|
|
* aggregate_and_proof.aggregator_index in get_beacon_committee(state, aggregate.data.slot,
|
|
|
|
* aggregate.data.index).
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate with unknown aggregator index",
|
|
|
|
|_, a| a.message.aggregator_index = VALIDATOR_COUNT as u64,
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
// Naively we should think this condition would trigger this error:
|
|
|
|
//
|
|
|
|
// AttnError::AggregatorPubkeyUnknown(unknown_validator)
|
|
|
|
//
|
|
|
|
// However the following error is triggered first:
|
|
|
|
AttnError::AggregatorNotInCommittee {
|
|
|
|
aggregator_index
|
|
|
|
}
|
|
|
|
if aggregator_index == VALIDATOR_COUNT as u64
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* aggregate_and_proof.selection_proof selects the validator as an aggregator for the slot --
|
|
|
|
* i.e. is_aggregator(state, aggregate.data.slot, aggregate.data.index,
|
|
|
|
* aggregate_and_proof.selection_proof) returns True.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate from non-aggregator",
|
|
|
|
|tester, a| {
|
|
|
|
let chain = &tester.harness.chain;
|
|
|
|
let (index, sk) = tester.non_aggregator();
|
|
|
|
*a = SignedAggregateAndProof::from_aggregate(
|
|
|
|
index as u64,
|
|
|
|
tester.valid_aggregate.message.aggregate.clone(),
|
|
|
|
None,
|
|
|
|
&sk,
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
&chain.canonical_head.cached_head().head_fork(),
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
chain.genesis_validators_root,
|
|
|
|
&chain.spec,
|
|
|
|
)
|
|
|
|
},
|
|
|
|
|tester, err| {
|
|
|
|
let (val_index, _) = tester.non_aggregator();
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::InvalidSelectionProof {
|
|
|
|
aggregator_index: index
|
|
|
|
}
|
|
|
|
if index == val_index as u64
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
// NOTE: from here on, the tests are stateful, and rely on the valid attestation having
|
|
|
|
// been seen.
|
|
|
|
.import_valid_aggregate()
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The valid aggregate attestation defined by hash_tree_root(aggregate) has not already been
|
|
|
|
* seen (via aggregate gossip, within a block, or through the creation of an equivalent
|
|
|
|
* aggregate locally).
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate that has already been seen",
|
|
|
|
|_, _| {},
|
|
|
|
|tester, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
2023-06-27 01:06:49 +00:00
|
|
|
AttnError::AttestationSupersetKnown(hash)
|
|
|
|
if hash == tester.valid_aggregate.message.aggregate.data.tree_hash_root()
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The aggregate is the first valid aggregate received for the aggregator with index
|
|
|
|
* aggregate_and_proof.aggregator_index for the epoch aggregate.data.target.epoch.
|
|
|
|
*/
|
|
|
|
.inspect_aggregate_err(
|
|
|
|
"aggregate from aggregator that has already been seen",
|
|
|
|
|_, a| a.message.aggregate.data.beacon_block_root = Hash256::repeat_byte(42),
|
|
|
|
|tester, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::AggregatorAlreadyKnown(index)
|
|
|
|
if index == tester.aggregator_validator_index as u64
|
|
|
|
))
|
|
|
|
},
|
|
|
|
);
|
2020-05-06 11:42:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/// Tests the verification conditions for an unaggregated attestation on the gossip network.
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
#[tokio::test]
|
|
|
|
async fn unaggregated_gossip_verification() {
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
GossipTester::new()
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
.await
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The committee index is within the expected range -- i.e. `data.index <
|
|
|
|
* get_committee_count_per_slot(state, data.target.epoch)`.
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation with invalid committee index",
|
|
|
|
|tester, a, _| {
|
|
|
|
a.data.index = tester
|
|
|
|
.harness
|
|
|
|
.chain
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
.head_snapshot()
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
.beacon_state
|
|
|
|
.get_committee_count_at_slot(a.data.slot)
|
|
|
|
.unwrap()
|
|
|
|
},
|
|
|
|
|_, err| assert!(matches!(err, AttnError::NoCommitteeForSlotAndIndex { .. })),
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The attestation is for the correct subnet (i.e. compute_subnet_for_attestation(state,
|
|
|
|
* attestation.data.slot, attestation.data.index) == subnet_id).
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation with invalid committee index",
|
|
|
|
|_, _, subnet_id| *subnet_id = SubnetId::new(42),
|
|
|
|
|tester, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::InvalidSubnetId {
|
|
|
|
received,
|
|
|
|
expected,
|
|
|
|
}
|
|
|
|
if received == SubnetId::new(42) && expected == tester.attestation_subnet_id
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following two tests ensure:
|
|
|
|
*
|
|
|
|
* attestation.data.slot is within the last ATTESTATION_PROPAGATION_SLOT_RANGE slots (within a
|
|
|
|
* MAXIMUM_GOSSIP_CLOCK_DISPARITY allowance) -- i.e. attestation.data.slot +
|
|
|
|
* ATTESTATION_PROPAGATION_SLOT_RANGE >= current_slot >= attestation.data.slot (a client MAY
|
|
|
|
* queue future attestations for processing at the appropriate slot).
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation from future slot",
|
|
|
|
|tester, a, _| a.data.slot = tester.slot() + 1,
|
|
|
|
|tester, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::FutureSlot {
|
|
|
|
attestation_slot,
|
|
|
|
latest_permissible_slot,
|
|
|
|
}
|
|
|
|
if attestation_slot == tester.slot() + 1 && latest_permissible_slot == tester.slot()
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation from past slot",
|
|
|
|
|tester, a, _| {
|
2023-08-09 19:44:47 +00:00
|
|
|
let too_early_slot = tester.earliest_valid_attestation_slot() - 1;
|
|
|
|
a.data.slot = too_early_slot;
|
|
|
|
a.data.target.epoch = too_early_slot.epoch(E::slots_per_epoch());
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
},
|
|
|
|
|tester, err| {
|
2023-08-09 19:44:47 +00:00
|
|
|
let valid_early_slot = tester.earliest_valid_attestation_slot();
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::PastSlot {
|
|
|
|
attestation_slot,
|
|
|
|
earliest_permissible_slot,
|
|
|
|
}
|
2023-08-09 19:44:47 +00:00
|
|
|
if attestation_slot == valid_early_slot - 1
|
|
|
|
&& earliest_permissible_slot == valid_early_slot
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The attestation's epoch matches its target -- i.e. `attestation.data.target.epoch ==
|
|
|
|
* compute_epoch_at_slot(attestation.data.slot)`
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation with invalid target epoch",
|
|
|
|
|_, a, _| a.data.target.epoch += 1,
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::InvalidTargetEpoch { .. }
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following two tests ensure:
|
|
|
|
*
|
|
|
|
* The attestation is unaggregated -- that is, it has exactly one participating validator
|
|
|
|
* (len([bit for bit in attestation.aggregation_bits if bit == 0b1]) == 1).
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation without any aggregation bits set",
|
|
|
|
|tester, a, _| {
|
|
|
|
a.aggregation_bits
|
|
|
|
.set(tester.attester_committee_index, false)
|
|
|
|
.expect("should unset aggregation bit");
|
|
|
|
assert_eq!(
|
|
|
|
a.aggregation_bits.num_set_bits(),
|
|
|
|
0,
|
|
|
|
"test requires no set bits"
|
|
|
|
);
|
|
|
|
},
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::NotExactlyOneAggregationBitSet(0)
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation with two aggregation bits set",
|
|
|
|
|tester, a, _| {
|
|
|
|
a.aggregation_bits
|
|
|
|
.set(tester.attester_committee_index + 1, true)
|
|
|
|
.expect("should set second aggregation bit");
|
|
|
|
},
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::NotExactlyOneAggregationBitSet(2)
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures:
|
|
|
|
*
|
|
|
|
* The number of aggregation bits matches the committee size -- i.e.
|
|
|
|
* `len(attestation.aggregation_bits) == len(get_beacon_committee(state, data.slot,
|
|
|
|
* data.index))`.
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation with invalid bitfield",
|
|
|
|
|_, a, _| {
|
|
|
|
let bits = a.aggregation_bits.iter().collect::<Vec<_>>();
|
|
|
|
a.aggregation_bits = BitList::with_capacity(bits.len() + 1).unwrap();
|
|
|
|
for (i, bit) in bits.into_iter().enumerate() {
|
|
|
|
a.aggregation_bits.set(i, bit).unwrap();
|
|
|
|
}
|
|
|
|
},
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::Invalid(AttestationValidationError::BeaconStateError(
|
|
|
|
BeaconStateError::InvalidBitfield
|
|
|
|
))
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures that:
|
|
|
|
*
|
|
|
|
* The block being voted for (attestation.data.beacon_block_root) passes validation.
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation with unknown head block",
|
|
|
|
|_, a, _| {
|
|
|
|
a.data.beacon_block_root = Hash256::repeat_byte(42);
|
|
|
|
},
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::UnknownHeadBlock {
|
|
|
|
beacon_block_root,
|
|
|
|
}
|
|
|
|
if beacon_block_root == Hash256::repeat_byte(42)
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures that:
|
|
|
|
*
|
|
|
|
* Spec v0.12.3
|
|
|
|
*
|
|
|
|
* The attestation's target block is an ancestor of the block named in the LMD vote
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation with invalid target root",
|
|
|
|
|_, a, _| {
|
|
|
|
a.data.target.root = Hash256::repeat_byte(42);
|
|
|
|
},
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::InvalidTargetRoot { .. }
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
/*
|
|
|
|
* The following test ensures that:
|
|
|
|
*
|
|
|
|
* The signature of attestation is valid.
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation with bad signature",
|
|
|
|
|tester, a, _| {
|
|
|
|
let mut agg_sig = AggregateSignature::infinity();
|
|
|
|
agg_sig.add_assign(&tester.attester_sk.sign(Hash256::repeat_byte(42)));
|
|
|
|
a.signature = agg_sig;
|
|
|
|
},
|
|
|
|
|_, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::InvalidSignature
|
|
|
|
))
|
|
|
|
},
|
|
|
|
)
|
|
|
|
// NOTE: from here on, the tests are stateful, and rely on the valid attestation having
|
|
|
|
// been seen.
|
|
|
|
.import_valid_unaggregate()
|
|
|
|
/*
|
|
|
|
* The following test ensures that:
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* There has been no other valid attestation seen on an attestation subnet that has an
|
|
|
|
* identical attestation.data.target.epoch and participating validator index.
|
|
|
|
*/
|
|
|
|
.inspect_unaggregate_err(
|
|
|
|
"attestation that has already been seen",
|
|
|
|
|_, _, _| {},
|
|
|
|
|tester, err| {
|
|
|
|
assert!(matches!(
|
|
|
|
err,
|
|
|
|
AttnError::PriorAttestationKnown {
|
|
|
|
validator_index,
|
|
|
|
epoch,
|
|
|
|
}
|
|
|
|
if validator_index == tester.attester_validator_index as u64 && epoch == tester.epoch()
|
|
|
|
))
|
|
|
|
},
|
|
|
|
);
|
2020-05-06 11:42:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/// Ensures that an attestation that skips epochs can still be processed.
|
|
|
|
///
|
|
|
|
/// This also checks that we can do a state lookup if we don't get a hit from the shuffling cache.
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
#[tokio::test]
|
|
|
|
async fn attestation_that_skips_epochs() {
|
2020-10-19 05:58:39 +00:00
|
|
|
let harness = get_harness(VALIDATOR_COUNT);
|
2020-05-06 11:42:56 +00:00
|
|
|
|
|
|
|
// Extend the chain out a few epochs so we have some chain depth to play with.
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize * 3 + 1,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::SomeValidators(vec![]),
|
|
|
|
)
|
|
|
|
.await;
|
2020-05-06 11:42:56 +00:00
|
|
|
|
2020-08-26 09:24:55 +00:00
|
|
|
let current_slot = harness.chain.slot().expect("should get slot");
|
|
|
|
let current_epoch = harness.chain.epoch().expect("should get epoch");
|
2020-05-06 11:42:56 +00:00
|
|
|
|
|
|
|
let earlier_slot = (current_epoch - 2).start_slot(MainnetEthSpec::slots_per_epoch());
|
2020-08-26 09:24:55 +00:00
|
|
|
let earlier_block = harness
|
|
|
|
.chain
|
Use the forwards iterator more often (#2376)
## Issue Addressed
NA
## Primary Change
When investigating memory usage, I noticed that retrieving a block from an early slot (e.g., slot 900) would cause a sharp increase in the memory footprint (from 400mb to 800mb+) which seemed to be ever-lasting.
After some investigation, I found that the reverse iteration from the head back to that slot was the likely culprit. To counter this, I've switched the `BeaconChain::block_root_at_slot` to use the forwards iterator, instead of the reverse one.
I also noticed that the networking stack is using `BeaconChain::root_at_slot` to check if a peer is relevant (`check_peer_relevance`). Perhaps the steep, seemingly-random-but-consistent increases in memory usage are caused by the use of this function.
Using the forwards iterator with the HTTP API alleviated the sharp increases in memory usage. It also made the response much faster (before it felt like to took 1-2s, now it feels instant).
## Additional Changes
In the process I also noticed that we have two functions for getting block roots:
- `BeaconChain::block_root_at_slot`: returns `None` for a skip slot.
- `BeaconChain::root_at_slot`: returns the previous root for a skip slot.
I unified these two functions into `block_root_at_slot` and added the `WhenSlotSkipped` enum. Now, the caller must be explicit about the skip-slot behaviour when requesting a root.
Additionally, I replaced `vec![]` with `Vec::with_capacity` in `store::chunked_vector::range_query`. I stumbled across this whilst debugging and made this modification to see what effect it would have (not much). It seems like a decent change to keep around, but I'm not concerned either way.
Also, `BeaconChain::get_ancestor_block_root` is unused, so I got rid of it :wastebasket:.
## Additional Info
I haven't also done the same for state roots here. Whilst it's possible and a good idea, it's more work since the fwds iterators are presently block-roots-specific.
Whilst there's a few places a reverse iteration of state roots could be triggered (e.g., attestation production, HTTP API), they're no where near as common as the `check_peer_relevance` call. As such, I think we should get this PR merged first, then come back for the state root iters. I made an issue here https://github.com/sigp/lighthouse/issues/2377.
2021-05-31 04:18:20 +00:00
|
|
|
.block_at_slot(earlier_slot, WhenSlotSkipped::Prev)
|
2020-05-06 11:42:56 +00:00
|
|
|
.expect("should not error getting block at slot")
|
|
|
|
.expect("should find block at slot");
|
|
|
|
|
2020-08-26 09:24:55 +00:00
|
|
|
let mut state = harness
|
|
|
|
.chain
|
2020-05-06 11:42:56 +00:00
|
|
|
.get_state(&earlier_block.state_root(), Some(earlier_slot))
|
|
|
|
.expect("should not error getting state")
|
|
|
|
.expect("should find state");
|
|
|
|
|
2021-07-09 06:15:32 +00:00
|
|
|
while state.slot() < current_slot {
|
2020-05-06 11:42:56 +00:00
|
|
|
per_slot_processing(&mut state, None, &harness.spec).expect("should process slot");
|
|
|
|
}
|
|
|
|
|
2021-03-17 05:09:57 +00:00
|
|
|
let state_root = state.update_tree_hash_cache().unwrap();
|
|
|
|
|
2020-06-18 09:11:03 +00:00
|
|
|
let (attestation, subnet_id) = harness
|
2020-05-06 11:42:56 +00:00
|
|
|
.get_unaggregated_attestations(
|
|
|
|
&AttestationStrategy::AllValidators,
|
|
|
|
&state,
|
2021-03-17 05:09:57 +00:00
|
|
|
state_root,
|
2020-05-06 11:42:56 +00:00
|
|
|
earlier_block.canonical_root(),
|
|
|
|
current_slot,
|
|
|
|
)
|
|
|
|
.first()
|
|
|
|
.expect("should have at least one committee")
|
|
|
|
.first()
|
|
|
|
.cloned()
|
|
|
|
.expect("should have at least one attestation in committee");
|
|
|
|
|
|
|
|
let block_root = attestation.data.beacon_block_root;
|
|
|
|
let block_slot = harness
|
|
|
|
.chain
|
|
|
|
.store
|
Separate execution payloads in the DB (#3157)
## Proposed Changes
Reduce post-merge disk usage by not storing finalized execution payloads in Lighthouse's database.
:warning: **This is achieved in a backwards-incompatible way for networks that have already merged** :warning:. Kiln users and shadow fork enjoyers will be unable to downgrade after running the code from this PR. The upgrade migration may take several minutes to run, and can't be aborted after it begins.
The main changes are:
- New column in the database called `ExecPayload`, keyed by beacon block root.
- The `BeaconBlock` column now stores blinded blocks only.
- Lots of places that previously used full blocks now use blinded blocks, e.g. analytics APIs, block replay in the DB, etc.
- On finalization:
- `prune_abanonded_forks` deletes non-canonical payloads whilst deleting non-canonical blocks.
- `migrate_db` deletes finalized canonical payloads whilst deleting finalized states.
- Conversions between blinded and full blocks are implemented in a compositional way, duplicating some work from Sean's PR #3134.
- The execution layer has a new `get_payload_by_block_hash` method that reconstructs a payload using the EE's `eth_getBlockByHash` call.
- I've tested manually that it works on Kiln, using Geth and Nethermind.
- This isn't necessarily the most efficient method, and new engine APIs are being discussed to improve this: https://github.com/ethereum/execution-apis/pull/146.
- We're depending on the `ethers` master branch, due to lots of recent changes. We're also using a workaround for https://github.com/gakonst/ethers-rs/issues/1134.
- Payload reconstruction is used in the HTTP API via `BeaconChain::get_block`, which is now `async`. Due to the `async` fn, the `blocking_json` wrapper has been removed.
- Payload reconstruction is used in network RPC to serve blocks-by-{root,range} responses. Here the `async` adjustment is messier, although I think I've managed to come up with a reasonable compromise: the handlers take the `SendOnDrop` by value so that they can drop it on _task completion_ (after the `fn` returns). Still, this is introducing disk reads onto core executor threads, which may have a negative performance impact (thoughts appreciated).
## Additional Info
- [x] For performance it would be great to remove the cloning of full blocks when converting them to blinded blocks to write to disk. I'm going to experiment with a `put_block` API that takes the block by value, breaks it into a blinded block and a payload, stores the blinded block, and then re-assembles the full block for the caller.
- [x] We should measure the latency of blocks-by-root and blocks-by-range responses.
- [x] We should add integration tests that stress the payload reconstruction (basic tests done, issue for more extensive tests: https://github.com/sigp/lighthouse/issues/3159)
- [x] We should (manually) test the schema v9 migration from several prior versions, particularly as blocks have changed on disk and some migrations rely on being able to load blocks.
Co-authored-by: Paul Hauner <paul@paulhauner.com>
2022-05-12 00:42:17 +00:00
|
|
|
.get_blinded_block(&block_root)
|
2020-05-06 11:42:56 +00:00
|
|
|
.expect("should not error getting block")
|
|
|
|
.expect("should find attestation block")
|
2021-07-09 06:15:32 +00:00
|
|
|
.message()
|
|
|
|
.slot();
|
2020-05-06 11:42:56 +00:00
|
|
|
|
|
|
|
assert!(
|
|
|
|
attestation.data.slot - block_slot > E::slots_per_epoch() * 2,
|
|
|
|
"the attestation must skip more than two epochs"
|
|
|
|
);
|
|
|
|
|
2020-06-03 03:34:01 +00:00
|
|
|
harness
|
|
|
|
.chain
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
.verify_unaggregated_attestation_for_gossip(&attestation, Some(subnet_id))
|
2020-06-03 03:34:01 +00:00
|
|
|
.expect("should gossip verify attestation that skips slots");
|
2020-05-06 11:42:56 +00:00
|
|
|
}
|
2021-08-09 02:43:03 +00:00
|
|
|
|
2023-05-09 10:48:15 +00:00
|
|
|
/// Ensures that an attestation can be processed when a validator receives proposer reward
|
|
|
|
/// in an epoch _and_ is scheduled for a withdrawal. This is a regression test for a scenario where
|
|
|
|
/// inconsistent state lookup could cause withdrawal root mismatch.
|
|
|
|
#[tokio::test]
|
|
|
|
async fn attestation_validator_receive_proposer_reward_and_withdrawals() {
|
2023-05-15 02:10:41 +00:00
|
|
|
let (harness, _) = get_harness_capella_spec(VALIDATOR_COUNT);
|
2023-05-09 10:48:15 +00:00
|
|
|
|
|
|
|
// Advance to a Capella block. Make sure the blocks have attestations.
|
|
|
|
let two_thirds = (VALIDATOR_COUNT / 3) * 2;
|
|
|
|
let attesters = (0..two_thirds).collect();
|
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
// To trigger the bug we need the proposer attestation reward to be signed at a block
|
|
|
|
// that isn't the first in the epoch.
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize + 1,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::SomeValidators(attesters),
|
|
|
|
)
|
|
|
|
.await;
|
|
|
|
|
|
|
|
// Add BLS change for the block proposer at slot 33. This sets up a withdrawal for the block proposer.
|
|
|
|
let proposer_index = harness
|
|
|
|
.chain
|
|
|
|
.block_at_slot(harness.get_current_slot(), WhenSlotSkipped::None)
|
|
|
|
.expect("should not error getting block at slot")
|
|
|
|
.expect("should find block at slot")
|
|
|
|
.message()
|
|
|
|
.proposer_index();
|
|
|
|
harness
|
|
|
|
.add_bls_to_execution_change(proposer_index, Address::from_low_u64_be(proposer_index))
|
|
|
|
.unwrap();
|
|
|
|
|
|
|
|
// Apply two blocks: one to process the BLS change, and another to process the withdrawal.
|
|
|
|
harness.advance_slot();
|
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
2,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::SomeValidators(vec![]),
|
|
|
|
)
|
|
|
|
.await;
|
|
|
|
let earlier_slot = harness.get_current_slot();
|
|
|
|
let earlier_block = harness
|
|
|
|
.chain
|
|
|
|
.block_at_slot(earlier_slot, WhenSlotSkipped::None)
|
|
|
|
.expect("should not error getting block at slot")
|
|
|
|
.expect("should find block at slot");
|
|
|
|
|
|
|
|
// Extend the chain out a few epochs so we have some chain depth to play with.
|
|
|
|
harness.advance_slot();
|
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize * 2,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::SomeValidators(vec![]),
|
|
|
|
)
|
|
|
|
.await;
|
|
|
|
|
|
|
|
let current_slot = harness.get_current_slot();
|
|
|
|
let mut state = harness
|
|
|
|
.chain
|
|
|
|
.get_state(&earlier_block.state_root(), Some(earlier_slot))
|
|
|
|
.expect("should not error getting state")
|
|
|
|
.expect("should find state");
|
|
|
|
|
|
|
|
while state.slot() < current_slot {
|
|
|
|
per_slot_processing(&mut state, None, &harness.spec).expect("should process slot");
|
|
|
|
}
|
|
|
|
|
|
|
|
let state_root = state.update_tree_hash_cache().unwrap();
|
|
|
|
|
|
|
|
// Get an attestation pointed to an old block (where we do not have its shuffling cached).
|
|
|
|
// Verifying the attestation triggers an inconsistent state replay.
|
|
|
|
let remaining_attesters = (two_thirds..VALIDATOR_COUNT).collect();
|
|
|
|
let (attestation, subnet_id) = harness
|
|
|
|
.get_unaggregated_attestations(
|
|
|
|
&AttestationStrategy::SomeValidators(remaining_attesters),
|
|
|
|
&state,
|
|
|
|
state_root,
|
|
|
|
earlier_block.canonical_root(),
|
|
|
|
current_slot,
|
|
|
|
)
|
|
|
|
.first()
|
|
|
|
.expect("should have at least one committee")
|
|
|
|
.first()
|
|
|
|
.cloned()
|
|
|
|
.expect("should have at least one attestation in committee");
|
|
|
|
|
|
|
|
harness
|
|
|
|
.chain
|
|
|
|
.verify_unaggregated_attestation_for_gossip(&attestation, Some(subnet_id))
|
|
|
|
.expect("should gossip verify attestation without checking withdrawals root");
|
|
|
|
}
|
|
|
|
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
#[tokio::test]
|
|
|
|
async fn attestation_to_finalized_block() {
|
2022-01-27 22:58:32 +00:00
|
|
|
let harness = get_harness(VALIDATOR_COUNT);
|
|
|
|
|
|
|
|
// Extend the chain out a few epochs so we have some chain depth to play with.
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize * 4 + 1,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::AllValidators,
|
|
|
|
)
|
|
|
|
.await;
|
2022-01-27 22:58:32 +00:00
|
|
|
|
|
|
|
let finalized_checkpoint = harness
|
|
|
|
.chain
|
|
|
|
.with_head(|head| Ok::<_, BeaconChainError>(head.beacon_state.finalized_checkpoint()))
|
|
|
|
.unwrap();
|
|
|
|
assert!(finalized_checkpoint.epoch > 0);
|
|
|
|
|
|
|
|
let current_slot = harness.get_current_slot();
|
|
|
|
|
|
|
|
let earlier_slot = finalized_checkpoint
|
|
|
|
.epoch
|
|
|
|
.start_slot(MainnetEthSpec::slots_per_epoch())
|
|
|
|
- 1;
|
|
|
|
let earlier_block = harness
|
|
|
|
.chain
|
|
|
|
.block_at_slot(earlier_slot, WhenSlotSkipped::Prev)
|
|
|
|
.expect("should not error getting block at slot")
|
|
|
|
.expect("should find block at slot");
|
|
|
|
let earlier_block_root = earlier_block.canonical_root();
|
|
|
|
assert_ne!(earlier_block_root, finalized_checkpoint.root);
|
|
|
|
|
|
|
|
let mut state = harness
|
|
|
|
.chain
|
|
|
|
.get_state(&earlier_block.state_root(), Some(earlier_slot))
|
|
|
|
.expect("should not error getting state")
|
|
|
|
.expect("should find state");
|
|
|
|
|
|
|
|
while state.slot() < current_slot {
|
|
|
|
per_slot_processing(&mut state, None, &harness.spec).expect("should process slot");
|
|
|
|
}
|
|
|
|
|
|
|
|
let state_root = state.update_tree_hash_cache().unwrap();
|
|
|
|
|
|
|
|
let (attestation, subnet_id) = harness
|
|
|
|
.get_unaggregated_attestations(
|
|
|
|
&AttestationStrategy::AllValidators,
|
|
|
|
&state,
|
|
|
|
state_root,
|
|
|
|
earlier_block_root,
|
|
|
|
current_slot,
|
|
|
|
)
|
|
|
|
.first()
|
|
|
|
.expect("should have at least one committee")
|
|
|
|
.first()
|
|
|
|
.cloned()
|
|
|
|
.expect("should have at least one attestation in committee");
|
|
|
|
assert_eq!(attestation.data.beacon_block_root, earlier_block_root);
|
|
|
|
|
|
|
|
// Attestation should be rejected for attesting to a pre-finalization block.
|
|
|
|
let res = harness
|
|
|
|
.chain
|
|
|
|
.verify_unaggregated_attestation_for_gossip(&attestation, Some(subnet_id));
|
|
|
|
assert!(
|
|
|
|
matches!(res, Err(AttnError:: HeadBlockFinalized { beacon_block_root })
|
|
|
|
if beacon_block_root == earlier_block_root
|
|
|
|
)
|
|
|
|
);
|
|
|
|
|
|
|
|
// Pre-finalization block cache should contain the block root.
|
|
|
|
assert!(harness
|
|
|
|
.chain
|
|
|
|
.pre_finalization_block_cache
|
|
|
|
.contains(earlier_block_root));
|
|
|
|
}
|
|
|
|
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
#[tokio::test]
|
|
|
|
async fn verify_aggregate_for_gossip_doppelganger_detection() {
|
2021-08-09 02:43:03 +00:00
|
|
|
let harness = get_harness(VALIDATOR_COUNT);
|
|
|
|
|
|
|
|
// Extend the chain out a few epochs so we have some chain depth to play with.
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize * 3 - 1,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::AllValidators,
|
|
|
|
)
|
|
|
|
.await;
|
2021-08-09 02:43:03 +00:00
|
|
|
|
|
|
|
// Advance into a slot where there have not been blocks or attestations produced.
|
|
|
|
harness.advance_slot();
|
|
|
|
|
|
|
|
let current_slot = harness.chain.slot().expect("should get slot");
|
|
|
|
|
|
|
|
assert_eq!(
|
|
|
|
current_slot % E::slots_per_epoch(),
|
|
|
|
0,
|
|
|
|
"the test requires a new epoch to avoid already-seen errors"
|
|
|
|
);
|
|
|
|
|
|
|
|
let (valid_attestation, _attester_index, _attester_committee_index, _, _) =
|
|
|
|
get_valid_unaggregated_attestation(&harness.chain);
|
|
|
|
let (valid_aggregate, _, _) =
|
|
|
|
get_valid_aggregated_attestation(&harness.chain, valid_attestation);
|
|
|
|
|
|
|
|
harness
|
|
|
|
.chain
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
.verify_aggregated_attestation_for_gossip(&valid_aggregate)
|
2021-08-09 02:43:03 +00:00
|
|
|
.expect("should verify aggregate attestation");
|
|
|
|
|
|
|
|
let epoch = valid_aggregate.message.aggregate.data.target.epoch;
|
|
|
|
let index = valid_aggregate.message.aggregator_index as usize;
|
|
|
|
assert!(harness.chain.validator_seen_at_epoch(index, epoch));
|
|
|
|
|
|
|
|
// Check the correct beacon cache is populated
|
|
|
|
assert!(!harness
|
|
|
|
.chain
|
|
|
|
.observed_block_attesters
|
|
|
|
.read()
|
|
|
|
.validator_has_been_observed(epoch, index)
|
|
|
|
.expect("should check if block attester was observed"));
|
|
|
|
assert!(!harness
|
|
|
|
.chain
|
|
|
|
.observed_gossip_attesters
|
|
|
|
.read()
|
|
|
|
.validator_has_been_observed(epoch, index)
|
|
|
|
.expect("should check if gossip attester was observed"));
|
|
|
|
assert!(harness
|
|
|
|
.chain
|
|
|
|
.observed_aggregators
|
|
|
|
.read()
|
|
|
|
.validator_has_been_observed(epoch, index)
|
|
|
|
.expect("should check if gossip aggregator was observed"));
|
|
|
|
}
|
|
|
|
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
#[tokio::test]
|
|
|
|
async fn verify_attestation_for_gossip_doppelganger_detection() {
|
2021-08-09 02:43:03 +00:00
|
|
|
let harness = get_harness(VALIDATOR_COUNT);
|
|
|
|
|
|
|
|
// Extend the chain out a few epochs so we have some chain depth to play with.
|
Use async code when interacting with EL (#3244)
## Overview
This rather extensive PR achieves two primary goals:
1. Uses the finalized/justified checkpoints of fork choice (FC), rather than that of the head state.
2. Refactors fork choice, block production and block processing to `async` functions.
Additionally, it achieves:
- Concurrent forkchoice updates to the EL and cache pruning after a new head is selected.
- Concurrent "block packing" (attestations, etc) and execution payload retrieval during block production.
- Concurrent per-block-processing and execution payload verification during block processing.
- The `Arc`-ification of `SignedBeaconBlock` during block processing (it's never mutated, so why not?):
- I had to do this to deal with sending blocks into spawned tasks.
- Previously we were cloning the beacon block at least 2 times during each block processing, these clones are either removed or turned into cheaper `Arc` clones.
- We were also `Box`-ing and un-`Box`-ing beacon blocks as they moved throughout the networking crate. This is not a big deal, but it's nice to avoid shifting things between the stack and heap.
- Avoids cloning *all the blocks* in *every chain segment* during sync.
- It also has the potential to clean up our code where we need to pass an *owned* block around so we can send it back in the case of an error (I didn't do much of this, my PR is already big enough :sweat_smile:)
- The `BeaconChain::HeadSafetyStatus` struct was removed. It was an old relic from prior merge specs.
For motivation for this change, see https://github.com/sigp/lighthouse/pull/3244#issuecomment-1160963273
## Changes to `canonical_head` and `fork_choice`
Previously, the `BeaconChain` had two separate fields:
```
canonical_head: RwLock<Snapshot>,
fork_choice: RwLock<BeaconForkChoice>
```
Now, we have grouped these values under a single struct:
```
canonical_head: CanonicalHead {
cached_head: RwLock<Arc<Snapshot>>,
fork_choice: RwLock<BeaconForkChoice>
}
```
Apart from ergonomics, the only *actual* change here is wrapping the canonical head snapshot in an `Arc`. This means that we no longer need to hold the `cached_head` (`canonical_head`, in old terms) lock when we want to pull some values from it. This was done to avoid deadlock risks by preventing functions from acquiring (and holding) the `cached_head` and `fork_choice` locks simultaneously.
## Breaking Changes
### The `state` (root) field in the `finalized_checkpoint` SSE event
Consider the scenario where epoch `n` is just finalized, but `start_slot(n)` is skipped. There are two state roots we might in the `finalized_checkpoint` SSE event:
1. The state root of the finalized block, which is `get_block(finalized_checkpoint.root).state_root`.
4. The state root at slot of `start_slot(n)`, which would be the state from (1), but "skipped forward" through any skip slots.
Previously, Lighthouse would choose (2). However, we can see that when [Teku generates that event](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/beaconrestapi/src/main/java/tech/pegasys/teku/beaconrestapi/handlers/v1/events/EventSubscriptionManager.java#L171-L182) it uses [`getStateRootFromBlockRoot`](https://github.com/ConsenSys/teku/blob/de2b2801c89ef5abf983d6bf37867c37fc47121f/data/provider/src/main/java/tech/pegasys/teku/api/ChainDataProvider.java#L336-L341) which uses (1).
I have switched Lighthouse from (2) to (1). I think it's a somewhat arbitrary choice between the two, where (1) is easier to compute and is consistent with Teku.
## Notes for Reviewers
I've renamed `BeaconChain::fork_choice` to `BeaconChain::recompute_head`. Doing this helped ensure I broke all previous uses of fork choice and I also find it more descriptive. It describes an action and can't be confused with trying to get a reference to the `ForkChoice` struct.
I've changed the ordering of SSE events when a block is received. It used to be `[block, finalized, head]` and now it's `[block, head, finalized]`. It was easier this way and I don't think we were making any promises about SSE event ordering so it's not "breaking".
I've made it so fork choice will run when it's first constructed. I did this because I wanted to have a cached version of the last call to `get_head`. Ensuring `get_head` has been run *at least once* means that the cached values doesn't need to wrapped in an `Option`. This was fairly simple, it just involved passing a `slot` to the constructor so it knows *when* it's being run. When loading a fork choice from the store and a slot clock isn't handy I've just used the `slot` that was saved in the `fork_choice_store`. That seems like it would be a faithful representation of the slot when we saved it.
I added the `genesis_time: u64` to the `BeaconChain`. It's small, constant and nice to have around.
Since we're using FC for the fin/just checkpoints, we no longer get the `0x00..00` roots at genesis. You can see I had to remove a work-around in `ef-tests` here: b56be3bc2. I can't find any reason why this would be an issue, if anything I think it'll be better since the genesis-alias has caught us out a few times (0x00..00 isn't actually a real root). Edit: I did find a case where the `network` expected the 0x00..00 alias and patched it here: 3f26ac3e2.
You'll notice a lot of changes in tests. Generally, tests should be functionally equivalent. Here are the things creating the most diff-noise in tests:
- Changing tests to be `tokio::async` tests.
- Adding `.await` to fork choice, block processing and block production functions.
- Refactor of the `canonical_head` "API" provided by the `BeaconChain`. E.g., `chain.canonical_head.cached_head()` instead of `chain.canonical_head.read()`.
- Wrapping `SignedBeaconBlock` in an `Arc`.
- In the `beacon_chain/tests/block_verification`, we can't use the `lazy_static` `CHAIN_SEGMENT` variable anymore since it's generated with an async function. We just generate it in each test, not so efficient but hopefully insignificant.
I had to disable `rayon` concurrent tests in the `fork_choice` tests. This is because the use of `rayon` and `block_on` was causing a panic.
Co-authored-by: Mac L <mjladson@pm.me>
2022-07-03 05:36:50 +00:00
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize * 3 - 1,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::AllValidators,
|
|
|
|
)
|
|
|
|
.await;
|
2021-08-09 02:43:03 +00:00
|
|
|
|
|
|
|
// Advance into a slot where there have not been blocks or attestations produced.
|
|
|
|
harness.advance_slot();
|
|
|
|
|
|
|
|
let current_slot = harness.chain.slot().expect("should get slot");
|
|
|
|
|
|
|
|
assert_eq!(
|
|
|
|
current_slot % E::slots_per_epoch(),
|
|
|
|
0,
|
|
|
|
"the test requires a new epoch to avoid already-seen errors"
|
|
|
|
);
|
|
|
|
|
|
|
|
let (valid_attestation, index, _attester_committee_index, _, subnet_id) =
|
|
|
|
get_valid_unaggregated_attestation(&harness.chain);
|
|
|
|
|
|
|
|
harness
|
|
|
|
.chain
|
Batch BLS verification for attestations (#2399)
## Issue Addressed
NA
## Proposed Changes
Adds the ability to verify batches of aggregated/unaggregated attestations from the network.
When the `BeaconProcessor` finds there are messages in the aggregated or unaggregated attestation queues, it will first check the length of the queue:
- `== 1` verify the attestation individually.
- `>= 2` take up to 64 of those attestations and verify them in a batch.
Notably, we only perform batch verification if the queue has a backlog. We don't apply any artificial delays to attestations to try and force them into batches.
### Batching Details
To assist with implementing batches we modify `beacon_chain::attestation_verification` to have two distinct categories for attestations:
- *Indexed* attestations: those which have passed initial validation and were valid enough for us to derive an `IndexedAttestation`.
- *Verified* attestations: those attestations which were indexed *and also* passed signature verification. These are well-formed, interesting messages which were signed by validators.
The batching functions accept `n` attestations and then return `n` attestation verification `Result`s, where those `Result`s can be any combination of `Ok` or `Err`. In other words, we attempt to verify as many attestations as possible and return specific per-attestation results so peer scores can be updated, if required.
When we batch verify attestations, we first try to map all those attestations to *indexed* attestations. If any of those attestations were able to be indexed, we then perform batch BLS verification on those indexed attestations. If the batch verification succeeds, we convert them into *verified* attestations, disabling individual signature checking. If the batch fails, we convert to verified attestations with individual signature checking enabled.
Ultimately, we optimistically try to do a batch verification of attestation signatures and fall-back to individual verification if it fails. This opens an attach vector for "poisoning" the attestations and causing us to waste a batch verification. I argue that peer scoring should do a good-enough job of defending against this and the typical-case gains massively outweigh the worst-case losses.
## Additional Info
Before this PR, attestation verification took the attestations by value (instead of by reference). It turns out that this was unnecessary and, in my opinion, resulted in some undesirable ergonomics (e.g., we had to pass the attestation back in the `Err` variant to avoid clones). In this PR I've modified attestation verification so that it now takes a reference.
I refactored the `beacon_chain/tests/attestation_verification.rs` tests so they use a builder-esque "tester" struct instead of a weird macro. It made it easier for me to test individual/batch with the same set of tests and I think it was a nice tidy-up. Notably, I did this last to try and make sure my new refactors to *actual* production code would pass under the existing test suite.
2021-09-22 08:49:41 +00:00
|
|
|
.verify_unaggregated_attestation_for_gossip(&valid_attestation, Some(subnet_id))
|
2021-08-09 02:43:03 +00:00
|
|
|
.expect("should verify attestation");
|
|
|
|
|
|
|
|
let epoch = valid_attestation.data.target.epoch;
|
|
|
|
assert!(harness.chain.validator_seen_at_epoch(index, epoch));
|
|
|
|
|
|
|
|
// Check the correct beacon cache is populated
|
|
|
|
assert!(!harness
|
|
|
|
.chain
|
|
|
|
.observed_block_attesters
|
|
|
|
.read()
|
|
|
|
.validator_has_been_observed(epoch, index)
|
|
|
|
.expect("should check if block attester was observed"));
|
|
|
|
assert!(harness
|
|
|
|
.chain
|
|
|
|
.observed_gossip_attesters
|
|
|
|
.read()
|
|
|
|
.validator_has_been_observed(epoch, index)
|
|
|
|
.expect("should check if gossip attester was observed"));
|
|
|
|
assert!(!harness
|
|
|
|
.chain
|
|
|
|
.observed_aggregators
|
|
|
|
.read()
|
|
|
|
.validator_has_been_observed(epoch, index)
|
|
|
|
.expect("should check if gossip aggregator was observed"));
|
|
|
|
}
|
2023-05-15 02:10:41 +00:00
|
|
|
|
|
|
|
#[tokio::test]
|
|
|
|
async fn attestation_verification_use_head_state_fork() {
|
|
|
|
let (harness, spec) = get_harness_capella_spec(VALIDATOR_COUNT);
|
|
|
|
|
|
|
|
// Advance to last block of the pre-Capella fork epoch. Capella is at slot 32.
|
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize * CAPELLA_FORK_EPOCH - 1,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::SomeValidators(vec![]),
|
|
|
|
)
|
|
|
|
.await;
|
|
|
|
|
|
|
|
// Assert our head is a block at slot 31 in the pre-Capella fork epoch.
|
|
|
|
let pre_capella_slot = harness.get_current_slot();
|
|
|
|
let pre_capella_block = harness
|
|
|
|
.chain
|
|
|
|
.block_at_slot(pre_capella_slot, WhenSlotSkipped::Prev)
|
|
|
|
.expect("should not error getting block at slot")
|
|
|
|
.expect("should find block at slot");
|
|
|
|
assert_eq!(pre_capella_block.fork_name(&spec).unwrap(), ForkName::Merge);
|
|
|
|
|
|
|
|
// Advance slot clock to Capella fork.
|
|
|
|
harness.advance_slot();
|
|
|
|
let first_capella_slot = harness.get_current_slot();
|
|
|
|
assert_eq!(
|
|
|
|
spec.fork_name_at_slot::<E>(first_capella_slot),
|
|
|
|
ForkName::Capella
|
|
|
|
);
|
|
|
|
|
|
|
|
let (state, state_root) = harness.get_current_state_and_root();
|
|
|
|
|
|
|
|
// Scenario 1: other node signed attestation using the Capella fork epoch.
|
|
|
|
{
|
|
|
|
let attesters = (0..VALIDATOR_COUNT / 2).collect::<Vec<_>>();
|
|
|
|
let capella_fork = spec.fork_for_name(ForkName::Capella).unwrap();
|
|
|
|
let committee_attestations = harness
|
|
|
|
.make_unaggregated_attestations_with_opts(
|
|
|
|
attesters.as_slice(),
|
|
|
|
&state,
|
|
|
|
state_root,
|
|
|
|
pre_capella_block.canonical_root().into(),
|
|
|
|
first_capella_slot,
|
|
|
|
MakeAttestationOptions {
|
|
|
|
fork: capella_fork,
|
|
|
|
limit: None,
|
|
|
|
},
|
|
|
|
)
|
|
|
|
.0
|
|
|
|
.first()
|
|
|
|
.cloned()
|
|
|
|
.expect("should have at least one committee");
|
|
|
|
let attestations_and_subnets = committee_attestations
|
|
|
|
.iter()
|
|
|
|
.map(|(attestation, subnet_id)| (attestation, Some(*subnet_id)));
|
|
|
|
|
|
|
|
assert!(
|
|
|
|
batch_verify_unaggregated_attestations(attestations_and_subnets, &harness.chain).is_ok(),
|
|
|
|
"should accept attestations with `data.slot` >= first capella slot signed using the Capella fork"
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Scenario 2: other node forgot to update their node and signed attestations using bellatrix fork
|
|
|
|
{
|
|
|
|
let attesters = (VALIDATOR_COUNT / 2..VALIDATOR_COUNT).collect::<Vec<_>>();
|
|
|
|
let merge_fork = spec.fork_for_name(ForkName::Merge).unwrap();
|
|
|
|
let committee_attestations = harness
|
|
|
|
.make_unaggregated_attestations_with_opts(
|
|
|
|
attesters.as_slice(),
|
|
|
|
&state,
|
|
|
|
state_root,
|
|
|
|
pre_capella_block.canonical_root().into(),
|
|
|
|
first_capella_slot,
|
|
|
|
MakeAttestationOptions {
|
|
|
|
fork: merge_fork,
|
|
|
|
limit: None,
|
|
|
|
},
|
|
|
|
)
|
|
|
|
.0
|
|
|
|
.first()
|
|
|
|
.cloned()
|
|
|
|
.expect("should have at least one committee");
|
|
|
|
let attestations_and_subnets = committee_attestations
|
|
|
|
.iter()
|
|
|
|
.map(|(attestation, subnet_id)| (attestation, Some(*subnet_id)));
|
|
|
|
|
|
|
|
let results =
|
|
|
|
batch_verify_unaggregated_attestations(attestations_and_subnets, &harness.chain)
|
|
|
|
.expect("should return attestation results");
|
|
|
|
let error = results
|
|
|
|
.into_iter()
|
|
|
|
.collect::<Result<Vec<_>, _>>()
|
|
|
|
.err()
|
|
|
|
.expect("should return an error");
|
|
|
|
assert!(
|
|
|
|
matches!(error, Error::InvalidSignature),
|
|
|
|
"should reject attestations with `data.slot` >= first capella slot signed using the pre-Capella fork"
|
|
|
|
);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
#[tokio::test]
|
|
|
|
async fn aggregated_attestation_verification_use_head_state_fork() {
|
|
|
|
let (harness, spec) = get_harness_capella_spec(VALIDATOR_COUNT);
|
|
|
|
|
|
|
|
// Advance to last block of the pre-Capella fork epoch. Capella is at slot 32.
|
|
|
|
harness
|
|
|
|
.extend_chain(
|
|
|
|
MainnetEthSpec::slots_per_epoch() as usize * CAPELLA_FORK_EPOCH - 1,
|
|
|
|
BlockStrategy::OnCanonicalHead,
|
|
|
|
AttestationStrategy::SomeValidators(vec![]),
|
|
|
|
)
|
|
|
|
.await;
|
|
|
|
|
|
|
|
// Assert our head is a block at slot 31 in the pre-Capella fork epoch.
|
|
|
|
let pre_capella_slot = harness.get_current_slot();
|
|
|
|
let pre_capella_block = harness
|
|
|
|
.chain
|
|
|
|
.block_at_slot(pre_capella_slot, WhenSlotSkipped::Prev)
|
|
|
|
.expect("should not error getting block at slot")
|
|
|
|
.expect("should find block at slot");
|
|
|
|
assert_eq!(pre_capella_block.fork_name(&spec).unwrap(), ForkName::Merge);
|
|
|
|
|
|
|
|
// Advance slot clock to Capella fork.
|
|
|
|
harness.advance_slot();
|
|
|
|
let first_capella_slot = harness.get_current_slot();
|
|
|
|
assert_eq!(
|
|
|
|
spec.fork_name_at_slot::<E>(first_capella_slot),
|
|
|
|
ForkName::Capella
|
|
|
|
);
|
|
|
|
|
|
|
|
let (state, state_root) = harness.get_current_state_and_root();
|
|
|
|
|
|
|
|
// Scenario 1: other node signed attestation using the Capella fork epoch.
|
|
|
|
{
|
|
|
|
let attesters = (0..VALIDATOR_COUNT / 2).collect::<Vec<_>>();
|
|
|
|
let capella_fork = spec.fork_for_name(ForkName::Capella).unwrap();
|
|
|
|
let aggregates = harness
|
|
|
|
.make_attestations_with_opts(
|
|
|
|
attesters.as_slice(),
|
|
|
|
&state,
|
|
|
|
state_root,
|
|
|
|
pre_capella_block.canonical_root().into(),
|
|
|
|
first_capella_slot,
|
|
|
|
MakeAttestationOptions {
|
|
|
|
fork: capella_fork,
|
|
|
|
limit: None,
|
|
|
|
},
|
|
|
|
)
|
|
|
|
.0
|
|
|
|
.into_iter()
|
|
|
|
.map(|(_, aggregate)| aggregate.expect("should have signed aggregate and proof"))
|
|
|
|
.collect::<Vec<_>>();
|
|
|
|
|
|
|
|
assert!(
|
|
|
|
batch_verify_aggregated_attestations(aggregates.iter(), &harness.chain).is_ok(),
|
|
|
|
"should accept aggregates with `data.slot` >= first capella slot signed using the Capella fork"
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Scenario 2: other node forgot to update their node and signed attestations using bellatrix fork
|
|
|
|
{
|
|
|
|
let attesters = (VALIDATOR_COUNT / 2..VALIDATOR_COUNT).collect::<Vec<_>>();
|
|
|
|
let merge_fork = spec.fork_for_name(ForkName::Merge).unwrap();
|
|
|
|
let aggregates = harness
|
|
|
|
.make_attestations_with_opts(
|
|
|
|
attesters.as_slice(),
|
|
|
|
&state,
|
|
|
|
state_root,
|
|
|
|
pre_capella_block.canonical_root().into(),
|
|
|
|
first_capella_slot,
|
|
|
|
MakeAttestationOptions {
|
|
|
|
fork: merge_fork,
|
|
|
|
limit: None,
|
|
|
|
},
|
|
|
|
)
|
|
|
|
.0
|
|
|
|
.into_iter()
|
|
|
|
.map(|(_, aggregate)| aggregate.expect("should have signed aggregate and proof"))
|
|
|
|
.collect::<Vec<_>>();
|
|
|
|
|
|
|
|
let results = batch_verify_aggregated_attestations(aggregates.iter(), &harness.chain)
|
|
|
|
.expect("should return attestation results");
|
|
|
|
let error = results
|
|
|
|
.into_iter()
|
|
|
|
.collect::<Result<Vec<_>, _>>()
|
|
|
|
.err()
|
|
|
|
.expect("should return an error");
|
|
|
|
assert!(
|
|
|
|
matches!(error, Error::InvalidSignature),
|
|
|
|
"should reject aggregates with `data.slot` >= first capella slot signed using the pre-Capella fork"
|
|
|
|
);
|
|
|
|
}
|
|
|
|
}
|