Merge pull request #826 from ethereum/vbuterin-patch-16
Simplify justification and finalization accounting logic
This commit is contained in:
commit
c978bb4a67
|
@ -1869,10 +1869,11 @@ def get_base_reward(state: BeaconState, index: ValidatorIndex) -> Gwei:
|
|||
|
||||
```python
|
||||
def get_inactivity_penalty(state: BeaconState, index: ValidatorIndex, epochs_since_finality: int) -> Gwei:
|
||||
return (
|
||||
get_base_reward(state, index) +
|
||||
get_effective_balance(state, index) * epochs_since_finality // INACTIVITY_PENALTY_QUOTIENT // 2
|
||||
)
|
||||
if epochs_since_finality <= 4:
|
||||
extra_penalty = 0
|
||||
else:
|
||||
extra_penalty = get_effective_balance(state, index) * epochs_since_finality // INACTIVITY_PENALTY_QUOTIENT // 2
|
||||
return get_base_reward(state, index) + extra_penalty
|
||||
```
|
||||
|
||||
Note: When applying penalties in the following balance recalculations implementers should make sure the `uint64` does not underflow.
|
||||
|
@ -1881,23 +1882,10 @@ Note: When applying penalties in the following balance recalculations implemente
|
|||
|
||||
```python
|
||||
def get_justification_and_finalization_deltas(state: BeaconState) -> Tuple[List[Gwei], List[Gwei]]:
|
||||
epochs_since_finality = get_current_epoch(state) + 1 - state.finalized_epoch
|
||||
if epochs_since_finality <= 4:
|
||||
return compute_normal_justification_and_finalization_deltas(state)
|
||||
else:
|
||||
return compute_inactivity_leak_deltas(state)
|
||||
```
|
||||
|
||||
When blocks are finalizing normally...
|
||||
|
||||
```python
|
||||
def compute_normal_justification_and_finalization_deltas(state: BeaconState) -> Tuple[List[Gwei], List[Gwei]]:
|
||||
# deltas[0] for rewards
|
||||
# deltas[1] for penalties
|
||||
deltas = [
|
||||
[0 for index in range(len(state.validator_registry))],
|
||||
[0 for index in range(len(state.validator_registry))]
|
||||
]
|
||||
current_epoch = get_current_epoch(state)
|
||||
epochs_since_finality = current_epoch + 1 - state.finalized_epoch
|
||||
rewards = [0 for index in range(len(state.validator_registry))]
|
||||
penalties = [0 for index in range(len(state.validator_registry))]
|
||||
# Some helper variables
|
||||
boundary_attestations = get_previous_epoch_boundary_attestations(state)
|
||||
boundary_attesting_balance = get_attesting_balance(state, boundary_attestations)
|
||||
|
@ -1905,76 +1893,44 @@ def compute_normal_justification_and_finalization_deltas(state: BeaconState) ->
|
|||
total_attesting_balance = get_attesting_balance(state, state.previous_epoch_attestations)
|
||||
matching_head_attestations = get_previous_epoch_matching_head_attestations(state)
|
||||
matching_head_balance = get_attesting_balance(state, matching_head_attestations)
|
||||
eligible_validators = [
|
||||
index for index, validator in enumerate(state.validator_registry)
|
||||
if (
|
||||
is_active_validator(validator, current_epoch) or
|
||||
(validator.slashed and current_epoch < validator.withdrawable_epoch)
|
||||
)
|
||||
]
|
||||
# Process rewards or penalties for all validators
|
||||
for index in get_active_validator_indices(state.validator_registry, get_previous_epoch(state)):
|
||||
for index in eligible_validators:
|
||||
base_reward = get_base_reward(state, index)
|
||||
# Expected FFG source
|
||||
if index in get_attesting_indices(state, state.previous_epoch_attestations):
|
||||
deltas[0][index] += get_base_reward(state, index) * total_attesting_balance // total_balance
|
||||
rewards[index] += base_reward * total_attesting_balance // total_balance
|
||||
# Inclusion speed bonus
|
||||
deltas[0][index] += (
|
||||
get_base_reward(state, index) * MIN_ATTESTATION_INCLUSION_DELAY //
|
||||
rewards[index] += (
|
||||
base_reward * MIN_ATTESTATION_INCLUSION_DELAY //
|
||||
inclusion_distance(state, index)
|
||||
)
|
||||
else:
|
||||
deltas[1][index] += get_base_reward(state, index)
|
||||
penalties[index] += base_reward
|
||||
# Expected FFG target
|
||||
if index in get_attesting_indices(state, boundary_attestations):
|
||||
deltas[0][index] += get_base_reward(state, index) * boundary_attesting_balance // total_balance
|
||||
rewards[index] += base_reward * boundary_attesting_balance // total_balance
|
||||
else:
|
||||
deltas[1][index] += get_base_reward(state, index)
|
||||
penalties[index] += get_inactivity_penalty(state, index, epochs_since_finality)
|
||||
# Expected head
|
||||
if index in get_attesting_indices(state, matching_head_attestations):
|
||||
deltas[0][index] += get_base_reward(state, index) * matching_head_balance // total_balance
|
||||
rewards[index] += base_reward * matching_head_balance // total_balance
|
||||
else:
|
||||
deltas[1][index] += get_base_reward(state, index)
|
||||
penalties[index] += base_reward
|
||||
# Proposer bonus
|
||||
if index in get_attesting_indices(state, state.previous_epoch_attestations):
|
||||
proposer_index = get_beacon_proposer_index(state, inclusion_slot(state, index))
|
||||
deltas[0][proposer_index] += get_base_reward(state, index) // ATTESTATION_INCLUSION_REWARD_QUOTIENT
|
||||
return deltas
|
||||
```
|
||||
|
||||
When blocks are not finalizing normally...
|
||||
|
||||
```python
|
||||
def compute_inactivity_leak_deltas(state: BeaconState) -> Tuple[List[Gwei], List[Gwei]]:
|
||||
# deltas[0] for rewards
|
||||
# deltas[1] for penalties
|
||||
deltas = [
|
||||
[0 for index in range(len(state.validator_registry))],
|
||||
[0 for index in range(len(state.validator_registry))]
|
||||
]
|
||||
boundary_attestations = get_previous_epoch_boundary_attestations(state)
|
||||
matching_head_attestations = get_previous_epoch_matching_head_attestations(state)
|
||||
active_validator_indices = get_active_validator_indices(state.validator_registry, get_previous_epoch(state))
|
||||
epochs_since_finality = get_current_epoch(state) + 1 - state.finalized_epoch
|
||||
for index in active_validator_indices:
|
||||
if index not in get_attesting_indices(state, state.previous_epoch_attestations):
|
||||
deltas[1][index] += get_inactivity_penalty(state, index, epochs_since_finality)
|
||||
else:
|
||||
# If a validator did attest, apply a small penalty for getting attestations included late
|
||||
deltas[0][index] += (
|
||||
get_base_reward(state, index) * MIN_ATTESTATION_INCLUSION_DELAY //
|
||||
inclusion_distance(state, index)
|
||||
)
|
||||
deltas[1][index] += get_base_reward(state, index)
|
||||
if index not in get_attesting_indices(state, boundary_attestations):
|
||||
deltas[1][index] += get_inactivity_penalty(state, index, epochs_since_finality)
|
||||
if index not in get_attesting_indices(state, matching_head_attestations):
|
||||
deltas[1][index] += get_base_reward(state, index)
|
||||
# Penalize slashed-but-inactive validators as though they were active but offline
|
||||
for index in range(len(state.validator_registry)):
|
||||
eligible = (
|
||||
index not in active_validator_indices and
|
||||
state.validator_registry[index].slashed and
|
||||
get_current_epoch(state) < state.validator_registry[index].withdrawable_epoch
|
||||
)
|
||||
if eligible:
|
||||
deltas[1][index] += (
|
||||
2 * get_inactivity_penalty(state, index, epochs_since_finality) +
|
||||
get_base_reward(state, index)
|
||||
)
|
||||
return deltas
|
||||
rewards[proposer_index] += base_reward // ATTESTATION_INCLUSION_REWARD_QUOTIENT
|
||||
# Take away max rewards if we're not finalizing
|
||||
if epochs_since_finality > 4:
|
||||
penalties[index] += base_reward * 4
|
||||
return [rewards, penalties]
|
||||
```
|
||||
|
||||
##### Crosslinks
|
||||
|
|
|
@ -108,6 +108,8 @@ def test_empty_epoch_transition_not_finalizing(state):
|
|||
|
||||
assert test_state.slot == block.slot
|
||||
assert test_state.finalized_epoch < get_current_epoch(test_state) - 4
|
||||
for index in range(len(test_state.validator_registry)):
|
||||
assert get_balance(test_state, index) < get_balance(state, index)
|
||||
|
||||
return state, [block], test_state
|
||||
|
||||
|
|
Loading…
Reference in New Issue