diff --git a/specs/core/0_beacon-chain.md b/specs/core/0_beacon-chain.md index 1e7eaccf4..e7794b70c 100644 --- a/specs/core/0_beacon-chain.md +++ b/specs/core/0_beacon-chain.md @@ -45,7 +45,6 @@ - [`BeaconState`](#beaconstate) - [`ValidatorRecord`](#validatorrecord) - [`CrosslinkRecord`](#crosslinkrecord) - - [`ShardCommittee`](#shardcommittee) - [`DepositRootVote`](#depositrootvote) - [`PendingAttestationRecord`](#pendingattestationrecord) - [`ForkData`](#forkdata) @@ -66,7 +65,10 @@ - [`get_active_validator_indices`](#get_active_validator_indices) - [`shuffle`](#shuffle) - [`split`](#split) + - [`get_committees_per_slot`](#get_committees_per_slot) - [`get_shuffling`](#get_shuffling) + - [`get_prev_epoch_committees_per_slot`](#get_prev_epoch_committees_per_slot) + - [`get_cur_epoch_committees_per_slot`](#get_cur_epoch_committees_per_slot) - [`get_shard_committees_at_slot`](#get_shard_committees_at_slot) - [`get_block_root`](#get_block_root) - [`get_beacon_proposer_index`](#get_beacon_proposer_index) @@ -478,7 +480,10 @@ Unless otherwise indicated, code appearing in `this style` is to be interpreted # Randomness and committees 'latest_randao_mixes': ['hash32'], 'latest_vdf_outputs': ['hash32'], - 'shard_committees_at_slots': [[ShardCommittee]], + 'prev_epoch_start_shard': 'uint64', + 'cur_epoch_start_shard': 'uint64', + 'prev_epoch_calculation_slot': 'uint64', + 'cur_epoch_calculation_slot': 'uint64', # Custody challenges 'custody_challenges': [CustodyChallenge], @@ -546,19 +551,6 @@ Unless otherwise indicated, code appearing in `this style` is to be interpreted } ``` -#### `ShardCommittee` - -```python -{ - # Shard number - 'shard': 'uint64', - # Validator indices - 'committee': ['uint24'], - # Total validator count (for custody challenges) - 'total_validator_count': 'uint64', -} -``` - #### `DepositRootVote` ```python @@ -854,15 +846,29 @@ def split(values: List[Any], split_count: int) -> List[Any]: ] ``` +#### `get_committees_per_slot` + +```python +def get_committees_per_slot(active_validator_count: int): + return max( + 1, + min( + SHARD_COUNT // EPOCH_LENGTH, + len(active_validator_indices) // EPOCH_LENGTH // TARGET_COMMITTEE_SIZE, + ) + ) +``` + #### `get_shuffling` ```python def get_shuffling(seed: Hash32, validators: List[ValidatorRecord], - crosslinking_start_shard: int, - slot: int) -> List[List[ShardCommittee]]: + slot: int) -> List[List[int]] """ - Shuffles ``validators`` into shard committees using ``seed`` as entropy. + Shuffles ``validators`` into shard committees using ``seed`` as entropy. Returns a list of + ``EPOCH_LENGTH * committees_per_slot`` committees where each committee is itself a list of + validator indices. """ # Normalizes slot to start of epoch boundary @@ -870,59 +876,67 @@ def get_shuffling(seed: Hash32, active_validator_indices = get_active_validator_indices(validators, slot) - committees_per_slot = max( - 1, - min( - SHARD_COUNT // EPOCH_LENGTH, - len(active_validator_indices) // EPOCH_LENGTH // TARGET_COMMITTEE_SIZE, - ) - ) + committees_per_slot = get_committees_per_slot(len(active_validator_indices)) # Shuffle with seed shuffled_active_validator_indices = shuffle(active_validator_indices, seed) - # Split the shuffled list into epoch_length pieces - validators_per_slot = split(shuffled_active_validator_indices, EPOCH_LENGTH) - - output = [] - for slot, slot_indices in enumerate(validators_per_slot): - # Split the shuffled list into committees_per_slot pieces - shard_indices = split(slot_indices, committees_per_slot) - - shard_id_start = crosslinking_start_shard + slot * committees_per_slot - - shard_committees = [ - ShardCommittee( - shard=(shard_id_start + shard_position) % SHARD_COUNT, - committee=indices, - total_validator_count=len(active_validator_indices), - ) - for shard_position, indices in enumerate(shard_indices) - ] - output.append(shard_committees) - - return output + # Split the shuffled list into epoch_length * committees_per_slot pieces + return split(shuffled_active_validator_indices, committees_per_slot * EPOCH_LENGTH) ``` **Invariant**: if `get_shuffling(seed, validators, shard, slot)` returns some value `x`, it should return the same value `x` for the same `seed` and `shard` and possible future modifications of `validators` forever in phase 0, and until the ~1 year deletion delay in phase 2 and in the future. -Here's a diagram of what is going on: +**Note**: this definition and the next few definitions will make heavy use of repetitive computing. Real life implementations are expected to appropriately use caching/memoization to avoid redoing work. -![](http://vitalik.ca/files/ShuffleAndAssign.png?1) +#### `get_prev_epoch_committees_per_slot` + +```python +def get_prev_epoch_committees_per_slot(state: BeaconState): + prev_active_validators = get_active_validator_indices(validators, state.prev_epoch_calculation_slot) + return get_committees_per_slot(len(prev_active_validators)) +``` + +#### `get_cur_epoch_committees_per_slot` + +```python +def get_cur_epoch_committees_per_slot(state: BeaconState): + cur_active_validators = get_active_validator_indices(validators, state.cur_epoch_calculation_slot) + return get_committees_per_slot(len(cur_active_validators)) +``` #### `get_shard_committees_at_slot` ```python def get_shard_committees_at_slot(state: BeaconState, - slot: int) -> List[ShardCommittee]: + slot: int) -> (List[List[int]], int): """ - Returns the ``ShardCommittee`` for the ``slot``. + Returns (i) the list of committees and (ii) the shard associated with the first committee for the ``slot``. """ - earliest_slot_in_array = state.slot - (state.slot % EPOCH_LENGTH) - EPOCH_LENGTH - assert earliest_slot_in_array <= slot < earliest_slot_in_array + EPOCH_LENGTH * 2 - return state.shard_committees_at_slots[slot - earliest_slot_in_array] + earliest_slot = state.slot - (state.slot % EPOCH_LENGTH) - EPOCH_LENGTH + assert earliest_slot <= slot < earliest_slot + EPOCH_LENGTH * 2 + offset = slot % EPOCH_LENGTH + + + if slot < earliest_slot + EPOCH_LENGTH: + committees_per_slot = get_prev_epoch_committees_per_slot(state) + shuffling = get_shuffling(state.latest_randao_mixes[state.prev_epoch_calculation_slot % LATEST_RANDAO_MIXES_LENGTH], + state.validator_registry, + state.slot - (state.slot % EPOCH_LENGTH) - EPOCH_LENGTH) + start_shard = state.prev_epoch_start_shard + else: + committees_per_slot = get_cur_epoch_committees_per_slot(state) + shuffling = get_shuffling(state.latest_randao_mixes[state.cur_epoch_calculation_slot % LATEST_RANDAO_MIXES_LENGTH], + state.validator_registry, + state.slot - (state.slot % EPOCH_LENGTH)) + start_shard = state.cur_epoch_start_shard + + return shuffling[committees_per_slot * offset: committees_per_slot * (offset + 1)], \ + (start_shard + committees_per_slot * offset) % SHARD_COUNT ``` +**Note**: we plan to replace the shuffling algorithm with a pointwise-evaluable shuffle (see https://github.com/ethereum/eth2.0-specs/issues/323), which will allow calculation of the committees for each slot individually. + #### `get_block_root` ```python @@ -946,8 +960,8 @@ def get_beacon_proposer_index(state: BeaconState, """ Returns the beacon proposer index for the ``slot``. """ - first_committee = get_shard_committees_at_slot(state, slot)[0].committee - return first_committee[slot % len(first_committee)] + committees, shard = get_shard_committees_at_slot(state, slot) + return committees[0][slot % len(committees[0])] ``` #### `merkle_root` @@ -974,13 +988,13 @@ def get_attestation_participants(state: BeaconState, """ # Find the relevant committee - shard_committees = get_shard_committees_at_slot(state, attestation_data.slot) - shard_committee = [x for x in shard_committees if x.shard == attestation_data.shard][0] - assert len(participation_bitfield) == ceil_div8(len(shard_committee.committee)) + shard_committees, start_shard = get_shard_committees_at_slot(state, attestation_data.slot) + shard_committee = shard_committee[(attestation_data.shard - start_shard) % SHARD_COUNT] + assert len(participation_bitfield) == ceil_div8(len(shard_committee)) # Find the participating attesters in the committee participants = [] - for i, validator_index in enumerate(shard_committee.committee): + for i, validator_index in enumerate(shard_committee): participation_bit = (participation_bitfield[i//8] >> (7 - (i % 8))) % 2 if participation_bit == 1: participants.append(validator_index) @@ -1165,7 +1179,10 @@ def get_initial_beacon_state(initial_validator_deposits: List[Deposit], # Randomness and committees latest_randao_mixes=[ZERO_HASH for _ in range(LATEST_RANDAO_MIXES_LENGTH)], latest_vdf_outputs=[ZERO_HASH for _ in range(LATEST_RANDAO_MIXES_LENGTH // EPOCH_LENGTH)], - shard_committees_at_slots=[], + prev_epoch_start_shard=0, + cur_epoch_start_shard=0, + prev_epoch_calculation_slot=GENESIS_SLOT, + cur_epoch_calculation_slot=GENESIS_SLOT, # Custody challenges custody_challenges=[], @@ -1205,10 +1222,6 @@ def get_initial_beacon_state(initial_validator_deposits: List[Deposit], if get_effective_balance(state, validator_index) >= MAX_DEPOSIT * GWEI_PER_ETH: activate_validator(state, validator_index, True) - # Set initial committee shuffling - initial_shuffling = get_shuffling(ZERO_HASH, state.validator_registry, 0, GENESIS_SLOT) - state.shard_committees_at_slots = initial_shuffling + initial_shuffling - return state ``` @@ -1554,14 +1567,14 @@ All [validators](#dfn-validator): **Note**: `previous_epoch_boundary_attesting_balance` balance might be marginally different than `current_epoch_boundary_attesting_balance` during the previous epoch transition. Due to the tight bound on validator churn each epoch and small per-epoch rewards/penalties, the potential balance difference is very low and only marginally affects consensus safety. -For every `shard_committee_at_slot` in `state.shard_committees_at_slots` and for every `shard_committee` in `shard_committee_at_slot`: +For every `slot in range(state.slot - 2 * EPOCH_LENGTH, state.slot)`, let `shard_committee_at_slot, start_shard = get_shard_committees_at_slot(slot)`. For every `index in range(len(shard_committee_at_slot))`, let `shard_committee = shard_committee_at_slot[index]`, `shard = (start_shard + index) % SHARD_COUNT`, and compute: -* Let `shard_block_root` be `state.latest_crosslinks[shard_committee.shard].shard_block_root` -* Let `attesting_validator_indices(shard_committee, shard_block_root)` be the union of the [validator](#dfn-validator) index sets given by `[get_attestation_participants(state, a.data, a.participation_bitfield) for a in current_epoch_attestations + previous_epoch_attestations if a.shard == shard_committee.shard and a.shard_block_root == shard_block_root]`. +* Let `shard_block_root` be `state.latest_crosslinks[shard].shard_block_root` +* Let `attesting_validator_indices(shard_committee, shard_block_root)` be the union of the [validator](#dfn-validator) index sets given by `[get_attestation_participants(state, a.data, a.participation_bitfield) for a in current_epoch_attestations + previous_epoch_attestations if a.shard == shard and a.shard_block_root == shard_block_root]`. * Let `winning_root(shard_committee)` be equal to the value of `shard_block_root` such that `sum([get_effective_balance(state, i) for i in attesting_validator_indices(shard_committee, shard_block_root)])` is maximized (ties broken by favoring lower `shard_block_root` values). * Let `attesting_validators(shard_committee)` be equal to `attesting_validator_indices(shard_committee, winning_root(shard_committee))` for convenience. * Let `total_attesting_balance(shard_committee) = sum([get_effective_balance(state, i) for i in attesting_validators(shard_committee)])`. -* Let `total_balance(shard_committee) = sum([get_effective_balance(state, i) for i in shard_committee.committee])`. +* Let `total_balance(shard_committee) = sum([get_effective_balance(state, i) for i in shard_committee])`. * Let `inclusion_slot(state, index) = a.slot_included` for the attestation `a` where `index` is in `get_attestation_participants(state, a.data, a.participation_bitfield)`. * Let `inclusion_distance(state, index) = a.slot_included - a.data.slot` where `a` is the above attestation. @@ -1587,9 +1600,9 @@ Set `state.finalized_slot = state.previous_justified_slot` if any of the followi ### Crosslinks -For every `shard_committee_at_slot` in `state.shard_committees_at_slots` and for every `shard_committee`in `shard_committee_at_slot`: +For every `slot in range(state.slot - 2 * EPOCH_LENGTH, state.slot)`, let `shard_committee_at_slot, start_shard = get_shard_committees_at_slot(slot)`. For every `index in range(len(shard_committee_at_slot))`, let `shard_committee = shard_committee_at_slot[index]`, `shard = (start_shard + index) % SHARD_COUNT`, and compute: -* Set `state.latest_crosslinks[shard_committee.shard] = CrosslinkRecord(slot=state.slot, shard_block_root=winning_root(shard_committee))` if `3 * total_attesting_balance(shard_committee) >= 2 * total_balance(shard_committee)`. +* Set `state.latest_crosslinks[shard] = CrosslinkRecord(slot=state.slot, shard_block_root=winning_root(shard_committee))` if `3 * total_attesting_balance(shard_committee) >= 2 * total_balance(shard_committee)`. ### Rewards and penalties @@ -1633,7 +1646,7 @@ For each `index` in `previous_epoch_attester_indices`, we determine the proposer #### Crosslinks -For every `shard_committee_at_slot` in `state.shard_committees_at_slots[:EPOCH_LENGTH]` (i.e. the objects corresponding to the epoch before the current one), for each `shard_committee` in `shard_committee_at_slot`, and for each `index` in `shard_committee.committee`, adjust balances as follows: +For every `i in range(state.slot - 2 * EPOCH_LENGTH, state.slot - EPOCH_LENGTH)`, let `shard_committee_at_slot, start_shard = get_shard_committees_at_slot(i)`. For every `j in range(len(shard_committee_at_slot))`, let `shard_committee = shard_committee_at_slot[j]`, `shard = (start_shard + j) % SHARD_COUNT`, and compute: * If `index in attesting_validators(shard_committee)`, `state.validator_balances[index] += base_reward(state, index) * total_attesting_balance(shard_committee) // total_balance(shard_committee))`. * If `index not in attesting_validators(shard_committee)`, `state.validator_balances[index] -= base_reward(state, index)`. @@ -1658,7 +1671,7 @@ def process_ejections(state: BeaconState) -> None: If the following are satisfied: * `state.finalized_slot > state.validator_registry_latest_change_slot` -* `state.latest_crosslinks[shard].slot > state.validator_registry_latest_change_slot` for every shard number `shard` in `state.shard_committees_at_slots` +* `state.latest_crosslinks[shard].slot > state.validator_registry_latest_change_slot` for every shard number `shard` in `[(state.cur_epoch_start_shard + i) % SHARD_COUNT for i in range(get_cur_epoch_committees_per_slot(state) * EPOCH_LENGTH)]` (that is, for every shard in the current committees) update the validator registry and associated fields by running @@ -1706,7 +1719,21 @@ def update_validator_registry(state: BeaconState) -> None: state.validator_registry_latest_change_slot = state.slot ``` -Regardless of whether the above conditions are satisfied, run the following: +and perform the following updates: + +* Set `state.prev_epoch_calculation_slot = state.cur_epoch_calculation_slot` +* Set `state.prev_epoch_start_shard = state.cur_epoch_start_shard` +* Set `state.cur_epoch_calculation_slot = state.slot` +* Set `state.cur_epoch_start_shard = (state.cur_epoch_start_shard + get_cur_epoch_committees_per_slot(state) * EPOCH_LENGTH) % SHARD_COUNT` + +If a validator registry update does _not_ happen do the following: + +* Set `state.prev_epoch_calculation_slot = state.cur_epoch_calculation_slot` +* Set `state.prev_epoch_start_shard = state.cur_epoch_start_shard` +* Let `epochs_since_last_registry_change = (state.slot - state.validator_registry_latest_change_slot) // EPOCH_LENGTH`. +* If `epochs_since_last_registry_change` is an exact power of 2, set `state.cur_epoch_calculation_slot = state.slot`. Note that `state.cur_epoch_start_shard` is left unchanged. + +Regardless of whether or not a validator set change happens, run the following: ```python def process_penalties_and_exits(state: BeaconState) -> None: @@ -1744,18 +1771,6 @@ def process_penalties_and_exits(state: BeaconState) -> None: break ``` -Also perform the following updates: - -* Set `state.shard_committees_at_slots[:EPOCH_LENGTH] = state.shard_committees_at_slots[EPOCH_LENGTH:]`. -* Set `state.shard_committees_at_slots[EPOCH_LENGTH:] = get_shuffling(state.latest_randao_mixes[(state.slot - SEED_LOOKAHEAD) % LATEST_RANDAO_MIXES_LENGTH], state.validator_registry, next_start_shard, state.slot)` where `next_start_shard = (state.shard_committees_at_slots[-1][-1].shard + 1) % SHARD_COUNT`. - -If a validator registry update does _not_ happen do the following: - -* Set `state.shard_committees_at_slots[:EPOCH_LENGTH] = state.shard_committees_at_slots[EPOCH_LENGTH:]`. -* Let `epochs_since_last_registry_change = (state.slot - state.validator_registry_latest_change_slot) // EPOCH_LENGTH`. -* Let `start_shard = state.shard_committees_at_slots[0][0].shard`. -* If `epochs_since_last_registry_change` is an exact power of 2, set `state.shard_committees_at_slots[EPOCH_LENGTH:] = get_shuffling(state.latest_randao_mixes[(state.slot - SEED_LOOKAHEAD) % LATEST_RANDAO_MIXES_LENGTH], state.validator_registry, start_shard, state.slot)`. Note that `start_shard` is not changed from the last epoch. - ### Final updates * Let `e = state.slot // EPOCH_LENGTH`. Set `state.latest_penalized_exit_balances[(e+1) % LATEST_PENALIZED_EXIT_LENGTH] = state.latest_penalized_exit_balances[e % LATEST_PENALIZED_EXIT_LENGTH]`