# Ethereum 2.0 Sharding -- Beacon Chain changes **Notice**: This document is a work-in-progress for researchers and implementers. ## Table of contents - [Introduction](#introduction) - [Custom types](#custom-types) - [Constants](#constants) - [Misc](#misc) - [Domain types](#domain-types) - [Shard Work Status](#shard-work-status) - [Preset](#preset) - [Misc](#misc-1) - [Shard block samples](#shard-block-samples) - [Precomputed size verification points](#precomputed-size-verification-points) - [Gwei values](#gwei-values) - [Configuration](#configuration) - [Updated containers](#updated-containers) - [`AttestationData`](#attestationdata) - [`BeaconBlockBody`](#beaconblockbody) - [`BeaconState`](#beaconstate) - [New containers](#new-containers) - [`DataCommitment`](#datacommitment) - [`ShardBlobBodySummary`](#shardblobbodysummary) - [`ShardBlobHeader`](#shardblobheader) - [`SignedShardBlobHeader`](#signedshardblobheader) - [`PendingShardHeader`](#pendingshardheader) - [`ShardBlobReference`](#shardblobreference) - [`SignedShardBlobReference`](#signedshardblobreference) - [`ShardProposerSlashing`](#shardproposerslashing) - [`ShardWork`](#shardwork) - [Helper functions](#helper-functions) - [Misc](#misc-2) - [`next_power_of_two`](#next_power_of_two) - [`compute_previous_slot`](#compute_previous_slot) - [`compute_updated_gasprice`](#compute_updated_gasprice) - [`compute_committee_source_epoch`](#compute_committee_source_epoch) - [Beacon state accessors](#beacon-state-accessors) - [Updated `get_committee_count_per_slot`](#updated-get_committee_count_per_slot) - [`get_active_shard_count`](#get_active_shard_count) - [`get_shard_committee`](#get_shard_committee) - [`compute_proposer_index`](#compute_proposer_index) - [`get_shard_proposer_index`](#get_shard_proposer_index) - [`get_start_shard`](#get_start_shard) - [`compute_shard_from_committee_index`](#compute_shard_from_committee_index) - [`compute_committee_index_from_shard`](#compute_committee_index_from_shard) - [Block processing](#block-processing) - [Operations](#operations) - [Extended Attestation processing](#extended-attestation-processing) - [`process_shard_header`](#process_shard_header) - [`process_shard_proposer_slashing`](#process_shard_proposer_slashing) - [Epoch transition](#epoch-transition) - [`process_pending_shard_confirmations`](#process_pending_shard_confirmations) - [`charge_confirmed_shard_fees`](#charge_confirmed_shard_fees) - [`reset_pending_shard_work`](#reset_pending_shard_work) - [`process_shard_epoch_increment`](#process_shard_epoch_increment) ## Introduction This document describes the extensions made to the Phase 0 design of The Beacon Chain to support data sharding, based on the ideas [here](https://hackmd.io/G-Iy5jqyT7CXWEz8Ssos8g) and more broadly [here](https://arxiv.org/abs/1809.09044), using KZG10 commitments to commit to data to remove any need for fraud proofs (and hence, safety-critical synchrony assumptions) in the design. ## Custom types We define the following Python custom types for type hinting and readability: | Name | SSZ equivalent | Description | | - | - | - | | `Shard` | `uint64` | A shard number | | `BLSCommitment` | `Bytes48` | A G1 curve point | | `BLSPoint` | `uint256` | A number `x` in the range `0 <= x < MODULUS` | ## Constants The following values are (non-configurable) constants used throughout the specification. ### Misc | Name | Value | Notes | | - | - | - | | `PRIMITIVE_ROOT_OF_UNITY` | `5` | Primitive root of unity of the BLS12_381 (inner) modulus | | `DATA_AVAILABILITY_INVERSE_CODING_RATE` | `2**1` (= 2) | Factor by which samples are extended for data availability encoding | | `POINTS_PER_SAMPLE` | `uint64(2**3)` (= 8) | 31 * 8 = 248 bytes | | `MODULUS` | `0x73eda753299d7d483339d80809a1d80553bda402fffe5bfeffffffff00000001` (curve order of BLS12_381) | ### Domain types | Name | Value | | - | - | | `DOMAIN_SHARD_PROPOSER` | `DomainType('0x80000000')` | | `DOMAIN_SHARD_COMMITTEE` | `DomainType('0x81000000')` | ### Shard Work Status | Name | Value | Notes | | - | - | - | | `SHARD_WORK_UNCONFIRMED` | `0` | Unconfirmed, nullified after confirmation time elapses | | `SHARD_WORK_CONFIRMED` | `1` | Confirmed, reduced to just the commitment | | `SHARD_WORK_PENDING` | `2` | Pending, a list of competing headers | ## Preset ### Misc | Name | Value | Notes | | - | - | - | | `MAX_SHARDS` | `uint64(2**10)` (= 1,024) | Theoretical max shard count (used to determine data structure sizes) | | `GASPRICE_ADJUSTMENT_COEFFICIENT` | `uint64(2**3)` (= 8) | Gasprice may decrease/increase by at most exp(1 / this value) *per epoch* | | `MAX_SHARD_PROPOSER_SLASHINGS` | `2**4` (= 16) | Maximum amount of shard proposer slashing operations per block | | `MAX_SHARD_HEADERS_PER_SHARD` | `4` | | | `SHARD_STATE_MEMORY_SLOTS` | `uint64(2**8)` (= 256) | Number of slots for which shard commitments and confirmation status is directly available in the state | ### Shard block samples | Name | Value | Notes | | - | - | - | | `MAX_SAMPLES_PER_BLOCK` | `uint64(2**11)` (= 2,048) | 248 * 2,048 = 507,904 bytes | | `TARGET_SAMPLES_PER_BLOCK` | `uint64(2**10)` (= 1,024) | 248 * 1,024 = 253,952 bytes | ### Precomputed size verification points | Name | Value | | - | - | | `G1_SETUP` | Type `List[G1]`. The G1-side trusted setup `[G, G*s, G*s**2....]`; note that the first point is the generator. | | `G2_SETUP` | Type `List[G2]`. The G2-side trusted setup `[G, G*s, G*s**2....]` | | `ROOT_OF_UNITY` | `pow(PRIMITIVE_ROOT_OF_UNITY, (MODULUS - 1) // int(MAX_SAMPLES_PER_BLOCK * POINTS_PER_SAMPLE), MODULUS)` | ### Gwei values | Name | Value | Unit | Description | | - | - | - | - | | `MAX_GASPRICE` | `Gwei(2**33)` (= 8,589,934,592) | Gwei | Max gasprice charged for a TARGET-sized shard block | | `MIN_GASPRICE` | `Gwei(2**3)` (= 8) | Gwei | Min gasprice charged for a TARGET-sized shard block | ## Configuration | Name | Value | Notes | | - | - | - | | `INITIAL_ACTIVE_SHARDS` | `uint64(2**6)` (= 64) | Initial shard count | ## Updated containers The following containers have updated definitions to support Sharding. ### `AttestationData` ```python class AttestationData(Container): slot: Slot index: CommitteeIndex # LMD GHOST vote beacon_block_root: Root # FFG vote source: Checkpoint target: Checkpoint # Shard header root shard_header_root: Root # [New in Sharding] ``` ### `BeaconBlockBody` ```python class BeaconBlockBody(merge.BeaconBlockBody): # [extends The Merge block body] shard_proposer_slashings: List[ShardProposerSlashing, MAX_SHARD_PROPOSER_SLASHINGS] shard_headers: List[SignedShardBlobHeader, MAX_SHARDS * MAX_SHARD_HEADERS_PER_SHARD] ``` ### `BeaconState` ```python class BeaconState(merge.BeaconState): # [extends The Merge state] # [Updated fields] (Warning: this changes with Altair, Sharding will rebase to use participation-flags) previous_epoch_attestations: List[PendingAttestation, MAX_ATTESTATIONS * SLOTS_PER_EPOCH] current_epoch_attestations: List[PendingAttestation, MAX_ATTESTATIONS * SLOTS_PER_EPOCH] # [New fields] # A ring buffer of the latest slots, with information per active shard. shard_buffer: Vector[List[ShardWork, MAX_SHARDS], SHARD_STATE_MEMORY_SLOTS] shard_gasprice: uint64 current_epoch_start_shard: Shard ``` ## New containers The shard data itself is network-layer only, and can be found in the [P2P specification](./p2p-interface.md). The beacon chain registers just the commitments of the shard data. ### `DataCommitment` ```python class DataCommitment(Container): # KZG10 commitment to the data point: BLSCommitment # Length of the data in samples length: uint64 ``` ### `ShardBlobBodySummary` ```python class ShardBlobBodySummary(Container): # The actual data commitment commitment: DataCommitment # Proof that the degree < commitment.length degree_proof: BLSCommitment # Hash-tree-root as summary of the data field data_root: Root # Latest block root of the Beacon Chain, before shard_blob.slot beacon_block_root: Root ``` ### `ShardBlobHeader` ```python class ShardBlobHeader(Container): # Slot and shard that this header is intended for slot: Slot shard: Shard # SSZ-summary of ShardBlobBody body_summary: ShardBlobBodySummary # Proposer of the shard-blob proposer_index: ValidatorIndex ``` ### `SignedShardBlobHeader` ```python class SignedShardBlobHeader(Container): message: ShardBlobHeader signature: BLSSignature ``` ### `PendingShardHeader` ```python class PendingShardHeader(Container): # KZG10 commitment to the data commitment: DataCommitment # hash_tree_root of the ShardHeader (stored so that attestations can be checked against it) root: Root # Who voted for the header votes: Bitlist[MAX_VALIDATORS_PER_COMMITTEE] # Sum of effective balances of votes weight: Gwei # When the header was last updated, as reference for weight accuracy update_slot: Slot ``` ### `ShardBlobReference` ```python class ShardBlobReference(Container): # Slot and shard that this reference is intended for slot: Slot shard: Shard # Hash-tree-root of ShardBlobBody body_root: Root # Proposer of the shard-blob proposer_index: ValidatorIndex ``` ### `SignedShardBlobReference` ```python class SignedShardBlobReference(Container): message: ShardBlobReference signature: BLSSignature ``` ### `ShardProposerSlashing` ```python class ShardProposerSlashing(Container): signed_reference_1: SignedShardBlobReference signed_reference_2: SignedShardBlobReference ``` ### `ShardWork` ```python class ShardWork(Container): # Upon confirmation the data is reduced to just the header. status: Union[ # See Shard Work Status enum None, # SHARD_WORK_UNCONFIRMED DataCommitment, # SHARD_WORK_CONFIRMED List[PendingShardHeader, MAX_SHARD_HEADERS_PER_SHARD] # SHARD_WORK_PENDING ] ``` ## Helper functions ### Misc #### `next_power_of_two` ```python def next_power_of_two(x: int) -> int: return 2 ** ((x - 1).bit_length()) ``` #### `compute_previous_slot` ```python def compute_previous_slot(slot: Slot) -> Slot: if slot > 0: return Slot(slot - 1) else: return Slot(0) ``` #### `compute_updated_gasprice` ```python def compute_updated_gasprice(prev_gasprice: Gwei, shard_block_length: uint64, adjustment_quotient: uint64) -> Gwei: if shard_block_length > TARGET_SAMPLES_PER_BLOCK: delta = max(1, prev_gasprice * (shard_block_length - TARGET_SAMPLES_PER_BLOCK) // TARGET_SAMPLES_PER_BLOCK // adjustment_quotient) return min(prev_gasprice + delta, MAX_GASPRICE) else: delta = max(1, prev_gasprice * (TARGET_SAMPLES_PER_BLOCK - shard_block_length) // TARGET_SAMPLES_PER_BLOCK // adjustment_quotient) return max(prev_gasprice, MIN_GASPRICE + delta) - delta ``` #### `compute_committee_source_epoch` ```python def compute_committee_source_epoch(epoch: Epoch, period: uint64) -> Epoch: """ Return the source epoch for computing the committee. """ source_epoch = Epoch(epoch - epoch % period) if source_epoch >= period: source_epoch -= period # `period` epochs lookahead return source_epoch ``` ### Beacon state accessors #### Updated `get_committee_count_per_slot` ```python def get_committee_count_per_slot(state: BeaconState, epoch: Epoch) -> uint64: """ Return the number of committees in each slot for the given ``epoch``. """ return max(uint64(1), min( get_active_shard_count(state, epoch), uint64(len(get_active_validator_indices(state, epoch))) // SLOTS_PER_EPOCH // TARGET_COMMITTEE_SIZE, )) ``` #### `get_active_shard_count` ```python def get_active_shard_count(state: BeaconState, epoch: Epoch) -> uint64: """ Return the number of active shards. Note that this puts an upper bound on the number of committees per slot. """ return INITIAL_ACTIVE_SHARDS ``` #### `get_shard_committee` ```python def get_shard_committee(beacon_state: BeaconState, epoch: Epoch, shard: Shard) -> Sequence[ValidatorIndex]: """ Return the shard committee of the given ``epoch`` of the given ``shard``. """ source_epoch = compute_committee_source_epoch(epoch, SHARD_COMMITTEE_PERIOD) active_validator_indices = get_active_validator_indices(beacon_state, source_epoch) seed = get_seed(beacon_state, source_epoch, DOMAIN_SHARD_COMMITTEE) return compute_committee( indices=active_validator_indices, seed=seed, index=shard, count=get_active_shard_count(beacon_state, epoch), ) ``` #### `compute_proposer_index` Updated version to get a proposer index that will only allow proposers with a certain minimum balance, ensuring that the balance is always sufficient to cover gas costs. ```python def compute_proposer_index(beacon_state: BeaconState, indices: Sequence[ValidatorIndex], seed: Bytes32, min_effective_balance: Gwei = Gwei(0)) -> ValidatorIndex: """ Return from ``indices`` a random index sampled by effective balance. """ assert len(indices) > 0 MAX_RANDOM_BYTE = 2**8 - 1 i = uint64(0) total = uint64(len(indices)) while True: candidate_index = indices[compute_shuffled_index(i % total, total, seed)] random_byte = hash(seed + uint_to_bytes(uint64(i // 32)))[i % 32] effective_balance = beacon_state.validators[candidate_index].effective_balance if effective_balance <= min_effective_balance: continue if effective_balance * MAX_RANDOM_BYTE >= MAX_EFFECTIVE_BALANCE * random_byte: return candidate_index i += 1 ``` #### `get_shard_proposer_index` ```python def get_shard_proposer_index(beacon_state: BeaconState, slot: Slot, shard: Shard) -> ValidatorIndex: """ Return the proposer's index of shard block at ``slot``. """ epoch = compute_epoch_at_slot(slot) committee = get_shard_committee(beacon_state, epoch, shard) seed = hash(get_seed(beacon_state, epoch, DOMAIN_SHARD_PROPOSER) + uint_to_bytes(slot)) # Proposer must have sufficient balance to pay for worst case fee burn EFFECTIVE_BALANCE_MAX_DOWNWARD_DEVIATION = ( EFFECTIVE_BALANCE_INCREMENT - EFFECTIVE_BALANCE_INCREMENT * HYSTERESIS_DOWNWARD_MULTIPLIER // HYSTERESIS_QUOTIENT ) min_effective_balance = ( beacon_state.shard_gasprice * MAX_SAMPLES_PER_BLOCK // TARGET_SAMPLES_PER_BLOCK + EFFECTIVE_BALANCE_MAX_DOWNWARD_DEVIATION ) return compute_proposer_index(beacon_state, committee, seed, min_effective_balance) ``` #### `get_start_shard` ```python def get_start_shard(state: BeaconState, slot: Slot) -> Shard: """ Return the start shard at ``slot``. """ current_epoch_start_slot = compute_start_slot_at_epoch(get_current_epoch(state)) shard = state.current_epoch_start_shard if slot > current_epoch_start_slot: # Current epoch or the next epoch lookahead for _slot in range(current_epoch_start_slot, slot): committee_count = get_committee_count_per_slot(state, compute_epoch_at_slot(Slot(_slot))) active_shard_count = get_active_shard_count(state, compute_epoch_at_slot(Slot(_slot))) shard = (shard + committee_count) % active_shard_count elif slot < current_epoch_start_slot: # Previous epoch for _slot in list(range(slot, current_epoch_start_slot))[::-1]: committee_count = get_committee_count_per_slot(state, compute_epoch_at_slot(Slot(_slot))) active_shard_count = get_active_shard_count(state, compute_epoch_at_slot(Slot(_slot))) # Ensure positive shard = (shard + active_shard_count - committee_count) % active_shard_count return Shard(shard) ``` #### `compute_shard_from_committee_index` ```python def compute_shard_from_committee_index(state: BeaconState, slot: Slot, index: CommitteeIndex) -> Shard: active_shards = get_active_shard_count(state, compute_epoch_at_slot(slot)) assert index < active_shards return Shard((index + get_start_shard(state, slot)) % active_shards) ``` #### `compute_committee_index_from_shard` ```python def compute_committee_index_from_shard(state: BeaconState, slot: Slot, shard: Shard) -> CommitteeIndex: epoch = compute_epoch_at_slot(slot) active_shards = get_active_shard_count(state, epoch) index = CommitteeIndex((active_shards + shard - get_start_shard(state, slot)) % active_shards) assert index < get_committee_count_per_slot(state, epoch) return index ``` ### Block processing ```python def process_block(state: BeaconState, block: BeaconBlock) -> None: process_block_header(state, block) process_randao(state, block.body) process_eth1_data(state, block.body) process_operations(state, block.body) # [Modified in Sharding] # Pre-merge, skip execution payload processing if is_execution_enabled(state, block): process_execution_payload(state, block.body.execution_payload, EXECUTION_ENGINE) # [New in Merge] ``` #### Operations ```python def process_operations(state: BeaconState, body: BeaconBlockBody) -> None: # Verify that outstanding deposits are processed up to the maximum number of deposits assert len(body.deposits) == min(MAX_DEPOSITS, state.eth1_data.deposit_count - state.eth1_deposit_index) def for_ops(operations: Sequence[Any], fn: Callable[[BeaconState, Any], None]) -> None: for operation in operations: fn(state, operation) for_ops(body.proposer_slashings, process_proposer_slashing) for_ops(body.attester_slashings, process_attester_slashing) # New shard proposer slashing processing for_ops(body.shard_proposer_slashings, process_shard_proposer_slashing) # Limit is dynamic based on active shard count assert len(body.shard_headers) <= MAX_SHARD_HEADERS_PER_SHARD * get_active_shard_count(state, get_current_epoch(state)) for_ops(body.shard_headers, process_shard_header) # New attestation processing for_ops(body.attestations, process_attestation) for_ops(body.deposits, process_deposit) for_ops(body.voluntary_exits, process_voluntary_exit) ``` ##### Extended Attestation processing ```python def process_attestation(state: BeaconState, attestation: Attestation) -> None: phase0.process_attestation(state, attestation) update_pending_shard_work(state, attestation) ``` ```python def update_pending_shard_work(state: BeaconState, attestation: Attestation) -> None: attestation_shard = compute_shard_from_committee_index( state, attestation.data.slot, attestation.data.index, ) buffer_index = attestation.data.slot % SHARD_STATE_MEMORY_SLOTS committee_work = state.shard_buffer[buffer_index][attestation_shard] # Skip attestation vote accounting if the header is not pending if committee_work.status.selector != SHARD_WORK_PENDING: # TODO In Altair: set participation bit flag, if attestation matches winning header. return current_headers: Sequence[PendingShardHeader] = committee_work.status.value # Find the corresponding header, abort if it cannot be found header_index = [header.root for header in current_headers].index(attestation.data.shard_header_root) pending_header: PendingShardHeader = current_headers[header_index] full_committee = get_beacon_committee(state, attestation.data.slot, attestation.data.index) # The weight may be outdated if it is not the initial weight, and from a previous epoch if pending_header.weight != 0 and compute_epoch_at_slot(pending_header.update_slot) < get_current_epoch(state): pending_header.weight = sum(state.validators[index].effective_balance for index, bit in zip(full_committee, pending_header.votes) if bit) pending_header.update_slot = state.slot full_committee_balance = Gwei(0) # Update votes bitfield in the state, update weights for i, bit in enumerate(attestation.aggregation_bits): weight = state.validators[full_committee[i]].effective_balance full_committee_balance += weight if bit: if not pending_header.votes[i]: pending_header.weight += weight pending_header.votes[i] = True # Check if the PendingShardHeader is eligible for expedited confirmation, requiring 2/3 of balance attesting if pending_header.weight * 3 >= full_committee_balance * 2: # TODO In Altair: set participation bit flag for voters of this early winning header if pending_header.commitment == DataCommitment(): # The committee voted to not confirm anything state.shard_buffer[buffer_index][attestation_shard].change( selector=SHARD_WORK_UNCONFIRMED, value=None, ) else: state.shard_buffer[buffer_index][attestation_shard].change( selector=SHARD_WORK_CONFIRMED, value=pending_header.commitment, ) ``` ##### `process_shard_header` ```python def process_shard_header(state: BeaconState, signed_header: SignedShardBlobHeader) -> None: header = signed_header.message # Verify the header is not 0, and not from the future. assert Slot(0) < header.slot <= state.slot header_epoch = compute_epoch_at_slot(header.slot) # Verify that the header is within the processing time window assert header_epoch in [get_previous_epoch(state), get_current_epoch(state)] # Verify that the shard is active assert header.shard < get_active_shard_count(state, header_epoch) # Verify that the block root matches, # to ensure the header will only be included in this specific Beacon Chain sub-tree. assert header.body_summary.beacon_block_root == get_block_root_at_slot(state, header.slot - 1) # Check that this data is still pending committee_work = state.shard_buffer[header.slot % SHARD_STATE_MEMORY_SLOTS][header.shard] assert committee_work.status.selector == SHARD_WORK_PENDING # Check that this header is not yet in the pending list current_headers: MutableSequence[PendingShardHeader] = committee_work.status.value header_root = hash_tree_root(header) assert header_root not in [pending_header.root for pending_header in current_headers] # Verify proposer assert header.proposer_index == get_shard_proposer_index(state, header.slot, header.shard) # Verify signature signing_root = compute_signing_root(header, get_domain(state, DOMAIN_SHARD_PROPOSER)) assert bls.Verify(state.validators[header.proposer_index].pubkey, signing_root, signed_header.signature) # Verify the length by verifying the degree. body_summary = header.body_summary if body_summary.commitment.length == 0: assert body_summary.degree_proof == G1_SETUP[0] assert ( bls.Pairing(body_summary.degree_proof, G2_SETUP[0]) == bls.Pairing(body_summary.commitment.point, G2_SETUP[-body_summary.commitment.length]) ) # Initialize the pending header index = compute_committee_index_from_shard(state, header.slot, header.shard) committee_length = len(get_beacon_committee(state, header.slot, index)) initial_votes = Bitlist[MAX_VALIDATORS_PER_COMMITTEE]([0] * committee_length) pending_header = PendingShardHeader( commitment=body_summary.commitment, root=header_root, votes=initial_votes, weight=0, update_slot=state.slot, ) # Include it in the pending list current_headers.append(pending_header) ``` The degree proof works as follows. For a block `B` with length `l` (so `l` values in `[0...l - 1]`, seen as a polynomial `B(X)` which takes these values), the length proof is the commitment to the polynomial `B(X) * X**(MAX_DEGREE + 1 - l)`, where `MAX_DEGREE` is the maximum power of `s` available in the setup, which is `MAX_DEGREE = len(G2_SETUP) - 1`. The goal is to ensure that a proof can only be constructed if `deg(B) < l` (there are not hidden higher-order terms in the polynomial, which would thwart reconstruction). ##### `process_shard_proposer_slashing` ```python def process_shard_proposer_slashing(state: BeaconState, proposer_slashing: ShardProposerSlashing) -> None: reference_1 = proposer_slashing.signed_reference_1.message reference_2 = proposer_slashing.signed_reference_2.message # Verify header slots match assert reference_1.slot == reference_2.slot # Verify header shards match assert reference_1.shard == reference_2.shard # Verify header proposer indices match assert reference_1.proposer_index == reference_2.proposer_index # Verify the headers are different (i.e. different body) assert reference_1 != reference_2 # Verify the proposer is slashable proposer = state.validators[reference_1.proposer_index] assert is_slashable_validator(proposer, get_current_epoch(state)) # Verify signatures for signed_header in (proposer_slashing.signed_reference_1, proposer_slashing.signed_reference_2): domain = get_domain(state, DOMAIN_SHARD_PROPOSER, compute_epoch_at_slot(signed_header.message.slot)) signing_root = compute_signing_root(signed_header.message, domain) assert bls.Verify(proposer.pubkey, signing_root, signed_header.signature) slash_validator(state, reference_1.proposer_index) ``` ### Epoch transition This epoch transition overrides the Merge epoch transition: ```python def process_epoch(state: BeaconState) -> None: # Sharding process_pending_shard_confirmations(state) charge_confirmed_shard_fees(state) reset_pending_shard_work(state) # Phase0 process_justification_and_finalization(state) process_rewards_and_penalties(state) process_registry_updates(state) process_slashings(state) # Final updates process_eth1_data_reset(state) process_effective_balance_updates(state) process_slashings_reset(state) process_randao_mixes_reset(state) process_historical_roots_update(state) process_participation_record_updates(state) process_shard_epoch_increment(state) ``` #### `process_pending_shard_confirmations` ```python def process_pending_shard_confirmations(state: BeaconState) -> None: # Pending header processing applies to the previous epoch. # Skip if `GENESIS_EPOCH` because no prior epoch to process. if get_current_epoch(state) == GENESIS_EPOCH: return previous_epoch = get_previous_epoch(state) previous_epoch_start_slot = compute_start_slot_at_epoch(previous_epoch) # Mark stale headers as unconfirmed for slot in range(previous_epoch_start_slot, previous_epoch_start_slot + SLOTS_PER_EPOCH): buffer_index = slot % SHARD_STATE_MEMORY_SLOTS for shard_index in range(len(state.shard_buffer[buffer_index])): committee_work = state.shard_buffer[buffer_index][shard_index] if committee_work.status.selector == SHARD_WORK_PENDING: winning_header = max(committee_work.status.value, key=lambda header: header.weight) # TODO In Altair: set participation bit flag of voters for winning header if winning_header.commitment == DataCommitment(): committee_work.status.change(selector=SHARD_WORK_UNCONFIRMED, value=None) else: committee_work.status.change(selector=SHARD_WORK_CONFIRMED, value=winning_header.commitment) ``` #### `charge_confirmed_shard_fees` ```python def charge_confirmed_shard_fees(state: BeaconState) -> None: new_gasprice = state.shard_gasprice previous_epoch = get_previous_epoch(state) previous_epoch_start_slot = compute_start_slot_at_epoch(previous_epoch) adjustment_quotient = ( get_active_shard_count(state, previous_epoch) * SLOTS_PER_EPOCH * GASPRICE_ADJUSTMENT_COEFFICIENT ) # Iterate through confirmed shard-headers for slot in range(previous_epoch_start_slot, previous_epoch_start_slot + SLOTS_PER_EPOCH): buffer_index = slot % SHARD_STATE_MEMORY_SLOTS for shard_index in range(len(state.shard_buffer[buffer_index])): committee_work = state.shard_buffer[buffer_index][shard_index] if committee_work.status.selector == SHARD_WORK_CONFIRMED: commitment: DataCommitment = committee_work.status.value # Charge EIP 1559 fee proposer = get_shard_proposer_index(state, slot, Shard(shard_index)) fee = ( (state.shard_gasprice * commitment.length) // TARGET_SAMPLES_PER_BLOCK ) decrease_balance(state, proposer, fee) # Track updated gas price new_gasprice = compute_updated_gasprice( new_gasprice, commitment.length, adjustment_quotient, ) state.shard_gasprice = new_gasprice ``` #### `reset_pending_shard_work` ```python def reset_pending_shard_work(state: BeaconState) -> None: # Add dummy "empty" PendingShardHeader (default vote if no shard header is available) next_epoch = get_current_epoch(state) + 1 next_epoch_start_slot = compute_start_slot_at_epoch(next_epoch) committees_per_slot = get_committee_count_per_slot(state, next_epoch) active_shards = get_active_shard_count(state, next_epoch) for slot in range(next_epoch_start_slot, next_epoch_start_slot + SLOTS_PER_EPOCH): buffer_index = slot % SHARD_STATE_MEMORY_SLOTS # Reset the shard work tracking state.shard_buffer[buffer_index] = [ShardWork() for _ in range(active_shards)] start_shard = get_start_shard(state, slot) for committee_index in range(committees_per_slot): shard = (start_shard + committee_index) % active_shards # a committee is available, initialize a pending shard-header list committee_length = len(get_beacon_committee(state, slot, committee_index)) state.shard_buffer[buffer_index][shard].change( selector=SHARD_WORK_PENDING, value=List[PendingShardHeader, MAX_SHARD_HEADERS_PER_SHARD]( PendingShardHeader( commitment=DataCommitment(), root=Root(), votes=Bitlist[MAX_VALIDATORS_PER_COMMITTEE]([0] * committee_length), weight=0, update_slot=slot, ) ) ) # a shard without committee available defaults to SHARD_WORK_UNCONFIRMED. ``` #### `process_shard_epoch_increment` ```python def process_shard_epoch_increment(state: BeaconState) -> None: # Update current_epoch_start_shard state.current_epoch_start_shard = get_start_shard(state, Slot(state.slot + 1)) ```