Further progress.....

This commit is contained in:
Vitalik Buterin 2018-06-13 01:18:26 -04:00
parent 15c2faf63b
commit ab0fd55511
3 changed files with 295 additions and 105 deletions

View File

@ -6,70 +6,98 @@ hash32 = Binary.fixed_length(32)
import rlp
import bls
import random
from bls import decompress_G1, aggregate_pubs, verify
from bls import decompress_G1, aggregate_pubs, verify, sign
from simpleserialize import deepcopy
privkeys = [int.from_bytes(blake(str(i).encode('utf-8')), 'big') for i in range(30)]
keymap = {bls.privtopub(k): k for k in privkeys}
SHARD_COUNT = 100
DEFAULT_BALANCE = 20000
class AggregateVote(rlp.Serializable):
fields = [
('shard_id', int256),
('checkpoint_hash', hash32),
('signer_bitmask', binary),
('aggregate_sig', int256)
]
class AggregateVote():
fields = {
'shard_id': 'int16',
'checkpoint_hash': 'hash32',
'signer_bitmask': 'bytes',
'aggregate_sig': 'int256'
}
defaults = {
'shard_id': 0,
'checkpoint_hash': b'\x00'*32,
'signer_bitmask': b'',
'aggregate_sig': 0,
}
def __init__(self, shard_id, checkpoint_hash, signer_bitmask, aggregate_sig):
# at the beginning of a method, locals() is a dict of all arguments
fields = {k: v for k, v in locals().items() if k != 'self'}
super(BlockHeader, self).__init__(**fields)
def __init__(self, **kwargs):
for k in self.fields.keys():
assert k in kwargs or k in self.defaults
setattr(self, k, kwargs.get(k, self.defaults.get(k)))
class BeaconBlock(rlp.Serializable):
class BeaconBlock():
fields = [
fields = {
# Hash of the parent block
('parent_hash', hash32),
'parent_hash': 'hash32',
# Number of skips (for the full PoS mechanism)
('skip_count', int256),
'skip_count': 'int64',
# Randao commitment reveal
('randao_reveal', hash32),
'randao_reveal': 'hash32',
# Bitmask of who participated in the block notarization committee
('attestation_bitmask', binary),
'attestation_bitmask': 'bytes',
# Their aggregate sig
('attestation_aggregate_sig', int256),
'attestation_aggregate_sig': 'int256',
# Shard aggregate votes
('shard_aggregate_votes', CountableList(AggregateVote)),
'shard_aggregate_votes': [AggregateVote],
# Reference to main chain block
('main_chain_ref', hash32),
'main_chain_ref': 'hash32',
# Hash of the state
('state_hash', hash32),
'state_hash': 'bytes',
# Block height
('height', int256),
'height': 'int64',
# Signature from signer
('sig', int256)
]
'sig': 'int256'
}
def __init__(self,
parent_hash=b'\x00'*32, skip_count=0, randao_reveal=b'\x00'*32,
attestation_bitmask=b'', attestation_aggregate_sig=0,
shard_aggregate_votes=[], main_chain_ref=b'\x00'*32,
state_hash=b'\x00'*32, height=0, sig=0):
# at the beginning of a method, locals() is a dict of all arguments
fields = {k: v for k, v in locals().items() if k != 'self'}
super(BlockHeader, self).__init__(**fields)
defaults = {
'parent_hash': b'\x00'*32,
'skip_count': 0,
'randao_reveal': b'\x00'*32,
'attestation_bitmask': b'',
'attestation_aggregate_sig': 0,
'shard_aggregate_votes': [],
'main_chain_ref': b'\x00'*32,
'state_hash': b'\x00'*32,
'height': 0,
'sig': 0
}
def __init__(self, **kwargs):
for k in self.fields.keys():
assert k in kwargs or k in self.defaults
setattr(self, k, kwargs.get(k, self.defaults.get(k)))
def sign(self, key):
self.sig = 0
self.sig = sign(serialize(self), key)
def verify(self, pub):
zig = self.sig
self.sig = 0
o = verify(serialize(self), pub, zig)
self.sig = zig
return o
def get_shuffling(seed, validator_count, sample=None):
assert validator_count <= 16777216
rand_max = 16777216 - 16777216 % validator_count
o = list(range(validator_count)); source = seed
i = 0
while i < sample if sample is not None else validator_count:
while i < (sample if sample is not None else validator_count):
source = blake(source)
for pos in range(0, 30, 3):
m = int.from_bytes(source[pos:pos+3], 'big')
remaining = validator_count - i
if remaining == 0:
break
if validator_count < rand_max:
replacement_pos = m % remaining + i
o[i], o[replacement_pos] = o[replacement_pos], o[i]
@ -89,7 +117,7 @@ class ValidatorRecord():
class CheckpointRecord():
fields = {'checkpoint_hash': 'hash32', 'voters': 'int16'}
fields = {'shard_id': 'int16', 'checkpoint_hash': 'hash32', 'voter_bitmask': 'bytes'}
defaults = {}
def __init__(self, **kwargs):
@ -114,34 +142,39 @@ class ActiveState():
setattr(self, k, kwargs.get(k, self.defaults.get(k)))
class CrystallizedState():
fields = {'active_validators': [ValidatorRecord],
'queued_validators': [ValidatorRecord],
'exited_validators': [ValidatorRecord],
'current_shuffling': ['int24'],
'current_epoch': 'int64',
'last_justified_epoch': 'int64',
'last_finalized_epoch': 'int64',
'dynasty': 'int64',
'next_shard': 'int16',
'current_checkpoint': 'hash32',
'checkpoint_last_crosslinked': ['int64'],
'total_deposits': 'int256'}
defaults = {'active_validators': [],
'queued_validators': [],
'exited_validators': [],
'current_shuffling': ['int24'],
'current_epoch': 0,
'last_justified_epoch': 0,
'last_finalized_epoch': 0,
'dynasty': 0,
'next_shard': 0,
'current_checkpoint': b'\x00'*32,
'checkpoint_last_crosslinked': [],
'total_deposits': 0}
def __init__(self, **kwargs):
self.fields = {'active_validators': [ValidatorRecord],
'queued_validators': [ValidatorRecord],
'exited_validators': [ValidatorRecord],
'current_shuffling': ['int24'],
'current_epoch': 'int64',
'last_justified_epoch': 'int64',
'last_finalized_epoch': 'int64',
'dynasty': 'int64',
'current_checkpoint': 'hash32',
'total_deposits': 'int256'}
self.defaults = {'active_validators': [],
'queued_validators': [],
'exited_validators': [],
'current_shuffling': ['int24'],
'current_epoch': 0,
'last_justified_epoch': 0,
'last_finalized_epoch': 0,
'dynasty': 0,
'current_checkpoint': b'\x00'*32,
'total_deposits': 0}
for k in self.fields.keys():
assert k in kwargs or k in self.defaults
setattr(self, k, kwargs.get(k, self.defaults.get(k)))
def get_checkpoint_aggvote_msg(aggvote, crystallized_state):
return aggvote.shard_id.to_bytes(2, 'big') + \
aggvote.checkpoint_hash + \
def get_checkpoint_aggvote_msg(shard_id, checkpoint_hash, crystallized_state):
return shard_id.to_bytes(2, 'big') + \
checkpoint_hash + \
crystallized_state.current_checkpoint + \
crystallized_state.current_epoch.to_bytes(8, 'big') + \
crystallized_state.last_justified_epoch.to_bytes(8, 'big')
@ -159,6 +192,106 @@ def get_shard_attesters(crystallized_state, shard_id):
def compute_state_transition(parent_state, parent_block, block):
crystallized_state, active_state = parent_state
# Possibly initialize a new epoch
if block.number % SHARD_COUNT == 1:
# Process rewards from FFG/crosslink votes
new_validator_records = deepcopy(crystallized_state.active_validators)
# Who voted in the last epoch
ffg_voter_bitmask = bytearray(active_state.ffg_voter_bitmask)
# Total deposit size
total_deposits = crystallized_state.total_deposits
# Number of epochs since last finality
finality_distance = crystallized_state.epoch - crystallized_state.last_finalized_epoch
online_reward = 3 if finality_distance == 2 else 0
offline_penalty = 2 * finality_distance
total_vote_count = 0
total_vote_deposits = 0
for i in range(len(active_validators)):
if ffg_voter_bitmask[i // 8] & (1 << (i % 8)):
total_vote_deposits += new_validator_records[i].balance
new_validator_records[i].balance += online_reward
total_vote_count += 1
else:
new_validator_records[i].balance -= offline_penalty
total_deposits += total_vote_count * online_reward - \
(len(active_validators) - total_vote_count) * online_penalty
# Find the most popular crosslink in each shard
main_crosslink = {}
for c in active_state.checkpoints:
vote_count = 0
mask = bytearray(c.voter_bitmask)
for byte in mask:
for j in range(8):
vote_count += (byte >> j) % 2
if vote_count > main_crosslink.get(c.shard_id, (b'', 0, b''))[1]:
main_crosslink[c.shard_id] = (c.checkpoint_hash, vote_count, mask)
# Adjust crosslinks
new_checkpoint_last_crosslinked = deepcopy(crystallized_state.checkpoint_last_crosslinked)
for shard in range(SHARD_COUNT):
h, votes, mask = main_crosslink.get(shard, (b'', 0))
crosslink_distance = crystallized_state.epoch - crystallized_state.checkpoint_last_crosslinked[shard]
indices = get_shard_attesters(crystallized_state, shard)
online_reward = 3 if crosslink_distance <= 2 else 0
offline_penalty = crosslink_distance * 2
for i, index in enumerate(indices):
if mask[i//8] & (1 << (i % 8)):
new_validator_records[index].balance += online_reward
else:
new_validator_records[index].balance -= offline_penalty
# New checkpoint last crosslinked record
if votes * 3 >= len(indices) * 2:
new_checkpoint_last_crosslinked[shard] = crystallized_state.epoch
# Process other balance deltas
for i in active_state.balance_deltas:
if i % 256 <= 128:
new_validator_records[i >> 8] += i % 256
else:
new_validator_records[i >> 8] += (i % 256) - 256
# Process finality and validator set changes
justify, finalize = False
if total_vote_deposits * 3 >= total_deposits * 2:
justify = True
if crystallized_state.last_justified_epoch == crystallized_state.current_epoch - 1:
finalize = True
if finalize:
new_active_validators = [v for v in crystallized_state.active_validators]
new_exited_validators = [v for v in crystallized_state.exited_validators]
i = 0
while i < len(new_active_validators):
if new_validator_records[i].balance <= DEFAULT_BALANCE // 2:
new_exited_validators.append(new_validator_records.pop(i))
elif new_validator_records.switch_dynasty == crystallized_state.dynasty + 1:
new_exited_validators.append(new_validator_records.pop(i))
else:
i += 1
induct = min(len(crystallized_state.queued_validators), crystallized_state.active_validators // 30 + 1)
for i in range(induct):
new_active_validators.append(crystallized_state.queued_validators[i])
new_queued_validators = crystallized_state.queued_validators[induct:]
else:
new_queued_validators = crystallized_state.queued_validators
new_active_validators = crystallized_state.active_validators
new_exited_validators = crystallized_state.exited_validators
crystallized_state = CrystallizedState(
queued_validators=new_queued_validators,
active_validators=new_active_validators,
exited_validators=new_exited_validators,
current_shuffling=get_shuffling(active_state.randao, len(new_active_validators)),
last_justified_epoch = crystallized_state.current_epoch if justified else crystallized_state.last_justified_epoch,
last_finalized_epoch = crystallized_state.current_epoch-1 if finalized else crystallized_state.last_finalized_epoch,
dynasty = crystallized_state.dynasty + (1 if finalized else 0),
next_shard = 0,
current_epoch = parent_block.hash,
checkpoint_last_crosslinked = new_checkpoint_last_crosslinked,
total_deposits = total_deposits
)
# Reset the active state
active_state = ActiveState(height=active_state.height,
randao=active_state.randao,
ffg_voter_bitmask=bytearray((len(crystallized_state.active_validators) + 7) // 8),
balance_deltas=[],
checkpoints=[],
total_skip_count=active_state.total_skip_count)
# Process the block-by-block stuff
# Verify the attestations of the parent
@ -176,10 +309,10 @@ def compute_state_transition(parent_state, parent_block, block):
assert verify(parent_block.hash, aggregate_pubs(pubs), block.aggregate_sig)
# Verify the attestations of checkpoint hashes
checkpoint_votes = {x.checkpoint_hash: x.votes for x in active_state.checkpoints}
checkpoint_votes = {x.checkpoint_hash + x.shard_id.to_bytes(2, 'big'): x.votes for x in active_state.checkpoints}
new_ffg_bitmask = bytearray(active_state.ffg_voter_bitmask)
for vote in block.shard_aggregate_votes:
attestation = get_checkpoint_aggvote_msg(vote, crystallized_state)
attestation = get_checkpoint_aggvote_msg(vote.shard_id, vote.checkpoint_hash, crystallized_state)
indices = get_shard_attesters(crystallized_state, vote.shard_id)
assert len(vote.signer_bitmask) == len(indices + 7) // 8
pubs = []
@ -192,56 +325,46 @@ def compute_state_transition(parent_state, parent_block, block):
voters += 1
assert verify(attestation, aggregate_pubs(pubs), vote.aggregate_sig)
balance_deltas.append((main_signer << 8) + (voters * 16 // len(indices)))
checkpoint_votes[vote.checkpoint_hash] = checkpoint_votes.get(vote.checkpoint_hash, 0) + voters
votekey = vote.checkpoint_hash + vote.shard_id.to_bytes(2, 'big')
checkpoint_votes[votekey] = checkpoint_votes.get(votekey, 0) + voters
o = ActiveState(height=active_state.height + 1,
randao=(int.from_bytes(active_state.randao, 'big') ^
int.from_bytes(block.randao_reveal)).to_bytes(32, 'big'),
total_skip_count=active_state.total_skip_count + block.skip_count,
checkpoints=[CheckpointRecord(checkpoint_hash=h, votes=checkpoint_votes[h])
checkpoints=[CheckpointRecord(shard_id=int.from_bytes(h[32:], 'big'),
checkpoint_hash=h[:32], votes=checkpoint_votes[h])
for h in sorted(checkpoint_votes.keys())],
ffg_voter_bitmask=new_ffg_bitmask,
balance_deltas=active_state.balance_deltas + balance_deltas)
return crystallized_state, o
def mock_make_child(parent_state, parent_hash, skips, attester_share=0.8, checkpoint_shards=[]):
parent_attestation_hash = parent_hash
validator_count = len(parent_state.active_validators)
attestation_count = min(parent_state.active_validators, 128)
indices = get_shuffling(parent.randao_state, validator_count,
attestation_count + skip_count + 1)
main_signer = indices[-1]
# Randomly pick indices to include
bitfield = [1 if random.random() < attester_share else 0 for i in indices]
# Attestations
sigs = [bls.sign(parent_attestation_hash, keymap[parent_state.active_validators[indices[i]].pubkey])
for i in range(len(indices)) if bitfield[i]]
attestation_aggregate_sig = bls.aggregate_sig(sigs)
attestation_bitmask = bytearray((len(bitfield)-1) // 8 + 1)
for i, b in enumerate(bitfield):
attestation_bitmask[i//8] ^= (128 >> (i % 8)) * b
# Randomly pick indices to include for checkpoints
shard_aggregate_votes = []
for shard, crosslinker_share in checkpoint_shards:
indices = parent_state.shuffling[(validator_count * shard) // 100: (validator_count * (shard + 1)) // 100]
bitfield = [1 if random.random() < crosslinker_share else 0 for i in indices]
bitmask = bytearray((len(bitfield)-1) // 8 + 1)
checkpoint = blake(bytes([shard]))
checkpoint_attestation_hash = checkpoint + \
parent_state.checkpoint_hash + \
parent_state.epoch.to_bytes(32, 'big') + \
parent_state.source_epoch.to_bytes(32, 'big')
sigs = [bls.sign(checkpoint, keymap[parent_state.active_validators[indices[i]].pubkey])
for i in range(len(indices)) if bitfield[i]]
shard_aggregate_votes.append(AggregateVote(shard, checkpoint, bitmask, bls.aggregate_sig(sigs)))
# State calculations
o = BlockHeader(parent.hash, skips, blake(str(random.random()).encode('utf-8')),
attestation_bitmask, attestation_aggregate_sig, shard_aggregate_votes,
b'\x00'*32, state.hash, state.height)
# Main signature
o.sign(keymap[parent_state.active_validators[indices[-1]].pubkey])
return o
def mk_genesis_state(pubkeys):
c = CrystallizedState(
active_validators=[ValidatorRecord(
pubkey=pub,
return_shard=0,
return_address=blake(pub.to_bytes(32, 'big'))[-20:],
randao_commitment=b'\x55'*32,
balance=DEFAULT_BALANCE,
switch_dynasty=9999999999999999999
) for pub in pubkeys],
queued_validators=[],
exited_validators=[],
current_shuffling=get_shuffling(b'\x35'*32, len(pubkeys)),
current_epoch=1,
last_justified_epoch=0,
last_finalized_epoch=0,
dynasty=1,
next_shard=0,
current_checkpoint=blake(b'insert EOS constitution here'),
checkpoint_last_crosslinked=[0] * SHARD_COUNT,
total_deposits=DEFAULT_BALANCE*len(pubkeys))
a = ActiveState(height=0,
randao=b'\x45'*32,
ffg_voter_bitmask=bytearray((len(c.active_validators) + 7) // 8),
balance_deltas=[],
checkpoints=[],
total_skip_count=0)
return c, a

View File

@ -62,3 +62,14 @@ def eq(x, y):
return True
else:
return x == y
def deepcopy(x):
if hasattr(x, 'fields'):
vals = {}
for f, val in x.fields.items():
vals[f] = deepcopy(getattr(val, f))
return x.__class__(**vals)
elif isinstance(x, list):
return [deepcopy(x) for x in list]
else:
return x

View File

@ -0,0 +1,56 @@
from full_pos import blake, mk_genesis_state
import bls
from simpleserialize import serialize, deserialize, eq, deepcopy
privkeys = [int.from_bytes(blake(str(i).encode('utf-8'))[:5], 'big') for i in range(400)]
print('Generated privkeys')
keymap = {}
for i,k in enumerate(privkeys):
keymap[bls.privtopub(k)] = k
if i%50 == 0:
print("Generated %d keys" % i)
def mock_make_child(parent_state, parent, skips, attester_share=0.8, checkpoint_shards=[]):
crystallized_state, active_state = parent_state
parent_attestation_hash = parent_hash
validator_count = len(parent_state.active_validators)
attestation_count = min(parent_state.active_validators, 128)
indices, main_signer = get_attesters_and_signer(crystallized_state, active_state, skips)
main_signer = indices[-1]
# Randomly pick indices to include
bitfield = [1 if random.random() < attester_share else 0 for i in indices]
# Attestations
sigs = [bls.sign(parent_attestation_hash, keymap[parent_state.active_validators[indices[i]].pubkey])
for i in range(len(indices)) if bitfield[i]]
attestation_aggregate_sig = bls.aggregate_sig(sigs)
attestation_bitmask = bytearray((len(bitfield)-1) // 8 + 1)
for i, b in enumerate(bitfield):
attestation_bitmask[i//8] ^= (128 >> (i % 8)) * b
# Randomly pick indices to include for checkpoints
shard_aggregate_votes = []
for shard, crosslinker_share in checkpoint_shards:
indices = parent_state.shuffling[(validator_count * shard) // 100: (validator_count * (shard + 1)) // 100]
bitfield = [1 if random.random() < crosslinker_share else 0 for i in indices]
bitmask = bytearray((len(bitfield)-1) // 8 + 1)
checkpoint = blake(bytes([shard]))
checkpoint_attestation_hash = get_checkpoint_aggvote_msg(shard, checkpoint, crystallized_state)
sigs = [bls.sign(checkpoint_attestation_hash, keymap[parent_state.active_validators[indices[i]].pubkey])
for i in range(len(indices)) if bitfield[i]]
shard_aggregate_votes.append(AggregateVote(shard, checkpoint, bitmask, bls.aggregate_sig(sigs)))
# State calculations
o = BlockHeader(parent.hash, skips, blake(str(random.random()).encode('utf-8')),
attestation_bitmask, attestation_aggregate_sig, shard_aggregate_votes,
b'\x00'*32, b'\x00'*32, state.height)
new_crystallized_state, new_active_state = compute_state_transition(crystallized_state, active_state, o)
if crystallized_state == new_crystallized_state:
o.state_hash = blake(parent.state_hash[:32] + blake(serialize(new_active_state)))
else:
o.state_hash = blake(blake(serialize(new_crystallized_state)) + blake(serialize(new_active_state)))
# Main signature
o.sign(keymap[parent_state.active_validators[indices[-1]].pubkey])
return o
c, a = mk_genesis_state(keymap.keys())
print('Generated genesis state')
print('Crystallized state length:', len(serialize(c)))
print('Active state length:', len(serialize(a)))