2021-02-17 10:33:18 +01:00
|
|
|
|
# standard imports
|
|
|
|
|
import logging
|
|
|
|
|
import math
|
|
|
|
|
|
|
|
|
|
# third-pary imports
|
|
|
|
|
import celery
|
|
|
|
|
import moolb
|
2021-03-29 15:27:53 +02:00
|
|
|
|
from chainlib.chain import ChainSpec
|
|
|
|
|
from chainlib.connection import RPCConnection
|
|
|
|
|
from chainlib.eth.tx import (
|
|
|
|
|
unpack,
|
|
|
|
|
transaction_by_block,
|
|
|
|
|
receipt,
|
|
|
|
|
)
|
|
|
|
|
from chainlib.eth.block import block_by_number
|
|
|
|
|
from chainlib.eth.contract import abi_decode_single
|
|
|
|
|
from chainlib.eth.erc20 import ERC20
|
2021-02-17 10:33:18 +01:00
|
|
|
|
from hexathon import strip_0x
|
2021-03-29 15:27:53 +02:00
|
|
|
|
from cic_eth_registry import CICRegistry
|
|
|
|
|
from cic_eth_registry.erc20 import ERC20Token
|
2021-04-04 14:40:59 +02:00
|
|
|
|
from chainqueue.db.models.otx import Otx
|
|
|
|
|
from chainqueue.db.enum import StatusEnum
|
|
|
|
|
from chainqueue.query import get_tx_cache
|
2021-02-17 10:33:18 +01:00
|
|
|
|
|
|
|
|
|
# local imports
|
|
|
|
|
from cic_eth.queue.time import tx_times
|
2021-03-29 15:27:53 +02:00
|
|
|
|
from cic_eth.task import BaseTask
|
2021-04-04 14:40:59 +02:00
|
|
|
|
from cic_eth.db.models.base import SessionBase
|
2021-02-17 10:33:18 +01:00
|
|
|
|
|
|
|
|
|
celery_app = celery.current_app
|
|
|
|
|
logg = logging.getLogger()
|
|
|
|
|
|
|
|
|
|
MAX_BLOCK_TX = 250
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# TODO: Make this method easier to read
|
2021-03-29 15:27:53 +02:00
|
|
|
|
@celery_app.task(bind=True, base=BaseTask)
|
|
|
|
|
def list_tx_by_bloom(self, bloomspec, address, chain_spec_dict):
|
2021-02-17 10:33:18 +01:00
|
|
|
|
"""Retrieve external transaction data matching the provided filter
|
|
|
|
|
|
|
|
|
|
The bloom filter representation with the following structure (the size of the filter will be inferred from the size of the provided filter data):
|
|
|
|
|
{
|
|
|
|
|
'alg': <str; hashing algorithm, currently only "sha256" is understood>,
|
|
|
|
|
'high': <number; highest block number in matched set>,
|
|
|
|
|
'low': <number; lowest block number in matched set>,
|
|
|
|
|
'filter_rounds': <number; hashing rounds used to generate filter entry>,
|
|
|
|
|
'block_filter': <hex; bloom filter data with block matches>,
|
|
|
|
|
'blocktx_filter': <hex; bloom filter data with block+tx matches>,
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
:param bloomspec: Bloom filter data
|
|
|
|
|
:type bloomspec: dict (see description above)
|
|
|
|
|
:param address: Recipient address to use in matching
|
|
|
|
|
:type address: str, 0x-hex
|
|
|
|
|
:param chain_str: Chain spec string representation
|
|
|
|
|
:type chain_str: str
|
|
|
|
|
:returns: dict of transaction data as dict, keyed by transaction hash
|
|
|
|
|
:rtype: dict of dict
|
|
|
|
|
"""
|
2021-03-29 15:27:53 +02:00
|
|
|
|
chain_spec = ChainSpec.from_dict(chain_spec_dict)
|
|
|
|
|
chain_str = str(chain_spec)
|
|
|
|
|
rpc = RPCConnection.connect(chain_spec, 'default')
|
|
|
|
|
registry = CICRegistry(chain_spec, rpc)
|
|
|
|
|
|
2021-02-17 10:33:18 +01:00
|
|
|
|
block_filter_data = bytes.fromhex(bloomspec['block_filter'])
|
|
|
|
|
tx_filter_data = bytes.fromhex(bloomspec['blocktx_filter'])
|
|
|
|
|
databitlen = len(block_filter_data)*8
|
|
|
|
|
block_filter = moolb.Bloom(databitlen, bloomspec['filter_rounds'], default_data=block_filter_data)
|
|
|
|
|
tx_filter = moolb.Bloom(databitlen, bloomspec['filter_rounds'], default_data=tx_filter_data)
|
|
|
|
|
|
|
|
|
|
txs = {}
|
|
|
|
|
for block_height in range(bloomspec['low'], bloomspec['high']):
|
|
|
|
|
block_height_bytes = block_height.to_bytes(4, 'big')
|
|
|
|
|
if block_filter.check(block_height_bytes):
|
|
|
|
|
logg.debug('filter matched block {}'.format(block_height))
|
2021-03-29 15:27:53 +02:00
|
|
|
|
o = block_by_number(block_height)
|
|
|
|
|
block = rpc.do(o)
|
|
|
|
|
logg.debug('block {}'.format(block))
|
2021-02-17 10:33:18 +01:00
|
|
|
|
|
2021-03-29 15:27:53 +02:00
|
|
|
|
for tx_index in range(0, len(block['transactions'])):
|
2021-02-17 10:33:18 +01:00
|
|
|
|
composite = tx_index + block_height
|
|
|
|
|
tx_index_bytes = composite.to_bytes(4, 'big')
|
|
|
|
|
if tx_filter.check(tx_index_bytes):
|
|
|
|
|
logg.debug('filter matched block {} tx {}'.format(block_height, tx_index))
|
|
|
|
|
|
|
|
|
|
try:
|
2021-03-29 15:27:53 +02:00
|
|
|
|
#tx = c.w3.eth.getTransactionByBlock(block_height, tx_index)
|
|
|
|
|
o = transaction_by_block(block['hash'], tx_index)
|
|
|
|
|
tx = rpc.do(o)
|
|
|
|
|
except Exception as e:
|
|
|
|
|
logg.debug('false positive on block {} tx {} ({})'.format(block_height, tx_index, e))
|
2021-02-17 10:33:18 +01:00
|
|
|
|
continue
|
|
|
|
|
tx_address = None
|
|
|
|
|
tx_token_value = 0
|
|
|
|
|
try:
|
2021-03-29 15:27:53 +02:00
|
|
|
|
transfer_data = ERC20.parse_transfer_request(tx['data'])
|
|
|
|
|
tx_address = transfer_data[0]
|
|
|
|
|
tx_token_value = transfer_data[1]
|
2021-02-17 10:33:18 +01:00
|
|
|
|
except ValueError:
|
|
|
|
|
logg.debug('not a transfer transaction, skipping {}'.format(tx))
|
|
|
|
|
continue
|
|
|
|
|
if address == tx_address:
|
|
|
|
|
status = StatusEnum.SENT
|
|
|
|
|
try:
|
2021-03-29 15:27:53 +02:00
|
|
|
|
o = receipt(tx['hash'])
|
|
|
|
|
rcpt = rpc.do(o)
|
2021-02-17 10:33:18 +01:00
|
|
|
|
if rcpt['status'] == 0:
|
|
|
|
|
pending = StatusEnum.REVERTED
|
|
|
|
|
else:
|
|
|
|
|
pending = StatusEnum.SUCCESS
|
2021-03-29 15:27:53 +02:00
|
|
|
|
except Exception as e:
|
|
|
|
|
logg.error('skipping receipt lookup for {}: {}'.format(tx['hash'], e))
|
2021-02-17 10:33:18 +01:00
|
|
|
|
pass
|
|
|
|
|
|
2021-03-29 15:27:53 +02:00
|
|
|
|
# TODO: pass through registry to validate declarator entry of token
|
|
|
|
|
#token = registry.by_address(tx['to'], sender_address=self.call_address)
|
|
|
|
|
token = ERC20Token(rpc, tx['to'])
|
|
|
|
|
token_symbol = token.symbol
|
|
|
|
|
token_decimals = token.decimals
|
|
|
|
|
times = tx_times(tx['hash'], chain_spec)
|
2021-02-17 10:33:18 +01:00
|
|
|
|
tx_r = {
|
2021-03-29 15:27:53 +02:00
|
|
|
|
'hash': tx['hash'],
|
2021-02-17 10:33:18 +01:00
|
|
|
|
'sender': tx['from'],
|
|
|
|
|
'recipient': tx_address,
|
|
|
|
|
'source_value': tx_token_value,
|
|
|
|
|
'destination_value': tx_token_value,
|
|
|
|
|
'source_token': tx['to'],
|
|
|
|
|
'destination_token': tx['to'],
|
|
|
|
|
'source_token_symbol': token_symbol,
|
|
|
|
|
'destination_token_symbol': token_symbol,
|
|
|
|
|
'source_token_decimals': token_decimals,
|
|
|
|
|
'destination_token_decimals': token_decimals,
|
|
|
|
|
'source_token_chain': chain_str,
|
|
|
|
|
'destination_token_chain': chain_str,
|
|
|
|
|
'nonce': tx['nonce'],
|
|
|
|
|
}
|
|
|
|
|
if times['queue'] != None:
|
|
|
|
|
tx_r['date_created'] = times['queue']
|
|
|
|
|
else:
|
|
|
|
|
tx_r['date_created'] = times['network']
|
2021-03-29 15:27:53 +02:00
|
|
|
|
txs[tx['hash']] = tx_r
|
2021-02-17 10:33:18 +01:00
|
|
|
|
break
|
|
|
|
|
return txs
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# TODO: Surely it must be possible to optimize this
|
|
|
|
|
# TODO: DRY this with callback filter in cic_eth/runnable/manager
|
|
|
|
|
# TODO: Remove redundant fields from end representation (timestamp, tx_hash)
|
|
|
|
|
@celery_app.task()
|
2021-03-29 15:27:53 +02:00
|
|
|
|
def tx_collate(tx_batches, chain_spec_dict, offset, limit, newest_first=True):
|
2021-02-17 10:33:18 +01:00
|
|
|
|
"""Merges transaction data from multiple sources and sorts them in chronological order.
|
|
|
|
|
|
|
|
|
|
:param tx_batches: Transaction data inputs
|
|
|
|
|
:type tx_batches: lists of lists of transaction data
|
|
|
|
|
:param chain_str: Chain spec string representation
|
|
|
|
|
:type chain_str: str
|
|
|
|
|
:param offset: Number of sorted results to skip (not yet implemented)
|
|
|
|
|
:type offset: number
|
|
|
|
|
:param limit: Maximum number of results to return (not yet implemented)
|
|
|
|
|
:type limit: number
|
|
|
|
|
:param newest_first: If True, returns results in reverse chronological order
|
|
|
|
|
:type newest_first: bool
|
|
|
|
|
:returns: Transactions
|
|
|
|
|
:rtype: list
|
|
|
|
|
"""
|
|
|
|
|
txs_by_block = {}
|
2021-03-29 15:27:53 +02:00
|
|
|
|
chain_spec = ChainSpec.from_dict(chain_spec_dict)
|
2021-02-17 10:33:18 +01:00
|
|
|
|
|
2021-03-04 17:47:13 +01:00
|
|
|
|
if isinstance(tx_batches, dict):
|
|
|
|
|
tx_batches = [tx_batches]
|
|
|
|
|
|
2021-04-04 14:40:59 +02:00
|
|
|
|
session = SessionBase.create_session()
|
|
|
|
|
|
2021-02-17 10:33:18 +01:00
|
|
|
|
for b in tx_batches:
|
|
|
|
|
for v in b.values():
|
|
|
|
|
tx = None
|
|
|
|
|
k = None
|
|
|
|
|
try:
|
|
|
|
|
hx = strip_0x(v)
|
2021-04-04 14:40:59 +02:00
|
|
|
|
tx = unpack(bytes.fromhex(hx), chain_spec)
|
|
|
|
|
txc = get_tx_cache(chain_spec, tx['hash'], session)
|
2021-02-17 10:33:18 +01:00
|
|
|
|
txc['timestamp'] = int(txc['date_created'].timestamp())
|
|
|
|
|
txc['hash'] = txc['tx_hash']
|
|
|
|
|
tx = txc
|
|
|
|
|
except TypeError:
|
|
|
|
|
tx = v
|
|
|
|
|
tx['timestamp'] = tx['date_created']
|
|
|
|
|
k = '{}.{}.{}'.format(tx['timestamp'], tx['sender'], tx['nonce'])
|
|
|
|
|
txs_by_block[k] = tx
|
|
|
|
|
|
2021-04-04 14:40:59 +02:00
|
|
|
|
session.close()
|
|
|
|
|
|
2021-02-17 10:33:18 +01:00
|
|
|
|
txs = []
|
|
|
|
|
ks = list(txs_by_block.keys())
|
|
|
|
|
ks.sort()
|
|
|
|
|
if newest_first:
|
|
|
|
|
ks.reverse()
|
|
|
|
|
for k in ks:
|
|
|
|
|
txs.append(txs_by_block[k])
|
2021-04-04 14:40:59 +02:00
|
|
|
|
|
2021-02-17 10:33:18 +01:00
|
|
|
|
return txs
|