_prepare_claim_for_sync generators

This commit is contained in:
Jack Robison 2021-06-17 21:19:31 -04:00
parent 7c34e4bb96
commit 65700e790e
No known key found for this signature in database
GPG key ID: DF25C68FE0239BB2
2 changed files with 184 additions and 150 deletions

View file

@ -26,11 +26,11 @@ from lbry.wallet.server.db.claimtrie import StagedClaimtrieItem, StagedClaimtrie
from lbry.wallet.server.db.claimtrie import get_takeover_name_ops, StagedActivation, get_add_effective_amount_ops
from lbry.wallet.server.db.claimtrie import get_remove_name_ops, get_remove_effective_amount_ops
from lbry.wallet.server.db.prefixes import ACTIVATED_SUPPORT_TXO_TYPE, ACTIVATED_CLAIM_TXO_TYPE
from lbry.wallet.server.db.prefixes import PendingActivationKey, PendingActivationValue
from lbry.wallet.server.db.prefixes import PendingActivationKey, PendingActivationValue, Prefixes
from lbry.wallet.server.udp import StatusServer
from lbry.wallet.server.db.revertable import RevertableOp, RevertablePut, RevertableDelete
if typing.TYPE_CHECKING:
from lbry.wallet.server.leveldb import LevelDB
from lbry.wallet.server.db.revertable import RevertableOp
class Prefetcher:
@ -259,22 +259,6 @@ class BlockProcessor:
self.amount_cache = {}
def claim_producer(self):
def get_claim_txo(tx_hash, nout):
raw = self.db.db.get(
DB_PREFIXES.TX_PREFIX.value + tx_hash
)
try:
output: TxOutput = self.coin.transaction(raw).outputs[nout]
script = OutputScript(output.pk_script)
script.parse()
return Claim.from_bytes(script.values['claim'])
except:
self.logger.error(
"tx parsing for ES went boom %s %s", tx_hash[::-1].hex(),
raw.hex()
)
return
if self.db.db_height <= 1:
return
@ -284,135 +268,8 @@ class BlockProcessor:
for claim_hash in self.removed_claims_to_send_es:
yield 'delete', claim_hash.hex()
for claim_hash in to_send_es:
claim = self.db._fs_get_claim_by_hash(claim_hash)
metadata = get_claim_txo(claim.tx_hash, claim.position)
if not metadata:
continue
reposted_claim_hash = None if not metadata.is_repost else metadata.repost.reference.claim_hash[::-1]
reposted_claim = None
reposted_metadata = None
if reposted_claim_hash:
reposted_claim = self.db.get_claim_txo(reposted_claim_hash)
if not reposted_claim:
continue
reposted_metadata = get_claim_txo(
self.db.total_transactions[reposted_claim.tx_num], reposted_claim.position
)
if not reposted_metadata:
continue
reposted_tags = []
reposted_languages = []
reposted_has_source = None
reposted_claim_type = None
if reposted_claim:
reposted_tx_hash = self.db.total_transactions[reposted_claim.tx_num]
raw_reposted_claim_tx = self.db.db.get(
DB_PREFIXES.TX_PREFIX.value + reposted_tx_hash
)
try:
reposted_claim_txo: TxOutput = self.coin.transaction(
raw_reposted_claim_tx
).outputs[reposted_claim.position]
reposted_script = OutputScript(reposted_claim_txo.pk_script)
reposted_script.parse()
except:
self.logger.error(
"repost tx parsing for ES went boom %s %s", reposted_tx_hash[::-1].hex(),
raw_reposted_claim_tx.hex()
)
continue
try:
reposted_metadata = Claim.from_bytes(reposted_script.values['claim'])
except:
self.logger.error(
"reposted claim parsing for ES went boom %s %s", reposted_tx_hash[::-1].hex(),
raw_reposted_claim_tx.hex()
)
continue
if reposted_metadata:
reposted_tags = [] if not reposted_metadata.is_stream else [tag for tag in reposted_metadata.stream.tags]
reposted_languages = [] if not reposted_metadata.is_stream else (
[lang.language or 'none' for lang in reposted_metadata.stream.languages] or ['none']
)
reposted_has_source = False if not reposted_metadata.is_stream else reposted_metadata.stream.has_source
reposted_claim_type = CLAIM_TYPES[reposted_metadata.claim_type]
claim_tags = [] if not metadata.is_stream else [tag for tag in metadata.stream.tags]
claim_languages = [] if not metadata.is_stream else (
[lang.language or 'none' for lang in metadata.stream.languages] or ['none']
)
tags = list(set(claim_tags).union(set(reposted_tags)))
languages = list(set(claim_languages).union(set(reposted_languages)))
canonical_url = f'{claim.name}#{claim.claim_hash.hex()}'
if metadata.is_signed:
channel_name = self._get_pending_claim_name(metadata.signing_channel_hash[::-1])
canonical_url = f'{channel_name}#{metadata.signing_channel_hash[::-1].hex()}/{canonical_url}'
value = {
'claim_hash': claim_hash[::-1],
# 'claim_id': claim_hash.hex(),
'claim_name': claim.name,
'normalized': claim.name,
'tx_id': claim.tx_hash[::-1].hex(),
'tx_num': claim.tx_num,
'tx_nout': claim.position,
'amount': claim.amount,
'timestamp': 0, # TODO: fix
'creation_timestamp': 0, # TODO: fix
'height': claim.height,
'creation_height': claim.creation_height,
'activation_height': claim.activation_height,
'expiration_height': claim.expiration_height,
'effective_amount': claim.effective_amount,
'support_amount': claim.support_amount,
'is_controlling': claim.is_controlling,
'last_take_over_height': claim.last_takeover_height,
'short_url': f'{claim.name}#{claim.claim_hash.hex()}', # TODO: fix
'canonical_url': canonical_url,
'title': None if not metadata.is_stream else metadata.stream.title,
'author': None if not metadata.is_stream else metadata.stream.author,
'description': None if not metadata.is_stream else metadata.stream.description,
'claim_type': CLAIM_TYPES[metadata.claim_type],
'has_source': None if not metadata.is_stream else metadata.stream.has_source,
'stream_type': None if not metadata.is_stream else STREAM_TYPES[guess_stream_type(metadata.stream.source.media_type)],
'media_type': None if not metadata.is_stream else metadata.stream.source.media_type,
'fee_amount': None if not metadata.is_stream or not metadata.stream.has_fee else int(
max(metadata.stream.fee.amount or 0, 0)*1000
),
'fee_currency': None if not metadata.is_stream else metadata.stream.fee.currency,
'reposted': self.db.get_reposted_count(claim_hash),
'reposted_claim_hash': reposted_claim_hash,
'reposted_claim_type': reposted_claim_type,
'reposted_has_source': reposted_has_source,
'channel_hash': metadata.signing_channel_hash,
'public_key_bytes': None if not metadata.is_channel else metadata.channel.public_key_bytes,
'public_key_hash': None if not metadata.is_channel else self.ledger.address_to_hash160(
self.ledger.public_key_to_address(metadata.channel.public_key_bytes)
),
'signature': metadata.signature,
'signature_digest': None, # TODO: fix
'signature_valid': claim.channel_hash is not None, # TODO: fix
'tags': tags,
'languages': languages,
'censor_type': 0, # TODO: fix
'censoring_channel_hash': None, # TODO: fix
# 'trending_group': 0,
# 'trending_mixed': 0,
# 'trending_local': 0,
# 'trending_global': 0,
}
if metadata.is_stream and (metadata.stream.video.duration or metadata.stream.audio.duration):
value['duration'] = metadata.stream.video.duration or metadata.stream.audio.duration
if metadata.is_stream and metadata.stream.release_time:
value['release_time'] = metadata.stream.release_time
if metadata.is_channel:
value['claims_in_channel'] = self.db.get_claims_in_channel_count(claim_hash)
yield 'update', value
for claim in self.db.claims_producer(to_send_es):
yield 'update', claim
async def run_in_thread_with_lock(self, func, *args):
# Run in a thread to prevent blocking. Shielded so that
@ -444,9 +301,9 @@ class BlockProcessor:
for block in blocks:
start = time.perf_counter()
await self.run_in_thread_with_lock(self.advance_block, block)
self.logger.info("advanced to %i in %ds", self.height, time.perf_counter() - start)
self.logger.info("advanced to %i in %0.3fs", self.height, time.perf_counter() - start)
# TODO: we shouldnt wait on the search index updating before advancing to the next block
# await self.db.search_index.claim_consumer(self.claim_producer())
await self.db.search_index.claim_consumer(self.claim_producer())
self.db.search_index.clear_caches()
self.touched_claims_to_send_es.clear()
self.removed_claims_to_send_es.clear()

View file

@ -36,11 +36,14 @@ from lbry.wallet.server.util import formatted_time, pack_be_uint16, unpack_be_ui
from lbry.wallet.server.storage import db_class
from lbry.wallet.server.db.revertable import RevertablePut, RevertableDelete, RevertableOp, delete_prefix
from lbry.wallet.server.db import DB_PREFIXES
from lbry.wallet.server.db.common import ResolveResult
from lbry.wallet.server.db.common import ResolveResult, STREAM_TYPES, CLAIM_TYPES
from lbry.wallet.server.db.prefixes import Prefixes, PendingActivationValue, ClaimTakeoverValue, ClaimToTXOValue
from lbry.wallet.server.db.prefixes import ACTIVATED_CLAIM_TXO_TYPE, ACTIVATED_SUPPORT_TXO_TYPE
from lbry.wallet.server.db.prefixes import PendingActivationKey, ClaimToTXOKey, TXOToClaimValue
from lbry.wallet.server.db.claimtrie import length_encoded_name
from lbry.wallet.transaction import OutputScript
from lbry.schema.claim import Claim, guess_stream_type
from lbry.wallet.ledger import Ledger, RegTestLedger, TestNetLedger
from lbry.wallet.server.db.elasticsearch import SearchIndex
@ -152,6 +155,13 @@ class LevelDB:
self.genesis_bytes = bytes.fromhex(self.coin.GENESIS_HASH)
if env.coin.NET == 'mainnet':
self.ledger = Ledger
elif env.coin.NET == 'testnet':
self.ledger = TestNetLedger
else:
self.ledger = RegTestLedger
def get_claim_from_txo(self, tx_num: int, tx_idx: int) -> Optional[TXOToClaimValue]:
claim_hash_and_name = self.db.get(Prefixes.txo_to_claim.pack_key(tx_num, tx_idx))
if not claim_hash_and_name:
@ -429,6 +439,168 @@ class LevelDB:
txos[claim_hash] = tx_num, nout
return txos
def get_claim_output_script(self, tx_hash, nout):
raw = self.db.get(
DB_PREFIXES.TX_PREFIX.value + tx_hash
)
try:
output = self.coin.transaction(raw).outputs[nout]
script = OutputScript(output.pk_script)
script.parse()
return Claim.from_bytes(script.values['claim'])
except:
self.logger.error(
"tx parsing for ES went boom %s %s", tx_hash[::-1].hex(),
raw.hex()
)
return
def _prepare_claim_for_sync(self, claim_hash: bytes):
claim = self._fs_get_claim_by_hash(claim_hash)
if not claim:
print("wat")
return
metadata = self.get_claim_output_script(claim.tx_hash, claim.position)
if not metadata:
return
reposted_claim_hash = None if not metadata.is_repost else metadata.repost.reference.claim_hash[::-1]
reposted_claim = None
reposted_metadata = None
if reposted_claim_hash:
reposted_claim = self.get_claim_txo(reposted_claim_hash)
if not reposted_claim:
return
reposted_metadata = self.get_claim_output_script(
self.total_transactions[reposted_claim.tx_num], reposted_claim.position
)
if not reposted_metadata:
return
reposted_tags = []
reposted_languages = []
reposted_has_source = None
reposted_claim_type = None
if reposted_claim:
reposted_tx_hash = self.total_transactions[reposted_claim.tx_num]
raw_reposted_claim_tx = self.db.get(
DB_PREFIXES.TX_PREFIX.value + reposted_tx_hash
)
try:
reposted_claim_txo = self.coin.transaction(
raw_reposted_claim_tx
).outputs[reposted_claim.position]
reposted_script = OutputScript(reposted_claim_txo.pk_script)
reposted_script.parse()
except:
self.logger.error(
"repost tx parsing for ES went boom %s %s", reposted_tx_hash[::-1].hex(),
raw_reposted_claim_tx.hex()
)
return
try:
reposted_metadata = Claim.from_bytes(reposted_script.values['claim'])
except:
self.logger.error(
"reposted claim parsing for ES went boom %s %s", reposted_tx_hash[::-1].hex(),
raw_reposted_claim_tx.hex()
)
return
if reposted_metadata:
reposted_tags = [] if not reposted_metadata.is_stream else [tag for tag in reposted_metadata.stream.tags]
reposted_languages = [] if not reposted_metadata.is_stream else (
[lang.language or 'none' for lang in reposted_metadata.stream.languages] or ['none']
)
reposted_has_source = False if not reposted_metadata.is_stream else reposted_metadata.stream.has_source
reposted_claim_type = CLAIM_TYPES[reposted_metadata.claim_type]
claim_tags = [] if not metadata.is_stream else [tag for tag in metadata.stream.tags]
claim_languages = [] if not metadata.is_stream else (
[lang.language or 'none' for lang in metadata.stream.languages] or ['none']
)
tags = list(set(claim_tags).union(set(reposted_tags)))
languages = list(set(claim_languages).union(set(reposted_languages)))
canonical_url = f'{claim.name}#{claim.claim_hash.hex()}'
if metadata.is_signed:
channel = self.get_claim_txo(metadata.signing_channel_hash[::-1])
if channel:
canonical_url = f'{channel.name}#{metadata.signing_channel_hash[::-1].hex()}/{canonical_url}'
value = {
'claim_hash': claim_hash[::-1],
# 'claim_id': claim_hash.hex(),
'claim_name': claim.name,
'normalized': claim.name,
'tx_id': claim.tx_hash[::-1].hex(),
'tx_num': claim.tx_num,
'tx_nout': claim.position,
'amount': claim.amount,
'timestamp': 0, # TODO: fix
'creation_timestamp': 0, # TODO: fix
'height': claim.height,
'creation_height': claim.creation_height,
'activation_height': claim.activation_height,
'expiration_height': claim.expiration_height,
'effective_amount': claim.effective_amount,
'support_amount': claim.support_amount,
'is_controlling': claim.is_controlling,
'last_take_over_height': claim.last_takeover_height,
'short_url': f'{claim.name}#{claim.claim_hash.hex()}', # TODO: fix
'canonical_url': canonical_url,
'title': None if not metadata.is_stream else metadata.stream.title,
'author': None if not metadata.is_stream else metadata.stream.author,
'description': None if not metadata.is_stream else metadata.stream.description,
'claim_type': CLAIM_TYPES[metadata.claim_type],
'has_source': None if not metadata.is_stream else metadata.stream.has_source,
'stream_type': None if not metadata.is_stream else STREAM_TYPES[
guess_stream_type(metadata.stream.source.media_type)],
'media_type': None if not metadata.is_stream else metadata.stream.source.media_type,
'fee_amount': None if not metadata.is_stream or not metadata.stream.has_fee else int(
max(metadata.stream.fee.amount or 0, 0) * 1000
),
'fee_currency': None if not metadata.is_stream else metadata.stream.fee.currency,
'reposted': self.get_reposted_count(claim_hash),
'reposted_claim_hash': reposted_claim_hash,
'reposted_claim_type': reposted_claim_type,
'reposted_has_source': reposted_has_source,
'channel_hash': metadata.signing_channel_hash,
'public_key_bytes': None if not metadata.is_channel else metadata.channel.public_key_bytes,
'public_key_hash': None if not metadata.is_channel else self.ledger.address_to_hash160(
self.ledger.public_key_to_address(metadata.channel.public_key_bytes)
),
'signature': metadata.signature,
'signature_digest': None, # TODO: fix
'signature_valid': claim.signature_valid,
'tags': tags,
'languages': languages,
'censor_type': 0, # TODO: fix
'censoring_channel_hash': None, # TODO: fix
'claims_in_channel': None if not metadata.is_channel else self.get_claims_in_channel_count(claim_hash)
# 'trending_group': 0,
# 'trending_mixed': 0,
# 'trending_local': 0,
# 'trending_global': 0,
}
if metadata.is_stream and (metadata.stream.video.duration or metadata.stream.audio.duration):
value['duration'] = metadata.stream.video.duration or metadata.stream.audio.duration
if metadata.is_stream and metadata.stream.release_time:
value['release_time'] = metadata.stream.release_time
return value
def all_claims_producer(self):
for claim_hash in self.db.iterator(prefix=Prefixes.claim_to_txo.prefix, include_value=False):
claim = self._prepare_claim_for_sync(claim_hash[1:])
if claim:
yield claim
def claims_producer(self, claim_hashes: Set[bytes]):
for claim_hash in claim_hashes:
result = self._prepare_claim_for_sync(claim_hash)
if result:
yield result
def get_activated_at_height(self, height: int) -> DefaultDict[PendingActivationValue, List[PendingActivationKey]]:
activated = defaultdict(list)
for _k, _v in self.db.iterator(prefix=Prefixes.pending_activation.pack_partial_key(height)):
@ -567,6 +739,11 @@ class LevelDB:
if height >= min_height:
break
keys.append(key)
if min_height > 0:
for key in self.db.iterator(start=DB_PREFIXES.undo_claimtrie.value,
stop=DB_PREFIXES.undo_claimtrie.value + util.pack_be_uint64(min_height),
include_value=False):
keys.append(key)
if keys:
with self.db.write_batch() as batch:
for key in keys: