513 lines
24 KiB
Python
513 lines
24 KiB
Python
import os
|
|
import asyncio
|
|
import typing
|
|
import binascii
|
|
import logging
|
|
import random
|
|
from decimal import Decimal
|
|
from lbrynet.error import ResolveError, InvalidStreamDescriptorError, KeyFeeAboveMaxAllowed, InsufficientFundsError, \
|
|
DownloadDataTimeout, DownloadSDTimeout
|
|
from lbrynet.utils import generate_id
|
|
from lbrynet.stream.descriptor import StreamDescriptor
|
|
from lbrynet.stream.downloader import StreamDownloader
|
|
from lbrynet.stream.managed_stream import ManagedStream
|
|
from lbrynet.schema.claim import Claim
|
|
from lbrynet.schema.uri import parse_lbry_uri
|
|
from lbrynet.extras.daemon.storage import lbc_to_dewies
|
|
if typing.TYPE_CHECKING:
|
|
from lbrynet.conf import Config
|
|
from lbrynet.blob.blob_manager import BlobFileManager
|
|
from lbrynet.dht.node import Node
|
|
from lbrynet.extras.daemon.analytics import AnalyticsManager
|
|
from lbrynet.extras.daemon.storage import SQLiteStorage, StoredStreamClaim
|
|
from lbrynet.wallet import LbryWalletManager
|
|
from lbrynet.extras.daemon.exchange_rate_manager import ExchangeRateManager
|
|
|
|
log = logging.getLogger(__name__)
|
|
|
|
filter_fields = [
|
|
'rowid',
|
|
'status',
|
|
'file_name',
|
|
'sd_hash',
|
|
'stream_hash',
|
|
'claim_name',
|
|
'claim_height',
|
|
'claim_id',
|
|
'outpoint',
|
|
'txid',
|
|
'nout',
|
|
'channel_claim_id',
|
|
'channel_name',
|
|
'full_status', # TODO: remove
|
|
'blobs_remaining',
|
|
'blobs_in_stream'
|
|
]
|
|
|
|
comparison_operators = {
|
|
'eq': lambda a, b: a == b,
|
|
'ne': lambda a, b: a != b,
|
|
'g': lambda a, b: a > b,
|
|
'l': lambda a, b: a < b,
|
|
'ge': lambda a, b: a >= b,
|
|
'le': lambda a, b: a <= b,
|
|
}
|
|
|
|
|
|
class StreamManager:
|
|
def __init__(self, loop: asyncio.BaseEventLoop, config: 'Config', blob_manager: 'BlobFileManager',
|
|
wallet: 'LbryWalletManager', storage: 'SQLiteStorage', node: typing.Optional['Node'],
|
|
analytics_manager: typing.Optional['AnalyticsManager'] = None):
|
|
self.loop = loop
|
|
self.config = config
|
|
self.blob_manager = blob_manager
|
|
self.wallet = wallet
|
|
self.storage = storage
|
|
self.node = node
|
|
self.analytics_manager = analytics_manager
|
|
self.streams: typing.Set[ManagedStream] = set()
|
|
self.starting_streams: typing.Dict[str, asyncio.Future] = {}
|
|
self.resume_downloading_task: asyncio.Task = None
|
|
self.re_reflect_task: asyncio.Task = None
|
|
self.update_stream_finished_futs: typing.List[asyncio.Future] = []
|
|
self.running_reflector_uploads: typing.List[asyncio.Task] = []
|
|
|
|
async def _update_content_claim(self, stream: ManagedStream):
|
|
claim_info = await self.storage.get_content_claim(stream.stream_hash)
|
|
stream.set_claim(claim_info, claim_info['value'])
|
|
|
|
async def start_stream(self, stream: ManagedStream) -> bool:
|
|
"""
|
|
Resume or rebuild a partial or completed stream
|
|
"""
|
|
if not stream.running and not stream.output_file_exists:
|
|
if stream.downloader:
|
|
stream.downloader.stop()
|
|
stream.downloader = None
|
|
|
|
# the directory is gone, can happen when the folder that contains a published file is deleted
|
|
# reset the download directory to the default and update the file name
|
|
if not os.path.isdir(stream.download_directory):
|
|
stream.download_directory = self.config.download_dir
|
|
|
|
stream.downloader = self.make_downloader(
|
|
stream.sd_hash, stream.download_directory, stream.descriptor.suggested_file_name
|
|
)
|
|
if stream.status != ManagedStream.STATUS_FINISHED:
|
|
await self.storage.change_file_status(stream.stream_hash, 'running')
|
|
stream.update_status('running')
|
|
stream.start_download(self.node)
|
|
try:
|
|
await asyncio.wait_for(self.loop.create_task(stream.downloader.wrote_bytes_event.wait()),
|
|
self.config.download_timeout)
|
|
except asyncio.TimeoutError:
|
|
await self.stop_stream(stream)
|
|
if stream in self.streams:
|
|
self.streams.remove(stream)
|
|
return False
|
|
file_name = os.path.basename(stream.downloader.output_path)
|
|
output_dir = os.path.dirname(stream.downloader.output_path)
|
|
await self.storage.change_file_download_dir_and_file_name(
|
|
stream.stream_hash, output_dir, file_name
|
|
)
|
|
stream._file_name = file_name
|
|
stream.download_directory = output_dir
|
|
self.wait_for_stream_finished(stream)
|
|
return True
|
|
return True
|
|
|
|
async def stop_stream(self, stream: ManagedStream):
|
|
stream.stop_download()
|
|
if not stream.finished and stream.output_file_exists:
|
|
try:
|
|
os.remove(stream.full_path)
|
|
except OSError as err:
|
|
log.warning("Failed to delete partial download %s from downloads directory: %s", stream.full_path,
|
|
str(err))
|
|
if stream.running:
|
|
stream.update_status(ManagedStream.STATUS_STOPPED)
|
|
await self.storage.change_file_status(stream.stream_hash, ManagedStream.STATUS_STOPPED)
|
|
|
|
def make_downloader(self, sd_hash: str, download_directory: str, file_name: str):
|
|
return StreamDownloader(
|
|
self.loop, self.config, self.blob_manager, sd_hash, download_directory, file_name
|
|
)
|
|
|
|
async def recover_streams(self, file_infos: typing.List[typing.Dict]):
|
|
to_restore = []
|
|
|
|
async def recover_stream(sd_hash: str, stream_hash: str, stream_name: str,
|
|
suggested_file_name: str, key: str) -> typing.Optional[StreamDescriptor]:
|
|
sd_blob = self.blob_manager.get_blob(sd_hash)
|
|
blobs = await self.storage.get_blobs_for_stream(stream_hash)
|
|
descriptor = await StreamDescriptor.recover(
|
|
self.blob_manager.blob_dir, sd_blob, stream_hash, stream_name, suggested_file_name, key, blobs
|
|
)
|
|
if not descriptor:
|
|
return
|
|
to_restore.append((descriptor, sd_blob))
|
|
|
|
await asyncio.gather(*[
|
|
recover_stream(
|
|
file_info['sd_hash'], file_info['stream_hash'], binascii.unhexlify(file_info['stream_name']).decode(),
|
|
binascii.unhexlify(file_info['suggested_file_name']).decode(), file_info['key']
|
|
) for file_info in file_infos
|
|
])
|
|
|
|
if to_restore:
|
|
await self.storage.recover_streams(to_restore, self.config.download_dir)
|
|
log.info("Recovered %i/%i attempted streams", len(to_restore), len(file_infos))
|
|
|
|
async def add_stream(self, rowid: int, sd_hash: str, file_name: str, download_directory: str, status: str,
|
|
claim: typing.Optional['StoredStreamClaim']):
|
|
sd_blob = self.blob_manager.get_blob(sd_hash)
|
|
if not sd_blob.get_is_verified():
|
|
return
|
|
try:
|
|
descriptor = await self.blob_manager.get_stream_descriptor(sd_blob.blob_hash)
|
|
except InvalidStreamDescriptorError as err:
|
|
log.warning("Failed to start stream for sd %s - %s", sd_hash, str(err))
|
|
return
|
|
if status == ManagedStream.STATUS_RUNNING:
|
|
downloader = self.make_downloader(descriptor.sd_hash, download_directory, file_name)
|
|
else:
|
|
downloader = None
|
|
stream = ManagedStream(
|
|
self.loop, self.blob_manager, rowid, descriptor, download_directory, file_name, downloader, status, claim
|
|
)
|
|
self.streams.add(stream)
|
|
self.storage.content_claim_callbacks[stream.stream_hash] = lambda: self._update_content_claim(stream)
|
|
|
|
async def load_streams_from_database(self):
|
|
to_recover = []
|
|
for file_info in await self.storage.get_all_lbry_files():
|
|
if not self.blob_manager.get_blob(file_info['sd_hash']).get_is_verified():
|
|
to_recover.append(file_info)
|
|
|
|
if to_recover:
|
|
log.info("Attempting to recover %i streams", len(to_recover))
|
|
await self.recover_streams(to_recover)
|
|
|
|
to_start = []
|
|
for file_info in await self.storage.get_all_lbry_files():
|
|
if self.blob_manager.get_blob(file_info['sd_hash']).get_is_verified():
|
|
to_start.append(file_info)
|
|
log.info("Initializing %i files", len(to_start))
|
|
|
|
await asyncio.gather(*[
|
|
self.add_stream(
|
|
file_info['rowid'], file_info['sd_hash'], binascii.unhexlify(file_info['file_name']).decode(),
|
|
binascii.unhexlify(file_info['download_directory']).decode(), file_info['status'],
|
|
file_info['claim']
|
|
) for file_info in to_start
|
|
])
|
|
log.info("Started stream manager with %i files", len(self.streams))
|
|
|
|
async def resume(self):
|
|
if self.node:
|
|
await self.node.joined.wait()
|
|
else:
|
|
log.warning("no DHT node given, resuming downloads trusting that we can contact reflector")
|
|
t = [
|
|
(stream.start_download(self.node), self.wait_for_stream_finished(stream))
|
|
for stream in self.streams if stream.status == ManagedStream.STATUS_RUNNING
|
|
]
|
|
if t:
|
|
log.info("resuming %i downloads", len(t))
|
|
|
|
async def reflect_streams(self):
|
|
while True:
|
|
if self.config.reflect_streams and self.config.reflector_servers:
|
|
sd_hashes = await self.storage.get_streams_to_re_reflect()
|
|
streams = list(filter(lambda s: s.sd_hash in sd_hashes, self.streams))
|
|
batch = []
|
|
while streams:
|
|
stream = streams.pop()
|
|
if not stream.fully_reflected.is_set():
|
|
host, port = random.choice(self.config.reflector_servers)
|
|
batch.append(stream.upload_to_reflector(host, port))
|
|
if len(batch) >= self.config.concurrent_reflector_uploads:
|
|
await asyncio.gather(*batch)
|
|
batch = []
|
|
if batch:
|
|
await asyncio.gather(*batch)
|
|
await asyncio.sleep(300, loop=self.loop)
|
|
|
|
async def start(self):
|
|
await self.load_streams_from_database()
|
|
self.resume_downloading_task = self.loop.create_task(self.resume())
|
|
self.re_reflect_task = self.loop.create_task(self.reflect_streams())
|
|
|
|
def stop(self):
|
|
if self.resume_downloading_task and not self.resume_downloading_task.done():
|
|
self.resume_downloading_task.cancel()
|
|
if self.re_reflect_task and not self.re_reflect_task.done():
|
|
self.re_reflect_task.cancel()
|
|
while self.streams:
|
|
stream = self.streams.pop()
|
|
stream.stop_download()
|
|
while self.update_stream_finished_futs:
|
|
self.update_stream_finished_futs.pop().cancel()
|
|
while self.running_reflector_uploads:
|
|
self.running_reflector_uploads.pop().cancel()
|
|
|
|
async def create_stream(self, file_path: str, key: typing.Optional[bytes] = None,
|
|
iv_generator: typing.Optional[typing.Generator[bytes, None, None]] = None) -> ManagedStream:
|
|
stream = await ManagedStream.create(self.loop, self.blob_manager, file_path, key, iv_generator)
|
|
self.streams.add(stream)
|
|
self.storage.content_claim_callbacks[stream.stream_hash] = lambda: self._update_content_claim(stream)
|
|
if self.config.reflect_streams and self.config.reflector_servers:
|
|
host, port = random.choice(self.config.reflector_servers)
|
|
task = self.loop.create_task(stream.upload_to_reflector(host, port))
|
|
self.running_reflector_uploads.append(task)
|
|
task.add_done_callback(
|
|
lambda _: None
|
|
if task not in self.running_reflector_uploads else self.running_reflector_uploads.remove(task)
|
|
)
|
|
return stream
|
|
|
|
async def delete_stream(self, stream: ManagedStream, delete_file: typing.Optional[bool] = False):
|
|
await self.stop_stream(stream)
|
|
if stream in self.streams:
|
|
self.streams.remove(stream)
|
|
blob_hashes = [stream.sd_hash] + [b.blob_hash for b in stream.descriptor.blobs[:-1]]
|
|
await self.blob_manager.delete_blobs(blob_hashes, delete_from_db=False)
|
|
await self.storage.delete_stream(stream.descriptor)
|
|
if delete_file and stream.output_file_exists:
|
|
os.remove(stream.full_path)
|
|
|
|
def get_stream_by_stream_hash(self, stream_hash: str) -> typing.Optional[ManagedStream]:
|
|
streams = tuple(filter(lambda stream: stream.stream_hash == stream_hash, self.streams))
|
|
if streams:
|
|
return streams[0]
|
|
|
|
def get_filtered_streams(self, sort_by: typing.Optional[str] = None, reverse: typing.Optional[bool] = False,
|
|
comparison: typing.Optional[str] = None,
|
|
**search_by) -> typing.List[ManagedStream]:
|
|
"""
|
|
Get a list of filtered and sorted ManagedStream objects
|
|
|
|
:param sort_by: field to sort by
|
|
:param reverse: reverse sorting
|
|
:param comparison: comparison operator used for filtering
|
|
:param search_by: fields and values to filter by
|
|
"""
|
|
if sort_by and sort_by not in filter_fields:
|
|
raise ValueError(f"'{sort_by}' is not a valid field to sort by")
|
|
if comparison and comparison not in comparison_operators:
|
|
raise ValueError(f"'{comparison}' is not a valid comparison")
|
|
if 'full_status' in search_by:
|
|
del search_by['full_status']
|
|
for search in search_by.keys():
|
|
if search not in filter_fields:
|
|
raise ValueError(f"'{search}' is not a valid search operation")
|
|
if search_by:
|
|
comparison = comparison or 'eq'
|
|
streams = []
|
|
for stream in self.streams:
|
|
for search, val in search_by.items():
|
|
if comparison_operators[comparison](getattr(stream, search), val):
|
|
streams.append(stream)
|
|
break
|
|
else:
|
|
streams = list(self.streams)
|
|
if sort_by:
|
|
streams.sort(key=lambda s: getattr(s, sort_by))
|
|
if reverse:
|
|
streams.reverse()
|
|
return streams
|
|
|
|
def wait_for_stream_finished(self, stream: ManagedStream):
|
|
async def _wait_for_stream_finished():
|
|
if stream.downloader and stream.running:
|
|
await stream.downloader.stream_finished_event.wait()
|
|
stream.update_status(ManagedStream.STATUS_FINISHED)
|
|
if self.analytics_manager:
|
|
self.loop.create_task(self.analytics_manager.send_download_finished(
|
|
stream.download_id, stream.claim_name, stream.sd_hash
|
|
))
|
|
|
|
task = self.loop.create_task(_wait_for_stream_finished())
|
|
self.update_stream_finished_futs.append(task)
|
|
task.add_done_callback(
|
|
lambda _: None if task not in self.update_stream_finished_futs else
|
|
self.update_stream_finished_futs.remove(task)
|
|
)
|
|
|
|
async def _store_stream(self, downloader: StreamDownloader) -> int:
|
|
file_name = os.path.basename(downloader.output_path)
|
|
download_directory = os.path.dirname(downloader.output_path)
|
|
if not await self.storage.stream_exists(downloader.sd_hash):
|
|
await self.storage.store_stream(downloader.sd_blob, downloader.descriptor)
|
|
if not await self.storage.file_exists(downloader.sd_hash):
|
|
return await self.storage.save_downloaded_file(
|
|
downloader.descriptor.stream_hash, file_name, download_directory,
|
|
0.0
|
|
)
|
|
else:
|
|
return await self.storage.rowid_for_stream(downloader.descriptor.stream_hash)
|
|
|
|
async def _check_update_or_replace(self, outpoint: str, claim_id: str, claim: ClaimDict) -> typing.Tuple[
|
|
typing.Optional[ManagedStream], typing.Optional[ManagedStream]]:
|
|
existing = self.get_filtered_streams(outpoint=outpoint)
|
|
if existing:
|
|
await self.start_stream(existing[0])
|
|
return existing[0], None
|
|
existing = self.get_filtered_streams(sd_hash=claim.source_hash.decode())
|
|
if existing and existing[0].claim_id != claim_id:
|
|
raise ResolveError(f"stream for {existing[0].claim_id} collides with existing "
|
|
f"download {claim_id}")
|
|
if existing:
|
|
log.info("claim contains a metadata only update to a stream we have")
|
|
await self.storage.save_content_claim(
|
|
existing[0].stream_hash, outpoint
|
|
)
|
|
await self._update_content_claim(existing[0])
|
|
await self.start_stream(existing[0])
|
|
return existing[0], None
|
|
else:
|
|
existing_for_claim_id = self.get_filtered_streams(claim_id=claim_id)
|
|
if existing_for_claim_id:
|
|
log.info("claim contains an update to a stream we have, downloading it")
|
|
return None, existing_for_claim_id[0]
|
|
return None, None
|
|
|
|
async def start_downloader(self, got_descriptor_time: asyncio.Future, downloader: StreamDownloader,
|
|
download_id: str, outpoint: str, claim: ClaimDict, resolved: typing.Dict,
|
|
file_name: typing.Optional[str] = None) -> ManagedStream:
|
|
start_time = self.loop.time()
|
|
downloader.download(self.node)
|
|
await downloader.got_descriptor.wait()
|
|
got_descriptor_time.set_result(self.loop.time() - start_time)
|
|
rowid = await self._store_stream(downloader)
|
|
await self.storage.save_content_claim(
|
|
downloader.descriptor.stream_hash, outpoint
|
|
)
|
|
stream = ManagedStream(self.loop, self.blob_manager, rowid, downloader.descriptor, self.config.download_dir,
|
|
file_name, downloader, ManagedStream.STATUS_RUNNING, download_id=download_id)
|
|
stream.set_claim(resolved, claim)
|
|
await stream.downloader.wrote_bytes_event.wait()
|
|
self.streams.add(stream)
|
|
return stream
|
|
|
|
async def _download_stream_from_uri(self, uri, timeout: float, exchange_rate_manager: 'ExchangeRateManager',
|
|
file_name: typing.Optional[str] = None) -> ManagedStream:
|
|
start_time = self.loop.time()
|
|
parsed_uri = parse_lbry_uri(uri)
|
|
if parsed_uri.is_channel:
|
|
raise ResolveError("cannot download a channel claim, specify a /path")
|
|
|
|
# resolve the claim
|
|
resolved = (await self.wallet.resolve(uri)).get(uri, {})
|
|
resolved = resolved if 'value' in resolved else resolved.get('claim')
|
|
|
|
if not resolved:
|
|
raise ResolveError(f"Failed to resolve stream at '{uri}'")
|
|
if 'error' in resolved:
|
|
raise ResolveError(f"error resolving stream: {resolved['error']}")
|
|
|
|
claim = Claim.from_bytes(binascii.unhexlify(resolved['hex']))
|
|
outpoint = f"{resolved['txid']}:{resolved['nout']}"
|
|
resolved_time = self.loop.time() - start_time
|
|
|
|
# resume or update an existing stream, if the stream changed download it and delete the old one after
|
|
updated_stream, to_replace = await self._check_update_or_replace(outpoint, resolved['claim_id'], claim)
|
|
if updated_stream:
|
|
return updated_stream
|
|
|
|
# check that the fee is payable
|
|
fee_amount, fee_address = None, None
|
|
if claim.stream.has_fee:
|
|
fee_amount = round(exchange_rate_manager.convert_currency(
|
|
claim.stream.fee.currency, "LBC", claim.stream.fee.amount
|
|
), 5)
|
|
max_fee_amount = round(exchange_rate_manager.convert_currency(
|
|
self.config.max_key_fee['currency'], "LBC", Decimal(self.config.max_key_fee['amount'])
|
|
), 5)
|
|
if fee_amount > max_fee_amount:
|
|
msg = f"fee of {fee_amount} exceeds max configured to allow of {max_fee_amount}"
|
|
log.warning(msg)
|
|
raise KeyFeeAboveMaxAllowed(msg)
|
|
balance = await self.wallet.default_account.get_balance()
|
|
if lbc_to_dewies(str(fee_amount)) > balance:
|
|
msg = f"fee of {fee_amount} exceeds max available balance"
|
|
log.warning(msg)
|
|
raise InsufficientFundsError(msg)
|
|
fee_address = claim.stream.fee.address
|
|
|
|
# download the stream
|
|
download_id = binascii.hexlify(generate_id()).decode()
|
|
downloader = StreamDownloader(self.loop, self.config, self.blob_manager, claim.stream.hash,
|
|
self.config.download_dir, file_name)
|
|
|
|
stream = None
|
|
descriptor_time_fut = self.loop.create_future()
|
|
start_download_time = self.loop.time()
|
|
time_to_descriptor = None
|
|
time_to_first_bytes = None
|
|
error = None
|
|
try:
|
|
stream = await asyncio.wait_for(
|
|
asyncio.ensure_future(
|
|
self.start_downloader(descriptor_time_fut, downloader, download_id, outpoint, claim, resolved,
|
|
file_name)
|
|
), timeout
|
|
)
|
|
time_to_descriptor = await descriptor_time_fut
|
|
time_to_first_bytes = self.loop.time() - start_download_time - time_to_descriptor
|
|
self.wait_for_stream_finished(stream)
|
|
if fee_address and fee_amount and not to_replace:
|
|
stream.tx = await self.wallet.send_amount_to_address(
|
|
lbc_to_dewies(str(fee_amount)), fee_address.encode('latin1'))
|
|
elif to_replace: # delete old stream now that the replacement has started downloading
|
|
await self.delete_stream(to_replace)
|
|
except asyncio.TimeoutError:
|
|
if descriptor_time_fut.done():
|
|
time_to_descriptor = descriptor_time_fut.result()
|
|
error = DownloadDataTimeout(downloader.sd_hash)
|
|
self.blob_manager.delete_blob(downloader.sd_hash)
|
|
await self.storage.delete_stream(downloader.descriptor)
|
|
else:
|
|
descriptor_time_fut.cancel()
|
|
error = DownloadSDTimeout(downloader.sd_hash)
|
|
if stream:
|
|
await self.stop_stream(stream)
|
|
else:
|
|
downloader.stop()
|
|
if error:
|
|
log.warning(error)
|
|
if self.analytics_manager:
|
|
self.loop.create_task(
|
|
self.analytics_manager.send_time_to_first_bytes(
|
|
resolved_time, self.loop.time() - start_time, download_id, parse_lbry_uri(uri).name, outpoint,
|
|
None if not stream else len(stream.downloader.blob_downloader.active_connections),
|
|
None if not stream else len(stream.downloader.blob_downloader.scores),
|
|
False if not downloader else downloader.added_fixed_peers,
|
|
self.config.fixed_peer_delay if not downloader else downloader.fixed_peers_delay,
|
|
claim.stream.hash, time_to_descriptor,
|
|
None if not (stream and stream.descriptor) else stream.descriptor.blobs[0].blob_hash,
|
|
None if not (stream and stream.descriptor) else stream.descriptor.blobs[0].length,
|
|
time_to_first_bytes, None if not error else error.__class__.__name__
|
|
)
|
|
)
|
|
if error:
|
|
raise error
|
|
return stream
|
|
|
|
async def download_stream_from_uri(self, uri, exchange_rate_manager: 'ExchangeRateManager',
|
|
file_name: typing.Optional[str] = None,
|
|
timeout: typing.Optional[float] = None) -> ManagedStream:
|
|
timeout = timeout or self.config.download_timeout
|
|
if uri in self.starting_streams:
|
|
return await self.starting_streams[uri]
|
|
fut = asyncio.Future(loop=self.loop)
|
|
self.starting_streams[uri] = fut
|
|
try:
|
|
stream = await self._download_stream_from_uri(uri, timeout, exchange_rate_manager, file_name)
|
|
fut.set_result(stream)
|
|
except Exception as err:
|
|
fut.set_exception(err)
|
|
try:
|
|
return await fut
|
|
finally:
|
|
del self.starting_streams[uri]
|