191 lines
8.6 KiB
Python
191 lines
8.6 KiB
Python
from random import choice
|
|
import logging
|
|
from twisted.internet import defer, task
|
|
from twisted.internet.error import ConnectingCancelledError
|
|
from twisted.web._newclient import ResponseNeverReceived
|
|
import treq
|
|
from lbrynet.core.utils import DeferredDict
|
|
from lbrynet.core.Error import DownloadCanceledError
|
|
|
|
log = logging.getLogger(__name__)
|
|
|
|
|
|
class HTTPBlobDownloader:
|
|
'''
|
|
A downloader that is able to get blobs from HTTP mirrors.
|
|
Note that when a blob gets downloaded from a mirror or from a peer, BlobManager will mark it as completed
|
|
and cause any other type of downloader to progress to the next missing blob. Also, BlobFile is naturally able
|
|
to cancel other writers when a writer finishes first. That's why there is no call to cancel/resume/stop between
|
|
different types of downloaders.
|
|
'''
|
|
def __init__(self, blob_manager, blob_hashes=None, servers=None, client=None, sd_hashes=None, retry=True,
|
|
clock=None):
|
|
if not clock:
|
|
from twisted.internet import reactor
|
|
self.clock = reactor
|
|
else:
|
|
self.clock = clock
|
|
self.blob_manager = blob_manager
|
|
self.servers = servers or []
|
|
self.client = client or treq
|
|
self.blob_hashes = blob_hashes or []
|
|
self.missing_blob_hashes = []
|
|
self.downloaded_blob_hashes = []
|
|
self.sd_hashes = sd_hashes or []
|
|
self.head_blob_hashes = []
|
|
self.max_failures = 3
|
|
self.semaphore = defer.DeferredSemaphore(2)
|
|
self.deferreds = []
|
|
self.writers = []
|
|
self.retry = retry
|
|
self.looping_call = task.LoopingCall(self._download_lc)
|
|
self.looping_call.clock = self.clock
|
|
self.finished_deferred = defer.Deferred()
|
|
self.finished_deferred.addErrback(lambda err: err.trap(defer.CancelledError))
|
|
self.short_delay = 30
|
|
self.long_delay = 600
|
|
self.delay = self.short_delay
|
|
self.last_missing = 10000000
|
|
self.lc_deferred = None
|
|
|
|
@defer.inlineCallbacks
|
|
def start(self):
|
|
if not self.looping_call.running:
|
|
self.lc_deferred = self.looping_call.start(self.short_delay, now=True)
|
|
self.lc_deferred.addErrback(lambda err: err.trap(defer.CancelledError))
|
|
yield self.finished_deferred
|
|
|
|
def stop(self):
|
|
for d in reversed(self.deferreds):
|
|
d.cancel()
|
|
while self.writers:
|
|
writer = self.writers.pop()
|
|
writer.close(DownloadCanceledError())
|
|
self.blob_hashes = []
|
|
if self.looping_call.running:
|
|
self.looping_call.stop()
|
|
if self.lc_deferred and not self.lc_deferred.called:
|
|
self.lc_deferred.cancel()
|
|
if not self.finished_deferred.called:
|
|
self.finished_deferred.cancel()
|
|
|
|
@defer.inlineCallbacks
|
|
def _download_lc(self):
|
|
delay = yield self._download_and_get_retry_delay()
|
|
log.debug("delay: %s, missing: %i, downloaded from mirror: %i", delay, len(self.missing_blob_hashes),
|
|
len(self.downloaded_blob_hashes))
|
|
while self.missing_blob_hashes:
|
|
self.blob_hashes.append(self.missing_blob_hashes.pop())
|
|
if not delay:
|
|
if self.looping_call.running:
|
|
self.looping_call.stop()
|
|
if not self.finished_deferred.called:
|
|
log.debug("mirror finished")
|
|
self.finished_deferred.callback(None)
|
|
elif delay and delay != self.delay:
|
|
if delay == self.long_delay:
|
|
log.debug("mirror not making progress, trying less frequently")
|
|
elif delay == self.short_delay:
|
|
log.debug("queueing retry of %i blobs", len(self.missing_blob_hashes))
|
|
if self.looping_call.running:
|
|
self.looping_call.stop()
|
|
self.delay = delay
|
|
self.looping_call = task.LoopingCall(self._download_lc)
|
|
self.looping_call.clock = self.clock
|
|
self.lc_deferred = self.looping_call.start(self.delay, now=False)
|
|
self.lc_deferred.addErrback(lambda err: err.trap(defer.CancelledError))
|
|
yield self.finished_deferred
|
|
|
|
@defer.inlineCallbacks
|
|
def _download_and_get_retry_delay(self):
|
|
if self.blob_hashes and self.servers:
|
|
if self.sd_hashes:
|
|
log.debug("trying to download stream from mirror (sd %s)", self.sd_hashes[0][:8])
|
|
else:
|
|
log.debug("trying to download %i blobs from mirror", len(self.blob_hashes))
|
|
blobs = yield DeferredDict(
|
|
{blob_hash: self.blob_manager.get_blob(blob_hash) for blob_hash in self.blob_hashes}
|
|
)
|
|
self.deferreds = [self.download_blob(blobs[blob_hash]) for blob_hash in self.blob_hashes]
|
|
yield defer.DeferredList(self.deferreds)
|
|
if self.retry and self.missing_blob_hashes:
|
|
if not self.downloaded_blob_hashes:
|
|
defer.returnValue(self.long_delay)
|
|
if len(self.missing_blob_hashes) < self.last_missing:
|
|
self.last_missing = len(self.missing_blob_hashes)
|
|
defer.returnValue(self.short_delay)
|
|
if self.retry and self.last_missing and len(self.missing_blob_hashes) == self.last_missing:
|
|
defer.returnValue(self.long_delay)
|
|
defer.returnValue(None)
|
|
|
|
@defer.inlineCallbacks
|
|
def _download_blob(self, blob):
|
|
for _ in range(self.max_failures):
|
|
writer, finished_deferred = blob.open_for_writing('mirror')
|
|
self.writers.append(writer)
|
|
try:
|
|
downloaded = yield self._write_blob(writer, blob)
|
|
if downloaded:
|
|
yield finished_deferred # yield for verification errors, so we log them
|
|
if blob.verified:
|
|
log.info('Mirror completed download for %s', blob.blob_hash)
|
|
should_announce = blob.blob_hash in self.sd_hashes or blob.blob_hash in self.head_blob_hashes
|
|
yield self.blob_manager.blob_completed(blob, should_announce=should_announce)
|
|
self.downloaded_blob_hashes.append(blob.blob_hash)
|
|
break
|
|
except (IOError, Exception, defer.CancelledError, ConnectingCancelledError, ResponseNeverReceived) as e:
|
|
if isinstance(
|
|
e, (DownloadCanceledError, defer.CancelledError, ConnectingCancelledError,
|
|
ResponseNeverReceived)
|
|
) or 'closed file' in str(e):
|
|
# some other downloader finished first or it was simply cancelled
|
|
log.info("Mirror download cancelled: %s", blob.blob_hash)
|
|
break
|
|
else:
|
|
log.exception('Mirror failed downloading')
|
|
finally:
|
|
finished_deferred.addBoth(lambda _: None) # suppress echoed errors
|
|
if 'mirror' in blob.writers:
|
|
writer.close()
|
|
self.writers.remove(writer)
|
|
|
|
def download_blob(self, blob):
|
|
if not blob.verified:
|
|
d = self.semaphore.run(self._download_blob, blob)
|
|
d.addErrback(lambda err: err.trap(defer.TimeoutError, defer.CancelledError))
|
|
return d
|
|
return defer.succeed(None)
|
|
|
|
@defer.inlineCallbacks
|
|
def _write_blob(self, writer, blob):
|
|
response = yield self.client.get(url_for(choice(self.servers), blob.blob_hash))
|
|
if response.code != 200:
|
|
log.debug('Missing a blob: %s', blob.blob_hash)
|
|
if blob.blob_hash in self.blob_hashes:
|
|
self.blob_hashes.remove(blob.blob_hash)
|
|
if blob.blob_hash not in self.missing_blob_hashes:
|
|
self.missing_blob_hashes.append(blob.blob_hash)
|
|
defer.returnValue(False)
|
|
|
|
log.debug('Download started: %s', blob.blob_hash)
|
|
blob.set_length(response.length)
|
|
yield self.client.collect(response, writer.write)
|
|
defer.returnValue(True)
|
|
|
|
@defer.inlineCallbacks
|
|
def download_stream(self, stream_hash, sd_hash):
|
|
stream_crypt_blobs = yield self.blob_manager.storage.get_blobs_for_stream(stream_hash)
|
|
self.blob_hashes.extend([
|
|
b.blob_hash for b in stream_crypt_blobs
|
|
if b.blob_hash and b.blob_hash not in self.blob_hashes
|
|
])
|
|
if sd_hash not in self.sd_hashes:
|
|
self.sd_hashes.append(sd_hash)
|
|
head_blob_hash = stream_crypt_blobs[0].blob_hash
|
|
if head_blob_hash not in self.head_blob_hashes:
|
|
self.head_blob_hashes.append(head_blob_hash)
|
|
yield self.start()
|
|
|
|
|
|
def url_for(server, blob_hash=''):
|
|
return 'http://{}/{}'.format(server, blob_hash)
|