PyBitmessage/src/inventory.py
Peter Surda e434825bb2
Pending download counter fix
- handles expired objects better
- counts objects that failed download more accurately
2017-02-06 11:02:03 +01:00

291 lines
13 KiB
Python

import collections
from threading import current_thread, enumerate as threadingEnumerate, RLock
import time
from helper_sql import *
from singleton import Singleton
@Singleton
class Inventory(collections.MutableMapping):
def __init__(self):
super(self.__class__, self).__init__()
self._inventory = {} #of objects (like msg payloads and pubkey payloads) Does not include protocol headers (the first 24 bytes of each packet).
self.numberOfInventoryLookupsPerformed = 0
self._streams = collections.defaultdict(set) # key = streamNumer, value = a set which holds the inventory object hashes that we are aware of. This is used whenever we receive an inv message from a peer to check to see what items are new to us. We don't delete things out of it; instead, the singleCleaner thread clears and refills it every couple hours.
self.lock = RLock() # Guarantees that two receiveDataThreads don't receive and process the same message concurrently (probably sent by a malicious individual)
self.InventoryItem = collections.namedtuple('InventoryItem', 'type stream payload expires tag')
def __contains__(self, hash):
with self.lock:
self.numberOfInventoryLookupsPerformed += 1
if hash in self._inventory:
return True
return bool(sqlQuery('SELECT 1 FROM inventory WHERE hash=?', hash))
def __getitem__(self, hash):
with self.lock:
if hash in self._inventory:
return self._inventory[hash]
rows = sqlQuery('SELECT objecttype, streamnumber, payload, expirestime, tag FROM inventory WHERE hash=?', hash)
if not rows:
raise KeyError(hash)
return self.InventoryItem(*rows[0])
def __setitem__(self, hash, value):
with self.lock:
value = self.InventoryItem(*value)
self._inventory[hash] = value
self._streams[value.stream].add(hash)
PendingDownload().delete(hash)
def __delitem__(self, hash):
raise NotImplementedError
def __iter__(self):
with self.lock:
hashes = self._inventory.keys()[:]
hashes += (x for x, in sqlQuery('SELECT hash FROM inventory'))
return hashes.__iter__()
def __len__(self):
with self.lock:
return len(self._inventory) + sqlQuery('SELECT count(*) FROM inventory')[0][0]
def by_type_and_tag(self, type, tag):
with self.lock:
values = [value for value in self._inventory.values() if value.type == type and value.tag == tag]
values += (self.InventoryItem(*value) for value in sqlQuery('SELECT objecttype, streamnumber, payload, expirestime, tag FROM inventory WHERE objecttype=? AND tag=?', type, tag))
return values
def hashes_by_stream(self, stream):
with self.lock:
return self._streams[stream]
def unexpired_hashes_by_stream(self, stream):
with self.lock:
t = int(time.time())
hashes = [x for x, value in self._inventory.items() if value.stream == stream and value.expires > t]
hashes += (payload for payload, in sqlQuery('SELECT hash FROM inventory WHERE streamnumber=? AND expirestime>?', stream, t))
return hashes
def flush(self):
with self.lock: # If you use both the inventoryLock and the sqlLock, always use the inventoryLock OUTSIDE of the sqlLock.
with SqlBulkExecute() as sql:
for objectHash, value in self._inventory.items():
sql.execute('INSERT INTO inventory VALUES (?, ?, ?, ?, ?, ?)', objectHash, *value)
self._inventory.clear()
def clean(self):
with self.lock:
sqlExecute('DELETE FROM inventory WHERE expirestime<?',int(time.time()) - (60 * 60 * 3))
self._streams.clear()
for objectHash, value in self.items():
self._streams[value.stream].add(objectHash)
@Singleton
class PendingDownload(object):
# keep a track of objects that have been advertised to us but we haven't downloaded them yet
def __init__(self):
super(self.__class__, self).__init__()
self.lock = RLock()
self.hashes = {}
self.stopped = False
# don't request the same object more frequently than this
self.frequency = 60
# after requesting and not receiving an object more than this times, consider it expired
self.maxRequestCount = 3
self.pending = {}
def add(self, objectHash):
if self.stopped:
return
with self.lock:
if objectHash not in self.hashes:
self.hashes[objectHash] = {'peers':[], 'requested':0, 'requestedCount':0}
self.hashes[objectHash]['peers'].append(current_thread().peer)
def addPending(self, objectHash=None):
if self.stopped:
return
if current_thread().peer not in self.pending:
self.pending[current_thread().peer] = {'objects':[], 'requested':0, 'received':0}
if objectHash not in self.pending[current_thread().peer]['objects'] and not objectHash is None:
self.pending[current_thread().peer]['objects'].append(objectHash)
self.pending[current_thread().peer]['requested'] = time.time()
def len(self):
with self.lock:
return sum(1 for x in self.hashes.values() if len(x) > 0)
def pull(self, count=1):
if count < 1:
raise ValueError("Must be at least one")
objectHashes = []
unreachableObjects = []
if self.stopped:
return objectHashes
start = time.time()
try:
for objectHash in self.hashes.keys():
with self.lock:
if len(objectHashes) >= count:
break
if current_thread().peer not in self.pending:
self.addPending()
if (self.pending[current_thread().peer]['requested'] >= time.time() - self.frequency or \
self.pending[current_thread().peer]['received'] >= time.time() - self.frequency) and \
len(self.pending[current_thread().peer]['objects']) >= count:
break
if len(self.hashes[objectHash]['peers']) == 0:
unreachableObjects.append(objectHash)
continue
# requested too long ago or not at all from any thread
if self.hashes[objectHash]['requested'] < time.time() - self.frequency:
# ready requested from this thread but haven't received yet
if objectHash in self.pending[current_thread().peer]['objects']:
# if still sending or receiving, request next
if self.pending[current_thread().peer]['received'] >= time.time() - self.frequency or \
self.pending[current_thread().peer]['requested'] >= time.time() - self.frequency:
continue
# haven't requested or received anything recently, re-request (i.e. continue)
# the current node doesn't have the object
elif current_thread().peer not in self.hashes[objectHash]['peers']:
continue
# already requested too many times, remove all signs of this object
if self.hashes[objectHash]['requestedCount'] >= self.maxRequestCount:
del self.hashes[objectHash]
for thread in self.pending.keys():
if objectHash in self.pending[thread]['objects']:
self.pending[thread]['objects'].remove(objectHash)
continue
# all ok, request
objectHashes.append(objectHash)
self.hashes[objectHash]['requested'] = time.time()
self.hashes[objectHash]['requestedCount'] += 1
self.pending[current_thread().peer]['requested'] = time.time()
self.addPending(objectHash)
except (RuntimeError, KeyError, ValueError):
# the for cycle sometimes breaks if you remove elements
pass
for objectHash in unreachableObjects:
with self.lock:
if objectHash in self.hashes:
del self.hashes[objectHash]
# logger.debug("Pull took %.3f seconds", time.time() - start)
return objectHashes
def delete(self, objectHash):
with self.lock:
if objectHash in self.hashes:
del self.hashes[objectHash]
if hasattr(current_thread(), 'peer') and current_thread().peer in self.pending:
self.pending[current_thread().peer]['received'] = time.time()
for thread in self.pending.keys():
with self.lock:
if thread in self.pending and objectHash in self.pending[thread]['objects']:
self.pending[thread]['objects'].remove(objectHash)
def stop(self):
with self.lock:
self.hashes = {}
self.pending = {}
def threadEnd(self):
while True:
try:
with self.lock:
if current_thread().peer in self.pending:
for objectHash in self.pending[current_thread().peer]['objects']:
if objectHash in self.hashes:
self.hashes[objectHash]['peers'].remove(current_thread().peer)
except (KeyError):
pass
else:
break
with self.lock:
try:
del self.pending[current_thread().peer]
except KeyError:
pass
class PendingUploadDeadlineException(Exception):
pass
@Singleton
class PendingUpload(object):
# keep a track of objects that we have created but haven't distributed yet
def __init__(self):
super(self.__class__, self).__init__()
self.lock = RLock()
self.hashes = {}
# end by this time in any case
self.deadline = 0
self.maxLen = 0
def add(self, objectHash = None):
with self.lock:
# add a new object into existing thread lists
if objectHash:
if objectHash not in self.hashes:
self.hashes[objectHash] = []
for thread in threadingEnumerate():
if thread.isAlive() and hasattr(thread, 'peer') and \
thread.peer not in self.hashes[objectHash]:
self.hashes[objectHash].append(thread.peer)
# add all objects into the current thread
else:
for objectHash in self.hashes:
if current_thread().peer not in self.hashes[objectHash]:
self.hashes[objectHash].append(current_thread().peer)
def len(self):
with self.lock:
return sum(len(self.hashes[x]) > 0 for x in self.hashes)
def _progress(self):
with self.lock:
return float(sum(len(self.hashes[x]) for x in self.hashes))
def progress(self, throwDeadline=True):
if self.maxLen < self._progress():
self.maxLen = self._progress()
if self.deadline < time.time():
if self.deadline > 0 and throwDeadline:
raise PendingUploadDeadlineException
self.deadline = time.time() + 20
try:
return 1.0 - self._progress() / self.maxLen
except ZeroDivisionError:
return 1.0
def delete(self, objectHash):
if not hasattr(current_thread(), 'peer'):
return
with self.lock:
if objectHash in self.hashes and current_thread().peer in self.hashes[objectHash]:
self.hashes[objectHash].remove(current_thread().peer)
if len(self.hashes[objectHash]) == 0:
del self.hashes[objectHash]
def stop(self):
with self.lock:
self.hashes = {}
def threadEnd(self):
while True:
try:
with self.lock:
for objectHash in self.hashes:
if current_thread().peer in self.hashes[objectHash]:
self.hashes[objectHash].remove(current_thread().peer)
if len(self.hashes[objectHash]) == 0:
del self.hashes[objectHash]
except (KeyError, RuntimeError):
pass
else:
break