PyBitmessage-2021-04-27/src/class_singleCleaner.py

205 lines
7.7 KiB
Python

"""
The singleCleaner class is a timer-driven thread that cleans data structures
to free memory, resends messages when a remote node doesn't respond, and
sends pong messages to keep connections alive if the network isn't busy.
It cleans these data structures in memory:
inventory (moves data to the on-disk sql database)
inventorySets (clears then reloads data out of sql database)
It cleans these tables on the disk:
inventory (clears expired objects)
pubkeys (clears pubkeys older than 4 weeks old which we have not used
personally)
knownNodes (clears addresses which have not been online for over 3 days)
It resends messages when there has been no response:
resends getpubkey messages in 5 days (then 10 days, then 20 days, etc...)
resends msg messages in 5 days (then 10 days, then 20 days, etc...)
"""
import gc
import os
import shared
import threading
import time
import tr
from bmconfigparser import BMConfigParser
from helper_sql import sqlQuery, sqlExecute
from helper_threading import StoppableThread
from inventory import Inventory
from network.connectionpool import BMConnectionPool
from debug import logger
import knownnodes
import queues
import state
class singleCleaner(threading.Thread, StoppableThread):
cycleLength = 300
expireDiscoveredPeers = 300
def __init__(self):
threading.Thread.__init__(self, name="singleCleaner")
self.initStop()
def run(self):
gc.disable()
timeWeLastClearedInventoryAndPubkeysTables = 0
try:
shared.maximumLengthOfTimeToBotherResendingMessages = (
float(BMConfigParser().get(
'bitmessagesettings', 'stopresendingafterxdays')) *
24 * 60 * 60
) + (
float(BMConfigParser().get(
'bitmessagesettings', 'stopresendingafterxmonths')) *
(60 * 60 * 24 * 365) / 12)
except:
# Either the user hasn't set stopresendingafterxdays and
# stopresendingafterxmonths yet or the options are missing
# from the config file.
shared.maximumLengthOfTimeToBotherResendingMessages = float('inf')
# initial wait
if state.shutdown == 0:
self.stop.wait(singleCleaner.cycleLength)
while state.shutdown == 0:
queues.UISignalQueue.put((
'updateStatusBar',
'Doing housekeeping (Flushing inventory in memory to disk...)'
))
Inventory().flush()
queues.UISignalQueue.put(('updateStatusBar', ''))
# If we are running as a daemon then we are going to fill up the UI
# queue which will never be handled by a UI. We should clear it to
# save memory.
# FIXME redundant?
if shared.thisapp.daemon or not state.enableGUI:
queues.UISignalQueue.queue.clear()
if timeWeLastClearedInventoryAndPubkeysTables < \
int(time.time()) - 7380:
timeWeLastClearedInventoryAndPubkeysTables = int(time.time())
Inventory().clean()
queues.workerQueue.put(('sendOnionPeerObj', ''))
# pubkeys
sqlExecute(
"DELETE FROM pubkeys WHERE time<? AND usedpersonally='no'",
int(time.time()) - shared.lengthOfTimeToHoldOnToAllPubkeys)
# Let us resend getpubkey objects if we have not yet heard
# a pubkey, and also msg objects if we have not yet heard
# an acknowledgement
queryreturn = sqlQuery(
"SELECT toaddress, ackdata, status FROM sent"
" WHERE ((status='awaitingpubkey' OR status='msgsent')"
" AND folder='sent' AND sleeptill<? AND senttime>?)",
int(time.time()), int(time.time()) -
shared.maximumLengthOfTimeToBotherResendingMessages
)
for row in queryreturn:
if len(row) < 2:
logger.error(
'Something went wrong in the singleCleaner thread:'
' a query did not return the requested fields. %r',
row
)
self.stop.wait(3)
break
toAddress, ackData, status = row
if status == 'awaitingpubkey':
resendPubkeyRequest(toAddress)
elif status == 'msgsent':
resendMsg(ackData)
try:
# Cleanup knownnodes and handle possible severe exception
# while writing it to disk
knownnodes.cleanupKnownNodes()
except Exception as err:
if "Errno 28" in str(err):
logger.fatal(
'(while writing knownnodes to disk)'
' Alert: Your disk or data storage volume is full.'
)
queues.UISignalQueue.put((
'alert',
(tr._translate("MainWindow", "Disk full"),
tr._translate(
"MainWindow",
'Alert: Your disk or data storage volume'
' is full. Bitmessage will now exit.'),
True)
))
# FIXME redundant?
if shared.daemon or not state.enableGUI:
os._exit(1)
# # clear download queues
# for thread in threading.enumerate():
# if thread.isAlive() and hasattr(thread, 'downloadQueue'):
# thread.downloadQueue.clear()
# inv/object tracking
for connection in \
BMConnectionPool().inboundConnections.values() + \
BMConnectionPool().outboundConnections.values():
connection.clean()
# discovery tracking
exp = time.time() - singleCleaner.expireDiscoveredPeers
reaper = (k for k, v in state.discoveredPeers.items() if v < exp)
for k in reaper:
try:
del state.discoveredPeers[k]
except KeyError:
pass
# TODO: cleanup pending upload / download
gc.collect()
if state.shutdown == 0:
self.stop.wait(singleCleaner.cycleLength)
def resendPubkeyRequest(address):
logger.debug(
'It has been a long time and we haven\'t heard a response to our'
' getpubkey request. Sending again.'
)
try:
# We need to take this entry out of the neededPubkeys structure
# because the queues.workerQueue checks to see whether the entry
# is already present and will not do the POW and send the message
# because it assumes that it has already done it recently.
del state.neededPubkeys[address]
except:
pass
queues.UISignalQueue.put((
'updateStatusBar',
'Doing work necessary to again attempt to request a public key...'
))
sqlExecute(
'''UPDATE sent SET status='msgqueued' WHERE toaddress=?''',
address)
queues.workerQueue.put(('sendmessage', ''))
def resendMsg(ackdata):
logger.debug(
'It has been a long time and we haven\'t heard an acknowledgement'
' to our msg. Sending again.'
)
sqlExecute(
'''UPDATE sent SET status='msgqueued' WHERE ackdata=?''',
ackdata)
queues.workerQueue.put(('sendmessage', ''))
queues.UISignalQueue.put((
'updateStatusBar',
'Doing work necessary to again attempt to deliver a message...'
))