hydrus/include/HydrusNetworking.py

734 lines
24 KiB
Python

import calendar
import collections
import datetime
import httplib
import HydrusConstants as HC
import HydrusData
import HydrusSerialisable
import socket
import ssl
import threading
import time
# The calendar portion of this works in GMT. A new 'day' or 'month' is calculated based on GMT time, so it won't tick over at midnight for most people.
# But this means a server can pass a bandwidth object to a lad and everyone can agree on when a new day is.
def ConvertBandwidthRuleToString( rule ):
( bandwidth_type, time_delta, max_allowed ) = rule
if max_allowed == 0:
return 'No requests currently permitted.'
if bandwidth_type == HC.BANDWIDTH_TYPE_DATA:
s = HydrusData.ConvertIntToBytes( max_allowed )
elif bandwidth_type == HC.BANDWIDTH_TYPE_REQUESTS:
s = HydrusData.ToHumanInt( max_allowed ) + ' rqs'
if time_delta is None:
s += ' per month'
else:
s += ' per ' + HydrusData.TimeDeltaToPrettyTimeDelta( time_delta )
return s
def GetLocalConnection( port, https = False ):
old_socket = httplib.socket.socket
httplib.socket.socket = socket._socketobject
try:
if https:
context = ssl.SSLContext( ssl.PROTOCOL_SSLv23 )
context.options |= ssl.OP_NO_SSLv2
context.options |= ssl.OP_NO_SSLv3
connection = httplib.HTTPSConnection( '127.0.0.1', port, timeout = 8, context = context )
else:
connection = httplib.HTTPConnection( '127.0.0.1', port, timeout = 8 )
connection.connect()
finally:
httplib.socket.socket = old_socket
return connection
class BandwidthRules( HydrusSerialisable.SerialisableBase ):
SERIALISABLE_TYPE = HydrusSerialisable.SERIALISABLE_TYPE_BANDWIDTH_RULES
SERIALISABLE_NAME = 'Bandwidth Rules'
SERIALISABLE_VERSION = 1
def __init__( self ):
HydrusSerialisable.SerialisableBase.__init__( self )
self._lock = threading.Lock()
self._rules = set()
def _GetSerialisableInfo( self ):
return list( self._rules )
def _InitialiseFromSerialisableInfo( self, serialisable_info ):
# tuples converted to lists via json
self._rules = set( ( tuple( rule_list ) for rule_list in serialisable_info ) )
def AddRule( self, bandwidth_type, time_delta, max_allowed ):
with self._lock:
rule = ( bandwidth_type, time_delta, max_allowed )
self._rules.add( rule )
def CanContinueDownload( self, bandwidth_tracker, threshold = 15 ):
with self._lock:
for ( bandwidth_type, time_delta, max_allowed ) in self._rules:
# Do not stop ongoing just because starts are throttled
requests_rule = bandwidth_type == HC.BANDWIDTH_TYPE_REQUESTS
# Do not block an ongoing jpg download because the current month is 100.03% used
wait_is_too_long = time_delta is None or time_delta > threshold
ignore_rule = requests_rule or wait_is_too_long
if ignore_rule:
continue
if bandwidth_tracker.GetUsage( bandwidth_type, time_delta ) >= max_allowed:
return False
return True
def CanDoWork( self, bandwidth_tracker, expected_requests, expected_bytes, threshold = 30 ):
with self._lock:
for ( bandwidth_type, time_delta, max_allowed ) in self._rules:
# Do not prohibit a raft of work starting or continuing because one small rule is over at this current second
if time_delta is not None and time_delta <= threshold:
continue
# we don't want to do a tiny amount of work, we want to do a decent whack
if bandwidth_type == HC.BANDWIDTH_TYPE_REQUESTS:
max_allowed -= expected_requests
elif bandwidth_type == HC.BANDWIDTH_TYPE_DATA:
max_allowed -= expected_bytes
if bandwidth_tracker.GetUsage( bandwidth_type, time_delta ) >= max_allowed:
return False
return True
def CanStartRequest( self, bandwidth_tracker, threshold = 5 ):
with self._lock:
for ( bandwidth_type, time_delta, max_allowed ) in self._rules:
# Do not prohibit a new job from starting just because the current download speed is 210/200KB/s
ignore_rule = bandwidth_type == HC.BANDWIDTH_TYPE_DATA and time_delta is not None and time_delta <= threshold
if ignore_rule:
continue
if bandwidth_tracker.GetUsage( bandwidth_type, time_delta ) >= max_allowed:
return False
return True
def GetWaitingEstimate( self, bandwidth_tracker ):
with self._lock:
estimates = []
for ( bandwidth_type, time_delta, max_allowed ) in self._rules:
if bandwidth_tracker.GetUsage( bandwidth_type, time_delta ) >= max_allowed:
estimates.append( bandwidth_tracker.GetWaitingEstimate( bandwidth_type, time_delta, max_allowed ) )
if len( estimates ) == 0:
return 0
else:
return max( estimates )
def GetBandwidthStringsAndGaugeTuples( self, bandwidth_tracker, threshold = 600 ):
with self._lock:
rows = []
rules_sorted = list( self._rules )
def key( ( bandwidth_type, time_delta, max_allowed ) ):
return time_delta
rules_sorted.sort( key = key )
for ( bandwidth_type, time_delta, max_allowed ) in rules_sorted:
time_is_less_than_threshold = time_delta is not None and time_delta <= threshold
if time_is_less_than_threshold or max_allowed == 0:
continue
usage = bandwidth_tracker.GetUsage( bandwidth_type, time_delta )
s = 'used '
if bandwidth_type == HC.BANDWIDTH_TYPE_DATA:
s += HydrusData.ConvertValueRangeToBytes( usage, max_allowed )
elif bandwidth_type == HC.BANDWIDTH_TYPE_REQUESTS:
s += HydrusData.ConvertValueRangeToPrettyString( usage, max_allowed ) + ' requests'
if time_delta is None:
s += ' this month'
else:
s += ' in the past ' + HydrusData.TimeDeltaToPrettyTimeDelta( time_delta )
rows.append( ( s, ( usage, max_allowed ) ) )
return rows
def GetRules( self ):
with self._lock:
return list( self._rules )
HydrusSerialisable.SERIALISABLE_TYPES_TO_OBJECT_TYPES[ HydrusSerialisable.SERIALISABLE_TYPE_BANDWIDTH_RULES ] = BandwidthRules
class BandwidthTracker( HydrusSerialisable.SerialisableBase ):
SERIALISABLE_TYPE = HydrusSerialisable.SERIALISABLE_TYPE_BANDWIDTH_TRACKER
SERIALISABLE_NAME = 'Bandwidth Tracker'
SERIALISABLE_VERSION = 1
# I want to track and query using smaller periods even when the total time delta is larger than the next step up to increase granularity
# for instance, querying minutes for 90 mins time delta is more smooth than watching a juddery sliding two hour window
MAX_SECONDS_TIME_DELTA = 240
MAX_MINUTES_TIME_DELTA = 180 * 60
MAX_HOURS_TIME_DELTA = 72 * 3600
MAX_DAYS_TIME_DELTA = 31 * 86400
CACHE_MAINTENANCE_TIME_DELTA = 120
MIN_TIME_DELTA_FOR_USER = 10
def __init__( self ):
HydrusSerialisable.SerialisableBase.__init__( self )
self._lock = threading.Lock()
self._next_cache_maintenance_timestamp = HydrusData.GetNow() + self.CACHE_MAINTENANCE_TIME_DELTA
self._months_bytes = collections.Counter()
self._days_bytes = collections.Counter()
self._hours_bytes = collections.Counter()
self._minutes_bytes = collections.Counter()
self._seconds_bytes = collections.Counter()
self._months_requests = collections.Counter()
self._days_requests = collections.Counter()
self._hours_requests = collections.Counter()
self._minutes_requests = collections.Counter()
self._seconds_requests = collections.Counter()
def _GetSerialisableInfo( self ):
dicts_flat = []
for d in ( self._months_bytes, self._days_bytes, self._hours_bytes, self._minutes_bytes, self._seconds_bytes, self._months_requests, self._days_requests, self._hours_requests, self._minutes_requests, self._seconds_requests ):
dicts_flat.append( d.items() )
return dicts_flat
def _InitialiseFromSerialisableInfo( self, serialisable_info ):
counters = [ collections.Counter( dict( flat_dict ) ) for flat_dict in serialisable_info ]
# unusual error someone reported by email--it came back an empty list, fugg
if len( counters ) != 10:
return
self._months_bytes = counters[ 0 ]
self._days_bytes = counters[ 1 ]
self._hours_bytes = counters[ 2 ]
self._minutes_bytes = counters[ 3 ]
self._seconds_bytes = counters[ 4 ]
self._months_requests = counters[ 5 ]
self._days_requests = counters[ 6 ]
self._hours_requests = counters[ 7 ]
self._minutes_requests = counters[ 8 ]
self._seconds_requests = counters[ 9 ]
def _GetCurrentDateTime( self ):
# keep getnow in here for the moment to aid in testing, which patches it to do time shifting
return datetime.datetime.utcfromtimestamp( HydrusData.GetNow() )
def _GetWindowAndCounter( self, bandwidth_type, time_delta ):
if bandwidth_type == HC.BANDWIDTH_TYPE_DATA:
if time_delta < self.MAX_SECONDS_TIME_DELTA:
window = 0
counter = self._seconds_bytes
elif time_delta < self.MAX_MINUTES_TIME_DELTA:
window = 60
counter = self._minutes_bytes
elif time_delta < self.MAX_HOURS_TIME_DELTA:
window = 3600
counter = self._hours_bytes
else:
window = 86400
counter = self._days_bytes
elif bandwidth_type == HC.BANDWIDTH_TYPE_REQUESTS:
if time_delta < self.MAX_SECONDS_TIME_DELTA:
window = 0
counter = self._seconds_requests
elif time_delta < self.MAX_MINUTES_TIME_DELTA:
window = 60
counter = self._minutes_requests
elif time_delta < self.MAX_HOURS_TIME_DELTA:
window = 3600
counter = self._hours_requests
else:
window = 86400
counter = self._days_requests
return ( window, counter )
def _GetMonthTime( self, dt ):
( year, month ) = ( dt.year, dt.month )
month_dt = datetime.datetime( year, month, 1 )
month_time = int( calendar.timegm( month_dt.timetuple() ) )
return month_time
def _GetRawUsage( self, bandwidth_type, time_delta ):
if time_delta is None:
dt = self._GetCurrentDateTime()
month_time = self._GetMonthTime( dt )
if bandwidth_type == HC.BANDWIDTH_TYPE_DATA:
return self._months_bytes[ month_time ]
elif bandwidth_type == HC.BANDWIDTH_TYPE_REQUESTS:
return self._months_requests[ month_time ]
( window, counter ) = self._GetWindowAndCounter( bandwidth_type, time_delta )
if time_delta == 1:
# the case of 1 poses a problem as our min block width is also 1. we can't have a window of 0.1s to make the transition smooth
# if we include the last second's data in an effort to span the whole previous 1000ms, we end up not doing anything until the next second rolls over
# this causes 50% consumption as we consume in the second after the one we verified was clear
# so, let's just check the current second and be happy with it
now = HydrusData.GetNow()
if now in counter:
return counter[ now ]
else:
return 0
else:
# we need the 'window' because this tracks brackets from the first timestamp and we want to include if 'since' lands anywhere in the bracket
# e.g. if it is 1200 and we want the past 1,000, we also need the bracket starting at 0, which will include 200-999
search_time_delta = time_delta + window
since = HydrusData.GetNow() - search_time_delta
return sum( ( value for ( timestamp, value ) in counter.items() if timestamp >= since ) )
def _GetTimes( self, dt ):
# collapse each time portion to the latest timestamp it covers
( year, month, day, hour, minute ) = ( dt.year, dt.month, dt.day, dt.hour, dt.minute )
month_dt = datetime.datetime( year, month, 1 )
day_dt = datetime.datetime( year, month, day )
hour_dt = datetime.datetime( year, month, day, hour )
minute_dt = datetime.datetime( year, month, day, hour, minute )
month_time = int( calendar.timegm( month_dt.timetuple() ) )
day_time = int( calendar.timegm( day_dt.timetuple() ) )
hour_time = int( calendar.timegm( hour_dt.timetuple() ) )
minute_time = int( calendar.timegm( minute_dt.timetuple() ) )
second_time = int( calendar.timegm( dt.timetuple() ) )
return ( month_time, day_time, hour_time, minute_time, second_time )
def _GetUsage( self, bandwidth_type, time_delta, for_user ):
if for_user and time_delta is not None and bandwidth_type == HC.BANDWIDTH_TYPE_DATA and time_delta <= self.MIN_TIME_DELTA_FOR_USER:
usage = self._GetWeightedApproximateUsage( time_delta )
else:
usage = self._GetRawUsage( bandwidth_type, time_delta )
self._MaintainCache()
return usage
def _GetWeightedApproximateUsage( self, time_delta ):
SEARCH_DELTA = self.MIN_TIME_DELTA_FOR_USER
counter = self._seconds_bytes
now = HydrusData.GetNow()
since = now - SEARCH_DELTA
valid_keys = [ key for key in counter.keys() if key >= since ]
if len( valid_keys ) == 0:
return 0
# If we want the average speed over past five secs but nothing has happened in sec 4 and 5, we don't want to count them
# otherwise your 1MB/s counts as 200KB/s
earliest_timestamp = min( valid_keys )
SAMPLE_DELTA = max( now - earliest_timestamp, 1 )
total_bytes = sum( ( counter[ key ] for key in valid_keys ) )
time_delta_average_per_sec = total_bytes / SAMPLE_DELTA
return time_delta_average_per_sec * time_delta
def _MaintainCache( self ):
if HydrusData.TimeHasPassed( self._next_cache_maintenance_timestamp ):
now = HydrusData.GetNow()
oldest_second = now - self.MAX_SECONDS_TIME_DELTA
oldest_minute = now - self.MAX_MINUTES_TIME_DELTA
oldest_hour = now - self.MAX_HOURS_TIME_DELTA
oldest_day = now - self.MAX_DAYS_TIME_DELTA
def clear_counter( counter, timestamp ):
bad_keys = [ key for key in counter.keys() if key < timestamp ]
for bad_key in bad_keys:
del counter[ bad_key ]
clear_counter( self._days_bytes, oldest_day )
clear_counter( self._days_requests, oldest_day )
clear_counter( self._hours_bytes, oldest_hour )
clear_counter( self._hours_requests, oldest_hour )
clear_counter( self._minutes_bytes, oldest_minute )
clear_counter( self._minutes_requests, oldest_minute )
clear_counter( self._seconds_bytes, oldest_second )
clear_counter( self._seconds_requests, oldest_second )
self._next_cache_maintenance_timestamp = HydrusData.GetNow() + self.CACHE_MAINTENANCE_TIME_DELTA
def GetCurrentMonthSummary( self ):
with self._lock:
num_bytes = self._GetUsage( HC.BANDWIDTH_TYPE_DATA, None, True )
num_requests = self._GetUsage( HC.BANDWIDTH_TYPE_REQUESTS, None, True )
return 'used ' + HydrusData.ConvertIntToBytes( num_bytes ) + ' in ' + HydrusData.ToHumanInt( num_requests ) + ' requests this month'
def GetMonthlyDataUsage( self ):
with self._lock:
result = []
for ( month_time, usage ) in self._months_bytes.items():
month_dt = datetime.datetime.utcfromtimestamp( month_time )
# this generates zero-padded month, to keep this lexicographically sortable at the gui level
date_str = month_dt.strftime( '%Y-%m' )
result.append( ( date_str, usage ) )
result.sort()
return result
def GetUsage( self, bandwidth_type, time_delta, for_user = False ):
with self._lock:
if time_delta == 0:
return 0
return self._GetUsage( bandwidth_type, time_delta, for_user )
def GetWaitingEstimate( self, bandwidth_type, time_delta, max_allowed ):
with self._lock:
if time_delta is None: # this is monthly
dt = self._GetCurrentDateTime()
( year, month ) = ( dt.year, dt.month )
next_month_year = year
if month == 12:
next_month_year += 1
next_month = ( month % 12 ) + 1
next_month_dt = datetime.datetime( next_month_year, next_month, 1 )
next_month_time = int( calendar.timegm( next_month_dt.timetuple() ) )
return HydrusData.GetTimeDeltaUntilTime( next_month_time )
else:
# we want the highest time_delta at which usage is >= than max_allowed
# time_delta subtract that amount is the time we have to wait for usage to be less than max_allowed
# e.g. if in the past 24 hours there was a bunch of usage 16 hours ago clogging it up, we'll have to wait ~8 hours
( window, counter ) = self._GetWindowAndCounter( bandwidth_type, time_delta )
time_delta_in_which_bandwidth_counts = time_delta + window
time_and_values = counter.items()
time_and_values.sort( reverse = True )
now = HydrusData.GetNow()
usage = 0
for ( timestamp, value ) in time_and_values:
current_search_time_delta = now - timestamp
if current_search_time_delta > time_delta_in_which_bandwidth_counts: # we are searching beyond our time delta. no need to wait
break
usage += value
if usage >= max_allowed:
return time_delta_in_which_bandwidth_counts - current_search_time_delta
return 0
def ReportDataUsed( self, num_bytes ):
with self._lock:
dt = self._GetCurrentDateTime()
( month_time, day_time, hour_time, minute_time, second_time ) = self._GetTimes( dt )
self._months_bytes[ month_time ] += num_bytes
self._days_bytes[ day_time ] += num_bytes
self._hours_bytes[ hour_time ] += num_bytes
self._minutes_bytes[ minute_time ] += num_bytes
self._seconds_bytes[ second_time ] += num_bytes
self._MaintainCache()
def ReportRequestUsed( self ):
with self._lock:
dt = self._GetCurrentDateTime()
( month_time, day_time, hour_time, minute_time, second_time ) = self._GetTimes( dt )
self._months_requests[ month_time ] += 1
self._days_requests[ day_time ] += 1
self._hours_requests[ hour_time ] += 1
self._minutes_requests[ minute_time ] += 1
self._seconds_requests[ second_time ] += 1
self._MaintainCache()
HydrusSerialisable.SERIALISABLE_TYPES_TO_OBJECT_TYPES[ HydrusSerialisable.SERIALISABLE_TYPE_BANDWIDTH_TRACKER ] = BandwidthTracker