%PDF- %PDF-
Direktori : /usr/lib/python3/dist-packages/twisted/protocols/ |
Current File : //usr/lib/python3/dist-packages/twisted/protocols/htb.py |
# -*- test-case-name: twisted.test.test_htb -*- # Copyright (c) Twisted Matrix Laboratories. # See LICENSE for details. """ Hierarchical Token Bucket traffic shaping. Patterned after U{Martin Devera's Hierarchical Token Bucket traffic shaper for the Linux kernel<http://luxik.cdi.cz/~devik/qos/htb/>}. @seealso: U{HTB Linux queuing discipline manual - user guide <http://luxik.cdi.cz/~devik/qos/htb/manual/userg.htm>} @seealso: U{Token Bucket Filter in Linux Advanced Routing & Traffic Control HOWTO<http://lartc.org/howto/lartc.qdisc.classless.html#AEN682>} """ # TODO: Investigate whether we should be using os.times()[-1] instead of # time.time. time.time, it has been pointed out, can go backwards. Is # the same true of os.times? from time import time from typing import Optional from zope.interface import Interface, implementer from twisted.protocols import pcp class Bucket: """ Implementation of a Token bucket. A bucket can hold a certain number of tokens and it drains over time. @cvar maxburst: The maximum number of tokens that the bucket can hold at any given time. If this is L{None}, the bucket has an infinite size. @type maxburst: C{int} @cvar rate: The rate at which the bucket drains, in number of tokens per second. If the rate is L{None}, the bucket drains instantaneously. @type rate: C{int} """ maxburst: Optional[int] = None rate: Optional[int] = None _refcount = 0 def __init__(self, parentBucket=None): """ Create a L{Bucket} that may have a parent L{Bucket}. @param parentBucket: If a parent Bucket is specified, all L{add} and L{drip} operations on this L{Bucket} will be applied on the parent L{Bucket} as well. @type parentBucket: L{Bucket} """ self.content = 0 self.parentBucket = parentBucket self.lastDrip = time() def add(self, amount): """ Adds tokens to the L{Bucket} and its C{parentBucket}. This will add as many of the C{amount} tokens as will fit into both this L{Bucket} and its C{parentBucket}. @param amount: The number of tokens to try to add. @type amount: C{int} @returns: The number of tokens that actually fit. @returntype: C{int} """ self.drip() if self.maxburst is None: allowable = amount else: allowable = min(amount, self.maxburst - self.content) if self.parentBucket is not None: allowable = self.parentBucket.add(allowable) self.content += allowable return allowable def drip(self): """ Let some of the bucket drain. The L{Bucket} drains at the rate specified by the class variable C{rate}. @returns: C{True} if the bucket is empty after this drip. @returntype: C{bool} """ if self.parentBucket is not None: self.parentBucket.drip() if self.rate is None: self.content = 0 else: now = time() deltaTime = now - self.lastDrip deltaTokens = deltaTime * self.rate self.content = max(0, self.content - deltaTokens) self.lastDrip = now return self.content == 0 class IBucketFilter(Interface): def getBucketFor(*somethings, **some_kw): """ Return a L{Bucket} corresponding to the provided parameters. @returntype: L{Bucket} """ @implementer(IBucketFilter) class HierarchicalBucketFilter: """ Filter things into buckets that can be nested. @cvar bucketFactory: Class of buckets to make. @type bucketFactory: L{Bucket} @cvar sweepInterval: Seconds between sweeping out the bucket cache. @type sweepInterval: C{int} """ bucketFactory = Bucket sweepInterval: Optional[int] = None def __init__(self, parentFilter=None): self.buckets = {} self.parentFilter = parentFilter self.lastSweep = time() def getBucketFor(self, *a, **kw): """ Find or create a L{Bucket} corresponding to the provided parameters. Any parameters are passed on to L{getBucketKey}, from them it decides which bucket you get. @returntype: L{Bucket} """ if (self.sweepInterval is not None) and ( (time() - self.lastSweep) > self.sweepInterval ): self.sweep() if self.parentFilter: parentBucket = self.parentFilter.getBucketFor(self, *a, **kw) else: parentBucket = None key = self.getBucketKey(*a, **kw) bucket = self.buckets.get(key) if bucket is None: bucket = self.bucketFactory(parentBucket) self.buckets[key] = bucket return bucket def getBucketKey(self, *a, **kw): """ Construct a key based on the input parameters to choose a L{Bucket}. The default implementation returns the same key for all arguments. Override this method to provide L{Bucket} selection. @returns: Something to be used as a key in the bucket cache. """ return None def sweep(self): """ Remove empty buckets. """ for key, bucket in self.buckets.items(): bucket_is_empty = bucket.drip() if (bucket._refcount == 0) and bucket_is_empty: del self.buckets[key] self.lastSweep = time() class FilterByHost(HierarchicalBucketFilter): """ A Hierarchical Bucket filter with a L{Bucket} for each host. """ sweepInterval = 60 * 20 def getBucketKey(self, transport): return transport.getPeer()[1] class FilterByServer(HierarchicalBucketFilter): """ A Hierarchical Bucket filter with a L{Bucket} for each service. """ sweepInterval = None def getBucketKey(self, transport): return transport.getHost()[2] class ShapedConsumer(pcp.ProducerConsumerProxy): """ Wraps a C{Consumer} and shapes the rate at which it receives data. """ # Providing a Pull interface means I don't have to try to schedule # traffic with callLaters. iAmStreaming = False def __init__(self, consumer, bucket): pcp.ProducerConsumerProxy.__init__(self, consumer) self.bucket = bucket self.bucket._refcount += 1 def _writeSomeData(self, data): # In practice, this actually results in obscene amounts of # overhead, as a result of generating lots and lots of packets # with twelve-byte payloads. We may need to do a version of # this with scheduled writes after all. amount = self.bucket.add(len(data)) return pcp.ProducerConsumerProxy._writeSomeData(self, data[:amount]) def stopProducing(self): pcp.ProducerConsumerProxy.stopProducing(self) self.bucket._refcount -= 1 class ShapedTransport(ShapedConsumer): """ Wraps a C{Transport} and shapes the rate at which it receives data. This is a L{ShapedConsumer} with a little bit of magic to provide for the case where the consumer it wraps is also a C{Transport} and people will be attempting to access attributes this does not proxy as a C{Consumer} (e.g. C{loseConnection}). """ # Ugh. We only wanted to filter IConsumer, not ITransport. iAmStreaming = False def __getattr__(self, name): # Because people will be doing things like .getPeer and # .loseConnection on me. return getattr(self.consumer, name) class ShapedProtocolFactory: """ Dispense C{Protocols} with traffic shaping on their transports. Usage:: myserver = SomeFactory() myserver.protocol = ShapedProtocolFactory(myserver.protocol, bucketFilter) Where C{SomeServerFactory} is a L{twisted.internet.protocol.Factory}, and C{bucketFilter} is an instance of L{HierarchicalBucketFilter}. """ def __init__(self, protoClass, bucketFilter): """ Tell me what to wrap and where to get buckets. @param protoClass: The class of C{Protocol} this will generate wrapped instances of. @type protoClass: L{Protocol<twisted.internet.interfaces.IProtocol>} class @param bucketFilter: The filter which will determine how traffic is shaped. @type bucketFilter: L{HierarchicalBucketFilter}. """ # More precisely, protoClass can be any callable that will return # instances of something that implements IProtocol. self.protocol = protoClass self.bucketFilter = bucketFilter def __call__(self, *a, **kw): """ Make a C{Protocol} instance with a shaped transport. Any parameters will be passed on to the protocol's initializer. @returns: A C{Protocol} instance with a L{ShapedTransport}. """ proto = self.protocol(*a, **kw) origMakeConnection = proto.makeConnection def makeConnection(transport): bucket = self.bucketFilter.getBucketFor(transport) shapedTransport = ShapedTransport(transport, bucket) return origMakeConnection(shapedTransport) proto.makeConnection = makeConnection return proto