-## Copyright 2002 Andrew Loewenstern, All Rights Reserved
+## Copyright 2002-2003 Andrew Loewenstern, All Rights Reserved
+# see LICENSE.txt for license information
from const import reactor
+import const
+
import time
+from sha import sha
+
from ktable import KTable, K
from knode import KNode as Node
-from hash import newID
+from khash import newID, newIDInRange
+
+from actions import FindNode, GetValue, KeyExpirer, StoreValue
+import krpc
-from actions import FindNode, GetValue
-from twisted.web import xmlrpc
from twisted.internet.defer import Deferred
+from twisted.internet import protocol
from twisted.python import threadable
-from twisted.internet.app import Application
+from twisted.application import service, internet
from twisted.web import server
threadable.init()
+import sys
-from bsddb3 import db ## find this at http://pybsddb.sf.net/
-from bsddb3._db import DBNotFoundError
-
-# don't ping unless it's been at least this many seconds since we've heard from a peer
-MAX_PING_INTERVAL = 60 * 15 # fifteen minutes
-
+import sqlite ## find this at http://pysqlite.sourceforge.net/
+class KhashmirDBExcept(Exception):
+ pass
# this is the main class!
-class Khashmir(xmlrpc.XMLRPC):
- __slots__ = ['listener', 'node', 'table', 'store', 'app']
- def __init__(self, host, port):
- self.node = Node(newID(), host, port)
- self.table = KTable(self.node)
- self.app = Application("xmlrpc")
- self.app.listenTCP(port, server.Site(self))
- self.store = db.DB()
- self.store.open(None, None, db.DB_BTREE)
-
-
- def render(self, request):
- """
- Override the built in render so we can have access to the request object!
- note, crequest is probably only valid on the initial call (not after deferred!)
- """
- self.crequest = request
- return xmlrpc.XMLRPC.render(self, request)
+class Khashmir(protocol.Factory):
+ __slots__ = ('listener', 'node', 'table', 'store', 'app', 'last', 'protocol')
+ def __init__(self, host, port, db='khashmir.db'):
+ self.setup(host, port, db)
+
+ def setup(self, host, port, db='khashmir.db'):
+ self._findDB(db)
+ self.port = port
+ self.node = self._loadSelfNode(host, port)
+ self.table = KTable(self.node)
+ self.app = service.Application("krpc")
+ self.udp = krpc.hostbroker(self)
+ self.udp.protocol = krpc.KRPC
+ self.listenport = reactor.listenUDP(port, self.udp)
+ self.last = time.time()
+ self._loadRoutingTable()
+ KeyExpirer(store=self.store)
+ #self.refreshTable(force=1)
+ reactor.callLater(60, self.checkpoint, (1,))
+
+ def __del__(self):
+ self.listenport.stopListening()
+
+ def _loadSelfNode(self, host, port):
+ c = self.store.cursor()
+ c.execute('select id from self where num = 0;')
+ if c.rowcount > 0:
+ id = c.fetchone()[0]
+ else:
+ id = newID()
+ return Node().init(id, host, port)
+
+ def _saveSelfNode(self):
+ self.store.autocommit = 0
+ c = self.store.cursor()
+ c.execute('delete from self where num = 0;')
+ c.execute("insert into self values (0, %s);", sqlite.encode(self.node.id))
+ self.store.commit()
+ self.store.autocommit = 1
+
+ def checkpoint(self, auto=0):
+ self._saveSelfNode()
+ self._dumpRoutingTable()
+ if auto:
+ reactor.callLater(const.CHECKPOINT_INTERVAL, self.checkpoint)
+
+ def _findDB(self, db):
+ import os
+ try:
+ os.stat(db)
+ except OSError:
+ self._createNewDB(db)
+ else:
+ self._loadDB(db)
+
+ def _loadDB(self, db):
+ try:
+ self.store = sqlite.connect(db=db)
+ self.store.autocommit = 1
+ except:
+ import traceback
+ raise KhashmirDBExcept, "Couldn't open DB", traceback.exc_traceback
+
+ def _createNewDB(self, db):
+ self.store = sqlite.connect(db=db)
+ self.store.autocommit = 1
+ s = """
+ create table kv (key binary, value binary, time timestamp, primary key (key, value));
+ create index kv_key on kv(key);
+ create index kv_timestamp on kv(time);
+
+ create table nodes (id binary primary key, host text, port number);
+
+ create table self (num number primary key, id binary);
+ """
+ c = self.store.cursor()
+ c.execute(s)
+
+ def _dumpRoutingTable(self):
+ """
+ save routing table nodes to the database
+ """
+ self.store.autocommit = 0;
+ c = self.store.cursor()
+ c.execute("delete from nodes where id not NULL;")
+ for bucket in self.table.buckets:
+ for node in bucket.l:
+ c.execute("insert into nodes values (%s, %s, %s);", (sqlite.encode(node.id), node.host, node.port))
+ self.store.commit()
+ self.store.autocommit = 1;
+
+ def _loadRoutingTable(self):
+ """
+ load routing table nodes from database
+ it's usually a good idea to call refreshTable(force=1) after loading the table
+ """
+ c = self.store.cursor()
+ c.execute("select * from nodes;")
+ for rec in c.fetchall():
+ n = Node().initWithDict({'id':rec[0], 'host':rec[1], 'port':int(rec[2])})
+ n.conn = self.udp.connectionForAddr((n.host, n.port))
+ self.table.insertNode(n, contacted=0)
+
-
#######
####### LOCAL INTERFACE - use these methods!
- def addContact(self, host, port):
- """
- ping this node and add the contact info to the table on pong!
- """
- n =Node(" "*20, host, port) # note, we
- self.sendPing(n)
-
+ def addContact(self, host, port, callback=None):
+ """
+ ping this node and add the contact info to the table on pong!
+ """
+ n =Node().init(const.NULL_ID, host, port)
+ n.conn = self.udp.connectionForAddr((n.host, n.port))
+ self.sendPing(n, callback=callback)
## this call is async!
def findNode(self, id, callback, errback=None):
- """ returns the contact info for node, or the k closest nodes, from the global table """
- # get K nodes out of local table/cache, or the node we want
- nodes = self.table.findNodes(id)
- d = Deferred()
- d.addCallbacks(callback, errback)
- if len(nodes) == 1 and nodes[0].id == id :
- d.callback(nodes)
- else:
- # create our search state
- state = FindNode(self, id, d.callback)
- reactor.callFromThread(state.goWithNodes, nodes)
+ """ returns the contact info for node, or the k closest nodes, from the global table """
+ # get K nodes out of local table/cache, or the node we want
+ nodes = self.table.findNodes(id)
+ d = Deferred()
+ if errback:
+ d.addCallbacks(callback, errback)
+ else:
+ d.addCallback(callback)
+ if len(nodes) == 1 and nodes[0].id == id :
+ d.callback(nodes)
+ else:
+ # create our search state
+ state = FindNode(self, id, d.callback)
+ reactor.callFromThread(state.goWithNodes, nodes)
## also async
- def valueForKey(self, key, callback):
- """ returns the values found for key in global table """
- nodes = self.table.findNodes(key)
- # create our search state
- state = GetValue(self, key, callback)
- reactor.callFromThread(state.goWithNodes, nodes)
-
-
- ## async, but in the current implementation there is no guarantee a store does anything so there is no callback right now
- def storeValueForKey(self, key, value):
- """ stores the value for key in the global table, returns immediately, no status
- in this implementation, peers respond but don't indicate status to storing values
- values are stored in peers on a first-come first-served basis
- this will probably change so more than one value can be stored under a key
- """
- def _storeValueForKey(nodes, key=key, value=value, response= self._storedValueHandler, default= lambda t: "didn't respond"):
- for node in nodes:
- if node.id != self.node.id:
- df = node.storeValue(key, value, self.node.senderDict())
- df.addCallbacks(response, default)
- # this call is asynch
- self.findNode(key, _storeValueForKey)
-
-
- def insertNode(self, n):
- """
- insert a node in our local table, pinging oldest contact in bucket, if necessary
-
- If all you have is a host/port, then use addContact, which calls this method after
- receiving the PONG from the remote node. The reason for the seperation is we can't insert
- a node into the table without it's peer-ID. That means of course the node passed into this
- method needs to be a properly formed Node object with a valid ID.
- """
- old = self.table.insertNode(n)
- if old and (time.time() - old.lastSeen) > MAX_PING_INTERVAL and old.id != self.node.id:
- # the bucket is full, check to see if old node is still around and if so, replace it
-
- ## these are the callbacks used when we ping the oldest node in a bucket
- def _staleNodeHandler(oldnode=old, newnode = n):
- """ called if the pinged node never responds """
- self.table.replaceStaleNode(old, newnode)
-
- def _notStaleNodeHandler(sender, old=old):
- """ called when we get a ping from the remote node """
- if sender['id'] == old.id:
- self.table.insertNode(old)
-
- df = old.ping()
- df.addCallbacks(_notStaleNodeHandler, self._staleNodeHandler)
-
-
- def sendPing(self, node):
- """
- ping a node
- """
- df = node.ping(self.node.senderDict())
- ## these are the callbacks we use when we issue a PING
- def _pongHandler(sender, id=node.id, host=node.host, port=node.port, table=self.table):
- if id != 20 * ' ' and id != sender['id']:
- # whoah, got response from different peer than we were expecting
- pass
- else:
- #print "Got PONG from %s at %s:%s" % (`msg['id']`, t.target.host, t.target.port)
- n = Node(sender['id'], host, port)
- table.insertNode(n)
- return
- def _defaultPong(err):
- # this should probably increment a failed message counter and dump the node if it gets over a threshold
- return
-
- df.addCallbacks(_pongHandler,_defaultPong)
-
-
- def findCloseNodes(self):
- """
- This does a findNode on the ID one away from our own.
- This will allow us to populate our table with nodes on our network closest to our own.
- This is called as soon as we start up with an empty table
- """
- id = self.node.id[:-1] + chr((ord(self.node.id[-1]) + 1) % 256)
- def callback(nodes):
- pass
- self.findNode(id, callback)
-
- def refreshTable(self):
- """
-
- """
- def callback(nodes):
- pass
-
- for bucket in self.table.buckets:
- if time.time() - bucket.lastAccessed >= 60 * 60:
- id = randRange(bucket.min, bucket.max)
- self.findNode(id, callback)
-
-
- #####
- ##### INCOMING MESSAGE HANDLERS
-
- def xmlrpc_ping(self, sender):
- """
- takes sender dict = {'id', <id>, 'port', port} optional keys = 'ip'
- returns sender dict
- """
- ip = self.crequest.getClientIP()
- n = Node(sender['id'], ip, sender['port'])
- self.insertNode(n)
- return self.node.senderDict()
-
- def xmlrpc_find_node(self, target, sender):
- nodes = self.table.findNodes(target)
- nodes = map(lambda node: node.senderDict(), nodes)
- ip = self.crequest.getClientIP()
- n = Node(sender['id'], ip, sender['port'])
- self.insertNode(n)
- return nodes, self.node.senderDict()
-
- def xmlrpc_store_value(self, key, value, sender):
- if not self.store.has_key(key):
- self.store.put(key, value)
- ip = self.crequest.getClientIP()
- n = Node(sender['id'], ip, sender['port'])
- self.insertNode(n)
- return self.node.senderDict()
-
- def xmlrpc_find_value(self, key, sender):
- ip = self.crequest.getClientIP()
- n = Node(sender['id'], ip, sender['port'])
- self.insertNode(n)
- if self.store.has_key(key):
- return {'values' : self.store[key]}, self.node.senderDict()
- else:
- nodes = self.table.findNodes(key)
- nodes = map(lambda node: node.senderDict(), nodes)
- return {'nodes' : nodes}, self.node.senderDict()
-
- ###
- ### message response callbacks
- # called when we get a response to store value
- def _storedValueHandler(self, sender):
- pass
-
-
-
-
-
-
-#------ testing
-
-def test_build_net(quiet=0, peers=256, pause=1):
- from whrandom import randrange
- import thread
- port = 2001
- l = []
+ def valueForKey(self, key, callback, searchlocal = 1):
+ """ returns the values found for key in global table
+ callback will be called with a list of values for each peer that returns unique values
+ final callback will be an empty list - probably should change to 'more coming' arg
+ """
+ nodes = self.table.findNodes(key)
- if not quiet:
- print "Building %s peer table." % peers
-
- for i in xrange(peers):
- a = Khashmir('localhost', port + i)
- l.append(a)
-
-
- thread.start_new_thread(l[0].app.run, ())
- time.sleep(1)
- for peer in l[1:]:
- peer.app.run()
- #time.sleep(.25)
-
- print "adding contacts...."
-
- for peer in l[1:]:
- n = l[randrange(0, len(l))].node
- peer.addContact(n.host, n.port)
- n = l[randrange(0, len(l))].node
- peer.addContact(n.host, n.port)
- n = l[randrange(0, len(l))].node
- peer.addContact(n.host, n.port)
- if pause:
- time.sleep(.30)
-
- time.sleep(5)
- print "finding close nodes...."
-
- for peer in l:
- peer.findCloseNodes()
- if pause:
- time.sleep(1)
-# for peer in l:
-# peer.refreshTable()
- return l
+ # get locals
+ if searchlocal:
+ l = self.retrieveValues(key)
+ if len(l) > 0:
+ reactor.callLater(0, callback, (l))
+ else:
+ l = []
+
+ # create our search state
+ state = GetValue(self, key, callback)
+ reactor.callFromThread(state.goWithNodes, nodes, l)
+
+ ## async, callback indicates nodes we got a response from (but no guarantee they didn't drop it on the floor)
+ def storeValueForKey(self, key, value, callback=None):
+ """ stores the value for key in the global table, returns immediately, no status
+ in this implementation, peers respond but don't indicate status to storing values
+ a key can have many values
+ """
+ def _storeValueForKey(nodes, key=key, value=value, response=callback , table=self.table):
+ if not response:
+ # default callback
+ def _storedValueHandler(sender):
+ pass
+ response=_storedValueHandler
+ action = StoreValue(self.table, key, value, response)
+ reactor.callFromThread(action.goWithNodes, nodes)
+
+ # this call is asynch
+ self.findNode(key, _storeValueForKey)
-def test_find_nodes(l, quiet=0):
- import threading, sys
- from whrandom import randrange
- flag = threading.Event()
- n = len(l)
+ def insertNode(self, n, contacted=1):
+ """
+ insert a node in our local table, pinging oldest contact in bucket, if necessary
+
+ If all you have is a host/port, then use addContact, which calls this method after
+ receiving the PONG from the remote node. The reason for the seperation is we can't insert
+ a node into the table without it's peer-ID. That means of course the node passed into this
+ method needs to be a properly formed Node object with a valid ID.
+ """
+ old = self.table.insertNode(n, contacted=contacted)
+ if old and (time.time() - old.lastSeen) > const.MIN_PING_INTERVAL and old.id != self.node.id:
+ # the bucket is full, check to see if old node is still around and if so, replace it
+
+ ## these are the callbacks used when we ping the oldest node in a bucket
+ def _staleNodeHandler(oldnode=old, newnode = n):
+ """ called if the pinged node never responds """
+ self.table.replaceStaleNode(old, newnode)
+
+ def _notStaleNodeHandler(dict, old=old):
+ """ called when we get a pong from the old node """
+ dict = dict['rsp']
+ if dict['id'] == old.id:
+ self.table.justSeenNode(old.id)
+
+ df = old.ping(self.node.id)
+ df.addCallbacks(_notStaleNodeHandler, _staleNodeHandler)
+
+ def sendPing(self, node, callback=None):
+ """
+ ping a node
+ """
+ df = node.ping(self.node.id)
+ ## these are the callbacks we use when we issue a PING
+ def _pongHandler(dict, node=node, table=self.table, callback=callback):
+ _krpc_sender = dict['_krpc_sender']
+ dict = dict['rsp']
+ sender = {'id' : dict['id']}
+ if node.id != const.NULL_ID and node.id != sender['id']:
+ # whoah, got response from different peer than we were expecting
+ self.table.invalidateNode(node)
+ else:
+ sender['host'] = _krpc_sender[0]
+ sender['port'] = _krpc_sender[1]
+ n = Node().initWithDict(sender)
+ n.conn = self.udp.connectionForAddr((n.host, n.port))
+ table.insertNode(n)
+ if callback:
+ callback()
+ def _defaultPong(err, node=node, table=self.table, callback=callback):
+ table.nodeFailed(node)
+ if callback:
+ callback()
+
+ df.addCallbacks(_pongHandler,_defaultPong)
+
+ def findCloseNodes(self, callback=lambda a: None):
+ """
+ This does a findNode on the ID one away from our own.
+ This will allow us to populate our table with nodes on our network closest to our own.
+ This is called as soon as we start up with an empty table
+ """
+ id = self.node.id[:-1] + chr((ord(self.node.id[-1]) + 1) % 256)
+ self.findNode(id, callback)
+
+ def refreshTable(self, force=0):
+ """
+ force=1 will refresh table regardless of last bucket access time
+ """
+ def callback(nodes):
+ pass
- a = l[randrange(0,n)]
- b = l[randrange(0,n)]
+ for bucket in self.table.buckets:
+ if force or (time.time() - bucket.lastAccessed >= const.BUCKET_STALENESS):
+ id = newIDInRange(bucket.min, bucket.max)
+ self.findNode(id, callback)
+
+
+ def retrieveValues(self, key):
+ c = self.store.cursor()
+ c.execute("select value from kv where key = %s;", sqlite.encode(key))
+ t = c.fetchone()
+ l = []
+ while t:
+ l.append(t['value'])
+ t = c.fetchone()
+ return l
- def callback(nodes, l=l, flag=flag):
- if (len(nodes) >0) and (nodes[0].id == b.node.id):
- print "test_find_nodes PASSED"
- else:
- print "test_find_nodes FAILED"
- flag.set()
- a.findNode(b.node.id, callback)
- flag.wait()
+ #####
+ ##### INCOMING MESSAGE HANDLERS
-def test_find_value(l, quiet=0):
- from whrandom import randrange
- from sha import sha
- import time, threading, sys
+ def krpc_ping(self, id, _krpc_sender):
+ sender = {'id' : id}
+ sender['host'] = _krpc_sender[0]
+ sender['port'] = _krpc_sender[1]
+ n = Node().initWithDict(sender)
+ n.conn = self.udp.connectionForAddr((n.host, n.port))
+ self.insertNode(n, contacted=0)
+ return {"id" : self.node.id}
+
+ def krpc_find_node(self, target, id, _krpc_sender):
+ nodes = self.table.findNodes(target)
+ nodes = map(lambda node: node.senderDict(), nodes)
+ sender = {'id' : id}
+ sender['host'] = _krpc_sender[0]
+ sender['port'] = _krpc_sender[1]
+ n = Node().initWithDict(sender)
+ n.conn = self.udp.connectionForAddr((n.host, n.port))
+ self.insertNode(n, contacted=0)
+ return {"nodes" : nodes, "id" : self.node.id}
+
+ def krpc_store_value(self, key, value, id, _krpc_sender):
+ t = "%0.6f" % time.time()
+ c = self.store.cursor()
+ try:
+ c.execute("insert into kv values (%s, %s, %s);", (sqlite.encode(key), sqlite.encode(value), t))
+ except sqlite.IntegrityError, reason:
+ # update last insert time
+ c.execute("update kv set time = %s where key = %s and value = %s;", (t, sqlite.encode(key), sqlite.encode(value)))
+ sender = {'id' : id}
+ sender['host'] = _krpc_sender[0]
+ sender['port'] = _krpc_sender[1]
+ n = Node().initWithDict(sender)
+ n.conn = self.udp.connectionForAddr((n.host, n.port))
+ self.insertNode(n, contacted=0)
+ return {"id" : self.node.id}
- fa = threading.Event()
- fb = threading.Event()
- fc = threading.Event()
+ def krpc_find_value(self, key, id, _krpc_sender):
+ sender = {'id' : id}
+ sender['host'] = _krpc_sender[0]
+ sender['port'] = _krpc_sender[1]
+ n = Node().initWithDict(sender)
+ n.conn = self.udp.connectionForAddr((n.host, n.port))
+ self.insertNode(n, contacted=0)
- n = len(l)
- a = l[randrange(0,n)]
- b = l[randrange(0,n)]
- c = l[randrange(0,n)]
- d = l[randrange(0,n)]
+ l = self.retrieveValues(key)
+ if len(l) > 0:
+ return {'values' : l, "id": self.node.id}
+ else:
+ nodes = self.table.findNodes(key)
+ nodes = map(lambda node: node.senderDict(), nodes)
+ return {'nodes' : nodes, "id": self.node.id}
- key = sha(`randrange(0,100000)`).digest()
- value = sha(`randrange(0,100000)`).digest()
- if not quiet:
- print "inserting value...",
- sys.stdout.flush()
- a.storeValueForKey(key, value)
- time.sleep(3)
- print "finding..."
-
- def mc(flag, value=value):
- def callback(values, f=flag, val=value):
- try:
- if(len(values) == 0):
- print "find FAILED"
- else:
- if values != val:
- print "find FAILED"
- else:
- print "find FOUND"
- finally:
- f.set()
- return callback
- b.valueForKey(key, mc(fa))
- fa.wait()
- c.valueForKey(key, mc(fb))
- fb.wait()
- d.valueForKey(key, mc(fc))
- fc.wait()
-
-if __name__ == "__main__":
- l = test_build_net()
- time.sleep(3)
- print "finding nodes..."
- test_find_nodes(l)
- test_find_nodes(l)
- test_find_nodes(l)
- print "inserting and fetching values..."
- test_find_value(l)
- test_find_value(l)
- test_find_value(l)
- test_find_value(l)
- test_find_value(l)
- test_find_value(l)