def add_asn_endpoints(vertex_list, linklist, datafile, count, endpointtype):
"""
Add endpoint nodes that connect to the graph based on ASNs.
@param vertex_list: An instance of VertexList
@type vertex_list: C{VertexList}
@param linklist: A tuple containing a list of links, and an attribute
dictionary
@type linklist: C{tuple}
@param datafile: Path to a datafile that contains at least
two columns called 'Number' and 'ASN'. The 'Number'
column should indicate the relative number of endpoints
that should connect to that ASN.
@type datafile: C{str}
@param count: The number of endpoints to add.
@type count: C{int}
@param endpointtype: The label for the endpoint. 'client' is a
good example.
@type endpointtype: C{str}
"""
r = connection.Redis()
try:
cdata_file = datautil.DataFile(datafile, sep="|")
cdata_file.add_index('ASN')
except IOError:
logging.error(Color.fail("Error reading %s" % datafile))
sys.exit(-1)
attach = dict()
for asn in cdata_file['ASN']:
try:
pop = find_pop_for_asn(asn)
attach[asn] = (pop, cdata_file['ASN'][asn][0]['Number'])
except graph_objects.ASNNotKnown:
pass
def node_id(asn, unique):
return "%s_%s_%s" % (endpointtype, asn, unique)
counter = 0
if len(attach) == 0:
sys.stderr.write(Color.fail(
"[failed] No %s could be attached.\n" % endpointtype))
logging.error(Color.fail(
"[failed] No %s could be attached." % endpointtype))
else:
total = sum(map(lambda x: int(x[1]), attach.itervalues()))
for asn, data in attach.iteritems():
num_to_attach = round(count * (float(data[1]) / float(total)))
for j in xrange(0, int(num_to_attach)):
vertex_list.add_vertex(node_id(asn, j), nodeid=node_id(asn, j),
nodetype=endpointtype, asn=asn)
linkkey = dbkeys.Link.intralink(data[0])
linkdelays = [
delay
for edge in r.smembers(linkkey)
for delay in r.smembers(dbkeys.delay_key(*eval(edge)))]
try:
latency = util.decile_transform(linkdelays)
except util.EmptyListError:
latency = [5 for x in xrange(10)]
linklist.append(EdgeLink(node_id(asn, j),
data[0],
{'latency': latency,
'med_latency': latency[len(latency)/2]}))
counter += 1
log.info(Color.wrapformat("Success [{0} attached]", Color.OKBLUE,
counter))
return (counter, len(attach))
def load_from_redis(r, args):
"""
Create a igraph graph from redis
"""
log.info("Loading from Redis")
linkdict = LinkDict(r)
vertices = VertexList()
tor_vertices = set()
graphlinks = []
graphattrs = dict()
graphattrs['latency'] = []
stats = Stats({'non-pop-trim': int,
'unattachable-poi-count': int,
'poi-latency-defaulted': int,
'unattachable-poi': set,
'num-pois': int,
'num-pops': int,
'num-links': int,
'num-clients': int,
'client-connect-points': int})
pipe = r.pipeline()
i = 0
#Obtain the set of Tor relay IPs
log.info("Reading Tor relays from %s... " % args.pointsofinterest)
try:
with open(args.pointsofinterest) as f:
PoIs = json.load(f)
except IOError as e:
log.info("Error: [%s]" % e)
raise
log.info(Color.wrap("Done", Color.OKBLUE))
log.info("Attaching clients to graph.")
#Add clients
if args.num_clients:
clients_attached, client_attach_points = add_asn_endpoints(
vertices,
graphlinks,
args.client_data,
args.num_clients,
'client')
log.info("Attached {0} clients to {1} attachment points".format(
clients_attached, client_attach_points))
log.info("Attaching destinations to graph.")
#Add dests
if args.num_dests:
dests_attached, dest_attach_points = add_alexa_destinations(
vertices,
graphlinks,
args.num_dests)
log.info("Attached {0} dests to {0} attachment points".format(
dests_attached, dest_attach_points))
protected = set()
protected.update([poi['pop'] for poi in PoIs])
protected.update(vertices.keys())
# We want to trim all of the hanging edges of the graph.
log.info("Trimming degree-1 vertices...")
found_hanging_edge = True
pass_ctr = 0
while found_hanging_edge:
pass_ctr += 1
found_hanging_edge = False
removed = set()
n = 0
timer = ProgressTimer(len(linkdict))
for pop in linkdict.keys():
if n % 100 == 0 or n == timer.total - 1:
timer.tick(100)
sys.stderr.write(
"{0}Pass {1}: {2} {3}".format(
Color.NEWL,
pass_ctr,
Color.wrapformat("[{0} processed, {1} trimmed]",
Color.HEADER,
n, stats['non-pop-trim']),
Color.wrapformat("[eta:{0}]",
Color.OKGREEN,
timer.eta())
))
n += 1
if pop in removed:
continue # we saw this already
if len(linkdict[pop]) >= 2:
continue # it can stay
if pop in protected:
continue # We need relay/client/dest connect point
# It's only connected to one
connected = linkdict[pop].pop()
removed.add(pop)
del linkdict[pop]
linkdict[connected].remove(pop)
if len(linkdict[connected]) == 0:
# This was a matched pair attached to nothing else
del linkdict[connected]
removed.add(connected)
stats.incr('non-pop-trim')
found_hanging_edge = True
sys.stderr.write("\n")
linkdict.collapse_degree_two(protected=protected)
log.info("Trimmed {non-pop-trim} degree two hops".format(**stats))
#Set vertex id's for all of the pops we have links for.
log.info("Adding PoPs...")
for pop in linkdict.iterkeys():
if pop in vertices:
continue # we have this one already.
vertices.add_vertex(pop,
nodeid=pop,
nodetype='pop',
asn=r.get(dbkeys.POP.asn(pop)),
countries=r.smembers(dbkeys.POP.countries(pop)))
stats.incr('num-pops')
i += 1
log.info(Color.wrapformat("Added [{0}]", Color.OKBLUE, stats['num-pops']))
#Attach the relays
for poi in PoIs:
if poi['pop'] not in vertices:
log.warn("Matched relay to {0}, but couldn't find it "
"in vertices".format(poi['pop']))
stats.incr('unattachable-poi-count')
stats.incr('unattachable-poi', poi['id'])
continue
vertices.add_vertex(poi['id'],
nodeid=poi['id'],
nodetype='relay',
**poi)
linkdelays = [
delay
for edge in r.smembers(dbkeys.Link.intralink(poi['pop']))
for delay in r.smembers(dbkeys.delay_key(*eval(edge)))]
try:
deciles = util.decile_transform(linkdelays)
except util.EmptyListError:
deciles = [5 for x in xrange(10)]
stats.incr('poi-latency-defaulted')
graphlinks.append(EdgeLink(poi['id'], poi['pop'],
{'latency': deciles,
'med_latency': deciles[len(deciles)/2]}))
stats.incr('num-pois')
tor_vertices.add(poi['id'])
i += 1
log.info("Added {0} PoIs. Did not attach {1} "
"whose connection point was not linked to anything."
.format(stats['num-pois'],
stats['unattachable-poi-count'])
)
log.info("PoIs defaulted to 5ms links: [{0}]".format(
stats['poi-latency-defaulted']))
pipe.execute()
already_processed = set()
log.info("Processing links... ")
i = 0
for pop1 in linkdict.iterkeys():
if pop1 not in vertices:
continue
for pop2 in linkdict[pop1]:
if (pop2 not in vertices
or dbkeys.Link.interlink(pop1, pop2) in already_processed):
continue
linkkey = dbkeys.Link.interlink(pop1, pop2)
linkdelays = [
delay
for edge in r.smembers(linkkey)
for delay in r.smembers(dbkeys.delay_key(*eval(edge)))]
try:
latency = util.decile_transform(linkdelays)
except util.EmptyListError:
latency = eval(r.get("graph:collapsed:%s" %
(dbkeys.Link.interlink(pop1, pop2))))
graphlinks.append(EdgeLink(pop1, pop2,
{'latency': latency, 'med_latency': latency[len(latency)/2]}))
stats.incr('num-links')
already_processed.add(dbkeys.Link.interlink(pop1, pop2))
i += 1
sys.stderr.write("{0}Processed links for {1} pops"
.format(Color.NEWL, i))
log.info("Processed {0} pop links "
.format(stats['num-links']))
log.info("Making Graph")
with open("vertices.dat", 'w') as vertout:
vertices.write(vertout)
gr = nx.Graph()
gr.add_nodes_from(vertices.nx_tuple_iter())
gr.add_edges_from([edge.nx_tuple() for edge in graphlinks])
try:
bfs_edges = nx.bfs_edges(gr, linkdict.max_degree())
except:
print "Something was wrong with: %s" % linkdict.max_degree()
raise
bfs_node_gen = (node for pair in bfs_edges for node in pair)
subgraph = gr.subgraph(bfs_node_gen)
assert nx.is_connected(subgraph)
log.info("BFS reduced graph from {0} to {1} vertices".format(
len(gr), len(subgraph)))
log.info("Writing data file")
nx.write_graphml(subgraph, args.reload)
log.info("Wrote files")
log.info("STATS:")
for key, val in stats.iteritems():
log.info("{0}: {1}".format(key, val))
return gr
def add_alexa_destinations(vertex_list, linklist, count):
"""
Add potential destination endpoints based on the top 10000 destinations
"""
r = connection.Redis()
aslookup = preprocess.MaxMindGeoIPReader.Instance()
attached = 0
failed = 0
pops = set()
with pkg_resources.resource_stream(
'inettopology_popmap.resources',
'alexa_top_dests.txt') as destlist:
for line in destlist:
if line[0] == '#':
continue
ip, url, matched_ip, matched_bits = line.split()
db_ip_pop = dbkeys.get_pop(matched_ip)
if db_ip_pop is None:
log.debug("Couldn't attach {0} with ip {1}. No matching IP found"
.format(url, matched_ip))
failed += 1
continue
nodeid = "dest_{0}".format(ip.replace('.', '_'))
if nodeid in vertex_list:
continue # Don't add the same url twice
countries = r.smembers(dbkeys.POP.countries(db_ip_pop))
if len(countries) == 1:
country = countries.pop()
else:
country = aslookup.lookup_country_codes(matched_ip)[0]
pops.add(db_ip_pop)
vertex_list.add_vertex(nodeid,
nodeid=nodeid,
nodetype="dest",
url=url,
ip=ip,
asn=r.get(dbkeys.POP.asn(db_ip_pop)),
country=country)
linkkey = dbkeys.Link.intralink(db_ip_pop)
linkdelays = [
delay
for edge in r.smembers(linkkey)
for delay in r.smembers(dbkeys.delay_key(*eval(edge)))]
try:
latency = util.decile_transform(linkdelays)
except util.EmptyListError:
latency = [5 for x in xrange(10)]
linklist.append(
EdgeLink(nodeid,
db_ip_pop,
{'latency': latency,
'med_latency': latency[len(latency)/2]}))
attached += 1
if attached % 10 == 0:
log.info("Attached {0} destinations. Couldn't attach {1}"
.format(attached, failed))
if attached >= count:
break
return (attached, len(pops))
def collapse_degree_two(self, protected=[]):
log.info("Cleaning up collapse dbkeys...")
r = connection.Redis()
p = r.pipeline()
for key in r.keys("graph:collapsed:*"):
p.delete(key)
write_failed(p.execute())
pass_ctr = 0
collapsable = True
ignoreable = set()
clogout = open('collapse.log', 'w')
while collapsable:
pass_ctr += 1
sys.stderr.write("\n")
collapsable = False
degree2nodes = filter(
lambda val: (len(val[1]) == 2 and val[0] not in ignoreable),
self.iteritems())
counter = 0
n = 0
deferred = 0
collapsed = set()
timer = ProgressTimer(len(degree2nodes))
for node, connections in degree2nodes:
if n % 50 == 0 or n == timer.total - 1:
timer.tick(50)
sys.stderr.write(
"{0}Pass {1}: {2} {3}".format(
Color.NEWL, pass_ctr,
Color.wrapformat(
"[{0} processed, {1} collapsed, {2} deferred]",
Color.HEADER, n, counter, deferred
),
Color.wrapformat(
"[eta: {0}]",
Color.OKGREEN, timer.eta()
))
)
n += 1
asns = [r.get(dbkeys.POP.asn(x)) for x in connections | set([node])]
countries = [r.smembers(dbkeys.POP.countries(x))
for x in connections | set([node])]
same_asn = reduce(lambda x, y: x if x == y else False, asns)
same_country = True
for x, y in pairwise(countries):
if x & y != x:
same_country = False
if (same_asn is False or same_country is False or node in protected):
ignoreable.update(connections | set([node]))
continue
if len(collapsed & (connections | set([node]))) != 0:
deferred += 1
continue
collapsed.update(connections | set([node]))
side1 = connections.pop()
side2 = connections.pop()
connections.update(set([side1, side2]))
try:
#side1_delay = median(get_delays(dbkeys.Link.interlink(node, side1)))
side1_delays = decile_transform(
[float(delay)
for edge in r.smembers(dbkeys.Link.interlink(node, side1))
for delay in r.smembers(dbkeys.delay_key(*eval(edge)))])
except:
side1_delays = eval(r.get("graph:collapsed:%s" %
(dbkeys.Link.interlink(node, side1))))
try:
#side2_delay = median(get_delays(dbkeys.Link.interlink(node, side2)))
side2_delays = decile_transform(
[float(delay)
for edge in r.smembers(dbkeys.Link.interlink(node, side2))
for delay in r.smembers(dbkeys.delay_key(*eval(edge)))])
except:
side2_delays = eval(r.get("graph:collapsed:%s" %
(dbkeys.Link.interlink(node, side2))))
combined_delays = [s1 + s2
for s1 in side1_delays
for s2 in side2_delays]
r.set('graph:collapsed:%s'
% (dbkeys.Link.interlink(*list(connections))),
decile_transform(combined_delays))
clogout.write("Collapsed %s <-> %s <-> %s\n" %
(side1, node, side2))
collapsable = True
del self[node]
self[side1].add(side2)
self[side2].add(side1)
self[side1].remove(node)
self[side2].remove(node)
counter += 1
clogout.close()