3 # Average bandwidth monitoring script. Run periodically via cron(8) to
4 # enforce a soft limit on daily bandwidth usage for each slice. If a
5 # slice is found to have exceeded its daily bandwidth usage when the
6 # script is run, its instantaneous rate will be capped at the desired
7 # average rate. Thus, in the worst case, a slice will only be able to
8 # send a little more than twice its average daily limit.
10 # Two separate limits are enforced, one for destinations exempt from
11 # the node bandwidth cap, and the other for all other destinations.
13 # Mark Huang <mlhuang@cs.princeton.edu>
14 # Andy Bavier <acb@cs.princeton.edu>
15 # Faiyaz Ahmed <faiyaza@cs.princeton.edu>
16 # Copyright (C) 2004-2006 The Trustees of Princeton University
18 # $Id: bwmon.py,v 1.5.2.2 2006/08/21 21:27:35 mlhuang Exp $
38 seconds_per_day = 24 * 60 * 60
44 datafile = "/var/lib/misc/bwmon.dat"
47 # Burst to line rate (or node cap). Set by NM.
48 default_maxrate = bwlimit.get_bwcap()
49 default_maxexemptrate = bwlimit.bwmax
51 # What we cap to when slices break the rules.
52 # 500 Kbit or 5.4 GB per day
53 #default_avgrate = 500000
54 # 1.5 Mbit or 16.4 GB per day
55 #default_avgexemptrate = 1500000
57 # 5.4 Gbyte per day. 5.4 * 1024 k * 1024M * 1024G
58 # 5.4 Gbyte per day max allowed transfered per recording period
59 default_ByteMax = 5798205850
60 default_ByteThresh = int(.8 * default_ByteMax)
61 # 16.4 Gbyte per day max allowed transfered per recording period to I2
62 default_ExemptByteMax = 17609365914
63 default_ExemptByteThresh = int(.8 * default_ExemptByteMax)
67 period = 1 * seconds_per_day
72 The slice %(slice)s has transmitted more than %(bytes)s from
73 %(hostname)s to %(class)s destinations
76 Its maximum %(class)s burst rate will be capped at %(new_maxrate)s
79 Please reduce the average %(class)s transmission rate
80 of the slice %(limit)s per %(period)s.
86 %(date)s %(hostname)s bwcap %(slice)s
91 Stores the last recorded bandwidth parameters of a slice.
93 xid - slice context/VServer ID
95 time - beginning of recording period in UNIX seconds
96 bytes - low bandwidth bytes transmitted at the beginning of the recording period
97 exemptbytes - high bandwidth bytes transmitted at the beginning of the recording period (for I2 -F)
98 ByteMax - total volume of data allowed
99 ByteThresh - After thresh, cap node to (maxbyte - bytes)/(time left in period)
100 ExemptByteMax - Same as above, but for i2.
101 ExemptByteThresh - i2 ByteThresh
102 maxrate - max_rate slice attribute.
103 maxexemptrate - max_exempt_rate slice attribute.
104 self.emailed = did we email during this recording period
108 def __init__(self, xid, name, maxrate, maxexemptrate, bytes, exemptbytes):
114 self.ByteMax = default_ByteMax
115 self.ByteThresh = default_ByteThresh
116 self.ExemptByteMax = default_ExemptByteMax
117 self.ExemptByteThresh = default_ExemptByteThresh
118 self.maxrate = default_maxrate
119 self.maxexemptrate = default_maxexemptrate
122 # Get real values where applicable
123 self.reset(maxrate, maxexemptrate, bytes, exemptbytes)
128 def updateSliceAttributes(self):
129 # Query Node Manager for max rate overrides
131 vals = nm.query(self.name,
132 [('nm_net_max_rate', self.maxrate),
133 ('nm_net_max_exempt_rate', self.maxexemptrate),
134 ("nm_net_max_byte", int(self.ByteMax / 1024)),
135 ("nm_net_max_exempt_byte", int(self.ExemptByteMax / 1024)),
136 ("nm_net_max_thresh_byte", int( .8 * self.ByteMax / 1024)),
137 ("nm_net_max_thresh_exempt_byte", int(.8 * self.ExemptByteMax / 1024)),
138 ("nm_net_avg_rate", 0),
139 ("nm_net_avg_exempt_rate", 0)])
148 avgexemptrate) = vals
150 # The shitty bit. Gotta bias the limits so as not to overflow xmlrpc
151 self.ByteMax = ByteMax * 1024
152 self.ByteThresh = ByteThresh * 1024
153 self.ExemptByteMax = ExemptByteMax * 1024
154 self.ExemptByteThresh = ExemptByteThresh * 1024
156 # The hack here is that when i pass 0 to the xmlrpc request to NM,
157 # for rate limits and it comes back non zero, then screw the byte limits.
158 # Mult by the period and recompute the byte limits. The thought is
159 # If/when PLC switches to byte limits, the avgrates wont be used as
160 # slice attributes and will return as 0
162 self.ByteMax = avgrate * period
163 self.ByteThresh = int(self.ByteMax * .8)
165 if (avgexemptrate != 0):
166 self.ExemptByteMax = avgexemptrate * period
167 self.ExemptByteThresh = int(self.ExemptByteMax * .8)
169 except Exception, err:
170 print "Warning: Exception received while querying NM:", err
172 def reset(self, maxrate, maxexemptrate, bytes, exemptbytes):
174 Begin a new recording period. Remove caps by restoring limits
175 to their default values.
178 # Query Node Manager for max rate overrides
179 self.updateSliceAttributes()
181 # Reset baseline time
182 self.time = time.time()
184 # Reset baseline byte coutns
186 self.exemptbytes = exemptbytes
191 if (self.maxrate != maxrate) or (self.maxexemptrate != maxexemptrate):
192 print "%s reset to %s/%s" % \
194 bwlimit.format_tc_rate(self.maxrate),
195 bwlimit.format_tc_rate(self.maxexemptrate))
196 bwlimit.set(xid = self.xid, maxrate = self.maxrate, maxexemptrate = self.maxexemptrate)
198 def update(self, maxrate, maxexemptrate, bytes, exemptbytes):
200 Update byte counts and check if byte limits have been
204 # Query Node Manager for max rate overrides
205 self.updateSliceAttributes()
208 print("\n%s slice attributes "\
209 "maxrate %s, maxexemptrate %s" % \
211 bwlimit.format_tc_rate(maxrate),
212 bwlimit.format_tc_rate(maxexemptrate)))
214 # Prepare message parameters from the template
216 params = {'slice': self.name, 'hostname': socket.gethostname(),
217 'since': time.asctime(time.gmtime(self.time)) + " GMT",
218 'until': time.asctime(time.gmtime(self.time + period)) + " GMT",
219 'date': time.asctime(time.gmtime()) + " GMT",
220 'period': format_period(period)}
222 if bytes >= (self.bytes + self.ByteThresh):
224 int((self.ByteMax - self.bytes + bytes)/(period - time.time() - self.time))
226 new_maxrate = maxrate
228 # Format template parameters for low bandwidth message
229 params['class'] = "low bandwidth"
230 params['bytes'] = format_bytes(bytes - self.bytes)
231 params['maxrate'] = bwlimit.format_tc_rate(maxrate)
232 params['limit'] = format_bytes(self.ByteMax)
233 params['new_maxrate'] = bwlimit.format_tc_rate(new_maxrate)
236 print "%(slice)s %(class)s " \
237 "%(bytes)s of %(limit)s (%(new_maxrate)s maxrate)" % \
240 # Cap low bandwidth burst rate
241 if new_maxrate != maxrate:
242 message += template % params
243 print "%(slice)s %(class)s capped at %(new_maxrate)s " % params
245 if exemptbytes >= (self.exemptbytes + self.ExemptByteThresh):
246 new_maxexemptrate = \
247 int((self.ExemptByteMax - (self.bytes + bytes))/(period - (time.time() - self.time)))
249 new_maxexemptrate = maxexemptrate
251 # Format template parameters for high bandwidth message
252 params['class'] = "high bandwidth"
253 params['bytes'] = format_bytes(exemptbytes - self.exemptbytes)
254 params['maxrate'] = bwlimit.format_tc_rate(maxexemptrate)
255 params['limit'] = format_bytes(self.ExemptByteMax)
256 params['new_maxexemptrate'] = bwlimit.format_tc_rate(new_maxexemptrate)
259 print "%(slice)s %(class)s " \
260 "%(bytes)s of %(limit)s (%(new_maxrate)s maxrate)" % params
262 # Cap high bandwidth burst rate
263 if new_maxexemptrate != maxexemptrate:
264 message += template % params
265 print "%(slice)s %(class)s capped at %(new_maxexemptrate)s" % params
268 if new_maxrate != maxrate or new_maxexemptrate != maxexemptrate:
269 bwlimit.set(xid = self.xid, maxrate = new_maxrate, maxexemptrate = new_maxexemptrate)
272 if message and self.emailed == False:
273 subject = "pl_mom capped bandwidth of slice %(slice)s on %(hostname)s" % params
276 print message + (footer % params)
279 slicemail(self.name, subject, message + (footer % params))
285 Usage: %s [OPTIONS]...
288 -d, --debug Enable debugging (default: %s)
289 -v, --verbose Increase verbosity level (default: %d)
290 -f, --file=FILE Data file (default: %s)
291 -s, --slice=SLICE Constrain monitoring to these slices (default: all)
292 -p, --period=SECONDS Interval in seconds over which to enforce average byte limits (default: %s)
293 -h, --help This message
294 """.lstrip() % (sys.argv[0], debug, verbose, datafile, format_period(period))
298 global debug, verbose, datafile, period, nm
303 longopts = ["debug", "verbose", "file=", "slice=", "period=", "help"]
304 (opts, argv) = getopt.getopt(sys.argv[1:], "dvf:s:p:h", longopts)
305 except getopt.GetoptError, err:
306 print "Error: " + err.msg
310 for (opt, optval) in opts:
311 if opt == "-d" or opt == "--debug":
313 elif opt == "-v" or opt == "--verbose":
315 bwlimit.verbose = verbose - 1
316 elif opt == "-f" or opt == "--file":
318 elif opt == "-s" or opt == "--slice":
320 elif opt == "-p" or opt == "--period":
326 # Check if we are already running
330 # Redirect stdout and stderr to syslog
331 syslog.openlog("bwmon")
332 sys.stdout = sys.stderr = Logger()
335 f = open(datafile, "r+")
337 print "Loading %s" % datafile
338 (version, slices) = pickle.load(f)
340 # Check version of data file
341 if version != "$Id: bwmon.py,v 1.5.2.2 2006/08/21 21:27:35 mlhuang Exp $":
342 print "Not using old version '%s' data file %s" % (version, datafile)
345 version = "$Id: bwmon.py,v 1.5.2.2 2006/08/21 21:27:35 mlhuang Exp $"
348 # Get special slice IDs
349 root_xid = bwlimit.get_xid("root")
350 default_xid = bwlimit.get_xid("default")
352 #Open connection to Node Manager. Global.
356 # Get actuall running values from tc.
357 for params in bwlimit.get():
360 minexemptrate, maxexemptrate,
361 bytes, exemptbytes) = params
364 # Ignore root and default buckets
365 if xid == root_xid or xid == default_xid:
368 name = bwlimit.get_slice(xid)
370 # Orphaned (not associated with a slice) class
373 # Monitor only the specified slices
374 if names and name not in names:
376 #slices is populated from the pickle file
377 #xid is populated from bwlimit (read from /etc/passwd)
378 if slices.has_key(xid):
380 if time.time() >= (slice.time + period) or \
381 bytes < slice.bytes or exemptbytes < slice.exemptbytes:
382 # Reset to defaults every 24 hours or if it appears
383 # that the byte counters have overflowed (or, more
384 # likely, the node was restarted or the HTB buckets
385 # were re-initialized).
386 slice.reset(maxrate, maxexemptrate, bytes, exemptbytes)
389 slice.update(maxrate, maxexemptrate, bytes, exemptbytes)
391 # New slice, initialize state
392 slice = slices[xid] = Slice(xid, name, maxrate, maxexemptrate, bytes, exemptbytes)
395 dead = Set(slices.keys()) - Set(live)
400 print "Saving %s" % datafile
401 f = open(datafile, "w")
402 pickle.dump((version, slices), f)
407 if __name__ == '__main__':