nfvbench/traffic_gen/trex.py

   1 # Copyright 2016 Cisco Systems, Inc.  All rights reserved.
   2 #
   3 #    Licensed under the Apache License, Version 2.0 (the "License"); you may
   4 #    not use this file except in compliance with the License. You may obtain
   5 #    a copy of the License at
   6 #
   7 #         http://www.apache.org/licenses/LICENSE-2.0
   8 #
   9 #    Unless required by applicable law or agreed to in writing, software
  10 #    distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
  11 #    WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
  12 #    License for the specific language governing permissions and limitations
  13 #    under the License.
  14
  15 from collections import defaultdict
  16 from itertools import count
  17 from nfvbench.log import LOG
  18 from nfvbench.specs import ChainType
  19 from nfvbench.traffic_server import TRexTrafficServer
  20 from nfvbench.utils import timeout
  21 from nfvbench.utils import TimeoutError
  22 import os
  23 import random
  24 import time
  25 import traceback
  26 from traffic_base import AbstractTrafficGenerator
  27 from traffic_base import TrafficGeneratorException
  28 import traffic_utils as utils
  29
  30
  31 from trex_stl_lib.api import CTRexVmInsFixHwCs
  32 from trex_stl_lib.api import Dot1Q
  33 from trex_stl_lib.api import Ether
  34 from trex_stl_lib.api import IP
  35 from trex_stl_lib.api import STLClient
  36 from trex_stl_lib.api import STLError
  37 from trex_stl_lib.api import STLFlowLatencyStats
  38 from trex_stl_lib.api import STLFlowStats
  39 from trex_stl_lib.api import STLPktBuilder
  40 from trex_stl_lib.api import STLScVmRaw
  41 from trex_stl_lib.api import STLStream
  42 from trex_stl_lib.api import STLTXCont
  43 from trex_stl_lib.api import STLVmFixChecksumHw
  44 from trex_stl_lib.api import STLVmFlowVar
  45 from trex_stl_lib.api import STLVmFlowVarRepetableRandom
  46 from trex_stl_lib.api import STLVmWrFlowVar
  47 from trex_stl_lib.api import UDP
  48 from trex_stl_lib.services.trex_stl_service_arp import STLServiceARP
  49
  50
  51 class TRex(AbstractTrafficGenerator):
  52
  53     LATENCY_PPS = 1000
  54
  55     def __init__(self, runner):
  56         AbstractTrafficGenerator.__init__(self, runner)
  57         self.client = None
  58         self.id = count()
  59         self.latencies = defaultdict(list)
  60         self.stream_ids = defaultdict(list)
  61         self.port_handle = []
  62         self.streamblock = defaultdict(list)
  63         self.rates = []
  64         self.arps = {}
  65
  66     def get_version(self):
  67         return self.client.get_server_version()
  68
  69     def extract_stats(self, in_stats):
  70         utils.nan_replace(in_stats)
  71         LOG.debug(in_stats)
  72
  73         result = {}
  74         for ph in self.port_handle:
  75             stats = self.__combine_stats(in_stats, ph)
  76             result[ph] = {
  77                 'tx': {
  78                     'total_pkts': stats['tx_pkts']['total'],
  79                     'total_pkt_bytes': stats['tx_bytes']['total'],
  80                     'pkt_rate': stats['tx_pps']['total'],
  81                     'pkt_bit_rate': stats['tx_bps']['total']
  82                 },
  83                 'rx': {
  84                     'total_pkts': stats['rx_pkts']['total'],
  85                     'total_pkt_bytes': stats['rx_bytes']['total'],
  86                     'pkt_rate': stats['rx_pps']['total'],
  87                     'pkt_bit_rate': stats['rx_bps']['total'],
  88                     'dropped_pkts': stats['tx_pkts']['total'] - stats['rx_pkts']['total']
  89                 }
  90             }
  91
  92             lat = self.__combine_latencies(in_stats, ph)
  93             result[ph]['rx']['max_delay_usec'] = lat.get('total_max', float('nan'))
  94             result[ph]['rx']['min_delay_usec'] = lat.get('total_min', float('nan'))
  95             result[ph]['rx']['avg_delay_usec'] = lat.get('average', float('nan'))
  96
  97         total_tx_pkts = result[0]['tx']['total_pkts'] + result[1]['tx']['total_pkts']
  98         result["total_tx_rate"] = total_tx_pkts / self.config.duration_sec
  99         return result
 100
 101     def __combine_stats(self, in_stats, port_handle):
 102         """Traverses TRex result dictionary and combines stream stats. Used for combining latency
 103         and regular streams together.
 104         """
 105         result = defaultdict(lambda: defaultdict(float))
 106
 107         for pg_id in [self.stream_ids[port_handle]] + self.latencies[port_handle]:
 108             record = in_stats['flow_stats'][pg_id]
 109             for stat_type, stat_type_values in record.iteritems():
 110                 for ph, value in stat_type_values.iteritems():
 111                     result[stat_type][ph] += value
 112
 113         return result
 114
 115     def __combine_latencies(self, in_stats, port_handle):
 116         """Traverses TRex result dictionary and combines chosen latency stats."""
 117         if not len(self.latencies[port_handle]):
 118             return {}
 119
 120         result = defaultdict(float)
 121         result['total_min'] = float("inf")
 122         for lat_id in self.latencies[port_handle]:
 123             lat = in_stats['latency'][lat_id]
 124             result['dropped_pkts'] += lat['err_cntrs']['dropped']
 125             result['total_max'] = max(lat['latency']['total_max'], result['total_max'])
 126             result['total_min'] = min(lat['latency']['total_min'], result['total_min'])
 127             result['average'] += lat['latency']['average']
 128
 129         result['average'] /= len(self.latencies[port_handle])
 130
 131         return result
 132
 133     def create_pkt(self, stream_cfg, l2frame_size):
 134         # 46 = 14 (Ethernet II) + 4 (CRC Checksum) + 20 (IPv4) + 8 (UDP)
 135         payload = 'x' * (max(64, int(l2frame_size)) - 46)
 136
 137         pkt_base = Ether(src=stream_cfg['mac_src'], dst=stream_cfg['mac_dst'])
 138
 139         if stream_cfg['vlan_tag'] is not None:
 140             pkt_base /= Dot1Q(vlan=stream_cfg['vlan_tag'])
 141
 142         pkt_base /= IP() / UDP()
 143
 144         if stream_cfg['ip_addrs_step'] == 'random':
 145             src_fv = STLVmFlowVarRepetableRandom(
 146                 name="ip_src",
 147                 min_value=stream_cfg['ip_src_addr'],
 148                 max_value=stream_cfg['ip_src_addr_max'],
 149                 size=4,
 150                 seed=random.randint(0, 32767),
 151                 limit=stream_cfg['ip_src_count'])
 152             dst_fv = STLVmFlowVarRepetableRandom(
 153                 name="ip_dst",
 154                 min_value=stream_cfg['ip_dst_addr'],
 155                 max_value=stream_cfg['ip_dst_addr_max'],
 156                 size=4,
 157                 seed=random.randint(0, 32767),
 158                 limit=stream_cfg['ip_dst_count'])
 159         else:
 160             src_fv = STLVmFlowVar(
 161                 name="ip_src",
 162                 min_value=stream_cfg['ip_src_addr'],
 163                 max_value=stream_cfg['ip_src_addr'],
 164                 size=4,
 165                 op="inc",
 166                 step=stream_cfg['ip_addrs_step'])
 167             dst_fv = STLVmFlowVar(
 168                 name="ip_dst",
 169                 min_value=stream_cfg['ip_dst_addr'],
 170                 max_value=stream_cfg['ip_dst_addr_max'],
 171                 size=4,
 172                 op="inc",
 173                 step=stream_cfg['ip_addrs_step'])
 174
 175         vm_param = [
 176             src_fv,
 177             STLVmWrFlowVar(fv_name="ip_src", pkt_offset="IP.src"),
 178             dst_fv,
 179             STLVmWrFlowVar(fv_name="ip_dst", pkt_offset="IP.dst"),
 180             STLVmFixChecksumHw(l3_offset="IP",
 181                                l4_offset="UDP",
 182                                l4_type=CTRexVmInsFixHwCs.L4_TYPE_UDP)
 183         ]
 184
 185         return STLPktBuilder(pkt=pkt_base / payload, vm=STLScVmRaw(vm_param))
 186
 187     def generate_streams(self, port_handle, stream_cfg, l2frame, isg=0.0, latency=True):
 188         idx_lat = None
 189         streams = []
 190         if l2frame == 'IMIX':
 191             for t, (ratio, l2_frame_size) in enumerate(zip(self.imix_ratios, self.imix_l2_sizes)):
 192                 pkt = self.create_pkt(stream_cfg, l2_frame_size)
 193                 streams.append(STLStream(packet=pkt,
 194                                          isg=0.1 * t,
 195                                          flow_stats=STLFlowStats(
 196                                              pg_id=self.stream_ids[port_handle]),
 197                                          mode=STLTXCont(pps=ratio)))
 198
 199             if latency:
 200                 idx_lat = self.id.next()
 201                 sl = STLStream(packet=pkt,
 202                                isg=isg,
 203                                flow_stats=STLFlowLatencyStats(pg_id=idx_lat),
 204                                mode=STLTXCont(pps=self.LATENCY_PPS))
 205                 streams.append(sl)
 206         else:
 207             pkt = self.create_pkt(stream_cfg, l2frame)
 208             streams.append(STLStream(packet=pkt,
 209                                      flow_stats=STLFlowStats(pg_id=self.stream_ids[port_handle]),
 210                                      mode=STLTXCont()))
 211
 212             if latency:
 213                 idx_lat = self.id.next()
 214                 streams.append(STLStream(packet=pkt,
 215                                          flow_stats=STLFlowLatencyStats(pg_id=idx_lat),
 216                                          mode=STLTXCont(pps=self.LATENCY_PPS)))
 217
 218         if latency:
 219             self.latencies[port_handle].append(idx_lat)
 220
 221         return streams
 222
 223     def init(self):
 224         pass
 225
 226     @timeout(5)
 227     def __connect(self, client):
 228         client.connect()
 229
 230     def __connect_after_start(self):
 231         # after start, Trex may take a bit of time to initialize
 232         # so we need to retry a few times
 233         for it in xrange(self.config.generic_retry_count):
 234             try:
 235                 time.sleep(1)
 236                 self.client.connect()
 237                 break
 238             except Exception as ex:
 239                 if it == (self.config.generic_retry_count - 1):
 240                     raise ex
 241                 LOG.info("Retrying connection to TRex (%s)...", ex.message)
 242
 243     def connect(self):
 244         LOG.info("Connecting to TRex...")
 245         server_ip = self.config.generator_config.ip
 246
 247         # Connect to TRex server
 248         self.client = STLClient(server=server_ip)
 249         try:
 250             self.__connect(self.client)
 251         except (TimeoutError, STLError) as e:
 252             if server_ip == '127.0.0.1':
 253                 try:
 254                     self.__start_server()
 255                     self.__connect_after_start()
 256                 except (TimeoutError, STLError) as e:
 257                     LOG.error('Cannot connect to TRex')
 258                     LOG.error(traceback.format_exc())
 259                     logpath = '/tmp/trex.log'
 260                     if os.path.isfile(logpath):
 261                         # Wait for TRex to finish writing error message
 262                         last_size = 0
 263                         for it in xrange(self.config.generic_retry_count):
 264                             size = os.path.getsize(logpath)
 265                             if size == last_size:
 266                                 # probably not writing anymore
 267                                 break
 268                             last_size = size
 269                             time.sleep(1)
 270                         with open(logpath, 'r') as f:
 271                             message = f.read()
 272                     else:
 273                         message = e.message
 274                     raise TrafficGeneratorException(message)
 275             else:
 276                 raise TrafficGeneratorException(e.message)
 277
 278         ports = list(self.config.generator_config.ports)
 279         self.port_handle = ports
 280         # Prepare the ports
 281         self.client.reset(ports)
 282
 283     def set_mode(self):
 284         if self.config.service_chain == ChainType.EXT and not self.config.no_arp:
 285             self.__set_l3_mode()
 286         else:
 287             self.__set_l2_mode()
 288
 289     def __set_l3_mode(self):
 290         self.client.set_service_mode(ports=self.port_handle, enabled=True)
 291         for port, device in zip(self.port_handle, self.config.generator_config.devices):
 292             try:
 293                 self.client.set_l3_mode(port=port,
 294                                         src_ipv4=device.tg_gateway_ip,
 295                                         dst_ipv4=device.dst.gateway_ip,
 296                                         vlan=device.vlan_tag if device.vlan_tagging else None)
 297             except STLError:
 298                 # TRex tries to resolve ARP already, doesn't have to be successful yet
 299                 continue
 300         self.client.set_service_mode(ports=self.port_handle, enabled=False)
 301
 302     def __set_l2_mode(self):
 303         self.client.set_service_mode(ports=self.port_handle, enabled=True)
 304         for port, device in zip(self.port_handle, self.config.generator_config.devices):
 305             for cfg in device.get_stream_configs(self.config.generator_config.service_chain):
 306                 self.client.set_l2_mode(port=port, dst_mac=cfg['mac_dst'])
 307         self.client.set_service_mode(ports=self.port_handle, enabled=False)
 308
 309     def __start_server(self):
 310         server = TRexTrafficServer()
 311         server.run_server(self.config.generator_config)
 312
 313     def resolve_arp(self):
 314         self.client.set_service_mode(ports=self.port_handle)
 315         LOG.info('Polling ARP until successful')
 316         resolved = 0
 317         attempt = 0
 318         for port, device in zip(self.port_handle, self.config.generator_config.devices):
 319             ctx = self.client.create_service_ctx(port=port)
 320
 321             arps = [
 322                 STLServiceARP(ctx,
 323                               src_ip=cfg['ip_src_tg_gw'],
 324                               dst_ip=cfg['mac_discovery_gw'],
 325                               vlan=device.vlan_tag if device.vlan_tagging else None)
 326                 for cfg in device.get_stream_configs(self.config.generator_config.service_chain)
 327             ]
 328
 329             for _ in xrange(self.config.generic_retry_count):
 330                 attempt += 1
 331                 try:
 332                     ctx.run(arps)
 333                 except STLError:
 334                     LOG.error(traceback.format_exc())
 335                     continue
 336
 337                 self.arps[port] = [arp.get_record().dst_mac for arp in arps
 338                                    if arp.get_record().dst_mac is not None]
 339
 340                 if len(self.arps[port]) == self.config.service_chain_count:
 341                     resolved += 1
 342                     LOG.info('ARP resolved successfully for port {}'.format(port))
 343                     break
 344                 else:
 345                     failed = [arp.get_record().dst_ip for arp in arps
 346                               if arp.get_record().dst_mac is None]
 347                     LOG.info('Retrying ARP for: {} ({} / {})'.format(
 348                         failed, attempt, self.config.generic_retry_count))
 349                     time.sleep(self.config.generic_poll_sec)
 350
 351         self.client.set_service_mode(ports=self.port_handle, enabled=False)
 352         return resolved == len(self.port_handle)
 353
 354     def config_interface(self):
 355         pass
 356
 357     def __is_rate_enough(self, l2frame_size, rates, bidirectional, latency):
 358         """Check if rate provided by user is above requirements. Applies only if latency is True."""
 359         intf_speed = self.config.generator_config.intf_speed
 360         if latency:
 361             if bidirectional:
 362                 mult = 2
 363                 total_rate = 0
 364                 for rate in rates:
 365                     r = utils.convert_rates(l2frame_size, rate, intf_speed)
 366                     total_rate += int(r['rate_pps'])
 367             else:
 368                 mult = 1
 369                 total_rate = utils.convert_rates(l2frame_size, rates[0], intf_speed)
 370             # rate must be enough for latency stream and at least 1 pps for base stream per chain
 371             required_rate = (self.LATENCY_PPS + 1) * self.config.service_chain_count * mult
 372             result = utils.convert_rates(l2frame_size,
 373                                          {'rate_pps': required_rate},
 374                                          intf_speed * mult)
 375             result['result'] = total_rate >= required_rate
 376             return result
 377
 378         return {'result': True}
 379
 380     def create_traffic(self, l2frame_size, rates, bidirectional, latency=True):
 381         r = self.__is_rate_enough(l2frame_size, rates, bidirectional, latency)
 382         if not r['result']:
 383             raise TrafficGeneratorException(
 384                 'Required rate in total is at least one of: \n{pps}pps \n{bps}bps \n{load}%.'
 385                 .format(pps=r['rate_pps'],
 386                         bps=r['rate_bps'],
 387                         load=r['rate_percent']))
 388
 389         stream_cfgs = [d.get_stream_configs(self.config.generator_config.service_chain)
 390                        for d in self.config.generator_config.devices]
 391         self.rates = map(lambda rate: utils.to_rate_str(rate), rates)
 392
 393         for ph in self.port_handle:
 394             # generate one pg_id for each direction
 395             self.stream_ids[ph] = self.id.next()
 396
 397         for i, (fwd_stream_cfg, rev_stream_cfg) in enumerate(zip(*stream_cfgs)):
 398             if self.config.service_chain == ChainType.EXT and not self.config.no_arp:
 399                 fwd_stream_cfg['mac_dst'] = self.arps[self.port_handle[0]][i]
 400                 rev_stream_cfg['mac_dst'] = self.arps[self.port_handle[1]][i]
 401
 402             self.streamblock[0].extend(self.generate_streams(self.port_handle[0],
 403                                                              fwd_stream_cfg,
 404                                                              l2frame_size,
 405                                                              latency=latency))
 406             if len(self.rates) > 1:
 407                 self.streamblock[1].extend(self.generate_streams(self.port_handle[1],
 408                                                                  rev_stream_cfg,
 409                                                                  l2frame_size,
 410                                                                  isg=10.0,
 411                                                                  latency=bidirectional and latency))
 412
 413         for ph in self.port_handle:
 414             self.client.add_streams(self.streamblock[ph], ports=ph)
 415             LOG.info('Created traffic stream for port %s.' % ph)
 416
 417     def modify_rate(self, rate, reverse):
 418         port_index = int(reverse)
 419         port = self.port_handle[port_index]
 420         self.rates[port_index] = utils.to_rate_str(rate)
 421         LOG.info('Modified traffic stream for %s, new rate=%s.' % (port, utils.to_rate_str(rate)))
 422
 423     def clear_streamblock(self):
 424         self.streamblock = defaultdict(list)
 425         self.latencies = defaultdict(list)
 426         self.stream_ids = defaultdict(list)
 427         self.rates = []
 428         self.client.reset(self.port_handle)
 429         LOG.info('Cleared all existing streams.')
 430
 431     def get_stats(self):
 432         stats = self.client.get_pgid_stats()
 433         return self.extract_stats(stats)
 434
 435     def get_macs(self):
 436         return [self.client.get_port_attr(port=port)['src_mac'] for port in self.port_handle]
 437
 438     def clear_stats(self):
 439         if self.port_handle:
 440             self.client.clear_stats()
 441
 442     def start_traffic(self):
 443         for port, rate in zip(self.port_handle, self.rates):
 444             self.client.start(ports=port, mult=rate, duration=self.config.duration_sec, force=True)
 445
 446     def stop_traffic(self):
 447         self.client.stop(ports=self.port_handle)
 448
 449     def cleanup(self):
 450         if self.client:
 451             try:
 452                 self.client.reset(self.port_handle)
 453                 self.client.disconnect()
 454             except STLError:
 455                 # TRex does not like a reset while in disconnected state
 456                 pass