summaryrefslogtreecommitdiffstats
path: root/scripts/automation
diff options
context:
space:
mode:
authorIdo Barnea <ibarnea@cisco.com>2017-03-27 15:37:21 +0300
committerIdo Barnea <ibarnea@cisco.com>2017-03-30 18:04:15 +0300
commitfa6068394a80766175b59a2b38b4881ec70e7760 (patch)
tree98969a757daf77cd89a0fcfbbe5140d654247c81 /scripts/automation
parent2c839ec5098bf1ad39ec574729891c622a485083 (diff)
Supporting more flows in stateless flow stats, according to NIC type, and running mode
Signed-off-by: Ido Barnea <ibarnea@cisco.com>
Diffstat (limited to 'scripts/automation')
-rw-r--r--scripts/automation/regression/stateless_tests/stl_rx_test.py389
1 files changed, 264 insertions, 125 deletions
diff --git a/scripts/automation/regression/stateless_tests/stl_rx_test.py b/scripts/automation/regression/stateless_tests/stl_rx_test.py
index a71322f5..ea379242 100644
--- a/scripts/automation/regression/stateless_tests/stl_rx_test.py
+++ b/scripts/automation/regression/stateless_tests/stl_rx_test.py
@@ -2,6 +2,7 @@
from .stl_general_test import CStlGeneral_Test, CTRexScenario
from trex_stl_lib.api import *
import os, sys
+import copy
ERROR_LATENCY_TOO_HIGH = 1
@@ -10,79 +11,96 @@ class STLRX_Test(CStlGeneral_Test):
def setUp(self):
per_driver_params = {
- 'net_vmxnet3': {
- 'rate_percent': 1,
- 'total_pkts': 50,
- 'rate_latency': 1,
- 'latency_9k_enable': False,
- },
- 'net_ixgbe': {
- 'rate_percent': 30,
- 'total_pkts': 1000,
- 'rate_latency': 1,
- 'latency_9k_enable': True,
- 'latency_9k_max_average': 300,
- 'latency_9k_max_latency': 400,
- },
- 'net_ixgbe_vf': {
- 'rate_percent': 20,
- 'total_pkts': 1000,
- 'rate_latency': 1,
- 'latency_9k_enable': False,
- },
-
- 'net_i40e': {
- 'rate_percent': 80,
- 'total_pkts': 1000,
- 'rate_latency': 1,
- 'latency_9k_enable': True,
- 'latency_9k_max_average': 100,
- 'latency_9k_max_latency': 250,
- },
- 'net_i40e_vf': {
- 'rate_percent': 10,
- 'total_pkts': 1000,
- 'rate_latency': 1,
- 'latency_9k_enable': False,
- },
- 'net_e1000_igb': {
- 'rate_percent': 80,
- 'total_pkts': 500,
- 'rate_latency': 1,
- 'latency_9k_enable': False,
- },
- 'net_e1000_em': {
- 'rate_percent': 1,
- 'total_pkts': 50,
- 'rate_latency': 1,
- 'latency_9k_enable': False,
- },
- 'net_virtio': {
- 'rate_percent': 1,
- 'total_pkts': 50,
- 'rate_latency': 1,
- 'latency_9k_enable': False,
- 'allow_packets_drop_num': 1, # allow 1 pkt drop
- },
-
- 'net_mlx5': {
- 'rate_percent': 80 if self.is_vf_nics else 5,
- 'total_pkts': 1000,
- 'rate_latency': 1,
- 'latency_9k_enable': False if self.is_vf_nics else True,
- 'latency_9k_max_average': 100,
- 'latency_9k_max_latency': 450, #see latency issue trex-261
- },
-
- 'net_enic': {
- 'rate_percent': 1,
- 'total_pkts': 50,
- 'rate_latency': 1,
- 'latency_9k_enable': False,
- },
-
-
- }
+ 'net_vmxnet3': {
+ 'rate_percent': 1,
+ 'total_pkts': 50,
+ 'rate_latency': 1,
+ 'latency_9k_enable': False,
+ 'no_vlan_even_in_software_mode': True,
+ },
+ 'net_ixgbe': {
+ 'rate_percent': 30,
+ 'total_pkts': 1000,
+ 'rate_latency': 1,
+ 'latency_9k_enable': True,
+ 'latency_9k_max_average': 300,
+ 'latency_9k_max_latency': 400,
+ 'no_vlan': True,
+ 'no_ipv6': True,
+ },
+ 'net_ixgbe_vf': {
+ 'rate_percent': 30,
+ 'total_pkts': 1000,
+ 'rate_latency': 1,
+ 'latency_9k_enable': False,
+ 'no_vlan': True,
+ 'no_ipv6': True,
+ 'no_vlan_even_in_software_mode': True,
+ 'max_pkt_size': 2000, # temporary, until we fix this
+ },
+
+ 'net_i40e': {
+ 'rate_percent': 80,
+ 'rate_percent_soft': 10,
+ 'total_pkts': 1000,
+ 'rate_latency': 1,
+ 'latency_9k_enable': True,
+ 'latency_9k_max_average': 100,
+ 'latency_9k_max_latency': 250,
+ },
+ 'net_i40e_vf': {
+ 'rate_percent': 80,
+ 'rate_percent_soft': 1,
+ 'total_pkts': 1000,
+ 'rate_latency': 1,
+ 'latency_9k_enable': False,
+ 'no_vlan_even_in_software_mode': True,
+ 'max_pkt_size': 2000, # temporary, until we fix this
+ 'allow_packets_drop_num': 5, # todo: fix
+ },
+ 'net_e1000_igb': {
+ 'rate_percent': 80,
+ 'total_pkts': 500,
+ 'rate_latency': 1,
+ 'latency_9k_enable': False,
+
+ },
+ 'net_e1000_em': {
+ 'rate_percent': 1,
+ 'total_pkts': 50,
+ 'rate_latency': 1,
+ 'latency_9k_enable': False,
+ 'no_vlan_even_in_software_mode': True,
+ },
+ 'net_virtio': {
+ 'rate_percent': 1,
+ 'total_pkts': 50,
+ 'rate_latency': 1,
+ 'latency_9k_enable': False,
+ 'allow_packets_drop_num': 1, # allow 1 pkt drop
+ },
+
+ 'net_mlx5': {
+ 'rate_percent': 80,
+ 'total_pkts': 1000,
+ 'rate_latency': 1,
+ 'latency_9k_enable': False if self.is_vf_nics else True,
+ 'latency_9k_max_average': 100,
+ 'latency_9k_max_latency': 450, #see latency issue trex-261
+ 'no_vlan_even_in_software_mode': True, # todo: fix - We see 1 or 2 packet drops from time to time with vlan
+ 'no_ipv6': True, #todo: fix - same issue with ipv6
+ 'allow_packets_drop_num': 15, # todo - remove
+ },
+
+ 'net_enic': {
+ 'rate_percent': 1,
+ 'total_pkts': 50,
+ 'rate_latency': 1,
+ 'latency_9k_enable': False,
+ 'rx_bytes_fix': True,
+ 'no_vlan_even_in_software_mode': True,
+ },
+ }
CStlGeneral_Test.setUp(self)
assert 'bi' in CTRexScenario.stl_ports_map
@@ -99,23 +117,62 @@ class STLRX_Test(CStlGeneral_Test):
self.skip('port {0} does not support RX'.format(self.rx_port))
self.cap = cap
+ self.max_flow_stats = port_info['rx']['counters']
+ if self.max_flow_stats == 1023:
+ # hack - to identify if --software flag was used on server
+ software_mode = True
+ else:
+ software_mode = False
+
+ software_mode = False # fix: need good way to identify software_mode
+
drv_name = port_info['driver']
self.drv_name = drv_name
- if drv_name == 'net_ixgbe':
+ if 'no_vlan' in per_driver_params[drv_name] and not software_mode:
+ self.vlan_support = False
+ else:
+ self.vlan_support = True
+
+ if 'no_ipv6' in per_driver_params[drv_name] and not software_mode:
self.ipv6_support = False
else:
self.ipv6_support = True
+
+ if 'max_pkt_size' in per_driver_params[drv_name]:
+ self.max_pkt_size = per_driver_params[drv_name]['max_pkt_size']
+ else:
+ self.max_pkt_size = 9000
+
self.rate_percent = per_driver_params[drv_name]['rate_percent']
self.total_pkts = per_driver_params[drv_name]['total_pkts']
self.rate_lat = per_driver_params[drv_name].get('rate_latency', self.rate_percent)
+ self.rate_fstat = per_driver_params[drv_name].get('rate_percent_soft', self.rate_percent)
self.latency_9k_enable = per_driver_params[drv_name]['latency_9k_enable']
self.latency_9k_max_average = per_driver_params[drv_name].get('latency_9k_max_average')
self.latency_9k_max_latency = per_driver_params[drv_name].get('latency_9k_max_latency')
self.allow_drop = per_driver_params[drv_name].get('allow_packets_drop_num', 0)
-
self.lat_pps = 1000
self.drops_expected = False
self.c.reset(ports = [self.tx_port, self.rx_port])
+ if 'rx_bytes_fix' in per_driver_params[drv_name] and per_driver_params[drv_name]['rx_bytes_fix'] == True:
+ self.fix_rx_byte_count = True
+ else:
+ self.fix_rx_byte_count = False
+
+ if software_mode:
+ self.qinq_support = True
+ else:
+ self.qinq_support = False
+
+ # hack for enic
+ if 'no_vlan_even_in_software_mode' in per_driver_params[drv_name]:
+ self.vlan_support = False
+ self.qinq_support = False
+
+ #trex25 has router which does not pass vlan
+ if CTRexScenario.setup_name == 'trex25':
+ self.vlan_support = False
+ self.qinq_support = False
vm = STLScVmRaw( [ STLVmFlowVar ( "ip_src", min_value="10.0.0.1",
max_value="10.0.0.255", size=4, step=1,op="inc"),
@@ -128,6 +185,8 @@ class STLRX_Test(CStlGeneral_Test):
);
self.pkt = STLPktBuilder(pkt = Ether()/IP(src="16.0.0.1",dst="48.0.0.1")/UDP(dport=12,sport=1025)/('Your_paylaod_comes_here'))
+ self.vlan_pkt = STLPktBuilder(pkt = Ether()/Dot1Q()/IP(src="16.0.0.1",dst="48.0.0.1")/UDP(dport=12,sport=1025)/('Your_paylaod_comes_here'))
+ self.qinq_pkt = STLPktBuilder(pkt = Ether(type=0x88A8)/Dot1Q(vlan=19)/Dot1Q(vlan=11)/IP(src="16.0.0.1",dst="48.0.0.1")/UDP(dport=12,sport=1025)/('Your_paylaod_comes_here'))
self.ipv6pkt = STLPktBuilder(pkt = Ether()/IPv6(dst="2001:0:4137:9350:8000:f12a:b9c8:2815",src="2001:4860:0:2001::68")
/UDP(dport=12,sport=1025)/('Your_paylaod_comes_here'))
self.large_pkt = STLPktBuilder(pkt = Ether()/IP(src="16.0.0.1",dst="48.0.0.1")/UDP(dport=12,sport=1025)/('a'*1000))
@@ -180,8 +239,13 @@ class STLRX_Test(CStlGeneral_Test):
return 0
+ def __exit_with_error(self, stats, err, pkt_len=0, pkt_type=""):
+ if pkt_len != 0:
+ print("Failed with packet: type {0}, len {1}".format(pkt_type, pkt_len))
+ pprint.pprint(stats)
+ assert False, err
- def __verify_flow (self, pg_id, total_pkts, pkt_len, stats):
+ def __verify_flow (self, pg_id, total_pkts, pkt_len, pkt_type, stats):
flow_stats = stats['flow_stats'].get(pg_id)
latency_stats = stats['latency'].get(pg_id)
@@ -189,6 +253,9 @@ class STLRX_Test(CStlGeneral_Test):
assert False, "no flow stats available"
tx_pkts = flow_stats['tx_pkts'].get(self.tx_port, 0)
+ # for continues tests, we do not know how many packets were sent
+ if total_pkts == 0:
+ total_pkts = tx_pkts
tx_bytes = flow_stats['tx_bytes'].get(self.tx_port, 0)
rx_pkts = flow_stats['rx_pkts'].get(self.rx_port, 0)
if latency_stats is not None:
@@ -199,49 +266,62 @@ class STLRX_Test(CStlGeneral_Test):
stl = latency_stats['err_cntrs']['seq_too_low']
lat = latency_stats['latency']
if ooo != 0 or dup != 0 or stl != 0:
- pprint.pprint(latency_stats)
- tmp='Error packets - dropped:{0}, ooo:{1} dup:{2} seq too high:{3} seq too low:{4}'.format(drops, ooo, dup, sth, stl)
- assert False, tmp
+ self.__exit_with_error(latency_stats,
+ 'Error packets - dropped:{0}, ooo:{1} dup:{2} seq too high:{3} seq too low:{4}'.format(drops, ooo, dup, sth, stl)
+ , pkt_len, pkt_type)
- if (drops > self.allow_drop or sth != 0) and not self.drops_expected:
- pprint.pprint(latency_stats)
- tmp='Error packets - dropped:{0}, ooo:{1} dup:{2} seq too high:{3} seq too low:{4}'.format(drops, ooo, dup, sth, stl)
- assert False, tmp
+ if (drops > self.allow_drop or sth > self.allow_drop) and not self.drops_expected:
+ self.__exit_with_error(latency_stats,
+ 'Error packets - dropped:{0}, ooo:{1} dup:{2} seq too high:{3} seq too low:{4}'.format(drops, ooo, dup, sth, stl)
+ , pkt_len, pkt_type)
if tx_pkts != total_pkts:
pprint.pprint(flow_stats)
- tmp = 'TX pkts mismatch - got: {0}, expected: {1}'.format(tx_pkts, total_pkts)
- assert False, tmp
+ self.__exit_with_error(flow_stats
+ , 'TX pkts mismatch - got: {0}, expected: {1}'.format(tx_pkts, total_pkts)
+ , pkt_len, pkt_type)
if tx_bytes != (total_pkts * pkt_len):
- pprint.pprint(flow_stats)
- tmp = 'TX bytes mismatch - got: {0}, expected: {1}'.format(tx_bytes, (total_pkts * pkt_len))
- assert False, tmp
+ self.__exit_with_error(flow_stats
+ , 'TX bytes mismatch - got: {0}, expected: {1}'.format(tx_bytes, (total_pkts * pkt_len))
+ , pkt_len, pkt_type)
if abs(total_pkts - rx_pkts) > self.allow_drop and not self.drops_expected:
- pprint.pprint(flow_stats)
- tmp = 'RX pkts mismatch - got: {0}, expected: {1}'.format(rx_pkts, total_pkts)
- assert False, tmp
+ self.__exit_with_error(flow_stats
+ , 'RX pkts mismatch - got: {0}, expected: {1}'.format(rx_pkts, total_pkts)
+ , pkt_len, pkt_type)
+
+ rx_pkt_len = pkt_len
+ if self.fix_rx_byte_count:
+ # Patch. Vic card always add vlan, so we should expect 4 extra bytes in each packet
+ rx_pkt_len += 4
if "rx_bytes" in self.cap:
rx_bytes = flow_stats['rx_bytes'].get(self.rx_port, 0)
- if abs(rx_bytes / pkt_len - total_pkts ) > self.allow_drop and not self.drops_expected:
- pprint.pprint(flow_stats)
- tmp = 'RX bytes mismatch - got: {0}, expected: {1}'.format(rx_bytes, (total_pkts * pkt_len))
- assert False, tmp
-
+ if abs(rx_bytes / rx_pkt_len - total_pkts ) > self.allow_drop and not self.drops_expected:
+ self.__exit_with_error(flow_stats
+ , 'RX bytes mismatch - got: {0}, expected: {1}'.format(rx_bytes, (total_pkts * rx_pkt_len))
+ , pkt_len, pkt_type)
# RX itreation
- def __rx_iteration (self, exp_list):
+ def __rx_iteration (self, exp_list, duration=0):
self.c.clear_stats()
- self.c.start(ports = [self.tx_port])
- self.c.wait_on_traffic(ports = [self.tx_port])
+ if duration != 0:
+ self.c.start(ports = [self.tx_port], duration=duration)
+ self.c.wait_on_traffic(ports = [self.tx_port],timeout = duration+10,rx_delay_ms = 100)
+ else:
+ self.c.start(ports = [self.tx_port])
+ self.c.wait_on_traffic(ports = [self.tx_port])
stats = self.c.get_stats()
for exp in exp_list:
- self.__verify_flow(exp['pg_id'], exp['total_pkts'], exp['pkt_len'], stats)
+ if 'pkt_type' in exp:
+ pkt_type = exp['pkt_type']
+ else:
+ pkt_type = "not specified"
+ self.__verify_flow(exp['pg_id'], exp['total_pkts'], exp['pkt_len'], pkt_type, stats)
# one stream on TX --> RX
@@ -271,42 +351,97 @@ class STLRX_Test(CStlGeneral_Test):
def test_multiple_streams(self):
+ self._test_multiple_streams(False)
+
+ def test_multiple_streams_random(self):
+ if self.drv_name == 'net_mlx5' or self.drv_name == 'net_i40e_vf':
+ self.skip('Not running on Mellanox cards currently')
+ self._test_multiple_streams(True)
+
+ def _test_multiple_streams(self, is_random):
if self.is_virt_nics:
self.skip('Skip this for virtual NICs')
- num_latency_streams = 128
- num_flow_stat_streams = 127
+ if is_random:
+ num_latency_streams = random.randint(1, 128);
+ num_flow_stat_streams = random.randint(1, self.max_flow_stats);
+ all_pkts = [self.pkt]
+ if self.ipv6_support:
+ all_pkts.append(self.ipv6pkt)
+ if self.vlan_support:
+ all_pkts.append(self.vlan_pkt)
+ if self.qinq_support:
+ all_pkts.append(self.qinq_pkt)
+ else:
+ num_latency_streams = 128
+ num_flow_stat_streams = self.max_flow_stats
total_pkts = int(self.total_pkts / (num_latency_streams + num_flow_stat_streams))
if total_pkts == 0:
total_pkts = 1
- percent = float(self.rate_lat) / (num_latency_streams + num_flow_stat_streams)
+ percent_lat = float(self.rate_lat) / num_latency_streams
+ percent_fstat = float(self.rate_fstat) / num_flow_stat_streams
+
+ print("num_latency_streams:{0}".format(num_latency_streams))
+ if is_random:
+ print(" total percent:{0} ({1} per stream)".format(percent_lat * num_latency_streams, percent_lat))
+
+ print("num_flow_stat_streams:{0}".format(num_flow_stat_streams))
+ if is_random:
+ print(" total percent:{0} ({1} per stream)".format(percent_fstat * num_flow_stat_streams, percent_fstat))
try:
streams = []
exp = []
- # 10 identical streams
for pg_id in range(1, num_latency_streams):
+ if is_random:
+ pkt = copy.deepcopy(all_pkts[random.randint(0, len(all_pkts) - 1)])
+ pkt.set_packet(pkt.pkt / ('a' * random.randint(0, self.max_pkt_size - len(pkt.pkt))))
+ send_mode = STLTXCont(percentage = percent_lat)
+ else:
+ pkt = self.pkt
+ send_mode = STLTXSingleBurst(total_pkts = total_pkts+pg_id, percentage = percent_lat)
+
streams.append(STLStream(name = 'rx {0}'.format(pg_id),
- packet = self.pkt,
+ packet = pkt,
flow_stats = STLFlowLatencyStats(pg_id = pg_id),
- mode = STLTXSingleBurst(total_pkts = total_pkts+pg_id, percentage = percent)))
+ mode = send_mode))
- exp.append({'pg_id': pg_id, 'total_pkts': total_pkts+pg_id, 'pkt_len': streams[-1].get_pkt_len()})
+ if is_random:
+ exp.append({'pg_id': pg_id, 'total_pkts': 0, 'pkt_len': streams[-1].get_pkt_len()
+ , 'pkt_type': pkt.pkt.sprintf("%Ether.type%")})
+ else:
+ exp.append({'pg_id': pg_id, 'total_pkts': total_pkts+pg_id, 'pkt_len': streams[-1].get_pkt_len()})
for pg_id in range(num_latency_streams + 1, num_latency_streams + num_flow_stat_streams):
+ if is_random:
+ pkt = copy.deepcopy(all_pkts[random.randint(0, len(all_pkts) - 1)])
+ pkt.set_packet(pkt.pkt / ('a' * random.randint(0, self.max_pkt_size - len(pkt.pkt))))
+ send_mode = STLTXCont(percentage = percent_fstat)
+ else:
+ pkt = self.pkt
+ send_mode = STLTXSingleBurst(total_pkts = total_pkts+pg_id, percentage = percent_fstat)
+
streams.append(STLStream(name = 'rx {0}'.format(pg_id),
- packet = self.pkt,
+ packet = pkt,
flow_stats = STLFlowStats(pg_id = pg_id),
- mode = STLTXSingleBurst(total_pkts = total_pkts+pg_id, percentage = percent)))
-
- exp.append({'pg_id': pg_id, 'total_pkts': total_pkts+pg_id, 'pkt_len': streams[-1].get_pkt_len()})
+ mode = send_mode))
+ if is_random:
+ exp.append({'pg_id': pg_id, 'total_pkts': 0, 'pkt_len': streams[-1].get_pkt_len()
+ , 'pkt_type': pkt.pkt.sprintf("%Ether.type%")})
+ else:
+ exp.append({'pg_id': pg_id, 'total_pkts': total_pkts+pg_id, 'pkt_len': streams[-1].get_pkt_len()})
# add both streams to ports
self.c.add_streams(streams, ports = [self.tx_port])
- self.__rx_iteration(exp)
+ if is_random:
+ duration = 60
+ print("Duration: {0}".format(duration))
+ else:
+ duration = 0
+ self.__rx_iteration(exp, duration = duration)
except STLError as e:
@@ -321,7 +456,15 @@ class STLRX_Test(CStlGeneral_Test):
{'name': 'Latency, no field engine', 'pkt': self.pkt, 'lat': True},
{'name': 'Latency, short packet with field engine', 'pkt': self.vm_pkt, 'lat': True},
{'name': 'Latency, large packet field engine', 'pkt': self.vm_large_pkt, 'lat': True}
+
+ # add here more stream types
]
+ if self.vlan_support:
+ streams_data.append({'name': 'Flow stat with vlan. No latency', 'pkt': self.vlan_pkt, 'lat': False})
+
+ if self.qinq_support:
+ streams_data.append({'name': 'Flow stat qinq. No latency', 'pkt': self.qinq_pkt, 'lat': False})
+
if self.latency_9k_enable:
streams_data.append({'name': 'Latency, 9k packet with field engine', 'pkt': self.vm_9k_pkt, 'lat': True})
@@ -410,10 +553,7 @@ class STLRX_Test(CStlGeneral_Test):
assert False , '{0}'.format(e)
-
-
-
- # check low latency when you have stream of 9K stream
+ # Verify that there is low latency with random packet size,duration and ports
def test_9k_stream(self):
if self.is_virt_nics:
self.skip('Skip this for virtual NICs')
@@ -428,8 +568,6 @@ class STLRX_Test(CStlGeneral_Test):
pgid=random.randint(1, 65000);
pkt_size=random.randint(1000, 9000);
all_ports = list(CTRexScenario.stl_ports_map['map'].keys());
-
-
s_port=random.sample(all_ports, random.randint(1, len(all_ports)) )
s_port=sorted(s_port)
@@ -515,7 +653,7 @@ class STLRX_Test(CStlGeneral_Test):
self.check_stats(ops["obytes"], bytes, "stats[%s][obytes]" % c_port)
self.check_stats(ops["opackets"], pkts, "stats[%s][opackets]" % c_port)
-
+
self.check_stats(ips["ibytes"], bytes, "stats[%s][ibytes]" % s_port)
self.check_stats(ips["ipackets"], pkts, "stats[%s][ipackets]" % s_port)
@@ -523,10 +661,10 @@ class STLRX_Test(CStlGeneral_Test):
ls = stats['flow_stats'][5 + c_port]
self.check_stats(ls['rx_pkts']['total'], pkts, "ls['rx_pkts']['total']")
self.check_stats(ls['rx_pkts'][s_port], pkts, "ls['rx_pkts'][%s]" % s_port)
-
+
self.check_stats(ls['tx_pkts']['total'], pkts, "ls['tx_pkts']['total']")
self.check_stats(ls['tx_pkts'][c_port], pkts, "ls['tx_pkts'][%s]" % c_port)
-
+
self.check_stats(ls['tx_bytes']['total'], bytes, "ls['tx_bytes']['total']")
self.check_stats(ls['tx_bytes'][c_port], bytes, "ls['tx_bytes'][%s]" % c_port)
@@ -556,9 +694,11 @@ class STLRX_Test(CStlGeneral_Test):
# this test sends 1 64 byte packet with latency and check that all counters are reported as 64 bytes
def test_fcs_stream(self):
-
- # in case of VM and vSwitch there are drop of packets in some cases, let retry number of times
- # in this case we just want to check functionality that packet of 64 is reported as 64 in all levels
+ if CTRexScenario.setup_name == 'trex19':
+ # todo: - fix
+ self.skip('Test does not run on trex19')
+ # In case of VM and vSwitch there is packet drop in some cases, so we retry few of times
+ # In this case we just want to check that packet of 64 bytes is reported correctly as 64 everywhere.
is_vm = self.is_virt_nics or self.is_vf_nics
tries = 1
@@ -639,4 +779,3 @@ class STLRX_Test(CStlGeneral_Test):
exp = {'pg_id': 5, 'total_pkts': total_pkts, 'pkt_len': s1.get_pkt_len()}
self.__rx_iteration( [exp] )
-