Age | Commit message (Collapse) | Author | Files | Lines |
|
Change-Id: I571289c744441a6114c6d45ac6cbd7cfc0dafef2
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
|
|
|
|
Change-Id: Ic384c4603e7482f150fd4c4d2d6a6448a45ddb9e
Signed-off-by: Trevor Tao <trevor.tao@arm.com>
|
|
|
|
|
|
To avoid the following ERROR when netperf test between
2 different subnets:
ERROR netperf: send_omni: send_data failed: Network is unreachable
For detail, please see:
https://serverfault.com/questions/802320/netperf-iptables-\
masquerade-network-unreachable
Or:
https://stackoverflow.com/questions/11981480/error-in-running-\
netperf-udp-stream-over-openvpn
Change-Id: I62b202844861440deaf3bf0f65b41561bd87ae87
Signed-off-by: Trevor Tao <trevor.tao@arm.com>
|
|
Instead of using a key_filename for Heat, we can
read the key as a string directly using pkg_resources.resource_string()
This will enable us to save Heat stacks as pod.yaml, because
we can embedded the key into the pod.yaml directly.
Change-Id: I16baaba17dab845ee0846f97678733bae33cb463
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Edward MacGillivray <edward.s.macgillivray@intel.com>
|
|
Also rename private to uplink, public to downlink
for scale-out template we need to count from 0
so we can use range() without +1/-1 errors
vnf_0, vnf_1
tg_0, tg_1
also fix Ixia defaults
Change-Id: I6aecfbb95f99af20f012a9df19c19be77d1b5b77
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Edward MacGillivray <edward.s.macgillivray@intel.com>
|
|
we assume the time it takes to start multiple
instances is proportional to the number of instances
so we scale the timeout based on the number of instances.
Change-Id: I6901890d3f184ac4e38e1d6823b96c291579e04a
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
We want to generate pod.yaml from Heat contexts so we can
re-use the context without destroying it.
But we don't have node role information and it doesn't
make sense in this case, so make the role optional.
Since we changed Heat to use pkey instead of key_filename,
we can embed the pkey into the pod.yaml, but we have
to make sure to convert the pkey to string, in case
it is a RSAKey object
Change-Id: Ibefcfbd8236e68013a704c39964cb870da825da8
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
|
|
|
|
Add a new PortPair class to resolve the
topology into list of public and private ports.
Before we were calculating public/private in multiple
locations and using different conventions.
In addition for all the DPDK test we need to use the DPDK
port number and no rely on interface ordering or interface naming
conventions.
We used to use xe0 -> 0, xe1 -> 1, etc. This is not the DPDK port
number.
Use the new dpdknicbind_helper class to parse the output of
dpdk-devbind.py to find the actual DPDK port number at runtime.
We then use this DPDK port number to correctly calculate the
port_mask_hex.
The port mask maps the DPDK port num (PMD ID) to the LINK ID
used in the pipeline config
We also need to make sure we only use the interfaces matched to the
topology and not use all the interfaces, because in some cases we will
have unused interfaces. In particular TRex always requires an even
number of interfaces, so for single port TRex tests we have to create
the second port and not use it.
Thus we had to modify the traffic generator stats code to only dump
stats for used ports and no unused ports.
Ixia was using interface ordering to map to Ixia ports, instead we use
the dpdk_port_num which must be hardcoded for Ixia.
Renamed traffic_profile.execute to traffic_profile.execute_traffic so
we can trace the code easier.
We pass the port used by the traffic profile to generate_samples so we
don't get stats for unused ports.
Fixed up vPE config creation and bring up issues.
Fixed up CGNAPT and UDP_Replay to work correctly.
Tested with 4-port scale-out
Change-Id: I2e4f328bff2904108081e92a4bf712333fa73869
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Edward MacGillivray <edward.s.macgillivray@intel.com>
|
|
Previsouly we added all servers to every network
in Heat in a full mesh.
To more closely replicate test topology and to limit
then number of ports we need to all each server
to specify which ports should be connected to each network.
This should also allow for some kind of multiport setup.
Add optional network_ports dict to each server with network to port_list
mapping
match inteface based on port name or vld_id
replace vld_id matching with network name matching, since network_name == vld_id
Change-Id: I5de46b8f673949e3c17d8df6fa96f055c43886ce
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
Change-Id: I744353f631cf1771d75f750543e8612f81be71ee
Signed-off-by: rexlee8776 <limingjiang@huawei.com>
|
|
Change-Id: I3ec1a6d3710d44df5ddac6bd8967d28ad58e8d33
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
|
|
|
|
- added common method to get relative paths
- added 'Ixia' APP_NAME
Change-Id: I7966798bab71af66d3efbeb1e13b07e8fbb41e88
Signed-off-by: Maciej Skrocki <maciej.skrocki@intel.com>
|
|
If we have a /32 or for some reason find a range of IPs
to use, we can default to the single IP specified on the interface.
Change-Id: Ieaa1d57b04e1d57e8cef344d5a53bbca05e7887f
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
return super result
Change-Id: I723a37281da15c1887ae1b3cf91d7e957b1924d1
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
|
|
- path should be defined via TREX_CLIENT_LIB environmental variable e.g. TREX_CLIENT_LIB=/opt/trex_client/stl
- refactored unit tests
Change-Id: I18767e48daf774432c010f1b88d18a4f0ee4e156
Signed-off-by: Maciej Skrocki <maciej.skrocki@intel.com>
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
|
|
|
|
This patch adds the framesize and flow into test option instead of
adding a seperate file to avoid the multiple file update incase of ip
change.
Change-Id: Ic473c73773ad36422ecc02618b8c646a5336b70a
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
|
|
Change-Id: I6649960dc9fbc61c22c9b7434805fc335634960b
Signed-off-by: rexlee8776 <limingjiang@huawei.com>
|
|
fix exception handling
Change-Id: I46fd0dbf21f3d3308e4f38fcfae5df2716ad7190
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
Change-Id: Ic1f13c0d28c1a1b01bbf3c8a6a618a5b3ab5bbeb
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
There are multiple issues wiht YAML loading.
1. Jinja2 renders None values as a string 'None'. This is not valid YAML
we need to render None values to '~' or 'null' which is the native YAML
None value.
2. Jinja2 renders dict and lists that contain unicode with
u'foo' values. This is not value YAML syntax.
Because we are serializing dict and lists into YAML, we
need to encode them as valid YAML. We can override Jinja2 finalize to
use yaml.dump to dump inline YAML.
We use yaml.safe_dump(elem, default_flow_style=True).replace('\n', '')
to generate valid single-line YAML dict and list values.
But this problem highlights the general difficulties with templating and
loading files.
We could avoid this Python->Jinja2->YAML->Python issue by directly
injecting the list or dict after the YAML is loaded.
I'm not sure of the real utility of these templates.
3. On Python 2 YAML loader is rendering all strings
as unicode. This does not work for Trex because Trex is broken
and badly coded. Trex does type checking against str() which
is different for Python 2 and Python 3.
The default YAML loader will return native string types, str() or unicode()
for Python 2 and Python 3 respectively.
The bad Trex codes is in convert_val:
https://github.com/cisco-system-traffic-generator/trex-core/blob/master/scripts/automation/trex_control_plane/stl/trex_stl_lib/trex_stl_packet_builder_scapy.py#L674
def convert_val (val):
if is_integer(val):
return val
if type(val) == str:
return ipv4_str_to_num (is_valid_ipv4(val))
raise CTRexPacketBuildException(-11,("init val invalid %s ") % val );
This code is doing type(val) == str. This is bad and broken.
We can't fix Trex, so we have to render all strings as native str() types
The bug here was that the Heat template loader template_format.py
was overriding the global YAML loader to always return unicode.
We don't want this global override.
To fix this we have to use local subclasses of the yaml.SafeLoader
class.
But in order to dynamically subclass from CSafeLoader or SafeLoader
we have to use the type() builtin to define a new class at runtime.
Once we have new classes defined, we can safely isolate different
YAML constructors and return unicode or not depending on the case.
To be consistent we implement a new yaml_loader.py module to centralize
all non-Heat template yaml loading to ensure correct uncode/str
conversion
Change-Id: Iebf9cf78fbda390977c390436b0869e7bbf503eb
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
untill we fix the sequence issue, we need to make the instantiate
blocking call to avoid hangs
Change-Id: Ie337372625d506e6b3c74a5248204a87fec1ca2c
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
|
|
Problem:
Running Vsperf in Tgen mode is supported but the integration is not complete at the code level
i.e. not ready-to-use, and dpdk loopback is not supported inside the VM.
Solution:
(1) Completely automates VM image generation and supports 1G huge pages.
(2) Adds a new test scenario VsperfDPDK for testpmd based loopback inside the VM.
Update 1-2: Fixed "line too long" issues not reported by local run_tests.sh (why?)
Update 3: Per comment change to use SSH.from_node() and add unit test cases
Update 4: Add more unit test cases for coverage and ready the code for merge
JIRA: YARDSTICK-661
Change-Id: Iea3014d4c83e1b0c079019a4ed27771d40a7eed8
Signed-off-by: Jing Zhang <jing.c.zhang@nokia.com>
|
|
JIRA: YARDSTICK-771
Change-Id: Ibcd2228505d341feb09b0d477e5f4ed6062c1e89
Signed-off-by: rexlee8776 <limingjiang@huawei.com>
|
|
remove unused docstring params
remove whitespace
fix sudo vnf_build command
ignore stdout
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
Change-Id: I6fcdbb0ed4bc6b7fc2f557529f5ff9fd960c05e9
|
|
Refactored main NSB VNF classes accroding to class diagram
https://wiki.opnfv.org/display/yardstick/NSB+class+diagram
All the SampleVNFs have been separated and placed under
the SampleVNF class.
Added AutoConnectSSH to automatically create SSH conneciton on demand.
Added VnfdHelper class to wrap the VNFD dictionary in prepartion for
class-based modeling.
Extracted DpdkVnfSetupEnvHelper for DPDK based VNF setup.
Extracted Stats and other client config to ResourceHelper
Had to replace dict_key_flatten with deepgetitem due to Python 2.7
Jinja2 infinite recursion.
Change-Id: Ia8840e9c44cdbdf39aab6b02e6d2176b31937dc9
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Edward MacGillivray <edward.s.macgillivray@intel.com>
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
apexlake is unmaintained, so remove it
From some reason orchestrator/heat.py started failing
so fixup those unittests
Change-Id: Ie06508b5ab7c9dcf9fdfca83e173a188a894d564
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
|
|
JIRA: YARDSTICK-755
The parameters that iperf3 testcase support now maybe insufficient in some usecase.
The test case should support parameters such as: windows size and length of buffer.
This work is about adding more parameters that can be specified in options section in the YAML file.
Change-Id: Ibc81cd1bbbc6fdd492aa67bd3acee7cd310c87ef
Signed-off-by: Ace Lee <liyin11@huawei.com>
|
|
Problem:
Neither OVS nor SRIOV multi-queue is not supported. Guest VM lacks tuning to reach high throughput.
Solution:
(1) Build SRIOV multi-queue capable guest image by recompiling the igxbevf driver (make CFLAGS_EXTRA=-DIXGBE_ENABLE_VF_MQ install).
(2) Change pktgen to send on multiple queues.
(3) Add tuning in guest VM (Disable irqbalance and setup vNIC interrupt affinity to vCPUs)
Update 1: Takes care comments plus adds unit test cases
Update 2: Jenkins reports code coverage 81%, local reports 96%, add more unit test cases
Update 3: Manually rebased to adapt to SSH.from_node()
Update 4: Takes care comments for vnic_type, vnic_name
Change-Id: Ieb15381c653b13697487d095efa4be6c3c49fa42
JIRA: YARDSTICK-619
Signed-off-by: Jing Zhang <jing.c.zhang@nokia.com>
|
|
we need to know which network each port is connected to
so we can find VLAN or VXLAN ID.
To do this we implement a new method for Contexts,
Context.get_network(). This method is similar to
Context.get_server(), it searches for a given
network name in all the contexts.
From this we generate a context_cfg["networks"]
dict that stores all the network info for the nodes
in the scenario.
Then when we generate the topology for VNFD, we can
lookup a given network by the vld_id and get the
network_type, segmentation_id, etc.
Then if we need to for example generated
traffic on a given VLAN or VXLAN, we have this
info available.
Define default nd_route_tbl for ACL VNF
we need default empty nd_route_tbl for IPv6 route.
Change-Id: I9f9cfbd6acabeb4ae4675ca7354390efa57b29e7
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Edward MacGillivray <edward.s.macgillivray@intel.com>
|
|
So it can easily be used by other testcase to unify result
JIRA: YARDSTICK-702
Change-Id: Id4fde38a9a0c2a87a6c870bdb7b0c8f3a3b371ac
Signed-off-by: rexlee8776 <limingjiang@huawei.com>
|
|
|
|
If we run sample/ping-hot.yaml, it will encounter an AttributeError, log
see below:
Traceback (most recent call last):
File
"/usr/local/lib/python2.7/dist-packages/yardstick/benchmark/runners/duration.py",
line 69, in _worker_process
method(data)
File
"/usr/local/lib/python2.7/dist-packages/yardstick/benchmark/scenarios/networking/ping.py",
line 94, in run
target_vm_name = target_vm.split('.')[0]
AttributeError: 'dict' object has no attribute 'split'
Because here host and target will be a dict
JIRA: YARDSTICK-561
Change-Id: I4b7628bf20050d6d516a80efe3785f750d27c05e
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
First we add mac_address, subnet_cidr to Heat template outputs
Then we convert those into a form for NSB and add vld_id.
NSB also requires PCI Bus ID, kernel driver
and dpdk_port_num. We get this by ssh-ing
into instance and dumping sysfs
We also need to fix allow for ssh key auth,
and implement relative path file loading
so NSB can find all its YAML files
JIRA: YARDSTICK-580
Change history:
don't hide heat create tracebacks we need tracebacks for debug
vnf_generic: add task_path to scenario so we can load relative paths
for vnf_generic we want to be able to load yaml relative to the
task path
For example:
traffic_profile: ../../traffic_profiles/fixed.yaml
topology: ping_tg_topology.yaml # TODO: look in relative path where the tc.yaml is found
These need to be relative to samples/vnf_samples/nsut/ping/tc_ping_heat_context.yaml
Add a scenario["task_path"] entry
heat: log actual exception
vnf_generic: replace list with set and iterate over values()
some general refactors to remove redundact lookups and
type conversions
heat: provide mac_address, device_id and network_id from outputs
We may need more information to dynamically
determine test topology.
Towards this end return more info in the heat template.
We can return mac_address, device_id and network_id.
Once we have this info we can add it to the context_cfg
as an interfaces dict.
add sample vnf ping multi-network test
this test requires 3 network, one for mgmt
and the other two for NSB traffic tests
We have to make sure we don't use DPDK
on mgmt interface because DPDK unbinds
the driver
heat: convert networks to OrderedDict
so we can lookups networks as well as
iterate over them in consisitent order
heat: and vld_id to networks for vnf_generic
vnf_generic uses vld_id Virtual Link Descriptor ID
to identify interfaces
Add the key to the networks dict
and store in Networks object
implement relative path file loading in vnf_generic
in multiple places we need to load a file
relative to the task path, so add
open_relative_file_path
and modify load_vnf_model to include the scenario_cfg
parameter so we have access to task_path
DRAFT: heat timeout support
Heat stack in CI job failed due to some Nova issue.
But then apparently yardstick kept running and took 180mins to timeout
https://build.opnfv.org/ci/view/bottlenecks/job/bottlenecks-compass-posca_stress_ping-baremetal-daily-master/16/console
We can add a Heat create timeout and fail faster if there is an error.
The question is how long should we wait for a Heat stack to deploy. We
can set a default and allow override in the heat context config, if
users make complicated stacks
heat: get netmask and gateway from heat outputs
we have do some tricky business with finding
the subnet cidr and converting it into netmask
vnf_generic: get vpci, driver and dpdk_port_num
use a big old find command to dump all the sysfs
netdev info nicely. This was re-used from autotest FCoE tests.
r"""find /sys/devices/pci* -type d -name net -exec sh -c '{ grep -sH ^ \
+$1/ifindex $1/address $1/operstate $1/device/vendor $1/device/device \
+$1/device/subsystem_vendor $1/device/subsystem_device ; \
+printf "%s/driver:" $1 ; basename $(readlink -s $1/device/driver); } \
+' sh \{\}/* \;
This finds all PCI devices that are network devices, then
dumps all the relevant info using /bin/sh.
Then we parse this into a 'netdevs' dict inside the node_dict
and also convert into VNF fields we need.
vnf_generic: set node name for kpis
node is a dict, so we have to use node_name
vnfdgen: we CANNOT use TaskTemplate.render
because it does not allow for missing variables, we need to allow
password for key_filename to be undefined
remove default ssh password hack, once rendering is fixed
add new example tc_external_ping_heat_context
Change-Id: If1fe0c1a2ab0a5be17e40790a66f28f706fa44d6
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Edward MacGillivray <edward.s.macgillivray@intel.com>
|
|
Need a fast path inside VM to verify full throughput of SRIOV and OVS-dpdk.
Update 1: Change newly added file names to avoid conflict
Update 2: Add more unit test cases
Update 3: Fix default parameter typo for testpmd
Update 4: Adapted to the pktgen-dpdk prompt change from "Pktgen>" to "Pktgen:/>", now just expect "Pktgen"
Update 5: Per comment, merge common functions between latency and throughput tests to utils.py
Update 6: Per comment, seperate the test case from TC008 to a new test case TC077
Change-Id: I1f7471d4ba77636a3a66c79c2652578321312185
JIRA: YARDSTICK-614
Signed-off-by: Jing Zhang <jing.c.zhang@nokia.com>
|
|
we need to be following defautl paramiko rules,
first use pkey, then key_filenames (autodetecting ~/.ssh/ keys),
then password
We have too much boilerplate redudant code everywhere, we need
to standardize on a factory function that takes a node dict.
Using Python3 ChainMap we can layer overrides and defaults.
VNF descriptors have to default key_filename, password to Python None.
The only way to do this is to omit key values if the variable is not
defined, this way the dict will not have the value and it will
default to Python None
Add python2 chainmap backport
Updated unittest mocking to use ssh.SSH.from_node
Change-Id: I80b0cb606e593b33e317c9e5e8ed0b74da591514
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
Change-Id: I36d93eacab2470f90af5653104ad5c07853411bf
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
|
|
This allows calculating throughput in Gbps.
Add: test_pktgen.py
Change-Id: I7e2132ec71985c14570ef18b952b2264ebf7abbc
JIRA: YARKSTICK-611
Signed-off-by: Jing Zhang <jing.c.zhang@nokia.com>
|
|
JIRA: YARDSTICK-586
When run tc072, there is a bug: network_utilization: parse error, see the log:
ERROR ('network_utilization: parse error', [], [u'05:10:46', u'IFACE',
u'rxpck/s', u'txpck/s', u'rxkB/s', u'txkB/s', u'rxcmp/s',
u'txcmp/s', u'rxmcst/s', u'%ifutil'])
Traceback (most recent call last):
File
"/usr/local/lib/python2.7/dist-packages/yardstick/benchmark/runners/duration.py",
line 69, in _worker_process
method(data)
File
"/usr/local/lib/python2.7/dist-packages/yardstick/benchmark/scenarios/networking/netutilization.py",
line 191, in run
result.update(self._get_network_utilization())
File
"/usr/local/lib/python2.7/dist-packages/yardstick/benchmark/scenarios/networking/netutilization.py",
line 182, in _get_network_utilization
result = self._filtrate_result(raw_result)
File
"/usr/local/lib/python2.7/dist-packages/yardstick/benchmark/scenarios/networking/netutilization.py",
line 146, in _filtrate_result
fields, line)
RuntimeError: ('network_utilization: parse error', [], [u'05:10:46',
u'IFACE', u'rxpck/s', u'txpck/s', u'rxkB/s', u'txkB/s',
u'rxcmp/s', u'txcmp/s', u'rxmcst/s', u'%ifutil'])
Maybe the tool has some changes.
Change-Id: I6c4fe3fe9f749ec942fb5fbd799b8f4ab9a5c16c
Signed-off-by: chenjiankun <chenjiankun1@huawei.com>
|
|
Change-Id: I063fd37fe25754c94d164ae5a209d15b69322093
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
|