diff options
author | Ace Lee <liyin11@huawei.com> | 2017-10-11 09:22:49 +0000 |
---|---|---|
committer | Gerrit Code Review <gerrit@opnfv.org> | 2017-10-11 09:22:49 +0000 |
commit | 1ddcbed821e841c032aa2f1a4dfc3714716e83b8 (patch) | |
tree | cf7289cc7066a5777b79eb6e70ae498414110398 /testsuites | |
parent | ba35a9ec913aaba9594cc402632d6fb1bbbbf548 (diff) | |
parent | c0dc0572a4bf4bfc308b9d9ff5072468b9995d3c (diff) |
Merge "Bottlenecks testpmd scale-up testcase."
Diffstat (limited to 'testsuites')
3 files changed, 181 insertions, 0 deletions
diff --git a/testsuites/posca/testcase_cfg/posca_feature_testpmd_scale_up.yaml b/testsuites/posca/testcase_cfg/posca_feature_testpmd_scale_up.yaml new file mode 100644 index 00000000..a686b9db --- /dev/null +++ b/testsuites/posca/testcase_cfg/posca_feature_testpmd_scale_up.yaml @@ -0,0 +1,39 @@ +############################################################################## +# Copyright (c) 2017 HUAWEI TECHNOLOGIES CO.,LTD and others. +# +# All rights reserved. This program and the accompanying materials +# are made available under the terms of the Apache License, Version 2.0 +# which accompanies this distribution, and is available at +# http://www.apache.org/licenses/LICENSE-2.0 +############################################################################## + +load_manager: + scenarios: + tool: testpmd + # TODO: The number of seconds to run when doing binary search for a throughput test.. + search_interval: 60 + # info that the cpus and memes have the same number of data. + cpus: 6 + mems: 12 + # this is pktsize of Moongen to generate + # pkt_size: 64,128,256,1024,1518 + pkt_size: 64,128,256,1024 + # this is multiqueue of Monngen to generate + multistream: 1, 1000 + + runners: + stack_create: yardstick + Debug: False + yardstick_test_dir: "samples" + yardstick_testcase: "netperf_bottlenecks" + + runner_exta: + # info this section is for yardstick do some exta env prepare. + installation_method: yardstick + installation_type: testpmd + +contexts: + # info that dashboard if have data, we will create the data dashboard. + dashboard: "Bottlenecks-ELK" + yardstick: "yardstick_pmd" + yardstick_envpre: Flase diff --git a/testsuites/posca/testcase_script/posca_factor_system_bandwidth.py b/testsuites/posca/testcase_script/posca_factor_system_bandwidth.py index 05ea61e2..1a54554c 100644 --- a/testsuites/posca/testcase_script/posca_factor_system_bandwidth.py +++ b/testsuites/posca/testcase_script/posca_factor_system_bandwidth.py @@ -94,6 +94,7 @@ def do_test(test_config, Use_Dashboard, context_conf): def run(test_config): con_dic = test_config["load_manager"] + Use_Dashboard = False env_pre(None) if test_config["contexts"]["yardstick_ip"] is None: con_dic["contexts"]["yardstick_ip"] =\ diff --git a/testsuites/posca/testcase_script/posca_feature_testpmd_scale_up.py b/testsuites/posca/testcase_script/posca_feature_testpmd_scale_up.py new file mode 100644 index 00000000..830ff73f --- /dev/null +++ b/testsuites/posca/testcase_script/posca_feature_testpmd_scale_up.py @@ -0,0 +1,141 @@ +#!/usr/bin/env python +############################################################################## +# Copyright (c) 2017 Huawei Technologies Co.,Ltd and others. +# +# All rights reserved. This program and the accompanying materials +# are made available under the terms of the Apache License, Version 2.0 +# which accompanies this distribution, and is available at +# http://www.apache.org/licenses/LICENSE-2.0 +############################################################################## +'''This file realize the function of run systembandwidth script. +for example this contain two part first run_script, +second is algorithm, this part is about how to judge the bottlenecks. +This test is using yardstick as a tool to begin test.''' + +import os +import time +import uuid +import json +import utils.logger as log +from utils.parser import Parser as conf_parser +import utils.env_prepare.stack_prepare as stack_prepare +import utils.infra_setup.runner.docker_env as docker_env +import utils.infra_setup.runner.yardstick as yardstick_task + +# -------------------------------------------------- +# logging configuration +# -------------------------------------------------- +LOG = log.Logger(__name__).getLogger() + +testfile = os.path.basename(__file__) +testcase, file_format = os.path.splitext(testfile) +cidr = "/home/opnfv/repos/yardstick/samples/pvp_throughput_bottlenecks.yaml" +runner_DEBUG = True + + +def env_pre(con_dic): + LOG.info("yardstick environment prepare!") + stack_prepare._prepare_env_daemon(True) + + +def config_to_result(test_config, test_result): + final_data = [] + print(test_result) + out_data = test_result["result"]["testcases"] + test_data = out_data["pvp_throughput_bottlenecks"]["tc_data"] + for result in test_data: + testdata = {} + testdata["vcpu"] = test_config["vcpu"] + testdata["memory"] = test_config["memory"] + testdata["nrFlows"] = result["data"]["nrFlows"] + testdata["packet_size"] = result["data"]["packet_size"] + testdata["throughput"] = result["data"]["throughput_rx_mbps"] + final_data.append(testdata) + return final_data + + +def testcase_parser(out_file="yardstick.out", **parameter_info): + cmd = yardstick_task.yardstick_command_parser(debug=runner_DEBUG, + cidr=cidr, + outfile=out_file, + parameter=parameter_info) + return cmd + + +def do_test(test_config, Use_Dashboard, context_conf): + yardstick_container = docker_env.yardstick_info['container'] + out_file = ("/tmp/yardstick_" + str(uuid.uuid4()) + ".out") + cmd = testcase_parser(out_file=out_file, **test_config) + print(cmd) + stdout = docker_env.docker_exec_cmd(yardstick_container, cmd) + LOG.info(stdout) + loop_value = 0 + while loop_value < 60: + time.sleep(2) + loop_value = loop_value + 1 + with open(out_file) as f: + data = json.load(f) + if data["status"] == 1: + LOG.info("yardstick run success") + break + elif data["status"] == 2: + LOG.error("yardstick error exit") + exit() + # data = json.load(output) + + save_data = config_to_result(test_config, data) + if Use_Dashboard is True: + print("use dashboard") + # DashBoard.dashboard_send_data(context_conf, save_data) + + # return save_data["data_body"] + return save_data + + +def run(test_config): + load_config = test_config["load_manager"] + scenarios_conf = load_config["scenarios"] + Use_Dashboard = False + + env_pre(None) + if test_config["contexts"]["yardstick_ip"] is None: + load_config["contexts"]["yardstick_ip"] =\ + conf_parser.ip_parser("yardstick_test_ip") + + if "dashboard" in test_config["contexts"].keys(): + if test_config["contexts"]["dashboard_ip"] is None: + test_config["contexts"]["dashboard_ip"] =\ + conf_parser.ip_parser("dashboard") + LOG.info("Create Dashboard data") + Use_Dashboard = True + # DashBoard.dashboard_system_bandwidth(test_config["contexts"]) + + cpus = conf_parser.str_to_list(scenarios_conf["cpus"]) + mems = conf_parser.str_to_list(scenarios_conf["mems"]) + pkt_size = conf_parser.str_to_list(scenarios_conf["pkt_size"]) + multistream = conf_parser.str_to_list(scenarios_conf["multistream"]) + search_interval = scenarios_conf["search_interval"] + + load_config["result_file"] = os.path.dirname( + os.path.abspath(__file__)) + "/test_case/result" + + if len(cpus) != len(mems): + LOG.error("the cpus and mems config data number is not same!") + os._exit() + + result = [] + + for i in range(0, len(cpus)): + case_config = {"vcpu": cpus[i], + "memory": int(mems[i]) * 1024, + "multistreams": multistream, + "pktsize": pkt_size, + "search_interval": search_interval} + + data_reply = do_test(case_config, Use_Dashboard, + test_config["contexts"]) + result.append(data_reply) + + LOG.info("Finished bottlenecks testcase") + LOG.info("The result data is %s", result) + return result |