aboutsummaryrefslogtreecommitdiffstats
path: root/yardstick/cmd/commands/task.py
diff options
context:
space:
mode:
Diffstat (limited to 'yardstick/cmd/commands/task.py')
-rw-r--r--yardstick/cmd/commands/task.py468
1 files changed, 5 insertions, 463 deletions
diff --git a/yardstick/cmd/commands/task.py b/yardstick/cmd/commands/task.py
index 9524778ba..bd018bcab 100644
--- a/yardstick/cmd/commands/task.py
+++ b/yardstick/cmd/commands/task.py
@@ -8,28 +8,12 @@
##############################################################################
""" Handler for yardstick command 'task' """
-
-import sys
-import os
-import yaml
-import atexit
-import ipaddress
-import time
-import logging
-import uuid
-import errno
-from itertools import ifilter
-
-from yardstick.benchmark.contexts.base import Context
-from yardstick.benchmark.runners import base as base_runner
-from yardstick.common.task_template import TaskTemplate
+from yardstick.benchmark.core.task import Task
from yardstick.common.utils import cliargs
-from yardstick.common.utils import source_env
-from yardstick.common import constants
+from yardstick.cmd.commands import change_osloobj_to_paras
+
output_file_default = "/tmp/yardstick.out"
-test_cases_dir_default = "tests/opnfv/test_cases/"
-LOG = logging.getLogger(__name__)
class TaskCommands(object):
@@ -55,447 +39,5 @@ class TaskCommands(object):
@cliargs("--suite", help="process test suite file instead of a task file",
action="store_true")
def do_start(self, args, **kwargs):
- '''Start a benchmark scenario.'''
-
- atexit.register(atexit_handler)
-
- self.task_id = kwargs.get('task_id', str(uuid.uuid4()))
-
- check_environment()
-
- total_start_time = time.time()
- parser = TaskParser(args.inputfile[0])
-
- if args.suite:
- # 1.parse suite, return suite_params info
- task_files, task_args, task_args_fnames = \
- parser.parse_suite()
- else:
- task_files = [parser.path]
- task_args = [args.task_args]
- task_args_fnames = [args.task_args_file]
-
- LOG.info("\ntask_files:%s, \ntask_args:%s, \ntask_args_fnames:%s",
- task_files, task_args, task_args_fnames)
-
- if args.parse_only:
- sys.exit(0)
-
- if os.path.isfile(args.output_file):
- os.remove(args.output_file)
- # parse task_files
- for i in range(0, len(task_files)):
- one_task_start_time = time.time()
- parser.path = task_files[i]
- scenarios, run_in_parallel, meet_precondition = parser.parse_task(
- self.task_id, task_args[i], task_args_fnames[i])
-
- if not meet_precondition:
- LOG.info("meet_precondition is %s, please check envrionment",
- meet_precondition)
- continue
-
- self._run(scenarios, run_in_parallel, args.output_file)
-
- if args.keep_deploy:
- # keep deployment, forget about stack
- # (hide it for exit handler)
- Context.list = []
- else:
- for context in Context.list:
- context.undeploy()
- Context.list = []
- one_task_end_time = time.time()
- LOG.info("task %s finished in %d secs", task_files[i],
- one_task_end_time - one_task_start_time)
-
- total_end_time = time.time()
- LOG.info("total finished in %d secs",
- total_end_time - total_start_time)
-
- print "Done, exiting"
-
- def _run(self, scenarios, run_in_parallel, output_file):
- '''Deploys context and calls runners'''
- for context in Context.list:
- context.deploy()
-
- background_runners = []
-
- # Start all background scenarios
- for scenario in ifilter(_is_background_scenario, scenarios):
- scenario["runner"] = dict(type="Duration", duration=1000000000)
- runner = run_one_scenario(scenario, output_file)
- background_runners.append(runner)
-
- runners = []
- if run_in_parallel:
- for scenario in scenarios:
- if not _is_background_scenario(scenario):
- runner = run_one_scenario(scenario, output_file)
- runners.append(runner)
-
- # Wait for runners to finish
- for runner in runners:
- runner_join(runner)
- print "Runner ended, output in", output_file
- else:
- # run serially
- for scenario in scenarios:
- if not _is_background_scenario(scenario):
- runner = run_one_scenario(scenario, output_file)
- runner_join(runner)
- print "Runner ended, output in", output_file
-
- # Abort background runners
- for runner in background_runners:
- runner.abort()
-
- # Wait for background runners to finish
- for runner in background_runners:
- if runner.join(timeout=60) is None:
- # Nuke if it did not stop nicely
- base_runner.Runner.terminate(runner)
- runner_join(runner)
- else:
- base_runner.Runner.release(runner)
- print "Background task ended"
-
-
-# TODO: Move stuff below into TaskCommands class !?
-
-
-class TaskParser(object):
- '''Parser for task config files in yaml format'''
- def __init__(self, path):
- self.path = path
-
- def _meet_constraint(self, task, cur_pod, cur_installer):
- if "constraint" in task:
- constraint = task.get('constraint', None)
- if constraint is not None:
- tc_fit_pod = constraint.get('pod', None)
- tc_fit_installer = constraint.get('installer', None)
- LOG.info("cur_pod:%s, cur_installer:%s,tc_constraints:%s",
- cur_pod, cur_installer, constraint)
- if cur_pod and tc_fit_pod and cur_pod not in tc_fit_pod:
- return False
- if cur_installer and tc_fit_installer and \
- cur_installer not in tc_fit_installer:
- return False
- return True
-
- def _get_task_para(self, task, cur_pod):
- task_args = task.get('task_args', None)
- if task_args is not None:
- task_args = task_args.get(cur_pod, None)
- task_args_fnames = task.get('task_args_fnames', None)
- if task_args_fnames is not None:
- task_args_fnames = task_args_fnames.get(cur_pod, None)
- return task_args, task_args_fnames
-
- def parse_suite(self):
- '''parse the suite file and return a list of task config file paths
- and lists of optional parameters if present'''
- LOG.info("\nParsing suite file:%s", self.path)
-
- try:
- with open(self.path) as stream:
- cfg = yaml.load(stream)
- except IOError as ioerror:
- sys.exit(ioerror)
-
- self._check_schema(cfg["schema"], "suite")
- LOG.info("\nStarting scenario:%s", cfg["name"])
-
- test_cases_dir = cfg.get("test_cases_dir", test_cases_dir_default)
- if test_cases_dir[-1] != os.sep:
- test_cases_dir += os.sep
-
- cur_pod = os.environ.get('NODE_NAME', None)
- cur_installer = os.environ.get('INSTALLER_TYPE', None)
-
- valid_task_files = []
- valid_task_args = []
- valid_task_args_fnames = []
-
- for task in cfg["test_cases"]:
- # 1.check file_name
- if "file_name" in task:
- task_fname = task.get('file_name', None)
- if task_fname is None:
- continue
- else:
- continue
- # 2.check constraint
- if self._meet_constraint(task, cur_pod, cur_installer):
- valid_task_files.append(test_cases_dir + task_fname)
- else:
- continue
- # 3.fetch task parameters
- task_args, task_args_fnames = self._get_task_para(task, cur_pod)
- valid_task_args.append(task_args)
- valid_task_args_fnames.append(task_args_fnames)
-
- return valid_task_files, valid_task_args, valid_task_args_fnames
-
- def parse_task(self, task_id, task_args=None, task_args_file=None):
- '''parses the task file and return an context and scenario instances'''
- print "Parsing task config:", self.path
-
- try:
- kw = {}
- if task_args_file:
- with open(task_args_file) as f:
- kw.update(parse_task_args("task_args_file", f.read()))
- kw.update(parse_task_args("task_args", task_args))
- except TypeError:
- raise TypeError()
-
- try:
- with open(self.path) as f:
- try:
- input_task = f.read()
- rendered_task = TaskTemplate.render(input_task, **kw)
- except Exception as e:
- print(("Failed to render template:\n%(task)s\n%(err)s\n")
- % {"task": input_task, "err": e})
- raise e
- print(("Input task is:\n%s\n") % rendered_task)
-
- cfg = yaml.load(rendered_task)
- except IOError as ioerror:
- sys.exit(ioerror)
-
- self._check_schema(cfg["schema"], "task")
- meet_precondition = self._check_precondition(cfg)
-
- # TODO: support one or many contexts? Many would simpler and precise
- # TODO: support hybrid context type
- if "context" in cfg:
- context_cfgs = [cfg["context"]]
- elif "contexts" in cfg:
- context_cfgs = cfg["contexts"]
- else:
- context_cfgs = [{"type": "Dummy"}]
-
- for cfg_attrs in context_cfgs:
- context_type = cfg_attrs.get("type", "Heat")
- if "Heat" == context_type and "networks" in cfg_attrs:
- # bugfix: if there are more than one network,
- # only add "external_network" on first one.
- # the name of netwrok should follow this rule:
- # test, test2, test3 ...
- # sort network with the length of network's name
- sorted_networks = sorted(cfg_attrs["networks"].keys())
- # config external_network based on env var
- cfg_attrs["networks"][sorted_networks[0]]["external_network"] \
- = os.environ.get("EXTERNAL_NETWORK", "net04_ext")
-
- context = Context.get(context_type)
- context.init(cfg_attrs)
-
- run_in_parallel = cfg.get("run_in_parallel", False)
-
- # add tc and task id for influxdb extended tags
- for scenario in cfg["scenarios"]:
- task_name = os.path.splitext(os.path.basename(self.path))[0]
- scenario["tc"] = task_name
- scenario["task_id"] = task_id
-
- # TODO we need something better here, a class that represent the file
- return cfg["scenarios"], run_in_parallel, meet_precondition
-
- def _check_schema(self, cfg_schema, schema_type):
- '''Check if config file is using the correct schema type'''
-
- if cfg_schema != "yardstick:" + schema_type + ":0.1":
- sys.exit("error: file %s has unknown schema %s" % (self.path,
- cfg_schema))
-
- def _check_precondition(self, cfg):
- '''Check if the envrionment meet the preconditon'''
-
- if "precondition" in cfg:
- precondition = cfg["precondition"]
- installer_type = precondition.get("installer_type", None)
- deploy_scenarios = precondition.get("deploy_scenarios", None)
- tc_fit_pods = precondition.get("pod_name", None)
- installer_type_env = os.environ.get('INSTALL_TYPE', None)
- deploy_scenario_env = os.environ.get('DEPLOY_SCENARIO', None)
- pod_name_env = os.environ.get('NODE_NAME', None)
-
- LOG.info("installer_type: %s, installer_type_env: %s",
- installer_type, installer_type_env)
- LOG.info("deploy_scenarios: %s, deploy_scenario_env: %s",
- deploy_scenarios, deploy_scenario_env)
- LOG.info("tc_fit_pods: %s, pod_name_env: %s",
- tc_fit_pods, pod_name_env)
- if installer_type and installer_type_env:
- if installer_type_env not in installer_type:
- return False
- if deploy_scenarios and deploy_scenario_env:
- deploy_scenarios_list = deploy_scenarios.split(',')
- for deploy_scenario in deploy_scenarios_list:
- if deploy_scenario_env.startswith(deploy_scenario):
- return True
- return False
- if tc_fit_pods and pod_name_env:
- if pod_name_env not in tc_fit_pods:
- return False
- return True
-
-
-def atexit_handler():
- '''handler for process termination'''
- base_runner.Runner.terminate_all()
-
- if len(Context.list) > 0:
- print "Undeploying all contexts"
- for context in Context.list:
- context.undeploy()
-
-
-def is_ip_addr(addr):
- '''check if string addr is an IP address'''
- try:
- ipaddress.ip_address(unicode(addr))
- return True
- except ValueError:
- return False
-
-
-def _is_same_heat_context(host_attr, target_attr):
- '''check if two servers are in the same heat context
- host_attr: either a name for a server created by yardstick or a dict
- with attribute name mapping when using external heat templates
- target_attr: either a name for a server created by yardstick or a dict
- with attribute name mapping when using external heat templates
- '''
- host = None
- target = None
- for context in Context.list:
- if context.__context_type__ != "Heat":
- continue
-
- host = context._get_server(host_attr)
- if host is None:
- continue
-
- target = context._get_server(target_attr)
- if target is None:
- return False
-
- # Both host and target is not None, then they are in the
- # same heat context.
- return True
-
- return False
-
-
-def _is_background_scenario(scenario):
- if "run_in_background" in scenario:
- return scenario["run_in_background"]
- else:
- return False
-
-
-def run_one_scenario(scenario_cfg, output_file):
- '''run one scenario using context'''
- runner_cfg = scenario_cfg["runner"]
- runner_cfg['output_filename'] = output_file
-
- # TODO support get multi hosts/vms info
- context_cfg = {}
- if "host" in scenario_cfg:
- context_cfg['host'] = Context.get_server(scenario_cfg["host"])
-
- if "target" in scenario_cfg:
- if is_ip_addr(scenario_cfg["target"]):
- context_cfg['target'] = {}
- context_cfg['target']["ipaddr"] = scenario_cfg["target"]
- else:
- context_cfg['target'] = Context.get_server(scenario_cfg["target"])
- if _is_same_heat_context(scenario_cfg["host"],
- scenario_cfg["target"]):
- context_cfg["target"]["ipaddr"] = \
- context_cfg["target"]["private_ip"]
- else:
- context_cfg["target"]["ipaddr"] = \
- context_cfg["target"]["ip"]
-
- if "targets" in scenario_cfg:
- ip_list = []
- for target in scenario_cfg["targets"]:
- if is_ip_addr(target):
- ip_list.append(target)
- context_cfg['target'] = {}
- else:
- context_cfg['target'] = Context.get_server(target)
- if _is_same_heat_context(scenario_cfg["host"], target):
- ip_list.append(context_cfg["target"]["private_ip"])
- else:
- ip_list.append(context_cfg["target"]["ip"])
- context_cfg['target']['ipaddr'] = ','.join(ip_list)
-
- if "nodes" in scenario_cfg:
- context_cfg["nodes"] = parse_nodes_with_context(scenario_cfg)
- runner = base_runner.Runner.get(runner_cfg)
-
- print "Starting runner of type '%s'" % runner_cfg["type"]
- runner.run(scenario_cfg, context_cfg)
-
- return runner
-
-
-def parse_nodes_with_context(scenario_cfg):
- '''paras the 'nodes' fields in scenario '''
- nodes = scenario_cfg["nodes"]
-
- nodes_cfg = {}
- for nodename in nodes:
- nodes_cfg[nodename] = Context.get_server(nodes[nodename])
-
- return nodes_cfg
-
-
-def runner_join(runner):
- '''join (wait for) a runner, exit process at runner failure'''
- status = runner.join()
- base_runner.Runner.release(runner)
- if status != 0:
- sys.exit("Runner failed")
-
-
-def print_invalid_header(source_name, args):
- print(("Invalid %(source)s passed:\n\n %(args)s\n")
- % {"source": source_name, "args": args})
-
-
-def parse_task_args(src_name, args):
- try:
- kw = args and yaml.safe_load(args)
- kw = {} if kw is None else kw
- except yaml.parser.ParserError as e:
- print_invalid_header(src_name, args)
- print(("%(source)s has to be YAML. Details:\n\n%(err)s\n")
- % {"source": src_name, "err": e})
- raise TypeError()
-
- if not isinstance(kw, dict):
- print_invalid_header(src_name, args)
- print(("%(src)s had to be dict, actually %(src_type)s\n")
- % {"src": src_name, "src_type": type(kw)})
- raise TypeError()
- return kw
-
-
-def check_environment():
- auth_url = os.environ.get('OS_AUTH_URL', None)
- if not auth_url:
- try:
- source_env(constants.OPENSTACK_RC_FILE)
- except IOError as e:
- if e.errno != errno.EEXIST:
- raise
- LOG.debug('OPENRC file not found')
+ param = change_osloobj_to_paras(args)
+ Task().start(param)