class Runner(RunnerInterface):
name = 'nrunner'
description = 'nrunner based implementation of job compliant runner'
def _save_to_file(self, filename, buff, mode='wb'):
with open(filename, mode) as fp:
fp.write(buff)
def _populate_task_logdir(self, base_path, task, statuses, debug=False):
# We are copying here to avoid printing duplicated information
local_statuses = copy(statuses)
last = local_statuses[-1]
try:
stdout = last.pop('stdout')
except KeyError:
stdout = None
try:
stderr = last.pop('stderr')
except KeyError:
stderr = None
# Create task dir
task_path = os.path.join(base_path, task.identifier.str_filesystem)
os.makedirs(task_path, exist_ok=True)
# Save stdout and stderr
if stdout is not None:
stdout_file = os.path.join(task_path, 'stdout')
self._save_to_file(stdout_file, stdout)
if stderr is not None:
stderr_file = os.path.join(task_path, 'stderr')
self._save_to_file(stderr_file, stderr)
# Save debug
if debug:
debug = os.path.join(task_path, 'debug')
with open(debug, 'w') as fp:
json.dump(local_statuses, fp)
data_file = os.path.join(task_path, 'data')
with open(data_file, 'w') as fp:
fp.write("{}\n".format(task.output_dir))
def _get_all_runtime_tasks(self, test_suite):
result = []
no_digits = len(str(len(test_suite)))
for index, task in enumerate(test_suite.tests, start=1):
task.known_runners = nrunner.RUNNERS_REGISTRY_PYTHON_CLASS
# this is all rubbish data
if test_suite.name:
prefix = "{}-{}".format(test_suite.name, index)
else:
prefix = index
test_id = TestID(prefix, task.runnable.uri, None, no_digits)
task.identifier = test_id
result.append(RuntimeTask(task))
return result
def _start_status_server(self, status_server_listen):
# pylint: disable=W0201
self.status_repo = StatusRepo()
# pylint: disable=W0201
self.status_server = StatusServer(status_server_listen,
self.status_repo)
asyncio.ensure_future(self.status_server.serve_forever())
async def _update_status(self, job):
tasks_by_id = {
str(runtime_task.task.identifier): runtime_task.task
for runtime_task in self.tasks
}
while True:
try:
(task_id, status,
_) = self.status_repo.status_journal_summary.pop(0)
except IndexError:
await asyncio.sleep(0.05)
continue
task = tasks_by_id.get(task_id)
early_state = {
'name': task.identifier,
'job_logdir': job.logdir,
'job_unique_id': job.unique_id
}
if status == 'started':
job.result.start_test(early_state)
job.result_events_dispatcher.map_method(
'start_test', job.result, early_state)
elif status == 'finished':
this_task_data = self.status_repo.get_task_data(task_id)
last_task_status = this_task_data[-1]
test_state = {'status': last_task_status.get('result').upper()}
test_state.update(early_state)
time_start = this_task_data[0]['time']
time_end = last_task_status['time']
time_elapsed = time_end - time_start
test_state['time_start'] = time_start
test_state['time_end'] = time_end
test_state['time_elapsed'] = time_elapsed
# fake log dir, needed by some result plugins such as HTML
test_state['logdir'] = ''
base_path = os.path.join(job.logdir, 'test-results')
self._populate_task_logdir(base_path, task, this_task_data,
job.config.get('core.debug'))
job.result.check_test(test_state)
job.result_events_dispatcher.map_method(
'end_test', job.result, test_state)
if not mapping[test_state['status']]:
self.summary.add("FAIL")
def run_suite(self, job, test_suite):
# pylint: disable=W0201
self.summary = set()
test_suite.tests, _ = nrunner.check_tasks_requirements(
test_suite.tests)
job.result.tests_total = test_suite.size # no support for variants yet
listen = test_suite.config.get('nrunner.status_server_listen')
self._start_status_server(listen)
# pylint: disable=W0201
self.tasks = self._get_all_runtime_tasks(test_suite)
if test_suite.config.get('nrunner.shuffle'):
random.shuffle(self.tasks)
tsm = TaskStateMachine(self.tasks)
spawner_name = test_suite.config.get('nrunner.spawner')
spawner = SpawnerDispatcher(test_suite.config)[spawner_name].obj
max_running = min(test_suite.config.get('nrunner.max_parallel_tasks'),
len(self.tasks))
workers = [
Worker(tsm, spawner, max_running=max_running).run()
for _ in range(max_running)
]
asyncio.ensure_future(self._update_status(job))
loop = asyncio.get_event_loop()
try:
loop.run_until_complete(
asyncio.wait_for(asyncio.gather(*workers), job.timeout
or None))
except (KeyboardInterrupt, asyncio.TimeoutError):
self.summary.add("INTERRUPTED")
# Wait until all messages may have been processed by the
# status_updater. This should be replaced by a mechanism
# that only waits if there are missing status messages to
# be processed, and, only for a given amount of time.
# Tests with non received status will always show as SKIP
# because of result reconciliation.
loop.run_until_complete(asyncio.sleep(0.05))
job.result.end_tests()
self.status_server.close()
return self.summary
class Runner(RunnerInterface):
name = 'nrunner'
description = 'nrunner based implementation of job compliant runner'
@staticmethod
def _get_requirements_runtime_tasks(runnable, prefix):
if runnable.requirements is None:
return
# creates the runnables for the requirements
requirements_runnables = RequirementsResolver.resolve(runnable)
requirements_runtime_tasks = []
# creates the tasks and runtime tasks for the requirements
for requirement_runnable in requirements_runnables:
name = '%s-%s' % (requirement_runnable.kind,
requirement_runnable.kwargs.get('name'))
# the human UI works with TestID objects, so we need to
# use it to name other tasks
task_id = TestID(prefix, name, None)
# with --dry-run we don't want to run requirement
if runnable.kind == 'dry-run':
requirement_runnable.kind = 'noop'
# creates the requirement task
requirement_task = nrunner.Task(requirement_runnable,
identifier=task_id,
category='requirement')
# make sure we track the dependencies of a task
# runtime_task.task.dependencies.add(requirement_task)
# created the requirement runtime task
requirements_runtime_tasks.append(RuntimeTask(requirement_task))
return requirements_runtime_tasks
@staticmethod
def _create_runtime_tasks_for_test(test_suite, runnable, no_digits, index,
variant):
"""Creates runtime tasks for both tests, and for its requirements."""
result = []
# test related operations
# create test ID
if test_suite.name:
prefix = "{}-{}".format(test_suite.name, index)
else:
prefix = index
test_id = TestID(prefix, runnable.uri, variant, no_digits)
# inject variant on runnable
runnable.variant = dump_variant(variant)
# handles the test task
task = nrunner.Task(
runnable,
identifier=test_id,
known_runners=nrunner.RUNNERS_REGISTRY_PYTHON_CLASS)
runtime_task = RuntimeTask(task)
result.append(runtime_task)
# handles the requirements
requirements_runtime_tasks = (Runner._get_requirements_runtime_tasks(
runnable, prefix))
# extend the list of tasks with the requirements runtime tasks
if requirements_runtime_tasks is not None:
for requirement_runtime_task in requirements_runtime_tasks:
# make sure we track the dependencies of a task
runtime_task.task.dependencies.add(
requirement_runtime_task.task)
result.extend(requirements_runtime_tasks)
return result
@staticmethod
def _get_all_runtime_tasks(test_suite):
runtime_tasks = []
test_result_total = test_suite.variants.get_number_of_tests(
test_suite.tests)
no_digits = len(str(test_result_total))
# define execution order
execution_order = test_suite.config.get('run.execution_order')
if execution_order == "variants-per-test":
test_variant = [(test, variant) for test in test_suite.tests
for variant in test_suite.variants.itertests()]
elif execution_order == "tests-per-variant":
test_variant = [(test, variant)
for variant in test_suite.variants.itertests()
for test in test_suite.tests]
# decide if a copy of the runnable is needed, in case of more
# variants than tests
copy_runnable = len(test_variant) > len(test_suite.tests)
# create runtime tasks
for index, (runnable, variant) in enumerate(test_variant, start=1):
if copy_runnable:
runnable = deepcopy(runnable)
runtime_tasks.extend(
Runner._create_runtime_tasks_for_test(test_suite, runnable,
no_digits, index,
variant))
return runtime_tasks
def _start_status_server(self, status_server_listen):
# pylint: disable=W0201
self.status_repo = StatusRepo()
# pylint: disable=W0201
self.status_server = StatusServer(status_server_listen,
self.status_repo)
asyncio.ensure_future(self.status_server.serve_forever())
async def _update_status(self, job):
tasks_by_id = {
str(runtime_task.task.identifier): runtime_task.task
for runtime_task in self.runtime_tasks
}
message_handler = MessageHandler()
while True:
try:
(task_id, _, _, index) = \
self.status_repo.status_journal_summary.pop(0)
except IndexError:
await asyncio.sleep(0.05)
continue
message = self.status_repo.get_task_data(task_id, index)
task = tasks_by_id.get(task_id)
message_handler.process_message(message, task, job)
def run_suite(self, job, test_suite):
summary = set()
test_suite.tests, _ = nrunner.check_runnables_runner_requirements(
test_suite.tests)
job.result.tests_total = test_suite.variants.get_number_of_tests(
test_suite.tests)
listen = test_suite.config.get('nrunner.status_server_listen')
self._start_status_server(listen)
# pylint: disable=W0201
self.runtime_tasks = self._get_all_runtime_tasks(test_suite)
if test_suite.config.get('nrunner.shuffle'):
random.shuffle(self.runtime_tasks)
test_ids = [
rt.task.identifier for rt in self.runtime_tasks
if rt.task.category == 'test'
]
tsm = TaskStateMachine(self.runtime_tasks, self.status_repo)
spawner_name = test_suite.config.get('nrunner.spawner')
spawner = SpawnerDispatcher(test_suite.config)[spawner_name].obj
max_running = min(test_suite.config.get('nrunner.max_parallel_tasks'),
len(self.runtime_tasks))
timeout = test_suite.config.get('task.timeout.running')
workers = [
Worker(state_machine=tsm,
spawner=spawner,
max_running=max_running,
task_timeout=timeout).run() for _ in range(max_running)
]
asyncio.ensure_future(self._update_status(job))
loop = asyncio.get_event_loop()
try:
loop.run_until_complete(
asyncio.wait_for(asyncio.gather(*workers), job.timeout
or None))
except (KeyboardInterrupt, asyncio.TimeoutError, TestFailFast) as ex:
LOG_JOB.info(str(ex))
job.interrupted_reason = str(ex)
summary.add("INTERRUPTED")
# Wait until all messages may have been processed by the
# status_updater. This should be replaced by a mechanism
# that only waits if there are missing status messages to
# be processed, and, only for a given amount of time.
# Tests with non received status will always show as SKIP
# because of result reconciliation.
loop.run_until_complete(asyncio.sleep(0.05))
job.result.end_tests()
self.status_server.close()
# Update the overall summary with found test statuses, which will
# determine the Avocado command line exit status
summary.update([
status.upper()
for status in self.status_repo.get_result_set_for_tasks(test_ids)
])
return summary
class Runner(RunnerInterface):
name = 'nrunner'
description = 'nrunner based implementation of job compliant runner'
def _determine_status_server_uri(self, test_suite):
# pylint: disable=W0201
self.status_server_dir = None
if test_suite.config.get('nrunner.status_server_auto'):
# no UNIX domain sockets on Windows
if platform.system() != 'Windows':
self.status_server_dir = tempfile.TemporaryDirectory(
prefix='avocado_')
return os.path.join(self.status_server_dir.name,
'.status_server.sock')
return test_suite.config.get('nrunner.status_server_listen')
def _create_status_server(self, test_suite, job):
listen = self._determine_status_server_uri(test_suite)
# pylint: disable=W0201
self.status_repo = StatusRepo(job.unique_id)
# pylint: disable=W0201
self.status_server = StatusServer(listen, self.status_repo)
async def _update_status(self, job):
tasks_by_id = {
str(runtime_task.task.identifier): runtime_task.task
for runtime_task in self.runtime_tasks
}
message_handler = MessageHandler()
while True:
try:
(task_id, _, _, index) = \
self.status_repo.status_journal_summary.pop(0)
except IndexError:
await asyncio.sleep(0.05)
continue
message = self.status_repo.get_task_data(task_id, index)
task = tasks_by_id.get(task_id)
message_handler.process_message(message, task, job)
@staticmethod
def _abort_if_missing_runners(runnables):
if runnables:
missing_kinds = set([runnable.kind for runnable in runnables])
msg = ("Could not find runners for runnable(s) of kind(s): %s" %
", ".join(missing_kinds))
raise JobError(msg)
def run_suite(self, job, test_suite):
summary = set()
if not test_suite.enabled:
job.interrupted_reason = f"Suite {test_suite.name} is disabled."
return summary
test_suite.tests, missing_requirements = nrunner.check_runnables_runner_requirements(
test_suite.tests)
self._abort_if_missing_runners(missing_requirements)
job.result.tests_total = test_suite.variants.get_number_of_tests(
test_suite.tests)
self._create_status_server(test_suite, job)
graph = RuntimeTaskGraph(test_suite.get_test_variants(),
test_suite.name, self.status_server.uri,
job.unique_id)
# pylint: disable=W0201
self.runtime_tasks = graph.get_tasks_in_topological_order()
# Start the status server
asyncio.ensure_future(self.status_server.serve_forever())
if test_suite.config.get('nrunner.shuffle'):
random.shuffle(self.runtime_tasks)
test_ids = [
rt.task.identifier for rt in self.runtime_tasks
if rt.task.category == 'test'
]
tsm = TaskStateMachine(self.runtime_tasks, self.status_repo)
spawner_name = test_suite.config.get('nrunner.spawner')
spawner = SpawnerDispatcher(test_suite.config, job)[spawner_name].obj
max_running = min(test_suite.config.get('nrunner.max_parallel_tasks'),
len(self.runtime_tasks))
timeout = test_suite.config.get('task.timeout.running')
failfast = test_suite.config.get('run.failfast')
workers = [
Worker(state_machine=tsm,
spawner=spawner,
max_running=max_running,
task_timeout=timeout,
failfast=failfast).run() for _ in range(max_running)
]
asyncio.ensure_future(self._update_status(job))
loop = asyncio.get_event_loop()
try:
loop.run_until_complete(
asyncio.wait_for(asyncio.gather(*workers), job.timeout
or None))
except (KeyboardInterrupt, asyncio.TimeoutError, TestFailFast) as ex:
LOG_JOB.info(str(ex))
job.interrupted_reason = str(ex)
summary.add("INTERRUPTED")
# Wait until all messages may have been processed by the
# status_updater. This should be replaced by a mechanism
# that only waits if there are missing status messages to
# be processed, and, only for a given amount of time.
# Tests with non received status will always show as SKIP
# because of result reconciliation.
loop.run_until_complete(asyncio.sleep(0.05))
job.result.end_tests()
self.status_server.close()
if self.status_server_dir is not None:
self.status_server_dir.cleanup()
# Update the overall summary with found test statuses, which will
# determine the Avocado command line exit status
summary.update([
status.upper()
for status in self.status_repo.get_result_set_for_tasks(test_ids)
])
return summary
class Runner(RunnerInterface):
name = 'nrunner'
description = 'nrunner based implementation of job compliant runner'
def _get_requirements_runtime_tasks(self, runnable, prefix, job_id):
if runnable.requirements is None:
return
# creates the runnables for the requirements
requirements_runnables = RequirementsResolver.resolve(runnable)
requirements_runtime_tasks = []
# creates the tasks and runtime tasks for the requirements
for requirement_runnable in requirements_runnables:
name = '%s-%s' % (requirement_runnable.kind,
requirement_runnable.kwargs.get('name'))
# the human UI works with TestID objects, so we need to
# use it to name other tasks
task_id = TestID(prefix,
name,
None)
# with --dry-run we don't want to run requirement
if runnable.kind == 'dry-run':
requirement_runnable.kind = 'noop'
# creates the requirement task
requirement_task = nrunner.Task(requirement_runnable,
identifier=task_id,
status_uris=[self.status_server.uri],
category='requirement',
job_id=job_id)
# make sure we track the dependencies of a task
# runtime_task.task.dependencies.add(requirement_task)
# created the requirement runtime task
requirements_runtime_tasks.append(RuntimeTask(requirement_task))
return requirements_runtime_tasks
def _create_runtime_tasks_for_test(self, test_suite, runnable, no_digits,
index, variant, job_id):
"""Creates runtime tasks for both tests, and for its requirements."""
result = []
# test related operations
# create test ID
if test_suite.name:
prefix = "{}-{}".format(test_suite.name, index)
else:
prefix = index
test_id = TestID(prefix,
runnable.identifier,
variant,
no_digits)
# inject variant on runnable
runnable.variant = dump_variant(variant)
# handles the test task
task = nrunner.Task(runnable,
identifier=test_id,
known_runners=nrunner.RUNNERS_REGISTRY_PYTHON_CLASS,
status_uris=[self.status_server.uri],
job_id=job_id)
runtime_task = RuntimeTask(task)
result.append(runtime_task)
# handles the requirements
requirements_runtime_tasks = (
self._get_requirements_runtime_tasks(runnable,
prefix,
job_id))
# extend the list of tasks with the requirements runtime tasks
if requirements_runtime_tasks is not None:
for requirement_runtime_task in requirements_runtime_tasks:
# make sure we track the dependencies of a task
runtime_task.task.dependencies.add(
requirement_runtime_task.task)
result.extend(requirements_runtime_tasks)
return result
def _get_all_runtime_tasks(self, test_suite, job_id):
runtime_tasks = []
test_result_total = test_suite.variants.get_number_of_tests(test_suite.tests)
no_digits = len(str(test_result_total))
if test_suite.test_parameters:
paths = ['/']
tree_nodes = TreeNode().get_node(paths[0], True)
tree_nodes.value = test_suite.test_parameters
variant = {"variant": tree_nodes, "variant_id": None, "paths": paths}
test_variant = [(test, variant) for test in test_suite.tests]
else:
# let's use variants when parameters are not available
# define execution order
execution_order = test_suite.config.get('run.execution_order')
if execution_order == "variants-per-test":
test_variant = [(test, variant) for test in test_suite.tests
for variant in test_suite.variants.itertests()]
elif execution_order == "tests-per-variant":
test_variant = [(test, variant)
for variant in test_suite.variants.itertests()
for test in test_suite.tests]
# decide if a copy of the runnable is needed, in case of more
# variants than tests
copy_runnable = len(test_variant) > len(test_suite.tests)
# create runtime tasks
for index, (runnable, variant) in enumerate(test_variant, start=1):
if copy_runnable:
runnable = deepcopy(runnable)
runtime_tasks.extend(self._create_runtime_tasks_for_test(
test_suite,
runnable,
no_digits,
index,
variant,
job_id))
return runtime_tasks
def _determine_status_server_uri(self, test_suite):
# pylint: disable=W0201
self.status_server_dir = None
if test_suite.config.get('nrunner.status_server_auto'):
# no UNIX domain sockets on Windows
if platform.system() != 'Windows':
self.status_server_dir = tempfile.TemporaryDirectory(
prefix='avocado_')
return os.path.join(self.status_server_dir.name,
'.status_server.sock')
return test_suite.config.get('nrunner.status_server_listen')
def _create_status_server(self, test_suite, job):
listen = self._determine_status_server_uri(test_suite)
# pylint: disable=W0201
self.status_repo = StatusRepo(job.unique_id)
# pylint: disable=W0201
self.status_server = StatusServer(listen,
self.status_repo)
async def _update_status(self, job):
tasks_by_id = {str(runtime_task.task.identifier): runtime_task.task
for runtime_task in self.runtime_tasks}
message_handler = MessageHandler()
while True:
try:
(task_id, _, _, index) = \
self.status_repo.status_journal_summary.pop(0)
except IndexError:
await asyncio.sleep(0.05)
continue
message = self.status_repo.get_task_data(task_id, index)
task = tasks_by_id.get(task_id)
message_handler.process_message(message, task, job)
@staticmethod
def _abort_if_missing_runners(runnables):
if runnables:
missing_kinds = set([runnable.kind for runnable in runnables])
msg = ("Could not find runners for runnable(s) of kind(s): %s"
% ", ".join(missing_kinds))
raise JobError(msg)
def run_suite(self, job, test_suite):
summary = set()
if not test_suite.enabled:
job.interrupted_reason = f"Suite {test_suite.name} is disabled."
return summary
test_suite.tests, missing_requirements = nrunner.check_runnables_runner_requirements(
test_suite.tests)
self._abort_if_missing_runners(missing_requirements)
job.result.tests_total = test_suite.variants.get_number_of_tests(test_suite.tests)
self._create_status_server(test_suite, job)
# pylint: disable=W0201
self.runtime_tasks = self._get_all_runtime_tasks(test_suite, job.unique_id)
# Start the status server
asyncio.ensure_future(self.status_server.serve_forever())
if test_suite.config.get('nrunner.shuffle'):
random.shuffle(self.runtime_tasks)
test_ids = [rt.task.identifier for rt in self.runtime_tasks
if rt.task.category == 'test']
tsm = TaskStateMachine(self.runtime_tasks, self.status_repo)
spawner_name = test_suite.config.get('nrunner.spawner')
spawner = SpawnerDispatcher(test_suite.config)[spawner_name].obj
spawner.job_output_dir = job.test_results_path
max_running = min(test_suite.config.get('nrunner.max_parallel_tasks'),
len(self.runtime_tasks))
timeout = test_suite.config.get('task.timeout.running')
failfast = test_suite.config.get('run.failfast')
workers = [Worker(state_machine=tsm,
spawner=spawner,
max_running=max_running,
task_timeout=timeout,
failfast=failfast).run()
for _ in range(max_running)]
asyncio.ensure_future(self._update_status(job))
loop = asyncio.get_event_loop()
try:
loop.run_until_complete(asyncio.wait_for(asyncio.gather(*workers),
job.timeout or None))
except (KeyboardInterrupt, asyncio.TimeoutError, TestFailFast) as ex:
LOG_JOB.info(str(ex))
job.interrupted_reason = str(ex)
summary.add("INTERRUPTED")
# Wait until all messages may have been processed by the
# status_updater. This should be replaced by a mechanism
# that only waits if there are missing status messages to
# be processed, and, only for a given amount of time.
# Tests with non received status will always show as SKIP
# because of result reconciliation.
loop.run_until_complete(asyncio.sleep(0.05))
job.result.end_tests()
self.status_server.close()
if self.status_server_dir is not None:
self.status_server_dir.cleanup()
# Update the overall summary with found test statuses, which will
# determine the Avocado command line exit status
summary.update([status.upper() for status in
self.status_repo.get_result_set_for_tasks(test_ids)])
return summary
class CartesianRunner(RunnerInterface):
"""Test runner for Cartesian graph traversal."""
name = 'traverser'
description = 'Runs tests through a Cartesian graph traversal'
@property
def all_tests_ok(self):
"""
Evaluate if all tests run under this runner have an ok status.
:returns: whether all tests ended with acceptable status
:rtype: bool
"""
mapped_status = {
STATUSES_MAPPING[t["status"]]
for t in self.job.result.tests
}
return all(mapped_status)
def __init__(self):
"""Construct minimal attributes for the Cartesian runner."""
self.tasks = []
self.slots = []
self.status_repo = None
self.status_server = None
"""running functionality"""
async def _update_status(self, job):
message_handler = MessageHandler()
while True:
try:
(task_id, _, _, index) = \
self.status_repo.status_journal_summary.pop(0)
except IndexError:
await asyncio.sleep(0.05)
continue
message = self.status_repo.get_task_data(task_id, index)
tasks_by_id = {
str(runtime_task.task.identifier): runtime_task.task
for runtime_task in self.tasks
}
task = tasks_by_id.get(task_id)
message_handler.process_message(message, task, job)
async def run_test(self, job, node):
"""
Run a test instance inside a subprocess.
:param job: job that includes the test suite
:type job: :py:class:`avocado.core.job.Job`
:param node: test node to run
:type node: :py:class:`TestNode`
"""
if node.spawner is None:
default_slot = self.slots[0] if len(self.slots) > 0 else ""
node.set_environment(job, default_slot)
# once the slot is set (here or earlier), the hostname reflects it
hostname = node.params["hostname"]
hostname = "localhost" if not hostname else hostname
logging.debug(f"Running {node.id} on {hostname}")
if not self.status_repo:
self.status_repo = StatusRepo(job.unique_id)
self.status_server = StatusServer(
job.config.get('nrunner.status_server_listen'),
self.status_repo)
asyncio.ensure_future(self.status_server.serve_forever())
# TODO: this needs more customization
asyncio.ensure_future(self._update_status(job))
raw_task = nrunner.Task(node.get_runnable(),
node.id_test,
[job.config.get('nrunner.status_server_uri')],
nrunner.RUNNERS_REGISTRY_PYTHON_CLASS,
job_id=self.job.unique_id)
task = RuntimeTask(raw_task)
self.tasks += [task]
# TODO: use a single state machine for all test nodes when we are able
# to at least add requested tasks to it safely (using its locks)
await Worker(
state_machine=TaskStateMachine([task], self.status_repo),
spawner=node.spawner,
max_running=1,
task_timeout=job.config.get('task.timeout.running')).run()
async def run_test_node(self, node, can_retry=False):
"""
Run a node once, and optionally re-run it depending on the parameters.
:param node: test node to run
:type node: :py:class:`TestNode`
:param bool can_retry: whether this node can be re-run
:returns: run status of :py:meth:`run_test`
:rtype: bool
:raises: :py:class:`AssertionError` if the ran test node contains no objects
The retry parameters are `retry_attempts` and `retry_stop`. The first is
the maximum number of retries, and the second indicates when to stop retrying.
The possible combinations of these values are:
- `retry_stop = error`: retry until error or a maximum of `retry_attempts` number of times
- `retry_stop = success`: retry until success or a maximum of `retry_attempts` number of times
- `retry_stop = none`: retry a maximum of `retry_attempts` number of times
Only tests with the status of pass, warning, error or failure will be retried.
Other statuses will be ignored and the test will run only once.
This method also works as a convenience wrapper around :py:meth:`run_test`,
providing some default arguments.
"""
if node.is_objectless():
raise AssertionError(
"Cannot run test nodes not using any test objects, here %s" %
node)
retry_stop = node.params.get("retry_stop", "none")
# ignore the retry parameters for nodes that cannot be re-run (need to run at least once)
runs_left = 1 + node.params.get_numeric("retry_attempts",
0) if can_retry else 1
# do not log when the user is not using the retry feature
if runs_left > 1:
logging.debug(
f"Running test with retry_stop={retry_stop} and retry_attempts={runs_left}"
)
assert runs_left >= 1, "retry_attempts cannot be less than zero"
assert retry_stop in [
"none", "error", "success"
], "retry_stop must be one of 'none', 'error' or 'success'"
original_prefix = node.prefix
for r in range(runs_left):
# appending a suffix to retries so we can tell them apart
if r > 0:
node.prefix = original_prefix + f"r{r}"
uid = node.long_prefix
name = node.params["name"]
await self.run_test(self.job, node)
try:
test_result = next(
(x for x in self.job.result.tests
if x["name"].name == name and x["name"].uid == uid))
test_status = test_result["status"]
except StopIteration:
test_status = "ERROR"
logging.info(
"Test result wasn't found and cannot be extracted")
if test_status not in ["PASS", "WARN", "ERROR", "FAIL"]:
# it doesn't make sense to retry with other status
logging.info(
f"Will not attempt to retry test with status {test_status}"
)
break
if retry_stop == "success" and test_status in ["PASS", "WARN"]:
logging.info("Stopping after first successful run")
break
if retry_stop == "error" and test_status in ["ERROR", "FAIL"]:
logging.info("Stopping after first failed run")
break
node.prefix = original_prefix
logging.info(f"Finished running test with status {test_status}")
# no need to log when test was not repeated
if runs_left > 1:
logging.info(f"Finished running test {r+1} times")
# FIX: as VT's retval is broken (always True), we fix its handling here
if test_status in ["ERROR", "FAIL"]:
return False
else:
return True
def _run_available_children(self, node, graph, params):
loop = asyncio.get_event_loop()
# TODO: parallelize only leaf nodes with just this setup node as parent for now
# but later on run together also internal nodes if they don't modify the same vm
run_children = [
n for n in node.cleanup_nodes if len(n.setup_nodes) == 1
and len(n.cleanup_nodes) == 0 and n.should_run
]
while len(run_children) > 0:
current_nodes = run_children[:len(self.slots)]
if len(current_nodes) == 0:
raise ValueError("Not enough container run slots")
logging.debug(
"Traversal advance running in parallel the tests:\n%s",
"\n".join([n.id for n in current_nodes]))
for i, n in enumerate(current_nodes):
current_nodes[i].set_environment(self.job, self.slots[i])
run_children.remove(current_nodes[i])
to_traverse = [
self._traverse_test_node(graph, n, params)
for n in current_nodes
]
loop.run_until_complete(
asyncio.wait_for(asyncio.gather(*to_traverse), self.job.timeout
or None))
def run_traversal(self, graph, params):
"""
Run all user and system defined tests optimizing the setup reuse and
minimizing the repetition of demanded tests.
:param graph: test graph to traverse
:type graph: :py:class:`TestGraph`
:param params: runtime parameters used for extra customization
:type params: {str, str}
:raises: :py:class:`AssertionError` if some traversal assertions are violated
The highest priority is at the setup tests (parents) since the test cannot be
run without the required setup, then the current test, then a single child of
its children (DFS), and finally the other children (tests that can benefit from
the fact that this test/setup was done) followed by the other siblings (tests
benefiting from its parent/setup.
Of course all possible children are restricted by the user-defined "only" and
the number of internal test nodes is minimized for achieving this goal.
"""
shared_roots = graph.get_nodes_by("shared_root", "yes")
assert len(
shared_roots
) == 1, "There can be only exactly one starting node (shared root)"
root = shared_roots[0]
if log.getLogger('graph').level <= log.DEBUG:
traverse_dir = os.path.join(self.job.logdir, "graph_traverse")
if not os.path.exists(traverse_dir):
os.makedirs(traverse_dir)
step = 0
loop = asyncio.get_event_loop()
traverse_path = [root]
while not root.is_cleanup_ready():
next = traverse_path[-1]
if len(traverse_path) > 1:
previous = traverse_path[-2]
else:
# since the loop is discontinued if len(traverse_path) == 0 or root.is_cleanup_ready()
# a valid current node with at least one child is guaranteed
traverse_path.append(next.pick_next_child())
continue
logging.debug(
"At test node %s which is %sready with setup, %sready with cleanup,"
" should %srun, should %sbe cleaned, and %sbe scanned",
next.params["shortname"],
"not " if not next.is_setup_ready() else "",
"not " if not next.is_cleanup_ready() else "",
"not " if not next.should_run else "",
"not " if not next.should_clean else "",
"not " if not next.should_scan else "")
logging.debug(
"Current traverse path/stack:\n%s",
"\n".join([n.params["shortname"] for n in traverse_path]))
# if previous in path is the child of the next, then the path is reversed
# looking for setup so if the next is setup ready and already run, remove
# the previous' reference to it and pop the current next from the path
if previous in next.cleanup_nodes or previous in next.visited_cleanup_nodes:
if next.is_setup_ready():
loop.run_until_complete(
self._traverse_test_node(graph, next, params))
previous.visit_node(next)
traverse_path.pop()
else:
# inverse DFS
traverse_path.append(next.pick_next_parent())
elif previous in next.setup_nodes or previous in next.visited_setup_nodes:
# stop if test is not a setup leaf since parents have higher priority than children
if not next.is_setup_ready():
traverse_path.append(next.pick_next_parent())
continue
else:
loop.run_until_complete(
self._traverse_test_node(graph, next, params))
if next.is_cleanup_ready():
loop.run_until_complete(
self._reverse_test_node(graph, next, params))
for setup in next.visited_setup_nodes:
setup.visit_node(next)
traverse_path.pop()
graph.report_progress()
else:
# parallel pocket lookahead
if next != root and len(self.slots) > 1:
self._run_available_children(next, graph, params)
graph.report_progress()
# normal DFS
traverse_path.append(next.pick_next_child())
else:
raise AssertionError(
"Discontinuous path in the test dependency graph detected")
if log.getLogger('graph').level <= log.DEBUG:
step += 1
graph.visualize(traverse_dir, step)
def run_suite(self, job, test_suite):
"""
Run one or more tests and report with test result.
:param job: job that includes the test suite
:type test_suite: :py:class:`avocado.core.job.Job`
:param test_suite: test suite with some tests to run
:type test_suite: :py:class:`avocado.core.suite.TestSuite`
:returns: a set with types of test failures
:rtype: :py:class:`set`
"""
self.job = job
self.status_repo = StatusRepo(job.unique_id)
self.status_server = StatusServer(
job.config.get('nrunner.status_server_listen'), self.status_repo)
asyncio.ensure_future(self.status_server.serve_forever())
# TODO: this needs more customization
asyncio.ensure_future(self._update_status(job))
graph = self._graph_from_suite(test_suite)
summary = set()
params = self.job.config["param_dict"]
self.tasks = []
self.slots = params.get("slots", "").split(" ")
# TODO: fix other run_traversal calls
try:
graph.visualize(self.job.logdir)
self.run_traversal(graph, params)
if not self.all_tests_ok:
# the summary is a set so only a single failed test is enough
summary.add('FAIL')
except KeyboardInterrupt:
summary.add('INTERRUPTED')
# TODO: the avocado implementation needs a workaround here:
# Wait until all messages may have been processed by the
# status_updater. This should be replaced by a mechanism
# that only waits if there are missing status messages to
# be processed, and, only for a given amount of time.
# Tests with non received status will always show as SKIP
# because of result reconciliation.
time.sleep(0.05)
self.job.result.end_tests()
self.job.funcatexit.run()
self.status_server.close()
signal.signal(signal.SIGTSTP, signal.SIG_IGN)
return summary
"""custom nodes"""
async def run_terminal_node(self, graph, object_name, params):
"""
Run the set of tests necessary for creating a given test object.
:param graph: test graph to run create node from
:type graph: :py:class:`TestGraph`
:param str object_name: name of the test object to be created
:param params: runtime parameters used for extra customization
:type params: {str, str}
:raises: :py:class:`NotImplementedError` if using incompatible installation variant
The current implementation with implicit knowledge on the types of test objects
internal spawns an original (otherwise unmodified) install test.
"""
object_suffix, object_variant = object_name.split(
"-")[:1][0], "-".join(object_name.split("-")[1:])
object_image, object_vm = object_suffix.split("_")
objects = graph.get_objects_by(param_val="^" + object_variant + "$",
subset=graph.get_objects_by(
"images",
object_suffix.split("_")[0]))
vms = [o for o in objects if o.key == "vms"]
assert len(
vms) == 1, "Test object %s's vm not existing or unique in: %s" % (
object_name, objects)
test_object = objects[0]
nodes = graph.get_nodes_by("object_root", object_name)
assert len(
nodes
) == 1, "There should exist one unique root for %s" % object_name
test_node = nodes[0]
if test_object.is_permanent(
) and not test_node.params.get_boolean("create_permanent_vm"):
raise AssertionError(
"Reached a permanent object root for %s due to incorrect setup"
% test_object.suffix)
logging.info("Configuring creation/installation for %s on %s",
object_vm, object_image)
setup_dict = test_node.params.copy()
setup_dict.update({} if params is None else params.copy())
setup_dict.update({
"type": "shared_configure_install",
"check_mode": "rr", # explicit root handling
# overwrite some params inherited from the modified install node
f"set_state_images_{object_image}_{object_vm}": "root",
"start_vm": "no"
})
install_config = test_object.config.get_copy()
install_config.parse_next_batch(base_file="sets.cfg",
ovrwrt_file=param.tests_ovrwrt_file(),
ovrwrt_str=param.re_str("all..noop"),
ovrwrt_dict=setup_dict)
status = await self.run_test_node(
TestNode("0t", install_config, test_node.objects[0]))
if not status:
logging.error("Could not configure the installation for %s on %s",
object_vm, object_image)
return status
logging.info("Installing virtual machine %s", test_object.suffix)
test_node.params["type"] = test_node.params["configure_install"]
return await self.run_test_node(test_node)
"""internals"""
async def _traverse_test_node(self, graph, test_node, params):
"""Run a single test according to user defined policy and state availability."""
if test_node.should_scan:
test_node.scan_states()
test_node.should_scan = False
if test_node.should_run:
# the primary setup nodes need special treatment
if params.get("dry_run", "no") == "yes":
logging.info("Running a dry %s", test_node.params["shortname"])
elif test_node.is_scan_node():
logging.debug("Test run started from the shared root")
elif test_node.is_object_root():
status = await self.run_terminal_node(
graph, test_node.params["object_root"], params)
if not status:
logging.error("Could not perform the installation from %s",
test_node)
else:
# finally, good old running of an actual test
status = await self.run_test_node(test_node, can_retry=True)
if not status:
logging.error("Got nonzero status from the test %s",
test_node)
for test_object in test_node.objects:
object_params = test_object.object_typed_params(
test_node.params)
# if a state was set it is final and the retrieved state was overwritten
object_state = object_params.get(
"set_state", object_params.get("get_state"))
if object_state is not None and object_state != "":
test_object.current_state = object_state
test_node.should_run = False
else:
logging.debug("Skipping test %s", test_node.params["shortname"])
async def _reverse_test_node(self, graph, test_node, params):
"""
Clean up any states that could be created by this node (will be skipped
by default but the states can be removed with "unset_mode=f.").
"""
if test_node.should_clean:
if params.get("dry_run", "no") == "yes":
logging.info("Cleaning a dry %s",
test_node.params["shortname"])
elif test_node.is_shared_root():
logging.debug("Test run ended at the shared root")
elif test_node.produces_setup():
setup_dict = {} if params is None else params.copy()
setup_dict["vm_action"] = "unset"
setup_dict["vms"] = test_node.params["vms"]
# the cleanup will be performed if at least one selected object has a cleanable state
has_selected_object_setup = False
for test_object in test_node.objects:
object_params = test_object.object_typed_params(
test_node.params)
object_state = object_params.get("set_state")
if not object_state:
continue
# avoid running any test unless the user really requires cleanup and such is needed
if object_params.get("unset_mode", "ri")[0] != "f":
continue
# avoid running any test for unselected vms
if test_object.key == "nets":
logging.warning("Net state cleanup is not supported")
continue
vm_name = test_object.suffix if test_object.key == "vms" else test_object.composites[
0].suffix
if vm_name in params.get("vms", param.all_objects("vms")):
has_selected_object_setup = True
else:
continue
# TODO: cannot remove ad-hoc root states, is this even needed?
if test_object.key == "vms":
vm_params = object_params
setup_dict["images_" + vm_name] = vm_params["images"]
for image_name in vm_params.objects("images"):
image_params = vm_params.object_params(image_name)
setup_dict[
f"image_name_{image_name}_{vm_name}"] = image_params[
"image_name"]
setup_dict[
f"image_format_{image_name}_{vm_name}"] = image_params[
"image_format"]
if image_params.get_boolean("create_image", False):
setup_dict[
f"remove_image_{image_name}_{vm_name}"] = "yes"
setup_dict["skip_image_processing"] = "no"
# reverse the state setup for the given test object
unset_suffixes = f"_{test_object.key}_{test_object.suffix}"
unset_suffixes += f"_{vm_name}" if test_object.key == "images" else ""
# NOTE: we are forcing the unset_mode to be the one defined for the test node because
# the unset manual step behaves differently now (all this extra complexity starts from
# the fact that it has different default value which is noninvasive
setup_dict.update({
f"unset_state{unset_suffixes}":
object_state,
f"unset_mode{unset_suffixes}":
object_params.get("unset_mode", "ri")
})
if has_selected_object_setup:
logging.info("Cleaning up %s", test_node)
setup_str = param.re_str("all..internal..manage.unchanged")
net = test_node.objects[0]
forward_config = net.config.get_copy()
forward_config.parse_next_batch(
base_file="sets.cfg",
ovrwrt_file=param.tests_ovrwrt_file(),
ovrwrt_str=setup_str,
ovrwrt_dict=setup_dict)
await self.run_test_node(
TestNode(test_node.prefix + "c", forward_config, net))
else:
logging.info("No need to clean up %s", test_node)
else:
logging.debug("The test %s should not be cleaned up",
test_node.params["shortname"])
def _graph_from_suite(self, test_suite):
"""
Restore a Cartesian graph from the digested list of test object factories.
"""
# HACK: pass the constructed graph to the runner using static attribute hack
# since the currently digested test suite contains factory arguments obtained
# from an irreversible (information destructive) approach
graph = TestGraph.REFERENCE
# validate the test suite refers to the same test graph
assert len(test_suite) == len(graph.nodes)
for node1, node2 in zip(test_suite.tests, graph.nodes):
assert node1.uri == node2.get_runnable().uri
return graph
class Runner(RunnerInterface):
name = 'nrunner'
description = 'nrunner based implementation of job compliant runner'
@staticmethod
def _get_all_runtime_tasks(test_suite):
runtime_tasks = []
no_digits = len(str(len(test_suite)))
status_uris = [test_suite.config.get('nrunner.status_server_uri')]
for index, runnable in enumerate(test_suite.tests, start=1):
# this is all rubbish data
if test_suite.name:
prefix = "{}-{}".format(test_suite.name, index)
else:
prefix = index
test_id = TestID(prefix, runnable.uri, None, no_digits)
task = nrunner.Task(runnable, test_id, status_uris,
nrunner.RUNNERS_REGISTRY_PYTHON_CLASS)
runtime_tasks.append(RuntimeTask(task))
return runtime_tasks
def _start_status_server(self, status_server_listen):
# pylint: disable=W0201
self.status_repo = StatusRepo()
# pylint: disable=W0201
self.status_server = StatusServer(status_server_listen,
self.status_repo)
asyncio.ensure_future(self.status_server.serve_forever())
async def _update_status(self, job):
tasks_by_id = {
str(runtime_task.task.identifier): runtime_task.task
for runtime_task in self.tasks
}
message_handler = MessageHandler()
while True:
try:
(task_id, _, _, index) = \
self.status_repo.status_journal_summary.pop(0)
except IndexError:
await asyncio.sleep(0.05)
continue
message = self.status_repo.get_task_data(task_id, index)
task = tasks_by_id.get(task_id)
message_handler.process_message(message, task, job)
def run_suite(self, job, test_suite):
# pylint: disable=W0201
self.summary = set()
test_suite.tests, _ = nrunner.check_runnables_runner_requirements(
test_suite.tests)
job.result.tests_total = test_suite.size # no support for variants yet
listen = test_suite.config.get('nrunner.status_server_listen')
self._start_status_server(listen)
# pylint: disable=W0201
self.tasks = self._get_all_runtime_tasks(test_suite)
if test_suite.config.get('nrunner.shuffle'):
random.shuffle(self.tasks)
tsm = TaskStateMachine(self.tasks)
spawner_name = test_suite.config.get('nrunner.spawner')
spawner = SpawnerDispatcher(test_suite.config)[spawner_name].obj
max_running = min(test_suite.config.get('nrunner.max_parallel_tasks'),
len(self.tasks))
timeout = test_suite.config.get('task.timeout.running')
workers = [
Worker(state_machine=tsm,
spawner=spawner,
max_running=max_running,
task_timeout=timeout).run() for _ in range(max_running)
]
asyncio.ensure_future(self._update_status(job))
loop = asyncio.get_event_loop()
try:
loop.run_until_complete(
asyncio.wait_for(asyncio.gather(*workers), job.timeout
or None))
except (KeyboardInterrupt, asyncio.TimeoutError):
self.summary.add("INTERRUPTED")
# Wait until all messages may have been processed by the
# status_updater. This should be replaced by a mechanism
# that only waits if there are missing status messages to
# be processed, and, only for a given amount of time.
# Tests with non received status will always show as SKIP
# because of result reconciliation.
loop.run_until_complete(asyncio.sleep(0.05))
job.result.end_tests()
self.status_server.close()
return self.summary