1
0
Fork 0
mirror of https://github.com/ansible-collections/community.general.git synced 2024-09-14 20:13:21 +02:00
community.general/test/runner/lib/cli.py

830 lines
30 KiB
Python
Raw Normal View History

"""Test runner for all Ansible tests."""
from __future__ import absolute_import, print_function
import errno
import os
import resource
import sys
from lib.util import (
ApplicationError,
display,
raw_command,
get_docker_completion,
generate_pip_command,
read_lines_without_comments,
)
from lib.delegation import (
delegate,
)
from lib.executor import (
command_posix_integration,
command_network_integration,
command_windows_integration,
command_units,
command_shell,
SUPPORTED_PYTHON_VERSIONS,
ApplicationWarning,
Delegate,
generate_pip_install,
check_startup,
)
from lib.config import (
IntegrationConfig,
PosixIntegrationConfig,
WindowsIntegrationConfig,
NetworkIntegrationConfig,
SanityConfig,
UnitsConfig,
ShellConfig,
)
from lib.env import (
EnvConfig,
command_env,
)
from lib.sanity import (
command_sanity,
sanity_init,
sanity_get_tests,
)
from lib.target import (
find_target_completion,
walk_posix_integration_targets,
walk_network_integration_targets,
walk_windows_integration_targets,
walk_units_targets,
walk_sanity_targets,
)
from lib.core_ci import (
AWS_ENDPOINTS,
)
from lib.cloud import (
initialize_cloud_plugins,
)
import lib.cover
def main():
"""Main program function."""
try:
git_root = os.path.abspath(os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', '..', '..'))
os.chdir(git_root)
initialize_cloud_plugins()
sanity_init()
args = parse_args()
config = args.config(args)
display.verbosity = config.verbosity
display.truncate = config.truncate
display.redact = config.redact
display.color = config.color
display.info_stderr = (isinstance(config, SanityConfig) and config.lint) or (isinstance(config, IntegrationConfig) and config.list_targets)
2017-01-24 19:31:39 +01:00
check_startup()
# to achieve a consistent nofile ulimit, set to 16k here, this can affect performance in subprocess.Popen when
# being called with close_fds=True on Python (8x the time on some environments)
nofile_limit = 16 * 1024
current_limit = resource.getrlimit(resource.RLIMIT_NOFILE)
new_limit = (nofile_limit, nofile_limit)
if current_limit > new_limit:
display.info('RLIMIT_NOFILE: %s -> %s' % (current_limit, new_limit), verbosity=2)
resource.setrlimit(resource.RLIMIT_NOFILE, (nofile_limit, nofile_limit))
else:
display.info('RLIMIT_NOFILE: %s' % (current_limit, ), verbosity=2)
try:
args.func(config)
except Delegate as ex:
delegate(config, ex.exclude, ex.require, ex.integration_targets)
display.review_warnings()
except ApplicationWarning as ex:
display.warning(u'%s' % ex)
exit(0)
except ApplicationError as ex:
display.error(u'%s' % ex)
exit(1)
except KeyboardInterrupt:
exit(2)
except IOError as ex:
if ex.errno == errno.EPIPE:
exit(3)
raise
def parse_args():
"""Parse command line arguments."""
try:
import argparse
except ImportError:
if '--requirements' not in sys.argv:
raise
raw_command(generate_pip_install(generate_pip_command(sys.executable), 'ansible-test'))
import argparse
try:
import argcomplete
except ImportError:
argcomplete = None
if argcomplete:
epilog = 'Tab completion available using the "argcomplete" python package.'
else:
epilog = 'Install the "argcomplete" python package to enable tab completion.'
parser = argparse.ArgumentParser(epilog=epilog)
common = argparse.ArgumentParser(add_help=False)
common.add_argument('-e', '--explain',
action='store_true',
help='explain commands that would be executed')
common.add_argument('-v', '--verbose',
dest='verbosity',
action='count',
default=0,
help='display more output')
common.add_argument('--color',
metavar='COLOR',
nargs='?',
help='generate color output: %(choices)s',
choices=('yes', 'no', 'auto'),
const='yes',
default='auto')
common.add_argument('--debug',
action='store_true',
help='run ansible commands in debug mode')
common.add_argument('--truncate',
dest='truncate',
metavar='COLUMNS',
type=int,
default=display.columns,
help='truncate some long output (0=disabled) (default: auto)')
common.add_argument('--redact',
dest='redact',
action='store_true',
help='redact sensitive values in output')
test = argparse.ArgumentParser(add_help=False, parents=[common])
test.add_argument('include',
metavar='TARGET',
nargs='*',
help='test the specified target').completer = complete_target
test.add_argument('--include',
metavar='TARGET',
action='append',
help='include the specified target').completer = complete_target
test.add_argument('--exclude',
metavar='TARGET',
action='append',
help='exclude the specified target').completer = complete_target
test.add_argument('--require',
metavar='TARGET',
action='append',
help='require the specified target').completer = complete_target
test.add_argument('--coverage',
action='store_true',
help='analyze code coverage when running tests')
test.add_argument('--coverage-label',
default='',
help='label to include in coverage output file names')
test.add_argument('--metadata',
help=argparse.SUPPRESS)
add_changes(test, argparse)
add_environments(test)
integration = argparse.ArgumentParser(add_help=False, parents=[test])
integration.add_argument('--python',
metavar='VERSION',
2017-06-30 23:53:32 +02:00
choices=SUPPORTED_PYTHON_VERSIONS + ('default',),
help='python version: %s' % ', '.join(SUPPORTED_PYTHON_VERSIONS))
integration.add_argument('--start-at',
metavar='TARGET',
help='start at the specified target').completer = complete_target
integration.add_argument('--start-at-task',
metavar='TASK',
help='start at the specified task')
integration.add_argument('--tags',
metavar='TAGS',
help='only run plays and tasks tagged with these values')
integration.add_argument('--skip-tags',
metavar='TAGS',
help='only run plays and tasks whose tags do not match these values')
integration.add_argument('--diff',
action='store_true',
help='show diff output')
integration.add_argument('--allow-destructive',
action='store_true',
help='allow destructive tests (--local and --tox only)')
integration.add_argument('--allow-root',
action='store_true',
help='allow tests requiring root when not root')
integration.add_argument('--allow-disabled',
action='store_true',
help='allow tests which have been marked as disabled')
integration.add_argument('--allow-unstable',
action='store_true',
help='allow tests which have been marked as unstable')
integration.add_argument('--allow-unstable-changed',
action='store_true',
help='allow tests which have been marked as unstable when focused changes are detected')
integration.add_argument('--allow-unsupported',
action='store_true',
help='allow tests which have been marked as unsupported')
integration.add_argument('--retry-on-error',
action='store_true',
help='retry failed test with increased verbosity')
integration.add_argument('--continue-on-error',
action='store_true',
help='continue after failed test')
integration.add_argument('--debug-strategy',
action='store_true',
help='run test playbooks using the debug strategy')
integration.add_argument('--changed-all-target',
metavar='TARGET',
default='all',
help='target to run when all tests are needed')
integration.add_argument('--changed-all-mode',
metavar='MODE',
choices=('default', 'include', 'exclude'),
help='include/exclude behavior with --changed-all-target: %(choices)s')
integration.add_argument('--list-targets',
action='store_true',
help='list matching targets instead of running tests')
subparsers = parser.add_subparsers(metavar='COMMAND')
subparsers.required = True # work-around for python 3 bug which makes subparsers optional
posix_integration = subparsers.add_parser('integration',
parents=[integration],
help='posix integration tests')
posix_integration.set_defaults(func=command_posix_integration,
targets=walk_posix_integration_targets,
config=PosixIntegrationConfig)
add_extra_docker_options(posix_integration)
add_httptester_options(posix_integration, argparse)
network_integration = subparsers.add_parser('network-integration',
parents=[integration],
help='network integration tests')
network_integration.set_defaults(func=command_network_integration,
targets=walk_network_integration_targets,
config=NetworkIntegrationConfig)
add_extra_docker_options(network_integration, integration=False)
network_integration.add_argument('--platform',
metavar='PLATFORM',
action='append',
help='network platform/version').completer = complete_network_platform
network_integration.add_argument('--inventory',
metavar='PATH',
help='path to inventory used for tests')
network_integration.add_argument('--testcase',
metavar='TESTCASE',
help='limit a test to a specified testcase').completer = complete_network_testcase
windows_integration = subparsers.add_parser('windows-integration',
parents=[integration],
help='windows integration tests')
windows_integration.set_defaults(func=command_windows_integration,
targets=walk_windows_integration_targets,
config=WindowsIntegrationConfig)
add_extra_docker_options(windows_integration, integration=False)
add_httptester_options(windows_integration, argparse)
windows_integration.add_argument('--windows',
metavar='VERSION',
action='append',
help='windows version').completer = complete_windows
units = subparsers.add_parser('units',
parents=[test],
help='unit tests')
units.set_defaults(func=command_units,
targets=walk_units_targets,
config=UnitsConfig)
units.add_argument('--python',
metavar='VERSION',
choices=SUPPORTED_PYTHON_VERSIONS + ('default',),
help='python version: %s' % ', '.join(SUPPORTED_PYTHON_VERSIONS))
units.add_argument('--collect-only',
action='store_true',
help='collect tests but do not execute them')
units.add_argument('--requirements-mode',
choices=('only', 'skip'),
help=argparse.SUPPRESS)
add_extra_docker_options(units, integration=False)
sanity = subparsers.add_parser('sanity',
parents=[test],
help='sanity tests')
sanity.set_defaults(func=command_sanity,
targets=walk_sanity_targets,
config=SanityConfig)
sanity.add_argument('--test',
metavar='TEST',
action='append',
choices=[test.name for test in sanity_get_tests()],
help='tests to run').completer = complete_sanity_test
sanity.add_argument('--skip-test',
metavar='TEST',
action='append',
choices=[test.name for test in sanity_get_tests()],
help='tests to skip').completer = complete_sanity_test
sanity.add_argument('--allow-disabled',
action='store_true',
help='allow tests to run which are disabled by default')
sanity.add_argument('--list-tests',
action='store_true',
help='list available tests')
sanity.add_argument('--python',
metavar='VERSION',
2017-06-30 23:53:32 +02:00
choices=SUPPORTED_PYTHON_VERSIONS + ('default',),
help='python version: %s' % ', '.join(SUPPORTED_PYTHON_VERSIONS))
sanity.add_argument('--base-branch',
help=argparse.SUPPRESS)
add_lint(sanity)
add_extra_docker_options(sanity, integration=False)
shell = subparsers.add_parser('shell',
parents=[common],
help='open an interactive shell')
shell.set_defaults(func=command_shell,
config=ShellConfig)
shell.add_argument('--raw',
action='store_true',
help='direct to shell with no setup')
add_environments(shell, tox_version=True)
add_extra_docker_options(shell)
add_httptester_options(shell, argparse)
coverage_common = argparse.ArgumentParser(add_help=False, parents=[common])
add_environments(coverage_common, tox_version=True, tox_only=True)
coverage = subparsers.add_parser('coverage',
help='code coverage management and reporting')
coverage_subparsers = coverage.add_subparsers(metavar='COMMAND')
coverage_subparsers.required = True # work-around for python 3 bug which makes subparsers optional
coverage_combine = coverage_subparsers.add_parser('combine',
parents=[coverage_common],
help='combine coverage data and rewrite remote paths')
coverage_combine.set_defaults(func=lib.cover.command_coverage_combine,
config=lib.cover.CoverageConfig)
add_extra_coverage_options(coverage_combine)
coverage_erase = coverage_subparsers.add_parser('erase',
parents=[coverage_common],
help='erase coverage data files')
coverage_erase.set_defaults(func=lib.cover.command_coverage_erase,
config=lib.cover.CoverageConfig)
coverage_report = coverage_subparsers.add_parser('report',
parents=[coverage_common],
help='generate console coverage report')
coverage_report.set_defaults(func=lib.cover.command_coverage_report,
config=lib.cover.CoverageReportConfig)
coverage_report.add_argument('--show-missing',
action='store_true',
help='show line numbers of statements not executed')
coverage_report.add_argument('--include',
metavar='PAT1,PAT2,...',
help='include only files whose paths match one of these '
'patterns. Accepts shell-style wildcards, which must be '
'quoted.')
coverage_report.add_argument('--omit',
metavar='PAT1,PAT2,...',
help='omit files whose paths match one of these patterns. '
'Accepts shell-style wildcards, which must be quoted.')
add_extra_coverage_options(coverage_report)
coverage_html = coverage_subparsers.add_parser('html',
parents=[coverage_common],
help='generate html coverage report')
coverage_html.set_defaults(func=lib.cover.command_coverage_html,
config=lib.cover.CoverageConfig)
add_extra_coverage_options(coverage_html)
coverage_xml = coverage_subparsers.add_parser('xml',
parents=[coverage_common],
help='generate xml coverage report')
coverage_xml.set_defaults(func=lib.cover.command_coverage_xml,
config=lib.cover.CoverageConfig)
add_extra_coverage_options(coverage_xml)
env = subparsers.add_parser('env',
parents=[common],
help='show information about the test environment')
env.set_defaults(func=command_env,
config=EnvConfig)
env.add_argument('--show',
action='store_true',
help='show environment on stdout')
env.add_argument('--dump',
action='store_true',
help='dump environment to disk')
if argcomplete:
argcomplete.autocomplete(parser, always_complete_options=False, validator=lambda i, k: True)
args = parser.parse_args()
if args.explain and not args.verbosity:
args.verbosity = 1
if args.color == 'yes':
args.color = True
elif args.color == 'no':
args.color = False
else:
args.color = sys.stdout.isatty()
return args
def add_lint(parser):
"""
:type parser: argparse.ArgumentParser
"""
parser.add_argument('--lint',
action='store_true',
help='write lint output to stdout, everything else stderr')
parser.add_argument('--junit',
action='store_true',
help='write test failures to junit xml files')
parser.add_argument('--failure-ok',
action='store_true',
help='exit successfully on failed tests after saving results')
def add_changes(parser, argparse):
"""
:type parser: argparse.ArgumentParser
:type argparse: argparse
"""
parser.add_argument('--changed', action='store_true', help='limit targets based on changes')
changes = parser.add_argument_group(title='change detection arguments')
changes.add_argument('--tracked', action='store_true', help=argparse.SUPPRESS)
changes.add_argument('--untracked', action='store_true', help='include untracked files')
changes.add_argument('--ignore-committed', dest='committed', action='store_false', help='exclude committed files')
changes.add_argument('--ignore-staged', dest='staged', action='store_false', help='exclude staged files')
changes.add_argument('--ignore-unstaged', dest='unstaged', action='store_false', help='exclude unstaged files')
changes.add_argument('--changed-from', metavar='PATH', help=argparse.SUPPRESS)
changes.add_argument('--changed-path', metavar='PATH', action='append', help=argparse.SUPPRESS)
def add_environments(parser, tox_version=False, tox_only=False):
"""
:type parser: argparse.ArgumentParser
:type tox_version: bool
:type tox_only: bool
"""
parser.add_argument('--requirements',
action='store_true',
help='install command requirements')
environments = parser.add_mutually_exclusive_group()
environments.add_argument('--local',
action='store_true',
help='run from the local environment')
if tox_version:
environments.add_argument('--tox',
metavar='VERSION',
nargs='?',
default=None,
const='.'.join(str(i) for i in sys.version_info[:2]),
choices=SUPPORTED_PYTHON_VERSIONS,
help='run from a tox virtualenv: %s' % ', '.join(SUPPORTED_PYTHON_VERSIONS))
else:
environments.add_argument('--tox',
action='store_true',
help='run from a tox virtualenv')
tox = parser.add_argument_group(title='tox arguments')
tox.add_argument('--tox-sitepackages',
action='store_true',
help='allow access to globally installed packages')
if tox_only:
environments.set_defaults(
docker=None,
remote=None,
remote_stage=None,
remote_provider=None,
remote_aws_region=None,
remote_terminate=None,
)
return
environments.add_argument('--docker',
metavar='IMAGE',
nargs='?',
default=None,
const='default',
help='run from a docker container').completer = complete_docker
environments.add_argument('--remote',
metavar='PLATFORM',
default=None,
help='run from a remote instance').completer = complete_remote_shell if parser.prog.endswith(' shell') else complete_remote
remote = parser.add_argument_group(title='remote arguments')
remote.add_argument('--remote-stage',
metavar='STAGE',
help='remote stage to use: %(choices)s',
choices=['prod', 'dev'],
default='prod')
remote.add_argument('--remote-provider',
metavar='PROVIDER',
help='remote provider to use: %(choices)s',
choices=['default', 'aws', 'azure', 'parallels'],
default='default')
remote.add_argument('--remote-aws-region',
metavar='REGION',
help='remote aws region to use: %(choices)s (default: auto)',
choices=sorted(AWS_ENDPOINTS),
default=None)
remote.add_argument('--remote-terminate',
metavar='WHEN',
help='terminate remote instance: %(choices)s (default: %(default)s)',
choices=['never', 'always', 'success'],
default='never')
def add_extra_coverage_options(parser):
"""
:type parser: argparse.ArgumentParser
"""
parser.add_argument('--group-by',
metavar='GROUP',
action='append',
choices=lib.cover.COVERAGE_GROUPS,
help='group output by: %s' % ', '.join(lib.cover.COVERAGE_GROUPS))
parser.add_argument('--all',
action='store_true',
help='include all python source files')
parser.add_argument('--stub',
action='store_true',
help='generate empty report of all python source files')
def add_httptester_options(parser, argparse):
"""
:type parser: argparse.ArgumentParser
:type argparse: argparse
"""
group = parser.add_mutually_exclusive_group()
group.add_argument('--httptester',
metavar='IMAGE',
default='quay.io/ansible/http-test-container:1.0.0',
help='docker image to use for the httptester container')
group.add_argument('--disable-httptester',
dest='httptester',
action='store_const',
const='',
help='do not use the httptester container')
parser.add_argument('--inject-httptester',
action='store_true',
help=argparse.SUPPRESS) # internal use only
def add_extra_docker_options(parser, integration=True):
"""
:type parser: argparse.ArgumentParser
:type integration: bool
"""
docker = parser.add_argument_group(title='docker arguments')
docker.add_argument('--docker-no-pull',
action='store_false',
dest='docker_pull',
help='do not explicitly pull the latest docker images')
docker.add_argument('--docker-keep-git',
action='store_true',
help='transfer git related files into the docker container')
2018-08-30 20:36:57 +02:00
docker.add_argument('--docker-seccomp',
metavar='SC',
choices=('default', 'unconfined'),
default=None,
help='set seccomp confinement for the test container: %(choices)s')
if not integration:
return
docker.add_argument('--docker-privileged',
action='store_true',
help='run docker container in privileged mode')
docker.add_argument('--docker-memory',
help='memory limit for docker in bytes', type=int)
def complete_target(prefix, parsed_args, **_):
"""
:type prefix: unicode
:type parsed_args: any
:rtype: list[str]
"""
return find_target_completion(parsed_args.targets, prefix)
def complete_remote(prefix, parsed_args, **_):
"""
:type prefix: unicode
:type parsed_args: any
:rtype: list[str]
"""
del parsed_args
images = read_lines_without_comments('test/runner/completion/remote.txt', remove_blank_lines=True)
return [i for i in images if i.startswith(prefix)]
def complete_remote_shell(prefix, parsed_args, **_):
"""
:type prefix: unicode
:type parsed_args: any
:rtype: list[str]
"""
del parsed_args
images = read_lines_without_comments('test/runner/completion/remote.txt', remove_blank_lines=True)
# 2008 doesn't support SSH so we do not add to the list of valid images
images.extend(["windows/%s" % i for i in read_lines_without_comments('test/runner/completion/windows.txt', remove_blank_lines=True) if i != '2008'])
return [i for i in images if i.startswith(prefix)]
def complete_docker(prefix, parsed_args, **_):
"""
:type prefix: unicode
:type parsed_args: any
:rtype: list[str]
"""
del parsed_args
images = sorted(get_docker_completion().keys())
return [i for i in images if i.startswith(prefix)]
def complete_windows(prefix, parsed_args, **_):
"""
:type prefix: unicode
:type parsed_args: any
:rtype: list[str]
"""
images = read_lines_without_comments('test/runner/completion/windows.txt', remove_blank_lines=True)
return [i for i in images if i.startswith(prefix) and (not parsed_args.windows or i not in parsed_args.windows)]
def complete_network_platform(prefix, parsed_args, **_):
"""
:type prefix: unicode
:type parsed_args: any
:rtype: list[str]
"""
images = read_lines_without_comments('test/runner/completion/network.txt', remove_blank_lines=True)
return [i for i in images if i.startswith(prefix) and (not parsed_args.platform or i not in parsed_args.platform)]
def complete_network_testcase(prefix, parsed_args, **_):
"""
:type prefix: unicode
:type parsed_args: any
:rtype: list[str]
"""
testcases = []
# since testcases are module specific, don't autocomplete if more than one
# module is specidied
if len(parsed_args.include) != 1:
return []
test_dir = 'test/integration/targets/%s/tests' % parsed_args.include[0]
connections = os.listdir(test_dir)
for conn in connections:
if os.path.isdir(os.path.join(test_dir, conn)):
for testcase in os.listdir(os.path.join(test_dir, conn)):
if testcase.startswith(prefix):
testcases.append(testcase.split('.')[0])
return testcases
def complete_sanity_test(prefix, parsed_args, **_):
"""
:type prefix: unicode
:type parsed_args: any
:rtype: list[str]
"""
del parsed_args
tests = sorted(t.name for t in sanity_get_tests())
return [i for i in tests if i.startswith(prefix)]