def _wait_child(self):
try:
# Don't block if no child processes have exited
pid, status = os.waitpid(0, os.WNOHANG)
if not pid:
return None
except OSError as exc:
if exc.errno not in (errno.EINTR, errno.ECHILD):
raise
return None
if os.WIFSIGNALED(status):
sig = os.WTERMSIG(status)
LOG.info(_LI('Child %(pid)d killed by signal %(sig)d'),
dict(pid=pid, sig=sig))
else:
code = os.WEXITSTATUS(status)
LOG.info(_LI('Child %(pid)s exited with status %(code)d'),
dict(pid=pid, code=code))
if pid not in self.children:
LOG.warning(_LW('pid %d not in child list'), pid)
return None
wrap = self.children.pop(pid)
wrap.children.remove(pid)
return wrap
def _start_child(self, wrap):
if len(wrap.forktimes) > wrap.workers:
# Limit ourselves to one process a second (over the period of
# number of workers * 1 second). This will allow workers to
# start up quickly but ensure we don't fork off children that
# die instantly too quickly.
if time.time() - wrap.forktimes[0] < wrap.workers:
LOG.info(_LI('Forking too fast, sleeping'))
time.sleep(1)
wrap.forktimes.pop(0)
wrap.forktimes.append(time.time())
pid = os.fork()
if pid == 0:
launcher = self._child_process(wrap.service)
while True:
self._child_process_handle_signal()
status, signo = self._child_wait_for_exit_or_signal(launcher)
if not _is_sighup_and_daemon(signo):
break
launcher.restart()
os._exit(status)
LOG.info(_LI('Started child %d'), pid)
wrap.children.add(pid)
self.children[pid] = wrap
return pid
def _wait_child(self):
try:
# Don't block if no child processes have exited
pid, status = os.waitpid(0, os.WNOHANG)
if not pid:
return None
except OSError as exc:
if exc.errno not in (errno.EINTR, errno.ECHILD):
raise
return None
if os.WIFSIGNALED(status):
sig = os.WTERMSIG(status)
LOG.info(_LI('Child %(pid)d killed by signal %(sig)d'),
dict(pid=pid, sig=sig))
else:
code = os.WEXITSTATUS(status)
LOG.info(_LI('Child %(pid)s exited with status %(code)d'),
dict(pid=pid, code=code))
if pid not in self.children:
LOG.warning(_LW('pid %d not in child list'), pid)
return None
wrap = self.children.pop(pid)
wrap.children.remove(pid)
return wrap
def _start_child(self, wrap):
if len(wrap.forktimes) > wrap.workers:
# Limit ourselves to one process a second (over the period of
# number of workers * 1 second). This will allow workers to
# start up quickly but ensure we don't fork off children that
# die instantly too quickly.
if time.time() - wrap.forktimes[0] < wrap.workers:
LOG.info(_LI('Forking too fast, sleeping'))
time.sleep(1)
wrap.forktimes.pop(0)
wrap.forktimes.append(time.time())
pid = os.fork()
if pid == 0:
launcher = self._child_process(wrap.service)
while True:
self._child_process_handle_signal()
status, signo = self._child_wait_for_exit_or_signal(launcher)
if not _is_sighup_and_daemon(signo):
break
launcher.restart()
os._exit(status)
LOG.info(_LI('Started child %d'), pid)
wrap.children.add(pid)
self.children[pid] = wrap
return pid
def _add_periodic_task(cls, task):
"""Add a periodic task to the list of periodic tasks.
The task should already be decorated by @periodic_task.
:return: whether task was actually enabled
"""
name = task._periodic_name
if task._periodic_spacing < 0:
LOG.info(_LI('Skipping periodic task %(task)s because '
'its interval is negative'),
{'task': name})
return False
if not task._periodic_enabled:
LOG.info(_LI('Skipping periodic task %(task)s because '
'it is disabled'),
{'task': name})
return False
# A periodic spacing of zero indicates that this task should
# be run on the default interval to avoid running too
# frequently.
if task._periodic_spacing == 0:
task._periodic_spacing = DEFAULT_INTERVAL
cls._periodic_tasks.append((name, task))
cls._periodic_spacing[name] = task._periodic_spacing
return True
def wait(self):
"""Loop waiting on children to die and respawning as necessary."""
systemd.notify_once()
LOG.debug('Full set of CONF:')
CONF.log_opt_values(LOG, logging.DEBUG)
try:
while True:
self.handle_signal()
self._respawn_children()
# No signal means that stop was called. Don't clean up here.
if not self.sigcaught:
return
signame = _signo_to_signame(self.sigcaught)
LOG.info(_LI('Caught %s, stopping children'), signame)
if not _is_sighup_and_daemon(self.sigcaught):
break
cfg.CONF.reload_config_files()
for service in set(
[wrap.service for wrap in self.children.values()]):
service.reset()
for pid in self.children:
os.kill(pid, signal.SIGHUP)
self.running = True
self.sigcaught = None
except eventlet.greenlet.GreenletExit:
LOG.info(_LI("Wait called after thread killed. Cleaning up."))
self.stop()
def main():
service.prepare_service(sys.argv)
app = api_app.setup_app()
# Create the WSGI server and start it
host, port = cfg.CONF.host, cfg.CONF.port
srv = simple_server.make_server(host,
port,
app,
handler_class=get_handler_cls())
LOG.info(_LI('Starting server in PID %s') % os.getpid())
LOG.debug("Configuration:")
cfg.CONF.log_opt_values(LOG, std_logging.DEBUG)
if host == '0.0.0.0':
LOG.info(
_LI('serving on 0.0.0.0:%(port)s, '
'view at http://127.0.0.1:%(port)s') % dict(port=port))
else:
LOG.info(
_LI('serving on http://%(host)s:%(port)s') %
dict(host=host, port=port))
srv.serve_forever()
def wait(self):
"""Loop waiting on children to die and respawning as necessary."""
systemd.notify_once()
LOG.debug('Full set of CONF:')
CONF.log_opt_values(LOG, logging.DEBUG)
try:
while True:
self.handle_signal()
self._respawn_children()
# No signal means that stop was called. Don't clean up here.
if not self.sigcaught:
return
signame = _signo_to_signame(self.sigcaught)
LOG.info(_LI('Caught %s, stopping children'), signame)
if not _is_sighup_and_daemon(self.sigcaught):
break
cfg.CONF.reload_config_files()
for service in set(
[wrap.service for wrap in self.children.values()]):
service.reset()
for pid in self.children:
os.kill(pid, signal.SIGHUP)
self.running = True
self.sigcaught = None
except eventlet.greenlet.GreenletExit:
LOG.info(_LI("Wait called after thread killed. Cleaning up."))
self.stop()
def _pipe_watcher(self):
# This will block until the write end is closed when the parent
# dies unexpectedly
self.readpipe.read(1)
LOG.info(_LI('Parent process has died unexpectedly, exiting'))
sys.exit(1)
def _pipe_watcher(self):
# This will block until the write end is closed when the parent
# dies unexpectedly
self.readpipe.read(1)
LOG.info(_LI('Parent process has died unexpectedly, exiting'))
sys.exit(1)
def timerfunc(*args, **kwargs):
context = None
acquireLock()
tests = objects.Cpulse.list(context)
releaseLock()
delete_old_entries()
for test in tests:
LOG.debug(_LI('Dumping REPFUNCTION %s') % test['uuid'])
if test['state'] == 'created' and test['testtype'] == 'manual':
methodtocall = getattr(sys.modules[__name__], 'test_run')
# methodtocall()
testthr = threading.Thread(name=test['name'],
target=methodtocall,
kwargs={'test': test})
testthreads.append(testthr)
testthr.start()
LOG.debug(_LI('REPFUNCTION, exec test %s') % test['name'])
def timerfunc(*args, **kwargs):
context = None
acquireLock()
tests = objects.Cpulse.list(context)
releaseLock()
delete_old_entries()
for test in tests:
LOG.debug(_LI('Dumping REPFUNCTION %s') % test['uuid'])
if test['state'] == 'scheduled' and test['testtype'] == 'manual':
methodtocall = getattr(sys.modules[__name__], 'test_run')
# methodtocall()
testthr = threading.Thread(name=test['name'],
target=methodtocall,
kwargs={'test': test})
testthreads.append(testthr)
testthr.start()
LOG.debug(_LI('REPFUNCTION, exec test %s') % test['name'])
def _service_admin_creds(self):
# Import auth_token to have keystone_authtoken settings setup.
importutils.import_module('keystonemiddleware.auth_token')
creds = {
'username': cfg.CONF.keystone_authtoken.admin_user,
'password': cfg.CONF.keystone_authtoken.admin_password,
'auth_url': self.v3_endpoint,
'endpoint': self.v3_endpoint,
'project_name': cfg.CONF.keystone_authtoken.admin_tenant_name}
LOG.info(_LI('admin creds %s') % creds)
return creds
def _service_admin_creds(self):
# Import auth_token to have keystone_authtoken settings setup.
importutils.import_module('keystonemiddleware.auth_token')
creds = {
'username': cfg.CONF.keystone_authtoken.admin_user,
'password': cfg.CONF.keystone_authtoken.admin_password,
'auth_url': self.v3_endpoint,
'endpoint': self.v3_endpoint,
'project_name': cfg.CONF.keystone_authtoken.admin_tenant_name}
LOG.info(_LI('admin creds %s') % creds)
return creds
def _service_admin_creds(self):
# Import auth_token to have keystone_authtoken settings setup.
importutils.import_module("keystonemiddleware.auth_token")
creds = {
"username": cfg.CONF.keystone_authtoken.admin_user,
"password": cfg.CONF.keystone_authtoken.admin_password,
"auth_url": self.v3_endpoint,
"endpoint": self.v3_endpoint,
"project_name": cfg.CONF.keystone_authtoken.admin_tenant_name,
}
LOG.info(_LI("admin creds %s") % creds)
return creds
def main():
service.prepare_service(sys.argv)
app = api_app.setup_app()
# Create the WSGI server and start it
host, port = cfg.CONF.host, cfg.CONF.port
srv = simple_server.make_server(host, port, app)
LOG.info(_LI('Starting server in PID %s') % os.getpid())
LOG.debug("Configuration:")
cfg.CONF.log_opt_values(LOG, std_logging.DEBUG)
if host == '0.0.0.0':
LOG.info(_LI('serving on 0.0.0.0:%(port)s, '
'view at http://127.0.0.1:%(port)s') %
dict(port=port))
else:
LOG.info(_LI('serving on http://%(host)s:%(port)s') %
dict(host=host, port=port))
srv.serve_forever()
def stop(self):
"""Terminate child processes and wait on each."""
self.running = False
for pid in self.children:
try:
os.kill(pid, signal.SIGTERM)
except OSError as exc:
if exc.errno != errno.ESRCH:
raise
# Wait for children to die
if self.children:
LOG.info(_LI('Waiting on %d children to exit'), len(self.children))
while self.children:
self._wait_child()
def create(self, context=None):
"""Create a Cpulse record in the DB.
:param context: Security context. NOTE: This should only
be used internally by the indirection_api.
Unfortunately, RPC requires context as the first
argument, even though we don't use it.
A context should be set when instantiating the
object, e.g.: Cpulse(context)
"""
values = self.obj_get_changes()
LOG.info(_LI('Dumping CREATE test datastructure %s') % str(values))
db = self.dbapi.create_test(values)
self._from_db_object(self, db)
def create(self, context=None):
"""Create a Cpulse record in the DB.
:param context: Security context. NOTE: This should only
be used internally by the indirection_api.
Unfortunately, RPC requires context as the first
argument, even though we don't use it.
A context should be set when instantiating the
object, e.g.: Cpulse(context)
"""
values = self.obj_get_changes()
LOG.info(_LI('Dumping CREATE test datastructure %s') % str(values))
db = self.dbapi.create_test(values)
self._from_db_object(self, db)
def stop(self):
"""Terminate child processes and wait on each."""
self.running = False
for pid in self.children:
try:
os.kill(pid, signal.SIGTERM)
except OSError as exc:
if exc.errno != errno.ESRCH:
raise
# Wait for children to die
if self.children:
LOG.info(_LI('Waiting on %d children to exit'), len(self.children))
while self.children:
self._wait_child()
def initialize_if_enabled():
backdoor_locals = {
'exit': _dont_use_this, # So we don't exit the entire process
'quit': _dont_use_this, # So we don't exit the entire process
'fo': _find_objects,
'pgt': _print_greenthreads,
'pnt': _print_nativethreads,
}
if CONF.backdoor_port is None:
return None
start_port, end_port = _parse_port_range(str(CONF.backdoor_port))
# NOTE(johannes): The standard sys.displayhook will print the value of
# the last expression and set it to __builtin__._, which overwrites
# the __builtin__._ that gettext sets. Let's switch to using pprint
# since it won't interact poorly with gettext, and it's easier to
# read the output too.
def displayhook(val):
if val is not None:
pprint.pprint(val)
sys.displayhook = displayhook
sock = _listen('localhost', start_port, end_port, eventlet.listen)
# In the case of backdoor port being zero, a port number is assigned by
# listen(). In any case, pull the port number out here.
port = sock.getsockname()[1]
LOG.info(
_LI('Eventlet backdoor listening on %(port)s for process %(pid)d'), {
'port': port,
'pid': os.getpid()
})
eventlet.spawn_n(eventlet.backdoor.backdoor_server,
sock,
locals=backdoor_locals)
return port
def _wait_for_exit_or_signal(self, ready_callback=None):
status = None
signo = 0
LOG.debug('Full set of CONF:')
CONF.log_opt_values(LOG, logging.DEBUG)
try:
if ready_callback:
ready_callback()
super(ServiceLauncher, self).wait()
except SignalExit as exc:
signame = _signo_to_signame(exc.signo)
LOG.info(_LI('Caught %s, exiting'), signame)
status = exc.code
signo = exc.signo
except SystemExit as exc:
status = exc.code
finally:
self.stop()
return status, signo
def _wait_for_exit_or_signal(self, ready_callback=None):
status = None
signo = 0
LOG.debug('Full set of CONF:')
CONF.log_opt_values(LOG, logging.DEBUG)
try:
if ready_callback:
ready_callback()
super(ServiceLauncher, self).wait()
except SignalExit as exc:
signame = _signo_to_signame(exc.signo)
LOG.info(_LI('Caught %s, exiting'), signame)
status = exc.code
signo = exc.signo
except SystemExit as exc:
status = exc.code
finally:
self.stop()
return status, signo
def _child_wait_for_exit_or_signal(self, launcher):
status = 0
signo = 0
# NOTE(johannes): All exceptions are caught to ensure this
# doesn't fallback into the loop spawning children. It would
# be bad for a child to spawn more children.
try:
launcher.wait()
except SignalExit as exc:
signame = _signo_to_signame(exc.signo)
LOG.info(_LI('Child caught %s, exiting'), signame)
status = exc.code
signo = exc.signo
except SystemExit as exc:
status = exc.code
except BaseException:
LOG.exception(_LE('Unhandled exception'))
status = 2
finally:
launcher.stop()
return status, signo
def _child_wait_for_exit_or_signal(self, launcher):
status = 0
signo = 0
# NOTE(johannes): All exceptions are caught to ensure this
# doesn't fallback into the loop spawning children. It would
# be bad for a child to spawn more children.
try:
launcher.wait()
except SignalExit as exc:
signame = _signo_to_signame(exc.signo)
LOG.info(_LI('Child caught %s, exiting'), signame)
status = exc.code
signo = exc.signo
except SystemExit as exc:
status = exc.code
except BaseException:
LOG.exception(_LE('Unhandled exception'))
status = 2
finally:
launcher.stop()
return status, signo
def initialize_if_enabled():
backdoor_locals = {
'exit': _dont_use_this, # So we don't exit the entire process
'quit': _dont_use_this, # So we don't exit the entire process
'fo': _find_objects,
'pgt': _print_greenthreads,
'pnt': _print_nativethreads,
}
if CONF.backdoor_port is None:
return None
start_port, end_port = _parse_port_range(str(CONF.backdoor_port))
# NOTE(johannes): The standard sys.displayhook will print the value of
# the last expression and set it to __builtin__._, which overwrites
# the __builtin__._ that gettext sets. Let's switch to using pprint
# since it won't interact poorly with gettext, and it's easier to
# read the output too.
def displayhook(val):
if val is not None:
pprint.pprint(val)
sys.displayhook = displayhook
sock = _listen('localhost', start_port, end_port, eventlet.listen)
# In the case of backdoor port being zero, a port number is assigned by
# listen(). In any case, pull the port number out here.
port = sock.getsockname()[1]
LOG.info(
_LI('Eventlet backdoor listening on %(port)s for process %(pid)d'),
{'port': port, 'pid': os.getpid()}
)
eventlet.spawn_n(eventlet.backdoor.backdoor_server, sock,
locals=backdoor_locals)
return port
def test_run(**kwargs):
LOG.debug(_LI('Running Openstack test%s') % kwargs['test']['name'])
test_manager.run(**kwargs)
def test_run(**kwargs):
LOG.debug(_LI('Running Openstack test%s') % kwargs['test']['name'])
test_manager.run(**kwargs)
def launch_service(self, service, workers=1):
wrap = ServiceWrapper(service, workers)
LOG.info(_LI('Starting %d workers'), wrap.workers)
while self.running and len(wrap.children) < wrap.workers:
self._start_child(wrap)
def launch_service(self, service, workers=1):
wrap = ServiceWrapper(service, workers)
LOG.info(_LI('Starting %d workers'), wrap.workers)
while self.running and len(wrap.children) < wrap.workers:
self._start_child(wrap)
