def register_server(url, ca, secret, address=None):
if _service_is_running() is True:
console_log.warning(
"chroma-agent service was running before registration, stopping.")
agent_service.stop()
crypto = Crypto(config.path)
# Call delete in case we are over-writing a previous configuration that wasn't removed properly
crypto.delete()
crypto.install_authority(ca)
agent_client = AgentClient(url + "register/%s/" % secret,
ActionPluginManager(), DevicePluginManager(),
ServerProperties(), crypto)
registration_result = agent_client.register(address)
crypto.install_certificate(registration_result['certificate'])
config.set('settings', 'server', {'url': url})
console_log.info("Enabling chroma-agent service")
agent_service.enable()
console_log.info("Starting chroma-agent service")
agent_service.start()
return registration_result
def main():
"""handle unexpected exceptions"""
parser = argparse.ArgumentParser(
description="Integrated Manager for Lustre software Agent")
parser.add_argument("--publish-zconf", action="store_true")
parser.parse_args()
signal.signal(signal.SIGHUP, signal.SIG_IGN)
daemon_log_setup()
console_log_setup()
daemon_log.info("Starting")
try:
daemon_log.info("Entering main loop")
try:
url = urljoin(os.environ["IML_MANAGER_URL"], "agent/message/")
except KeyError as e:
daemon_log.error(
"No configuration found (must be registered before running the agent service), "
"details: %s" % e)
return
if config.profile_managed is False:
# This is kind of terrible. The design of DevicePluginManager is
# such that it can be called with either class methods or
# instantiated and then called with instance methods. As such,
# we can't pass in a list of excluded plugins to the instance
# constructor. Well, we could, but it would only work some
# of the time and that would be even more awful.
import chroma_agent.plugin_manager
chroma_agent.plugin_manager.EXCLUDED_PLUGINS += ['corosync']
agent_client = AgentClient(url, ActionPluginManager(),
DevicePluginManager(), ServerProperties(),
Crypto(ENV_PATH))
def teardown_callback(*args, **kwargs):
agent_client.stop()
agent_client.join()
[function() for function in agent_daemon_teardown_functions]
signal.signal(signal.SIGINT, teardown_callback)
signal.signal(signal.SIGTERM, teardown_callback)
signal.signal(signal.SIGUSR1, decrease_loglevel)
signal.signal(signal.SIGUSR2, increase_loglevel)
# Call any agent daemon startup methods that were registered.
[function() for function in agent_daemon_startup_functions]
agent_client.start()
# Waking-wait to pick up signals
while not agent_client.stopped.is_set():
agent_client.stopped.wait(timeout=10)
agent_client.join()
except Exception, e:
backtrace = '\n'.join(traceback.format_exception(*(sys.exc_info())))
daemon_log.error("Unhandled exception: %s" % backtrace)
def reregister_server(url, address):
""" Update manager url and register agent address with manager """
if _service_is_running() is True:
console_log.warning(
"chroma-agent service was running before registration, stopping.")
agent_service.stop()
config.set('settings', 'server', {'url': url})
crypto = Crypto(config.path)
agent_client = AgentClient(url + 'reregister/', ActionPluginManager(),
DevicePluginManager(), ServerProperties(),
crypto)
data = {'address': address, 'fqdn': agent_client._fqdn}
try:
result = agent_client.post(data)
except HttpError:
console_log.error("Reregistration failed to %s with request %s" %
(agent_client.url, data))
raise
console_log.info("Starting chroma-agent service")
agent_service.start()
return result
def reregister_server(url, address):
""" Update manager url and register agent address with manager """
if _service_is_running() is True:
console_log.warning(
"chroma-agent service was running before registration, stopping.")
agent_service.stop()
conf.set_server_url(url)
crypto = Crypto(conf.ENV_PATH)
agent_client = AgentClient(
url + "reregister/",
ActionPluginManager(),
DevicePluginManager(),
ServerProperties(),
crypto,
)
data = {"address": address, "fqdn": agent_client._fqdn}
try:
result = agent_client.post(data)
except HttpError:
console_log.error("Reregistration failed to %s with request %s" %
(agent_client.url, data))
raise
console_log.info("Starting chroma-agent service")
agent_service.start()
return result
def configure_repo(filename, file_contents):
crypto = Crypto(config.path)
full_filename = os.path.join(REPO_PATH, filename)
temp_full_filename = full_filename + '.tmp'
file_contents = file_contents.format(crypto.AUTHORITY_FILE, crypto.PRIVATE_KEY_FILE, crypto.CERTIFICATE_FILE)
try:
file_handle = os.fdopen(os.open(temp_full_filename, os.O_WRONLY | os.O_CREAT, 0644), 'w')
file_handle.write(file_contents)
file_handle.close()
os.rename(temp_full_filename, full_filename)
except OSError as error:
return agent_error(str(error))
return agent_result_ok
def configure_repo(filename, file_contents):
crypto = Crypto(ENV_PATH)
full_filename = os.path.join(REPO_PATH, filename)
temp_full_filename = full_filename + ".tmp"
# this format needs to match create_repo() in manager agent-bootstrap-script
file_contents = file_contents.format(crypto.AUTHORITY_FILE,
crypto.PRIVATE_KEY_FILE,
crypto.CERTIFICATE_FILE)
try:
file_handle = os.fdopen(
os.open(temp_full_filename, os.O_WRONLY | os.O_CREAT, 0o644), "w")
file_handle.write(file_contents)
file_handle.close()
os.rename(temp_full_filename, full_filename)
except OSError as error:
return agent_error(str(error))
return agent_result_ok
def main():
parser = ArgumentParser(
description="Intel® Manager for Lustre* software Copytool Monitor")
parser.add_argument("copytool_id", action=GetCopytoolAction)
args = parser.parse_args()
copytool_log_setup()
try:
manager_url = config.get('settings',
'server')['url'] + "copytool_event/"
except KeyError:
copytool_log.error(
"No configuration found (must be configured before starting a copytool monitor)"
)
sys.exit(1)
client = CryptoClient(manager_url, Crypto(config.path))
monitor = CopytoolMonitor(client, args.copytool)
def teardown_callback(*args, **kwargs):
monitor.stop()
signal.signal(signal.SIGTERM, teardown_callback)
signal.signal(signal.SIGINT, teardown_callback)
signal.signal(signal.SIGUSR1, decrease_loglevel)
signal.signal(signal.SIGUSR2, increase_loglevel)
try:
monitor.start()
while not monitor.stopping.is_set():
monitor.stopping.wait(timeout=10)
monitor.join()
except Exception as e:
copytool_log.exception()
sys.stderr.write("Unhandled exception: %s\n" % e)
sys.exit(1)
copytool_log.info("Terminating")
return
if config.profile_managed is False:
# This is kind of terrible. The design of DevicePluginManager is
# such that it can be called with either class methods or
# instantiated and then called with instance methods. As such,
# we can't pass in a list of excluded plugins to the instance
# constructor. Well, we could, but it would only work some
# of the time and that would be even more awful.
import chroma_agent.plugin_manager
chroma_agent.plugin_manager.EXCLUDED_PLUGINS += ['corosync']
agent_client = AgentClient(conf['url'] + "message/",
ActionPluginManager(),
DevicePluginManager(), ServerProperties(),
Crypto(config.path))
def teardown_callback(*args, **kwargs):
agent_client.stop()
agent_client.join()
[function() for function in agent_daemon_teardown_functions]
if not args.foreground:
handlers = {
signal.SIGTERM: teardown_callback,
signal.SIGUSR1: decrease_loglevel,
signal.SIGUSR2: increase_loglevel
}
set_signal_handlers(handlers)
else:
signal.signal(signal.SIGINT, teardown_callback)
def __init__(self,
simulator,
fake_cluster,
server_id,
fqdn,
nodename,
nid_tuples_or_network_interfaces,
worker=False,
client_mounts=None):
self.fqdn = fqdn
super(FakeServer, self).__init__(simulator.folder)
self.nodename = nodename
# When reloaded from a file the network interfaces are presented as a dict but when created from code they
# are represented as nid_tuples.
if type(nid_tuples_or_network_interfaces) is dict:
self.network_interfaces = nid_tuples_or_network_interfaces
else:
self.network_interfaces = {}
interface_no = 0
for nid in nid_tuples_or_network_interfaces:
self.network_interfaces[nid[0]] = {
'interface_no': interface_no,
'type': nid[1],
'lnd_type': nid[1],
'lnd_network': nid[2]
}
interface_no += 1
self.boot_time = IMLDateTime.utcnow()
self.id = server_id
self.log_rate = 0
self._log_messages = []
self._simulator = simulator
self._devices = simulator.devices
self._cluster = fake_cluster
self._lock = threading.Lock()
self._agent_client = None
self._cluster.join(nodename, fqdn=fqdn)
if simulator.folder:
# If simulator is persistent, save Crypto in same place
crypto_folder = os.path.join(simulator.folder,
"%s_crypto" % self.fqdn)
else:
# If simulator is ephemeral, create temp dir for Crypto
crypto_folder = tempfile.mkdtemp()
if not os.path.exists(crypto_folder):
os.makedirs(crypto_folder)
self.crypto = Crypto(crypto_folder)
if not client_mounts:
client_mounts = {}
self.state['id'] = server_id
self.state['network_interfaces'] = self.network_interfaces
self.state['nodename'] = nodename
self.state['fqdn'] = fqdn
self.state['worker'] = worker
self.state['client_mounts'] = client_mounts
self.state['packages'][
'chroma-agent'] = self._simulator.available_packages(
self.node_type)[
'chroma-agent'] # The agent is installed at this point.
self.save()
