def ip_address(self):
"""Getter for the service IP address, queries Swarm as the IP address changes outside our control."""
if self.docker_status != self.DOCKER_START_STATUS:
return {}
swarm = SwarmClient(get_conf())
s_info = swarm.inspect_container(self.docker_id)
return s_info['ip_address'][get_conf().overlay_network_name]
def ip_address(self):
"""Getter for the service IP address, queries Swarm as the IP address changes outside our control."""
if self.docker_status != self.DOCKER_START_STATUS:
return {}
swarm = SwarmClient(get_conf())
s_info = swarm.inspect_container(self.docker_id)
return s_info['ip_address'][get_conf().overlay_network_name]
def loop(self):
assert isinstance(config.singletons['sql_manager'],
zoe_lib.sql_manager.SQLManager)
while True:
message = self.zmq_s.recv_json()
self.debug_has_replied = False
start_time = time.time()
if message['command'] == 'execution_start':
exec_id = message['exec_id']
execution = config.singletons['sql_manager'].execution_list(
id=exec_id, only_one=True)
if execution is None:
self._reply_error('Execution ID {} not found'.format(
message['exec_id']))
else:
execution.set_scheduled()
self._reply_ok()
zoe_master.execution_manager.execution_submit(execution)
elif message['command'] == 'execution_terminate':
exec_id = message['exec_id']
execution = config.singletons['sql_manager'].execution_list(
id=exec_id, only_one=True)
if execution is None:
self._reply_error('Execution ID {} not found'.format(
message['exec_id']))
else:
execution.set_cleaning_up()
self._reply_ok()
zoe_master.execution_manager.execution_terminate(execution)
elif message['command'] == 'execution_delete':
exec_id = message['exec_id']
execution = config.singletons['sql_manager'].execution_list(
id=exec_id, only_one=True)
if execution is not None:
zoe_master.execution_manager.execution_delete(execution)
self._reply_ok()
elif message['command'] == 'service_inspect':
service_id = message['service_id']
service = config.singletons['sql_manager'].service_list(
id=service_id, only_one=True)
if service is None:
self._reply_error('no such service')
else:
swarm = SwarmClient(config.get_conf())
info = swarm.inspect_container(service.docker_id)
self._reply_ok(info)
else:
log.error('Unknown command: {}'.format(message['command']))
self._reply_error('unknown command')
if not self.debug_has_replied:
self._reply_error('bug')
raise ZoeException('BUG: command {} does not fill a reply')
config.singletons['metric'].metric_api_call(
start_time, message['command'])
def loop(self):
assert isinstance(config.singletons['sql_manager'], zoe_lib.sql_manager.SQLManager)
while True:
message = self.zmq_s.recv_json()
self.debug_has_replied = False
start_time = time.time()
if message['command'] == 'execution_start':
exec_id = message['exec_id']
execution = config.singletons['sql_manager'].execution_list(id=exec_id, only_one=True)
if execution is None:
self._reply_error('Execution ID {} not found'.format(message['exec_id']))
else:
execution.set_scheduled()
self._reply_ok()
zoe_master.execution_manager.execution_submit(execution)
elif message['command'] == 'execution_terminate':
exec_id = message['exec_id']
execution = config.singletons['sql_manager'].execution_list(id=exec_id, only_one=True)
if execution is None:
self._reply_error('Execution ID {} not found'.format(message['exec_id']))
else:
execution.set_cleaning_up()
self._reply_ok()
zoe_master.execution_manager.execution_terminate(execution)
elif message['command'] == 'execution_delete':
exec_id = message['exec_id']
execution = config.singletons['sql_manager'].execution_list(id=exec_id, only_one=True)
if execution is not None:
zoe_master.execution_manager.execution_delete(execution)
self._reply_ok()
elif message['command'] == 'service_inspect':
service_id = message['service_id']
service = config.singletons['sql_manager'].service_list(id=service_id, only_one=True)
if service is None:
self._reply_error('no such service')
else:
swarm = SwarmClient(config.get_conf())
info = swarm.inspect_container(service.docker_id)
self._reply_ok(info)
else:
log.error('Unknown command: {}'.format(message['command']))
self._reply_error('unknown command')
if not self.debug_has_replied:
self._reply_error('bug')
raise ZoeException('BUG: command {} does not fill a reply')
config.singletons['metric'].metric_api_call(start_time, message['command'])
class PlatformManager:
"""
:type swarm: SwarmClient
:type scheduler: ZoeScheduler
:type state_manager: StateManager
"""
def __init__(self, sched_policy_class):
self.swarm = SwarmClient(get_conf())
self.scheduler = ZoeScheduler(self, sched_policy_class)
self.state_manager = None
def execution_submitted(self, execution: execution_module.Execution):
execution.set_scheduled()
self.scheduler.incoming(execution)
def execution_start(self, execution: execution_module.Execution) -> bool:
try:
self._application_to_containers(execution)
except ZoeException:
self.execution_terminate(execution, reason='error')
raise
execution.set_started()
self.state_manager.state_updated()
def _application_to_containers(self, execution: execution_module.Execution):
for process in execution.application.processes:
self._spawn_process(execution, process)
def _spawn_process(self, execution: execution_module.Execution, process_description: application_module.Process) -> bool:
copts = ContainerOptions()
copts.name = get_conf().container_name_prefix + '-' + process_description.name + "-{}".format(execution.id)
copts.set_memory_limit(process_description.required_resources['memory'])
copts.network_name = '{}-usernet-{}'.format(get_conf().container_name_prefix, execution.owner.id)
container_id = self.state_manager.gen_id()
copts.labels = {
'zoe.{}'.format(get_conf().container_name_prefix): '',
'zoe.execution.id': str(execution.id),
'zoe.execution.name': execution.name,
'zoe.container.id': str(container_id),
'zoe.container.name': process_description.name,
'zoe.owner': execution.owner.name,
'zoe.prefix': get_conf().container_name_prefix,
'zoe.type': 'app_process'
}
if process_description.monitor:
copts.labels['zoe.monitor'] = ''
else:
copts.labels['zoe.normal'] = ''
copts.restart = not process_description.monitor # Monitor containers should not restart
# Generate a dictionary containing the current cluster status (before the new container is spawned)
# This information is used to substitute template strings in the environment variables
subst_dict = {
"execution_id": str(execution.id),
"user_id": str(execution.owner.id),
'user_name': execution.owner.name,
'name_prefix': get_conf().container_name_prefix
}
for env_name, env_value in process_description.environment:
try:
env_value = env_value.format(**subst_dict)
except KeyError:
raise ZoeException("cannot find variable to substitute in expression {}".format(env_value))
copts.add_env_variable(env_name, env_value)
# The same dictionary is used for templates in the command
if process_description.command is not None:
copts.set_command(process_description.command.format(**subst_dict))
cont_info = self.swarm.spawn_container(process_description.docker_image, copts)
container = container_module.Container(self.state_manager)
container.docker_id = cont_info["docker_id"]
container.ip_address = cont_info["ip_address"]
container.name = copts.name
container.is_monitor = process_description.monitor
container.ports = [p.to_dict() for p in process_description.ports]
container.id = container_id
execution.containers.append(container)
container.execution = execution
self.swarm.connect_to_network(container.docker_id, 'eeef9754c16790a29d5210c5d9ad8e66614ee8a6229b6dc6f779019d46cec792')
self.state_manager.new('container', container)
return True
def execution_terminate(self, execution: execution_module.Execution, reason):
"""
:param execution: The execution to be terminated
:param reason: termination reason
:return:
"""
logs = []
if len(execution.containers) > 0:
containers = execution.containers.copy()
for c in containers:
assert isinstance(c, container_module.Container)
l = self.log_get(c.id)
if l is not None:
logs.append((c.name, l))
self.swarm.terminate_container(c.docker_id, delete=True)
self.state_manager.delete('container', c.id)
log.info('Container {} terminated'.format(c.name))
execution.store_logs(logs)
if reason == 'error':
execution.set_error()
elif reason == 'finished':
execution.set_finished()
else:
execution.set_terminated()
self.scheduler.execution_terminate(execution)
def start_gateway_container(self, user):
copts = ContainerOptions()
copts.name = '{}-gateway-{}'.format(get_conf().container_name_prefix, user.id)
copts.network_name = '{}-usernet-{}'.format(get_conf().container_name_prefix, user.id)
copts.ports.append(1080)
copts.labels = {
'zoe.{}.gateway'.format(get_conf().container_name_prefix): '',
'zoe.owner': user.name,
'zoe.prefix': get_conf().container_name_prefix,
'zoe.type': 'gateway'
}
copts.restart = True
if user.role == 'guest':
image = get_conf().private_registry + '/zoerepo/guest-gateway'
else:
image = get_conf().private_registry + '/zoerepo/guest-gateway' # TODO: create an image with ssh
cont_info = self.swarm.spawn_container(image, copts)
if cont_info is None:
raise ZoeException('Cannot create user gateway container')
user.gateway_docker_id = cont_info['docker_id']
user.set_gateway_urls(cont_info)
self.swarm.connect_to_network(user.gateway_docker_id, 'eeef9754c16790a29d5210c5d9ad8e66614ee8a6229b6dc6f779019d46cec792')
def kill_gateway_container(self, user):
self.swarm.terminate_container(user.gateway_docker_id, delete=True)
user.gateway_docker_id = None
user.gateway_urls = []
def create_user_network(self, user):
log.info('Creating a new network for user {}'.format(user.id))
net_name = '{}-usernet-{}'.format(get_conf().container_name_prefix, user.id)
net_id = self.swarm.network_create(net_name)
user.network_id = net_id
def remove_user_network(self, user):
log.info('Removing network for user {}'.format(user.name))
self.swarm.network_remove(user.network_id)
def log_get(self, container_id: int) -> str:
container = self.state_manager.get_one('container', id=container_id)
if container is None:
return ''
else:
return self.swarm.log_get(container.docker_id)
def container_stats(self, container_id: int) -> ContainerStats:
container = self.state_manager.get_one('container', id=container_id)
return self.swarm.stats(container.docker_id)
def is_container_alive(self, container: container_module.Container) -> bool:
ret = self.swarm.inspect_container(container.docker_id)
if ret is None:
return False
return ret["running"]
def swarm_stats(self) -> SwarmStats:
# TODO implement some caching
return self.swarm.info()
def scheduler_stats(self) -> SchedulerStats:
return self.scheduler.scheduler_policy.stats()
def check_state_swarm_consistency(self):
state_changed = False
users = self.state_manager.get('user')
networks = self.swarm.network_list('{}-usernet-'.format(get_conf().container_name_prefix))
gateways = self.swarm.list(['zoe.{}.gateway'.format(get_conf().container_name_prefix)])
users_no_network = []
users_no_gateway = []
networks_to_delete = []
gateways_to_delete = []
for u in users:
if u.network_id is None:
log.error('state inconsistency: user {} has no network'.format(u.name))
users_no_network.append(u)
elif u.network_id not in [x['id'] for x in networks]:
log.error('state inconsistency: user {} has an invalid network'.format(u.name))
u.network_id = None
users_no_network.append(u)
if u.gateway_docker_id is None:
log.error('state inconsistency: user {} has no gateway'.format(u.name))
users_no_gateway.append(u)
elif u.gateway_docker_id not in [x['id'] for x in gateways]:
log.error('state inconsistency: user {} has an invalid gateway container ID'.format(u.name))
u.gateway_docker_id = None
users_no_gateway.append(u)
duplicate_check = set()
for n in networks:
try:
uid = int(n['name'][len('{}-usernet-'.format(get_conf().container_name_prefix)):])
except ValueError:
log.error('network {} does not belong to Zoe, bug?'.format(n['name']))
networks_to_delete.append(n['id'])
continue
if uid in duplicate_check:
log.warning('state inconsistency: found two networks for the same user')
networks_to_delete.append(n['id'])
continue
duplicate_check.add(uid)
user = self.state_manager.get_one('user', id=uid)
if user is not None and user in users_no_network:
user.network_id = n['id']
users_no_network.remove(user)
log.error('fixed: user {} linked to network {}'.format(user.name, n['name']))
state_changed = True
continue
elif user is None:
log.error('state inconsistency: found a network for user {} who no longer exists'.format(uid))
networks_to_delete.append(n['id'])
for g in gateways:
try:
uid = int(g['name'][len('{}-gateway-'.format(get_conf().container_name_prefix)):])
except ValueError:
log.error('container {} does not belong to Zoe, bug?'.format(g['name']))
gateways_to_delete.append(g['id'])
continue
user = self.state_manager.get_one('user', id=uid)
if user is not None and user in users_no_gateway:
user.gateway_docker_id = g['id']
users_no_gateway.remove(user)
cont_info = self.swarm.inspect_container(g['id'])
user.set_gateway_urls(cont_info)
log.error('fixed: user {} linked to gateway {}'.format(user.name, g['name']))
state_changed = True
continue
elif user is None:
log.error('state inconsistency: found a gateway for user {} who no longer exists'.format(uid))
gateways_to_delete.append(g['id'])
# Fix all inconsistencies found
for g in gateways_to_delete:
log.error('fixed: terminating orphan gateway container {}'.format(g[:8]))
self.swarm.terminate_container(g, delete=True)
for n in networks_to_delete:
log.error('fixed: terminating orphan network {}'.format(n[:8]))
self.swarm.network_remove(n)
for u in users_no_network:
log.error('fixed: creating network for user {}'.format(u.name))
self.create_user_network(u)
for u in users_no_gateway:
log.error('fixed: creating gateway for user {}'.format(u.name))
self.start_gateway_container(u)
# ### Check executions and container consistency
swarm_containers = self.swarm.list(only_label='zoe.{}'.format(get_conf().container_name_prefix))
conts_state_to_delete = []
for c_id, c in self.state_manager.containers.items():
if c.docker_id not in [x['id'] for x in swarm_containers]:
log.error('fixed: removing from state container {} that does not exist in Swarm'.format(c.name))
conts_state_to_delete.append(c_id)
for c_id in conts_state_to_delete:
self.state_manager.delete('container', c_id)
if state_changed or len(users_no_gateway) > 0 or len(users_no_network) > 0:
self.state_manager.state_updated()
