def __init__(self, config):
# Logging.
self.logger = logging.getLogger(__name__)
# The overall configuration.
self.config = config
# The proxy being used to talk to HAProxy.
self.proxy = Proxy()
# The components, by name.
self.components = {}
# Build the components map.
for component_config in config.components:
self.components[component_config.name] = Component(self, component_config)
# Create the lock for the watcher thread and the notification event.
self.watcher_lock = threading.Lock()
self.watcher_event = threading.Event()
# The set of containers which should be terminated by the terminating workers.
self.containers_to_terminate = Queue()
# Start the thread used to watch and stop containers that are no longer needed.
self.pool = ThreadPool()
# Place to collect the results of the monitor
self.monitor_futures = Queue()
def __init__(self, config, daemon_mode = False):
# Logging.
self.logger = logging.getLogger(__name__)
# Whether gantry is running in daemon mode.
self.daemon_mode = daemon_mode
# The overall configuration.
self.config = config
# The proxy being used to talk to HAProxy.
self.proxy = Proxy()
# The components, by name.
self.components = {}
# Build the components map.
for component_config in config.components:
self.components[component_config.name] = Component(self, component_config)
# Create the lock for the watcher thread and the notification event.
self.watcher_lock = threading.Lock()
self.watcher_event = threading.Event()
# The set of containers watched. Must be accessed under the watcher_lock.
# When the connections to the ports used by a container are not longer there,
# then the container is stopped.
self.containers_watched = []
# Start the thread used to watch and stop containers that are no longer needed.
self.watcher_thread = threading.Thread(target = self.checkProxy, args = [])
self.watcher_thread.daemon = True
self.watcher_thread.start()
def run(self, container, report):
container_ip = self.getContainerIPAddress(container)
for connection in Proxy.get_connections():
if not connection.laddr or not connection.raddr:
continue
if connection.raddr[0] == container_ip:
report('Container still has existing connections: %s' % container['Id'][0:12],
level=ReportLevels.EXTRA)
return False
report('Container has no remaining connections: %s' % container['Id'][0:12],
level=ReportLevels.EXTRA)
return True
class RuntimeManager(object):
""" Manager class which handles tracking of all the components and other runtime
information.
"""
def __init__(self, config, daemon_mode = False):
# Logging.
self.logger = logging.getLogger(__name__)
# Whether gantry is running in daemon mode.
self.daemon_mode = daemon_mode
# The overall configuration.
self.config = config
# The proxy being used to talk to HAProxy.
self.proxy = Proxy()
# The components, by name.
self.components = {}
# Build the components map.
for component_config in config.components:
self.components[component_config.name] = Component(self, component_config)
# Create the lock for the watcher thread and the notification event.
self.watcher_lock = threading.Lock()
self.watcher_event = threading.Event()
# The set of containers watched. Must be accessed under the watcher_lock.
# When the connections to the ports used by a container are not longer there,
# then the container is stopped.
self.containers_watched = []
# Start the thread used to watch and stop containers that are no longer needed.
self.watcher_thread = threading.Thread(target = self.checkProxy, args = [])
self.watcher_thread.daemon = True
self.watcher_thread.start()
def getComponent(self, name):
""" Returns the component with the given name defined or None if none. """
if not name in self.components:
return None
return self.components[name]
def adjustForUpdatingComponent(self, component, started_container):
""" Adjusts the runtime for a component which has been started in the given
container.
"""
self.logger.debug('Adjusting runtime for updating component: %s', component.getName())
self.updateProxy()
def adjustForStoppingComponent(self, component):
""" Adjusts the runtime for a component which has been stopped.
"""
self.logger.debug('Adjusting runtime for stopped component: %s', component.getName())
self.updateProxy()
def findConnectionLessContainers(self, containers):
""" Determines which containers no longer have any valid connections to the
outside world.
"""
# Build the set of active connections from all the running proxy processes.
connections = self.proxy.get_connections()
# Build the set of container-local ports used.
ports = set([])
for connection in connections:
laddr = connection.laddr
raddr = connection.raddr
if not laddr or not raddr:
continue
laddress = laddr[0]
raddress = raddr[0]
lport = laddr[1]
rport = raddr[1]
if laddress == '127.0.0.1' and raddress == '127.0.0.1':
ports.add(rport)
# For each draining container, if the port set contains one of the known mappings, then
# the container is still being used.
connectionless = list(containers)
for container in containers:
if getContainerStatus(container) == 'draining':
container_local_ports = containerutil.getLocalPorts(container)
if ports.intersection(container_local_ports):
connectionless.remove(container)
return connectionless
def checkProxy(self):
""" Checks to see if a draining container can be shutdown. """
counter = 0
client = getDockerClient()
while True:
# Wait until something of interest is avaliable to check.
self.watcher_event.wait()
self.watcher_event.clear()
while True:
# Load the containers to check (under the lock).
containers = None
with self.watcher_lock:
containers = list(self.containers_watched)
# If none, we're done.
if not containers:
break
# Find the containers that no longer need to be running. Any container with no
# valid connections coming in and a status of 'draining', can be shutdown.
report('Monitor check started', level = ReportLevels.BACKGROUND)
containers_to_shutdown = self.findConnectionLessContainers(containers)
if len(containers_to_shutdown) > 0:
with self.watcher_lock:
for container in containers_to_shutdown:
self.containers_watched.remove(container)
for container in containers_to_shutdown:
setContainerStatus(container, 'shutting-down')
report('Shutting down container: ' + container['Id'][0:12], level = ReportLevels.BACKGROUND)
client.stop(container)
removeContainerMetadata(container)
# Determine how many residual containers are left over.
difference = len(containers) - len(containers_to_shutdown)
if difference > 0:
report(str(difference) + ' additional containers to monitor. Sleeping for 10 seconds', level = ReportLevels.BACKGROUND)
time.sleep(10)
counter = counter + 1
report('Monitor check finished', level = ReportLevels.BACKGROUND)
if not self.daemon_mode:
# Quit now that we're done.
return
def updateProxy(self):
""" Updates the proxy used for port mapping to conform to the current running container
list.
"""
client = getDockerClient()
# Clear all routes in the proxy.
# TODO: When this is in daemon mode, don't need do this. We could selectively
# edit it instead.
self.proxy.clear_routes()
# Add routes for the non-draining containers and collect the draining containers to
# watch.
report('Finding running containers...', level = ReportLevels.EXTRA)
draining_containers = []
starting_containers = []
for component in self.components.values():
for container in component.getAllContainers(client):
if getContainerStatus(container) != 'draining':
starting_containers.append(container)
for mapping in component.config.ports:
local_port = containerutil.getLocalPort(client, container, mapping.container)
route = Route(mapping.kind == 'http', mapping.external, 'localhost', local_port)
self.proxy.add_route(route)
else:
draining_containers.append(container)
# Commit the changes to the proxy.
if draining_containers or starting_containers:
report('Updating proxy...', level = ReportLevels.EXTRA)
self.proxy.commit()
else:
report('Shutting down proxy...', level = ReportLevels.EXTRA)
self.proxy.shutdown()
# Mark the starting containers as running.
for container in starting_containers:
setContainerStatus(container, 'running')
if draining_containers:
report('Starting monitoring...', level = ReportLevels.EXTRA)
# If there are any draining containers, add them to the watcher thread.
with self.watcher_lock:
self.containers_watched.extend(draining_containers)
# Call the event to wakeup the watcher thread.
if draining_containers:
self.watcher_event.set()
# If in local mode, then wait until all the containers have drained. This
# prevents the python debugger from shutting down, since the other threads
# are all daemon threads.
if not self.daemon_mode and draining_containers:
while True:
self.watcher_thread.join(10)
if not self.watcher_thread.isAlive():
break
class RuntimeManager(object):
""" Manager class which handles tracking of all the components and other runtime
information.
"""
def __init__(self, config):
# Logging.
self.logger = logging.getLogger(__name__)
# The overall configuration.
self.config = config
# The proxy being used to talk to HAProxy.
self.proxy = Proxy()
# The components, by name.
self.components = {}
# Build the components map.
for component_config in config.components:
self.components[component_config.name] = Component(self, component_config)
# Create the lock for the watcher thread and the notification event.
self.watcher_lock = threading.Lock()
self.watcher_event = threading.Event()
# The set of containers which should be terminated by the terminating workers.
self.containers_to_terminate = Queue()
# Start the thread used to watch and stop containers that are no longer needed.
self.pool = ThreadPool()
# Place to collect the results of the monitor
self.monitor_futures = Queue()
def getComponent(self, name):
""" Returns the component with the given name defined or None if none. """
if not name in self.components:
return None
return self.components[name]
def lookupComponentLink(self, link_name):
""" Looks up the component link with the given name defined or None if none. """
for component_name, component in self.components.items():
defined_links = component.config.getDefinedComponentLinks()
if link_name in defined_links:
return ComponentLinkInformation(self, component, defined_links[link_name])
return None
def adjustForUpdatingComponent(self, component, started_container):
""" Adjusts the runtime for a component which has been started in the given
container.
"""
self.logger.debug('Adjusting runtime for updating component: %s', component.getName())
self.updateProxy()
def adjustForStoppingComponent(self, component):
""" Adjusts the runtime for a component which has been stopped.
"""
self.logger.debug('Adjusting runtime for stopped component: %s', component.getName())
self.updateProxy()
def watchTermination(self, container, component):
report('Monitor check started', level=ReportLevels.BACKGROUND)
client = getDockerClient()
# Send the termination signal(s) to the container
signals = []
for signal in component.config.termination_signals:
signals.append((signal, buildTerminationSignal(signal)))
report('Sending %s termination signals' % len(signals), component=component)
for (config, signal) in signals:
report('Sending termination signal: ' + config.getTitle(), component=component)
result = signal.run(container, report)
if not result:
report('Termination signal failed', component=component)
# Now wait until all of the termination conditions are met
checks = []
for check in component.config.termination_checks:
checks.append((check, buildHealthCheck(check)))
report('Waiting for %s termination checks' % len(checks), component=component)
for (config, check) in checks:
check_passed = False
while not check_passed:
report('Running termination check: ' + config.getTitle(), component=component)
result = check.run(container, report)
if not result:
report('Termination check failed', component=component)
report('Sleeping ' + str(config.timeout) + ' second(s)...', component=component)
time.sleep(config.timeout)
else:
check_passed = True
report('Monitor check finished', level=ReportLevels.BACKGROUND)
setContainerStatus(container, 'shutting-down')
report('Shutting down container: ' + container['Id'][0:12], level=ReportLevels.BACKGROUND)
client.stop(container)
removeContainerMetadata(container)
def terminateContainer(self, container, component):
""" Adds the given container to the list of containers which should be terminated.
"""
report('Terminating container: %s' % container['Id'][:12], component=component)
self.monitor_futures.put(self.pool.apply_async(self.watchTermination, (container, component)))
def updateProxy(self):
""" Updates the proxy used for port mapping to conform to the current running container
list.
"""
client = getDockerClient()
# Clear all routes in the proxy.
# TODO: When this is in daemon mode, don't need do this. We could selectively
# edit it instead.
self.proxy.clear_routes()
# Add routes for the non-draining containers and collect the draining containers to
# watch.
report('Finding running containers...', level=ReportLevels.EXTRA)
draining_containers = []
starting_containers = []
for component in self.components.values():
for container in component.getAllContainers(client):
if getContainerStatus(container) != 'draining':
container_ip = containerutil.getContainerIPAddress(client, container)
starting_containers.append(container)
# Add the normal exposed ports.
for mapping in component.config.ports:
route = Route(mapping.kind == 'http', mapping.external, container_ip,
mapping.container)
self.proxy.add_route(route)
# Add the container link ports.
for link in component.config.defined_component_links:
route = Route(link.kind == 'http', link.getHostPort(), container_ip, link.port)
self.proxy.add_route(route)
else:
draining_containers.append(container)
# Commit the changes to the proxy.
if draining_containers or starting_containers:
report('Updating proxy...', level=ReportLevels.EXTRA)
self.proxy.commit()
else:
report('Shutting down proxy...', level=ReportLevels.EXTRA)
self.proxy.shutdown()
# Mark the starting containers as running.
for container in starting_containers:
setContainerStatus(container, 'running')
def join(self):
self.pool.close()
while not self.monitor_futures.empty():
# If any of the futures threw and exception we'll get it now
self.monitor_futures.get().get()
self.pool.join()
