2 # NEPI, a framework to manage network experiments
3 # Copyright (C) 2013 INRIA
5 # This program is free software: you can redistribute it and/or modify
6 # it under the terms of the GNU General Public License as published by
7 # the Free Software Foundation, either version 3 of the License, or
8 # (at your option) any later version.
10 # This program is distributed in the hope that it will be useful,
11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 # GNU General Public License for more details.
15 # You should have received a copy of the GNU General Public License
16 # along with this program. If not, see <http://www.gnu.org/licenses/>.
18 # Author: Alina Quereilhac <alina.quereilhac@inria.fr>
20 from nepi.util.timefuncs import tnow, tdiff, tdiffsec, stabsformat
21 from nepi.util.logger import Logger
22 from nepi.execution.attribute import Attribute, Flags, Types
23 from nepi.execution.trace import TraceAttr
34 reschedule_delay = "1s"
37 """ Action that a user can order to a Resource Manager
45 """ State of a Resource Manager
57 ResourceState2str = dict({
58 ResourceState.NEW : "NEW",
59 ResourceState.DISCOVERED : "DISCOVERED",
60 ResourceState.PROVISIONED : "PROVISIONED",
61 ResourceState.READY : "READY",
62 ResourceState.STARTED : "STARTED",
63 ResourceState.STOPPED : "STOPPED",
64 ResourceState.FAILED : "FAILED",
65 ResourceState.RELEASED : "RELEASED",
69 """ Initializes template information (i.e. attributes and traces)
70 on classes derived from the ResourceManager class.
72 It is used as a decorator in the class declaration as follows:
75 class MyResourceManager(ResourceManager):
84 def clsinit_copy(cls):
85 """ Initializes template information (i.e. attributes and traces)
86 on classes direved from the ResourceManager class.
87 It differs from the clsinit method in that it forces inheritance
88 of attributes and traces from the parent class.
90 It is used as a decorator in the class declaration as follows:
93 class MyResourceManager(ResourceManager):
98 clsinit_copy should be prefered to clsinit when creating new
99 ResourceManager child classes.
107 """ Decorator function for instance methods that should set the
108 RM state to FAILED when an error is raised. The methods that must be
109 decorated are: discover, provision, deploy, start, stop.
112 def wrapped(self, *args, **kwargs):
114 return func(self, *args, **kwargs)
117 err = traceback.format_exc()
119 self.debug("SETTING guid %d to state FAILED" % self.guid)
126 class ResourceManager(Logger):
127 """ Base clase for all ResourceManagers.
129 A ResourceManger is specific to a resource type (e.g. Node,
130 Switch, Application, etc) on a specific backend (e.g. PlanetLab,
133 The ResourceManager instances are responsible for interacting with
134 and controlling concrete (physical or virtual) resources in the
135 experimental backends.
145 def _register_attribute(cls, attr):
146 """ Resource subclasses will invoke this method to add a
151 cls._attributes[attr.name] = attr
154 def _remove_attribute(cls, name):
155 """ Resource subclasses will invoke this method to remove a
160 del cls._attributes[name]
163 def _register_trace(cls, trace):
164 """ Resource subclasses will invoke this method to add a
169 cls._traces[trace.name] = trace
172 def _remove_trace(cls, name):
173 """ Resource subclasses will invoke this method to remove a
178 del cls._traces[name]
181 def _register_attributes(cls):
182 """ Resource subclasses will invoke this method to register
185 This method should be overriden in the RMs that define
190 critical = Attribute("critical",
191 "Defines whether the resource is critical. "
192 "A failure on a critical resource will interrupt "
196 flags = Flags.ExecReadOnly)
198 cls._register_attribute(critical)
201 def _register_traces(cls):
202 """ Resource subclasses will invoke this method to register
205 This method should be overriden in the RMs that define traces.
213 """ ResourceManager classes have different attributes and traces.
214 Attribute and traces are stored in 'class attribute' dictionaries.
215 When a new ResourceManager class is created, the _clsinit method is
216 called to create a new instance of those dictionaries and initialize
219 The _clsinit method is called by the clsinit decorator method.
223 # static template for resource attributes
224 cls._attributes = dict()
225 cls._register_attributes()
227 # static template for resource traces
229 cls._register_traces()
232 def _clsinit_copy(cls):
233 """ Same as _clsinit, except that after creating new instances of the
234 dictionaries it copies all the attributes and traces from the parent
237 The _clsinit_copy method is called by the clsinit_copy decorator method.
240 # static template for resource attributes
241 cls._attributes = copy.deepcopy(cls._attributes)
242 cls._register_attributes()
244 # static template for resource traces
245 cls._traces = copy.deepcopy(cls._traces)
246 cls._register_traces()
250 """ Returns the type of the Resource Manager
256 def get_attributes(cls):
257 """ Returns a copy of the attributes
260 return copy.deepcopy(cls._attributes.values())
263 def get_attribute(cls, name):
264 """ Returns a copy of the attribute with name 'name'
267 return copy.deepcopy(cls._attributes[name])
272 """ Returns a copy of the traces
275 return copy.deepcopy(cls._traces.values())
279 """ Returns the description of the type of Resource
285 def get_backend(cls):
286 """ Returns the identified of the backend (i.e. testbed, environment)
292 def __init__(self, ec, guid):
293 super(ResourceManager, self).__init__(self.get_rtype())
296 self._ec = weakref.ref(ec)
297 self._connections = set()
298 self._conditions = dict()
300 # the resource instance gets a copy of all attributes
301 self._attrs = copy.deepcopy(self._attributes)
303 # the resource instance gets a copy of all traces
304 self._trcs = copy.deepcopy(self._traces)
306 # Each resource is placed on a deployment group by the EC
308 self.deployment_group = None
310 self._start_time = None
311 self._stop_time = None
312 self._discover_time = None
313 self._provision_time = None
314 self._ready_time = None
315 self._release_time = None
316 self._failed_time = None
318 self._state = ResourceState.NEW
320 # instance lock to synchronize exclusive state change methods (such
321 # as deploy and release methods), in order to prevent them from being
322 # executed at the same time
323 self._release_lock = threading.Lock()
327 """ Returns the global unique identifier of the RM """
332 """ Returns the Experiment Controller of the RM """
336 def connections(self):
337 """ Returns the set of guids of connected RMs """
338 return self._connections
341 def conditions(self):
342 """ Returns the conditions to which the RM is subjected to.
344 This method returns a dictionary of conditions lists indexed by
348 return self._conditions
351 def start_time(self):
352 """ Returns the start time of the RM as a timestamp """
353 return self._start_time
357 """ Returns the stop time of the RM as a timestamp """
358 return self._stop_time
361 def discover_time(self):
362 """ Returns the discover time of the RM as a timestamp """
363 return self._discover_time
366 def provision_time(self):
367 """ Returns the provision time of the RM as a timestamp """
368 return self._provision_time
371 def ready_time(self):
372 """ Returns the deployment time of the RM as a timestamp """
373 return self._ready_time
376 def release_time(self):
377 """ Returns the release time of the RM as a timestamp """
378 return self._release_time
381 def failed_time(self):
382 """ Returns the time failure occured for the RM as a timestamp """
383 return self._failed_time
387 """ Get the current state of the RM """
390 def log_message(self, msg):
391 """ Returns the log message formatted with added information.
393 :param msg: text message
398 return " %s guid: %d - %s " % (self._rtype, self.guid, msg)
400 def register_connection(self, guid):
401 """ Registers a connection to the RM identified by guid
403 This method should not be overriden. Specific functionality
404 should be added in the do_connect method.
406 :param guid: Global unique identified of the RM to connect to
410 if self.valid_connection(guid):
411 self.do_connect(guid)
412 self._connections.add(guid)
414 def unregister_connection(self, guid):
415 """ Removes a registered connection to the RM identified by guid
417 This method should not be overriden. Specific functionality
418 should be added in the do_disconnect method.
420 :param guid: Global unique identified of the RM to connect to
424 if guid in self._connections:
425 self.do_disconnect(guid)
426 self._connections.remove(guid)
430 """ Performs resource discovery.
432 This method is responsible for selecting an individual resource
433 matching user requirements.
435 This method should not be overriden directly. Specific functionality
436 should be added in the do_discover method.
439 with self._release_lock:
440 if self._state != ResourceState.RELEASED:
445 """ Performs resource provisioning.
447 This method is responsible for provisioning one resource.
448 After this method has been successfully invoked, the resource
449 should be accessible/controllable by the RM.
451 This method should not be overriden directly. Specific functionality
452 should be added in the do_provision method.
455 with self._release_lock:
456 if self._state != ResourceState.RELEASED:
461 """ Starts the RM (e.g. launch remote process).
463 There is no standard start behavior. Some RMs will not need to perform
464 any actions upon start.
466 This method should not be overriden directly. Specific functionality
467 should be added in the do_start method.
471 if not self.state in [ResourceState.READY, ResourceState.STOPPED]:
472 self.error("Wrong state %s for start" % self.state)
475 with self._release_lock:
476 if self._state != ResourceState.RELEASED:
481 """ Interrupts the RM, stopping any tasks the RM was performing.
483 There is no standard stop behavior. Some RMs will not need to perform
484 any actions upon stop.
486 This method should not be overriden directly. Specific functionality
487 should be added in the do_stop method.
490 if not self.state in [ResourceState.STARTED]:
491 self.error("Wrong state %s for stop" % self.state)
494 with self._release_lock:
499 """ Execute all steps required for the RM to reach the state READY.
501 This method is responsible for deploying the resource (and invoking
502 the discover and provision methods).
504 This method should not be overriden directly. Specific functionality
505 should be added in the do_deploy method.
508 if self.state > ResourceState.READY:
509 self.error("Wrong state %s for deploy" % self.state)
512 with self._release_lock:
513 if self._state != ResourceState.RELEASED:
515 self.debug("----- READY ---- ")
518 """ Perform actions to free resources used by the RM.
520 This method is responsible for releasing resources that were
521 used during the experiment by the RM.
523 This method should not be overriden directly. Specific functionality
524 should be added in the do_release method.
527 with self._release_lock:
532 err = traceback.format_exc()
536 self.debug("----- RELEASED ---- ")
539 """ Sets the RM to state FAILED.
541 This method should not be overriden directly. Specific functionality
542 should be added in the do_fail method.
545 with self._release_lock:
546 if self._state != ResourceState.RELEASED:
549 def set(self, name, value):
550 """ Set the value of the attribute
552 :param name: Name of the attribute
554 :param name: Value of the attribute
557 attr = self._attrs[name]
561 """ Returns the value of the attribute
563 :param name: Name of the attribute
567 attr = self._attrs[name]
570 def enable_trace(self, name):
571 """ Explicitly enable trace generation
573 :param name: Name of the trace
576 trace = self._trcs[name]
579 def trace_enabled(self, name):
580 """Returns True if trace is enables
582 :param name: Name of the trace
585 trace = self._trcs[name]
588 def trace(self, name, attr = TraceAttr.ALL, block = 512, offset = 0):
589 """ Get information on collected trace
591 :param name: Name of the trace
594 :param attr: Can be one of:
595 - TraceAttr.ALL (complete trace content),
596 - TraceAttr.STREAM (block in bytes to read starting at offset),
597 - TraceAttr.PATH (full path to the trace file),
598 - TraceAttr.SIZE (size of trace file).
601 :param block: Number of bytes to retrieve from trace, when attr is TraceAttr.STREAM
604 :param offset: Number of 'blocks' to skip, when attr is TraceAttr.STREAM
611 def register_condition(self, action, group, state, time = None):
612 """ Registers a condition on the resource manager to allow execution
613 of 'action' only after 'time' has elapsed from the moment all resources
614 in 'group' reached state 'state'
616 :param action: Action to restrict to condition (either 'START' or 'STOP')
618 :param group: Group of RMs to wait for (list of guids)
619 :type group: int or list of int
620 :param state: State to wait for on all RM in group. (either 'STARTED', 'STOPPED' or 'READY')
622 :param time: Time to wait after 'state' is reached on all RMs in group. (e.g. '2s')
627 if not action in self.conditions:
628 self._conditions[action] = list()
630 conditions = self.conditions.get(action)
632 # For each condition to register a tuple of (group, state, time) is
633 # added to the 'action' list
634 if not isinstance(group, list):
637 conditions.append((group, state, time))
639 def unregister_condition(self, group, action = None):
640 """ Removed conditions for a certain group of guids
642 :param action: Action to restrict to condition (either 'START', 'STOP' or 'READY')
645 :param group: Group of RMs to wait for (list of guids)
646 :type group: int or list of int
649 # For each condition a tuple of (group, state, time) is
650 # added to the 'action' list
651 if not isinstance(group, list):
654 for act, conditions in self.conditions.iteritems():
655 if action and act != action:
658 for condition in list(conditions):
659 (grp, state, time) = condition
661 # If there is an intersection between grp and group,
662 # then remove intersected elements
663 intsec = set(group).intersection(set(grp))
665 idx = conditions.index(condition)
667 newgrp.difference_update(intsec)
668 conditions[idx] = (newgrp, state, time)
670 def get_connected(self, rtype = None):
671 """ Returns the list of RM with the type 'rtype'
673 :param rtype: Type of the RM we look for
675 :return: list of guid
678 rclass = ResourceFactory.get_resource_type(rtype)
679 for guid in self.connections:
680 rm = self.ec.get_resource(guid)
681 if not rtype or isinstance(rm, rclass):
686 def _needs_reschedule(self, group, state, time):
687 """ Internal method that verify if 'time' has elapsed since
688 all elements in 'group' have reached state 'state'.
690 :param group: Group of RMs to wait for (list of guids)
691 :type group: int or list of int
692 :param state: State to wait for on all RM in group. (either 'STARTED', 'STOPPED' or 'READY')
694 :param time: Time to wait after 'state' is reached on all RMs in group. (e.g. '2s')
697 .. note : time should be written like "2s" or "3m" with s for seconds, m for minutes, h for hours, ...
698 If for example, you need to wait 2min 30sec, time could be "150s" or "2.5m".
699 For the moment, 2m30s is not a correct syntax.
703 delay = reschedule_delay
705 # check state and time elapsed on all RMs
707 rm = self.ec.get_resource(guid)
709 # If one of the RMs this resource needs to wait for has FAILED
710 # and is critical we raise an exception
711 if rm.state == ResourceState.FAILED:
712 if not rm.get('critical'):
714 msg = "Resource can not wait for FAILED RM %d. Setting Resource to FAILED"
715 raise RuntimeError, msg
717 # If the RM state is lower than the requested state we must
718 # reschedule (e.g. if RM is READY but we required STARTED).
723 # If there is a time restriction, we must verify the
724 # restriction is satisfied
726 if state == ResourceState.DISCOVERED:
728 if state == ResourceState.PROVISIONED:
729 t = rm.provision_time
730 elif state == ResourceState.READY:
732 elif state == ResourceState.STARTED:
734 elif state == ResourceState.STOPPED:
736 elif state == ResourceState.RELEASED:
741 # time already elapsed since RM changed state
742 waited = "%fs" % tdiffsec(tnow(), t)
745 wait = tdiffsec(stabsformat(time), stabsformat(waited))
752 return reschedule, delay
754 def set_with_conditions(self, name, value, group, state, time):
755 """ Set value 'value' on attribute with name 'name' when 'time'
756 has elapsed since all elements in 'group' have reached state
759 :param name: Name of the attribute to set
761 :param name: Value of the attribute to set
763 :param group: Group of RMs to wait for (list of guids)
764 :type group: int or list of int
765 :param state: State to wait for on all RM in group. (either 'STARTED', 'STOPPED' or 'READY')
767 :param time: Time to wait after 'state' is reached on all RMs in group. (e.g. '2s')
772 delay = reschedule_delay
774 ## evaluate if set conditions are met
776 # only can set with conditions after the RM is started
777 if self.state != ResourceState.STARTED:
780 reschedule, delay = self._needs_reschedule(group, state, time)
783 callback = functools.partial(self.set_with_conditions,
784 name, value, group, state, time)
785 self.ec.schedule(delay, callback)
787 self.set(name, value)
789 def start_with_conditions(self):
790 """ Starts RM when all the conditions in self.conditions for
791 action 'START' are satisfied.
794 #import pdb;pdb.set_trace()
797 delay = reschedule_delay
800 ## evaluate if conditions to start are met
804 # Can only start when RM is either STOPPED or READY
805 if self.state not in [ResourceState.STOPPED, ResourceState.READY]:
807 self.debug("---- RESCHEDULING START ---- state %s " % self.state )
809 start_conditions = self.conditions.get(ResourceAction.START, [])
811 self.debug("---- START CONDITIONS ---- %s" % start_conditions)
813 # Verify all start conditions are met
814 for (group, state, time) in start_conditions:
815 # Uncomment for debug
818 rm = self.ec.get_resource(guid)
819 unmet.append((guid, rm._state))
821 self.debug("---- WAITED STATES ---- %s" % unmet )
823 reschedule, delay = self._needs_reschedule(group, state, time)
828 self.ec.schedule(delay, self.start_with_conditions)
830 self.debug("----- STARTING ---- ")
833 def stop_with_conditions(self):
834 """ Stops RM when all the conditions in self.conditions for
835 action 'STOP' are satisfied.
839 delay = reschedule_delay
841 ## evaluate if conditions to stop are met
845 # only can stop when RM is STARTED
846 if self.state != ResourceState.STARTED:
848 self.debug("---- RESCHEDULING STOP ---- state %s " % self.state )
850 self.debug(" ---- STOP CONDITIONS ---- %s" %
851 self.conditions.get(ResourceAction.STOP))
853 stop_conditions = self.conditions.get(ResourceAction.STOP, [])
854 for (group, state, time) in stop_conditions:
855 reschedule, delay = self._needs_reschedule(group, state, time)
860 callback = functools.partial(self.stop_with_conditions)
861 self.ec.schedule(delay, callback)
863 self.debug(" ----- STOPPING ---- ")
866 def deploy_with_conditions(self):
867 """ Deploy RM when all the conditions in self.conditions for
868 action 'READY' are satisfied.
872 delay = reschedule_delay
874 ## evaluate if conditions to deploy are met
878 # only can deploy when RM is either NEW, DISCOVERED or PROVISIONED
879 if self.state not in [ResourceState.NEW, ResourceState.DISCOVERED,
880 ResourceState.PROVISIONED]:
882 self.debug("---- RESCHEDULING DEPLOY ---- state %s " % self.state )
884 deploy_conditions = self.conditions.get(ResourceAction.DEPLOY, [])
886 self.debug("---- DEPLOY CONDITIONS ---- %s" % deploy_conditions)
888 # Verify all start conditions are met
889 for (group, state, time) in deploy_conditions:
890 # Uncomment for debug
893 # rm = self.ec.get_resource(guid)
894 # unmet.append((guid, rm._state))
896 #self.debug("---- WAITED STATES ---- %s" % unmet )
898 reschedule, delay = self._needs_reschedule(group, state, time)
903 self.ec.schedule(delay, self.deploy_with_conditions)
905 self.debug("----- DEPLOYING ---- ")
908 def do_connect(self, guid):
909 """ Performs actions that need to be taken upon associating RMs.
910 This method should be redefined when necessary in child classes.
914 def do_disconnect(self, guid):
915 """ Performs actions that need to be taken upon disassociating RMs.
916 This method should be redefined when necessary in child classes.
920 def valid_connection(self, guid):
921 """Checks whether a connection with the other RM
923 This method need to be redefined by each new Resource Manager.
925 :param guid: Guid of the current Resource Manager
933 def do_discover(self):
934 self.set_discovered()
936 def do_provision(self):
937 self.set_provisioned()
948 def do_release(self):
954 def set_started(self):
955 """ Mark ResourceManager as STARTED """
956 self.set_state(ResourceState.STARTED, "_start_time")
958 def set_stopped(self):
959 """ Mark ResourceManager as STOPPED """
960 self.set_state(ResourceState.STOPPED, "_stop_time")
963 """ Mark ResourceManager as READY """
964 self.set_state(ResourceState.READY, "_ready_time")
966 def set_released(self):
967 """ Mark ResourceManager as REALEASED """
968 self.set_state(ResourceState.RELEASED, "_release_time")
970 def set_failed(self):
971 """ Mark ResourceManager as FAILED """
972 self.set_state(ResourceState.FAILED, "_failed_time")
974 def set_discovered(self):
975 """ Mark ResourceManager as DISCOVERED """
976 self.set_state(ResourceState.DISCOVERED, "_discover_time")
978 def set_provisioned(self):
979 """ Mark ResourceManager as PROVISIONED """
980 self.set_state(ResourceState.PROVISIONED, "_provision_time")
982 def set_state(self, state, state_time_attr):
983 """ Set the state of the RM while keeping a trace of the time """
985 # Ensure that RM state will not change after released
986 if self._state == ResourceState.RELEASED:
989 setattr(self, state_time_attr, tnow())
992 class ResourceFactory(object):
993 _resource_types = dict()
996 def resource_types(cls):
997 """Return the type of the Class"""
998 return cls._resource_types
1001 def get_resource_type(cls, rtype):
1002 """Return the type of the Class"""
1003 return cls._resource_types.get(rtype)
1006 def register_type(cls, rclass):
1007 """Register a new Ressource Manager"""
1008 cls._resource_types[rclass.get_rtype()] = rclass
1011 def create(cls, rtype, ec, guid):
1012 """Create a new instance of a Ressource Manager"""
1013 rclass = cls._resource_types[rtype]
1014 return rclass(ec, guid)
1016 def populate_factory():
1017 """Register all the possible RM that exists in the current version of Nepi.
1019 # Once the factory is populated, don't repopulate
1020 if not ResourceFactory.resource_types():
1021 for rclass in find_types():
1022 ResourceFactory.register_type(rclass)
1025 """Look into the different folders to find all the
1026 availables Resources Managers
1028 search_path = os.environ.get("NEPI_SEARCH_PATH", "")
1029 search_path = set(search_path.split(" "))
1032 import nepi.resources
1033 path = os.path.dirname(nepi.resources.__file__)
1034 search_path.add(path)
1038 for importer, modname, ispkg in pkgutil.walk_packages(search_path,
1039 prefix = "nepi.resources."):
1041 loader = importer.find_module(modname)
1044 # Notice: Repeated calls to load_module will act as a reload of teh module
1045 if modname in sys.modules:
1046 module = sys.modules.get(modname)
1048 module = loader.load_module(modname)
1050 for attrname in dir(module):
1051 if attrname.startswith("_"):
1054 attr = getattr(module, attrname)
1056 if attr == ResourceManager:
1059 if not inspect.isclass(attr):
1062 if issubclass(attr, ResourceManager):
1065 if not modname in sys.modules:
1066 sys.modules[modname] = module
1071 err = traceback.format_exc()
1072 logger = logging.getLogger("Resource.find_types()")
1073 logger.error("Error while loading Resource Managers %s" % err)