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 version 2 as
7 # published by the Free Software Foundation;
9 # This program is distributed in the hope that it will be useful,
10 # but WITHOUT ANY WARRANTY; without even the implied warranty of
11 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 # GNU General Public License for more details.
14 # You should have received a copy of the GNU General Public License
15 # along with this program. If not, see <http://www.gnu.org/licenses/>.
17 # Author: Alina Quereilhac <alina.quereilhac@inria.fr>
21 from nepi.util.parallel import ParallelRun
22 from nepi.util.timefuncs import tnow, tdiffsec, stabsformat, tsformat
23 from nepi.execution.resource import ResourceFactory, ResourceAction, \
24 ResourceState, ResourceState2str
25 from nepi.execution.scheduler import HeapScheduler, Task, TaskStatus
26 from nepi.execution.trace import TraceAttr
27 from nepi.util.serializer import ECSerializer, SFormats
28 from nepi.util.plotter import ECPlotter, PFormats
29 from nepi.util.netgraph import NetGraph, TopologyType
31 # TODO: use multiprocessing instead of threading
32 # TODO: Allow to reconnect to a running experiment instance! (reconnect mode vs deploy mode)
43 class FailureLevel(object):
44 """ Possible failure states for the experiment """
49 class FailureManager(object):
50 """ The FailureManager is responsible for handling errors
51 and deciding whether an experiment should be aborted or not
56 self._failure_level = FailureLevel.OK
60 self._ec = weakref.ref(ec)
64 """ Returns the ExperimentController associated to this FailureManager
72 def eval_failure(self, guid):
73 """ Implements failure policy and sets the abort state of the
74 experiment based on the failure state and criticality of
77 :param guid: Guid of the RM upon which the failure of the experiment
82 if self._failure_level == FailureLevel.OK:
83 rm = self.ec.get_resource(guid)
85 critical = rm.get("critical")
87 if state == ResourceState.FAILED and critical:
88 self._failure_level = FailureLevel.RM_FAILURE
90 self.ec.logger.debug("RM critical failure occurred on guid %d." \
91 " Setting EC FAILURE LEVEL to RM_FAILURE" % guid)
93 def set_ec_failure(self):
94 self._failure_level = FailureLevel.EC_FAILURE
96 class ECState(object):
97 """ Possible states of the ExperimentController
105 # historical note: this class used to be in util/guid.py but is used only here
106 # FIXME: This class is not thread-safe. Should it be made thread-safe?
107 class GuidGenerator(object):
111 # historical note: this used to be called `next`
112 # which confused 2to3 - and me - while it has
113 # nothing to do at all with the iteration protocol
114 def generate(self, guid = None):
116 guid = self._last_guid + 1
118 self._last_guid = self._last_guid if guid <= self._last_guid else guid
122 class ExperimentController(object):
126 An experiment, or scenario, is defined by a concrete set of resources,
127 and the behavior, configuration and interconnection of those resources.
128 The Experiment Description (ED) is a detailed representation of a
129 single experiment. It contains all the necessary information to
130 allow repeating the experiment. NEPI allows to describe
131 experiments by registering components (resources), configuring them
132 and interconnecting them.
134 A same experiment (scenario) can be executed many times, generating
135 different results. We call an experiment execution (instance) a 'run'.
137 The ExperimentController (EC), is the entity responsible of
138 managing an experiment run. The same scenario can be
139 recreated (and re-run) by instantiating an EC and recreating
140 the same experiment description.
142 An experiment is represented as a graph of interconnected
143 resources. A resource is a generic concept in the sense that any
144 component taking part of an experiment, whether physical of
145 virtual, is considered a resource. A resources could be a host,
146 a virtual machine, an application, a simulator, a IP address.
148 A ResourceManager (RM), is the entity responsible for managing a
149 single resource. ResourceManagers are specific to a resource
150 type (i.e. An RM to control a Linux application will not be
151 the same as the RM used to control a ns-3 simulation).
152 To support a new type of resource, a new RM must be implemented.
153 NEPI already provides a variety of RMs to control basic resources,
154 and new can be extended from the existing ones.
156 Through the EC interface the user can create ResourceManagers (RMs),
157 configure them and interconnect them, to describe an experiment.
158 Describing an experiment through the EC does not run the experiment.
159 Only when the 'deploy()' method is invoked on the EC, the EC will take
160 actions to transform the 'described' experiment into a 'running' experiment.
162 While the experiment is running, it is possible to continue to
163 create/configure/connect RMs, and to deploy them to involve new
164 resources in the experiment (this is known as 'interactive' deployment).
166 An experiments in NEPI is identified by a string id,
167 which is either given by the user, or automatically generated by NEPI.
168 The purpose of this identifier is to separate files and results that
169 belong to different experiment scenarios.
170 However, since a same 'experiment' can be run many times, the experiment
171 id is not enough to identify an experiment instance (run).
172 For this reason, the ExperimentController has two identifier, the
173 exp_id, which can be re-used in different ExperimentController,
174 and the run_id, which is unique to one ExperimentController instance, and
175 is automatically generated by NEPI.
180 def load(cls, filepath, format = SFormats.XML):
181 serializer = ECSerializer()
182 ec = serializer.load(filepath)
185 def __init__(self, exp_id = None, local_dir = None, persist = False,
186 fm = None, add_node_callback = None, add_edge_callback = None,
188 """ ExperimentController entity to model an execute a network
191 :param exp_id: Human readable name to identify the experiment
194 :param local_dir: Path to local directory where to store experiment
198 :param persist: Save an XML description of the experiment after
199 completion at local_dir
202 :param fm: FailureManager object. If None is given, the default
203 FailureManager class will be used
204 :type fm: FailureManager
206 :param add_node_callback: Callback to invoke for node instantiation
207 when automatic topology creation mode is used
208 :type add_node_callback: function
210 :param add_edge_callback: Callback to invoke for edge instantiation
211 when automatic topology creation mode is used
212 :type add_edge_callback: function
215 super(ExperimentController, self).__init__()
218 self._logger = logging.getLogger("ExperimentController")
220 # Run identifier. It identifies a concrete execution instance (run)
222 # Since a same experiment (same configuration) can be executed many
223 # times, this run_id permits to separate result files generated on
224 # different experiment executions
225 self._run_id = tsformat()
227 # Experiment identifier. Usually assigned by the user
228 # Identifies the experiment scenario (i.e. configuration,
229 # resources used, etc)
230 self._exp_id = exp_id or "exp-%s" % os.urandom(8).encode('hex')
232 # Local path where to store experiment related files (results, etc)
234 local_dir = tempfile.gettempdir() # /tmp
236 self._local_dir = local_dir
237 self._exp_dir = os.path.join(local_dir, self.exp_id)
238 self._run_dir = os.path.join(self.exp_dir, self.run_id)
240 # If True persist the experiment controller in XML format, after completion
241 self._persist = persist
243 # generator of globally unique ids
244 self._guid_generator = GuidGenerator()
247 self._resources = dict()
249 # Scheduler. It a queue that holds tasks scheduled for
250 # execution, and yields the next task to be executed
251 # ordered by execution and arrival time
252 self._scheduler = HeapScheduler()
257 # RM groups (for deployment)
258 self._groups = dict()
260 # generator of globally unique id for groups
261 self._group_id_generator = GuidGenerator()
263 # Flag to stop processing thread
266 # Entity in charge of managing system failures
268 self._fm = FailureManager()
269 self._fm.set_ec(self)
272 self._state = ECState.RUNNING
274 # Automatically construct experiment description
275 self._netgraph = None
276 if add_node_callback or add_edge_callback or kwargs.get("topology"):
277 self._build_from_netgraph(add_node_callback, add_edge_callback,
280 # The runner is a pool of threads used to parallelize
285 # Event processing thread
286 self._cond = threading.Condition()
287 self._thread = threading.Thread(target = self._process)
288 self._thread.setDaemon(True)
293 """ Returns the logger instance of the Experiment Controller
300 """ Returns the failure manager
307 def failure_level(self):
308 """ Returns the level of FAILURE of th experiment
312 return self._fm._failure_level
316 """ Returns the state of the Experiment Controller
323 """ Returns the experiment id assigned by the user
330 """ Returns the experiment instance (run) identifier (automatically
338 """ Returns the number of processing nthreads used
341 return self._nthreads
345 """ Root local directory for experiment files
348 return self._local_dir
352 """ Local directory to store results and other files related to the
360 """ Local directory to store results and other files related to the
368 """ If True, persists the ExperimentController to XML format upon
369 experiment completion
376 """ Return NetGraph instance if experiment description was automatically
380 return self._netgraph
384 """ Returns True if the experiment has failed and should be interrupted,
388 return self._fm.abort
390 def inform_failure(self, guid):
391 """ Reports a failure in a RM to the EC for evaluation
393 :param guid: Resource id
398 return self._fm.eval_failure(guid)
400 def wait_finished(self, guids):
401 """ Blocking method that waits until all RMs in the 'guids' list
402 have reached a state >= STOPPED (i.e. STOPPED, FAILED or
403 RELEASED ), or until a failure in the experiment occurs
406 :param guids: List of guids
414 return self.wait(guids, state = ResourceState.STOPPED,
417 def wait_started(self, guids):
418 """ Blocking method that waits until all RMs in the 'guids' list
419 have reached a state >= STARTED, or until a failure in the
420 experiment occurs (i.e. abort == True)
422 :param guids: List of guids
430 return self.wait(guids, state = ResourceState.STARTED,
433 def wait_released(self, guids):
434 """ Blocking method that waits until all RMs in the 'guids' list
435 have reached a state == RELEASED, or until the EC fails
437 :param guids: List of guids
443 return self._state == ECState.FAILED
445 return self.wait(guids, state = ResourceState.RELEASED,
448 def wait_deployed(self, guids):
449 """ Blocking method that waits until all RMs in the 'guids' list
450 have reached a state >= READY, or until a failure in the
451 experiment occurs (i.e. abort == True)
453 :param guids: List of guids
461 return self.wait(guids, state = ResourceState.READY,
464 def wait(self, guids, state, quit):
465 """ Blocking method that waits until all RMs in the 'guids' list
466 have reached a state >= 'state', or until the 'quit' callback
469 :param guids: List of guids
473 if isinstance(guids, int):
476 # Make a copy to avoid modifying the original guids list
480 # If there are no more guids to wait for
481 # or the quit function returns True, exit the loop
482 if len(guids) == 0 or quit():
485 # If a guid reached one of the target states, remove it from list
487 rm = self.get_resource(guid)
491 self.logger.debug(" %s guid %d DONE - state is %s, required is >= %s " % (
492 rm.get_rtype(), guid, rstate, state))
495 self.logger.debug(" WAITING FOR guid %d - state is %s, required is >= %s " % (
496 guid, rstate, state))
502 def plot(self, dirpath = None, format= PFormats.FIGURE, show = False):
503 plotter = ECPlotter()
504 fpath = plotter.plot(self, dirpath = dirpath, format= format,
508 def serialize(self, format = SFormats.XML):
509 serializer = ECSerializer()
510 sec = serializer.load(self, format = format)
513 def save(self, dirpath = None, format = SFormats.XML):
515 dirpath = self.run_dir
522 serializer = ECSerializer()
523 path = serializer.save(self, dirpath, format = format)
526 def get_task(self, tid):
527 """ Returns a task by its id
529 :param tid: Id of the task
535 return self._tasks.get(tid)
537 def get_resource(self, guid):
538 """ Returns a registered ResourceManager by its guid
540 :param guid: Id of the resource
543 :rtype: ResourceManager
546 rm = self._resources.get(guid)
549 def get_resources_by_type(self, rtype):
550 """ Returns the ResourceManager objects of type rtype
552 :param rtype: Resource type
555 :rtype: list of ResourceManagers
559 for guid, rm in self._resources.items():
560 if rm.get_rtype() == rtype:
564 def remove_resource(self, guid):
565 del self._resources[guid]
569 """ Returns the guids of all ResourceManagers
571 :return: Set of all RM guids
575 keys = list(self._resources.keys())
579 def filter_resources(self, rtype):
580 """ Returns the guids of all ResourceManagers of type rtype
582 :param rtype: Resource type
585 :rtype: list of guids
589 for guid, rm in self._resources.items():
590 if rm.get_rtype() == rtype:
594 def register_resource(self, rtype, guid = None, **keywords):
595 """ Registers a new ResourceManager of type 'rtype' in the experiment
597 This method will assign a new 'guid' for the RM, if no guid
600 :param rtype: Type of the RM
603 :return: Guid of the RM
607 # Get next available guid
609 guid = self._guid_generator.generate(guid)
612 rm = ResourceFactory.create(rtype, self, guid)
615 self._resources[guid] = rm
617 ### so we can do something like
618 # node = ec.register_resource("linux::Node",
622 # node = ec.register_resource("linux::Node")
623 # ec.set(node, "username", user)
624 # ec.set(node, "hostname", host)
626 for name, value in keywords.items():
627 self.set(guid, name, value)
631 def get_attributes(self, guid):
632 """ Returns all the attributes of the RM with guid 'guid'
634 :param guid: Guid of the RM
637 :return: List of attributes
641 rm = self.get_resource(guid)
642 return rm.get_attributes()
644 def get_attribute(self, guid, name):
645 """ Returns the attribute 'name' of the RM with guid 'guid'
647 :param guid: Guid of the RM
650 :param name: Name of the attribute
653 :return: The attribute with name 'name'
657 rm = self.get_resource(guid)
658 return rm.get_attribute(name)
660 def register_connection(self, guid1, guid2):
661 """ Registers a connection between a RM with guid 'guid1'
662 and another RM with guid 'guid2'.
664 The order of the in which the two guids are provided is not
665 important, since the connection relationship is symmetric.
667 :param guid1: First guid to connect
668 :type guid1: ResourceManager
670 :param guid2: Second guid to connect
671 :type guid: ResourceManager
674 rm1 = self.get_resource(guid1)
675 rm2 = self.get_resource(guid2)
677 rm1.register_connection(guid2)
678 rm2.register_connection(guid1)
680 def register_condition(self, guids1, action, guids2, state,
682 """ Registers an action START, STOP or DEPLOY for all RM on list
683 guids1 to occur at time 'time' after all elements in list guids2
684 have reached state 'state'.
686 :param guids1: List of guids of RMs subjected to action
689 :param action: Action to perform (either START, STOP or DEPLOY)
690 :type action: ResourceAction
692 :param guids2: List of guids of RMs to we waited for
695 :param state: State to wait for on RMs of list guids2 (STARTED,
697 :type state: ResourceState
699 :param time: Time to wait after guids2 has reached status
703 if isinstance(guids1, int):
705 if isinstance(guids2, int):
709 rm = self.get_resource(guid1)
710 rm.register_condition(action, guids2, state, time)
712 def enable_trace(self, guid, name):
713 """ Enables a trace to be collected during the experiment run
715 :param name: Name of the trace
719 rm = self.get_resource(guid)
720 rm.enable_trace(name)
722 def trace_enabled(self, guid, name):
723 """ Returns True if the trace of name 'name' is enabled
725 :param name: Name of the trace
729 rm = self.get_resource(guid)
730 return rm.trace_enabled(name)
732 def trace(self, guid, name, attr = TraceAttr.ALL, block = 512, offset = 0):
733 """ Returns information on a collected trace, the trace stream or
734 blocks (chunks) of the trace stream
736 :param name: Name of the trace
739 :param attr: Can be one of:
740 - TraceAttr.ALL (complete trace content),
741 - TraceAttr.STREAM (block in bytes to read starting
743 - TraceAttr.PATH (full path to the trace file),
744 - TraceAttr.SIZE (size of trace file).
747 :param block: Number of bytes to retrieve from trace, when attr is
751 :param offset: Number of 'blocks' to skip, when attr is TraceAttr.STREAM
757 rm = self.get_resource(guid)
758 return rm.trace(name, attr, block, offset)
760 def get_traces(self, guid):
761 """ Returns the list of the trace names of the RM with guid 'guid'
763 :param guid: Guid of the RM
766 :return: List of trace names
770 rm = self.get_resource(guid)
771 return rm.get_traces()
774 def discover(self, guid):
775 """ Discovers an available resource matching the criteria defined
776 by the RM with guid 'guid', and associates that resource to the RM
778 Not all RM types require (or are capable of) performing resource
779 discovery. For the RM types which are not capable of doing so,
780 invoking this method does not have any consequences.
782 :param guid: Guid of the RM
786 rm = self.get_resource(guid)
789 def provision(self, guid):
790 """ Provisions the resource associated to the RM with guid 'guid'.
792 Provisioning means making a resource 'accessible' to the user.
793 Not all RM types require (or are capable of) performing resource
794 provisioning. For the RM types which are not capable of doing so,
795 invoking this method does not have any consequences.
797 :param guid: Guid of the RM
801 rm = self.get_resource(guid)
802 return rm.provision()
804 def get(self, guid, name):
805 """ Returns the value of the attribute with name 'name' on the
808 :param guid: Guid of the RM
811 :param name: Name of the attribute
814 :return: The value of the attribute with name 'name'
817 rm = self.get_resource(guid)
820 def set(self, guid, name, value):
821 """ Modifies the value of the attribute with name 'name' on the
824 :param guid: Guid of the RM
827 :param name: Name of the attribute
830 :param value: Value of the attribute
833 rm = self.get_resource(guid)
836 def get_global(self, rtype, name):
837 """ Returns the value of the global attribute with name 'name' on the
838 RMs of rtype 'rtype'.
840 :param guid: Guid of the RM
843 :param name: Name of the attribute
846 :return: The value of the attribute with name 'name'
849 rclass = ResourceFactory.get_resource_type(rtype)
850 return rclass.get_global(name)
852 def set_global(self, rtype, name, value):
853 """ Modifies the value of the global attribute with name 'name' on the
854 RMs of with rtype 'rtype'.
856 :param guid: Guid of the RM
859 :param name: Name of the attribute
862 :param value: Value of the attribute
865 rclass = ResourceFactory.get_resource_type(rtype)
866 return rclass.set_global(name, value)
868 def state(self, guid, hr = False):
869 """ Returns the state of a resource
871 :param guid: Resource guid
874 :param hr: Human readable. Forces return of a
875 status string instead of a number
879 rm = self.get_resource(guid)
883 return ResourceState2str.get(state)
887 def stop(self, guid):
888 """ Stops the RM with guid 'guid'
890 Stopping a RM means that the resource it controls will
891 no longer take part of the experiment.
893 :param guid: Guid of the RM
897 rm = self.get_resource(guid)
900 def start(self, guid):
901 """ Starts the RM with guid 'guid'
903 Starting a RM means that the resource it controls will
904 begin taking part of the experiment.
906 :param guid: Guid of the RM
910 rm = self.get_resource(guid)
913 def get_start_time(self, guid):
914 """ Returns the start time of the RM as a timestamp """
915 rm = self.get_resource(guid)
918 def get_stop_time(self, guid):
919 """ Returns the stop time of the RM as a timestamp """
920 rm = self.get_resource(guid)
923 def get_discover_time(self, guid):
924 """ Returns the discover time of the RM as a timestamp """
925 rm = self.get_resource(guid)
926 return rm.discover_time
928 def get_provision_time(self, guid):
929 """ Returns the provision time of the RM as a timestamp """
930 rm = self.get_resource(guid)
931 return rm.provision_time
933 def get_ready_time(self, guid):
934 """ Returns the deployment time of the RM as a timestamp """
935 rm = self.get_resource(guid)
938 def get_release_time(self, guid):
939 """ Returns the release time of the RM as a timestamp """
940 rm = self.get_resource(guid)
941 return rm.release_time
943 def get_failed_time(self, guid):
944 """ Returns the time failure occured for the RM as a timestamp """
945 rm = self.get_resource(guid)
946 return rm.failed_time
948 def set_with_conditions(self, name, value, guids1, guids2, state,
950 """ Modifies the value of attribute with name 'name' on all RMs
951 on the guids1 list when time 'time' has elapsed since all
952 elements in guids2 list have reached state 'state'.
954 :param name: Name of attribute to set in RM
957 :param value: Value of attribute to set in RM
960 :param guids1: List of guids of RMs subjected to action
963 :param action: Action to register (either START or STOP)
964 :type action: ResourceAction
966 :param guids2: List of guids of RMs to we waited for
969 :param state: State to wait for on RMs (STARTED, STOPPED, etc)
970 :type state: ResourceState
972 :param time: Time to wait after guids2 has reached status
976 if isinstance(guids1, int):
978 if isinstance(guids2, int):
982 rm = self.get_resource(guid)
983 rm.set_with_conditions(name, value, guids2, state, time)
985 def deploy(self, guids = None, wait_all_ready = True, group = None):
986 """ Deploys all ResourceManagers in the guids list.
988 If the argument 'guids' is not given, all RMs with state NEW
991 :param guids: List of guids of RMs to deploy
994 :param wait_all_ready: Wait until all RMs are ready in
995 order to start the RMs
998 :param group: Id of deployment group in which to deploy RMs
1002 self.logger.debug(" ------- DEPLOY START ------ ")
1005 # If no guids list was passed, all 'NEW' RMs will be deployed
1007 for guid, rm in self._resources.items():
1008 if rm.state == ResourceState.NEW:
1011 if isinstance(guids, int):
1014 # Create deployment group
1015 # New guids can be added to a same deployment group later on
1020 group = self._group_id_generator.generate()
1022 if group not in self._groups:
1023 self._groups[group] = []
1025 self._groups[group].extend(guids)
1027 def wait_all_and_start(group):
1028 # Function that checks if all resources are READY
1029 # before scheduling a start_with_conditions for each RM
1032 # Get all guids in group
1033 guids = self._groups[group]
1036 if self.state(guid) < ResourceState.READY:
1041 callback = functools.partial(wait_all_and_start, group)
1042 self.schedule("1s", callback)
1044 # If all resources are ready, we schedule the start
1046 rm = self.get_resource(guid)
1047 self.schedule("0s", rm.start_with_conditions)
1049 if rm.conditions.get(ResourceAction.STOP):
1050 # Only if the RM has STOP conditions we
1051 # schedule a stop. Otherwise the RM will stop immediately
1052 self.schedule("0s", rm.stop_with_conditions)
1054 if wait_all_ready and new_group:
1055 # Schedule a function to check that all resources are
1056 # READY, and only then schedule the start.
1057 # This aims at reducing the number of tasks looping in the
1059 # Instead of having many start tasks, we will have only one for
1061 callback = functools.partial(wait_all_and_start, group)
1062 self.schedule("0s", callback)
1065 rm = self.get_resource(guid)
1066 rm.deployment_group = group
1067 self.schedule("0s", rm.deploy_with_conditions)
1069 if not wait_all_ready:
1070 self.schedule("0s", rm.start_with_conditions)
1072 if rm.conditions.get(ResourceAction.STOP):
1073 # Only if the RM has STOP conditions we
1074 # schedule a stop. Otherwise the RM will stop immediately
1075 self.schedule("0s", rm.stop_with_conditions)
1077 def release(self, guids = None):
1078 """ Releases all ResourceManagers in the guids list.
1080 If the argument 'guids' is not given, all RMs registered
1081 in the experiment are released.
1083 :param guids: List of RM guids
1087 if self._state == ECState.RELEASED:
1090 if isinstance(guids, int):
1094 guids = self.resources
1097 rm = self.get_resource(guid)
1098 self.schedule("0s", rm.release)
1100 self.wait_released(guids)
1106 if self.get(guid, "hardRelease"):
1107 self.remove_resource(guid)\
1109 # Mark the EC state as RELEASED
1110 self._state = ECState.RELEASED
1113 """ Releases all resources and stops the ExperimentController
1116 # If there was a major failure we can't exit gracefully
1117 if self._state == ECState.FAILED:
1118 raise RuntimeError("EC failure. Can not exit gracefully")
1120 # Remove all pending tasks from the scheduler queue
1121 for tid in list(self._scheduler.pending):
1122 self._scheduler.remove(tid)
1124 # Remove pending tasks from the workers queue
1125 self._runner.empty()
1129 # Mark the EC state as TERMINATED
1130 self._state = ECState.TERMINATED
1132 # Stop processing thread
1135 # Notify condition to wake up the processing thread
1138 if self._thread.is_alive():
1141 def schedule(self, date, callback, track = False):
1142 """ Schedules a callback to be executed at time 'date'.
1144 :param date: string containing execution time for the task.
1145 It can be expressed as an absolute time, using
1146 timestamp format, or as a relative time matching
1147 ^\d+.\d+(h|m|s|ms|us)$
1149 :param callback: code to be executed for the task. Must be a
1150 Python function, and receives args and kwargs
1153 :param track: if set to True, the task will be retrievable with
1154 the get_task() method
1156 :return : The Id of the task
1160 timestamp = stabsformat(date)
1161 task = Task(timestamp, callback)
1162 task = self._scheduler.schedule(task)
1165 self._tasks[task.id] = task
1167 # Notify condition to wake up the processing thread
1173 """ Process scheduled tasks.
1177 Tasks are scheduled by invoking the schedule method with a target
1178 callback and an execution time.
1179 The schedule method creates a new Task object with that callback
1180 and execution time, and pushes it into the '_scheduler' queue.
1181 The execution time and the order of arrival of tasks are used
1182 to order the tasks in the queue.
1184 The _process method is executed in an independent thread held by
1185 the ExperimentController for as long as the experiment is running.
1186 This method takes tasks from the '_scheduler' queue in a loop
1187 and processes them in parallel using multithreading.
1188 The environmental variable NEPI_NTHREADS can be used to control
1189 the number of threads used to process tasks. The default value is
1192 To execute tasks in parallel, a ParallelRunner (PR) object is used.
1193 This object keeps a pool of threads (workers), and a queue of tasks
1194 scheduled for 'immediate' execution.
1196 On each iteration, the '_process' loop will take the next task that
1197 is scheduled for 'future' execution from the '_scheduler' queue,
1198 and if the execution time of that task is >= to the current time,
1199 it will push that task into the PR for 'immediate execution'.
1200 As soon as a worker is free, the PR will assign the next task to
1203 Upon receiving a task to execute, each PR worker (thread) will
1204 invoke the _execute method of the EC, passing the task as
1206 The _execute method will then invoke task.callback inside a
1207 try/except block. If an exception is raised by the tasks.callback,
1208 it will be trapped by the try block, logged to standard error
1209 (usually the console), and the task will be marked as failed.
1213 self._nthreads = int(os.environ.get("NEPI_NTHREADS", str(self._nthreads)))
1214 self._runner = ParallelRun(maxthreads = self.nthreads)
1215 self._runner.start()
1217 while not self._stop:
1219 self._cond.acquire()
1221 task = next(self._scheduler)
1224 # No task to execute. Wait for a new task to be scheduled.
1227 # The task timestamp is in the future. Wait for timeout
1228 # or until another task is scheduled.
1230 if now < task.timestamp:
1231 # Calculate timeout in seconds
1232 timeout = tdiffsec(task.timestamp, now)
1234 # Re-schedule task with the same timestamp
1235 self._scheduler.schedule(task)
1239 # Wait timeout or until a new task awakes the condition
1240 self._cond.wait(timeout)
1242 self._cond.release()
1245 # Process tasks in parallel
1246 self._runner.put(self._execute, task)
1249 err = traceback.format_exc()
1250 self.logger.error("Error while processing tasks in the EC: %s" % err)
1252 # Set the EC to FAILED state
1253 self._state = ECState.FAILED
1255 # Set the FailureManager failure level to EC failure
1256 self._fm.set_ec_failure()
1258 self.logger.debug("Exiting the task processing loop ... ")
1261 self._runner.destroy()
1263 def _execute(self, task):
1264 """ Executes a single task.
1266 :param task: Object containing the callback to execute
1272 task.result = task.callback()
1273 task.status = TaskStatus.DONE
1276 err = traceback.format_exc()
1278 task.status = TaskStatus.ERROR
1280 self.logger.error("Error occurred while executing task: %s" % err)
1283 """ Awakes the processing thread if it is blocked waiting
1284 for new tasks to arrive
1287 self._cond.acquire()
1289 self._cond.release()
1291 def _build_from_netgraph(self, add_node_callback, add_edge_callback,
1293 """ Automates experiment description using a NetGraph instance.
1295 self._netgraph = NetGraph(**kwargs)
1297 if add_node_callback:
1298 ### Add resources to the EC
1299 for nid in self.netgraph.nodes():
1300 add_node_callback(self, nid)
1302 if add_edge_callback:
1303 #### Add connections between resources
1304 for nid1, nid2 in self.netgraph.edges():
1305 add_edge_callback(self, nid1, nid2)