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",
623 # node = ec.register_resource("linux::Node")
624 # ec.set(node, "username", user)
625 # ec.set(node, "hostname", host)
628 auto_deploy = 'autoDeploy' in keywords and keywords['autoDeploy']
630 # now we can do all the calls to 'set'
631 for name, value in keywords.items():
632 # autoDeploy is handled locally and not propagated to 'set'
633 if name != 'autoDeploy':
634 self.set(guid, name, value)
641 def get_attributes(self, guid):
642 """ Returns all the attributes of the RM with guid 'guid'
644 :param guid: Guid of the RM
647 :return: List of attributes
651 rm = self.get_resource(guid)
652 return rm.get_attributes()
654 def get_attribute(self, guid, name):
655 """ Returns the attribute 'name' of the RM with guid 'guid'
657 :param guid: Guid of the RM
660 :param name: Name of the attribute
663 :return: The attribute with name 'name'
667 rm = self.get_resource(guid)
668 return rm.get_attribute(name)
670 def register_connection(self, guid1, guid2):
671 """ Registers a connection between a RM with guid 'guid1'
672 and another RM with guid 'guid2'.
674 The order of the in which the two guids are provided is not
675 important, since the connection relationship is symmetric.
677 :param guid1: First guid to connect
678 :type guid1: ResourceManager
680 :param guid2: Second guid to connect
681 :type guid: ResourceManager
684 rm1 = self.get_resource(guid1)
685 rm2 = self.get_resource(guid2)
687 rm1.register_connection(guid2)
688 rm2.register_connection(guid1)
690 def register_condition(self, guids1, action, guids2, state,
692 """ Registers an action START, STOP or DEPLOY for all RM on list
693 guids1 to occur at time 'time' after all elements in list guids2
694 have reached state 'state'.
696 :param guids1: List of guids of RMs subjected to action
699 :param action: Action to perform (either START, STOP or DEPLOY)
700 :type action: ResourceAction
702 :param guids2: List of guids of RMs to we waited for
705 :param state: State to wait for on RMs of list guids2 (STARTED,
707 :type state: ResourceState
709 :param time: Time to wait after guids2 has reached status
713 if isinstance(guids1, int):
715 if isinstance(guids2, int):
719 rm = self.get_resource(guid1)
720 rm.register_condition(action, guids2, state, time)
722 def enable_trace(self, guid, name):
723 """ Enables a trace to be collected during the experiment run
725 :param name: Name of the trace
729 rm = self.get_resource(guid)
730 rm.enable_trace(name)
732 def trace_enabled(self, guid, name):
733 """ Returns True if the trace of name 'name' is enabled
735 :param name: Name of the trace
739 rm = self.get_resource(guid)
740 return rm.trace_enabled(name)
742 def trace(self, guid, name, attr = TraceAttr.ALL, block = 512, offset = 0):
743 """ Returns information on a collected trace, the trace stream or
744 blocks (chunks) of the trace stream
746 :param name: Name of the trace
749 :param attr: Can be one of:
750 - TraceAttr.ALL (complete trace content),
751 - TraceAttr.STREAM (block in bytes to read starting
753 - TraceAttr.PATH (full path to the trace file),
754 - TraceAttr.SIZE (size of trace file).
757 :param block: Number of bytes to retrieve from trace, when attr is
761 :param offset: Number of 'blocks' to skip, when attr is TraceAttr.STREAM
767 rm = self.get_resource(guid)
768 return rm.trace(name, attr, block, offset)
770 def get_traces(self, guid):
771 """ Returns the list of the trace names of the RM with guid 'guid'
773 :param guid: Guid of the RM
776 :return: List of trace names
780 rm = self.get_resource(guid)
781 return rm.get_traces()
784 def discover(self, guid):
785 """ Discovers an available resource matching the criteria defined
786 by the RM with guid 'guid', and associates that resource to the RM
788 Not all RM types require (or are capable of) performing resource
789 discovery. For the RM types which are not capable of doing so,
790 invoking this method does not have any consequences.
792 :param guid: Guid of the RM
796 rm = self.get_resource(guid)
799 def provision(self, guid):
800 """ Provisions the resource associated to the RM with guid 'guid'.
802 Provisioning means making a resource 'accessible' to the user.
803 Not all RM types require (or are capable of) performing resource
804 provisioning. For the RM types which are not capable of doing so,
805 invoking this method does not have any consequences.
807 :param guid: Guid of the RM
811 rm = self.get_resource(guid)
812 return rm.provision()
814 def get(self, guid, name):
815 """ Returns the value of the attribute with name 'name' on the
818 :param guid: Guid of the RM
821 :param name: Name of the attribute
824 :return: The value of the attribute with name 'name'
827 rm = self.get_resource(guid)
830 def set(self, guid, name, value):
831 """ Modifies the value of the attribute with name 'name' on the
834 :param guid: Guid of the RM
837 :param name: Name of the attribute
840 :param value: Value of the attribute
843 rm = self.get_resource(guid)
846 def get_global(self, rtype, name):
847 """ Returns the value of the global attribute with name 'name' on the
848 RMs of rtype 'rtype'.
850 :param guid: Guid of the RM
853 :param name: Name of the attribute
856 :return: The value of the attribute with name 'name'
859 rclass = ResourceFactory.get_resource_type(rtype)
860 return rclass.get_global(name)
862 def set_global(self, rtype, name, value):
863 """ Modifies the value of the global attribute with name 'name' on the
864 RMs of with rtype 'rtype'.
866 :param guid: Guid of the RM
869 :param name: Name of the attribute
872 :param value: Value of the attribute
875 rclass = ResourceFactory.get_resource_type(rtype)
876 return rclass.set_global(name, value)
878 def state(self, guid, hr = False):
879 """ Returns the state of a resource
881 :param guid: Resource guid
884 :param hr: Human readable. Forces return of a
885 status string instead of a number
889 rm = self.get_resource(guid)
893 return ResourceState2str.get(state)
897 def stop(self, guid):
898 """ Stops the RM with guid 'guid'
900 Stopping a RM means that the resource it controls will
901 no longer take part of the experiment.
903 :param guid: Guid of the RM
907 rm = self.get_resource(guid)
910 def start(self, guid):
911 """ Starts the RM with guid 'guid'
913 Starting a RM means that the resource it controls will
914 begin taking part of the experiment.
916 :param guid: Guid of the RM
920 rm = self.get_resource(guid)
923 def get_start_time(self, guid):
924 """ Returns the start time of the RM as a timestamp """
925 rm = self.get_resource(guid)
928 def get_stop_time(self, guid):
929 """ Returns the stop time of the RM as a timestamp """
930 rm = self.get_resource(guid)
933 def get_discover_time(self, guid):
934 """ Returns the discover time of the RM as a timestamp """
935 rm = self.get_resource(guid)
936 return rm.discover_time
938 def get_provision_time(self, guid):
939 """ Returns the provision time of the RM as a timestamp """
940 rm = self.get_resource(guid)
941 return rm.provision_time
943 def get_ready_time(self, guid):
944 """ Returns the deployment time of the RM as a timestamp """
945 rm = self.get_resource(guid)
948 def get_release_time(self, guid):
949 """ Returns the release time of the RM as a timestamp """
950 rm = self.get_resource(guid)
951 return rm.release_time
953 def get_failed_time(self, guid):
954 """ Returns the time failure occured for the RM as a timestamp """
955 rm = self.get_resource(guid)
956 return rm.failed_time
958 def set_with_conditions(self, name, value, guids1, guids2, state,
960 """ Modifies the value of attribute with name 'name' on all RMs
961 on the guids1 list when time 'time' has elapsed since all
962 elements in guids2 list have reached state 'state'.
964 :param name: Name of attribute to set in RM
967 :param value: Value of attribute to set in RM
970 :param guids1: List of guids of RMs subjected to action
973 :param action: Action to register (either START or STOP)
974 :type action: ResourceAction
976 :param guids2: List of guids of RMs to we waited for
979 :param state: State to wait for on RMs (STARTED, STOPPED, etc)
980 :type state: ResourceState
982 :param time: Time to wait after guids2 has reached status
986 if isinstance(guids1, int):
988 if isinstance(guids2, int):
992 rm = self.get_resource(guid)
993 rm.set_with_conditions(name, value, guids2, state, time)
995 def deploy(self, guids = None, wait_all_ready = True, group = None):
996 """ Deploys all ResourceManagers in the guids list.
998 If the argument 'guids' is not given, all RMs with state NEW
1001 :param guids: List of guids of RMs to deploy
1004 :param wait_all_ready: Wait until all RMs are ready in
1005 order to start the RMs
1008 :param group: Id of deployment group in which to deploy RMs
1012 self.logger.debug(" ------- DEPLOY START ------ ")
1015 # If no guids list was passed, all 'NEW' RMs will be deployed
1017 for guid, rm in self._resources.items():
1018 if rm.state == ResourceState.NEW:
1021 if isinstance(guids, int):
1024 # Create deployment group
1025 # New guids can be added to a same deployment group later on
1030 group = self._group_id_generator.generate()
1032 if group not in self._groups:
1033 self._groups[group] = []
1035 self._groups[group].extend(guids)
1037 def wait_all_and_start(group):
1038 # Function that checks if all resources are READY
1039 # before scheduling a start_with_conditions for each RM
1042 # Get all guids in group
1043 guids = self._groups[group]
1046 if self.state(guid) < ResourceState.READY:
1051 callback = functools.partial(wait_all_and_start, group)
1052 self.schedule("1s", callback)
1054 # If all resources are ready, we schedule the start
1056 rm = self.get_resource(guid)
1057 self.schedule("0s", rm.start_with_conditions)
1059 if rm.conditions.get(ResourceAction.STOP):
1060 # Only if the RM has STOP conditions we
1061 # schedule a stop. Otherwise the RM will stop immediately
1062 self.schedule("0s", rm.stop_with_conditions)
1064 if wait_all_ready and new_group:
1065 # Schedule a function to check that all resources are
1066 # READY, and only then schedule the start.
1067 # This aims at reducing the number of tasks looping in the
1069 # Instead of having many start tasks, we will have only one for
1071 callback = functools.partial(wait_all_and_start, group)
1072 self.schedule("0s", callback)
1075 rm = self.get_resource(guid)
1076 rm.deployment_group = group
1077 self.schedule("0s", rm.deploy_with_conditions)
1079 if not wait_all_ready:
1080 self.schedule("0s", rm.start_with_conditions)
1082 if rm.conditions.get(ResourceAction.STOP):
1083 # Only if the RM has STOP conditions we
1084 # schedule a stop. Otherwise the RM will stop immediately
1085 self.schedule("0s", rm.stop_with_conditions)
1087 def release(self, guids = None):
1088 """ Releases all ResourceManagers in the guids list.
1090 If the argument 'guids' is not given, all RMs registered
1091 in the experiment are released.
1093 :param guids: List of RM guids
1097 if self._state == ECState.RELEASED:
1100 if isinstance(guids, int):
1104 guids = self.resources
1107 rm = self.get_resource(guid)
1108 self.schedule("0s", rm.release)
1110 self.wait_released(guids)
1116 if self.get(guid, "hardRelease"):
1117 self.remove_resource(guid)\
1119 # Mark the EC state as RELEASED
1120 self._state = ECState.RELEASED
1123 """ Releases all resources and stops the ExperimentController
1126 # If there was a major failure we can't exit gracefully
1127 if self._state == ECState.FAILED:
1128 raise RuntimeError("EC failure. Can not exit gracefully")
1130 # Remove all pending tasks from the scheduler queue
1131 for tid in list(self._scheduler.pending):
1132 self._scheduler.remove(tid)
1134 # Remove pending tasks from the workers queue
1135 self._runner.empty()
1139 # Mark the EC state as TERMINATED
1140 self._state = ECState.TERMINATED
1142 # Stop processing thread
1145 # Notify condition to wake up the processing thread
1148 if self._thread.is_alive():
1151 def schedule(self, date, callback, track = False):
1152 """ Schedules a callback to be executed at time 'date'.
1154 :param date: string containing execution time for the task.
1155 It can be expressed as an absolute time, using
1156 timestamp format, or as a relative time matching
1157 ^\d+.\d+(h|m|s|ms|us)$
1159 :param callback: code to be executed for the task. Must be a
1160 Python function, and receives args and kwargs
1163 :param track: if set to True, the task will be retrievable with
1164 the get_task() method
1166 :return : The Id of the task
1170 timestamp = stabsformat(date)
1171 task = Task(timestamp, callback)
1172 task = self._scheduler.schedule(task)
1175 self._tasks[task.id] = task
1177 # Notify condition to wake up the processing thread
1183 """ Process scheduled tasks.
1187 Tasks are scheduled by invoking the schedule method with a target
1188 callback and an execution time.
1189 The schedule method creates a new Task object with that callback
1190 and execution time, and pushes it into the '_scheduler' queue.
1191 The execution time and the order of arrival of tasks are used
1192 to order the tasks in the queue.
1194 The _process method is executed in an independent thread held by
1195 the ExperimentController for as long as the experiment is running.
1196 This method takes tasks from the '_scheduler' queue in a loop
1197 and processes them in parallel using multithreading.
1198 The environmental variable NEPI_NTHREADS can be used to control
1199 the number of threads used to process tasks. The default value is
1202 To execute tasks in parallel, a ParallelRunner (PR) object is used.
1203 This object keeps a pool of threads (workers), and a queue of tasks
1204 scheduled for 'immediate' execution.
1206 On each iteration, the '_process' loop will take the next task that
1207 is scheduled for 'future' execution from the '_scheduler' queue,
1208 and if the execution time of that task is >= to the current time,
1209 it will push that task into the PR for 'immediate execution'.
1210 As soon as a worker is free, the PR will assign the next task to
1213 Upon receiving a task to execute, each PR worker (thread) will
1214 invoke the _execute method of the EC, passing the task as
1216 The _execute method will then invoke task.callback inside a
1217 try/except block. If an exception is raised by the tasks.callback,
1218 it will be trapped by the try block, logged to standard error
1219 (usually the console), and the task will be marked as failed.
1223 self._nthreads = int(os.environ.get("NEPI_NTHREADS", str(self._nthreads)))
1224 self._runner = ParallelRun(maxthreads = self.nthreads)
1225 self._runner.start()
1227 while not self._stop:
1229 self._cond.acquire()
1231 task = next(self._scheduler)
1234 # No task to execute. Wait for a new task to be scheduled.
1237 # The task timestamp is in the future. Wait for timeout
1238 # or until another task is scheduled.
1240 if now < task.timestamp:
1241 # Calculate timeout in seconds
1242 timeout = tdiffsec(task.timestamp, now)
1244 # Re-schedule task with the same timestamp
1245 self._scheduler.schedule(task)
1249 # Wait timeout or until a new task awakes the condition
1250 self._cond.wait(timeout)
1252 self._cond.release()
1255 # Process tasks in parallel
1256 self._runner.put(self._execute, task)
1259 err = traceback.format_exc()
1260 self.logger.error("Error while processing tasks in the EC: %s" % err)
1262 # Set the EC to FAILED state
1263 self._state = ECState.FAILED
1265 # Set the FailureManager failure level to EC failure
1266 self._fm.set_ec_failure()
1268 self.logger.debug("Exiting the task processing loop ... ")
1271 self._runner.destroy()
1273 def _execute(self, task):
1274 """ Executes a single task.
1276 :param task: Object containing the callback to execute
1282 task.result = task.callback()
1283 task.status = TaskStatus.DONE
1286 err = traceback.format_exc()
1288 task.status = TaskStatus.ERROR
1290 self.logger.error("Error occurred while executing task: %s" % err)
1293 """ Awakes the processing thread if it is blocked waiting
1294 for new tasks to arrive
1297 self._cond.acquire()
1299 self._cond.release()
1301 def _build_from_netgraph(self, add_node_callback, add_edge_callback,
1303 """ Automates experiment description using a NetGraph instance.
1305 self._netgraph = NetGraph(**kwargs)
1307 if add_node_callback:
1308 ### Add resources to the EC
1309 for nid in self.netgraph.nodes():
1310 add_node_callback(self, nid)
1312 if add_edge_callback:
1313 #### Add connections between resources
1314 for nid1, nid2 in self.netgraph.edges():
1315 add_edge_callback(self, nid1, nid2)