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>
28 from nepi.util import guid
29 from nepi.util.parallel import ParallelRun
30 from nepi.util.timefuncs import tnow, tdiffsec, stabsformat, tsformat
31 from nepi.execution.resource import ResourceFactory, ResourceAction, \
32 ResourceState, ResourceState2str
33 from nepi.execution.scheduler import HeapScheduler, Task, TaskStatus
34 from nepi.execution.trace import TraceAttr
36 # TODO: use multiprocessing instead of threading
37 # TODO: When a failure occurs during deployment, scp and ssh processes are left running behind!!
38 # TODO: Allow to reconnect to a running experiment instance! (reconnect mode vs deploy mode)
40 class ECState(object):
41 """ State of the Experiment Controller
48 class ExperimentController(object):
50 .. class:: Class Args :
52 :param exp_id: Human readable identifier for the experiment scenario.
53 It will be used in the name of the directory
54 where experiment related information is stored
59 An experiment, or scenario, is defined by a concrete use, behavior,
60 configuration and interconnection of resources that describe a single
61 experiment case (We call this the experiment description).
62 A same experiment (scenario) can be run many times.
64 The ExperimentController (EC), is the entity responsible for
65 managing an experiment instance (run). The same scenario can be
66 recreated (and re-run) by instantiating an EC and recreating
67 the same experiment description.
69 In NEPI, an experiment is represented as a graph of interconnected
70 resources. A resource is a generic concept in the sense that any
71 component taking part of an experiment, whether physical of
72 virtual, is considered a resource. A resources could be a host,
73 a virtual machine, an application, a simulator, a IP address.
75 A ResourceManager (RM), is the entity responsible for managing a
76 single resource. ResourceManagers are specific to a resource
77 type (i.e. An RM to control a Linux application will not be
78 the same as the RM used to control a ns-3 simulation).
79 In order for a new type of resource to be supported in NEPI
80 a new RM must be implemented. NEPI already provides different
81 RMs to control basic resources, and new can be extended from
84 Through the EC interface the user can create ResourceManagers (RMs),
85 configure them and interconnect them, in order to describe an experiment.
86 Describing an experiment through the EC does not run the experiment.
87 Only when the 'deploy()' method is invoked on the EC, will the EC take
88 actions to transform the 'described' experiment into a 'running' experiment.
90 While the experiment is running, it is possible to continue to
91 create/configure/connect RMs, and to deploy them to involve new
92 resources in the experiment (this is known as 'interactive' deployment).
94 An experiments in NEPI is identified by a string id,
95 which is either given by the user, or automatically generated by NEPI.
96 The purpose of this identifier is to separate files and results that
97 belong to different experiment scenarios.
98 However, since a same 'experiment' can be run many times, the experiment
99 id is not enough to identify an experiment instance (run).
100 For this reason, the ExperimentController has two identifier, the
101 exp_id, which can be re-used by different ExperimentController instances,
102 and the run_id, which unique to a ExperimentController instance, and
103 is automatically generated by NEPI.
107 def __init__(self, exp_id = None):
108 super(ExperimentController, self).__init__()
109 # root directory to store files
111 # Run identifier. It identifies a concrete instance (run) of an experiment.
112 # Since a same experiment (same configuration) can be run many times,
113 # this id permits to identify concrete exoeriment run
114 self._run_id = tsformat()
116 # Experiment identifier. Usually assigned by the user
117 self._exp_id = exp_id or "exp-%s" % os.urandom(8).encode('hex')
119 # generator of globally unique ids
120 self._guid_generator = guid.GuidGenerator()
123 self._resources = dict()
126 self._scheduler = HeapScheduler()
131 # Event processing thread
132 self._cond = threading.Condition()
133 self._thread = threading.Thread(target = self._process)
134 self._thread.setDaemon(True)
138 self._state = ECState.RUNNING
141 self._logger = logging.getLogger("ExperimentController")
145 """ Return the logger of the Experiment Controller
152 """ Return the state of the Experiment Controller
159 """ Return the experiment id assigned by the user
166 """ Return the experiment instance (run) identifier
173 """ Put the state of the Experiment Controller into a final state :
174 Either TERMINATED or FAILED
177 return self.ecstate in [ECState.FAILED, ECState.TERMINATED]
179 def wait_finished(self, guids):
180 """ Blocking method that wait until all the RM from the 'guid' list
181 reached the state FINISHED
183 :param guids: List of guids
186 return self.wait(guids)
188 def wait_started(self, guids):
189 """ Blocking method that wait until all the RM from the 'guid' list
190 reached the state STARTED
192 :param guids: List of guids
195 return self.wait(guids, states = [ResourceState.STARTED,
196 ResourceState.STOPPED,
197 ResourceState.FINISHED])
199 def wait(self, guids, states = [ResourceState.FINISHED,
200 ResourceState.STOPPED]):
201 """ Blocking method that waits until all the RM from the 'guid' list
202 reached state 'state' or until a failure occurs
204 :param guids: List of guids
207 if isinstance(guids, int):
210 while not all([self.state(guid) in states for guid in guids]) and \
211 not any([self.state(guid) in [
212 ResourceState.FAILED] for guid in guids]) and \
217 waited += "guid %d - %s \n" % (guid, self.state(guid, hr = True))
218 self.logger.debug(" WAITING FOR %s " % waited )
220 # We keep the sleep big to decrease the number of RM state queries
223 def get_task(self, tid):
224 """ Get a specific task
226 :param tid: Id of the task
230 return self._tasks.get(tid)
232 def get_resource(self, guid):
233 """ Get a specific Resource Manager
235 :param guid: Id of the task
237 :rtype: ResourceManager
239 return self._resources.get(guid)
243 """ Returns the list of all the Resource Manager Id
248 return self._resources.keys()
250 def register_resource(self, rtype, guid = None):
251 """ Register a Resource Manager. It creates a new 'guid', if it is not specified,
252 for the RM of type 'rtype' and add it to the list of Resources.
254 :param rtype: Type of the RM
256 :return: Id of the RM
259 # Get next available guid
260 guid = self._guid_generator.next(guid)
263 rm = ResourceFactory.create(rtype, self, guid)
266 self._resources[guid] = rm
270 def get_attributes(self, guid):
271 """ Return all the attibutes of a specific RM
273 :param guid: Guid of the RM
275 :return: List of attributes
278 rm = self.get_resource(guid)
279 return rm.get_attributes()
281 def register_connection(self, guid1, guid2):
282 """ Registers a guid1 with a guid2.
283 The declaration order is not important
285 :param guid1: First guid to connect
286 :type guid1: ResourceManager
288 :param guid2: Second guid to connect
289 :type guid: ResourceManager
291 rm1 = self.get_resource(guid1)
292 rm2 = self.get_resource(guid2)
294 rm1.register_connection(guid2)
295 rm2.register_connection(guid1)
297 def register_condition(self, group1, action, group2, state,
299 """ Registers an action START or STOP for all RM on group1 to occur
300 time 'time' after all elements in group2 reached state 'state'.
302 :param group1: List of guids of RMs subjected to action
305 :param action: Action to register (either START or STOP)
306 :type action: ResourceAction
308 :param group2: List of guids of RMs to we waited for
311 :param state: State to wait for on RMs (STARTED, STOPPED, etc)
312 :type state: ResourceState
314 :param time: Time to wait after group2 has reached status
318 if isinstance(group1, int):
320 if isinstance(group2, int):
324 rm = self.get_resource(guid1)
325 rm.register_condition(action, group2, state, time)
327 def register_trace(self, guid, name):
330 :param name: Name of the trace
333 rm = self.get_resource(guid)
334 rm.register_trace(name)
336 def trace(self, guid, name, attr = TraceAttr.ALL, block = 512, offset = 0):
337 """ Get information on collected trace
339 :param name: Name of the trace
342 :param attr: Can be one of:
343 - TraceAttr.ALL (complete trace content),
344 - TraceAttr.STREAM (block in bytes to read starting at offset),
345 - TraceAttr.PATH (full path to the trace file),
346 - TraceAttr.SIZE (size of trace file).
349 :param block: Number of bytes to retrieve from trace, when attr is TraceAttr.STREAM
352 :param offset: Number of 'blocks' to skip, when attr is TraceAttr.STREAM
357 rm = self.get_resource(guid)
358 return rm.trace(name, attr, block, offset)
360 def discover(self, guid):
361 """ Discover a specific RM defined by its 'guid'
363 :param guid: Guid of the RM
367 rm = self.get_resource(guid)
370 def provision(self, guid):
371 """ Provision a specific RM defined by its 'guid'
373 :param guid: Guid of the RM
377 rm = self.get_resource(guid)
378 return rm.provision()
380 def get(self, guid, name):
381 """ Get a specific attribute 'name' from the RM 'guid'
383 :param guid: Guid of the RM
386 :param name: attribute's name
390 rm = self.get_resource(guid)
393 def set(self, guid, name, value):
394 """ Set a specific attribute 'name' from the RM 'guid'
395 with the value 'value'
397 :param guid: Guid of the RM
400 :param name: attribute's name
403 :param value: attribute's value
406 rm = self.get_resource(guid)
407 return rm.set(name, value)
409 def state(self, guid, hr = False):
410 """ Returns the state of a resource
412 :param guid: Resource guid
415 :param hr: Human readable. Forces return of a
416 status string instead of a number
420 rm = self.get_resource(guid)
424 return ResourceState2str.get(state)
428 def stop(self, guid):
429 """ Stop a specific RM defined by its 'guid'
431 :param guid: Guid of the RM
435 rm = self.get_resource(guid)
438 def start(self, guid):
439 """ Start a specific RM defined by its 'guid'
441 :param guid: Guid of the RM
445 rm = self.get_resource(guid)
448 def set_with_conditions(self, name, value, group1, group2, state,
450 """ Set value 'value' on attribute with name 'name' on all RMs of
451 group1 when 'time' has elapsed since all elements in group2
452 have reached state 'state'.
454 :param name: Name of attribute to set in RM
457 :param value: Value of attribute to set in RM
460 :param group1: List of guids of RMs subjected to action
463 :param action: Action to register (either START or STOP)
464 :type action: ResourceAction
466 :param group2: List of guids of RMs to we waited for
469 :param state: State to wait for on RMs (STARTED, STOPPED, etc)
470 :type state: ResourceState
472 :param time: Time to wait after group2 has reached status
476 if isinstance(group1, int):
478 if isinstance(group2, int):
482 rm = self.get_resource(guid)
483 rm.set_with_conditions(name, value, group2, state, time)
485 def stop_with_conditions(self, guid):
486 """ Stop a specific RM defined by its 'guid' only if all the conditions are true
488 :param guid: Guid of the RM
492 rm = self.get_resource(guid)
493 return rm.stop_with_conditions()
495 def start_with_conditions(self, guid):
496 """ Start a specific RM defined by its 'guid' only if all the conditions are true
498 :param guid: Guid of the RM
502 rm = self.get_resource(guid)
503 return rm.start_with_condition()
505 def deploy(self, group = None, wait_all_ready = True):
506 """ Deploy all resource manager in group
508 :param group: List of guids of RMs to deploy
511 :param wait_all_ready: Wait until all RMs are ready in
512 order to start the RMs
516 self.logger.debug(" ------- DEPLOY START ------ ")
519 # By default, if not deployment group is indicated,
520 # all RMs that are undeployed will be deployed
522 for guid in self.resources:
523 if self.state(guid) == ResourceState.NEW:
526 if isinstance(group, int):
529 # Before starting deployment we disorder the group list with the
530 # purpose of speeding up the whole deployment process.
531 # It is likely that the user inserted in the 'group' list closely
532 # resources one after another (e.g. all applications
533 # connected to the same node can likely appear one after another).
534 # This can originate a slow down in the deployment since the N
535 # threads the parallel runner uses to processes tasks may all
536 # be taken up by the same family of resources waiting for the
537 # same conditions (e.g. LinuxApplications running on a same
538 # node share a single lock, so they will tend to be serialized).
539 # If we disorder the group list, this problem can be mitigated.
540 random.shuffle(group)
542 def wait_all_and_start(group):
545 if self.state(guid) < ResourceState.READY:
550 callback = functools.partial(wait_all_and_start, group)
551 self.schedule("1s", callback)
553 # If all resources are read, we schedule the start
555 rm = self.get_resource(guid)
556 self.schedule("0s", rm.start_with_conditions)
559 # Schedule the function that will check all resources are
560 # READY, and only then it will schedule the start.
561 # This is aimed to reduce the number of tasks looping in the scheduler.
562 # Intead of having N start tasks, we will have only one
563 callback = functools.partial(wait_all_and_start, group)
564 self.schedule("1s", callback)
567 rm = self.get_resource(guid)
568 self.schedule("0s", rm.deploy)
570 if not wait_all_ready:
571 self.schedule("1s", rm.start_with_conditions)
573 if rm.conditions.get(ResourceAction.STOP):
574 # Only if the RM has STOP conditions we
575 # schedule a stop. Otherwise the RM will stop immediately
576 self.schedule("2s", rm.stop_with_conditions)
578 def release(self, group = None):
579 """ Release the elements of the list 'group' or
580 all the resources if any group is specified
582 :param group: List of RM
587 group = self.resources
591 rm = self.get_resource(guid)
592 thread = threading.Thread(target=rm.release)
593 threads.append(thread)
594 thread.setDaemon(True)
597 while list(threads) and not self.finished:
599 # Time out after 5 seconds to check EC not terminated
601 if not thread.is_alive():
602 threads.remove(thread)
605 """ Shutdown the Experiment Controller.
606 Releases all the resources and stops task processing thread
611 # Mark the EC state as TERMINATED
612 self._state = ECState.TERMINATED
614 # Notify condition to wake up the processing thread
617 if self._thread.is_alive():
620 def schedule(self, date, callback, track = False):
621 """ Schedule a callback to be executed at time date.
623 :param date: string containing execution time for the task.
624 It can be expressed as an absolute time, using
625 timestamp format, or as a relative time matching
626 ^\d+.\d+(h|m|s|ms|us)$
628 :param callback: code to be executed for the task. Must be a
629 Python function, and receives args and kwargs
632 :param track: if set to True, the task will be retrivable with
633 the get_task() method
635 :return : The Id of the task
637 timestamp = stabsformat(date)
638 task = Task(timestamp, callback)
639 task = self._scheduler.schedule(task)
642 self._tasks[task.id] = task
644 # Notify condition to wake up the processing thread
650 """ Process scheduled tasks.
654 The _process method is executed in an independent thread held by the
655 ExperimentController for as long as the experiment is running.
657 Tasks are scheduled by invoking the schedule method with a target callback.
658 The schedule method is given a execution time which controls the
659 order in which tasks are processed.
661 Tasks are processed in parallel using multithreading.
662 The environmental variable NEPI_NTHREADS can be used to control
663 the number of threads used to process tasks. The default value is 50.
667 To execute tasks in parallel, an ParallelRunner (PR) object, holding
668 a pool of threads (workers), is used.
669 For each available thread in the PR, the next task popped from
670 the scheduler queue is 'put' in the PR.
671 Upon receiving a task to execute, each PR worker (thread) invokes the
672 _execute method of the EC, passing the task as argument.
673 This method, calls task.callback inside a try/except block. If an
674 exception is raised by the tasks.callback, it will be trapped by the
675 try block, logged to standard error (usually the console), and the EC
676 state will be set to ECState.FAILED.
677 The invocation of _notify immediately after, forces the processing
678 loop in the _process method, to wake up if it was blocked waiting for new
679 tasks to arrived, and to check the EC state.
680 As the EC is in FAILED state, the processing loop exits and the
681 'finally' block is invoked. In the 'finally' block, the 'sync' method
682 of the PR is invoked, which forces the PR to raise any unchecked errors
683 that might have been raised by the workers.
686 nthreads = int(os.environ.get("NEPI_NTHREADS", "50"))
688 runner = ParallelRun(maxthreads = nthreads)
692 while not self.finished:
695 task = self._scheduler.next()
698 # No task to execute. Wait for a new task to be scheduled.
701 # The task timestamp is in the future. Wait for timeout
702 # or until another task is scheduled.
704 if now < task.timestamp:
705 # Calculate timeout in seconds
706 timeout = tdiffsec(task.timestamp, now)
708 # Re-schedule task with the same timestamp
709 self._scheduler.schedule(task)
713 # Wait timeout or until a new task awakes the condition
714 self._cond.wait(timeout)
719 # Process tasks in parallel
720 runner.put(self._execute, task)
723 err = traceback.format_exc()
724 self.logger.error("Error while processing tasks in the EC: %s" % err)
726 self._state = ECState.FAILED
728 self.logger.debug("Exiting the task processing loop ... ")
731 def _execute(self, task):
732 """ Executes a single task.
734 :param task: Object containing the callback to execute
739 If the invokation of the task callback raises an
740 exception, the processing thread of the ExperimentController
741 will be stopped and the experiment will be aborted.
745 task.status = TaskStatus.DONE
748 task.result = task.callback()
751 err = traceback.format_exc()
753 task.status = TaskStatus.ERROR
755 self.logger.error("Error occurred while executing task: %s" % err)
757 # Set the EC to FAILED state (this will force to exit the task
759 self._state = ECState.FAILED
761 # Notify condition to wake up the processing thread
764 # Propage error to the ParallelRunner
768 """ Awakes the processing thread in case it is blocked waiting
769 for a new task to be scheduled.