def run_scheduler_worker(env: simpy.Environment, queue: simpy.Store, context: ClusterContext, scheduler: Scheduler,
oracles: List[Oracle], log: List[LoggingRow]):
while True:
logging.debug('Scheduler waiting for pod...')
pod = yield queue.get()
# TODO fix time not changing (env.now)
logging.debug('Pod received by scheduler at %.2f', env.now)
log.append(LoggingRow(env.now, EventType.POD_RECEIVED, pod.name))
# execute scheduling algorithm
then = time.time()
result = scheduler.schedule(pod)
duration = time.time() - then
yield env.timeout(duration)
logging.debug('Pod scheduling took %.2f ms, and yielded %s', duration * 1000, result)
# weight the startup
metadata = dict([o.estimate(context, pod, result) for o in oracles])
# also add the image name and the selected node to the metadata
metadata['image'] = pod.spec.containers[0].image
metadata['suggested_host'] = None if result.suggested_host is None else result.suggested_host.name
log.append(LoggingRow(env.now, EventType.POD_SCHEDULED, pod.name, metadata))
def drop_object(obj_name, me):
pos = _WORLD.component_for_entity(me, Position)
inventory = _WORLD.component_for_entity(me, Inventory)
skeleton = inventory.objects.get(obj_name, None)
success: bool = False
msg: str = "Something went wrong"
if skeleton is None:
msg = f'Not holding object {obj_name}'
else:
reply_channel = Store(_ENV)
drop_offset = (pos.center[0], pos.center[1] + pos.h)
drop_payload = ObjectManager.DropPayload(
obj_name, ObjectManager.ObjectManagerOps.RECREATE, skeleton,
drop_offset, reply_channel)
_EVENT_STORE.put(EVENT(ObjectManager.ManagerTag, drop_payload))
# Wait for reply
response = yield reply_channel.get()
if response.get('success', False):
success = True
msg = ''
response = RESPONSE_ClawPayload(op=ClawOps.DROP,
ent=me,
success=success,
msg=msg)
event_to_put = EVENT(ClawDoneTag, response)
_EVENT_STORE.put(event_to_put)
return success, msg
def pick_object(obj_name: str, me: int):
pos = _WORLD.component_for_entity(me, Position)
claw = _WORLD.component_for_entity(me, Claw)
success: bool = False
msg: str = f'Object {obj_name} not found.'
# Create boundaries, if necessary
if claw.boundaries is None:
span = Ellipse(pos.center, claw.max_range, claw.max_range)
col = Collidable(shape=helpers.collision_from_points(span, pos.center))
claw.boundaries = col
# For every pickable component, see if it's within range
for _, (pick, col) in _WORLD.get_components(Pickable, Collidable):
if pick.name == obj_name:
# This is the object we want. Let's see if it's in range and under limit weight
for s1 in col.shapes:
if collide(claw.boundaries.shapes[0], s1):
if pick.weight <= claw.max_weight:
# Take the object
reply_channel = Store(_ENV)
payload = ObjectManager.GrabPayload(
obj_name, ObjectManager.ObjectManagerOps.REMOVE,
reply_channel)
event = EVENT(ObjectManager.ManagerTag, payload)
_EVENT_STORE.put(event)
# Wait for reply
response = yield reply_channel.get()
if response.get('success', False):
success = True
# Add removed component to my inventory
if not _WORLD.has_component(me, Inventory):
_WORLD.add_component(me, Inventory())
inventory = _WORLD.component_for_entity(
me, Inventory)
inventory.objects[obj_name] = pick.skeleton
msg = f'Picked {obj_name}. My inventory: {inventory.objects}'
else:
success = False
msg = response.get('msg', '')
else:
msg = f'Pickable {obj_name} too heavy. Max weight:{claw.max_weight}. Object weight: {pick.weight}'
success = False
else:
msg = f'Pickable {obj_name} not within claw range!'
success = False
response = RESPONSE_ClawPayload(op=ClawOps.GRAB,
ent=me,
success=success,
msg=msg)
event_to_put = EVENT(ClawDoneTag, response)
_EVENT_STORE.put(event_to_put)
return success, msg
def machine(env: simpy.Environment, last_q: simpy.Store, next_q: simpy.Store,
machine_id: str):
"""模拟一个机器, 一个机器就可以同时处理多少物件 取决资源数(工人数)"""
workers = simpy.Resource(env, capacity=WORKER_NUM)
def process(item):
"""模拟一个工人的工作进程"""
with workers.request() as req:
yield req
yield env.timeout(PROCESS_TIME)
env.process(con_belt_process(env, CON_BELT_TIME, item, next_q))
print(
f'{round(env.now, 2)} - item: {item} - machine工人: {machine_id} - processed正在处理'
)
while True:
item = yield last_q.get()
env.process(process(item))
class Connection:
"""This class represents the propagation through a Connection."""
def __init__(self, env, origin_node, destination_node):
self.env = env
self.store = Store(env)
self.origin_node = origin_node
self.destination_node = destination_node
def latency(self, envelope):
latency_delay = get_latency_delay(self.env, self.origin_node.location,
self.destination_node.location)
yield self.env.timeout(latency_delay)
self.store.put(envelope)
def put(self, envelope):
print(
f'{envelope.origin.address} at {envelope.timestamp}: Message (ID: {envelope.msg["id"]}) sent with {envelope.msg["size"]} MB with a destination: {envelope.destination.address}'
)
self.env.process(self.latency(envelope))
def get(self):
return self.store.get()
class Storage:
def __init__(self, sim, buffer_params):
"""Returns a buffer object for a Buffer object in MOCA
:param sim: Reference to DES simulation object
:param buffer_params: Dictionary of parameters
"""
# Name and size of the buffer object
self.name = buffer_params['name']
self.size = buffer_params['size']
# The parent simulation object
self.procflow_sim = sim
# References to upstream and downstream processes
self.upstream_process = None
self.downstream_process = None
# Internal buffer
self.buffer = Store(self.procflow_sim.simpy_env, capacity=self.size)
# Buffer state variables
self.buffer_metrics = {'usage': 0}
def store(self, part):
"""Stores the given part in the buffer
:param part: Part object to be stored
:return: The event of the buffer
"""
yield self.buffer.put(part)
def retrieve(self):
"""Retrieves the first part from the buffer
:return: The part from the buffer (or waits for the buffer till a part arrives)
"""
yield self.buffer.get()
class Comm(Component):
def __init__(self, nname, env):
super().__init__(nname)
conf = Configuration()
self.queue = Store(env)
self.tcomm = conf.get('[comm]t_comm')
self.perr = conf.get('[comm]p_err')
self.unavail = conf.get('[comm]unavail')
def put(self, msg):
guess = uniform(0, 1)
if (guess > self.unavail):
yield self.env.timeout(self.tcomm)
while (random() < self.perr):
self.debug('errorInMessage;;')
yield self.env.timeout(self.tcomm)
self.queue.put(msg)
else:
self.debug('unavailabilityOfCommunication;;')
def get(self):
msg = yield self.queue.get()
return msg
class Worker:
DOWNLOAD_PRIORITY_BOOST_FOR_READY_TASK = 100000
def __init__(self, cpus=1, max_downloads=4, max_downloads_per_worker=2):
self.cpus = cpus
self.assignments = []
self.ready_store = None
self.data = set()
self.running_tasks = {}
self.scheduled_downloads = {}
self.running_downloads = []
self.free_cpus = cpus
self.max_downloads = max_downloads
self.max_downloads_per_worker = max_downloads_per_worker
self.id = None
def copy(self):
return Worker(cpus=self.cpus,
max_downloads=self.max_downloads,
max_downloads_per_worker=self.max_downloads_per_worker)
def _download(self, output, priority):
logger.info("Worker %s: scheduled downloading %s, priority=%s", self,
output, priority)
assert output not in self.data
download = Download(Event(self.env), output, priority)
self.scheduled_downloads[output] = download
return download
def assign_tasks(self, assignments):
for assignment in assignments:
assert assignment.worker == self
# assert assignment not in self.assignments
if assignment.task.cpus > self.cpus:
raise Exception(
"Task {} allocated on worker with {} cpus".format(
assignment.task, self.cpus))
self.assignments.append(assignment)
self._init_downloads(assignment)
if not self.download_wakeup.triggered:
self.download_wakeup.succeed()
def update_tasks(self, tasks):
for task in tasks:
for assignment in self.assignments:
if task == assignment.task:
self._init_downloads(assignment)
break
else:
raise Exception(
"Updating non assigned task {}, worker={}".format(
task, self))
if not self.download_wakeup.triggered:
self.download_wakeup.succeed()
@property
def assigned_tasks(self):
return [a.task for a in self.assignments]
def _init_downloads(self, assignment):
deps = []
not_complete = False
runtime_state = self.simulator.runtime_state
for input in assignment.task.inputs:
if input in self.data:
continue
priority = assignment.priority
if runtime_state.task_info(assignment.task).is_ready:
priority += self.DOWNLOAD_PRIORITY_BOOST_FOR_READY_TASK
d = self.scheduled_downloads.get(input)
if d is None:
info = runtime_state.output_info(input)
if info.placing:
if input.size == 0:
self.data.add(input)
continue
d = self._download(input, priority)
deps.append(d.event)
else:
not_complete = True
else:
d.update_priority(priority)
deps.append(d.event)
if not_complete:
return
if not deps:
self.ready_store.put(assignment)
else:
e = self.env.all_of(deps)
e.callbacks.append(lambda _: self.ready_store.put(assignment))
def _download_process(self):
events = [self.download_wakeup]
env = self.env
runtime_state = self.simulator.runtime_state
while True:
finished = yield env.any_of(events)
for event in finished.keys():
if event == events[0]:
self.download_wakeup = Event(self.simulator.env)
events[0] = self.download_wakeup
downloads = None
continue
events.remove(event)
download = event.value
assert download.output not in self.data
self.data.add(download.output)
self.running_downloads.remove(download)
del self.scheduled_downloads[download.output]
download.event.succeed(download)
self.simulator.add_trace_event(
FetchEndTraceEvent(self.env.now, self, download.source,
download.output))
if len(self.running_downloads) < self.max_downloads:
# We need to sort any time, as it priority may changed in background
if downloads is None:
downloads = list(
o for o in self.scheduled_downloads.values()
if o not in self.running_downloads)
downloads.sort(key=lambda d: d.priority, reverse=True)
for d in downloads[:]:
count = 0
worker = runtime_state.output_info(d.output).placing[0]
for rd in self.running_downloads:
if worker == rd.source:
count += 1
if count >= self.max_downloads_per_worker:
continue
downloads.remove(d)
d.start_time = self.env.now
d.source = worker
self.running_downloads.append(d)
event = self.netmodel.download(worker, self, d.output.size,
d)
events.append(event)
self.simulator.add_trace_event(
FetchStartTraceEvent(self.env.now, self, worker,
d.output))
if len(self.running_downloads) >= self.max_downloads:
break
def run(self, env, simulator, netmodel):
self.env = env
self.simulator = simulator
self.netmodel = netmodel
self.ready_store = Store(env)
self.download_wakeup = Event(self.simulator.env)
self.free_cpus = self.cpus
env.process(self._download_process())
prepared_assignments = []
events = [self.ready_store.get()]
while True:
finished = yield env.any_of(events)
for event in finished.keys():
if event == events[0]:
prepared_assignments.append(event.value)
prepared_assignments.sort(key=lambda a: a.priority,
reverse=True)
events[0] = self.ready_store.get()
continue
assignment = event.value
task = assignment.task
self.free_cpus += task.cpus
self.assignments.remove(assignment)
events.remove(event)
del self.running_tasks[task]
simulator.add_trace_event(
TaskEndTraceEvent(self.env.now, self, task))
for output in task.outputs:
self.data.add(output)
simulator.on_task_finished(self, task)
block = float("-inf")
for assignment in prepared_assignments[:]:
if assignment.priority < block:
continue
task = assignment.task
if task.cpus <= self.free_cpus:
self.free_cpus -= task.cpus
self.running_tasks[task] = RunningTask(task, self.env.now)
simulator.add_trace_event(
TaskStartTraceEvent(self.env.now, self, task))
events.append(env.timeout(task.duration, assignment))
prepared_assignments.remove(assignment)
else:
block = max(block, assignment.block)
def __repr__(self):
return "<Worker {}>".format(self.id)
class Peer:
def __init__(self, sim, peer_id: int, peer_type: str,
location: str, bandwidth_ul: float, bandwidth_dl: float, **kwargs):
"""
Physical representation of a Peer
sim: Simulation environment
name: Info about peer type and peer id
location: Physical location of peer
bandwidth_ul: Uplink bandwidth
bandwidth_dl: Downlink bandwidth
"""
self.sim = sim
self.env = sim.env
self.peer_type = peer_type
self.peer_id = peer_id
self.name = str(peer_id) + ":" + str(peer_type)
self.location = location
self.bandwidth_ul = bandwidth_ul
self.bandwidth_dl = bandwidth_dl
self.__dict__.update(kwargs)
peer_repr = repr(self)
# Define log file path for results of the simulation
if sim.sim_dir:
self.log_name = os.path.join(sim.sim_dir, peer_repr + ".log")
self.logger = setup_logger(peer_repr, self.log_name)
else:
self.log_name = None
self.logger = None
# Message queue for the received messages
self.msg_queue = Store(self.env)
# Network connections that are online
self.online = True
self.connections = dict()
# Known peers
self.disconnect_callbacks = []
self.last_seen = dict()
# Peer services
self.handlers = {} # Service.handle_message(self, msg) called on message
self.mh_map = {} # Message -> Handler map
self.runners = {} # Peer service runners
self.mprt_map = {} # Message -> Pre-Receive Trigger
# Storage
self.storage = {}
# Monitoring services
self.bytes_load = {} # Overhead on bytes per sec
self.msg_count_load = {} # Msg per sec
# Start peer as it is created
self.env.process(self.run())
def __repr__(self):
return '%s_%s' % (self.__class__.__name__, self.name)
def __lt__(self, other):
return self.name < other.name
def run(self):
while True:
# Receive message
msg = yield self.msg_queue.get()
num_bytes = msg.size
sender = msg.sender
delay = num_bytes / self.bandwidth_dl
yield self.env.timeout(delay)
# Trigger pre-receive tasks if any
if msg.pre_task:
val = yield self.env.process(msg.pre_task(msg, self))
if not msg.pre_task or val:
self.receive(msg)
# Trigger post-receive tasks if any
# Check for connection with any particular peer object
def is_connected(self, other):
return other in self.connections
def bootstrap_connect(self, other):
"""
Create ad-hoc connection and send Hello
other: peer object
"""
#
cnx = Connection(self, other)
cnx.send(Hello(self), connect=True)
def connect(self, other):
"""
Add peer to the connections and repeat the same with other peer
other: peer object
"""
if not self.is_connected(other):
if self.logger:
self.logger.info("%s: Connecting to %s", self.env.now, repr(other))
self.connections[other] = Connection(self, other)
# We create bilateral connection
if not other.is_connected(self):
other.connect(self)
def disconnect(self, other):
"""
Disconnect with previously connected peer
other: peer object
"""
if self.is_connected(other):
if self.logger:
self.logger.warning("%s: Breaking connection with %s", self.env.now, repr(other))
del self.connections[other]
if other.is_connected(self):
other.disconnect(self)
for cb in self.disconnect_callbacks:
cb(self, other)
def receive(self, msg):
"""
Receive message, will trigger handlers on the message
msg: message object
"""
if self.online:
msg_sender = msg.sender
# Monitor the overhead of the message size
now_sec = int(self.env.now / 1000)
self.bytes_load[now_sec] = self.bytes_load.get(now_sec, 0) + msg.size
self.msg_count_load[now_sec] = self.msg_count_load.get(now_sec, 0) + 1
# Update peer connection data
self.last_seen[msg_sender] = self.env.now
if self.logger:
self.logger.info("%s: Received message <%s> from %s", self.env.now, repr(msg), msg_sender)
# Find the services that should be triggered
services = set()
for msg_type in self.mh_map:
if isinstance(msg, msg_type):
services.update(self.mh_map[msg_type])
if not services:
if self.logger:
self.logger.error("No handler for the message %s", type(msg))
raise Exception("No handler for the message ", type(msg), repr(self))
else:
for service_id in services:
self.handlers[service_id].handle_message(msg)
def send(self, receiver, msg):
"""
Send to a receiver peer in a fire-and-forget fashion.
If receiver is not connected will raise and exception
"""
# fire and forget
if self.online:
if receiver not in self.connections:
if self.logger:
self.logger.error("%s: Sending message to a not connected peer %s",
self.env.now, repr(receiver))
raise Exception("Not connected")
if self.logger:
self.logger.info("%s: Sending message <%s> to %s", self.env.now, repr(msg), receiver)
self.connections[receiver].send(msg)
# Get all peer connections
def _get_connections(self, exclude_bootstrap=True, except_set: set = None, except_type: set = None):
if except_set is None:
except_set = set()
if except_type is None:
except_type = set()
if exclude_bootstrap:
except_type.add('bootstrap')
conn_set = set(self.connections.keys()) - except_set
return (p for p in conn_set if p.peer_type not in except_type)
def gossip(self, msg, f, exclude_bootstrap=True, except_peers: set = None, except_type: set = None):
"""
Send to f neighbours selected randomly
msg: Message object
f: the fanout parameter (number of peers to gossip to)
exclude_bootstrap: Exclude bootstrap from gossip
except_peers: connected peers to exclude from gossip
except_type: exclude from gossip type of peers
"""
if not self.online:
return None
gossip_set = list(self._get_connections(exclude_bootstrap, except_peers, except_type))
selected = random.sample(gossip_set, min(f, len(gossip_set)))
for other in selected:
self.send(other, msg)
return selected
def broadcast(self, msg, exclude_bootstrap=True, except_set: set = None, except_type: set = None):
"""Send to all connected peers except given """
for other in self._get_connections(exclude_bootstrap, except_set, except_type):
self.send(other, msg)
def add_service(self, service):
"""
Add a service to the peer
"""
serv_name = type(service).__name__
if isinstance(service, BaseHandler):
self.handlers[serv_name] = service
for m in service.messages:
if m not in self.mh_map:
self.mh_map[m] = set()
self.mh_map[m].add(serv_name)
if isinstance(service, BaseRunner):
self.runners[serv_name] = service
# Start all peer serice runners
def start_all_runners(self):
for runner in self.runners.values():
runner.start()
# Get storage used by the peer
def get_storage(self, storage_name):
return self.storage.get(storage_name)
# Store message in peer storage
def store(self, storage_name, msg_id, msg):
if storage_name not in self.storage:
if self.logger:
self.logger.error("No storage %s found", storage_name)
raise Exception("No storage {} found" % storage_name)
self.storage[storage_name].add(msg_id, msg)
# Add new storage for the peer
def add_storage(self, storage_name, storage):
self.storage[storage_name] = storage
class MainController(Behaviour):
def __init__(self, nname, agvcommchannel):
super().__init__(nname)
b = Blackboard()
self.rbs = RuleBasedSystem(b.get('[RBS]maxattempts'))
envy = b.get('enviro')
self.accuracyParameter = b.get('[vision]accuracy')
self.toagv = agvcommchannel
#robot
self.robotcomm = Store(envy)
self.robot = Robot('robot', self.robotcomm)
# stations
ninputs = b.get('[stations]inumber')
noutputs = b.get('[stations]onumber')
nspas = b.get('[stations]spa')
self.stations = dict()
b.put('[Shared]packages', dict())
b.put('[Shared]packages.dirtybit', dict())
for sindex in range(0, ninputs):
sname, stat = self.makeStation(sindex, envy)
self.stations[sname] = (stat, StationDir.INPUT)
for sindex in range(0, noutputs):
sname, stat = self.makeStation(sindex + ninputs, envy)
self.stations[sname] = (stat, StationDir.OUTPUT)
def makeStation(self, index, envy):
name = 'STAT_' + str(index)
cin = Store(envy)
cout = Store(envy)
v = Vision('VIS_' + str(index), self.accuracyParameter)
s = Station(name, cin, cout)
s.setVision(v)
Blackboard().get('[Shared]packages')[name] = None
Blackboard().get('[Shared]packages.dirtybit')[name] = True
return name, s
def boot(self):
for name in self.stations.keys():
self.load(name)
def load(self, sname):
(stat, dir) = self.stations[sname]
order = MovementOrder.INBD if (dir == StationDir.INPUT) else (
MovementOrder.OUTD)
msg = (order, stat.channelIN, stat.channelOUT)
self.toagv.put(msg)
def reload(self, sname):
(stat, dir) = self.stations[sname]
order = MovementOrder.DRPI if (dir == StationDir.INPUT) else (
MovementOrder.DRPO)
msg = (order, stat.channelIN, stat.channelOUT)
self.toagv.put(msg)
def checkFullEmptyPallets(self):
for name in self.stations:
stat, dir = self.stations[name]
if stat.operative():
p = stat.getPallet()
flag = (dir == StationDir.INPUT) and (p.isEmpty())
flagout = ((dir == StationDir.OUTPUT) and (p.isSatisfied()))
flag = flag or flagout
if flag == True:
stat.dirt()
if (flagout):
self.log('vault;' + str(p.lenght()) + ';', 2)
Recorder().add('hit', p.lenght())
self.reload(name)
def do(self):
if (self.onrun == True):
action = None
while (action == None):
action, resetcondition, stationtoreset = self.rbs.computeNextStep(
self.stations)
if (action == None):
if resetcondition == False:
yield self.env.timeout(10)
else:
stationtoreset.dirt()
p = stationtoreset.getPallet()
pname = stationtoreset.getName()
l = p.lenght()
self.log(
'MISS;' + stationtoreset.getName() + ',' + str(l) +
';', 2)
Recorder().add('miss', l)
self.reload(pname)
yield self.env.timeout(1)
self.robotcomm.put(action)
yield self.robotcomm.get()
self.checkFullEmptyPallets()
class Worker:
DOWNLOAD_PRIORITY_BOOST_FOR_READY_TASK = 100000
def __init__(self, cpus=1, max_downloads=4, max_downloads_per_worker=2):
self.cpus = cpus
self.assignments = {}
self.ready_store = None
self.data = set()
self.running_tasks = {}
self.scheduled_downloads = {}
self.running_downloads = []
self.free_cpus = cpus
self.max_downloads = max_downloads
self.max_downloads_per_worker = max_downloads_per_worker
self.id = None
def to_dict(self):
return {"id": self.id, "cpus": self.cpus}
def copy(self):
return Worker(cpus=self.cpus,
max_downloads=self.max_downloads,
max_downloads_per_worker=self.max_downloads_per_worker)
def try_retract_task(self, task):
if task in self.running_tasks:
logging.debug(
"Retracting task %s from worker %s cancelled because task is running",
task, self)
return False
logging.debug("Retracting task %s from worker %s", task, self)
a = self.assignments[task]
a.cancelled = True
for inp in task.inputs:
d = self.scheduled_downloads.get(inp)
if d is None:
continue
d.consumer_count -= 1
if d.consumer_count <= 0:
logging.debug("Cancelling download of %s", inp)
assert d.consumer_count == 0
if d.source is None: # is not running
assert d not in self.running_downloads
del self.scheduled_downloads[inp]
# This is necessary to cleanup cache
if not self.download_wakeup.triggered:
self.download_wakeup.succeed()
del self.assignments[a.task]
return True
def assign_tasks(self, assignments):
runtime_state = self.simulator.runtime_state
for assignment in assignments:
assert assignment.worker == self
# assert assignment not in self.assignments
if assignment.task.cpus > self.cpus:
raise Exception(
"Task {} allocated on worker with {} cpus".format(
assignment.task, self.cpus))
self.assignments[assignment.task] = assignment
need_inputs = 0
for inp in assignment.task.inputs:
if inp in self.data:
continue
if runtime_state.object_info(inp).placing:
self._schedule_download(
assignment, inp,
runtime_state.task_info(assignment.task).is_ready)
need_inputs += 1
assignment.remaining_inputs_count = need_inputs
if need_inputs == 0:
self.ready_store.put(assignment)
logger.info("Task %s scheduled on %s (%s ri)", assignment.task,
self, need_inputs)
def update_tasks(self, updates):
runtime_state = self.simulator.runtime_state
for task, obj in updates:
a = self.assignments[task]
if obj not in self.data:
if obj.size == 0:
self._add_data(obj)
else:
self._schedule_download(
a, obj,
runtime_state.task_info(task).is_ready)
@property
def assigned_tasks(self):
return iter(self.assignments)
def _add_data(self, obj):
if obj in self.data:
raise Exception("Object {} is already on worker {}".format(
obj, self))
self.data.add(obj)
for t in obj.consumers:
a = self.assignments.get(t)
if a is None:
continue
a.remaining_inputs_count -= 1
if a.remaining_inputs_count <= 0:
assert a.remaining_inputs_count == 0
self.ready_store.put(a)
def _schedule_download(self, assignment, obj, ready):
priority = assignment.priority
if ready:
priority += self.DOWNLOAD_PRIORITY_BOOST_FOR_READY_TASK
d = self.scheduled_downloads.get(obj)
if d is None:
logger.info("Worker %s: scheduled downloading %s, priority=%s",
self, obj, priority)
assert obj not in self.data
d = Download(obj, priority)
self.scheduled_downloads[obj] = d
else:
d.update_priority(priority)
d.consumer_count += 1
if not self.download_wakeup.triggered:
self.download_wakeup.succeed()
def _download_process(self):
events = [self.download_wakeup]
env = self.env
runtime_state = self.simulator.runtime_state
while True:
finished = yield env.any_of(events)
for event in finished.keys():
if event == events[0]:
self.download_wakeup = Event(self.simulator.env)
events[0] = self.download_wakeup
downloads = None
continue
events.remove(event)
download = event.value
self._add_data(download.output)
self.running_downloads.remove(download)
del self.scheduled_downloads[download.output]
self.simulator.fetch_finished(self, download.source,
download.output)
if len(self.running_downloads) < self.max_downloads:
# We need to sort any time, as it priority may changed in background
if downloads is None:
downloads = list(
o for o in self.scheduled_downloads.values()
if o not in self.running_downloads)
downloads.sort(key=lambda d: d.priority, reverse=True)
for d in downloads[:]:
count = 0
worker = runtime_state.object_info(d.output).placing[0]
for rd in self.running_downloads:
if worker == rd.source:
count += 1
if count >= self.max_downloads_per_worker:
continue
downloads.remove(d)
assert d.start_time is None
d.start_time = self.env.now
d.source = worker
self.running_downloads.append(d)
event = self.netmodel.download(worker, self, d.output.size,
d)
events.append(event)
self.simulator.add_trace_event(
FetchStartTraceEvent(self.env.now, self, worker,
d.output))
if len(self.running_downloads) >= self.max_downloads:
break
def run(self, env, simulator, netmodel):
self.env = env
self.simulator = simulator
self.netmodel = netmodel
self.ready_store = Store(env)
self.download_wakeup = Event(self.simulator.env)
self.free_cpus = self.cpus
env.process(self._download_process())
prepared_assignments = []
events = [self.ready_store.get()]
while True:
finished = yield env.any_of(events)
for event in finished.keys():
if event == events[0]:
events[0] = self.ready_store.get()
assignment = event.value
if assignment.cancelled:
continue
prepared_assignments.append(assignment)
prepared_assignments.sort(key=lambda a: a.priority,
reverse=True)
continue
assignment = event.value
task = assignment.task
self.free_cpus += task.cpus
assert not assignment.cancelled
del self.assignments[assignment.task]
events.remove(event)
del self.running_tasks[task]
simulator.add_trace_event(
TaskEndTraceEvent(self.env.now, self, task))
for output in task.outputs:
self._add_data(output)
simulator.on_task_finished(self, task)
block = float("-inf")
for assignment in prepared_assignments[:]:
if assignment.priority < block:
continue
task = assignment.task
if task.cpus <= self.free_cpus:
prepared_assignments.remove(assignment)
if assignment.cancelled:
continue
self.free_cpus -= task.cpus
self.running_tasks[task] = RunningTask(task, self.env.now)
simulator.add_trace_event(
TaskStartTraceEvent(self.env.now, self, task))
events.append(env.timeout(task.duration, assignment))
self.simulator.on_task_start(self, assignment.task)
else:
block = max(block, assignment.block)
def __repr__(self):
return "<Worker {}>".format(self.id)
class Worker:
def __init__(self):
self.assigned_tasks = []
self.ready_tasks = None
self.data = set()
self.downloads = {}
def _download(self, worker, task):
def _helper():
yield self.connector.download(worker, self, task.size)
self.simulator.add_trace_event(
FetchEndTraceEvent(self.env.now, self, worker, task))
del self.downloads[task]
self.data.add(task)
assert worker != self
self.simulator.add_trace_event(
FetchStartTraceEvent(self.env.now, self, worker, task))
p = self.env.process(_helper())
self.downloads[task] = p
return p
def assign_task(self, task):
self.assigned_tasks.append(task)
self._init_downloads(task)
def update_task(self, task):
assert task in self.assigned_tasks
self._init_downloads(task)
def _init_downloads(self, task):
deps = []
not_complete = False
for input in task.inputs:
if input in self.data:
continue
d = self.downloads.get(input)
if d is None:
if input.info.is_finished:
worker = input.info.assigned_workers[0]
d = self._download(worker, input)
else:
not_complete = True
deps.append(d)
def _helper():
yield self.env.all_of(deps)
self.ready_tasks.put(task)
if not_complete:
return
if not deps:
self.ready_tasks.put(task)
else:
self.env.process(_helper())
def run(self, env, simulator, connector):
self.env = env
self.simulator = simulator
self.connector = connector
self.ready_tasks = Store(env)
while True:
task = yield self.ready_tasks.get()
simulator.add_trace_event(
TaskStartTraceEvent(self.env.now, self, task))
yield env.timeout(task.duration)
simulator.add_trace_event(
TaskEndTraceEvent(self.env.now, self, task))
self.assigned_tasks.remove(task)
self.data.add(task)
simulator.on_task_finished(self, task)
def __repr__(self):
return "<Worker {}>".format(id(self))
class Component:
def __init__(self, env, cp_params):
self.env = env
self.in_pipe = Store(self.env) # TODO: define capacity in cp_params
self.pending_jobs = {}
self.response_times = []
self.interarrivals = []
self.queue_times = []
self.num_cores = cp_params['num_cores']
self.used_cores = 0
self.core_speed = random_number(cp_params['core_speed'])
self.cores = Resource(self.num_cores)
self.name = cp_params['name']
self.jobs_completed = 0
self.sim_components = None
self.data_res = []
self.idle_time = 0
self.time_last_arrival = None
def __str__(self):
return f'[{self.name}]'
def logger(self, message, job_id):
#print(f'{self} {message} {job_id} time_{self.env.now}')
pass
def run(self):
while True:
if len(self.in_pipe.items) < 1 or self.used_cores >= self.num_cores:
yield self.env.timeout(1)
self.idle_time += 1
continue
self.used_cores += 1
job = yield self.in_pipe.get()
job.stats[self.name]
enqueue_time = job.stats[self.name]['enqueue_time']
queue_time = self.env.now - enqueue_time
self.queue_times.append(queue_time)
self.env.process(self.process_job(job))
def get_stats(self):
stats = dict(
avg_response_time = 9999 if len(self.response_times) == 0 else mean(self.response_times),
avg_queue_time = 9999 if len(self.queue_times) == 0 else mean(self.queue_times),
avg_interarrival_time = 9999 if len(self.interarrivals) == 0 else mean(self.interarrivals),
queue_size=len(self.in_pipe.items),
idle_time = self.idle_time,
used_cores = self.used_cores,
jobs_completed = self.jobs_completed,
name = self.name,
sim_time = self.env.now
)
return stats
def init_job_stats(self, job):
job.stats[self.name] = {}
job.stats[self.name]['arrival_time'] = self.env.now
if self.time_last_arrival is None:
self.time_last_arrival = self.env.now
else:
interarrival = self.env.now - self.time_last_arrival
self.interarrivals.append(interarrival)
def receive_request(self, job):
self.init_job_stats(job)
self.logger("received", job.id)
yield self.env.process(self.enqueue_job(job))
def make_request(self, job, component):
job.response_method = self.receive_response
self.env.process(component.receive_request(job))
yield self.env.timeout(0) # NOTE: make request delay
self.pending_jobs[job.id] = job
def receive_response(self, response_job, type_of_response='response'):
self.logger(f'received_{type_of_response}_for', response_job.id)
if response_job.id in list(self.pending_jobs.keys()):
self.env.process(self.enqueue_job(self.pending_jobs[response_job.id]))
del self.pending_jobs[response_job.id]
yield self.env.timeout(0)
return "OK"
return "INVALID JOB"
def process_job(self,job):
self.logger('processing', job.id)
processing_time = ceil(job.size / self.core_speed)
yield self.env.timeout(processing_time)
self.used_cores -= 1
self.complete_job(job)
def complete_job(self,job):
job.stats[self.name]['finish_time'] = self.env.now
self.jobs_completed += 1
response_time = self.env.now - job.stats[self.name]['arrival_time']
self.response_times.append(response_time)
def enqueue_job(self, job):
job_stats = {}
try:
job_stats['arrival_time'] = self.env.now
job_stats['estimated_processing_time'] = job.size / self.core_speed
job_stats['enqueue_time'] = self.env.now
yield self.in_pipe.put(job)
job.stats[self.name] = job_stats
except Exception as e:
self.logger("error_at_enqueue", job.id)