def mapping(self, sorted_jobs, existing_plan, live_nodes, commcost,
compcost):
"""def allocate(job, orders, jobson, prec, compcost, commcost):"""
""" Allocate job to the machine with earliest finish time
Operates in place
"""
## TODO: add finished tasks
jobson = dict()
for (node, items) in existing_plan.items():
for item in items:
if item.state == ScheduleItem.FINISHED or item.state == ScheduleItem.EXECUTING:
jobson[item.job] = node
new_plan = existing_plan
def ft(machine):
#cost = st(machine)
runtime = compcost(task, machine)
cost = st(machine, runtime) + runtime
##print("machine: %s job:%s cost: %s" % (machine.name, task.id, cost))
##print("machine: " + str(machine.name) + " cost: " + str(cost))
return cost
if len(live_nodes) != 0:
## in case if there is not any live nodes we just return the same cleaned schedule
for wf, tasks in sorted_jobs:
##wf_dag = self.convert_to_parent_children_map(wf)
wf_dag = HeftHelper.convert_to_parent_children_map(wf)
prec = reverse_dict(wf_dag)
for task in tasks:
st = partial(self.start_time, wf, task, new_plan, jobson,
prec, commcost)
# ress = [(key, ft(key)) for key in new_plan.keys()]
# agent_pair = min(ress, key=lambda x: x[1][0])
# agent = agent_pair[0]
# start = agent_pair[1][0]
# end = agent_pair[1][1]
# agent = min(new_plan.keys(), key=ft)
agent = min(live_nodes, key=ft)
runtime = compcost(task, agent)
start = st(agent, runtime)
end = ft(agent)
# new_plan[agent].append(ScheduleItem(task, start, end))
Schedule.insert_item(new_plan, agent,
ScheduleItem(task, start, end))
jobson[task] = agent
new_sched = Schedule(new_plan)
return new_sched
def build_schedule(workflow, estimator, resource_manager, solution):
"""
the solution consists all parts necessary to build whole solution
For the moment, it is mentioned that all species taking part in algorithm
are necessary to build complete solution
solution = {
s1.name: val1,
s2.name: val2,
....
}
"""
ms = solution[MAPPING_SPECIE]
os = solution[ORDERING_SPECIE]
assert check_precedence(workflow, os), "Precedence is violated"
ms = {t: resource_manager.byName(n) for t, n in ms}
schedule_mapping = {n: [] for n in set(ms.values())}
task_to_node = {}
for t in os:
node = ms[t]
t = workflow.byId(t)
(start_time,
end_time) = place_task_to_schedule(workflow, estimator,
schedule_mapping, task_to_node, ms,
t, node, 0)
task_to_node[t.id] = (node, start_time, end_time)
schedule = Schedule(schedule_mapping)
return schedule
def default_fixed_schedule_part(resource_manager):
fix_schedule_part = Schedule({
node: []
for node in HeftHelper.to_nodes(
resource_manager.get_resources())
})
return fix_schedule_part
def __init__(self,
workflow,
resource_manager,
estimator,
oct,
ranking=None):
self.current_schedule = Schedule(dict())
self.workflow = workflow
self.resource_manager = resource_manager
self.estimator = estimator
self.oct = oct
self.current_time = 0
nodes = self.get_nodes()
self.wf_jobs = self.make_ranking(self.workflow,
nodes) if ranking is None else ranking
# print("A: " + str(self.wf_jobs))
#TODO: remove it later
# to_print = ''
# for job in self.wf_jobs:
# to_print = to_print + str(job.id) + " "
# print(to_print)
pass
def schedule(self):
"""
create inter-priority
"""
def byPriority(wf):
return 0 if wf.priority is None else wf.priority
##simple inter priority sorting
sorted_wfs = sorted(self.workflows, key=byPriority)
wf_jobs = {wf: [] for wf in sorted_wfs}
resources = self.resource_manager.get_resources()
##print("common nodes count:" + str(len(toNodes(resources))))
nodes = HeftHelper.to_nodes(resources)
wf_jobs = {wf: self.make_ranking(wf, nodes) for wf in sorted_wfs}
##new_schedule = self.get_unchanged_schedule(self.old_schedule, time)
new_schedule = Schedule({node: [] for node in nodes})
new_plan = new_schedule.mapping
for (wf, jobs) in wf_jobs.items():
new_schedule = self.mapping([(wf, jobs)], new_plan, nodes,
self.commcost, self.compcost)
new_plan = new_schedule.mapping
return new_schedule
def run_heft(workflow, resource_manager, estimator):
"""
It simply runs src with empty initial schedule
and returns complete schedule
"""
heft = DynamicHeft(workflow, resource_manager, estimator)
nodes = resource_manager.get_nodes()
init_schedule = Schedule({node: [] for node in nodes})
return heft.run(init_schedule)
def run_peft(workflow, resource_manager, estimator):
"""
It simply runs peft with empty initial schedule
and returns complete schedule
"""
oct = PeftHelper.get_OCT(workflow, resource_manager, estimator)
peft = DynamicPeft(workflow, resource_manager, estimator, oct)
nodes = resource_manager.get_nodes()
init_schedule = Schedule({node: [] for node in nodes})
return peft.run(init_schedule)
def __init__(self, workflow, resource_manager, estimator, ranking=None):
self.current_schedule = Schedule(dict())
self.workflow = workflow
self.resource_manager = resource_manager
self.estimator = estimator
self.ranking = ranking
self.current_time = 0
nodes = self.get_nodes()
pass
def clean_unfinished(schedule):
def clean(items):
return [
item for item in items if item.state == ScheduleItem.FINISHED
or item.state == ScheduleItem.EXECUTING
]
new_mapping = {
node: clean(items)
for (node, items) in schedule.mapping.items()
}
return Schedule(new_mapping)
def _run_heft():
dynamic_planner = DynamicHeft(wf, resource_manager, estimator)
nodes = HeftHelper.to_nodes(resource_manager.resources)
current_cleaned_schedule = Schedule({node: [] for node in nodes})
schedule_dynamic_heft = dynamic_planner.run(
current_cleaned_schedule)
self._validate(wf, estimator, schedule_dynamic_heft)
if is_visualized:
viz.visualize_task_node_mapping(wf, schedule_dynamic_heft)
# Utility.create_jedule_visualization(schedule_dynamic_heft, wf_name+'_heft')
pass
return schedule_dynamic_heft
def __call__(self):
_wf = wf(self.wf_name)
rm = ExperimentResourceManager(rg.r([10, 15, 25, 30]))
estimator = ModelTimeEstimator(bandwidth=10)
empty_fixed_schedule_part = Schedule({node: [] for node in rm.get_nodes()})
heft_schedule = run_heft(_wf, rm, estimator)
fixed_schedule = empty_fixed_schedule_part
ga_functions = GAFunctions2(_wf, rm, estimator)
generate = partial(ga_generate, ga_functions=ga_functions,
fixed_schedule_part=fixed_schedule,
current_time=0.0, init_sched_percent=0.05,
initial_schedule=heft_schedule)
stats = tools.Statistics(lambda ind: ind.fitness.values[0])
stats.register("avg", numpy.mean)
stats.register("std", numpy.std)
stats.register("min", numpy.min)
stats.register("max", numpy.max)
logbook = tools.Logbook()
logbook.header = ["gen", "evals"] + stats.fields
toolbox = Toolbox()
toolbox.register("generate", generate)
toolbox.register("evaluate", fit_converter(ga_functions.build_fitness(empty_fixed_schedule_part, 0.0)))
toolbox.register("clone", deepcopy)
toolbox.register("mate", ga_functions.crossover)
toolbox.register("sweep_mutation", ga_functions.sweep_mutation)
toolbox.register("mutate", ga_functions.mutation)
# toolbox.register("select_parents", )
# toolbox.register("select", tools.selTournament, tournsize=4)
toolbox.register("select", tools.selRoulette)
pop, logbook, best = run_ga(toolbox=toolbox,
logbook=logbook,
stats=stats,
**self.GA_PARAMS)
resulted_schedule = ga_functions.build_schedule(best, empty_fixed_schedule_part, 0.0)
ga_makespan = Utility.makespan(resulted_schedule)
return (ga_makespan, logbook)
def generate(wf, rm, estimator, schedule=None, fixed_schedule_part=None, current_time=0.0):
sched = schedule if schedule is not None else SimpleRandomizedHeuristic(wf, rm.get_nodes(), estimator).schedule(fixed_schedule_part, current_time)
if fixed_schedule_part is not None:
un_tasks = unmoveable_tasks(fixed_schedule_part)
clean_sched = Schedule({node: [item for item in items if item.job.id not in un_tasks and item.state != ScheduleItem.FAILED]
for node, items in sched.mapping.items()})
else:
clean_sched = sched
mapping, ordering = ord_and_map(clean_sched)
ordering_numseq = ordering_to_numseq(ordering)
ordering_map = {task_id: val for task_id, val in zip(ordering, ordering_numseq)}
ord_p, map_p = OrderingParticle(ordering_map), MappingParticle(mapping)
ord_p.velocity = OrderingParticle.Velocity({})
map_p.velocity = MappingParticle.Velocity({})
result = CompoundParticle(map_p, ord_p)
if schedule is None and not validate_mapping_with_alive_nodes(result.mapping.entity, rm):
raise Exception("found invalid solution in generated array")
return result
def schedule(self, fixed_schedule_part=None, current_time=0.0):
estimate = self.estimator.estimate_transfer_time
# TODO: make common utility function with ScheduleBuilder
def is_last_version_of_task_executing(item):
return item.state == ScheduleItem.EXECUTING or item.state == ScheduleItem.FINISHED or item.state == ScheduleItem.UNSTARTED
def _get_ready_tasks(children, finished_tasks):
def _is_child_ready(child):
ids = set([p.id for p in child.parents])
result = False in [id in finished_tasks for id in ids]
return not result
ready_children = [child for child in children if _is_child_ready(child)]
return ready_children
if fixed_schedule_part is None:
schedule_mapping = {node: [] for node in self.nodes}
ready_tasks = [child.id for child in self.workflow.head_task.children]
task_to_node = dict()
finished_tasks = set()
else:
schedule_mapping = {node: [item for item in items] for (node, items) in fixed_schedule_part.mapping.items()}
finished_tasks = [item.job.id for (node, items) in fixed_schedule_part.mapping.items() for item in items if is_last_version_of_task_executing(item)]
finished_tasks = set([self.workflow.head_task.id] + finished_tasks)
unfinished = [task for task in self.workflow.get_all_unique_tasks() if not task.id in finished_tasks]
ready_tasks = [task.id for task in _get_ready_tasks(unfinished, finished_tasks)]
task_to_node = {item.job.id: (node, item.start_time, item.end_time) for (node, items) in fixed_schedule_part.mapping.items() for item in items if is_last_version_of_task_executing(item)}
def is_child_ready(child):
ids = set([p.id for p in child.parents])
result = False in [id in finished_tasks for id in ids]
return not result
def find_slots(node, comm_ready, runtime):
node_schedule = schedule_mapping.get(node, list())
free_time = 0 if len(node_schedule) == 0 else node_schedule[-1].end_time
## TODO: refactor it later
f_time = max(free_time, comm_ready)
f_time = max(f_time, current_time)
base_variant = [(f_time, f_time + runtime + 1)]
zero_interval = [] if len(node_schedule) == 0 else [(0, node_schedule[0].start_time)]
middle_intervals = [(node_schedule[i].end_time, node_schedule[i + 1].start_time) for i in range(len(node_schedule) - 1)]
intervals = zero_interval + middle_intervals + base_variant
#result = [(st, end) for (st, end) in intervals if st >= comm_ready and end - st >= runtime]
## TODO: rethink rounding
result = [(st, end) for (st, end) in intervals if (current_time < st or abs((current_time - st)) < 0.01) and st >= comm_ready and (runtime < (end - st) or abs((end - st) - runtime) < 0.01)]
return result
def comm_ready_func(task, node):
##TODO: remake this stub later.
if len(task.parents) == 1 and self.workflow.head_task.id == list(task.parents)[0].id:
return 0
return max([task_to_node[p.id][2] + estimate(node, task_to_node[p.id][0], task, p) for p in task.parents])
def get_possible_execution_times(task, node):
## pay attention to the last element in the resulted seq
## it represents all available time of node after it completes all its work
## (if such interval can exist)
## time_slots = [(st1, end1),(st2, end2,...,(st_last, st_last + runtime)]
runtime = self.estimator.estimate_runtime(task, node)
comm_ready = comm_ready_func(task, node)
time_slots = find_slots(node, comm_ready, runtime)
return time_slots, runtime
while len(ready_tasks) > 0:
choosed_index = random.randint(0, len(ready_tasks) - 1)
task = self.task_map[ready_tasks[choosed_index]]
#TODO: make checking for all nodes are dead.(It's a very rare situation so it is not consider for now)
alive_nodes = [node for node in self.nodes if node.state != Node.Down]
choosed_node_index = random.randint(0, len(alive_nodes) - 1)
node = alive_nodes[choosed_node_index]
time_slots, runtime = get_possible_execution_times(task, node)
choosed_time_index = 0 if len(time_slots) == 1 else random.randint(0, len(time_slots) - 1)
time_slot = time_slots[choosed_time_index]
start_time = time_slot[0]
end_time = start_time + runtime
item = ScheduleItem(task, start_time, end_time)
##schedule_mapping[node].append(item)
Schedule.insert_item(schedule_mapping, node, item)
task_to_node[task.id] = (node, start_time, end_time)
##print('I am here')
ready_tasks.remove(task.id)
finished_tasks.add(task.id)
ready_children = [child for child in task.children if is_child_ready(child)]
for child in ready_children:
ready_tasks.append(child.id)
schedule = Schedule(schedule_mapping)
return schedule
def __call__(self, chromo, current_time):
count_of_tasks = lambda mapping: reduce(operator.add, (
len(tasks) for node, tasks in mapping.items()), 0)
alive_nodes = [node for node in self.nodes if node.state != Node.Down]
alive_nodes_names = [node.name for node in alive_nodes]
for node_name, tasks in chromo.items():
if node_name not in alive_nodes_names and len(tasks) > 0:
raise ValueError(
"Chromo is invalid. There is a task assigned to a dead node"
)
if count_of_tasks(chromo) + len(
self.fixed_schedule_part.get_unfailed_tasks_ids()) != len(
self.workflow.get_all_unique_tasks()):
print("==Chromosome==================================")
print(chromo)
print("=fixed_schedule_part===================================")
print(self.fixed_schedule_part)
raise Exception(
"The chromosome not a full. Chromo length: {0}, Fixed part length: {1}, workflow size: {2}"
.format(count_of_tasks(chromo),
len(self.fixed_schedule_part.get_unfailed_tasks_ids()),
len(self.workflow.get_all_unique_tasks())))
# TODO: add not to schedule
#if count_of_tasks(chromo) + count_of_tasks(self.fixed_schedule_part.mapping) !=
(schedule_mapping, finished_tasks, ready_tasks, chrmo_mapping,
task_to_node) = self._create_helping_structures(chromo)
# print("SCHEDULE_MAPPING")
# print("AlIVE_NODES", alive_nodes)
# pprint(schedule_mapping)
#chromo_copy = {nd_name: [item for item in items] for (nd_name, items) in chromo.items()}
chromo_copy = deepcopy(chromo)
if len(alive_nodes) == 0:
raise Exception("There are not alive nodes")
#print("Building started...")
while len(ready_tasks) > 0:
# ## TODO: only for debug. Remove it later.
# print("alive nodes: {0}".format(alive_nodes))
# for node_name, tasks in chromo_copy.items():
# print("Node: {0}, tasks count: {1}".format(node_name, len(tasks)))
count_before = count_of_tasks(chromo_copy)
if len(alive_nodes) == 0:
raise ValueError("Count of alive_nodes is zero")
for node in alive_nodes:
if len(chromo_copy[node.name]) == 0:
continue
## TODO: Urgent! completely rethink this procedure
tsk_id = None
for i in range(len(chromo_copy[node.name])):
if chromo_copy[node.name][i] in ready_tasks:
tsk_id = chromo_copy[node.name][i]
break
if tsk_id is not None:
task = self.task_map[tsk_id]
#del chromo_copy[node.name][0]
chromo_copy[node.name].remove(tsk_id)
ready_tasks.remove(tsk_id)
(start_time, end_time) = place_task_to_schedule(
self.workflow, self.estimator, schedule_mapping,
task_to_node, chrmo_mapping, task, node, current_time)
task_to_node[task.id] = (node, start_time, end_time)
finished_tasks.add(task.id)
ready_children = self._get_ready_tasks(
task.children, finished_tasks)
for child in ready_children:
ready_tasks.append(child.id)
count_after = count_of_tasks(chromo_copy)
if count_before == count_after:
raise Exception(
"Unable to properly process a chromosome."
" Perhaps, due to invalid fixed_schedule_part or the chromosome."
)
pass
schedule = Schedule(schedule_mapping)
return schedule
def __call__(self, chromo, current_time):
(schedule_mapping, finished_tasks, ready_tasks, chrmo_mapping, task_to_node) = self._create_helping_structures(chromo)
chromo_copy = dict()
for (nd_name, items) in chromo.items():
chromo_copy[nd_name] = []
for item in items:
chromo_copy[nd_name].append(item)
alive_nodes = [node for node in self.nodes if node.state != Node.Down]
if len(alive_nodes) == 0:
raise Exception("There are not alive nodes")
while len(ready_tasks) > 0:
for node in alive_nodes:
if len(chromo_copy[node.name]) == 0:
continue
if node.state == Node.Down:
continue
## TODO: Urgent! completely rethink this procedure
tsk_id = None
for i in range(len(chromo_copy[node.name])):
if chromo_copy[node.name][i] in ready_tasks:
tsk_id = chromo_copy[node.name][i]
break
if tsk_id is not None:
task = self.task_map[tsk_id]
#del chromo_copy[node.name][0]
chromo_copy[node.name].remove(tsk_id)
ready_tasks.remove(tsk_id)
time_slots, runtime = self._get_possible_execution_times(
schedule_mapping,
task_to_node,
chrmo_mapping,
task,
node,
current_time)
time_slot = next(time_slots)
start_time = time_slot[0]
end_time = start_time + runtime
item = ScheduleItem(task, start_time, end_time)
# need to account current time
Schedule.insert_item(schedule_mapping, node, item)
task_to_node[task.id] = (node, start_time, end_time)
finished_tasks.add(task.id)
#ready_children = [child for child in task.children if self._is_child_ready(finished_tasks, child)]
ready_children = self._get_ready_tasks(task.children, finished_tasks)
for child in ready_children:
ready_tasks.append(child.id)
schedule = Schedule(schedule_mapping)
return schedule