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 __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
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