1
|
|
2
|
|
3
|
|
4
|
|
5
|
|
6
|
|
7
|
|
8
|
|
9
|
|
10
|
|
11
|
|
12
|
|
13
|
|
14
|
|
15
|
|
16
|
|
17
|
|
18
|
|
19
|
"""This module provides the main classes for Haizea's VM Scheduler. All the
|
20
|
scheduling code that decides when and where a lease is scheduled is contained
|
21
|
in the VMScheduler class (except for the code that specifically decides
|
22
|
what physical machines each virtual machine is mapped to, which is factored out
|
23
|
into the "mapper" module). This module also provides the classes for the
|
24
|
reservations that will be placed in the slot table and correspond to VMs.
|
25
|
"""
|
26
|
|
27
|
import haizea.common.constants as constants
|
28
|
from haizea.common.utils import round_datetime_delta, round_datetime, estimate_transfer_time, pretty_nodemap, get_config, get_clock, get_policy
|
29
|
from haizea.core.leases import Lease, Capacity
|
30
|
from haizea.core.scheduler.slottable import ResourceReservation, ResourceTuple
|
31
|
from haizea.core.scheduler import ReservationEventHandler, RescheduleLeaseException, NormalEndLeaseException, EnactmentError, NotSchedulableException, InconsistentScheduleError, InconsistentLeaseStateError, MigrationResourceReservation
|
32
|
from operator import attrgetter, itemgetter
|
33
|
from mx.DateTime import TimeDelta
|
34
|
|
35
|
import logging
|
36
|
|
37
|
|
38
|
class VMScheduler(object):
|
39
|
"""The Haizea VM Scheduler
|
40
|
|
41
|
This class is responsible for taking a lease and scheduling VMs to satisfy
|
42
|
the requirements of that lease.
|
43
|
"""
|
44
|
|
45
|
def __init__(self, slottable, resourcepool, mapper):
|
46
|
"""Constructor
|
47
|
|
48
|
The constructor does little more than create the VM scheduler's
|
49
|
attributes. However, it does expect (in the arguments) a fully-constructed
|
50
|
SlotTable, ResourcePool, and Mapper (these are constructed in the
|
51
|
Manager's constructor).
|
52
|
|
53
|
Arguments:
|
54
|
slottable -- Slot table
|
55
|
resourcepool -- Resource pool where enactment commands will be sent to
|
56
|
mapper -- Mapper
|
57
|
"""
|
58
|
self.slottable = slottable
|
59
|
self.resourcepool = resourcepool
|
60
|
self.mapper = mapper
|
61
|
self.logger = logging.getLogger("VMSCHED")
|
62
|
|
63
|
|
64
|
|
65
|
self.handlers = {}
|
66
|
self.handlers[VMResourceReservation] = ReservationEventHandler(
|
67
|
sched = self,
|
68
|
on_start = VMScheduler._handle_start_vm,
|
69
|
on_end = VMScheduler._handle_end_vm)
|
70
|
|
71
|
self.handlers[ShutdownResourceReservation] = ReservationEventHandler(
|
72
|
sched = self,
|
73
|
on_start = VMScheduler._handle_start_shutdown,
|
74
|
on_end = VMScheduler._handle_end_shutdown)
|
75
|
|
76
|
self.handlers[SuspensionResourceReservation] = ReservationEventHandler(
|
77
|
sched = self,
|
78
|
on_start = VMScheduler._handle_start_suspend,
|
79
|
on_end = VMScheduler._handle_end_suspend)
|
80
|
|
81
|
self.handlers[ResumptionResourceReservation] = ReservationEventHandler(
|
82
|
sched = self,
|
83
|
on_start = VMScheduler._handle_start_resume,
|
84
|
on_end = VMScheduler._handle_end_resume)
|
85
|
|
86
|
self.handlers[MemImageMigrationResourceReservation] = ReservationEventHandler(
|
87
|
sched = self,
|
88
|
on_start = VMScheduler._handle_start_migrate,
|
89
|
on_end = VMScheduler._handle_end_migrate)
|
90
|
|
91
|
|
92
|
|
93
|
backfilling = get_config().get("backfilling")
|
94
|
if backfilling == constants.BACKFILLING_OFF:
|
95
|
self.max_in_future = 0
|
96
|
elif backfilling == constants.BACKFILLING_AGGRESSIVE:
|
97
|
self.max_in_future = 1
|
98
|
elif backfilling == constants.BACKFILLING_CONSERVATIVE:
|
99
|
self.max_in_future = -1
|
100
|
elif backfilling == constants.BACKFILLING_INTERMEDIATE:
|
101
|
self.max_in_future = get_config().get("backfilling-reservations")
|
102
|
self.future_leases = set()
|
103
|
|
104
|
|
105
|
def schedule(self, lease, nexttime, earliest):
|
106
|
""" The scheduling function
|
107
|
|
108
|
This particular function doesn't do much except call __schedule_asap
|
109
|
and __schedule_exact (which do all the work).
|
110
|
|
111
|
Arguments:
|
112
|
lease -- Lease to schedule
|
113
|
nexttime -- The next time at which the scheduler can allocate resources.
|
114
|
earliest -- The earliest possible starting times on each physical node
|
115
|
"""
|
116
|
if lease.get_type() == Lease.BEST_EFFORT:
|
117
|
return self.__schedule_asap(lease, nexttime, earliest, allow_in_future = self.can_schedule_in_future())
|
118
|
elif lease.get_type() == Lease.ADVANCE_RESERVATION:
|
119
|
return self.__schedule_exact(lease, nexttime, earliest)
|
120
|
elif lease.get_type() == Lease.IMMEDIATE:
|
121
|
return self.__schedule_asap(lease, nexttime, earliest, allow_in_future = False)
|
122
|
|
123
|
|
124
|
def estimate_migration_time(self, lease):
|
125
|
""" Estimates the time required to migrate a lease's VMs
|
126
|
|
127
|
This function conservatively estimates that all the VMs are going to
|
128
|
be migrated to other nodes. Since all the transfers are intra-node,
|
129
|
the bottleneck is the transfer from whatever node has the most
|
130
|
memory to transfer.
|
131
|
|
132
|
Note that this method only estimates the time to migrate the memory
|
133
|
state files for the VMs. Migrating the software environment (which may
|
134
|
or may not be a disk image) is the responsibility of the preparation
|
135
|
scheduler, which has it's own set of migration scheduling methods.
|
136
|
|
137
|
Arguments:
|
138
|
lease -- Lease that might be migrated
|
139
|
"""
|
140
|
migration = get_config().get("migration")
|
141
|
if migration == constants.MIGRATE_YES:
|
142
|
vmrr = lease.get_last_vmrr()
|
143
|
mem_in_pnode = dict([(pnode,0) for pnode in set(vmrr.nodes.values())])
|
144
|
for (vnode,pnode) in vmrr.nodes.items():
|
145
|
mem = vmrr.resources_in_pnode[pnode].get_by_type(constants.RES_MEM)
|
146
|
mem_in_pnode[pnode] += mem
|
147
|
max_mem_to_transfer = max(mem_in_pnode.values())
|
148
|
bandwidth = self.resourcepool.info.get_migration_bandwidth()
|
149
|
return estimate_transfer_time(max_mem_to_transfer, bandwidth)
|
150
|
elif migration == constants.MIGRATE_YES_NOTRANSFER:
|
151
|
return TimeDelta(seconds=0)
|
152
|
|
153
|
def schedule_migration(self, lease, vmrr, nexttime):
|
154
|
""" Schedules migrations for a lease
|
155
|
|
156
|
Arguments:
|
157
|
lease -- Lease being migrated
|
158
|
vmrr -- The VM reservation before which the migration will take place
|
159
|
nexttime -- The next time at which the scheduler can allocate resources.
|
160
|
"""
|
161
|
|
162
|
|
163
|
|
164
|
last_vmrr = lease.get_last_vmrr()
|
165
|
vnode_migrations = dict([(vnode, (last_vmrr.nodes[vnode], vmrr.nodes[vnode])) for vnode in vmrr.nodes])
|
166
|
|
167
|
|
168
|
mustmigrate = False
|
169
|
for vnode in vnode_migrations:
|
170
|
if vnode_migrations[vnode][0] != vnode_migrations[vnode][1]:
|
171
|
mustmigrate = True
|
172
|
break
|
173
|
|
174
|
if not mustmigrate:
|
175
|
return []
|
176
|
|
177
|
|
178
|
|
179
|
if get_config().get("migration") == constants.MIGRATE_YES_NOTRANSFER:
|
180
|
start = nexttime
|
181
|
end = nexttime
|
182
|
res = {}
|
183
|
migr_rr = MemImageMigrationResourceReservation(lease, start, end, res, vmrr, vnode_migrations)
|
184
|
migr_rr.state = ResourceReservation.STATE_SCHEDULED
|
185
|
return [migr_rr]
|
186
|
|
187
|
|
188
|
migrations = []
|
189
|
while len(vnode_migrations) > 0:
|
190
|
pnodes = set()
|
191
|
migration = {}
|
192
|
for vnode in vnode_migrations:
|
193
|
origin = vnode_migrations[vnode][0]
|
194
|
dest = vnode_migrations[vnode][1]
|
195
|
if not origin in pnodes and not dest in pnodes:
|
196
|
migration[vnode] = vnode_migrations[vnode]
|
197
|
pnodes.add(origin)
|
198
|
pnodes.add(dest)
|
199
|
for vnode in migration:
|
200
|
del vnode_migrations[vnode]
|
201
|
migrations.append(migration)
|
202
|
|
203
|
|
204
|
start = max(last_vmrr.post_rrs[-1].end, nexttime)
|
205
|
bandwidth = self.resourcepool.info.get_migration_bandwidth()
|
206
|
migr_rrs = []
|
207
|
for m in migrations:
|
208
|
vnodes_to_migrate = m.keys()
|
209
|
max_mem_to_migrate = max([lease.requested_resources[vnode].get_quantity(constants.RES_MEM) for vnode in vnodes_to_migrate])
|
210
|
migr_time = estimate_transfer_time(max_mem_to_migrate, bandwidth)
|
211
|
end = start + migr_time
|
212
|
res = {}
|
213
|
for (origin,dest) in m.values():
|
214
|
resorigin = Capacity([constants.RES_NETOUT])
|
215
|
resorigin.set_quantity(constants.RES_NETOUT, bandwidth)
|
216
|
resdest = Capacity([constants.RES_NETIN])
|
217
|
resdest.set_quantity(constants.RES_NETIN, bandwidth)
|
218
|
res[origin] = self.slottable.create_resource_tuple_from_capacity(resorigin)
|
219
|
res[dest] = self.slottable.create_resource_tuple_from_capacity(resdest)
|
220
|
migr_rr = MemImageMigrationResourceReservation(lease, start, start + migr_time, res, vmrr, m)
|
221
|
migr_rr.state = ResourceReservation.STATE_SCHEDULED
|
222
|
migr_rrs.append(migr_rr)
|
223
|
start = end
|
224
|
|
225
|
return migr_rrs
|
226
|
|
227
|
def cancel_vm(self, vmrr):
|
228
|
""" Cancels a VM resource reservation
|
229
|
|
230
|
Arguments:
|
231
|
vmrr -- VM RR to be cancelled
|
232
|
"""
|
233
|
|
234
|
|
235
|
|
236
|
if vmrr.lease in self.future_leases:
|
237
|
self.future_leases.remove(vmrr.lease)
|
238
|
|
239
|
|
240
|
for rr in vmrr.pre_rrs:
|
241
|
if rr.state == ResourceReservation.STATE_SCHEDULED:
|
242
|
self.slottable.remove_reservation(rr)
|
243
|
|
244
|
|
245
|
for rr in vmrr.post_rrs:
|
246
|
self.slottable.remove_reservation(rr)
|
247
|
|
248
|
|
249
|
self.slottable.remove_reservation(vmrr)
|
250
|
|
251
|
|
252
|
def can_suspend_at(self, lease, t):
|
253
|
""" Determines if it is possible to suspend a lease before a given time
|
254
|
|
255
|
Arguments:
|
256
|
vmrr -- VM RR to be preempted
|
257
|
t -- Time by which the VM must be preempted
|
258
|
"""
|
259
|
|
260
|
|
261
|
|
262
|
vmrr = lease.get_last_vmrr()
|
263
|
time_until_suspend = t - vmrr.start
|
264
|
min_duration = self.__compute_scheduling_threshold(lease)
|
265
|
can_suspend = time_until_suspend >= min_duration
|
266
|
return can_suspend
|
267
|
|
268
|
|
269
|
def preempt_vm(self, vmrr, t):
|
270
|
""" Preempts a VM reservation at a given time
|
271
|
|
272
|
This method assumes that the lease is, in fact, preemptable,
|
273
|
that the VMs are running at the given time, and that there is
|
274
|
enough time to suspend the VMs before the given time (all these
|
275
|
checks are done in the lease scheduler).
|
276
|
|
277
|
Arguments:
|
278
|
vmrr -- VM RR to be preempted
|
279
|
t -- Time by which the VM must be preempted
|
280
|
"""
|
281
|
|
282
|
|
283
|
old_start = vmrr.start
|
284
|
old_end = vmrr.end
|
285
|
|
286
|
|
287
|
self.__schedule_suspension(vmrr, t)
|
288
|
|
289
|
|
290
|
self.slottable.update_reservation_with_key_change(vmrr, old_start, old_end)
|
291
|
|
292
|
|
293
|
for susprr in vmrr.post_rrs:
|
294
|
self.slottable.add_reservation(susprr)
|
295
|
|
296
|
|
297
|
def get_future_reschedulable_leases(self):
|
298
|
""" Returns a list of future leases that are reschedulable.
|
299
|
|
300
|
Currently, this list is just the best-effort leases scheduled
|
301
|
in the future as determined by the backfilling algorithm.
|
302
|
Advance reservation leases, by their nature, cannot be
|
303
|
rescheduled to find a "better" starting time.
|
304
|
"""
|
305
|
return list(self.future_leases)
|
306
|
|
307
|
|
308
|
def can_schedule_in_future(self):
|
309
|
""" Returns True if the backfilling algorithm would allow a lease
|
310
|
to be scheduled in the future.
|
311
|
|
312
|
"""
|
313
|
if self.max_in_future == -1:
|
314
|
return True
|
315
|
else:
|
316
|
return len(self.future_leases) < self.max_in_future
|
317
|
|
318
|
|
319
|
def get_utilization(self, time):
|
320
|
""" Computes resource utilization (currently just CPU-based)
|
321
|
|
322
|
Arguments:
|
323
|
time -- Time at which to determine utilization
|
324
|
"""
|
325
|
total = self.slottable.get_total_capacity(restype = constants.RES_CPU)
|
326
|
util = {}
|
327
|
reservations = self.slottable.get_reservations_at(time)
|
328
|
for r in reservations:
|
329
|
for node in r.resources_in_pnode:
|
330
|
if isinstance(r, VMResourceReservation):
|
331
|
use = r.resources_in_pnode[node].get_by_type(constants.RES_CPU)
|
332
|
util[type(r)] = use + util.setdefault(type(r),0.0)
|
333
|
elif isinstance(r, SuspensionResourceReservation) or isinstance(r, ResumptionResourceReservation) or isinstance(r, ShutdownResourceReservation):
|
334
|
use = r.vmrr.resources_in_pnode[node].get_by_type(constants.RES_CPU)
|
335
|
util[type(r)] = use + util.setdefault(type(r),0.0)
|
336
|
util[None] = total - sum(util.values())
|
337
|
for k in util:
|
338
|
util[k] /= total
|
339
|
|
340
|
return util
|
341
|
|
342
|
|
343
|
def __schedule_exact(self, lease, nexttime, earliest):
|
344
|
""" Schedules VMs that must start at an exact time
|
345
|
|
346
|
This type of lease is "easy" to schedule because we know the exact
|
347
|
start time, which means that's the only starting time we have to
|
348
|
check. So, this method does little more than call the mapper.
|
349
|
|
350
|
Arguments:
|
351
|
lease -- Lease to schedule
|
352
|
nexttime -- The next time at which the scheduler can allocate resources.
|
353
|
earliest -- The earliest possible starting times on each physical node
|
354
|
"""
|
355
|
|
356
|
|
357
|
start = lease.start.requested
|
358
|
end = start + lease.duration.requested
|
359
|
|
360
|
|
361
|
|
362
|
requested_resources = dict([(k,self.slottable.create_resource_tuple_from_capacity(v)) for k,v in lease.requested_resources.items()])
|
363
|
|
364
|
|
365
|
mapping, actualend, preemptions = self.mapper.map(lease,
|
366
|
requested_resources,
|
367
|
start,
|
368
|
end,
|
369
|
strictend = True)
|
370
|
|
371
|
|
372
|
if mapping == None:
|
373
|
raise NotSchedulableException, "Not enough resources in specified interval"
|
374
|
|
375
|
|
376
|
res = {}
|
377
|
|
378
|
for (vnode,pnode) in mapping.items():
|
379
|
vnode_res = requested_resources[vnode]
|
380
|
if res.has_key(pnode):
|
381
|
res[pnode].incr(vnode_res)
|
382
|
else:
|
383
|
res[pnode] = ResourceTuple.copy(vnode_res)
|
384
|
|
385
|
vmrr = VMResourceReservation(lease, start, end, mapping, res)
|
386
|
vmrr.state = ResourceReservation.STATE_SCHEDULED
|
387
|
|
388
|
|
389
|
self.__schedule_shutdown(vmrr)
|
390
|
|
391
|
return vmrr, preemptions
|
392
|
|
393
|
|
394
|
def __schedule_asap(self, lease, nexttime, earliest, allow_in_future = None):
|
395
|
""" Schedules VMs as soon as possible
|
396
|
|
397
|
This method is a bit more complex that __schedule_exact because
|
398
|
we need to figure out what "as soon as possible" actually is.
|
399
|
This involves attempting several mappings, at different points
|
400
|
in time, before we can schedule the lease.
|
401
|
|
402
|
This method will always check, at least, if the lease can be scheduled
|
403
|
at the earliest possible moment at which the lease could be prepared
|
404
|
(e.g., if the lease can't start until 1 hour in the future because that's
|
405
|
the earliest possible time at which the disk images it requires can
|
406
|
be transferred, then that's when the scheduler will check). Note, however,
|
407
|
that this "earliest possible moment" is determined by the preparation
|
408
|
scheduler.
|
409
|
|
410
|
Additionally, if the lease can't be scheduled at the earliest
|
411
|
possible moment, it can also check if the lease can be scheduled
|
412
|
in the future. This partially implements a backfilling algorithm
|
413
|
(the maximum number of future leases is stored in the max_in_future
|
414
|
attribute of VMScheduler), the other part being implemented in the
|
415
|
__process_queue method of LeaseScheduler.
|
416
|
|
417
|
Note that, if the method is allowed to scheduled in the future,
|
418
|
and assuming that the lease doesn't request more resources than
|
419
|
the site itself, this method will always schedule the VMs succesfully
|
420
|
(since there's always an empty spot somewhere in the future).
|
421
|
|
422
|
|
423
|
Arguments:
|
424
|
lease -- Lease to schedule
|
425
|
nexttime -- The next time at which the scheduler can allocate resources.
|
426
|
earliest -- The earliest possible starting times on each physical node
|
427
|
allow_in_future -- Boolean indicating whether the scheduler is
|
428
|
allowed to schedule the VMs in the future.
|
429
|
"""
|
430
|
|
431
|
|
432
|
|
433
|
|
434
|
|
435
|
|
436
|
|
437
|
lease_id = lease.id
|
438
|
remaining_duration = lease.duration.get_remaining_duration()
|
439
|
shutdown_time = self.__estimate_shutdown_time(lease)
|
440
|
|
441
|
|
442
|
|
443
|
|
444
|
mustresume = (lease.get_state() in (Lease.STATE_SUSPENDED_PENDING, Lease.STATE_SUSPENDED_QUEUED, Lease.STATE_SUSPENDED_SCHEDULED))
|
445
|
|
446
|
|
447
|
|
448
|
min_duration = self.__compute_scheduling_threshold(lease)
|
449
|
|
450
|
|
451
|
|
452
|
|
453
|
|
454
|
|
455
|
|
456
|
|
457
|
|
458
|
|
459
|
|
460
|
|
461
|
|
462
|
|
463
|
|
464
|
|
465
|
|
466
|
if not mustresume:
|
467
|
|
468
|
|
469
|
cps = [(node, e.time) for node, e in earliest.items()]
|
470
|
cps.sort(key=itemgetter(1))
|
471
|
curcp = None
|
472
|
changepoints = []
|
473
|
nodes = []
|
474
|
for node, time in cps:
|
475
|
nodes.append(node)
|
476
|
if time != curcp:
|
477
|
changepoints.append([time, set(nodes)])
|
478
|
curcp = time
|
479
|
else:
|
480
|
changepoints[-1][1] = set(nodes)
|
481
|
else:
|
482
|
|
483
|
|
484
|
|
485
|
if get_config().get("migration") == constants.MIGRATE_NO:
|
486
|
vmrr = lease.get_last_vmrr()
|
487
|
onlynodes = set(vmrr.nodes.values())
|
488
|
else:
|
489
|
onlynodes = None
|
490
|
changepoints = list(set([x.time for x in earliest.values()]))
|
491
|
changepoints.sort()
|
492
|
changepoints = [(x, onlynodes) for x in changepoints]
|
493
|
|
494
|
|
495
|
|
496
|
|
497
|
if allow_in_future:
|
498
|
res = self.slottable.get_reservations_ending_after(changepoints[-1][0])
|
499
|
|
500
|
|
501
|
futurecp = [r.get_final_end() for r in res if isinstance(r, VMResourceReservation)]
|
502
|
|
503
|
|
504
|
futurecp += [r.end for r in res if isinstance(r, ShutdownResourceReservation) and not r.vmrr in res]
|
505
|
if not mustresume:
|
506
|
futurecp = [(p,None) for p in futurecp]
|
507
|
else:
|
508
|
futurecp = [(p,onlynodes) for p in futurecp]
|
509
|
else:
|
510
|
futurecp = []
|
511
|
|
512
|
|
513
|
|
514
|
|
515
|
|
516
|
|
517
|
|
518
|
|
519
|
|
520
|
|
521
|
|
522
|
|
523
|
if mustresume:
|
524
|
duration = remaining_duration + self.__estimate_resume_time(lease)
|
525
|
else:
|
526
|
duration = remaining_duration
|
527
|
|
528
|
duration += shutdown_time
|
529
|
|
530
|
in_future = False
|
531
|
|
532
|
|
533
|
|
534
|
requested_resources = dict([(k,self.slottable.create_resource_tuple_from_capacity(v)) for k,v in lease.requested_resources.items()])
|
535
|
|
536
|
|
537
|
start, end, mapping, preemptions = self.__find_fit_at_points(lease,
|
538
|
requested_resources,
|
539
|
changepoints,
|
540
|
duration,
|
541
|
min_duration)
|
542
|
|
543
|
if start == None and not allow_in_future:
|
544
|
|
545
|
|
546
|
raise NotSchedulableException, "Could not find enough resources for this request"
|
547
|
|
548
|
|
549
|
|
550
|
if start == None and allow_in_future:
|
551
|
start, end, mapping, preemptions = self.__find_fit_at_points(lease,
|
552
|
requested_resources,
|
553
|
futurecp,
|
554
|
duration,
|
555
|
min_duration
|
556
|
)
|
557
|
|
558
|
|
559
|
|
560
|
if start == None:
|
561
|
raise InconsistentScheduleError, "Could not find a mapping in the future (this should not happen)"
|
562
|
|
563
|
in_future = True
|
564
|
|
565
|
|
566
|
|
567
|
|
568
|
|
569
|
|
570
|
|
571
|
|
572
|
|
573
|
|
574
|
res = {}
|
575
|
|
576
|
for (vnode,pnode) in mapping.items():
|
577
|
vnode_res = requested_resources[vnode]
|
578
|
if res.has_key(pnode):
|
579
|
res[pnode].incr(vnode_res)
|
580
|
else:
|
581
|
res[pnode] = ResourceTuple.copy(vnode_res)
|
582
|
|
583
|
vmrr = VMResourceReservation(lease, start, end, mapping, res)
|
584
|
vmrr.state = ResourceReservation.STATE_SCHEDULED
|
585
|
|
586
|
|
587
|
if mustresume:
|
588
|
self.__schedule_resumption(vmrr, start)
|
589
|
|
590
|
|
591
|
|
592
|
mustsuspend = (vmrr.end - vmrr.start) < remaining_duration
|
593
|
if mustsuspend:
|
594
|
self.__schedule_suspension(vmrr, end)
|
595
|
else:
|
596
|
|
597
|
if (vmrr.end - vmrr.start) > remaining_duration + shutdown_time:
|
598
|
vmrr.end = vmrr.start + remaining_duration + shutdown_time
|
599
|
self.__schedule_shutdown(vmrr)
|
600
|
|
601
|
if in_future:
|
602
|
self.future_leases.add(lease)
|
603
|
|
604
|
susp_str = res_str = ""
|
605
|
if mustresume:
|
606
|
res_str = " (resuming)"
|
607
|
if mustsuspend:
|
608
|
susp_str = " (suspending)"
|
609
|
self.logger.info("Lease #%i has been scheduled on nodes %s from %s%s to %s%s" % (lease.id, mapping.values(), start, res_str, end, susp_str))
|
610
|
|
611
|
return vmrr, preemptions
|
612
|
|
613
|
|
614
|
def __find_fit_at_points(self, lease, requested_resources, changepoints, duration, min_duration):
|
615
|
""" Tries to map a lease in a given list of points in time
|
616
|
|
617
|
This method goes through a given list of points in time and tries
|
618
|
to find the earliest time at which that lease can be allocated
|
619
|
resources.
|
620
|
|
621
|
Arguments:
|
622
|
lease -- Lease to schedule
|
623
|
requested_resources -- A dictionary of lease node -> ResourceTuple.
|
624
|
changepoints -- The list of changepoints
|
625
|
duration -- The amount of time requested
|
626
|
min_duration -- The minimum amount of time that should be allocated
|
627
|
|
628
|
Returns:
|
629
|
start -- The time at which resources have been found for the lease
|
630
|
actualend -- The time at which the resources won't be available. Note
|
631
|
that this is not necessarily (start + duration) since the mapper
|
632
|
might be unable to find enough resources for the full requested duration.
|
633
|
mapping -- A mapping of lease nodes to physical nodes
|
634
|
preemptions -- A list of
|
635
|
(if no mapping is found, all these values are set to None)
|
636
|
"""
|
637
|
found = False
|
638
|
|
639
|
for time, onlynodes in changepoints:
|
640
|
start = time
|
641
|
end = start + duration
|
642
|
self.logger.debug("Attempting to map from %s to %s" % (start, end))
|
643
|
|
644
|
|
645
|
|
646
|
susptype = get_config().get("suspension")
|
647
|
if susptype == constants.SUSPENSION_NONE or (lease.numnodes > 1 and susptype == constants.SUSPENSION_SERIAL):
|
648
|
strictend = True
|
649
|
else:
|
650
|
strictend = False
|
651
|
|
652
|
|
653
|
mapping, actualend, preemptions = self.mapper.map(lease,
|
654
|
requested_resources,
|
655
|
start,
|
656
|
end,
|
657
|
strictend = strictend,
|
658
|
onlynodes = onlynodes)
|
659
|
|
660
|
|
661
|
|
662
|
if mapping != None:
|
663
|
if actualend < end:
|
664
|
actualduration = actualend - start
|
665
|
if actualduration >= min_duration:
|
666
|
self.logger.debug("This lease can be scheduled from %s to %s (will require suspension)" % (start, actualend))
|
667
|
found = True
|
668
|
break
|
669
|
else:
|
670
|
self.logger.debug("This starting time does not allow for the requested minimum duration (%s < %s)" % (actualduration, min_duration))
|
671
|
else:
|
672
|
self.logger.debug("This lease can be scheduled from %s to %s (full duration)" % (start, end))
|
673
|
found = True
|
674
|
break
|
675
|
|
676
|
if found:
|
677
|
return start, actualend, mapping, preemptions
|
678
|
else:
|
679
|
return None, None, None, None
|
680
|
|
681
|
|
682
|
def __compute_susprem_times(self, vmrr, time, direction, exclusion, rate, override = None):
|
683
|
""" Computes the times at which suspend/resume operations would have to start
|
684
|
|
685
|
When suspending or resuming a VM, the VM's memory is dumped to a
|
686
|
file on disk. To correctly estimate the time required to suspend
|
687
|
a lease with multiple VMs, Haizea makes sure that no two
|
688
|
suspensions/resumptions happen at the same time (e.g., if eight
|
689
|
memory files were being saved at the same time to disk, the disk's
|
690
|
performance would be reduced in a way that is not as easy to estimate
|
691
|
as if only one file were being saved at a time). Based on a number
|
692
|
of parameters, this method estimates the times at which the
|
693
|
suspend/resume commands would have to be sent to guarantee this
|
694
|
exclusion.
|
695
|
|
696
|
Arguments:
|
697
|
vmrr -- The VM reservation that will be suspended/resumed
|
698
|
time -- The time at which the suspend should end or the resume should start.
|
699
|
direction -- DIRECTION_BACKWARD: start at "time" and compute the times going
|
700
|
backward (for suspensions) DIRECTION_FORWARD: start at time "time" and compute
|
701
|
the times going forward.
|
702
|
exclusion -- SUSPRES_EXCLUSION_GLOBAL (memory is saved to global filesystem)
|
703
|
or SUSPRES_EXCLUSION_LOCAL (saved to local filesystem)
|
704
|
rate -- The rate at which an individual VM is suspended/resumed
|
705
|
override -- If specified, then instead of computing the time to
|
706
|
suspend/resume VM based on its memory and the "rate" parameter,
|
707
|
use this override value.
|
708
|
|
709
|
"""
|
710
|
times = []
|
711
|
enactment_overhead = get_config().get("enactment-overhead")
|
712
|
|
713
|
if exclusion == constants.SUSPRES_EXCLUSION_GLOBAL:
|
714
|
|
715
|
|
716
|
|
717
|
|
718
|
t = time
|
719
|
t_prev = None
|
720
|
|
721
|
for (vnode,pnode) in vmrr.nodes.items():
|
722
|
if override == None:
|
723
|
mem = vmrr.lease.requested_resources.get_by_type(constants.RES_MEM)
|
724
|
op_time = self.__compute_suspend_resume_time(mem, rate)
|
725
|
else:
|
726
|
op_time = override
|
727
|
|
728
|
op_time += enactment_overhead
|
729
|
|
730
|
t_prev = t
|
731
|
|
732
|
if direction == constants.DIRECTION_FORWARD:
|
733
|
t += op_time
|
734
|
times.append((t_prev, t, {pnode:[vnode]}))
|
735
|
elif direction == constants.DIRECTION_BACKWARD:
|
736
|
t -= op_time
|
737
|
times.append((t, t_prev, {pnode:[vnode]}))
|
738
|
|
739
|
elif exclusion == constants.SUSPRES_EXCLUSION_LOCAL:
|
740
|
|
741
|
|
742
|
|
743
|
pervnode_times = []
|
744
|
vnodes_in_pnode = {}
|
745
|
for (vnode,pnode) in vmrr.nodes.items():
|
746
|
vnodes_in_pnode.setdefault(pnode, []).append(vnode)
|
747
|
for pnode in vnodes_in_pnode:
|
748
|
t = time
|
749
|
t_prev = None
|
750
|
for vnode in vnodes_in_pnode[pnode]:
|
751
|
if override == None:
|
752
|
mem = vmrr.lease.requested_resources[vnode].get_quantity(constants.RES_MEM)
|
753
|
op_time = self.__compute_suspend_resume_time(mem, rate)
|
754
|
else:
|
755
|
op_time = override
|
756
|
|
757
|
t_prev = t
|
758
|
|
759
|
if direction == constants.DIRECTION_FORWARD:
|
760
|
t += op_time
|
761
|
pervnode_times.append((t_prev, t, vnode))
|
762
|
elif direction == constants.DIRECTION_BACKWARD:
|
763
|
t -= op_time
|
764
|
pervnode_times.append((t, t_prev, vnode))
|
765
|
|
766
|
|
767
|
uniq_times = set([(start, end) for (start, end, vnode) in pervnode_times])
|
768
|
for (start, end) in uniq_times:
|
769
|
vnodes = [x[2] for x in pervnode_times if x[0] == start and x[1] == end]
|
770
|
node_mappings = {}
|
771
|
for vnode in vnodes:
|
772
|
pnode = vmrr.nodes[vnode]
|
773
|
node_mappings.setdefault(pnode, []).append(vnode)
|
774
|
times.append([start,end,node_mappings])
|
775
|
|
776
|
|
777
|
for t in times:
|
778
|
num_vnodes = sum([len(vnodes) for vnodes in t[2].values()])
|
779
|
overhead = TimeDelta(seconds = num_vnodes * enactment_overhead)
|
780
|
if direction == constants.DIRECTION_FORWARD:
|
781
|
t[1] += overhead
|
782
|
elif direction == constants.DIRECTION_BACKWARD:
|
783
|
t[0] -= overhead
|
784
|
|
785
|
|
786
|
if direction == constants.DIRECTION_FORWARD:
|
787
|
times.sort(key=itemgetter(0))
|
788
|
elif direction == constants.DIRECTION_BACKWARD:
|
789
|
times.sort(key=itemgetter(1))
|
790
|
times.reverse()
|
791
|
|
792
|
prev_start = None
|
793
|
prev_end = None
|
794
|
for t in times:
|
795
|
if prev_start != None:
|
796
|
start = t[0]
|
797
|
end = t[1]
|
798
|
if direction == constants.DIRECTION_FORWARD:
|
799
|
if start < prev_end:
|
800
|
diff = prev_end - start
|
801
|
t[0] += diff
|
802
|
t[1] += diff
|
803
|
elif direction == constants.DIRECTION_BACKWARD:
|
804
|
if end > prev_start:
|
805
|
diff = end - prev_start
|
806
|
t[0] -= diff
|
807
|
t[1] -= diff
|
808
|
prev_start = t[0]
|
809
|
prev_end = t[1]
|
810
|
|
811
|
return times
|
812
|
|
813
|
|
814
|
def __schedule_shutdown(self, vmrr):
|
815
|
""" Schedules the shutdown of a VM reservation
|
816
|
|
817
|
Arguments:
|
818
|
vmrr -- The VM reservation that will be shutdown
|
819
|
|
820
|
"""
|
821
|
config = get_config()
|
822
|
shutdown_time = self.__estimate_shutdown_time(vmrr.lease)
|
823
|
|
824
|
start = vmrr.end - shutdown_time
|
825
|
end = vmrr.end
|
826
|
|
827
|
shutdown_rr = ShutdownResourceReservation(vmrr.lease, start, end, vmrr.resources_in_pnode, vmrr.nodes, vmrr)
|
828
|
shutdown_rr.state = ResourceReservation.STATE_SCHEDULED
|
829
|
|
830
|
vmrr.update_end(start)
|
831
|
|
832
|
|
833
|
for rr in vmrr.post_rrs:
|
834
|
self.slottable.remove_reservation(rr)
|
835
|
vmrr.post_rrs = []
|
836
|
|
837
|
vmrr.post_rrs.append(shutdown_rr)
|
838
|
|
839
|
|
840
|
def __schedule_suspension(self, vmrr, suspend_by):
|
841
|
""" Schedules the suspension of a VM reservation
|
842
|
|
843
|
Most of the work is done in __compute_susprem_times. See that
|
844
|
method's documentation for more details.
|
845
|
|
846
|
Arguments:
|
847
|
vmrr -- The VM reservation that will be suspended
|
848
|
suspend_by -- The time by which the VMs should be suspended.
|
849
|
|
850
|
"""
|
851
|
config = get_config()
|
852
|
susp_exclusion = config.get("suspendresume-exclusion")
|
853
|
override = get_config().get("override-suspend-time")
|
854
|
rate = config.get("suspend-rate")
|
855
|
|
856
|
if suspend_by < vmrr.start or suspend_by > vmrr.end:
|
857
|
raise InconsistentScheduleError, "Tried to schedule a suspension by %s, which is outside the VMRR's duration (%s-%s)" % (suspend_by, vmrr.start, vmrr.end)
|
858
|
|
859
|
|
860
|
times = self.__compute_susprem_times(vmrr, suspend_by, constants.DIRECTION_BACKWARD, susp_exclusion, rate, override)
|
861
|
|
862
|
|
863
|
suspend_rrs = []
|
864
|
for (start, end, node_mappings) in times:
|
865
|
suspres = {}
|
866
|
all_vnodes = []
|
867
|
for (pnode,vnodes) in node_mappings.items():
|
868
|
num_vnodes = len(vnodes)
|
869
|
r = Capacity([constants.RES_MEM,constants.RES_DISK])
|
870
|
mem = 0
|
871
|
for vnode in vnodes:
|
872
|
mem += vmrr.lease.requested_resources[vnode].get_quantity(constants.RES_MEM)
|
873
|
r.set_quantity(constants.RES_MEM, mem * num_vnodes)
|
874
|
r.set_quantity(constants.RES_DISK, mem * num_vnodes)
|
875
|
suspres[pnode] = self.slottable.create_resource_tuple_from_capacity(r)
|
876
|
all_vnodes += vnodes
|
877
|
|
878
|
susprr = SuspensionResourceReservation(vmrr.lease, start, end, suspres, all_vnodes, vmrr)
|
879
|
susprr.state = ResourceReservation.STATE_SCHEDULED
|
880
|
suspend_rrs.append(susprr)
|
881
|
|
882
|
suspend_rrs.sort(key=attrgetter("start"))
|
883
|
|
884
|
susp_start = suspend_rrs[0].start
|
885
|
if susp_start < vmrr.start:
|
886
|
raise InconsistentScheduleError, "Determined suspension should start at %s, before the VMRR's start (%s) -- Suspend time not being properly estimated?" % (susp_start, vmrr.start)
|
887
|
|
888
|
vmrr.update_end(susp_start)
|
889
|
|
890
|
|
891
|
for rr in vmrr.post_rrs:
|
892
|
self.slottable.remove_reservation(rr)
|
893
|
vmrr.post_rrs = []
|
894
|
|
895
|
|
896
|
for susprr in suspend_rrs:
|
897
|
vmrr.post_rrs.append(susprr)
|
898
|
|
899
|
|
900
|
def __schedule_resumption(self, vmrr, resume_at):
|
901
|
""" Schedules the resumption of a VM reservation
|
902
|
|
903
|
Most of the work is done in __compute_susprem_times. See that
|
904
|
method's documentation for more details.
|
905
|
|
906
|
Arguments:
|
907
|
vmrr -- The VM reservation that will be resumed
|
908
|
resume_at -- The time at which the resumption should start
|
909
|
|
910
|
"""
|
911
|
config = get_config()
|
912
|
resm_exclusion = config.get("suspendresume-exclusion")
|
913
|
override = get_config().get("override-resume-time")
|
914
|
rate = config.get("resume-rate")
|
915
|
|
916
|
if resume_at < vmrr.start or resume_at > vmrr.end:
|
917
|
raise InconsistentScheduleError, "Tried to schedule a resumption at %s, which is outside the VMRR's duration (%s-%s)" % (resume_at, vmrr.start, vmrr.end)
|
918
|
|
919
|
|
920
|
times = self.__compute_susprem_times(vmrr, resume_at, constants.DIRECTION_FORWARD, resm_exclusion, rate, override)
|
921
|
|
922
|
|
923
|
resume_rrs = []
|
924
|
for (start, end, node_mappings) in times:
|
925
|
resmres = {}
|
926
|
all_vnodes = []
|
927
|
for (pnode,vnodes) in node_mappings.items():
|
928
|
num_vnodes = len(vnodes)
|
929
|
r = Capacity([constants.RES_MEM,constants.RES_DISK])
|
930
|
mem = 0
|
931
|
for vnode in vnodes:
|
932
|
mem += vmrr.lease.requested_resources[vnode].get_quantity(constants.RES_MEM)
|
933
|
r.set_quantity(constants.RES_MEM, mem * num_vnodes)
|
934
|
r.set_quantity(constants.RES_DISK, mem * num_vnodes)
|
935
|
resmres[pnode] = self.slottable.create_resource_tuple_from_capacity(r)
|
936
|
all_vnodes += vnodes
|
937
|
resmrr = ResumptionResourceReservation(vmrr.lease, start, end, resmres, all_vnodes, vmrr)
|
938
|
resmrr.state = ResourceReservation.STATE_SCHEDULED
|
939
|
resume_rrs.append(resmrr)
|
940
|
|
941
|
resume_rrs.sort(key=attrgetter("start"))
|
942
|
|
943
|
resm_end = resume_rrs[-1].end
|
944
|
if resm_end > vmrr.end:
|
945
|
raise InconsistentScheduleError, "Determined resumption would end at %s, after the VMRR's end (%s) -- Resume time not being properly estimated?" % (resm_end, vmrr.end)
|
946
|
|
947
|
vmrr.update_start(resm_end)
|
948
|
|
949
|
|
950
|
for resmrr in resume_rrs:
|
951
|
vmrr.pre_rrs.append(resmrr)
|
952
|
|
953
|
|
954
|
def __compute_suspend_resume_time(self, mem, rate):
|
955
|
""" Compute the time to suspend/resume a single VM
|
956
|
|
957
|
Arguments:
|
958
|
mem -- Amount of memory used by the VM
|
959
|
rate -- The rate at which an individual VM is suspended/resumed
|
960
|
|
961
|
"""
|
962
|
time = float(mem) / rate
|
963
|
time = round_datetime_delta(TimeDelta(seconds = time))
|
964
|
return time
|
965
|
|
966
|
|
967
|
def __estimate_suspend_time(self, lease):
|
968
|
""" Estimate the time to suspend an entire lease
|
969
|
|
970
|
Most of the work is done in __estimate_suspend_resume_time. See
|
971
|
that method's documentation for more details.
|
972
|
|
973
|
Arguments:
|
974
|
lease -- Lease that is going to be suspended
|
975
|
|
976
|
"""
|
977
|
rate = get_config().get("suspend-rate")
|
978
|
override = get_config().get("override-suspend-time")
|
979
|
if override != None:
|
980
|
return override
|
981
|
else:
|
982
|
return self.__estimate_suspend_resume_time(lease, rate)
|
983
|
|
984
|
|
985
|
def __estimate_resume_time(self, lease):
|
986
|
""" Estimate the time to resume an entire lease
|
987
|
|
988
|
Most of the work is done in __estimate_suspend_resume_time. See
|
989
|
that method's documentation for more details.
|
990
|
|
991
|
Arguments:
|
992
|
lease -- Lease that is going to be resumed
|
993
|
|
994
|
"""
|
995
|
rate = get_config().get("resume-rate")
|
996
|
override = get_config().get("override-resume-time")
|
997
|
if override != None:
|
998
|
return override
|
999
|
else:
|
1000
|
return self.__estimate_suspend_resume_time(lease, rate)
|
1001
|
|
1002
|
|
1003
|
def __estimate_suspend_resume_time(self, lease, rate):
|
1004
|
""" Estimate the time to suspend/resume an entire lease
|
1005
|
|
1006
|
Note that, unlike __compute_suspend_resume_time, this estimates
|
1007
|
the time to suspend/resume an entire lease (which may involve
|
1008
|
suspending several VMs)
|
1009
|
|
1010
|
Arguments:
|
1011
|
lease -- Lease that is going to be suspended/resumed
|
1012
|
rate -- The rate at which an individual VM is suspended/resumed
|
1013
|
|
1014
|
"""
|
1015
|
susp_exclusion = get_config().get("suspendresume-exclusion")
|
1016
|
enactment_overhead = get_config().get("enactment-overhead")
|
1017
|
mem = 0
|
1018
|
for vnode in lease.requested_resources:
|
1019
|
mem += lease.requested_resources[vnode].get_quantity(constants.RES_MEM)
|
1020
|
if susp_exclusion == constants.SUSPRES_EXCLUSION_GLOBAL:
|
1021
|
return lease.numnodes * (self.__compute_suspend_resume_time(mem, rate) + enactment_overhead)
|
1022
|
elif susp_exclusion == constants.SUSPRES_EXCLUSION_LOCAL:
|
1023
|
|
1024
|
return lease.numnodes * (self.__compute_suspend_resume_time(mem, rate) + enactment_overhead)
|
1025
|
|
1026
|
|
1027
|
def __estimate_shutdown_time(self, lease):
|
1028
|
""" Estimate the time to shutdown an entire lease
|
1029
|
|
1030
|
Arguments:
|
1031
|
lease -- Lease that is going to be shutdown
|
1032
|
|
1033
|
"""
|
1034
|
enactment_overhead = get_config().get("enactment-overhead").seconds
|
1035
|
return get_config().get("shutdown-time") + (enactment_overhead * lease.numnodes)
|
1036
|
|
1037
|
|
1038
|
def __compute_scheduling_threshold(self, lease):
|
1039
|
""" Compute the scheduling threshold (the 'minimum duration') of a lease
|
1040
|
|
1041
|
To avoid thrashing, Haizea will not schedule a lease unless all overheads
|
1042
|
can be correctly scheduled (which includes image transfers, suspensions, etc.).
|
1043
|
However, this can still result in situations where a lease is prepared,
|
1044
|
and then immediately suspended because of a blocking lease in the future.
|
1045
|
The scheduling threshold is used to specify that a lease must
|
1046
|
not be scheduled unless it is guaranteed to run for a minimum amount of
|
1047
|
time (the rationale behind this is that you ideally don't want leases
|
1048
|
to be scheduled if they're not going to be active for at least as much time
|
1049
|
as was spent in overheads).
|
1050
|
|
1051
|
An important part of computing this value is the "scheduling threshold factor".
|
1052
|
The default value is 1, meaning that the lease will be active for at least
|
1053
|
as much time T as was spent on overheads (e.g., if preparing the lease requires
|
1054
|
60 seconds, and we know that it will have to be suspended, requiring 30 seconds,
|
1055
|
Haizea won't schedule the lease unless it can run for at least 90 minutes).
|
1056
|
In other words, a scheduling factor of F required a minimum duration of
|
1057
|
F*T. A value of 0 could lead to thrashing, since Haizea could end up with
|
1058
|
situations where a lease starts and immediately gets suspended.
|
1059
|
|
1060
|
Arguments:
|
1061
|
lease -- Lease for which we want to find the scheduling threshold
|
1062
|
"""
|
1063
|
|
1064
|
config = get_config()
|
1065
|
threshold = config.get("force-scheduling-threshold")
|
1066
|
if threshold != None:
|
1067
|
|
1068
|
return threshold
|
1069
|
else:
|
1070
|
factor = config.get("scheduling-threshold-factor")
|
1071
|
|
1072
|
|
1073
|
|
1074
|
|
1075
|
susp_overhead = self.__estimate_suspend_time(lease)
|
1076
|
safe_duration = susp_overhead
|
1077
|
|
1078
|
if lease.get_state() == Lease.STATE_SUSPENDED_QUEUED:
|
1079
|
resm_overhead = self.__estimate_resume_time(lease)
|
1080
|
safe_duration += resm_overhead
|
1081
|
|
1082
|
|
1083
|
min_duration = safe_duration
|
1084
|
|
1085
|
|
1086
|
|
1087
|
|
1088
|
|
1089
|
|
1090
|
|
1091
|
|
1092
|
|
1093
|
|
1094
|
threshold = safe_duration + (min_duration * factor)
|
1095
|
return threshold
|
1096
|
|
1097
|
|
1098
|
|
1099
|
|
1100
|
|
1101
|
|
1102
|
|
1103
|
|
1104
|
def _handle_start_vm(self, l, rr):
|
1105
|
""" Handles the start of a VMResourceReservation
|
1106
|
|
1107
|
Arguments:
|
1108
|
l -- Lease the VMResourceReservation belongs to
|
1109
|
rr -- THe VMResourceReservation
|
1110
|
"""
|
1111
|
self.logger.debug("LEASE-%i Start of handleStartVM" % l.id)
|
1112
|
l.print_contents()
|
1113
|
lease_state = l.get_state()
|
1114
|
if lease_state == Lease.STATE_READY:
|
1115
|
l.set_state(Lease.STATE_ACTIVE)
|
1116
|
rr.state = ResourceReservation.STATE_ACTIVE
|
1117
|
now_time = get_clock().get_time()
|
1118
|
l.start.actual = now_time
|
1119
|
|
1120
|
try:
|
1121
|
self.resourcepool.start_vms(l, rr)
|
1122
|
except EnactmentError, exc:
|
1123
|
self.logger.error("Enactment error when starting VMs.")
|
1124
|
|
1125
|
|
1126
|
|
1127
|
|
1128
|
raise
|
1129
|
|
1130
|
elif lease_state == Lease.STATE_RESUMED_READY:
|
1131
|
l.set_state(Lease.STATE_ACTIVE)
|
1132
|
rr.state = ResourceReservation.STATE_ACTIVE
|
1133
|
|
1134
|
|
1135
|
else:
|
1136
|
raise InconsistentLeaseStateError(l, doing = "starting a VM")
|
1137
|
|
1138
|
|
1139
|
|
1140
|
if rr.lease in self.future_leases:
|
1141
|
self.future_leases.remove(l)
|
1142
|
|
1143
|
l.print_contents()
|
1144
|
self.logger.debug("LEASE-%i End of handleStartVM" % l.id)
|
1145
|
self.logger.info("Started VMs for lease %i on nodes %s" % (l.id, rr.nodes.values()))
|
1146
|
|
1147
|
|
1148
|
def _handle_end_vm(self, l, rr):
|
1149
|
""" Handles the end of a VMResourceReservation
|
1150
|
|
1151
|
Arguments:
|
1152
|
l -- Lease the VMResourceReservation belongs to
|
1153
|
rr -- THe VMResourceReservation
|
1154
|
"""
|
1155
|
self.logger.debug("LEASE-%i Start of handleEndVM" % l.id)
|
1156
|
self.logger.vdebug("LEASE-%i Before:" % l.id)
|
1157
|
l.print_contents()
|
1158
|
now_time = round_datetime(get_clock().get_time())
|
1159
|
diff = now_time - rr.start
|
1160
|
l.duration.accumulate_duration(diff)
|
1161
|
rr.state = ResourceReservation.STATE_DONE
|
1162
|
|
1163
|
self.logger.vdebug("LEASE-%i After:" % l.id)
|
1164
|
l.print_contents()
|
1165
|
self.logger.debug("LEASE-%i End of handleEndVM" % l.id)
|
1166
|
self.logger.info("Stopped VMs for lease %i on nodes %s" % (l.id, rr.nodes.values()))
|
1167
|
|
1168
|
|
1169
|
def _handle_unscheduled_end_vm(self, l, vmrr):
|
1170
|
""" Handles the unexpected end of a VMResourceReservation
|
1171
|
|
1172
|
Arguments:
|
1173
|
l -- Lease the VMResourceReservation belongs to
|
1174
|
rr -- THe VMResourceReservation
|
1175
|
"""
|
1176
|
|
1177
|
self.logger.info("LEASE-%i The VM has ended prematurely." % l.id)
|
1178
|
for rr in vmrr.post_rrs:
|
1179
|
self.slottable.remove_reservation(rr)
|
1180
|
vmrr.post_rrs = []
|
1181
|
vmrr.end = get_clock().get_time()
|
1182
|
self._handle_end_vm(l, vmrr)
|
1183
|
|
1184
|
|
1185
|
def _handle_start_suspend(self, l, rr):
|
1186
|
""" Handles the start of a SuspensionResourceReservation
|
1187
|
|
1188
|
Arguments:
|
1189
|
l -- Lease the SuspensionResourceReservation belongs to
|
1190
|
rr -- The SuspensionResourceReservation
|
1191
|
|
1192
|
"""
|
1193
|
self.logger.debug("LEASE-%i Start of handleStartSuspend" % l.id)
|
1194
|
l.print_contents()
|
1195
|
rr.state = ResourceReservation.STATE_ACTIVE
|
1196
|
|
1197
|
try:
|
1198
|
self.resourcepool.suspend_vms(l, rr)
|
1199
|
except EnactmentError, exc:
|
1200
|
self.logger.error("Enactment error when suspending VMs.")
|
1201
|
|
1202
|
|
1203
|
|
1204
|
|
1205
|
raise
|
1206
|
|
1207
|
if rr.is_first():
|
1208
|
l.set_state(Lease.STATE_SUSPENDING)
|
1209
|
l.print_contents()
|
1210
|
self.logger.info("Suspending lease %i..." % (l.id))
|
1211
|
self.logger.debug("LEASE-%i End of handleStartSuspend" % l.id)
|
1212
|
|
1213
|
|
1214
|
def _handle_end_suspend(self, l, rr):
|
1215
|
""" Handles the end of a SuspensionResourceReservation
|
1216
|
|
1217
|
Arguments:
|
1218
|
l -- Lease the SuspensionResourceReservation belongs to
|
1219
|
rr -- The SuspensionResourceReservation
|
1220
|
"""
|
1221
|
self.logger.debug("LEASE-%i Start of handleEndSuspend" % l.id)
|
1222
|
l.print_contents()
|
1223
|
|
1224
|
self.resourcepool.verify_suspend(l, rr)
|
1225
|
rr.state = ResourceReservation.STATE_DONE
|
1226
|
if rr.is_last():
|
1227
|
l.set_state(Lease.STATE_SUSPENDED_PENDING)
|
1228
|
l.print_contents()
|
1229
|
self.logger.debug("LEASE-%i End of handleEndSuspend" % l.id)
|
1230
|
self.logger.info("Lease %i suspended." % (l.id))
|
1231
|
|
1232
|
if l.get_state() == Lease.STATE_SUSPENDED_PENDING:
|
1233
|
raise RescheduleLeaseException
|
1234
|
|
1235
|
|
1236
|
def _handle_start_resume(self, l, rr):
|
1237
|
""" Handles the start of a ResumptionResourceReservation
|
1238
|
|
1239
|
Arguments:
|
1240
|
l -- Lease the ResumptionResourceReservation belongs to
|
1241
|
rr -- The ResumptionResourceReservation
|
1242
|
|
1243
|
"""
|
1244
|
self.logger.debug("LEASE-%i Start of handleStartResume" % l.id)
|
1245
|
l.print_contents()
|
1246
|
|
1247
|
try:
|
1248
|
self.resourcepool.resume_vms(l, rr)
|
1249
|
except EnactmentError, exc:
|
1250
|
self.logger.error("Enactment error when resuming VMs.")
|
1251
|
|
1252
|
|
1253
|
|
1254
|
|
1255
|
raise
|
1256
|
|
1257
|
rr.state = ResourceReservation.STATE_ACTIVE
|
1258
|
if rr.is_first():
|
1259
|
l.set_state(Lease.STATE_RESUMING)
|
1260
|
l.print_contents()
|
1261
|
self.logger.info("Resuming lease %i..." % (l.id))
|
1262
|
self.logger.debug("LEASE-%i End of handleStartResume" % l.id)
|
1263
|
|
1264
|
|
1265
|
def _handle_end_resume(self, l, rr):
|
1266
|
""" Handles the end of a ResumptionResourceReservation
|
1267
|
|
1268
|
Arguments:
|
1269
|
l -- Lease the ResumptionResourceReservation belongs to
|
1270
|
rr -- The ResumptionResourceReservation
|
1271
|
|
1272
|
"""
|
1273
|
self.logger.debug("LEASE-%i Start of handleEndResume" % l.id)
|
1274
|
l.print_contents()
|
1275
|
|
1276
|
self.resourcepool.verify_resume(l, rr)
|
1277
|
rr.state = ResourceReservation.STATE_DONE
|
1278
|
if rr.is_last():
|
1279
|
l.set_state(Lease.STATE_RESUMED_READY)
|
1280
|
self.logger.info("Resumed lease %i" % (l.id))
|
1281
|
for vnode, pnode in rr.vmrr.nodes.items():
|
1282
|
self.resourcepool.remove_ramfile(pnode, l.id, vnode)
|
1283
|
l.print_contents()
|
1284
|
self.logger.debug("LEASE-%i End of handleEndResume" % l.id)
|
1285
|
|
1286
|
|
1287
|
def _handle_start_shutdown(self, l, rr):
|
1288
|
""" Handles the start of a ShutdownResourceReservation
|
1289
|
|
1290
|
Arguments:
|
1291
|
l -- Lease the SuspensionResourceReservation belongs to
|
1292
|
rr -- The SuspensionResourceReservation
|
1293
|
"""
|
1294
|
|
1295
|
self.logger.debug("LEASE-%i Start of handleStartShutdown" % l.id)
|
1296
|
l.print_contents()
|
1297
|
rr.state = ResourceReservation.STATE_ACTIVE
|
1298
|
try:
|
1299
|
self.resourcepool.stop_vms(l, rr)
|
1300
|
except EnactmentError, exc:
|
1301
|
self.logger.error("Enactment error when shutting down VMs.")
|
1302
|
|
1303
|
|
1304
|
|
1305
|
|
1306
|
raise
|
1307
|
|
1308
|
l.print_contents()
|
1309
|
self.logger.debug("LEASE-%i End of handleStartShutdown" % l.id)
|
1310
|
|
1311
|
|
1312
|
def _handle_end_shutdown(self, l, rr):
|
1313
|
""" Handles the end of a SuspensionResourceReservation
|
1314
|
|
1315
|
Arguments:
|
1316
|
l -- Lease the SuspensionResourceReservation belongs to
|
1317
|
rr -- The SuspensionResourceReservation
|
1318
|
|
1319
|
"""
|
1320
|
self.logger.debug("LEASE-%i Start of handleEndShutdown" % l.id)
|
1321
|
l.print_contents()
|
1322
|
rr.state = ResourceReservation.STATE_DONE
|
1323
|
l.print_contents()
|
1324
|
self.logger.debug("LEASE-%i End of handleEndShutdown" % l.id)
|
1325
|
self.logger.info("Lease %i's VMs have shutdown." % (l.id))
|
1326
|
raise NormalEndLeaseException
|
1327
|
|
1328
|
|
1329
|
def _handle_start_migrate(self, l, rr):
|
1330
|
""" Handles the start of a MemImageMigrationResourceReservation
|
1331
|
|
1332
|
Arguments:
|
1333
|
l -- Lease the MemImageMigrationResourceReservation belongs to
|
1334
|
rr -- The MemImageMigrationResourceReservation
|
1335
|
|
1336
|
"""
|
1337
|
self.logger.debug("LEASE-%i Start of handleStartMigrate" % l.id)
|
1338
|
l.print_contents()
|
1339
|
rr.state = ResourceReservation.STATE_ACTIVE
|
1340
|
l.print_contents()
|
1341
|
self.logger.debug("LEASE-%i End of handleStartMigrate" % l.id)
|
1342
|
self.logger.info("Migrating lease %i..." % (l.id))
|
1343
|
|
1344
|
|
1345
|
def _handle_end_migrate(self, l, rr):
|
1346
|
""" Handles the end of a MemImageMigrationResourceReservation
|
1347
|
|
1348
|
Arguments:
|
1349
|
l -- Lease the MemImageMigrationResourceReservation belongs to
|
1350
|
rr -- The MemImageMigrationResourceReservation
|
1351
|
|
1352
|
"""
|
1353
|
self.logger.debug("LEASE-%i Start of handleEndMigrate" % l.id)
|
1354
|
l.print_contents()
|
1355
|
|
1356
|
for vnode in rr.transfers:
|
1357
|
origin = rr.transfers[vnode][0]
|
1358
|
dest = rr.transfers[vnode][1]
|
1359
|
|
1360
|
|
1361
|
self.resourcepool.remove_ramfile(origin, l.id, vnode)
|
1362
|
self.resourcepool.add_ramfile(dest, l.id, vnode, l.requested_resources[vnode].get_quantity(constants.RES_MEM))
|
1363
|
|
1364
|
rr.state = ResourceReservation.STATE_DONE
|
1365
|
l.print_contents()
|
1366
|
self.logger.debug("LEASE-%i End of handleEndMigrate" % l.id)
|
1367
|
self.logger.info("Migrated lease %i..." % (l.id))
|
1368
|
|
1369
|
|
1370
|
|
1371
|
class VMResourceReservation(ResourceReservation):
|
1372
|
def __init__(self, lease, start, end, nodes, res):
|
1373
|
ResourceReservation.__init__(self, lease, start, end, res)
|
1374
|
self.nodes = nodes
|
1375
|
self.pre_rrs = []
|
1376
|
self.post_rrs = []
|
1377
|
|
1378
|
|
1379
|
self.__update_prematureend()
|
1380
|
|
1381
|
def update_start(self, time):
|
1382
|
self.start = time
|
1383
|
|
1384
|
self.__update_prematureend()
|
1385
|
|
1386
|
def update_end(self, time):
|
1387
|
self.end = time
|
1388
|
|
1389
|
self.__update_prematureend()
|
1390
|
|
1391
|
|
1392
|
def __update_prematureend(self):
|
1393
|
if self.lease.duration.known != None:
|
1394
|
remdur = self.lease.duration.get_remaining_known_duration()
|
1395
|
rrdur = self.end - self.start
|
1396
|
if remdur < rrdur:
|
1397
|
self.prematureend = self.start + remdur
|
1398
|
|
1399
|
|
1400
|
|
1401
|
|
1402
|
if self.prematureend == self.start:
|
1403
|
self.prematureend += 1
|
1404
|
else:
|
1405
|
self.prematureend = None
|
1406
|
else:
|
1407
|
self.prematureend = None
|
1408
|
|
1409
|
def get_final_end(self):
|
1410
|
if len(self.post_rrs) == 0:
|
1411
|
return self.end
|
1412
|
else:
|
1413
|
return self.post_rrs[-1].end
|
1414
|
|
1415
|
def is_suspending(self):
|
1416
|
return len(self.post_rrs) > 0 and isinstance(self.post_rrs[0], SuspensionResourceReservation)
|
1417
|
|
1418
|
def is_shutting_down(self):
|
1419
|
return len(self.post_rrs) > 0 and isinstance(self.post_rrs[0], ShutdownResourceReservation)
|
1420
|
|
1421
|
def print_contents(self, loglevel=constants.LOGLEVEL_VDEBUG):
|
1422
|
for resmrr in self.pre_rrs:
|
1423
|
resmrr.print_contents(loglevel)
|
1424
|
self.logger.log(loglevel, "--")
|
1425
|
self.logger.log(loglevel, "Type : VM")
|
1426
|
self.logger.log(loglevel, "Nodes : %s" % pretty_nodemap(self.nodes))
|
1427
|
if self.prematureend != None:
|
1428
|
self.logger.log(loglevel, "Premature end : %s" % self.prematureend)
|
1429
|
ResourceReservation.print_contents(self, loglevel)
|
1430
|
for susprr in self.post_rrs:
|
1431
|
self.logger.log(loglevel, "--")
|
1432
|
susprr.print_contents(loglevel)
|
1433
|
|
1434
|
|
1435
|
class SuspensionResourceReservation(ResourceReservation):
|
1436
|
def __init__(self, lease, start, end, res, vnodes, vmrr):
|
1437
|
ResourceReservation.__init__(self, lease, start, end, res)
|
1438
|
self.vmrr = vmrr
|
1439
|
self.vnodes = vnodes
|
1440
|
|
1441
|
def print_contents(self, loglevel=constants.LOGLEVEL_VDEBUG):
|
1442
|
self.logger.log(loglevel, "Type : SUSPEND")
|
1443
|
self.logger.log(loglevel, "Vnodes : %s" % self.vnodes)
|
1444
|
ResourceReservation.print_contents(self, loglevel)
|
1445
|
|
1446
|
def is_first(self):
|
1447
|
return (self == self.vmrr.post_rrs[0])
|
1448
|
|
1449
|
def is_last(self):
|
1450
|
return (self == self.vmrr.post_rrs[-1])
|
1451
|
|
1452
|
|
1453
|
class ResumptionResourceReservation(ResourceReservation):
|
1454
|
def __init__(self, lease, start, end, res, vnodes, vmrr):
|
1455
|
ResourceReservation.__init__(self, lease, start, end, res)
|
1456
|
self.vmrr = vmrr
|
1457
|
self.vnodes = vnodes
|
1458
|
|
1459
|
def print_contents(self, loglevel=constants.LOGLEVEL_VDEBUG):
|
1460
|
self.logger.log(loglevel, "Type : RESUME")
|
1461
|
self.logger.log(loglevel, "Vnodes : %s" % self.vnodes)
|
1462
|
ResourceReservation.print_contents(self, loglevel)
|
1463
|
|
1464
|
def is_first(self):
|
1465
|
resm_rrs = [r for r in self.vmrr.pre_rrs if isinstance(r, ResumptionResourceReservation)]
|
1466
|
return (self == resm_rrs[0])
|
1467
|
|
1468
|
def is_last(self):
|
1469
|
resm_rrs = [r for r in self.vmrr.pre_rrs if isinstance(r, ResumptionResourceReservation)]
|
1470
|
return (self == resm_rrs[-1])
|
1471
|
|
1472
|
|
1473
|
class ShutdownResourceReservation(ResourceReservation):
|
1474
|
def __init__(self, lease, start, end, res, vnodes, vmrr):
|
1475
|
ResourceReservation.__init__(self, lease, start, end, res)
|
1476
|
self.vmrr = vmrr
|
1477
|
self.vnodes = vnodes
|
1478
|
|
1479
|
def print_contents(self, loglevel=constants.LOGLEVEL_VDEBUG):
|
1480
|
self.logger.log(loglevel, "Type : SHUTDOWN")
|
1481
|
ResourceReservation.print_contents(self, loglevel)
|
1482
|
|
1483
|
|
1484
|
class MemImageMigrationResourceReservation(MigrationResourceReservation):
|
1485
|
def __init__(self, lease, start, end, res, vmrr, transfers):
|
1486
|
MigrationResourceReservation.__init__(self, lease, start, end, res, vmrr, transfers)
|
1487
|
|
1488
|
def print_contents(self, loglevel=constants.LOGLEVEL_VDEBUG):
|
1489
|
self.logger.log(loglevel, "Type : MEM IMAGE MIGRATION")
|
1490
|
self.logger.log(loglevel, "Transfers : %s" % self.transfers)
|
1491
|
ResourceReservation.print_contents(self, loglevel)
|