Statistics
| Branch: | Tag: | Revision:

root / snf-cyclades-app / synnefo / logic / backend.py @ fae6e5f0

History | View | Annotate | Download (37.8 kB)

1
# Copyright 2011-2013 GRNET S.A. All rights reserved.
2
#
3
# Redistribution and use in source and binary forms, with or
4
# without modification, are permitted provided that the following
5
# conditions are met:
6
#
7
#   1. Redistributions of source code must retain the above
8
#      copyright notice, this list of conditions and the following
9
#      disclaimer.
10
#
11
#   2. Redistributions in binary form must reproduce the above
12
#      copyright notice, this list of conditions and the following
13
#      disclaimer in the documentation and/or other materials
14
#      provided with the distribution.
15
#
16
# THIS SOFTWARE IS PROVIDED BY GRNET S.A. ``AS IS'' AND ANY EXPRESS
17
# OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18
# WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL GRNET S.A OR
20
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23
# USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
24
# AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25
# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
26
# ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27
# POSSIBILITY OF SUCH DAMAGE.
28
#
29
# The views and conclusions contained in the software and
30
# documentation are those of the authors and should not be
31
# interpreted as representing official policies, either expressed
32
# or implied, of GRNET S.A.
33
from django.conf import settings
34
from django.db import transaction
35
from datetime import datetime, timedelta
36

    
37
from synnefo.db.models import (Backend, VirtualMachine, Network,
38
                               BackendNetwork, BACKEND_STATUSES,
39
                               pooled_rapi_client, VirtualMachineDiagnostic,
40
                               Flavor, IPAddressLog)
41
from synnefo.logic import utils
42
from synnefo import quotas
43
from synnefo.api.util import release_resource, allocate_ip
44
from synnefo.util.mac2eui64 import mac2eui64
45
from synnefo.logic.rapi import GanetiApiError
46

    
47
from logging import getLogger
48
log = getLogger(__name__)
49

    
50

    
51
_firewall_tags = {
52
    'ENABLED': settings.GANETI_FIREWALL_ENABLED_TAG,
53
    'DISABLED': settings.GANETI_FIREWALL_DISABLED_TAG,
54
    'PROTECTED': settings.GANETI_FIREWALL_PROTECTED_TAG}
55

    
56
_reverse_tags = dict((v.split(':')[3], k) for k, v in _firewall_tags.items())
57

    
58
# Timeout in seconds for building NICs. After this period the NICs considered
59
# stale and removed from DB.
60
BUILDING_NIC_TIMEOUT = timedelta(seconds=180)
61

    
62
SIMPLE_NIC_FIELDS = ["state", "mac", "network", "firewall_profile", "index"]
63
COMPLEX_NIC_FIELDS = ["ipv4_address", "ipv6_address"]
64
NIC_FIELDS = SIMPLE_NIC_FIELDS + COMPLEX_NIC_FIELDS
65
UNKNOWN_NIC_PREFIX = "unknown-"
66

    
67

    
68
def handle_vm_quotas(vm, job_id, job_opcode, job_status, job_fields):
69
    """Handle quotas for updated VirtualMachine.
70

71
    Update quotas for the updated VirtualMachine based on the job that run on
72
    the Ganeti backend. If a commission has been already issued for this job,
73
    then this commission is just accepted or rejected based on the job status.
74
    Otherwise, a new commission for the given change is issued, that is also in
75
    force and auto-accept mode. In this case, previous commissions are
76
    rejected, since they reflect a previous state of the VM.
77

78
    """
79
    if job_status not in ["success", "error", "canceled"]:
80
        return vm
81

    
82
    # Check successful completion of a job will trigger any quotable change in
83
    # the VM state.
84
    action = utils.get_action_from_opcode(job_opcode, job_fields)
85
    if action == "BUILD":
86
        # Quotas for new VMs are automatically accepted by the API
87
        return vm
88
    commission_info = quotas.get_commission_info(vm, action=action,
89
                                                 action_fields=job_fields)
90

    
91
    if vm.task_job_id == job_id and vm.serial is not None:
92
        # Commission for this change has already been issued. So just
93
        # accept/reject it. Special case is OP_INSTANCE_CREATE, which even
94
        # if fails, must be accepted, as the user must manually remove the
95
        # failed server
96
        serial = vm.serial
97
        if job_status == "success":
98
            quotas.accept_serial(serial)
99
        elif job_status in ["error", "canceled"]:
100
            log.debug("Job %s failed. Rejecting related serial %s", job_id,
101
                      serial)
102
            quotas.reject_serial(serial)
103
        vm.serial = None
104
    elif job_status == "success" and commission_info is not None:
105
        log.debug("Expected job was %s. Processing job %s. Commission for"
106
                  " this job: %s", vm.task_job_id, job_id, commission_info)
107
        # Commission for this change has not been issued, or the issued
108
        # commission was unaware of the current change. Reject all previous
109
        # commissions and create a new one in forced mode!
110
        commission_name = ("client: dispatcher, resource: %s, ganeti_job: %s"
111
                           % (vm, job_id))
112
        quotas.handle_resource_commission(vm, action,
113
                                          commission_info=commission_info,
114
                                          commission_name=commission_name,
115
                                          force=True,
116
                                          auto_accept=True)
117
        log.debug("Issued new commission: %s", vm.serial)
118

    
119
    return vm
120

    
121

    
122
@transaction.commit_on_success
123
def process_op_status(vm, etime, jobid, opcode, status, logmsg, nics=None,
124
                      job_fields=None):
125
    """Process a job progress notification from the backend
126

127
    Process an incoming message from the backend (currently Ganeti).
128
    Job notifications with a terminating status (sucess, error, or canceled),
129
    also update the operating state of the VM.
130

131
    """
132
    # See #1492, #1031, #1111 why this line has been removed
133
    #if (opcode not in [x[0] for x in VirtualMachine.BACKEND_OPCODES] or
134
    if status not in [x[0] for x in BACKEND_STATUSES]:
135
        raise VirtualMachine.InvalidBackendMsgError(opcode, status)
136

    
137
    vm.backendjobid = jobid
138
    vm.backendjobstatus = status
139
    vm.backendopcode = opcode
140
    vm.backendlogmsg = logmsg
141

    
142
    if status in ["queued", "waiting", "running"]:
143
        vm.save()
144
        return
145

    
146
    if job_fields is None:
147
        job_fields = {}
148
    state_for_success = VirtualMachine.OPER_STATE_FROM_OPCODE.get(opcode)
149

    
150
    # Notifications of success change the operating state
151
    if status == "success":
152
        if state_for_success is not None:
153
            vm.operstate = state_for_success
154
        beparams = job_fields.get("beparams", None)
155
        if beparams:
156
            # Change the flavor of the VM
157
            _process_resize(vm, beparams)
158
        # Update backendtime only for jobs that have been successfully
159
        # completed, since only these jobs update the state of the VM. Else a
160
        # "race condition" may occur when a successful job (e.g.
161
        # OP_INSTANCE_REMOVE) completes before an error job and messages arrive
162
        # in reversed order.
163
        vm.backendtime = etime
164

    
165
    if status in ["success", "error", "canceled"] and nics is not None:
166
        # Update the NICs of the VM
167
        _process_net_status(vm, etime, nics)
168

    
169
    # Special case: if OP_INSTANCE_CREATE fails --> ERROR
170
    if opcode == 'OP_INSTANCE_CREATE' and status in ('canceled', 'error'):
171
        vm.operstate = 'ERROR'
172
        vm.backendtime = etime
173
        # Update state of associated NICs
174
        vm.nics.all().update(state="ERROR")
175
    elif opcode == 'OP_INSTANCE_REMOVE':
176
        # Special case: OP_INSTANCE_REMOVE fails for machines in ERROR,
177
        # when no instance exists at the Ganeti backend.
178
        # See ticket #799 for all the details.
179
        if status == 'success' or (status == 'error' and
180
                                   not vm_exists_in_backend(vm)):
181
            # VM has been deleted
182
            for nic in vm.nics.all():
183
                # Release the IP
184
                remove_nic_ips(nic)
185
                # And delete the NIC.
186
                nic.delete()
187
            vm.deleted = True
188
            vm.operstate = state_for_success
189
            vm.backendtime = etime
190
            status = "success"
191

    
192
    if status in ["success", "error", "canceled"]:
193
        # Job is finalized: Handle quotas/commissioning
194
        vm = handle_vm_quotas(vm, job_id=jobid, job_opcode=opcode,
195
                              job_status=status, job_fields=job_fields)
196
        # and clear task fields
197
        if vm.task_job_id == jobid:
198
            vm.task = None
199
            vm.task_job_id = None
200

    
201
    vm.save()
202

    
203

    
204
def _process_resize(vm, beparams):
205
    """Change flavor of a VirtualMachine based on new beparams."""
206
    old_flavor = vm.flavor
207
    vcpus = beparams.get("vcpus", old_flavor.cpu)
208
    ram = beparams.get("maxmem", old_flavor.ram)
209
    if vcpus == old_flavor.cpu and ram == old_flavor.ram:
210
        return
211
    try:
212
        new_flavor = Flavor.objects.get(cpu=vcpus, ram=ram,
213
                                        disk=old_flavor.disk,
214
                                        disk_template=old_flavor.disk_template)
215
    except Flavor.DoesNotExist:
216
        raise Exception("Can not find flavor for VM")
217
    vm.flavor = new_flavor
218
    vm.save()
219

    
220

    
221
@transaction.commit_on_success
222
def process_net_status(vm, etime, nics):
223
    """Wrap _process_net_status inside transaction."""
224
    _process_net_status(vm, etime, nics)
225

    
226

    
227
def _process_net_status(vm, etime, nics):
228
    """Process a net status notification from the backend
229

230
    Process an incoming message from the Ganeti backend,
231
    detailing the NIC configuration of a VM instance.
232

233
    Update the state of the VM in the DB accordingly.
234

235
    """
236
    ganeti_nics = process_ganeti_nics(nics)
237
    db_nics = dict([(nic.id, nic)
238
                    for nic in vm.nics.prefetch_related("ips__subnet")])
239

    
240
    # Get X-Lock on backend before getting X-Lock on network IP pools, to
241
    # guarantee that no deadlock will occur with Backend allocator.
242
    Backend.objects.select_for_update().get(id=vm.backend_id)
243

    
244
    for nic_name in set(db_nics.keys()) | set(ganeti_nics.keys()):
245
        db_nic = db_nics.get(nic_name)
246
        ganeti_nic = ganeti_nics.get(nic_name)
247
        if ganeti_nic is None:
248
            # NIC exists in DB but not in Ganeti. If the NIC is in 'building'
249
            # state for more than 5 minutes, then we remove the NIC.
250
            # TODO: This is dangerous as the job may be stack in the queue, and
251
            # releasing the IP may lead to duplicate IP use.
252
            if db_nic.state != "BUILD" or\
253
                (db_nic.state == "BUILD" and
254
                 etime > db_nic.created + BUILDING_NIC_TIMEOUT):
255
                remove_nic_ips(db_nic)
256
                db_nic.delete()
257
            else:
258
                log.warning("Ignoring recent building NIC: %s", db_nic)
259
        elif db_nic is None:
260
            msg = ("NIC/%s of VM %s does not exist in DB! Cannot automatically"
261
                   " fix this issue!" % (nic_name, vm))
262
            log.error(msg)
263
            continue
264
        elif not nics_are_equal(db_nic, ganeti_nic):
265
            for f in SIMPLE_NIC_FIELDS:
266
                # Update the NIC in DB with the values from Ganeti NIC
267
                setattr(db_nic, f, ganeti_nic[f])
268
                db_nic.save()
269
            # Special case where the IPv4 address has changed, because you
270
            # need to release the old IPv4 address and reserve the new one
271
            ipv4_address = ganeti_nic["ipv4_address"]
272
            if db_nic.ipv4_address != ipv4_address:
273
                remove_nic_ips(db_nic)
274
                if ipv4_address:
275
                    network = ganeti_nic["network"]
276
                    ipaddress = allocate_ip(network, vm.userid,
277
                                            address=ipv4_address)
278
                    ipaddress.nic = nic
279
                    ipaddress.save()
280

    
281
    vm.backendtime = etime
282
    vm.save()
283

    
284

    
285
def nics_are_equal(db_nic, gnt_nic):
286
    for field in NIC_FIELDS:
287
        if getattr(db_nic, field) != gnt_nic[field]:
288
            return False
289
    return True
290

    
291

    
292
def process_ganeti_nics(ganeti_nics):
293
    """Process NIC dict from ganeti"""
294
    new_nics = []
295
    for index, gnic in enumerate(ganeti_nics):
296
        nic_name = gnic.get("name", None)
297
        if nic_name is not None:
298
            nic_id = utils.id_from_nic_name(nic_name)
299
        else:
300
            # Put as default value the index. If it is an unknown NIC to
301
            # synnefo it will be created automaticaly.
302
            nic_id = UNKNOWN_NIC_PREFIX + str(index)
303
        network_name = gnic.get('network', '')
304
        network_id = utils.id_from_network_name(network_name)
305
        network = Network.objects.get(id=network_id)
306

    
307
        # Get the new nic info
308
        mac = gnic.get('mac')
309
        ipv4 = gnic.get('ip')
310
        subnet6 = network.subnet6
311
        ipv6 = mac2eui64(mac, subnet6) if subnet6 else None
312

    
313
        firewall = gnic.get('firewall')
314
        firewall_profile = _reverse_tags.get(firewall)
315
        if not firewall_profile and network.public:
316
            firewall_profile = settings.DEFAULT_FIREWALL_PROFILE
317

    
318
        nic_info = {
319
            'index': index,
320
            'network': network,
321
            'mac': mac,
322
            'ipv4_address': ipv4,
323
            'ipv6_address': ipv6,
324
            'firewall_profile': firewall_profile,
325
            'state': 'ACTIVE'}
326

    
327
        new_nics.append((nic_id, nic_info))
328
    return dict(new_nics)
329

    
330

    
331
def remove_nic_ips(nic):
332
    """Remove IP addresses associated with a NetworkInterface.
333

334
    Remove all IP addresses that are associated with the NetworkInterface
335
    object, by returning them to the pool and deleting the IPAddress object. If
336
    the IP is a floating IP, then it is just disassociated from the NIC.
337

338
    """
339

    
340
    for ip in nic.ips.all():
341
        # Update the DB table holding the logging of all IP addresses
342
        update_ip_address_log(nic, ip)
343

    
344
        if ip.ipversion == 4:
345
            if ip.floating_ip:
346
                ip.nic = None
347
                ip.save()
348
            else:
349
                ip.release_address()
350
        if not ip.floating_ip:
351
            ip.delete()
352

    
353

    
354
def update_ip_address_log(nic, ip):
355
    """Update DB logging entry for this IP address."""
356
    if not ip.network.public or nic.machine is None:
357
        return
358
    try:
359
        ip_log, created = \
360
            IPAddressLog.objects.get_or_create(server_id=nic.machine_id,
361
                                               network_id=ip.network_id,
362
                                               address=ip.address,
363
                                               active=True)
364
    except IPAddressLog.MultipleObjectsReturned:
365
        logmsg = ("Multiple active log entries for IP %s, Network %s,"
366
                  "Server %s. Can not proceed!"
367
                  % (ip.address, ip.network, nic.machine))
368
        log.error(logmsg)
369
        raise
370

    
371
    if created:
372
        logmsg = ("No log entry for IP %s, Network %s, Server %s. Created new"
373
                  " but with wrong creation timestamp."
374
                  % (ip.address, ip.network, nic.machine))
375
        log.error(logmsg)
376
    ip_log.released_at = datetime.now()
377
    ip_log.active = False
378
    ip_log.save()
379

    
380

    
381
@transaction.commit_on_success
382
def process_network_status(back_network, etime, jobid, opcode, status, logmsg):
383
    if status not in [x[0] for x in BACKEND_STATUSES]:
384
        raise Network.InvalidBackendMsgError(opcode, status)
385

    
386
    back_network.backendjobid = jobid
387
    back_network.backendjobstatus = status
388
    back_network.backendopcode = opcode
389
    back_network.backendlogmsg = logmsg
390

    
391
    network = back_network.network
392

    
393
    # Notifications of success change the operating state
394
    state_for_success = BackendNetwork.OPER_STATE_FROM_OPCODE.get(opcode, None)
395
    if status == 'success' and state_for_success is not None:
396
        back_network.operstate = state_for_success
397

    
398
    if status in ('canceled', 'error') and opcode == 'OP_NETWORK_ADD':
399
        back_network.operstate = 'ERROR'
400
        back_network.backendtime = etime
401

    
402
    if opcode == 'OP_NETWORK_REMOVE':
403
        network_is_deleted = (status == "success")
404
        if network_is_deleted or (status == "error" and not
405
                                  network_exists_in_backend(back_network)):
406
            back_network.operstate = state_for_success
407
            back_network.deleted = True
408
            back_network.backendtime = etime
409

    
410
    if status == 'success':
411
        back_network.backendtime = etime
412
    back_network.save()
413
    # Also you must update the state of the Network!!
414
    update_network_state(network)
415

    
416

    
417
def update_network_state(network):
418
    """Update the state of a Network based on BackendNetwork states.
419

420
    Update the state of a Network based on the operstate of the networks in the
421
    backends that network exists.
422

423
    The state of the network is:
424
    * ACTIVE: If it is 'ACTIVE' in at least one backend.
425
    * DELETED: If it is is 'DELETED' in all backends that have been created.
426

427
    This function also releases the resources (MAC prefix or Bridge) and the
428
    quotas for the network.
429

430
    """
431
    if network.deleted:
432
        # Network has already been deleted. Just assert that state is also
433
        # DELETED
434
        if not network.state == "DELETED":
435
            network.state = "DELETED"
436
            network.save()
437
        return
438

    
439
    backend_states = [s.operstate for s in network.backend_networks.all()]
440
    if not backend_states and network.action != "DESTROY":
441
        if network.state != "ACTIVE":
442
            network.state = "ACTIVE"
443
            network.save()
444
            return
445

    
446
    # Network is deleted when all BackendNetworks go to "DELETED" operstate
447
    deleted = reduce(lambda x, y: x == y and "DELETED", backend_states,
448
                     "DELETED")
449

    
450
    # Release the resources on the deletion of the Network
451
    if deleted:
452
        log.info("Network %r deleted. Releasing link %r mac_prefix %r",
453
                 network.id, network.mac_prefix, network.link)
454
        network.deleted = True
455
        network.state = "DELETED"
456
        if network.mac_prefix:
457
            if network.FLAVORS[network.flavor]["mac_prefix"] == "pool":
458
                release_resource(res_type="mac_prefix",
459
                                 value=network.mac_prefix)
460
        if network.link:
461
            if network.FLAVORS[network.flavor]["link"] == "pool":
462
                release_resource(res_type="bridge", value=network.link)
463

    
464
        # Issue commission
465
        if network.userid:
466
            quotas.issue_and_accept_commission(network, delete=True)
467
            # the above has already saved the object and committed;
468
            # a second save would override others' changes, since the
469
            # object is now unlocked
470
            return
471
        elif not network.public:
472
            log.warning("Network %s does not have an owner!", network.id)
473

    
474
        # TODO!!!!!
475
        # Set all subnets as deleted
476
        network.subnets.update(deleted=True)
477
        # And delete the IP pools
478
        network.subnets.ip_pools.all().delete()
479
    network.save()
480

    
481

    
482
@transaction.commit_on_success
483
def process_network_modify(back_network, etime, jobid, opcode, status,
484
                           job_fields):
485
    assert (opcode == "OP_NETWORK_SET_PARAMS")
486
    if status not in [x[0] for x in BACKEND_STATUSES]:
487
        raise Network.InvalidBackendMsgError(opcode, status)
488

    
489
    back_network.backendjobid = jobid
490
    back_network.backendjobstatus = status
491
    back_network.opcode = opcode
492

    
493
    add_reserved_ips = job_fields.get("add_reserved_ips")
494
    if add_reserved_ips:
495
        network = back_network.network
496
        for ip in add_reserved_ips:
497
            network.reserve_address(ip, external=True)
498

    
499
    if status == 'success':
500
        back_network.backendtime = etime
501
    back_network.save()
502

    
503

    
504
@transaction.commit_on_success
505
def process_create_progress(vm, etime, progress):
506

    
507
    percentage = int(progress)
508

    
509
    # The percentage may exceed 100%, due to the way
510
    # snf-image:copy-progress tracks bytes read by image handling processes
511
    percentage = 100 if percentage > 100 else percentage
512
    if percentage < 0:
513
        raise ValueError("Percentage cannot be negative")
514

    
515
    # FIXME: log a warning here, see #1033
516
#   if last_update > percentage:
517
#       raise ValueError("Build percentage should increase monotonically " \
518
#                        "(old = %d, new = %d)" % (last_update, percentage))
519

    
520
    # This assumes that no message of type 'ganeti-create-progress' is going to
521
    # arrive once OP_INSTANCE_CREATE has succeeded for a Ganeti instance and
522
    # the instance is STARTED.  What if the two messages are processed by two
523
    # separate dispatcher threads, and the 'ganeti-op-status' message for
524
    # successful creation gets processed before the 'ganeti-create-progress'
525
    # message? [vkoukis]
526
    #
527
    #if not vm.operstate == 'BUILD':
528
    #    raise VirtualMachine.IllegalState("VM is not in building state")
529

    
530
    vm.buildpercentage = percentage
531
    vm.backendtime = etime
532
    vm.save()
533

    
534

    
535
@transaction.commit_on_success
536
def create_instance_diagnostic(vm, message, source, level="DEBUG", etime=None,
537
                               details=None):
538
    """
539
    Create virtual machine instance diagnostic entry.
540

541
    :param vm: VirtualMachine instance to create diagnostic for.
542
    :param message: Diagnostic message.
543
    :param source: Diagnostic source identifier (e.g. image-helper).
544
    :param level: Diagnostic level (`DEBUG`, `INFO`, `WARNING`, `ERROR`).
545
    :param etime: The time the message occured (if available).
546
    :param details: Additional details or debug information.
547
    """
548
    VirtualMachineDiagnostic.objects.create_for_vm(vm, level, source=source,
549
                                                   source_date=etime,
550
                                                   message=message,
551
                                                   details=details)
552

    
553

    
554
def create_instance(vm, nics, flavor, image):
555
    """`image` is a dictionary which should contain the keys:
556
            'backend_id', 'format' and 'metadata'
557

558
        metadata value should be a dictionary.
559
    """
560

    
561
    # Handle arguments to CreateInstance() as a dictionary,
562
    # initialize it based on a deployment-specific value.
563
    # This enables the administrator to override deployment-specific
564
    # arguments, such as the disk template to use, name of os provider
565
    # and hypervisor-specific parameters at will (see Synnefo #785, #835).
566
    #
567
    kw = vm.backend.get_create_params()
568
    kw['mode'] = 'create'
569
    kw['name'] = vm.backend_vm_id
570
    # Defined in settings.GANETI_CREATEINSTANCE_KWARGS
571

    
572
    kw['disk_template'] = flavor.disk_template
573
    kw['disks'] = [{"size": flavor.disk * 1024}]
574
    provider = flavor.disk_provider
575
    if provider:
576
        kw['disks'][0]['provider'] = provider
577
        kw['disks'][0]['origin'] = flavor.disk_origin
578

    
579
    kw['nics'] = [{"name": nic.backend_uuid,
580
                   "network": nic.network.backend_id,
581
                   "ip": nic.ipv4_address}
582
                  for nic in nics]
583
    backend = vm.backend
584
    depend_jobs = []
585
    for nic in nics:
586
        network = Network.objects.select_for_update().get(id=nic.network_id)
587
        bnet, created = BackendNetwork.objects.get_or_create(backend=backend,
588
                                                             network=network)
589
        if bnet.operstate != "ACTIVE":
590
            depend_jobs = create_network(network, backend, connect=True)
591
    kw["depends"] = [[job, ["success", "error", "canceled"]]
592
                     for job in depend_jobs]
593

    
594
    # Defined in settings.GANETI_CREATEINSTANCE_KWARGS
595
    # kw['os'] = settings.GANETI_OS_PROVIDER
596
    kw['ip_check'] = False
597
    kw['name_check'] = False
598

    
599
    # Do not specific a node explicitly, have
600
    # Ganeti use an iallocator instead
601
    #kw['pnode'] = rapi.GetNodes()[0]
602

    
603
    kw['dry_run'] = settings.TEST
604

    
605
    kw['beparams'] = {
606
        'auto_balance': True,
607
        'vcpus': flavor.cpu,
608
        'memory': flavor.ram}
609

    
610
    kw['osparams'] = {
611
        'config_url': vm.config_url,
612
        # Store image id and format to Ganeti
613
        'img_id': image['backend_id'],
614
        'img_format': image['format']}
615

    
616
    # Use opportunistic locking
617
    kw['opportunistic_locking'] = True
618

    
619
    # Defined in settings.GANETI_CREATEINSTANCE_KWARGS
620
    # kw['hvparams'] = dict(serial_console=False)
621

    
622
    log.debug("Creating instance %s", utils.hide_pass(kw))
623
    with pooled_rapi_client(vm) as client:
624
        return client.CreateInstance(**kw)
625

    
626

    
627
def delete_instance(vm):
628
    with pooled_rapi_client(vm) as client:
629
        return client.DeleteInstance(vm.backend_vm_id, dry_run=settings.TEST)
630

    
631

    
632
def reboot_instance(vm, reboot_type):
633
    assert reboot_type in ('soft', 'hard')
634
    kwargs = {"instance": vm.backend_vm_id,
635
              "reboot_type": "hard"}
636
    # XXX: Currently shutdown_timeout parameter is not supported from the
637
    # Ganeti RAPI. Until supported, we will fallback for both reboot types
638
    # to the default shutdown timeout of Ganeti (120s). Note that reboot
639
    # type of Ganeti job must be always hard. The 'soft' and 'hard' type
640
    # of OS API is different from the one in Ganeti, and maps to
641
    # 'shutdown_timeout'.
642
    #if reboot_type == "hard":
643
    #    kwargs["shutdown_timeout"] = 0
644
    if settings.TEST:
645
        kwargs["dry_run"] = True
646
    with pooled_rapi_client(vm) as client:
647
        return client.RebootInstance(**kwargs)
648

    
649

    
650
def startup_instance(vm):
651
    with pooled_rapi_client(vm) as client:
652
        return client.StartupInstance(vm.backend_vm_id, dry_run=settings.TEST)
653

    
654

    
655
def shutdown_instance(vm):
656
    with pooled_rapi_client(vm) as client:
657
        return client.ShutdownInstance(vm.backend_vm_id, dry_run=settings.TEST)
658

    
659

    
660
def resize_instance(vm, vcpus, memory):
661
    beparams = {"vcpus": int(vcpus),
662
                "minmem": int(memory),
663
                "maxmem": int(memory)}
664
    with pooled_rapi_client(vm) as client:
665
        return client.ModifyInstance(vm.backend_vm_id, beparams=beparams)
666

    
667

    
668
def get_instance_console(vm):
669
    # RAPI GetInstanceConsole() returns endpoints to the vnc_bind_address,
670
    # which is a cluster-wide setting, either 0.0.0.0 or 127.0.0.1, and pretty
671
    # useless (see #783).
672
    #
673
    # Until this is fixed on the Ganeti side, construct a console info reply
674
    # directly.
675
    #
676
    # WARNING: This assumes that VNC runs on port network_port on
677
    #          the instance's primary node, and is probably
678
    #          hypervisor-specific.
679
    #
680
    log.debug("Getting console for vm %s", vm)
681

    
682
    console = {}
683
    console['kind'] = 'vnc'
684

    
685
    with pooled_rapi_client(vm) as client:
686
        i = client.GetInstance(vm.backend_vm_id)
687

    
688
    if vm.backend.hypervisor == "kvm" and i['hvparams']['serial_console']:
689
        raise Exception("hv parameter serial_console cannot be true")
690
    console['host'] = i['pnode']
691
    console['port'] = i['network_port']
692

    
693
    return console
694

    
695

    
696
def get_instance_info(vm):
697
    with pooled_rapi_client(vm) as client:
698
        return client.GetInstance(vm.backend_vm_id)
699

    
700

    
701
def vm_exists_in_backend(vm):
702
    try:
703
        get_instance_info(vm)
704
        return True
705
    except GanetiApiError as e:
706
        if e.code == 404:
707
            return False
708
        raise e
709

    
710

    
711
def get_network_info(backend_network):
712
    with pooled_rapi_client(backend_network) as client:
713
        return client.GetNetwork(backend_network.network.backend_id)
714

    
715

    
716
def network_exists_in_backend(backend_network):
717
    try:
718
        get_network_info(backend_network)
719
        return True
720
    except GanetiApiError as e:
721
        if e.code == 404:
722
            return False
723

    
724

    
725
def create_network(network, backend, connect=True):
726
    """Create a network in a Ganeti backend"""
727
    log.debug("Creating network %s in backend %s", network, backend)
728

    
729
    job_id = _create_network(network, backend)
730

    
731
    if connect:
732
        job_ids = connect_network(network, backend, depends=[job_id])
733
        return job_ids
734
    else:
735
        return [job_id]
736

    
737

    
738
def _create_network(network, backend):
739
    """Create a network."""
740

    
741
    tags = network.backend_tag
742
    subnet = None
743
    subnet6 = None
744
    gateway = None
745
    gateway6 = None
746
    for _subnet in network.subnets.all():
747
        if _subnet.ipversion == 4:
748
            if _subnet.dhcp:
749
                tags.append('nfdhcpd')
750
            subnet = _subnet.cidr
751
            gateway = _subnet.gateway
752
        elif _subnet.ipversion == 6:
753
            subnet6 = _subnet.cidr
754
            gateway6 = _subnet.gateway
755

    
756
    if network.public:
757
        conflicts_check = True
758
        tags.append('public')
759
    else:
760
        conflicts_check = False
761
        tags.append('private')
762

    
763
    # Use a dummy network subnet for IPv6 only networks. Currently Ganeti does
764
    # not support IPv6 only networks. To bypass this limitation, we create the
765
    # network with a dummy network subnet, and make Cyclades connect instances
766
    # to such networks, with address=None.
767
    if subnet is None:
768
        subnet = "10.0.0.0/24"
769

    
770
    try:
771
        bn = BackendNetwork.objects.get(network=network, backend=backend)
772
        mac_prefix = bn.mac_prefix
773
    except BackendNetwork.DoesNotExist:
774
        raise Exception("BackendNetwork for network '%s' in backend '%s'"
775
                        " does not exist" % (network.id, backend.id))
776

    
777
    with pooled_rapi_client(backend) as client:
778
        return client.CreateNetwork(network_name=network.backend_id,
779
                                    network=subnet,
780
                                    network6=subnet6,
781
                                    gateway=gateway,
782
                                    gateway6=gateway6,
783
                                    mac_prefix=mac_prefix,
784
                                    conflicts_check=conflicts_check,
785
                                    tags=tags)
786

    
787

    
788
def connect_network(network, backend, depends=[], group=None):
789
    """Connect a network to nodegroups."""
790
    log.debug("Connecting network %s to backend %s", network, backend)
791

    
792
    if network.public:
793
        conflicts_check = True
794
    else:
795
        conflicts_check = False
796

    
797
    depends = [[job, ["success", "error", "canceled"]] for job in depends]
798
    with pooled_rapi_client(backend) as client:
799
        groups = [group] if group is not None else client.GetGroups()
800
        job_ids = []
801
        for group in groups:
802
            job_id = client.ConnectNetwork(network.backend_id, group,
803
                                           network.mode, network.link,
804
                                           conflicts_check,
805
                                           depends=depends)
806
            job_ids.append(job_id)
807
    return job_ids
808

    
809

    
810
def delete_network(network, backend, disconnect=True):
811
    log.debug("Deleting network %s from backend %s", network, backend)
812

    
813
    depends = []
814
    if disconnect:
815
        depends = disconnect_network(network, backend)
816
    _delete_network(network, backend, depends=depends)
817

    
818

    
819
def _delete_network(network, backend, depends=[]):
820
    depends = [[job, ["success", "error", "canceled"]] for job in depends]
821
    with pooled_rapi_client(backend) as client:
822
        return client.DeleteNetwork(network.backend_id, depends)
823

    
824

    
825
def disconnect_network(network, backend, group=None):
826
    log.debug("Disconnecting network %s to backend %s", network, backend)
827

    
828
    with pooled_rapi_client(backend) as client:
829
        groups = [group] if group is not None else client.GetGroups()
830
        job_ids = []
831
        for group in groups:
832
            job_id = client.DisconnectNetwork(network.backend_id, group)
833
            job_ids.append(job_id)
834
    return job_ids
835

    
836

    
837
def connect_to_network(vm, nic):
838
    network = nic.network
839
    backend = vm.backend
840
    network = Network.objects.select_for_update().get(id=network.id)
841
    bnet, created = BackendNetwork.objects.get_or_create(backend=backend,
842
                                                         network=network)
843
    depend_jobs = []
844
    if bnet.operstate != "ACTIVE":
845
        depend_jobs = create_network(network, backend, connect=True)
846

    
847
    depends = [[job, ["success", "error", "canceled"]] for job in depend_jobs]
848

    
849
    nic = {'name': nic.backend_uuid,
850
           'network': network.backend_id,
851
           'ip': nic.ipv4_address}
852

    
853
    log.debug("Adding NIC %s to VM %s", nic, vm)
854

    
855
    kwargs = {
856
        "instance": vm.backend_vm_id,
857
        "nics": [("add", "-1", nic)],
858
        "depends": depends,
859
    }
860
    if vm.backend.use_hotplug():
861
        kwargs["hotplug"] = True
862
    if settings.TEST:
863
        kwargs["dry_run"] = True
864

    
865
    with pooled_rapi_client(vm) as client:
866
        return client.ModifyInstance(**kwargs)
867

    
868

    
869
def disconnect_from_network(vm, nic):
870
    log.debug("Removing NIC %s of VM %s", nic, vm)
871

    
872
    kwargs = {
873
        "instance": vm.backend_vm_id,
874
        "nics": [("remove", nic.backend_uuid, {})],
875
    }
876
    if vm.backend.use_hotplug():
877
        kwargs["hotplug"] = True
878
    if settings.TEST:
879
        kwargs["dry_run"] = True
880

    
881
    with pooled_rapi_client(vm) as client:
882
        jobID = client.ModifyInstance(**kwargs)
883
        firewall_profile = nic.firewall_profile
884
        if firewall_profile and firewall_profile != "DISABLED":
885
            tag = _firewall_tags[firewall_profile] % nic.backend_uuid
886
            client.DeleteInstanceTags(vm.backend_vm_id, [tag],
887
                                      dry_run=settings.TEST)
888

    
889
        return jobID
890

    
891

    
892
def set_firewall_profile(vm, profile, nic):
893
    uuid = nic.backend_uuid
894
    try:
895
        tag = _firewall_tags[profile] % uuid
896
    except KeyError:
897
        raise ValueError("Unsopported Firewall Profile: %s" % profile)
898

    
899
    log.debug("Setting tag of VM %s, NIC %s, to %s", vm, nic, profile)
900

    
901
    with pooled_rapi_client(vm) as client:
902
        # Delete previous firewall tags
903
        old_tags = client.GetInstanceTags(vm.backend_vm_id)
904
        delete_tags = [(t % uuid) for t in _firewall_tags.values()
905
                       if (t % uuid) in old_tags]
906
        if delete_tags:
907
            client.DeleteInstanceTags(vm.backend_vm_id, delete_tags,
908
                                      dry_run=settings.TEST)
909

    
910
        if profile != "DISABLED":
911
            client.AddInstanceTags(vm.backend_vm_id, [tag],
912
                                   dry_run=settings.TEST)
913

    
914
        # XXX NOP ModifyInstance call to force process_net_status to run
915
        # on the dispatcher
916
        os_name = settings.GANETI_CREATEINSTANCE_KWARGS['os']
917
        client.ModifyInstance(vm.backend_vm_id,
918
                              os_name=os_name)
919
    return None
920

    
921

    
922
def get_instances(backend, bulk=True):
923
    with pooled_rapi_client(backend) as c:
924
        return c.GetInstances(bulk=bulk)
925

    
926

    
927
def get_nodes(backend, bulk=True):
928
    with pooled_rapi_client(backend) as c:
929
        return c.GetNodes(bulk=bulk)
930

    
931

    
932
def get_jobs(backend, bulk=True):
933
    with pooled_rapi_client(backend) as c:
934
        return c.GetJobs(bulk=bulk)
935

    
936

    
937
def get_physical_resources(backend):
938
    """ Get the physical resources of a backend.
939

940
    Get the resources of a backend as reported by the backend (not the db).
941

942
    """
943
    nodes = get_nodes(backend, bulk=True)
944
    attr = ['mfree', 'mtotal', 'dfree', 'dtotal', 'pinst_cnt', 'ctotal']
945
    res = {}
946
    for a in attr:
947
        res[a] = 0
948
    for n in nodes:
949
        # Filter out drained, offline and not vm_capable nodes since they will
950
        # not take part in the vm allocation process
951
        can_host_vms = n['vm_capable'] and not (n['drained'] or n['offline'])
952
        if can_host_vms and n['cnodes']:
953
            for a in attr:
954
                res[a] += int(n[a] or 0)
955
    return res
956

    
957

    
958
def update_backend_resources(backend, resources=None):
959
    """ Update the state of the backend resources in db.
960

961
    """
962

    
963
    if not resources:
964
        resources = get_physical_resources(backend)
965

    
966
    backend.mfree = resources['mfree']
967
    backend.mtotal = resources['mtotal']
968
    backend.dfree = resources['dfree']
969
    backend.dtotal = resources['dtotal']
970
    backend.pinst_cnt = resources['pinst_cnt']
971
    backend.ctotal = resources['ctotal']
972
    backend.updated = datetime.now()
973
    backend.save()
974

    
975

    
976
def get_memory_from_instances(backend):
977
    """ Get the memory that is used from instances.
978

979
    Get the used memory of a backend. Note: This is different for
980
    the real memory used, due to kvm's memory de-duplication.
981

982
    """
983
    with pooled_rapi_client(backend) as client:
984
        instances = client.GetInstances(bulk=True)
985
    mem = 0
986
    for i in instances:
987
        mem += i['oper_ram']
988
    return mem
989

    
990

    
991
def get_available_disk_templates(backend):
992
    """Get the list of available disk templates of a Ganeti backend.
993

994
    The list contains the disk templates that are enabled in the Ganeti backend
995
    and also included in ipolicy-disk-templates.
996

997
    """
998
    with pooled_rapi_client(backend) as c:
999
        info = c.GetInfo()
1000
    ipolicy_disk_templates = info["ipolicy"]["disk-templates"]
1001
    try:
1002
        enabled_disk_templates = info["enabled_disk_templates"]
1003
        return [dp for dp in enabled_disk_templates
1004
                if dp in ipolicy_disk_templates]
1005
    except KeyError:
1006
        # Ganeti < 2.8 does not have 'enabled_disk_templates'
1007
        return ipolicy_disk_templates
1008

    
1009

    
1010
def update_backend_disk_templates(backend):
1011
    disk_templates = get_available_disk_templates(backend)
1012
    backend.disk_templates = disk_templates
1013
    backend.save()
1014

    
1015

    
1016
##
1017
## Synchronized operations for reconciliation
1018
##
1019

    
1020

    
1021
def create_network_synced(network, backend):
1022
    result = _create_network_synced(network, backend)
1023
    if result[0] != 'success':
1024
        return result
1025
    result = connect_network_synced(network, backend)
1026
    return result
1027

    
1028

    
1029
def _create_network_synced(network, backend):
1030
    with pooled_rapi_client(backend) as client:
1031
        job = _create_network(network, backend)
1032
        result = wait_for_job(client, job)
1033
    return result
1034

    
1035

    
1036
def connect_network_synced(network, backend):
1037
    with pooled_rapi_client(backend) as client:
1038
        for group in client.GetGroups():
1039
            job = client.ConnectNetwork(network.backend_id, group,
1040
                                        network.mode, network.link)
1041
            result = wait_for_job(client, job)
1042
            if result[0] != 'success':
1043
                return result
1044

    
1045
    return result
1046

    
1047

    
1048
def wait_for_job(client, jobid):
1049
    result = client.WaitForJobChange(jobid, ['status', 'opresult'], None, None)
1050
    status = result['job_info'][0]
1051
    while status not in ['success', 'error', 'cancel']:
1052
        result = client.WaitForJobChange(jobid, ['status', 'opresult'],
1053
                                         [result], None)
1054
        status = result['job_info'][0]
1055

    
1056
    if status == 'success':
1057
        return (status, None)
1058
    else:
1059
        error = result['job_info'][1]
1060
        return (status, error)