root / snf-cyclades-app / synnefo / logic / backend.py @ bc37fa88
History | View | Annotate | Download (35.5 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 |
36 |
|
37 |
from synnefo.db.models import (Backend, VirtualMachine, Network, |
38 |
FloatingIP, |
39 |
BackendNetwork, BACKEND_STATUSES, |
40 |
pooled_rapi_client, VirtualMachineDiagnostic, |
41 |
Flavor) |
42 |
from synnefo.logic import utils |
43 |
from synnefo import quotas |
44 |
from synnefo.api.util import release_resource |
45 |
from synnefo.util.mac2eui64 import mac2eui64 |
46 |
from synnefo.logic.rapi import GanetiApiError |
47 |
|
48 |
from logging import getLogger |
49 |
log = getLogger(__name__) |
50 |
|
51 |
|
52 |
_firewall_tags = { |
53 |
'ENABLED': settings.GANETI_FIREWALL_ENABLED_TAG,
|
54 |
'DISABLED': settings.GANETI_FIREWALL_DISABLED_TAG,
|
55 |
'PROTECTED': settings.GANETI_FIREWALL_PROTECTED_TAG}
|
56 |
|
57 |
_reverse_tags = dict((v.split(':')[3], k) for k, v in _firewall_tags.items()) |
58 |
|
59 |
|
60 |
def handle_vm_quotas(vm, job_id, job_opcode, job_status, job_fields): |
61 |
"""Handle quotas for updated VirtualMachine.
|
62 |
|
63 |
Update quotas for the updated VirtualMachine based on the job that run on
|
64 |
the Ganeti backend. If a commission has been already issued for this job,
|
65 |
then this commission is just accepted or rejected based on the job status.
|
66 |
Otherwise, a new commission for the given change is issued, that is also in
|
67 |
force and auto-accept mode. In this case, previous commissions are
|
68 |
rejected, since they reflect a previous state of the VM.
|
69 |
|
70 |
"""
|
71 |
if job_status not in ["success", "error", "canceled"]: |
72 |
return vm
|
73 |
|
74 |
# Check successful completion of a job will trigger any quotable change in
|
75 |
# the VM state.
|
76 |
action = utils.get_action_from_opcode(job_opcode, job_fields) |
77 |
if action == "BUILD": |
78 |
# Quotas for new VMs are automatically accepted by the API
|
79 |
return vm
|
80 |
commission_info = quotas.get_commission_info(vm, action=action, |
81 |
action_fields=job_fields) |
82 |
|
83 |
if vm.task_job_id == job_id and vm.serial is not None: |
84 |
# Commission for this change has already been issued. So just
|
85 |
# accept/reject it. Special case is OP_INSTANCE_CREATE, which even
|
86 |
# if fails, must be accepted, as the user must manually remove the
|
87 |
# failed server
|
88 |
serial = vm.serial |
89 |
if job_status == "success": |
90 |
quotas.accept_serial(serial) |
91 |
elif job_status in ["error", "canceled"]: |
92 |
log.debug("Job %s failed. Rejecting related serial %s", job_id,
|
93 |
serial) |
94 |
quotas.reject_serial(serial) |
95 |
vm.serial = None
|
96 |
elif job_status == "success" and commission_info is not None: |
97 |
log.debug("Expected job was %s. Processing job %s. Commission for"
|
98 |
" this job: %s", vm.task_job_id, job_id, commission_info)
|
99 |
# Commission for this change has not been issued, or the issued
|
100 |
# commission was unaware of the current change. Reject all previous
|
101 |
# commissions and create a new one in forced mode!
|
102 |
commission_name = ("client: dispatcher, resource: %s, ganeti_job: %s"
|
103 |
% (vm, job_id)) |
104 |
quotas.handle_resource_commission(vm, action, |
105 |
commission_info=commission_info, |
106 |
commission_name=commission_name, |
107 |
force=True,
|
108 |
auto_accept=True)
|
109 |
log.debug("Issued new commission: %s", vm.serial)
|
110 |
|
111 |
return vm
|
112 |
|
113 |
|
114 |
@transaction.commit_on_success
|
115 |
def process_op_status(vm, etime, jobid, opcode, status, logmsg, nics=None, |
116 |
beparams=None):
|
117 |
"""Process a job progress notification from the backend
|
118 |
|
119 |
Process an incoming message from the backend (currently Ganeti).
|
120 |
Job notifications with a terminating status (sucess, error, or canceled),
|
121 |
also update the operating state of the VM.
|
122 |
|
123 |
"""
|
124 |
# See #1492, #1031, #1111 why this line has been removed
|
125 |
#if (opcode not in [x[0] for x in VirtualMachine.BACKEND_OPCODES] or
|
126 |
if status not in [x[0] for x in BACKEND_STATUSES]: |
127 |
raise VirtualMachine.InvalidBackendMsgError(opcode, status)
|
128 |
|
129 |
vm.backendjobid = jobid |
130 |
vm.backendjobstatus = status |
131 |
vm.backendopcode = opcode |
132 |
vm.backendlogmsg = logmsg |
133 |
|
134 |
if status in ["queued", "waiting", "running"]: |
135 |
vm.save() |
136 |
return
|
137 |
|
138 |
state_for_success = VirtualMachine.OPER_STATE_FROM_OPCODE.get(opcode) |
139 |
|
140 |
# Notifications of success change the operating state
|
141 |
if status == "success": |
142 |
if state_for_success is not None: |
143 |
vm.operstate = state_for_success |
144 |
if beparams:
|
145 |
# Change the flavor of the VM
|
146 |
_process_resize(vm, beparams) |
147 |
# Update backendtime only for jobs that have been successfully
|
148 |
# completed, since only these jobs update the state of the VM. Else a
|
149 |
# "race condition" may occur when a successful job (e.g.
|
150 |
# OP_INSTANCE_REMOVE) completes before an error job and messages arrive
|
151 |
# in reversed order.
|
152 |
vm.backendtime = etime |
153 |
|
154 |
if status in ["success", "error", "canceled"] and nics is not None: |
155 |
# Update the NICs of the VM
|
156 |
_process_net_status(vm, etime, nics) |
157 |
|
158 |
# Special case: if OP_INSTANCE_CREATE fails --> ERROR
|
159 |
if opcode == 'OP_INSTANCE_CREATE' and status in ('canceled', 'error'): |
160 |
vm.operstate = 'ERROR'
|
161 |
vm.backendtime = etime |
162 |
elif opcode == 'OP_INSTANCE_REMOVE': |
163 |
# Special case: OP_INSTANCE_REMOVE fails for machines in ERROR,
|
164 |
# when no instance exists at the Ganeti backend.
|
165 |
# See ticket #799 for all the details.
|
166 |
if status == 'success' or (status == 'error' and |
167 |
not vm_exists_in_backend(vm)):
|
168 |
# VM has been deleted. Release the instance IPs
|
169 |
release_instance_ips(vm, []) |
170 |
# And delete the releated NICs (must be performed after release!)
|
171 |
vm.nics.all().delete() |
172 |
vm.deleted = True
|
173 |
vm.operstate = state_for_success |
174 |
vm.backendtime = etime |
175 |
status = "success"
|
176 |
|
177 |
if status in ["success", "error", "canceled"]: |
178 |
# Job is finalized: Handle quotas/commissioning
|
179 |
job_fields = {"nics": nics, "beparams": beparams} |
180 |
vm = handle_vm_quotas(vm, job_id=jobid, job_opcode=opcode, |
181 |
job_status=status, job_fields=job_fields) |
182 |
# and clear task fields
|
183 |
if vm.task_job_id == jobid:
|
184 |
vm.task = None
|
185 |
vm.task_job_id = None
|
186 |
|
187 |
vm.save() |
188 |
|
189 |
|
190 |
def _process_resize(vm, beparams): |
191 |
"""Change flavor of a VirtualMachine based on new beparams."""
|
192 |
old_flavor = vm.flavor |
193 |
vcpus = beparams.get("vcpus", old_flavor.cpu)
|
194 |
ram = beparams.get("maxmem", old_flavor.ram)
|
195 |
if vcpus == old_flavor.cpu and ram == old_flavor.ram: |
196 |
return
|
197 |
try:
|
198 |
new_flavor = Flavor.objects.get(cpu=vcpus, ram=ram, |
199 |
disk=old_flavor.disk, |
200 |
disk_template=old_flavor.disk_template) |
201 |
except Flavor.DoesNotExist:
|
202 |
raise Exception("Can not find flavor for VM") |
203 |
vm.flavor = new_flavor |
204 |
vm.save() |
205 |
|
206 |
|
207 |
@transaction.commit_on_success
|
208 |
def process_net_status(vm, etime, nics): |
209 |
"""Wrap _process_net_status inside transaction."""
|
210 |
_process_net_status(vm, etime, nics) |
211 |
|
212 |
|
213 |
def _process_net_status(vm, etime, nics): |
214 |
"""Process a net status notification from the backend
|
215 |
|
216 |
Process an incoming message from the Ganeti backend,
|
217 |
detailing the NIC configuration of a VM instance.
|
218 |
|
219 |
Update the state of the VM in the DB accordingly.
|
220 |
"""
|
221 |
|
222 |
ganeti_nics = process_ganeti_nics(nics) |
223 |
if not nics_changed(vm.nics.order_by('index'), ganeti_nics): |
224 |
log.debug("NICs for VM %s have not changed", vm)
|
225 |
return
|
226 |
|
227 |
# Get X-Lock on backend before getting X-Lock on network IP pools, to
|
228 |
# guarantee that no deadlock will occur with Backend allocator.
|
229 |
Backend.objects.select_for_update().get(id=vm.backend_id) |
230 |
|
231 |
# NICs have changed. Release the instance IPs
|
232 |
release_instance_ips(vm, ganeti_nics) |
233 |
# And delete the releated NICs (must be performed after release!)
|
234 |
vm.nics.all().delete() |
235 |
|
236 |
for nic in ganeti_nics: |
237 |
ipv4 = nic["ipv4"]
|
238 |
net = nic['network']
|
239 |
if ipv4:
|
240 |
net.reserve_address(ipv4) |
241 |
|
242 |
nic['dirty'] = False |
243 |
vm.nics.create(**nic) |
244 |
# Dummy save the network, because UI uses changed-since for VMs
|
245 |
# and Networks in order to show the VM NICs
|
246 |
net.save() |
247 |
|
248 |
vm.backendtime = etime |
249 |
vm.save() |
250 |
|
251 |
|
252 |
def process_ganeti_nics(ganeti_nics): |
253 |
"""Process NIC dict from ganeti hooks."""
|
254 |
new_nics = [] |
255 |
for i, new_nic in enumerate(ganeti_nics): |
256 |
network = new_nic.get('network', '') |
257 |
n = str(network)
|
258 |
pk = utils.id_from_network_name(n) |
259 |
|
260 |
net = Network.objects.get(pk=pk) |
261 |
|
262 |
# Get the new nic info
|
263 |
mac = new_nic.get('mac')
|
264 |
ipv4 = new_nic.get('ip')
|
265 |
ipv6 = mac2eui64(mac, net.subnet6) if net.subnet6 is not None else None |
266 |
|
267 |
firewall = new_nic.get('firewall')
|
268 |
firewall_profile = _reverse_tags.get(firewall) |
269 |
if not firewall_profile and net.public: |
270 |
firewall_profile = settings.DEFAULT_FIREWALL_PROFILE |
271 |
|
272 |
nic = { |
273 |
'index': i,
|
274 |
'network': net,
|
275 |
'mac': mac,
|
276 |
'ipv4': ipv4,
|
277 |
'ipv6': ipv6,
|
278 |
'firewall_profile': firewall_profile,
|
279 |
'state': 'ACTIVE'} |
280 |
|
281 |
new_nics.append(nic) |
282 |
return new_nics
|
283 |
|
284 |
|
285 |
def nics_changed(old_nics, new_nics): |
286 |
"""Return True if NICs have changed in any way."""
|
287 |
if len(old_nics) != len(new_nics): |
288 |
return True |
289 |
fields = ["ipv4", "ipv6", "mac", "firewall_profile", "index", "network"] |
290 |
for old_nic, new_nic in zip(old_nics, new_nics): |
291 |
for field in fields: |
292 |
if getattr(old_nic, field) != new_nic[field]: |
293 |
return True |
294 |
return False |
295 |
|
296 |
|
297 |
def release_instance_ips(vm, ganeti_nics): |
298 |
old_addresses = set(vm.nics.values_list("network", "ipv4")) |
299 |
new_addresses = set(map(lambda nic: (nic["network"].id, nic["ipv4"]), |
300 |
ganeti_nics)) |
301 |
to_release = old_addresses - new_addresses |
302 |
for (network_id, ipv4) in to_release: |
303 |
if ipv4:
|
304 |
# Get X-Lock before searching floating IP, to exclusively search
|
305 |
# and release floating IP. Otherwise you may release a floating IP
|
306 |
# that has been just reserved.
|
307 |
net = Network.objects.select_for_update().get(id=network_id) |
308 |
if net.floating_ip_pool:
|
309 |
try:
|
310 |
floating_ip = net.floating_ips.select_for_update()\ |
311 |
.get(ipv4=ipv4, machine=vm, |
312 |
deleted=False)
|
313 |
floating_ip.machine = None
|
314 |
floating_ip.save() |
315 |
except FloatingIP.DoesNotExist:
|
316 |
net.release_address(ipv4) |
317 |
else:
|
318 |
net.release_address(ipv4) |
319 |
|
320 |
|
321 |
@transaction.commit_on_success
|
322 |
def process_network_status(back_network, etime, jobid, opcode, status, logmsg): |
323 |
if status not in [x[0] for x in BACKEND_STATUSES]: |
324 |
raise Network.InvalidBackendMsgError(opcode, status)
|
325 |
|
326 |
back_network.backendjobid = jobid |
327 |
back_network.backendjobstatus = status |
328 |
back_network.backendopcode = opcode |
329 |
back_network.backendlogmsg = logmsg |
330 |
|
331 |
network = back_network.network |
332 |
|
333 |
# Notifications of success change the operating state
|
334 |
state_for_success = BackendNetwork.OPER_STATE_FROM_OPCODE.get(opcode, None)
|
335 |
if status == 'success' and state_for_success is not None: |
336 |
back_network.operstate = state_for_success |
337 |
|
338 |
if status in ('canceled', 'error') and opcode == 'OP_NETWORK_ADD': |
339 |
back_network.operstate = 'ERROR'
|
340 |
back_network.backendtime = etime |
341 |
|
342 |
if opcode == 'OP_NETWORK_REMOVE': |
343 |
network_is_deleted = (status == "success")
|
344 |
if network_is_deleted or (status == "error" and not |
345 |
network_exists_in_backend(back_network)): |
346 |
back_network.operstate = state_for_success |
347 |
back_network.deleted = True
|
348 |
back_network.backendtime = etime |
349 |
|
350 |
if status == 'success': |
351 |
back_network.backendtime = etime |
352 |
back_network.save() |
353 |
# Also you must update the state of the Network!!
|
354 |
update_network_state(network) |
355 |
|
356 |
|
357 |
def update_network_state(network): |
358 |
"""Update the state of a Network based on BackendNetwork states.
|
359 |
|
360 |
Update the state of a Network based on the operstate of the networks in the
|
361 |
backends that network exists.
|
362 |
|
363 |
The state of the network is:
|
364 |
* ACTIVE: If it is 'ACTIVE' in at least one backend.
|
365 |
* DELETED: If it is is 'DELETED' in all backends that have been created.
|
366 |
|
367 |
This function also releases the resources (MAC prefix or Bridge) and the
|
368 |
quotas for the network.
|
369 |
|
370 |
"""
|
371 |
if network.deleted:
|
372 |
# Network has already been deleted. Just assert that state is also
|
373 |
# DELETED
|
374 |
if not network.state == "DELETED": |
375 |
network.state = "DELETED"
|
376 |
network.save() |
377 |
return
|
378 |
|
379 |
backend_states = [s.operstate for s in network.backend_networks.all()] |
380 |
if not backend_states and network.action != "DESTROY": |
381 |
if network.state != "ACTIVE": |
382 |
network.state = "ACTIVE"
|
383 |
network.save() |
384 |
return
|
385 |
|
386 |
# Network is deleted when all BackendNetworks go to "DELETED" operstate
|
387 |
deleted = reduce(lambda x, y: x == y and "DELETED", backend_states, |
388 |
"DELETED")
|
389 |
|
390 |
# Release the resources on the deletion of the Network
|
391 |
if deleted:
|
392 |
log.info("Network %r deleted. Releasing link %r mac_prefix %r",
|
393 |
network.id, network.mac_prefix, network.link) |
394 |
network.deleted = True
|
395 |
network.state = "DELETED"
|
396 |
if network.mac_prefix:
|
397 |
if network.FLAVORS[network.flavor]["mac_prefix"] == "pool": |
398 |
release_resource(res_type="mac_prefix",
|
399 |
value=network.mac_prefix) |
400 |
if network.link:
|
401 |
if network.FLAVORS[network.flavor]["link"] == "pool": |
402 |
release_resource(res_type="bridge", value=network.link)
|
403 |
|
404 |
# Issue commission
|
405 |
if network.userid:
|
406 |
quotas.issue_and_accept_commission(network, delete=True)
|
407 |
# the above has already saved the object and committed;
|
408 |
# a second save would override others' changes, since the
|
409 |
# object is now unlocked
|
410 |
return
|
411 |
elif not network.public: |
412 |
log.warning("Network %s does not have an owner!", network.id)
|
413 |
network.save() |
414 |
|
415 |
|
416 |
@transaction.commit_on_success
|
417 |
def process_network_modify(back_network, etime, jobid, opcode, status, |
418 |
add_reserved_ips): |
419 |
assert (opcode == "OP_NETWORK_SET_PARAMS") |
420 |
if status not in [x[0] for x in BACKEND_STATUSES]: |
421 |
raise Network.InvalidBackendMsgError(opcode, status)
|
422 |
|
423 |
back_network.backendjobid = jobid |
424 |
back_network.backendjobstatus = status |
425 |
back_network.opcode = opcode |
426 |
|
427 |
if add_reserved_ips:
|
428 |
net = back_network.network |
429 |
pool = net.get_pool() |
430 |
if add_reserved_ips:
|
431 |
for ip in add_reserved_ips: |
432 |
pool.reserve(ip, external=True)
|
433 |
pool.save() |
434 |
|
435 |
if status == 'success': |
436 |
back_network.backendtime = etime |
437 |
back_network.save() |
438 |
|
439 |
|
440 |
@transaction.commit_on_success
|
441 |
def process_create_progress(vm, etime, progress): |
442 |
|
443 |
percentage = int(progress)
|
444 |
|
445 |
# The percentage may exceed 100%, due to the way
|
446 |
# snf-image:copy-progress tracks bytes read by image handling processes
|
447 |
percentage = 100 if percentage > 100 else percentage |
448 |
if percentage < 0: |
449 |
raise ValueError("Percentage cannot be negative") |
450 |
|
451 |
# FIXME: log a warning here, see #1033
|
452 |
# if last_update > percentage:
|
453 |
# raise ValueError("Build percentage should increase monotonically " \
|
454 |
# "(old = %d, new = %d)" % (last_update, percentage))
|
455 |
|
456 |
# This assumes that no message of type 'ganeti-create-progress' is going to
|
457 |
# arrive once OP_INSTANCE_CREATE has succeeded for a Ganeti instance and
|
458 |
# the instance is STARTED. What if the two messages are processed by two
|
459 |
# separate dispatcher threads, and the 'ganeti-op-status' message for
|
460 |
# successful creation gets processed before the 'ganeti-create-progress'
|
461 |
# message? [vkoukis]
|
462 |
#
|
463 |
#if not vm.operstate == 'BUILD':
|
464 |
# raise VirtualMachine.IllegalState("VM is not in building state")
|
465 |
|
466 |
vm.buildpercentage = percentage |
467 |
vm.backendtime = etime |
468 |
vm.save() |
469 |
|
470 |
|
471 |
@transaction.commit_on_success
|
472 |
def create_instance_diagnostic(vm, message, source, level="DEBUG", etime=None, |
473 |
details=None):
|
474 |
"""
|
475 |
Create virtual machine instance diagnostic entry.
|
476 |
|
477 |
:param vm: VirtualMachine instance to create diagnostic for.
|
478 |
:param message: Diagnostic message.
|
479 |
:param source: Diagnostic source identifier (e.g. image-helper).
|
480 |
:param level: Diagnostic level (`DEBUG`, `INFO`, `WARNING`, `ERROR`).
|
481 |
:param etime: The time the message occured (if available).
|
482 |
:param details: Additional details or debug information.
|
483 |
"""
|
484 |
VirtualMachineDiagnostic.objects.create_for_vm(vm, level, source=source, |
485 |
source_date=etime, |
486 |
message=message, |
487 |
details=details) |
488 |
|
489 |
|
490 |
def create_instance(vm, nics, flavor, image): |
491 |
"""`image` is a dictionary which should contain the keys:
|
492 |
'backend_id', 'format' and 'metadata'
|
493 |
|
494 |
metadata value should be a dictionary.
|
495 |
"""
|
496 |
|
497 |
# Handle arguments to CreateInstance() as a dictionary,
|
498 |
# initialize it based on a deployment-specific value.
|
499 |
# This enables the administrator to override deployment-specific
|
500 |
# arguments, such as the disk template to use, name of os provider
|
501 |
# and hypervisor-specific parameters at will (see Synnefo #785, #835).
|
502 |
#
|
503 |
kw = vm.backend.get_create_params() |
504 |
kw['mode'] = 'create' |
505 |
kw['name'] = vm.backend_vm_id
|
506 |
# Defined in settings.GANETI_CREATEINSTANCE_KWARGS
|
507 |
|
508 |
kw['disk_template'] = flavor.disk_template
|
509 |
kw['disks'] = [{"size": flavor.disk * 1024}] |
510 |
provider = flavor.disk_provider |
511 |
if provider:
|
512 |
kw['disks'][0]['provider'] = provider |
513 |
kw['disks'][0]['origin'] = flavor.disk_origin |
514 |
|
515 |
kw['nics'] = [{"network": nic.network.backend_id, "ip": nic.ipv4} |
516 |
for nic in nics] |
517 |
backend = vm.backend |
518 |
depend_jobs = [] |
519 |
for nic in nics: |
520 |
network = Network.objects.select_for_update().get(id=nic.network.id) |
521 |
bnet, created = BackendNetwork.objects.get_or_create(backend=backend, |
522 |
network=network) |
523 |
if bnet.operstate != "ACTIVE": |
524 |
if network.public:
|
525 |
msg = "Can not connect instance to network %s. Network is not"\
|
526 |
" ACTIVE in backend %s." % (network, backend)
|
527 |
raise Exception(msg) |
528 |
else:
|
529 |
jobs = create_network(network, backend, connect=True)
|
530 |
if isinstance(jobs, list): |
531 |
depend_jobs.extend(jobs) |
532 |
else:
|
533 |
depend_jobs.append(jobs) |
534 |
kw["depends"] = [[job, ["success", "error", "canceled"]] |
535 |
for job in depend_jobs] |
536 |
|
537 |
if vm.backend.use_hotplug():
|
538 |
kw['hotplug'] = True |
539 |
# Defined in settings.GANETI_CREATEINSTANCE_KWARGS
|
540 |
# kw['os'] = settings.GANETI_OS_PROVIDER
|
541 |
kw['ip_check'] = False |
542 |
kw['name_check'] = False |
543 |
|
544 |
# Do not specific a node explicitly, have
|
545 |
# Ganeti use an iallocator instead
|
546 |
#kw['pnode'] = rapi.GetNodes()[0]
|
547 |
|
548 |
kw['dry_run'] = settings.TEST
|
549 |
|
550 |
kw['beparams'] = {
|
551 |
'auto_balance': True, |
552 |
'vcpus': flavor.cpu,
|
553 |
'memory': flavor.ram}
|
554 |
|
555 |
kw['osparams'] = {
|
556 |
'config_url': vm.config_url,
|
557 |
# Store image id and format to Ganeti
|
558 |
'img_id': image['backend_id'], |
559 |
'img_format': image['format']} |
560 |
|
561 |
# Use opportunistic locking
|
562 |
kw['opportunistic_locking'] = True |
563 |
|
564 |
# Defined in settings.GANETI_CREATEINSTANCE_KWARGS
|
565 |
# kw['hvparams'] = dict(serial_console=False)
|
566 |
|
567 |
log.debug("Creating instance %s", utils.hide_pass(kw))
|
568 |
with pooled_rapi_client(vm) as client: |
569 |
return client.CreateInstance(**kw)
|
570 |
|
571 |
|
572 |
def delete_instance(vm): |
573 |
with pooled_rapi_client(vm) as client: |
574 |
return client.DeleteInstance(vm.backend_vm_id, dry_run=settings.TEST)
|
575 |
|
576 |
|
577 |
def reboot_instance(vm, reboot_type): |
578 |
assert reboot_type in ('soft', 'hard') |
579 |
kwargs = {"instance": vm.backend_vm_id,
|
580 |
"reboot_type": "hard"} |
581 |
# XXX: Currently shutdown_timeout parameter is not supported from the
|
582 |
# Ganeti RAPI. Until supported, we will fallback for both reboot types
|
583 |
# to the default shutdown timeout of Ganeti (120s). Note that reboot
|
584 |
# type of Ganeti job must be always hard. The 'soft' and 'hard' type
|
585 |
# of OS API is different from the one in Ganeti, and maps to
|
586 |
# 'shutdown_timeout'.
|
587 |
#if reboot_type == "hard":
|
588 |
# kwargs["shutdown_timeout"] = 0
|
589 |
if settings.TEST:
|
590 |
kwargs["dry_run"] = True |
591 |
with pooled_rapi_client(vm) as client: |
592 |
return client.RebootInstance(**kwargs)
|
593 |
|
594 |
|
595 |
def startup_instance(vm): |
596 |
with pooled_rapi_client(vm) as client: |
597 |
return client.StartupInstance(vm.backend_vm_id, dry_run=settings.TEST)
|
598 |
|
599 |
|
600 |
def shutdown_instance(vm): |
601 |
with pooled_rapi_client(vm) as client: |
602 |
return client.ShutdownInstance(vm.backend_vm_id, dry_run=settings.TEST)
|
603 |
|
604 |
|
605 |
def resize_instance(vm, vcpus, memory): |
606 |
beparams = {"vcpus": int(vcpus), |
607 |
"minmem": int(memory), |
608 |
"maxmem": int(memory)} |
609 |
with pooled_rapi_client(vm) as client: |
610 |
return client.ModifyInstance(vm.backend_vm_id, beparams=beparams)
|
611 |
|
612 |
|
613 |
def get_instance_console(vm): |
614 |
# RAPI GetInstanceConsole() returns endpoints to the vnc_bind_address,
|
615 |
# which is a cluster-wide setting, either 0.0.0.0 or 127.0.0.1, and pretty
|
616 |
# useless (see #783).
|
617 |
#
|
618 |
# Until this is fixed on the Ganeti side, construct a console info reply
|
619 |
# directly.
|
620 |
#
|
621 |
# WARNING: This assumes that VNC runs on port network_port on
|
622 |
# the instance's primary node, and is probably
|
623 |
# hypervisor-specific.
|
624 |
#
|
625 |
log.debug("Getting console for vm %s", vm)
|
626 |
|
627 |
console = {} |
628 |
console['kind'] = 'vnc' |
629 |
|
630 |
with pooled_rapi_client(vm) as client: |
631 |
i = client.GetInstance(vm.backend_vm_id) |
632 |
|
633 |
if vm.backend.hypervisor == "kvm" and i['hvparams']['serial_console']: |
634 |
raise Exception("hv parameter serial_console cannot be true") |
635 |
console['host'] = i['pnode'] |
636 |
console['port'] = i['network_port'] |
637 |
|
638 |
return console
|
639 |
|
640 |
|
641 |
def get_instance_info(vm): |
642 |
with pooled_rapi_client(vm) as client: |
643 |
return client.GetInstance(vm.backend_vm_id)
|
644 |
|
645 |
|
646 |
def vm_exists_in_backend(vm): |
647 |
try:
|
648 |
get_instance_info(vm) |
649 |
return True |
650 |
except GanetiApiError as e: |
651 |
if e.code == 404: |
652 |
return False |
653 |
raise e
|
654 |
|
655 |
|
656 |
def get_network_info(backend_network): |
657 |
with pooled_rapi_client(backend_network) as client: |
658 |
return client.GetNetwork(backend_network.network.backend_id)
|
659 |
|
660 |
|
661 |
def network_exists_in_backend(backend_network): |
662 |
try:
|
663 |
get_network_info(backend_network) |
664 |
return True |
665 |
except GanetiApiError as e: |
666 |
if e.code == 404: |
667 |
return False |
668 |
|
669 |
|
670 |
def create_network(network, backend, connect=True): |
671 |
"""Create a network in a Ganeti backend"""
|
672 |
log.debug("Creating network %s in backend %s", network, backend)
|
673 |
|
674 |
job_id = _create_network(network, backend) |
675 |
|
676 |
if connect:
|
677 |
job_ids = connect_network(network, backend, depends=[job_id]) |
678 |
return job_ids
|
679 |
else:
|
680 |
return [job_id]
|
681 |
|
682 |
|
683 |
def _create_network(network, backend): |
684 |
"""Create a network."""
|
685 |
|
686 |
tags = network.backend_tag |
687 |
if network.dhcp:
|
688 |
tags.append('nfdhcpd')
|
689 |
|
690 |
if network.public:
|
691 |
conflicts_check = True
|
692 |
else:
|
693 |
conflicts_check = False
|
694 |
|
695 |
# Use a dummy network subnet for IPv6 only networks. Currently Ganeti does
|
696 |
# not support IPv6 only networks. To bypass this limitation, we create the
|
697 |
# network with a dummy network subnet, and make Cyclades connect instances
|
698 |
# to such networks, with address=None.
|
699 |
subnet = network.subnet |
700 |
if subnet is None: |
701 |
subnet = "10.0.0.0/24"
|
702 |
|
703 |
try:
|
704 |
bn = BackendNetwork.objects.get(network=network, backend=backend) |
705 |
mac_prefix = bn.mac_prefix |
706 |
except BackendNetwork.DoesNotExist:
|
707 |
raise Exception("BackendNetwork for network '%s' in backend '%s'" |
708 |
" does not exist" % (network.id, backend.id))
|
709 |
|
710 |
with pooled_rapi_client(backend) as client: |
711 |
return client.CreateNetwork(network_name=network.backend_id,
|
712 |
network=subnet, |
713 |
network6=network.subnet6, |
714 |
gateway=network.gateway, |
715 |
gateway6=network.gateway6, |
716 |
mac_prefix=mac_prefix, |
717 |
conflicts_check=conflicts_check, |
718 |
tags=tags) |
719 |
|
720 |
|
721 |
def connect_network(network, backend, depends=[], group=None): |
722 |
"""Connect a network to nodegroups."""
|
723 |
log.debug("Connecting network %s to backend %s", network, backend)
|
724 |
|
725 |
if network.public:
|
726 |
conflicts_check = True
|
727 |
else:
|
728 |
conflicts_check = False
|
729 |
|
730 |
depends = [[job, ["success", "error", "canceled"]] for job in depends] |
731 |
with pooled_rapi_client(backend) as client: |
732 |
groups = [group] if group is not None else client.GetGroups() |
733 |
job_ids = [] |
734 |
for group in groups: |
735 |
job_id = client.ConnectNetwork(network.backend_id, group, |
736 |
network.mode, network.link, |
737 |
conflicts_check, |
738 |
depends=depends) |
739 |
job_ids.append(job_id) |
740 |
return job_ids
|
741 |
|
742 |
|
743 |
def delete_network(network, backend, disconnect=True): |
744 |
log.debug("Deleting network %s from backend %s", network, backend)
|
745 |
|
746 |
depends = [] |
747 |
if disconnect:
|
748 |
depends = disconnect_network(network, backend) |
749 |
_delete_network(network, backend, depends=depends) |
750 |
|
751 |
|
752 |
def _delete_network(network, backend, depends=[]): |
753 |
depends = [[job, ["success", "error", "canceled"]] for job in depends] |
754 |
with pooled_rapi_client(backend) as client: |
755 |
return client.DeleteNetwork(network.backend_id, depends)
|
756 |
|
757 |
|
758 |
def disconnect_network(network, backend, group=None): |
759 |
log.debug("Disconnecting network %s to backend %s", network, backend)
|
760 |
|
761 |
with pooled_rapi_client(backend) as client: |
762 |
groups = [group] if group is not None else client.GetGroups() |
763 |
job_ids = [] |
764 |
for group in groups: |
765 |
job_id = client.DisconnectNetwork(network.backend_id, group) |
766 |
job_ids.append(job_id) |
767 |
return job_ids
|
768 |
|
769 |
|
770 |
def connect_to_network(vm, nic): |
771 |
network = nic.network |
772 |
backend = vm.backend |
773 |
network = Network.objects.select_for_update().get(id=network.id) |
774 |
bnet, created = BackendNetwork.objects.get_or_create(backend=backend, |
775 |
network=network) |
776 |
depend_jobs = [] |
777 |
if bnet.operstate != "ACTIVE": |
778 |
depend_jobs = create_network(network, backend, connect=True)
|
779 |
|
780 |
depends = [[job, ["success", "error", "canceled"]] for job in depend_jobs] |
781 |
|
782 |
nic = {'ip': nic.ipv4, 'network': network.backend_id} |
783 |
|
784 |
log.debug("Connecting NIC %s to VM %s", nic, vm)
|
785 |
|
786 |
kwargs = { |
787 |
"instance": vm.backend_vm_id,
|
788 |
"nics": [("add", nic)], |
789 |
"depends": depends,
|
790 |
} |
791 |
if vm.backend.use_hotplug():
|
792 |
kwargs["hotplug"] = True |
793 |
if settings.TEST:
|
794 |
kwargs["dry_run"] = True |
795 |
|
796 |
with pooled_rapi_client(vm) as client: |
797 |
return client.ModifyInstance(**kwargs)
|
798 |
|
799 |
|
800 |
def disconnect_from_network(vm, nic): |
801 |
log.debug("Removing nic of VM %s, with index %s", vm, str(nic.index)) |
802 |
|
803 |
kwargs = { |
804 |
"instance": vm.backend_vm_id,
|
805 |
"nics": [("remove", nic.index, {})], |
806 |
} |
807 |
if vm.backend.use_hotplug():
|
808 |
kwargs["hotplug"] = True |
809 |
if settings.TEST:
|
810 |
kwargs["dry_run"] = True |
811 |
|
812 |
with pooled_rapi_client(vm) as client: |
813 |
jobID = client.ModifyInstance(**kwargs) |
814 |
# If the NIC has a tag for a firewall profile it must be deleted,
|
815 |
# otherwise it may affect another NIC. XXX: Deleting the tag should
|
816 |
# depend on the removing the NIC, but currently RAPI client does not
|
817 |
# support this, this may result in clearing the firewall profile
|
818 |
# without successfully removing the NIC. This issue will be fixed with
|
819 |
# use of NIC UUIDs.
|
820 |
firewall_profile = nic.firewall_profile |
821 |
if firewall_profile and firewall_profile != "DISABLED": |
822 |
tag = _firewall_tags[firewall_profile] % nic.index |
823 |
client.DeleteInstanceTags(vm.backend_vm_id, [tag], |
824 |
dry_run=settings.TEST) |
825 |
|
826 |
return jobID
|
827 |
|
828 |
|
829 |
def set_firewall_profile(vm, profile, index=0): |
830 |
try:
|
831 |
tag = _firewall_tags[profile] % index |
832 |
except KeyError: |
833 |
raise ValueError("Unsopported Firewall Profile: %s" % profile) |
834 |
|
835 |
log.debug("Setting tag of VM %s, NIC index %d, to %s", vm, index, profile)
|
836 |
|
837 |
with pooled_rapi_client(vm) as client: |
838 |
# Delete previous firewall tags
|
839 |
old_tags = client.GetInstanceTags(vm.backend_vm_id) |
840 |
delete_tags = [(t % index) for t in _firewall_tags.values() |
841 |
if (t % index) in old_tags] |
842 |
if delete_tags:
|
843 |
client.DeleteInstanceTags(vm.backend_vm_id, delete_tags, |
844 |
dry_run=settings.TEST) |
845 |
|
846 |
if profile != "DISABLED": |
847 |
client.AddInstanceTags(vm.backend_vm_id, [tag], |
848 |
dry_run=settings.TEST) |
849 |
|
850 |
# XXX NOP ModifyInstance call to force process_net_status to run
|
851 |
# on the dispatcher
|
852 |
os_name = settings.GANETI_CREATEINSTANCE_KWARGS['os']
|
853 |
client.ModifyInstance(vm.backend_vm_id, |
854 |
os_name=os_name) |
855 |
return None |
856 |
|
857 |
|
858 |
def get_instances(backend, bulk=True): |
859 |
with pooled_rapi_client(backend) as c: |
860 |
return c.GetInstances(bulk=bulk)
|
861 |
|
862 |
|
863 |
def get_nodes(backend, bulk=True): |
864 |
with pooled_rapi_client(backend) as c: |
865 |
return c.GetNodes(bulk=bulk)
|
866 |
|
867 |
|
868 |
def get_jobs(backend, bulk=True): |
869 |
with pooled_rapi_client(backend) as c: |
870 |
return c.GetJobs(bulk=bulk)
|
871 |
|
872 |
|
873 |
def get_physical_resources(backend): |
874 |
""" Get the physical resources of a backend.
|
875 |
|
876 |
Get the resources of a backend as reported by the backend (not the db).
|
877 |
|
878 |
"""
|
879 |
nodes = get_nodes(backend, bulk=True)
|
880 |
attr = ['mfree', 'mtotal', 'dfree', 'dtotal', 'pinst_cnt', 'ctotal'] |
881 |
res = {} |
882 |
for a in attr: |
883 |
res[a] = 0
|
884 |
for n in nodes: |
885 |
# Filter out drained, offline and not vm_capable nodes since they will
|
886 |
# not take part in the vm allocation process
|
887 |
can_host_vms = n['vm_capable'] and not (n['drained'] or n['offline']) |
888 |
if can_host_vms and n['cnodes']: |
889 |
for a in attr: |
890 |
res[a] += int(n[a])
|
891 |
return res
|
892 |
|
893 |
|
894 |
def update_backend_resources(backend, resources=None): |
895 |
""" Update the state of the backend resources in db.
|
896 |
|
897 |
"""
|
898 |
|
899 |
if not resources: |
900 |
resources = get_physical_resources(backend) |
901 |
|
902 |
backend.mfree = resources['mfree']
|
903 |
backend.mtotal = resources['mtotal']
|
904 |
backend.dfree = resources['dfree']
|
905 |
backend.dtotal = resources['dtotal']
|
906 |
backend.pinst_cnt = resources['pinst_cnt']
|
907 |
backend.ctotal = resources['ctotal']
|
908 |
backend.updated = datetime.now() |
909 |
backend.save() |
910 |
|
911 |
|
912 |
def get_memory_from_instances(backend): |
913 |
""" Get the memory that is used from instances.
|
914 |
|
915 |
Get the used memory of a backend. Note: This is different for
|
916 |
the real memory used, due to kvm's memory de-duplication.
|
917 |
|
918 |
"""
|
919 |
with pooled_rapi_client(backend) as client: |
920 |
instances = client.GetInstances(bulk=True)
|
921 |
mem = 0
|
922 |
for i in instances: |
923 |
mem += i['oper_ram']
|
924 |
return mem
|
925 |
|
926 |
|
927 |
def get_available_disk_templates(backend): |
928 |
"""Get the list of available disk templates of a Ganeti backend.
|
929 |
|
930 |
The list contains the disk templates that are enabled in the Ganeti backend
|
931 |
and also included in ipolicy-disk-templates.
|
932 |
|
933 |
"""
|
934 |
with pooled_rapi_client(backend) as c: |
935 |
info = c.GetInfo() |
936 |
ipolicy_disk_templates = info["ipolicy"]["disk-templates"] |
937 |
try:
|
938 |
enabled_disk_templates = info["enabled_disk_templates"]
|
939 |
return [dp for dp in enabled_disk_templates |
940 |
if dp in ipolicy_disk_templates] |
941 |
except KeyError: |
942 |
# Ganeti < 2.8 does not have 'enabled_disk_templates'
|
943 |
return ipolicy_disk_templates
|
944 |
|
945 |
|
946 |
def update_backend_disk_templates(backend): |
947 |
disk_templates = get_available_disk_templates(backend) |
948 |
backend.disk_templates = disk_templates |
949 |
backend.save() |
950 |
|
951 |
|
952 |
##
|
953 |
## Synchronized operations for reconciliation
|
954 |
##
|
955 |
|
956 |
|
957 |
def create_network_synced(network, backend): |
958 |
result = _create_network_synced(network, backend) |
959 |
if result[0] != 'success': |
960 |
return result
|
961 |
result = connect_network_synced(network, backend) |
962 |
return result
|
963 |
|
964 |
|
965 |
def _create_network_synced(network, backend): |
966 |
with pooled_rapi_client(backend) as client: |
967 |
job = _create_network(network, backend) |
968 |
result = wait_for_job(client, job) |
969 |
return result
|
970 |
|
971 |
|
972 |
def connect_network_synced(network, backend): |
973 |
with pooled_rapi_client(backend) as client: |
974 |
for group in client.GetGroups(): |
975 |
job = client.ConnectNetwork(network.backend_id, group, |
976 |
network.mode, network.link) |
977 |
result = wait_for_job(client, job) |
978 |
if result[0] != 'success': |
979 |
return result
|
980 |
|
981 |
return result
|
982 |
|
983 |
|
984 |
def wait_for_job(client, jobid): |
985 |
result = client.WaitForJobChange(jobid, ['status', 'opresult'], None, None) |
986 |
status = result['job_info'][0] |
987 |
while status not in ['success', 'error', 'cancel']: |
988 |
result = client.WaitForJobChange(jobid, ['status', 'opresult'], |
989 |
[result], None)
|
990 |
status = result['job_info'][0] |
991 |
|
992 |
if status == 'success': |
993 |
return (status, None) |
994 |
else:
|
995 |
error = result['job_info'][1] |
996 |
return (status, error)
|