root / snf-cyclades-app / synnefo / logic / reconciliation.py @ 6e9255ab
History | View | Annotate | Download (27.2 kB)
1 |
# -*- coding: utf-8 -*-
|
---|---|
2 |
#
|
3 |
# Copyright 2011-2013 GRNET S.A. All rights reserved.
|
4 |
#
|
5 |
# Redistribution and use in source and binary forms, with or
|
6 |
# without modification, are permitted provided that the following
|
7 |
# conditions are met:
|
8 |
#
|
9 |
# 1. Redistributions of source code must retain the above
|
10 |
# copyright notice, this list of conditions and the following
|
11 |
# disclaimer.
|
12 |
#
|
13 |
# 2. Redistributions in binary form must reproduce the above
|
14 |
# copyright notice, this list of conditions and the following
|
15 |
# disclaimer in the documentation and/or other materials
|
16 |
# provided with the distribution.
|
17 |
#
|
18 |
# THIS SOFTWARE IS PROVIDED BY GRNET S.A. ``AS IS'' AND ANY EXPRESS
|
19 |
# OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
|
20 |
# WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
21 |
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL GRNET S.A OR
|
22 |
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
23 |
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
24 |
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
|
25 |
# USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
|
26 |
# AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
27 |
# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
|
28 |
# ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
29 |
# POSSIBILITY OF SUCH DAMAGE.
|
30 |
#
|
31 |
# The views and conclusions contained in the software and
|
32 |
# documentation are those of the authors and should not be
|
33 |
# interpreted as representing official policies, either expressed
|
34 |
# or implied, of GRNET S.A.
|
35 |
#
|
36 |
"""Business logic for reconciliation
|
37 |
|
38 |
Reconcile the contents of the DB with the actual state of the
|
39 |
Ganeti backend.
|
40 |
|
41 |
Let D be the set of VMs in the DB, G the set of VMs in Ganeti.
|
42 |
RULES:
|
43 |
R1. Stale servers in DB:
|
44 |
For any v in D but not in G:
|
45 |
Set deleted=True.
|
46 |
R2. Orphan instances in Ganet:
|
47 |
For any v in G with deleted=True in D:
|
48 |
Issue OP_INSTANCE_DESTROY.
|
49 |
R3. Unsynced operstate:
|
50 |
For any v whose operating state differs between G and V:
|
51 |
Set the operating state in D based on the state in G.
|
52 |
In the code, D, G are Python dicts mapping instance ids to operating state.
|
53 |
For D, the operating state is chosen from VirtualMachine.OPER_STATES.
|
54 |
For G, the operating state is True if the machine is up, False otherwise.
|
55 |
|
56 |
"""
|
57 |
|
58 |
|
59 |
from django.core.management import setup_environ |
60 |
try:
|
61 |
from synnefo import settings |
62 |
except ImportError: |
63 |
raise Exception("Cannot import settings, make sure PYTHONPATH contains " |
64 |
"the parent directory of the Synnefo Django project.")
|
65 |
setup_environ(settings) |
66 |
|
67 |
|
68 |
import logging |
69 |
import itertools |
70 |
import bitarray |
71 |
from datetime import datetime, timedelta |
72 |
|
73 |
from django.db import transaction |
74 |
from synnefo.db.models import (Backend, VirtualMachine, Flavor, |
75 |
pooled_rapi_client, Network, |
76 |
BackendNetwork) |
77 |
from synnefo.db.pools import IPPool |
78 |
from synnefo.logic import utils, backend as backend_mod |
79 |
from synnefo.logic.rapi import GanetiApiError |
80 |
|
81 |
logger = logging.getLogger() |
82 |
logging.basicConfig() |
83 |
|
84 |
try:
|
85 |
CHECK_INTERVAL = settings.RECONCILIATION_CHECK_INTERVAL |
86 |
except AttributeError: |
87 |
CHECK_INTERVAL = 60
|
88 |
|
89 |
|
90 |
class BackendReconciler(object): |
91 |
def __init__(self, backend, logger, options=None): |
92 |
self.backend = backend
|
93 |
self.log = logger
|
94 |
self.client = backend.get_client()
|
95 |
if options is None: |
96 |
self.options = {}
|
97 |
else:
|
98 |
self.options = options
|
99 |
|
100 |
def close(self): |
101 |
self.backend.put_client(self.client) |
102 |
|
103 |
@transaction.commit_on_success
|
104 |
def reconcile(self): |
105 |
log = self.log
|
106 |
backend = self.backend
|
107 |
log.debug("Reconciling backend %s", backend)
|
108 |
|
109 |
self.db_servers = get_database_servers(backend)
|
110 |
self.db_servers_keys = set(self.db_servers.keys()) |
111 |
log.debug("Got servers info from database.")
|
112 |
|
113 |
self.gnt_servers = get_ganeti_servers(backend)
|
114 |
self.gnt_servers_keys = set(self.gnt_servers.keys()) |
115 |
log.debug("Got servers info from Ganeti backend.")
|
116 |
|
117 |
self.event_time = datetime.now()
|
118 |
|
119 |
self.stale_servers = self.reconcile_stale_servers() |
120 |
self.orphan_servers = self.reconcile_orphan_servers() |
121 |
self.unsynced_servers = self.reconcile_unsynced_servers() |
122 |
self.close()
|
123 |
|
124 |
def get_build_status(self, db_server): |
125 |
job = db_server.backendjobid |
126 |
if job is None: |
127 |
created = db_server.created |
128 |
# Job has not yet been enqueued.
|
129 |
if self.event_time < created + timedelta(seconds=60): |
130 |
return "RUNNING" |
131 |
else:
|
132 |
return "ERROR" |
133 |
else:
|
134 |
updated = db_server.backendtime |
135 |
if self.event_time >= updated + timedelta(seconds=60): |
136 |
try:
|
137 |
job_info = self.client.GetJobStatus(job_id=job)
|
138 |
finalized = ["success", "error", "cancelled"] |
139 |
if job_info["status"] == "error": |
140 |
return "ERROR" |
141 |
elif job_info["status"] not in finalized: |
142 |
return "RUNNING" |
143 |
else:
|
144 |
return "FINALIZED" |
145 |
except GanetiApiError:
|
146 |
return "ERROR" |
147 |
else:
|
148 |
self.log.debug("Pending build for server '%s'", db_server.id) |
149 |
return "RUNNING" |
150 |
|
151 |
def reconcile_stale_servers(self): |
152 |
# Detect stale servers
|
153 |
stale = [] |
154 |
stale_keys = self.db_servers_keys - self.gnt_servers_keys |
155 |
for server_id in stale_keys: |
156 |
db_server = self.db_servers[server_id]
|
157 |
if db_server.operstate == "BUILD": |
158 |
build_status = self.get_build_status(db_server)
|
159 |
if build_status == "ERROR": |
160 |
# Special handling of BUILD eerrors
|
161 |
self.reconcile_building_server(db_server)
|
162 |
elif build_status != "RUNNING": |
163 |
stale.append(server_id) |
164 |
else:
|
165 |
stale.append(server_id) |
166 |
|
167 |
# Report them
|
168 |
if stale:
|
169 |
self.log.info("Found stale servers %s at backend %s", |
170 |
", ".join(map(str, stale)), self.backend) |
171 |
else:
|
172 |
self.log.debug("No stale servers at backend %s", self.backend) |
173 |
|
174 |
# Fix them
|
175 |
if stale and self.options["fix_stale"]: |
176 |
for server_id in stale: |
177 |
db_server = self.db_servers[server_id]
|
178 |
backend_mod.process_op_status( |
179 |
vm=db_server, |
180 |
etime=self.event_time,
|
181 |
jobid=-0,
|
182 |
opcode='OP_INSTANCE_REMOVE', status='success', |
183 |
logmsg='Reconciliation: simulated Ganeti event')
|
184 |
self.log.debug("Simulated Ganeti removal for stale servers.") |
185 |
|
186 |
def reconcile_orphan_servers(self): |
187 |
orphans = self.gnt_servers_keys - self.db_servers_keys |
188 |
if orphans:
|
189 |
self.log.info("Found orphan servers %s at backend %s", |
190 |
", ".join(map(str, orphans)), self.backend) |
191 |
else:
|
192 |
self.log.debug("No orphan servers at backend %s", self.backend) |
193 |
|
194 |
if orphans and self.options["fix_orphans"]: |
195 |
for server_id in orphans: |
196 |
server_name = utils.id_to_instance_name(server_id) |
197 |
self.client.DeleteInstance(server_name)
|
198 |
self.log.debug("Issued OP_INSTANCE_REMOVE for orphan servers.") |
199 |
|
200 |
def reconcile_unsynced_servers(self): |
201 |
#log = self.log
|
202 |
for server_id in self.db_servers_keys & self.gnt_servers_keys: |
203 |
db_server = self.db_servers[server_id]
|
204 |
gnt_server = self.gnt_servers[server_id]
|
205 |
if db_server.operstate == "BUILD": |
206 |
build_status = self.get_build_status(db_server)
|
207 |
if build_status == "RUNNING": |
208 |
# Do not reconcile building VMs
|
209 |
continue
|
210 |
elif build_status == "ERROR": |
211 |
# Special handling of build errors
|
212 |
self.reconcile_building_server(db_server)
|
213 |
continue
|
214 |
|
215 |
self.reconcile_unsynced_operstate(server_id, db_server,
|
216 |
gnt_server) |
217 |
self.reconcile_unsynced_flavor(server_id, db_server,
|
218 |
gnt_server) |
219 |
self.reconcile_unsynced_nics(server_id, db_server, gnt_server)
|
220 |
self.reconcile_unsynced_disks(server_id, db_server, gnt_server)
|
221 |
|
222 |
def reconcile_building_server(self, db_server): |
223 |
self.log.info("Server '%s' is BUILD in DB, but 'ERROR' in Ganeti.", |
224 |
db_server.id) |
225 |
if self.options["fix_unsynced"]: |
226 |
fix_opcode = "OP_INSTANCE_CREATE"
|
227 |
backend_mod.process_op_status( |
228 |
vm=db_server, |
229 |
etime=self.event_time,
|
230 |
jobid=-0,
|
231 |
opcode=fix_opcode, status='error',
|
232 |
logmsg='Reconciliation: simulated Ganeti event')
|
233 |
self.log.debug("Simulated Ganeti error build event for" |
234 |
" server '%s'", db_server.id)
|
235 |
|
236 |
def reconcile_unsynced_operstate(self, server_id, db_server, gnt_server): |
237 |
if db_server.operstate != gnt_server["state"]: |
238 |
self.log.info("Server '%s' is '%s' in DB and '%s' in Ganeti.", |
239 |
server_id, db_server.operstate, gnt_server["state"])
|
240 |
if self.options["fix_unsynced"]: |
241 |
# If server is in building state, you will have first to
|
242 |
# reconcile it's creation, to avoid wrong quotas
|
243 |
if db_server.operstate == "BUILD": |
244 |
backend_mod.process_op_status( |
245 |
vm=db_server, etime=self.event_time, jobid=-0, |
246 |
opcode="OP_INSTANCE_CREATE", status='success', |
247 |
logmsg='Reconciliation: simulated Ganeti event')
|
248 |
fix_opcode = "OP_INSTANCE_STARTUP"\
|
249 |
if gnt_server["state"] == "STARTED"\ |
250 |
else "OP_INSTANCE_SHUTDOWN" |
251 |
backend_mod.process_op_status( |
252 |
vm=db_server, etime=self.event_time, jobid=-0, |
253 |
opcode=fix_opcode, status='success',
|
254 |
logmsg='Reconciliation: simulated Ganeti event')
|
255 |
self.log.debug("Simulated Ganeti state event for server '%s'", |
256 |
server_id) |
257 |
|
258 |
def reconcile_unsynced_flavor(self, server_id, db_server, gnt_server): |
259 |
db_flavor = db_server.flavor |
260 |
gnt_flavor = gnt_server["flavor"]
|
261 |
if (db_flavor.ram != gnt_flavor["ram"] or |
262 |
db_flavor.cpu != gnt_flavor["vcpus"]):
|
263 |
try:
|
264 |
gnt_flavor = Flavor.objects.get( |
265 |
ram=gnt_flavor["ram"],
|
266 |
cpu=gnt_flavor["vcpus"],
|
267 |
disk=db_flavor.disk, |
268 |
disk_template=db_flavor.disk_template) |
269 |
except Flavor.DoesNotExist:
|
270 |
self.log.warning("Server '%s' has unknown flavor.", server_id) |
271 |
return
|
272 |
|
273 |
self.log.info("Server '%s' has flavor '%' in DB and '%s' in" |
274 |
" Ganeti", server_id, db_flavor, gnt_flavor)
|
275 |
if self.options["fix_unsynced_flavors"]: |
276 |
old_state = db_server.operstate |
277 |
opcode = "OP_INSTANCE_SET_PARAMS"
|
278 |
beparams = {"vcpus": gnt_flavor.cpu,
|
279 |
"minmem": gnt_flavor.ram,
|
280 |
"maxmem": gnt_flavor.ram}
|
281 |
backend_mod.process_op_status( |
282 |
vm=db_server, etime=self.event_time, jobid=-0, |
283 |
opcode=opcode, status='success',
|
284 |
beparams=beparams, |
285 |
logmsg='Reconciliation: simulated Ganeti event')
|
286 |
# process_op_status with beparams will set the vmstate to
|
287 |
# shutdown. Fix this be returning it to old state
|
288 |
vm = VirtualMachine.objects.get(pk=server_id) |
289 |
vm.operstate = old_state |
290 |
vm.save() |
291 |
self.log.debug("Simulated Ganeti flavor event for server '%s'", |
292 |
server_id) |
293 |
|
294 |
def reconcile_unsynced_nics(self, server_id, db_server, gnt_server): |
295 |
db_nics = db_server.nics.order_by("index")
|
296 |
gnt_nics = gnt_server["nics"]
|
297 |
gnt_nics_parsed = backend_mod.process_ganeti_nics(gnt_nics) |
298 |
if backend_mod.nics_changed(db_nics, gnt_nics_parsed):
|
299 |
msg = "Found unsynced NICs for server '%s'.\n\t"\
|
300 |
"DB: %s\n\tGaneti: %s"
|
301 |
db_nics_str = ", ".join(map(format_db_nic, db_nics)) |
302 |
gnt_nics_str = ", ".join(map(format_gnt_nic, gnt_nics_parsed)) |
303 |
self.log.info(msg, server_id, db_nics_str, gnt_nics_str)
|
304 |
if self.options["fix_unsynced_nics"]: |
305 |
backend_mod.process_net_status(vm=db_server, |
306 |
etime=self.event_time,
|
307 |
nics=gnt_nics) |
308 |
|
309 |
def reconcile_unsynced_disks(self, server_id, db_server, gnt_server): |
310 |
pass
|
311 |
|
312 |
|
313 |
def format_db_nic(nic): |
314 |
return "Index: %s IP: %s Network: %s MAC: %s Firewall: %s" % (nic.index, |
315 |
nic.ipv4, nic.network_id, nic.mac, nic.firewall_profile) |
316 |
|
317 |
|
318 |
def format_gnt_nic(nic): |
319 |
return "Index: %s IP: %s Network: %s MAC: %s Firewall: %s" %\ |
320 |
(nic["index"], nic["ipv4"], nic["network"], nic["mac"], |
321 |
nic["firewall_profile"])
|
322 |
|
323 |
|
324 |
#
|
325 |
# Networks
|
326 |
#
|
327 |
|
328 |
|
329 |
def get_networks_from_ganeti(backend): |
330 |
prefix = settings.BACKEND_PREFIX_ID + 'net-'
|
331 |
|
332 |
networks = {} |
333 |
with pooled_rapi_client(backend) as c: |
334 |
for net in c.GetNetworks(bulk=True): |
335 |
if net['name'].startswith(prefix): |
336 |
id = utils.id_from_network_name(net['name'])
|
337 |
networks[id] = net
|
338 |
|
339 |
return networks
|
340 |
|
341 |
|
342 |
def hanging_networks(backend, GNets): |
343 |
"""Get networks that are not connected to all Nodegroups.
|
344 |
|
345 |
"""
|
346 |
def get_network_groups(group_list): |
347 |
groups = set()
|
348 |
for g in group_list: |
349 |
g_name = g.split('(')[0] |
350 |
groups.add(g_name) |
351 |
return groups
|
352 |
|
353 |
with pooled_rapi_client(backend) as c: |
354 |
groups = set(c.GetGroups())
|
355 |
|
356 |
hanging = {} |
357 |
for id, info in GNets.items(): |
358 |
group_list = get_network_groups(info['group_list'])
|
359 |
if group_list != groups:
|
360 |
hanging[id] = groups - group_list
|
361 |
return hanging
|
362 |
|
363 |
|
364 |
def get_online_backends(): |
365 |
return Backend.objects.filter(offline=False) |
366 |
|
367 |
|
368 |
def get_database_servers(backend): |
369 |
servers = backend.virtual_machines.select_related("nics", "flavor")\ |
370 |
.filter(deleted=False)
|
371 |
return dict([(s.id, s) for s in servers]) |
372 |
|
373 |
|
374 |
def get_ganeti_servers(backend): |
375 |
gnt_instances = backend_mod.get_instances(backend) |
376 |
# Filter out non-synnefo instances
|
377 |
snf_backend_prefix = settings.BACKEND_PREFIX_ID |
378 |
gnt_instances = filter(lambda i: i["name"].startswith(snf_backend_prefix), |
379 |
gnt_instances) |
380 |
gnt_instances = map(parse_gnt_instance, gnt_instances)
|
381 |
return dict([(i["id"], i) for i in gnt_instances if i["id"] is not None]) |
382 |
|
383 |
|
384 |
def parse_gnt_instance(instance): |
385 |
try:
|
386 |
instance_id = utils.id_from_instance_name(instance['name'])
|
387 |
except Exception: |
388 |
logger.error("Ignoring instance with malformed name %s",
|
389 |
instance['name'])
|
390 |
return (None, None) |
391 |
|
392 |
beparams = instance["beparams"]
|
393 |
|
394 |
vcpus = beparams["vcpus"]
|
395 |
ram = beparams["maxmem"]
|
396 |
state = instance["oper_state"] and "STARTED" or "STOPPED" |
397 |
|
398 |
return {
|
399 |
"id": instance_id,
|
400 |
"state": state, # FIX |
401 |
"updated": datetime.fromtimestamp(instance["mtime"]), |
402 |
"disks": disks_from_instance(instance),
|
403 |
"nics": nics_from_instance(instance),
|
404 |
"flavor": {"vcpus": vcpus, |
405 |
"ram": ram},
|
406 |
"tags": instance["tags"] |
407 |
} |
408 |
|
409 |
|
410 |
def nics_from_instance(i): |
411 |
ips = zip(itertools.repeat('ip'), i['nic.ips']) |
412 |
macs = zip(itertools.repeat('mac'), i['nic.macs']) |
413 |
networks = zip(itertools.repeat('network'), i['nic.networks']) |
414 |
# modes = zip(itertools.repeat('mode'), i['nic.modes'])
|
415 |
# links = zip(itertools.repeat('link'), i['nic.links'])
|
416 |
# nics = zip(ips,macs,modes,networks,links)
|
417 |
nics = zip(ips, macs, networks)
|
418 |
nics = map(lambda x: dict(x), nics) |
419 |
#nics = dict(enumerate(nics))
|
420 |
tags = i["tags"]
|
421 |
for tag in tags: |
422 |
t = tag.split(":")
|
423 |
if t[0:2] == ["synnefo", "network"]: |
424 |
if len(t) != 4: |
425 |
logger.error("Malformed synefo tag %s", tag)
|
426 |
continue
|
427 |
try:
|
428 |
index = int(t[2]) |
429 |
nics[index]['firewall'] = t[3] |
430 |
except ValueError: |
431 |
logger.error("Malformed synnefo tag %s", tag)
|
432 |
except IndexError: |
433 |
logger.error("Found tag %s for non-existent NIC %d",
|
434 |
tag, index) |
435 |
return nics
|
436 |
|
437 |
|
438 |
def disks_from_instance(i): |
439 |
return dict([(index, {"size": size}) |
440 |
for index, size in enumerate(i["disk.sizes"])]) |
441 |
|
442 |
|
443 |
class NetworkReconciler(object): |
444 |
def __init__(self, logger, fix=False, conflicting_ips=False): |
445 |
self.log = logger
|
446 |
self.conflicting_ips = conflicting_ips
|
447 |
self.fix = fix
|
448 |
|
449 |
@transaction.commit_on_success
|
450 |
def reconcile_networks(self): |
451 |
# Get models from DB
|
452 |
backends = Backend.objects.exclude(offline=True)
|
453 |
networks = Network.objects.filter(deleted=False)
|
454 |
|
455 |
self.event_time = datetime.now()
|
456 |
|
457 |
# Get info from all ganeti backends
|
458 |
ganeti_networks = {} |
459 |
ganeti_hanging_networks = {} |
460 |
for b in backends: |
461 |
g_nets = get_networks_from_ganeti(b) |
462 |
ganeti_networks[b] = g_nets |
463 |
g_hanging_nets = hanging_networks(b, g_nets) |
464 |
ganeti_hanging_networks[b] = g_hanging_nets |
465 |
|
466 |
# Perform reconciliation for each network
|
467 |
for network in networks: |
468 |
ip_available_maps = [] |
469 |
ip_reserved_maps = [] |
470 |
for bend in backends: |
471 |
bnet = get_backend_network(network, bend) |
472 |
gnet = ganeti_networks[bend].get(network.id) |
473 |
if not bnet: |
474 |
if network.floating_ip_pool:
|
475 |
# Network is a floating IP pool and does not exist in
|
476 |
# backend. We need to create it
|
477 |
bnet = self.reconcile_parted_network(network, bend)
|
478 |
elif not gnet: |
479 |
# Network does not exist either in Ganeti nor in BD.
|
480 |
continue
|
481 |
else:
|
482 |
# Network exists in Ganeti and not in DB.
|
483 |
if network.action != "DESTROY" and not network.public: |
484 |
bnet = self.reconcile_parted_network(network, bend)
|
485 |
else:
|
486 |
continue
|
487 |
|
488 |
if not gnet: |
489 |
# Network does not exist in Ganeti. If the network action
|
490 |
# is DESTROY, we have to mark as deleted in DB, else we
|
491 |
# have to create it in Ganeti.
|
492 |
if network.action == "DESTROY": |
493 |
if bnet.operstate != "DELETED": |
494 |
self.reconcile_stale_network(bnet)
|
495 |
else:
|
496 |
self.reconcile_missing_network(network, bend)
|
497 |
# Skip rest reconciliation!
|
498 |
continue
|
499 |
|
500 |
try:
|
501 |
hanging_groups = ganeti_hanging_networks[bend][network.id] |
502 |
except KeyError: |
503 |
# Network is connected to all nodegroups
|
504 |
hanging_groups = [] |
505 |
|
506 |
if hanging_groups:
|
507 |
# CASE-3: Ganeti networks not connected to all nodegroups
|
508 |
self.reconcile_hanging_groups(network, bend,
|
509 |
hanging_groups) |
510 |
continue
|
511 |
|
512 |
if bnet.operstate != 'ACTIVE': |
513 |
# CASE-4: Unsynced network state. At this point the network
|
514 |
# exists and is connected to all nodes so is must be
|
515 |
# active!
|
516 |
self.reconcile_unsynced_network(network, bend, bnet)
|
517 |
|
518 |
# Get ganeti IP Pools
|
519 |
available_map, reserved_map = get_network_pool(gnet) |
520 |
ip_available_maps.append(available_map) |
521 |
ip_reserved_maps.append(reserved_map) |
522 |
|
523 |
if ip_available_maps or ip_reserved_maps: |
524 |
# CASE-5: Unsynced IP Pools
|
525 |
self.reconcile_ip_pools(network, ip_available_maps,
|
526 |
ip_reserved_maps) |
527 |
|
528 |
if self.conflicting_ips: |
529 |
self.detect_conflicting_ips()
|
530 |
|
531 |
# CASE-6: Orphan networks
|
532 |
self.reconcile_orphan_networks(networks, ganeti_networks)
|
533 |
|
534 |
def reconcile_parted_network(self, network, backend): |
535 |
self.log.info("D: Missing DB entry for network %s in backend %s", |
536 |
network, backend) |
537 |
if self.fix: |
538 |
network.create_backend_network(backend) |
539 |
self.log.info("F: Created DB entry") |
540 |
bnet = get_backend_network(network, backend) |
541 |
return bnet
|
542 |
|
543 |
def reconcile_stale_network(self, backend_network): |
544 |
self.log.info("D: Stale DB entry for network %s in backend %s", |
545 |
backend_network.network, backend_network.backend) |
546 |
if self.fix: |
547 |
backend_mod.process_network_status( |
548 |
backend_network, self.event_time, 0, |
549 |
"OP_NETWORK_REMOVE",
|
550 |
"success",
|
551 |
"Reconciliation simulated event")
|
552 |
self.log.info("F: Reconciled event: OP_NETWORK_REMOVE") |
553 |
|
554 |
def reconcile_missing_network(self, network, backend): |
555 |
self.log.info("D: Missing Ganeti network %s in backend %s", |
556 |
network, backend) |
557 |
if self.fix: |
558 |
backend_mod.create_network(network, backend) |
559 |
self.log.info("F: Issued OP_NETWORK_CONNECT") |
560 |
|
561 |
def reconcile_hanging_groups(self, network, backend, hanging_groups): |
562 |
self.log.info('D: Network %s in backend %s is not connected to ' |
563 |
'the following groups:', network, backend)
|
564 |
self.log.info('- ' + '\n- '.join(hanging_groups)) |
565 |
if self.fix: |
566 |
for group in hanging_groups: |
567 |
self.log.info('F: Connecting network %s to nodegroup %s', |
568 |
network, group) |
569 |
backend_mod.connect_network(network, backend, depends=[], |
570 |
group=group) |
571 |
|
572 |
def reconcile_unsynced_network(self, network, backend, backend_network): |
573 |
self.log.info("D: Unsynced network %s in backend %s", network, backend) |
574 |
if self.fix: |
575 |
self.log.info("F: Issuing OP_NETWORK_CONNECT") |
576 |
backend_mod.process_network_status( |
577 |
backend_network, self.event_time, 0, |
578 |
"OP_NETWORK_CONNECT",
|
579 |
"success",
|
580 |
"Reconciliation simulated eventd")
|
581 |
|
582 |
def reconcile_ip_pools(self, network, available_maps, reserved_maps): |
583 |
available_map = reduce(lambda x, y: x & y, available_maps) |
584 |
reserved_map = reduce(lambda x, y: x & y, reserved_maps) |
585 |
|
586 |
pool = network.get_pool() |
587 |
# Temporary release unused floating IPs
|
588 |
temp_pool = network.get_pool() |
589 |
used_ips = network.nics.values_list("ipv4", flat=True) |
590 |
unused_static_ips = network.floating_ips.exclude(ipv4__in=used_ips) |
591 |
map(lambda ip: temp_pool.put(ip.ipv4), unused_static_ips) |
592 |
if temp_pool.available != available_map:
|
593 |
self.log.info("D: Unsynced available map of network %s:\n" |
594 |
"\tDB: %r\n\tGB: %r", network,
|
595 |
temp_pool.available.to01(), |
596 |
available_map.to01()) |
597 |
if self.fix: |
598 |
pool.available = available_map |
599 |
# Release unsued floating IPs, as they are not included in the
|
600 |
# available map
|
601 |
map(lambda ip: pool.reserve(ip.ipv4), unused_static_ips) |
602 |
pool.save() |
603 |
if pool.reserved != reserved_map:
|
604 |
self.log.info("D: Unsynced reserved map of network %s:\n" |
605 |
"\tDB: %r\n\tGB: %r", network, pool.reserved.to01(),
|
606 |
reserved_map.to01()) |
607 |
if self.fix: |
608 |
pool.reserved = reserved_map |
609 |
pool.save() |
610 |
|
611 |
def detect_conflicting_ips(self, network): |
612 |
"""Detect NIC's that have the same IP in the same network."""
|
613 |
machine_ips = network.nics.all().values_list('ipv4', 'machine') |
614 |
ips = map(lambda x: x[0], machine_ips) |
615 |
distinct_ips = set(ips)
|
616 |
if len(distinct_ips) < len(ips): |
617 |
for i in distinct_ips: |
618 |
ips.remove(i) |
619 |
for i in ips: |
620 |
machines = [utils.id_to_instance_name(x[1])
|
621 |
for x in machine_ips if x[0] == i] |
622 |
self.log.info('D: Conflicting IP:%s Machines: %s', |
623 |
i, ', '.join(machines))
|
624 |
|
625 |
def reconcile_orphan_networks(self, db_networks, ganeti_networks): |
626 |
# Detect Orphan Networks in Ganeti
|
627 |
db_network_ids = set([net.id for net in db_networks]) |
628 |
for back_end, ganeti_networks in ganeti_networks.items(): |
629 |
ganeti_network_ids = set(ganeti_networks.keys())
|
630 |
orphans = ganeti_network_ids - db_network_ids |
631 |
|
632 |
if len(orphans) > 0: |
633 |
self.log.info('D: Orphan Networks in backend %s:', |
634 |
back_end.clustername) |
635 |
self.log.info('- ' + '\n- '.join([str(o) for o in orphans])) |
636 |
if self.fix: |
637 |
for net_id in orphans: |
638 |
self.log.info('Disconnecting and deleting network %d', |
639 |
net_id) |
640 |
try:
|
641 |
network = Network.objects.get(id=net_id) |
642 |
backend_mod.delete_network(network, |
643 |
backend=back_end) |
644 |
except Network.DoesNotExist:
|
645 |
self.log.info("Not entry for network %s in DB !!", |
646 |
net_id) |
647 |
|
648 |
|
649 |
def get_backend_network(network, backend): |
650 |
try:
|
651 |
return BackendNetwork.objects.get(network=network, backend=backend)
|
652 |
except BackendNetwork.DoesNotExist:
|
653 |
return None |
654 |
|
655 |
|
656 |
def get_network_pool(gnet): |
657 |
"""Return available and reserved IP maps.
|
658 |
|
659 |
Extract the available and reserved IP map from the info return from Ganeti
|
660 |
for a network.
|
661 |
|
662 |
"""
|
663 |
converter = IPPool(Foo(gnet['network']))
|
664 |
a_map = bitarray_from_map(gnet['map'])
|
665 |
a_map.invert() |
666 |
reserved = gnet['external_reservations']
|
667 |
r_map = a_map.copy() |
668 |
r_map.setall(True)
|
669 |
if reserved:
|
670 |
for address in reserved.split(','): |
671 |
index = converter.value_to_index(address.strip()) |
672 |
a_map[index] = True
|
673 |
r_map[index] = False
|
674 |
return a_map, r_map
|
675 |
|
676 |
|
677 |
def bitarray_from_map(bitmap): |
678 |
return bitarray.bitarray(bitmap.replace("X", "1").replace(".", "0")) |
679 |
|
680 |
|
681 |
class Foo(): |
682 |
def __init__(self, subnet): |
683 |
self.available_map = '' |
684 |
self.reserved_map = '' |
685 |
self.size = 0 |
686 |
self.network = Foo.Foo1(subnet)
|
687 |
|
688 |
class Foo1(): |
689 |
def __init__(self, subnet): |
690 |
self.subnet = subnet
|
691 |
self.gateway = None |