root / lib / config.py @ 8a113c7a
History | View | Annotate | Download (34.5 kB)
1 |
#
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007 Google Inc.
|
5 |
#
|
6 |
# This program is free software; you can redistribute it and/or modify
|
7 |
# it under the terms of the GNU General Public License as published by
|
8 |
# the Free Software Foundation; either version 2 of the License, or
|
9 |
# (at your option) any later version.
|
10 |
#
|
11 |
# This program is distributed in the hope that it will be useful, but
|
12 |
# WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 |
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 |
# General Public License for more details.
|
15 |
#
|
16 |
# You should have received a copy of the GNU General Public License
|
17 |
# along with this program; if not, write to the Free Software
|
18 |
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
19 |
# 02110-1301, USA.
|
20 |
|
21 |
|
22 |
"""Configuration management for Ganeti
|
23 |
|
24 |
This module provides the interface to the Ganeti cluster configuration.
|
25 |
|
26 |
The configuration data is stored on every node but is updated on the master
|
27 |
only. After each update, the master distributes the data to the other nodes.
|
28 |
|
29 |
Currently, the data storage format is JSON. YAML was slow and consuming too
|
30 |
much memory.
|
31 |
|
32 |
"""
|
33 |
|
34 |
import os |
35 |
import tempfile |
36 |
import random |
37 |
import logging |
38 |
|
39 |
from ganeti import errors |
40 |
from ganeti import locking |
41 |
from ganeti import utils |
42 |
from ganeti import constants |
43 |
from ganeti import rpc |
44 |
from ganeti import objects |
45 |
from ganeti import serializer |
46 |
|
47 |
|
48 |
_config_lock = locking.SharedLock() |
49 |
|
50 |
|
51 |
def _ValidateConfig(data): |
52 |
"""Verifies that a configuration objects looks valid.
|
53 |
|
54 |
This only verifies the version of the configuration.
|
55 |
|
56 |
@raise errors.ConfigurationError: if the version differs from what
|
57 |
we expect
|
58 |
|
59 |
"""
|
60 |
if data.version != constants.CONFIG_VERSION:
|
61 |
raise errors.ConfigurationError("Cluster configuration version" |
62 |
" mismatch, got %s instead of %s" %
|
63 |
(data.version, |
64 |
constants.CONFIG_VERSION)) |
65 |
|
66 |
|
67 |
class ConfigWriter: |
68 |
"""The interface to the cluster configuration.
|
69 |
|
70 |
"""
|
71 |
def __init__(self, cfg_file=None, offline=False): |
72 |
self.write_count = 0 |
73 |
self._lock = _config_lock
|
74 |
self._config_data = None |
75 |
self._offline = offline
|
76 |
if cfg_file is None: |
77 |
self._cfg_file = constants.CLUSTER_CONF_FILE
|
78 |
else:
|
79 |
self._cfg_file = cfg_file
|
80 |
self._temporary_ids = set() |
81 |
self._temporary_drbds = {}
|
82 |
# Note: in order to prevent errors when resolving our name in
|
83 |
# _DistributeConfig, we compute it here once and reuse it; it's
|
84 |
# better to raise an error before starting to modify the config
|
85 |
# file than after it was modified
|
86 |
self._my_hostname = utils.HostInfo().name
|
87 |
self._last_cluster_serial = -1 |
88 |
self._OpenConfig()
|
89 |
|
90 |
# this method needs to be static, so that we can call it on the class
|
91 |
@staticmethod
|
92 |
def IsCluster(): |
93 |
"""Check if the cluster is configured.
|
94 |
|
95 |
"""
|
96 |
return os.path.exists(constants.CLUSTER_CONF_FILE)
|
97 |
|
98 |
@locking.ssynchronized(_config_lock, shared=1) |
99 |
def GenerateMAC(self): |
100 |
"""Generate a MAC for an instance.
|
101 |
|
102 |
This should check the current instances for duplicates.
|
103 |
|
104 |
"""
|
105 |
prefix = self._config_data.cluster.mac_prefix
|
106 |
all_macs = self._AllMACs()
|
107 |
retries = 64
|
108 |
while retries > 0: |
109 |
byte1 = random.randrange(0, 256) |
110 |
byte2 = random.randrange(0, 256) |
111 |
byte3 = random.randrange(0, 256) |
112 |
mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3)
|
113 |
if mac not in all_macs: |
114 |
break
|
115 |
retries -= 1
|
116 |
else:
|
117 |
raise errors.ConfigurationError("Can't generate unique MAC") |
118 |
return mac
|
119 |
|
120 |
@locking.ssynchronized(_config_lock, shared=1) |
121 |
def IsMacInUse(self, mac): |
122 |
"""Predicate: check if the specified MAC is in use in the Ganeti cluster.
|
123 |
|
124 |
This only checks instances managed by this cluster, it does not
|
125 |
check for potential collisions elsewhere.
|
126 |
|
127 |
"""
|
128 |
all_macs = self._AllMACs()
|
129 |
return mac in all_macs |
130 |
|
131 |
@locking.ssynchronized(_config_lock, shared=1) |
132 |
def GenerateDRBDSecret(self): |
133 |
"""Generate a DRBD secret.
|
134 |
|
135 |
This checks the current disks for duplicates.
|
136 |
|
137 |
"""
|
138 |
all_secrets = self._AllDRBDSecrets()
|
139 |
retries = 64
|
140 |
while retries > 0: |
141 |
secret = utils.GenerateSecret() |
142 |
if secret not in all_secrets: |
143 |
break
|
144 |
retries -= 1
|
145 |
else:
|
146 |
raise errors.ConfigurationError("Can't generate unique DRBD secret") |
147 |
return secret
|
148 |
|
149 |
def _ComputeAllLVs(self): |
150 |
"""Compute the list of all LVs.
|
151 |
|
152 |
"""
|
153 |
lvnames = set()
|
154 |
for instance in self._config_data.instances.values(): |
155 |
node_data = instance.MapLVsByNode() |
156 |
for lv_list in node_data.values(): |
157 |
lvnames.update(lv_list) |
158 |
return lvnames
|
159 |
|
160 |
@locking.ssynchronized(_config_lock, shared=1) |
161 |
def GenerateUniqueID(self, exceptions=None): |
162 |
"""Generate an unique disk name.
|
163 |
|
164 |
This checks the current node, instances and disk names for
|
165 |
duplicates.
|
166 |
|
167 |
@param exceptions: a list with some other names which should be checked
|
168 |
for uniqueness (used for example when you want to get
|
169 |
more than one id at one time without adding each one in
|
170 |
turn to the config file)
|
171 |
|
172 |
@rtype: string
|
173 |
@return: the unique id
|
174 |
|
175 |
"""
|
176 |
existing = set()
|
177 |
existing.update(self._temporary_ids)
|
178 |
existing.update(self._ComputeAllLVs())
|
179 |
existing.update(self._config_data.instances.keys())
|
180 |
existing.update(self._config_data.nodes.keys())
|
181 |
if exceptions is not None: |
182 |
existing.update(exceptions) |
183 |
retries = 64
|
184 |
while retries > 0: |
185 |
unique_id = utils.NewUUID() |
186 |
if unique_id not in existing and unique_id is not None: |
187 |
break
|
188 |
else:
|
189 |
raise errors.ConfigurationError("Not able generate an unique ID" |
190 |
" (last tried ID: %s" % unique_id)
|
191 |
self._temporary_ids.add(unique_id)
|
192 |
return unique_id
|
193 |
|
194 |
def _AllMACs(self): |
195 |
"""Return all MACs present in the config.
|
196 |
|
197 |
@rtype: list
|
198 |
@return: the list of all MACs
|
199 |
|
200 |
"""
|
201 |
result = [] |
202 |
for instance in self._config_data.instances.values(): |
203 |
for nic in instance.nics: |
204 |
result.append(nic.mac) |
205 |
|
206 |
return result
|
207 |
|
208 |
def _AllDRBDSecrets(self): |
209 |
"""Return all DRBD secrets present in the config.
|
210 |
|
211 |
@rtype: list
|
212 |
@return: the list of all DRBD secrets
|
213 |
|
214 |
"""
|
215 |
def helper(disk, result): |
216 |
"""Recursively gather secrets from this disk."""
|
217 |
if disk.dev_type == constants.DT_DRBD8:
|
218 |
result.append(disk.logical_id[5])
|
219 |
if disk.children:
|
220 |
for child in disk.children: |
221 |
helper(child, result) |
222 |
|
223 |
result = [] |
224 |
for instance in self._config_data.instances.values(): |
225 |
for disk in instance.disks: |
226 |
helper(disk, result) |
227 |
|
228 |
return result
|
229 |
|
230 |
@locking.ssynchronized(_config_lock, shared=1) |
231 |
def VerifyConfig(self): |
232 |
"""Verify function.
|
233 |
|
234 |
"""
|
235 |
result = [] |
236 |
seen_macs = [] |
237 |
ports = {} |
238 |
data = self._config_data
|
239 |
for instance_name in data.instances: |
240 |
instance = data.instances[instance_name] |
241 |
if instance.primary_node not in data.nodes: |
242 |
result.append("instance '%s' has invalid primary node '%s'" %
|
243 |
(instance_name, instance.primary_node)) |
244 |
for snode in instance.secondary_nodes: |
245 |
if snode not in data.nodes: |
246 |
result.append("instance '%s' has invalid secondary node '%s'" %
|
247 |
(instance_name, snode)) |
248 |
for idx, nic in enumerate(instance.nics): |
249 |
if nic.mac in seen_macs: |
250 |
result.append("instance '%s' has NIC %d mac %s duplicate" %
|
251 |
(instance_name, idx, nic.mac)) |
252 |
else:
|
253 |
seen_macs.append(nic.mac) |
254 |
|
255 |
# gather the drbd ports for duplicate checks
|
256 |
for dsk in instance.disks: |
257 |
if dsk.dev_type in constants.LDS_DRBD: |
258 |
tcp_port = dsk.logical_id[2]
|
259 |
if tcp_port not in ports: |
260 |
ports[tcp_port] = [] |
261 |
ports[tcp_port].append((instance.name, "drbd disk %s" % dsk.iv_name))
|
262 |
# gather network port reservation
|
263 |
net_port = getattr(instance, "network_port", None) |
264 |
if net_port is not None: |
265 |
if net_port not in ports: |
266 |
ports[net_port] = [] |
267 |
ports[net_port].append((instance.name, "network port"))
|
268 |
|
269 |
# cluster-wide pool of free ports
|
270 |
for free_port in data.cluster.tcpudp_port_pool: |
271 |
if free_port not in ports: |
272 |
ports[free_port] = [] |
273 |
ports[free_port].append(("cluster", "port marked as free")) |
274 |
|
275 |
# compute tcp/udp duplicate ports
|
276 |
keys = ports.keys() |
277 |
keys.sort() |
278 |
for pnum in keys: |
279 |
pdata = ports[pnum] |
280 |
if len(pdata) > 1: |
281 |
txt = ", ".join(["%s/%s" % val for val in pdata]) |
282 |
result.append("tcp/udp port %s has duplicates: %s" % (pnum, txt))
|
283 |
|
284 |
# highest used tcp port check
|
285 |
if keys:
|
286 |
if keys[-1] > data.cluster.highest_used_port: |
287 |
result.append("Highest used port mismatch, saved %s, computed %s" %
|
288 |
(data.cluster.highest_used_port, keys[-1]))
|
289 |
|
290 |
if not data.nodes[data.cluster.master_node].master_candidate: |
291 |
result.append("Master node is not a master candidate")
|
292 |
|
293 |
mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
|
294 |
if mc_now < mc_max:
|
295 |
result.append("Not enough master candidates: actual %d, target %d" %
|
296 |
(mc_now, mc_max)) |
297 |
|
298 |
return result
|
299 |
|
300 |
def _UnlockedSetDiskID(self, disk, node_name): |
301 |
"""Convert the unique ID to the ID needed on the target nodes.
|
302 |
|
303 |
This is used only for drbd, which needs ip/port configuration.
|
304 |
|
305 |
The routine descends down and updates its children also, because
|
306 |
this helps when the only the top device is passed to the remote
|
307 |
node.
|
308 |
|
309 |
This function is for internal use, when the config lock is already held.
|
310 |
|
311 |
"""
|
312 |
if disk.children:
|
313 |
for child in disk.children: |
314 |
self._UnlockedSetDiskID(child, node_name)
|
315 |
|
316 |
if disk.logical_id is None and disk.physical_id is not None: |
317 |
return
|
318 |
if disk.dev_type == constants.LD_DRBD8:
|
319 |
pnode, snode, port, pminor, sminor, secret = disk.logical_id |
320 |
if node_name not in (pnode, snode): |
321 |
raise errors.ConfigurationError("DRBD device not knowing node %s" % |
322 |
node_name) |
323 |
pnode_info = self._UnlockedGetNodeInfo(pnode)
|
324 |
snode_info = self._UnlockedGetNodeInfo(snode)
|
325 |
if pnode_info is None or snode_info is None: |
326 |
raise errors.ConfigurationError("Can't find primary or secondary node" |
327 |
" for %s" % str(disk)) |
328 |
p_data = (pnode_info.secondary_ip, port) |
329 |
s_data = (snode_info.secondary_ip, port) |
330 |
if pnode == node_name:
|
331 |
disk.physical_id = p_data + s_data + (pminor, secret) |
332 |
else: # it must be secondary, we tested above |
333 |
disk.physical_id = s_data + p_data + (sminor, secret) |
334 |
else:
|
335 |
disk.physical_id = disk.logical_id |
336 |
return
|
337 |
|
338 |
@locking.ssynchronized(_config_lock)
|
339 |
def SetDiskID(self, disk, node_name): |
340 |
"""Convert the unique ID to the ID needed on the target nodes.
|
341 |
|
342 |
This is used only for drbd, which needs ip/port configuration.
|
343 |
|
344 |
The routine descends down and updates its children also, because
|
345 |
this helps when the only the top device is passed to the remote
|
346 |
node.
|
347 |
|
348 |
"""
|
349 |
return self._UnlockedSetDiskID(disk, node_name) |
350 |
|
351 |
@locking.ssynchronized(_config_lock)
|
352 |
def AddTcpUdpPort(self, port): |
353 |
"""Adds a new port to the available port pool.
|
354 |
|
355 |
"""
|
356 |
if not isinstance(port, int): |
357 |
raise errors.ProgrammerError("Invalid type passed for port") |
358 |
|
359 |
self._config_data.cluster.tcpudp_port_pool.add(port)
|
360 |
self._WriteConfig()
|
361 |
|
362 |
@locking.ssynchronized(_config_lock, shared=1) |
363 |
def GetPortList(self): |
364 |
"""Returns a copy of the current port list.
|
365 |
|
366 |
"""
|
367 |
return self._config_data.cluster.tcpudp_port_pool.copy() |
368 |
|
369 |
@locking.ssynchronized(_config_lock)
|
370 |
def AllocatePort(self): |
371 |
"""Allocate a port.
|
372 |
|
373 |
The port will be taken from the available port pool or from the
|
374 |
default port range (and in this case we increase
|
375 |
highest_used_port).
|
376 |
|
377 |
"""
|
378 |
# If there are TCP/IP ports configured, we use them first.
|
379 |
if self._config_data.cluster.tcpudp_port_pool: |
380 |
port = self._config_data.cluster.tcpudp_port_pool.pop()
|
381 |
else:
|
382 |
port = self._config_data.cluster.highest_used_port + 1 |
383 |
if port >= constants.LAST_DRBD_PORT:
|
384 |
raise errors.ConfigurationError("The highest used port is greater" |
385 |
" than %s. Aborting." %
|
386 |
constants.LAST_DRBD_PORT) |
387 |
self._config_data.cluster.highest_used_port = port
|
388 |
|
389 |
self._WriteConfig()
|
390 |
return port
|
391 |
|
392 |
def _ComputeDRBDMap(self, instance): |
393 |
"""Compute the used DRBD minor/nodes.
|
394 |
|
395 |
@return: dictionary of node_name: dict of minor: instance_name;
|
396 |
the returned dict will have all the nodes in it (even if with
|
397 |
an empty list).
|
398 |
|
399 |
"""
|
400 |
def _AppendUsedPorts(instance_name, disk, used): |
401 |
if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5: |
402 |
nodeA, nodeB, dummy, minorA, minorB = disk.logical_id[:5]
|
403 |
for node, port in ((nodeA, minorA), (nodeB, minorB)): |
404 |
assert node in used, "Instance node not found in node list" |
405 |
if port in used[node]: |
406 |
raise errors.ProgrammerError("DRBD minor already used:" |
407 |
" %s/%s, %s/%s" %
|
408 |
(node, port, instance_name, |
409 |
used[node][port])) |
410 |
|
411 |
used[node][port] = instance_name |
412 |
if disk.children:
|
413 |
for child in disk.children: |
414 |
_AppendUsedPorts(instance_name, child, used) |
415 |
|
416 |
my_dict = dict((node, {}) for node in self._config_data.nodes) |
417 |
for (node, minor), instance in self._temporary_drbds.iteritems(): |
418 |
my_dict[node][minor] = instance |
419 |
for instance in self._config_data.instances.itervalues(): |
420 |
for disk in instance.disks: |
421 |
_AppendUsedPorts(instance.name, disk, my_dict) |
422 |
return my_dict
|
423 |
|
424 |
@locking.ssynchronized(_config_lock)
|
425 |
def AllocateDRBDMinor(self, nodes, instance): |
426 |
"""Allocate a drbd minor.
|
427 |
|
428 |
The free minor will be automatically computed from the existing
|
429 |
devices. A node can be given multiple times in order to allocate
|
430 |
multiple minors. The result is the list of minors, in the same
|
431 |
order as the passed nodes.
|
432 |
|
433 |
"""
|
434 |
d_map = self._ComputeDRBDMap(instance)
|
435 |
result = [] |
436 |
for nname in nodes: |
437 |
ndata = d_map[nname] |
438 |
if not ndata: |
439 |
# no minors used, we can start at 0
|
440 |
result.append(0)
|
441 |
ndata[0] = instance
|
442 |
self._temporary_drbds[(nname, 0)] = instance |
443 |
continue
|
444 |
keys = ndata.keys() |
445 |
keys.sort() |
446 |
ffree = utils.FirstFree(keys) |
447 |
if ffree is None: |
448 |
# return the next minor
|
449 |
# TODO: implement high-limit check
|
450 |
minor = keys[-1] + 1 |
451 |
else:
|
452 |
minor = ffree |
453 |
result.append(minor) |
454 |
ndata[minor] = instance |
455 |
assert (nname, minor) not in self._temporary_drbds, \ |
456 |
"Attempt to reuse reserved DRBD minor"
|
457 |
self._temporary_drbds[(nname, minor)] = instance
|
458 |
logging.debug("Request to allocate drbd minors, input: %s, returning %s",
|
459 |
nodes, result) |
460 |
return result
|
461 |
|
462 |
@locking.ssynchronized(_config_lock)
|
463 |
def ReleaseDRBDMinors(self, instance): |
464 |
"""Release temporary drbd minors allocated for a given instance.
|
465 |
|
466 |
This should be called on both the error paths and on the success
|
467 |
paths (after the instance has been added or updated).
|
468 |
|
469 |
@type instance: string
|
470 |
@param instance: the instance for which temporary minors should be
|
471 |
released
|
472 |
|
473 |
"""
|
474 |
for key, name in self._temporary_drbds.items(): |
475 |
if name == instance:
|
476 |
del self._temporary_drbds[key] |
477 |
|
478 |
@locking.ssynchronized(_config_lock, shared=1) |
479 |
def GetConfigVersion(self): |
480 |
"""Get the configuration version.
|
481 |
|
482 |
@return: Config version
|
483 |
|
484 |
"""
|
485 |
return self._config_data.version |
486 |
|
487 |
@locking.ssynchronized(_config_lock, shared=1) |
488 |
def GetClusterName(self): |
489 |
"""Get cluster name.
|
490 |
|
491 |
@return: Cluster name
|
492 |
|
493 |
"""
|
494 |
return self._config_data.cluster.cluster_name |
495 |
|
496 |
@locking.ssynchronized(_config_lock, shared=1) |
497 |
def GetMasterNode(self): |
498 |
"""Get the hostname of the master node for this cluster.
|
499 |
|
500 |
@return: Master hostname
|
501 |
|
502 |
"""
|
503 |
return self._config_data.cluster.master_node |
504 |
|
505 |
@locking.ssynchronized(_config_lock, shared=1) |
506 |
def GetMasterIP(self): |
507 |
"""Get the IP of the master node for this cluster.
|
508 |
|
509 |
@return: Master IP
|
510 |
|
511 |
"""
|
512 |
return self._config_data.cluster.master_ip |
513 |
|
514 |
@locking.ssynchronized(_config_lock, shared=1) |
515 |
def GetMasterNetdev(self): |
516 |
"""Get the master network device for this cluster.
|
517 |
|
518 |
"""
|
519 |
return self._config_data.cluster.master_netdev |
520 |
|
521 |
@locking.ssynchronized(_config_lock, shared=1) |
522 |
def GetFileStorageDir(self): |
523 |
"""Get the file storage dir for this cluster.
|
524 |
|
525 |
"""
|
526 |
return self._config_data.cluster.file_storage_dir |
527 |
|
528 |
@locking.ssynchronized(_config_lock, shared=1) |
529 |
def GetHypervisorType(self): |
530 |
"""Get the hypervisor type for this cluster.
|
531 |
|
532 |
"""
|
533 |
return self._config_data.cluster.default_hypervisor |
534 |
|
535 |
@locking.ssynchronized(_config_lock, shared=1) |
536 |
def GetHostKey(self): |
537 |
"""Return the rsa hostkey from the config.
|
538 |
|
539 |
@rtype: string
|
540 |
@return: the rsa hostkey
|
541 |
|
542 |
"""
|
543 |
return self._config_data.cluster.rsahostkeypub |
544 |
|
545 |
@locking.ssynchronized(_config_lock)
|
546 |
def AddInstance(self, instance): |
547 |
"""Add an instance to the config.
|
548 |
|
549 |
This should be used after creating a new instance.
|
550 |
|
551 |
@type instance: L{objects.Instance}
|
552 |
@param instance: the instance object
|
553 |
|
554 |
"""
|
555 |
if not isinstance(instance, objects.Instance): |
556 |
raise errors.ProgrammerError("Invalid type passed to AddInstance") |
557 |
|
558 |
if instance.disk_template != constants.DT_DISKLESS:
|
559 |
all_lvs = instance.MapLVsByNode() |
560 |
logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs)
|
561 |
|
562 |
instance.serial_no = 1
|
563 |
self._config_data.instances[instance.name] = instance
|
564 |
self._WriteConfig()
|
565 |
|
566 |
def _SetInstanceStatus(self, instance_name, status): |
567 |
"""Set the instance's status to a given value.
|
568 |
|
569 |
"""
|
570 |
if status not in ("up", "down"): |
571 |
raise errors.ProgrammerError("Invalid status '%s' passed to" |
572 |
" ConfigWriter._SetInstanceStatus()" %
|
573 |
status) |
574 |
|
575 |
if instance_name not in self._config_data.instances: |
576 |
raise errors.ConfigurationError("Unknown instance '%s'" % |
577 |
instance_name) |
578 |
instance = self._config_data.instances[instance_name]
|
579 |
if instance.status != status:
|
580 |
instance.status = status |
581 |
instance.serial_no += 1
|
582 |
self._WriteConfig()
|
583 |
|
584 |
@locking.ssynchronized(_config_lock)
|
585 |
def MarkInstanceUp(self, instance_name): |
586 |
"""Mark the instance status to up in the config.
|
587 |
|
588 |
"""
|
589 |
self._SetInstanceStatus(instance_name, "up") |
590 |
|
591 |
@locking.ssynchronized(_config_lock)
|
592 |
def RemoveInstance(self, instance_name): |
593 |
"""Remove the instance from the configuration.
|
594 |
|
595 |
"""
|
596 |
if instance_name not in self._config_data.instances: |
597 |
raise errors.ConfigurationError("Unknown instance '%s'" % instance_name) |
598 |
del self._config_data.instances[instance_name] |
599 |
self._WriteConfig()
|
600 |
|
601 |
@locking.ssynchronized(_config_lock)
|
602 |
def RenameInstance(self, old_name, new_name): |
603 |
"""Rename an instance.
|
604 |
|
605 |
This needs to be done in ConfigWriter and not by RemoveInstance
|
606 |
combined with AddInstance as only we can guarantee an atomic
|
607 |
rename.
|
608 |
|
609 |
"""
|
610 |
if old_name not in self._config_data.instances: |
611 |
raise errors.ConfigurationError("Unknown instance '%s'" % old_name) |
612 |
inst = self._config_data.instances[old_name]
|
613 |
del self._config_data.instances[old_name] |
614 |
inst.name = new_name |
615 |
|
616 |
for disk in inst.disks: |
617 |
if disk.dev_type == constants.LD_FILE:
|
618 |
# rename the file paths in logical and physical id
|
619 |
file_storage_dir = os.path.dirname(os.path.dirname(disk.logical_id[1]))
|
620 |
disk.physical_id = disk.logical_id = (disk.logical_id[0],
|
621 |
os.path.join(file_storage_dir, |
622 |
inst.name, |
623 |
disk.iv_name)) |
624 |
|
625 |
self._config_data.instances[inst.name] = inst
|
626 |
self._WriteConfig()
|
627 |
|
628 |
@locking.ssynchronized(_config_lock)
|
629 |
def MarkInstanceDown(self, instance_name): |
630 |
"""Mark the status of an instance to down in the configuration.
|
631 |
|
632 |
"""
|
633 |
self._SetInstanceStatus(instance_name, "down") |
634 |
|
635 |
def _UnlockedGetInstanceList(self): |
636 |
"""Get the list of instances.
|
637 |
|
638 |
This function is for internal use, when the config lock is already held.
|
639 |
|
640 |
"""
|
641 |
return self._config_data.instances.keys() |
642 |
|
643 |
@locking.ssynchronized(_config_lock, shared=1) |
644 |
def GetInstanceList(self): |
645 |
"""Get the list of instances.
|
646 |
|
647 |
@return: array of instances, ex. ['instance2.example.com',
|
648 |
'instance1.example.com']
|
649 |
|
650 |
"""
|
651 |
return self._UnlockedGetInstanceList() |
652 |
|
653 |
@locking.ssynchronized(_config_lock, shared=1) |
654 |
def ExpandInstanceName(self, short_name): |
655 |
"""Attempt to expand an incomplete instance name.
|
656 |
|
657 |
"""
|
658 |
return utils.MatchNameComponent(short_name,
|
659 |
self._config_data.instances.keys())
|
660 |
|
661 |
def _UnlockedGetInstanceInfo(self, instance_name): |
662 |
"""Returns informations about an instance.
|
663 |
|
664 |
This function is for internal use, when the config lock is already held.
|
665 |
|
666 |
"""
|
667 |
if instance_name not in self._config_data.instances: |
668 |
return None |
669 |
|
670 |
return self._config_data.instances[instance_name] |
671 |
|
672 |
@locking.ssynchronized(_config_lock, shared=1) |
673 |
def GetInstanceInfo(self, instance_name): |
674 |
"""Returns informations about an instance.
|
675 |
|
676 |
It takes the information from the configuration file. Other informations of
|
677 |
an instance are taken from the live systems.
|
678 |
|
679 |
@param instance_name: name of the instance, e.g.
|
680 |
I{instance1.example.com}
|
681 |
|
682 |
@rtype: L{objects.Instance}
|
683 |
@return: the instance object
|
684 |
|
685 |
"""
|
686 |
return self._UnlockedGetInstanceInfo(instance_name) |
687 |
|
688 |
@locking.ssynchronized(_config_lock, shared=1) |
689 |
def GetAllInstancesInfo(self): |
690 |
"""Get the configuration of all instances.
|
691 |
|
692 |
@rtype: dict
|
693 |
@returns: dict of (instance, instance_info), where instance_info is what
|
694 |
would GetInstanceInfo return for the node
|
695 |
|
696 |
"""
|
697 |
my_dict = dict([(instance, self._UnlockedGetInstanceInfo(instance)) |
698 |
for instance in self._UnlockedGetInstanceList()]) |
699 |
return my_dict
|
700 |
|
701 |
@locking.ssynchronized(_config_lock)
|
702 |
def AddNode(self, node): |
703 |
"""Add a node to the configuration.
|
704 |
|
705 |
@type node: L{objects.Node}
|
706 |
@param node: a Node instance
|
707 |
|
708 |
"""
|
709 |
logging.info("Adding node %s to configuration" % node.name)
|
710 |
|
711 |
node.serial_no = 1
|
712 |
self._config_data.nodes[node.name] = node
|
713 |
self._config_data.cluster.serial_no += 1 |
714 |
self._WriteConfig()
|
715 |
|
716 |
@locking.ssynchronized(_config_lock)
|
717 |
def RemoveNode(self, node_name): |
718 |
"""Remove a node from the configuration.
|
719 |
|
720 |
"""
|
721 |
logging.info("Removing node %s from configuration" % node_name)
|
722 |
|
723 |
if node_name not in self._config_data.nodes: |
724 |
raise errors.ConfigurationError("Unknown node '%s'" % node_name) |
725 |
|
726 |
del self._config_data.nodes[node_name] |
727 |
self._config_data.cluster.serial_no += 1 |
728 |
self._WriteConfig()
|
729 |
|
730 |
@locking.ssynchronized(_config_lock, shared=1) |
731 |
def ExpandNodeName(self, short_name): |
732 |
"""Attempt to expand an incomplete instance name.
|
733 |
|
734 |
"""
|
735 |
return utils.MatchNameComponent(short_name,
|
736 |
self._config_data.nodes.keys())
|
737 |
|
738 |
def _UnlockedGetNodeInfo(self, node_name): |
739 |
"""Get the configuration of a node, as stored in the config.
|
740 |
|
741 |
This function is for internal use, when the config lock is already
|
742 |
held.
|
743 |
|
744 |
@param node_name: the node name, e.g. I{node1.example.com}
|
745 |
|
746 |
@rtype: L{objects.Node}
|
747 |
@return: the node object
|
748 |
|
749 |
"""
|
750 |
if node_name not in self._config_data.nodes: |
751 |
return None |
752 |
|
753 |
return self._config_data.nodes[node_name] |
754 |
|
755 |
|
756 |
@locking.ssynchronized(_config_lock, shared=1) |
757 |
def GetNodeInfo(self, node_name): |
758 |
"""Get the configuration of a node, as stored in the config.
|
759 |
|
760 |
This is just a locked wrapper over L{_UnlockedGetNodeInfo}.
|
761 |
|
762 |
@param node_name: the node name, e.g. I{node1.example.com}
|
763 |
|
764 |
@rtype: L{objects.Node}
|
765 |
@return: the node object
|
766 |
|
767 |
"""
|
768 |
return self._UnlockedGetNodeInfo(node_name) |
769 |
|
770 |
def _UnlockedGetNodeList(self): |
771 |
"""Return the list of nodes which are in the configuration.
|
772 |
|
773 |
This function is for internal use, when the config lock is already
|
774 |
held.
|
775 |
|
776 |
@rtype: list
|
777 |
|
778 |
"""
|
779 |
return self._config_data.nodes.keys() |
780 |
|
781 |
|
782 |
@locking.ssynchronized(_config_lock, shared=1) |
783 |
def GetNodeList(self): |
784 |
"""Return the list of nodes which are in the configuration.
|
785 |
|
786 |
"""
|
787 |
return self._UnlockedGetNodeList() |
788 |
|
789 |
@locking.ssynchronized(_config_lock, shared=1) |
790 |
def GetOnlineNodeList(self): |
791 |
"""Return the list of nodes which are online.
|
792 |
|
793 |
"""
|
794 |
all_nodes = [self._UnlockedGetNodeInfo(node)
|
795 |
for node in self._UnlockedGetNodeList()] |
796 |
return [node.name for node in all_nodes if not node.offline] |
797 |
|
798 |
@locking.ssynchronized(_config_lock, shared=1) |
799 |
def GetAllNodesInfo(self): |
800 |
"""Get the configuration of all nodes.
|
801 |
|
802 |
@rtype: dict
|
803 |
@return: dict of (node, node_info), where node_info is what
|
804 |
would GetNodeInfo return for the node
|
805 |
|
806 |
"""
|
807 |
my_dict = dict([(node, self._UnlockedGetNodeInfo(node)) |
808 |
for node in self._UnlockedGetNodeList()]) |
809 |
return my_dict
|
810 |
|
811 |
def _UnlockedGetMasterCandidateStats(self): |
812 |
"""Get the number of current and maximum desired and possible candidates.
|
813 |
|
814 |
@rtype: tuple
|
815 |
@return: tuple of (current, desired and possible)
|
816 |
|
817 |
"""
|
818 |
mc_now = mc_max = 0
|
819 |
for node in self._config_data.nodes.itervalues(): |
820 |
if not node.offline: |
821 |
mc_max += 1
|
822 |
if node.master_candidate:
|
823 |
mc_now += 1
|
824 |
mc_max = min(mc_max, self._config_data.cluster.candidate_pool_size) |
825 |
return (mc_now, mc_max)
|
826 |
|
827 |
@locking.ssynchronized(_config_lock, shared=1) |
828 |
def GetMasterCandidateStats(self): |
829 |
"""Get the number of current and maximum possible candidates.
|
830 |
|
831 |
This is just a wrapper over L{_UnlockedGetMasterCandidateStats}.
|
832 |
|
833 |
@rtype: tuple
|
834 |
@return: tuple of (current, max)
|
835 |
|
836 |
"""
|
837 |
return self._UnlockedGetMasterCandidateStats() |
838 |
|
839 |
@locking.ssynchronized(_config_lock)
|
840 |
def MaintainCandidatePool(self): |
841 |
"""Try to grow the candidate pool to the desired size.
|
842 |
|
843 |
@rtype: list
|
844 |
@return: list with the adjusted nodes (L{objects.Node} instances)
|
845 |
|
846 |
"""
|
847 |
mc_now, mc_max = self._UnlockedGetMasterCandidateStats()
|
848 |
mod_list = [] |
849 |
if mc_now < mc_max:
|
850 |
node_list = self._config_data.nodes.keys()
|
851 |
random.shuffle(node_list) |
852 |
for name in node_list: |
853 |
if mc_now >= mc_max:
|
854 |
break
|
855 |
node = self._config_data.nodes[name]
|
856 |
if node.master_candidate or node.offline: |
857 |
continue
|
858 |
mod_list.append(node) |
859 |
node.master_candidate = True
|
860 |
node.serial_no += 1
|
861 |
mc_now += 1
|
862 |
if mc_now != mc_max:
|
863 |
# this should not happen
|
864 |
logging.warning("Warning: MaintainCandidatePool didn't manage to"
|
865 |
" fill the candidate pool (%d/%d)", mc_now, mc_max)
|
866 |
if mod_list:
|
867 |
self._config_data.cluster.serial_no += 1 |
868 |
self._WriteConfig()
|
869 |
|
870 |
return mod_list
|
871 |
|
872 |
def _BumpSerialNo(self): |
873 |
"""Bump up the serial number of the config.
|
874 |
|
875 |
"""
|
876 |
self._config_data.serial_no += 1 |
877 |
|
878 |
def _OpenConfig(self): |
879 |
"""Read the config data from disk.
|
880 |
|
881 |
"""
|
882 |
f = open(self._cfg_file, 'r') |
883 |
try:
|
884 |
try:
|
885 |
data = objects.ConfigData.FromDict(serializer.Load(f.read())) |
886 |
except Exception, err: |
887 |
raise errors.ConfigurationError(err)
|
888 |
finally:
|
889 |
f.close() |
890 |
|
891 |
# Make sure the configuration has the right version
|
892 |
_ValidateConfig(data) |
893 |
|
894 |
if (not hasattr(data, 'cluster') or |
895 |
not hasattr(data.cluster, 'rsahostkeypub')): |
896 |
raise errors.ConfigurationError("Incomplete configuration" |
897 |
" (missing cluster.rsahostkeypub)")
|
898 |
self._config_data = data
|
899 |
# reset the last serial as -1 so that the next write will cause
|
900 |
# ssconf update
|
901 |
self._last_cluster_serial = -1 |
902 |
|
903 |
def _DistributeConfig(self): |
904 |
"""Distribute the configuration to the other nodes.
|
905 |
|
906 |
Currently, this only copies the configuration file. In the future,
|
907 |
it could be used to encapsulate the 2/3-phase update mechanism.
|
908 |
|
909 |
"""
|
910 |
if self._offline: |
911 |
return True |
912 |
bad = False
|
913 |
|
914 |
node_list = [] |
915 |
addr_list = [] |
916 |
myhostname = self._my_hostname
|
917 |
# we can skip checking whether _UnlockedGetNodeInfo returns None
|
918 |
# since the node list comes from _UnlocketGetNodeList, and we are
|
919 |
# called with the lock held, so no modifications should take place
|
920 |
# in between
|
921 |
for node_name in self._UnlockedGetNodeList(): |
922 |
if node_name == myhostname:
|
923 |
continue
|
924 |
node_info = self._UnlockedGetNodeInfo(node_name)
|
925 |
if not node_info.master_candidate: |
926 |
continue
|
927 |
node_list.append(node_info.name) |
928 |
addr_list.append(node_info.primary_ip) |
929 |
|
930 |
result = rpc.RpcRunner.call_upload_file(node_list, self._cfg_file,
|
931 |
address_list=addr_list) |
932 |
for node in node_list: |
933 |
if not result[node]: |
934 |
logging.error("copy of file %s to node %s failed",
|
935 |
self._cfg_file, node)
|
936 |
bad = True
|
937 |
return not bad |
938 |
|
939 |
def _WriteConfig(self, destination=None): |
940 |
"""Write the configuration data to persistent storage.
|
941 |
|
942 |
"""
|
943 |
if destination is None: |
944 |
destination = self._cfg_file
|
945 |
self._BumpSerialNo()
|
946 |
txt = serializer.Dump(self._config_data.ToDict())
|
947 |
dir_name, file_name = os.path.split(destination) |
948 |
fd, name = tempfile.mkstemp('.newconfig', file_name, dir_name)
|
949 |
f = os.fdopen(fd, 'w')
|
950 |
try:
|
951 |
f.write(txt) |
952 |
os.fsync(f.fileno()) |
953 |
finally:
|
954 |
f.close() |
955 |
# we don't need to do os.close(fd) as f.close() did it
|
956 |
os.rename(name, destination) |
957 |
self.write_count += 1 |
958 |
|
959 |
# and redistribute the config file to master candidates
|
960 |
self._DistributeConfig()
|
961 |
|
962 |
# Write ssconf files on all nodes (including locally)
|
963 |
if self._last_cluster_serial < self._config_data.cluster.serial_no: |
964 |
if not self._offline: |
965 |
rpc.RpcRunner.call_write_ssconf_files(self._UnlockedGetNodeList(),
|
966 |
self._UnlockedGetSsconfValues())
|
967 |
self._last_cluster_serial = self._config_data.cluster.serial_no |
968 |
|
969 |
def _UnlockedGetSsconfValues(self): |
970 |
"""Return the values needed by ssconf.
|
971 |
|
972 |
@rtype: dict
|
973 |
@return: a dictionary with keys the ssconf names and values their
|
974 |
associated value
|
975 |
|
976 |
"""
|
977 |
fn = "\n".join
|
978 |
node_names = utils.NiceSort(self._UnlockedGetNodeList())
|
979 |
node_info = [self._UnlockedGetNodeInfo(name) for name in node_names] |
980 |
|
981 |
off_data = fn(node.name for node in node_info if node.offline) |
982 |
mc_data = fn(node.name for node in node_info if node.master_candidate) |
983 |
node_data = fn(node_names) |
984 |
|
985 |
cluster = self._config_data.cluster
|
986 |
return {
|
987 |
constants.SS_CLUSTER_NAME: cluster.cluster_name, |
988 |
constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir, |
989 |
constants.SS_MASTER_CANDIDATES: mc_data, |
990 |
constants.SS_MASTER_IP: cluster.master_ip, |
991 |
constants.SS_MASTER_NETDEV: cluster.master_netdev, |
992 |
constants.SS_MASTER_NODE: cluster.master_node, |
993 |
constants.SS_NODE_LIST: node_data, |
994 |
constants.SS_OFFLINE_NODES: off_data, |
995 |
constants.SS_RELEASE_VERSION: constants.RELEASE_VERSION, |
996 |
} |
997 |
|
998 |
@locking.ssynchronized(_config_lock)
|
999 |
def InitConfig(self, version, cluster_config, master_node_config): |
1000 |
"""Create the initial cluster configuration.
|
1001 |
|
1002 |
It will contain the current node, which will also be the master
|
1003 |
node, and no instances.
|
1004 |
|
1005 |
@type version: int
|
1006 |
@param version: Configuration version
|
1007 |
@type cluster_config: objects.Cluster
|
1008 |
@param cluster_config: Cluster configuration
|
1009 |
@type master_node_config: objects.Node
|
1010 |
@param master_node_config: Master node configuration
|
1011 |
|
1012 |
"""
|
1013 |
nodes = { |
1014 |
master_node_config.name: master_node_config, |
1015 |
} |
1016 |
|
1017 |
self._config_data = objects.ConfigData(version=version,
|
1018 |
cluster=cluster_config, |
1019 |
nodes=nodes, |
1020 |
instances={}, |
1021 |
serial_no=1)
|
1022 |
self._WriteConfig()
|
1023 |
|
1024 |
@locking.ssynchronized(_config_lock, shared=1) |
1025 |
def GetVGName(self): |
1026 |
"""Return the volume group name.
|
1027 |
|
1028 |
"""
|
1029 |
return self._config_data.cluster.volume_group_name |
1030 |
|
1031 |
@locking.ssynchronized(_config_lock)
|
1032 |
def SetVGName(self, vg_name): |
1033 |
"""Set the volume group name.
|
1034 |
|
1035 |
"""
|
1036 |
self._config_data.cluster.volume_group_name = vg_name
|
1037 |
self._config_data.cluster.serial_no += 1 |
1038 |
self._WriteConfig()
|
1039 |
|
1040 |
@locking.ssynchronized(_config_lock, shared=1) |
1041 |
def GetDefBridge(self): |
1042 |
"""Return the default bridge.
|
1043 |
|
1044 |
"""
|
1045 |
return self._config_data.cluster.default_bridge |
1046 |
|
1047 |
@locking.ssynchronized(_config_lock, shared=1) |
1048 |
def GetMACPrefix(self): |
1049 |
"""Return the mac prefix.
|
1050 |
|
1051 |
"""
|
1052 |
return self._config_data.cluster.mac_prefix |
1053 |
|
1054 |
@locking.ssynchronized(_config_lock, shared=1) |
1055 |
def GetClusterInfo(self): |
1056 |
"""Returns informations about the cluster
|
1057 |
|
1058 |
@rtype: L{objects.Cluster}
|
1059 |
@return: the cluster object
|
1060 |
|
1061 |
"""
|
1062 |
return self._config_data.cluster |
1063 |
|
1064 |
@locking.ssynchronized(_config_lock)
|
1065 |
def Update(self, target): |
1066 |
"""Notify function to be called after updates.
|
1067 |
|
1068 |
This function must be called when an object (as returned by
|
1069 |
GetInstanceInfo, GetNodeInfo, GetCluster) has been updated and the
|
1070 |
caller wants the modifications saved to the backing store. Note
|
1071 |
that all modified objects will be saved, but the target argument
|
1072 |
is the one the caller wants to ensure that it's saved.
|
1073 |
|
1074 |
@param target: an instance of either L{objects.Cluster},
|
1075 |
L{objects.Node} or L{objects.Instance} which is existing in
|
1076 |
the cluster
|
1077 |
|
1078 |
"""
|
1079 |
if self._config_data is None: |
1080 |
raise errors.ProgrammerError("Configuration file not read," |
1081 |
" cannot save.")
|
1082 |
update_serial = False
|
1083 |
if isinstance(target, objects.Cluster): |
1084 |
test = target == self._config_data.cluster
|
1085 |
elif isinstance(target, objects.Node): |
1086 |
test = target in self._config_data.nodes.values() |
1087 |
update_serial = True
|
1088 |
elif isinstance(target, objects.Instance): |
1089 |
test = target in self._config_data.instances.values() |
1090 |
else:
|
1091 |
raise errors.ProgrammerError("Invalid object type (%s) passed to" |
1092 |
" ConfigWriter.Update" % type(target)) |
1093 |
if not test: |
1094 |
raise errors.ConfigurationError("Configuration updated since object" |
1095 |
" has been read or unknown object")
|
1096 |
target.serial_no += 1
|
1097 |
|
1098 |
if update_serial:
|
1099 |
# for node updates, we need to increase the cluster serial too
|
1100 |
self._config_data.cluster.serial_no += 1 |
1101 |
|
1102 |
self._WriteConfig()
|