root / lib / bootstrap.py @ 7b3a8fb5
History | View | Annotate | Download (13.2 kB)
1 |
#
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007, 2008 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 |
"""Functions to bootstrap a new cluster.
|
23 |
|
24 |
"""
|
25 |
|
26 |
import os |
27 |
import os.path |
28 |
import sha |
29 |
import re |
30 |
import logging |
31 |
|
32 |
from ganeti import rpc |
33 |
from ganeti import ssh |
34 |
from ganeti import utils |
35 |
from ganeti import errors |
36 |
from ganeti import config |
37 |
from ganeti import constants |
38 |
from ganeti import objects |
39 |
from ganeti import ssconf |
40 |
|
41 |
|
42 |
def _InitSSHSetup(node): |
43 |
"""Setup the SSH configuration for the cluster.
|
44 |
|
45 |
|
46 |
This generates a dsa keypair for root, adds the pub key to the
|
47 |
permitted hosts and adds the hostkey to its own known hosts.
|
48 |
|
49 |
Args:
|
50 |
node: the name of this host as a fqdn
|
51 |
|
52 |
"""
|
53 |
priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS) |
54 |
|
55 |
for name in priv_key, pub_key: |
56 |
if os.path.exists(name):
|
57 |
utils.CreateBackup(name) |
58 |
utils.RemoveFile(name) |
59 |
|
60 |
result = utils.RunCmd(["ssh-keygen", "-t", "dsa", |
61 |
"-f", priv_key,
|
62 |
"-q", "-N", ""]) |
63 |
if result.failed:
|
64 |
raise errors.OpExecError("Could not generate ssh keypair, error %s" % |
65 |
result.output) |
66 |
|
67 |
f = open(pub_key, 'r') |
68 |
try:
|
69 |
utils.AddAuthorizedKey(auth_keys, f.read(8192))
|
70 |
finally:
|
71 |
f.close() |
72 |
|
73 |
|
74 |
def _InitGanetiServerSetup(): |
75 |
"""Setup the necessary configuration for the initial node daemon.
|
76 |
|
77 |
This creates the nodepass file containing the shared password for
|
78 |
the cluster and also generates the SSL certificate.
|
79 |
|
80 |
"""
|
81 |
# Create pseudo random password
|
82 |
randpass = utils.GenerateSecret() |
83 |
|
84 |
# and write it into the config file
|
85 |
utils.WriteFile(constants.CLUSTER_PASSWORD_FILE, |
86 |
data="%s\n" % randpass, mode=0400) |
87 |
|
88 |
result = utils.RunCmd(["openssl", "req", "-new", "-newkey", "rsa:1024", |
89 |
"-days", str(365*5), "-nodes", "-x509", |
90 |
"-keyout", constants.SSL_CERT_FILE,
|
91 |
"-out", constants.SSL_CERT_FILE, "-batch"]) |
92 |
if result.failed:
|
93 |
raise errors.OpExecError("could not generate server ssl cert, command" |
94 |
" %s had exitcode %s and error message %s" %
|
95 |
(result.cmd, result.exit_code, result.output)) |
96 |
|
97 |
os.chmod(constants.SSL_CERT_FILE, 0400)
|
98 |
|
99 |
result = utils.RunCmd([constants.NODE_INITD_SCRIPT, "restart"])
|
100 |
|
101 |
if result.failed:
|
102 |
raise errors.OpExecError("Could not start the node daemon, command %s" |
103 |
" had exitcode %s and error %s" %
|
104 |
(result.cmd, result.exit_code, result.output)) |
105 |
|
106 |
|
107 |
def InitCluster(cluster_name, hypervisor_type, mac_prefix, def_bridge, |
108 |
master_netdev, file_storage_dir, |
109 |
secondary_ip=None,
|
110 |
vg_name=None):
|
111 |
"""Initialise the cluster.
|
112 |
|
113 |
"""
|
114 |
if config.ConfigWriter.IsCluster():
|
115 |
raise errors.OpPrereqError("Cluster is already initialised") |
116 |
|
117 |
if hypervisor_type == constants.HT_XEN_HVM:
|
118 |
if not os.path.exists(constants.VNC_PASSWORD_FILE): |
119 |
raise errors.OpPrereqError("Please prepare the cluster VNC" |
120 |
"password file %s" %
|
121 |
constants.VNC_PASSWORD_FILE) |
122 |
|
123 |
hostname = utils.HostInfo() |
124 |
|
125 |
if hostname.ip.startswith("127."): |
126 |
raise errors.OpPrereqError("This host's IP resolves to the private" |
127 |
" range (%s). Please fix DNS or %s." %
|
128 |
(hostname.ip, constants.ETC_HOSTS)) |
129 |
|
130 |
if not utils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT, |
131 |
source=constants.LOCALHOST_IP_ADDRESS): |
132 |
raise errors.OpPrereqError("Inconsistency: this host's name resolves" |
133 |
" to %s,\nbut this ip address does not"
|
134 |
" belong to this host."
|
135 |
" Aborting." % hostname.ip)
|
136 |
|
137 |
clustername = utils.HostInfo(cluster_name) |
138 |
|
139 |
if utils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT,
|
140 |
timeout=5):
|
141 |
raise errors.OpPrereqError("Cluster IP already active. Aborting.") |
142 |
|
143 |
if secondary_ip:
|
144 |
if not utils.IsValidIP(secondary_ip): |
145 |
raise errors.OpPrereqError("Invalid secondary ip given") |
146 |
if (secondary_ip != hostname.ip and |
147 |
(not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
|
148 |
source=constants.LOCALHOST_IP_ADDRESS))): |
149 |
raise errors.OpPrereqError("You gave %s as secondary IP," |
150 |
" but it does not belong to this host." %
|
151 |
secondary_ip) |
152 |
else:
|
153 |
secondary_ip = hostname.ip |
154 |
|
155 |
if vg_name is not None: |
156 |
# Check if volume group is valid
|
157 |
vgstatus = utils.CheckVolumeGroupSize(utils.ListVolumeGroups(), vg_name, |
158 |
constants.MIN_VG_SIZE) |
159 |
if vgstatus:
|
160 |
raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if" |
161 |
" you are not using lvm" % vgstatus)
|
162 |
|
163 |
file_storage_dir = os.path.normpath(file_storage_dir) |
164 |
|
165 |
if not os.path.isabs(file_storage_dir): |
166 |
raise errors.OpPrereqError("The file storage directory you passed is" |
167 |
" not an absolute path.")
|
168 |
|
169 |
if not os.path.exists(file_storage_dir): |
170 |
try:
|
171 |
os.makedirs(file_storage_dir, 0750)
|
172 |
except OSError, err: |
173 |
raise errors.OpPrereqError("Cannot create file storage directory" |
174 |
" '%s': %s" %
|
175 |
(file_storage_dir, err)) |
176 |
|
177 |
if not os.path.isdir(file_storage_dir): |
178 |
raise errors.OpPrereqError("The file storage directory '%s' is not" |
179 |
" a directory." % file_storage_dir)
|
180 |
|
181 |
if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", mac_prefix): |
182 |
raise errors.OpPrereqError("Invalid mac prefix given '%s'" % mac_prefix) |
183 |
|
184 |
if hypervisor_type not in constants.HYPER_TYPES: |
185 |
raise errors.OpPrereqError("Invalid hypervisor type given '%s'" % |
186 |
hypervisor_type) |
187 |
|
188 |
result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev]) |
189 |
if result.failed:
|
190 |
raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" % |
191 |
(master_netdev, |
192 |
result.output.strip())) |
193 |
|
194 |
if not (os.path.isfile(constants.NODE_INITD_SCRIPT) and |
195 |
os.access(constants.NODE_INITD_SCRIPT, os.X_OK)): |
196 |
raise errors.OpPrereqError("Init.d script '%s' missing or not" |
197 |
" executable." % constants.NODE_INITD_SCRIPT)
|
198 |
|
199 |
# set up the inter-node password and certificate
|
200 |
_InitGanetiServerSetup() |
201 |
|
202 |
# set up ssh config and /etc/hosts
|
203 |
f = open(constants.SSH_HOST_RSA_PUB, 'r') |
204 |
try:
|
205 |
sshline = f.read() |
206 |
finally:
|
207 |
f.close() |
208 |
sshkey = sshline.split(" ")[1] |
209 |
|
210 |
utils.AddHostToEtcHosts(hostname.name) |
211 |
_InitSSHSetup(hostname.name) |
212 |
|
213 |
# init of cluster config file
|
214 |
cluster_config = objects.Cluster( |
215 |
serial_no=1,
|
216 |
rsahostkeypub=sshkey, |
217 |
highest_used_port=(constants.FIRST_DRBD_PORT - 1),
|
218 |
mac_prefix=mac_prefix, |
219 |
volume_group_name=vg_name, |
220 |
default_bridge=def_bridge, |
221 |
tcpudp_port_pool=set(),
|
222 |
hypervisor=hypervisor_type, |
223 |
master_node=hostname.name, |
224 |
master_ip=clustername.ip, |
225 |
master_netdev=master_netdev, |
226 |
cluster_name=clustername.name, |
227 |
file_storage_dir=file_storage_dir, |
228 |
) |
229 |
master_node_config = objects.Node(name=hostname.name, |
230 |
primary_ip=hostname.ip, |
231 |
secondary_ip=secondary_ip) |
232 |
|
233 |
cfg = InitConfig(constants.CONFIG_VERSION, |
234 |
cluster_config, master_node_config) |
235 |
ssh.WriteKnownHostsFile(cfg, constants.SSH_KNOWN_HOSTS_FILE) |
236 |
|
237 |
# start the master ip
|
238 |
# TODO: Review rpc call from bootstrap
|
239 |
rpc.call_node_start_master(hostname.name, True)
|
240 |
|
241 |
|
242 |
def InitConfig(version, cluster_config, master_node_config, |
243 |
cfg_file=constants.CLUSTER_CONF_FILE): |
244 |
"""Create the initial cluster configuration.
|
245 |
|
246 |
It will contain the current node, which will also be the master
|
247 |
node, and no instances.
|
248 |
|
249 |
@type version: int
|
250 |
@param version: Configuration version
|
251 |
@type cluster_config: objects.Cluster
|
252 |
@param cluster_config: Cluster configuration
|
253 |
@type master_node_config: objects.Node
|
254 |
@param master_node_config: Master node configuration
|
255 |
@type file_name: string
|
256 |
@param file_name: Configuration file path
|
257 |
|
258 |
@rtype: ssconf.SimpleConfigWriter
|
259 |
@returns: Initialized config instance
|
260 |
|
261 |
"""
|
262 |
nodes = { |
263 |
master_node_config.name: master_node_config, |
264 |
} |
265 |
|
266 |
config_data = objects.ConfigData(version=version, |
267 |
cluster=cluster_config, |
268 |
nodes=nodes, |
269 |
instances={}, |
270 |
serial_no=1)
|
271 |
cfg = ssconf.SimpleConfigWriter.FromDict(config_data.ToDict(), cfg_file) |
272 |
cfg.Save() |
273 |
|
274 |
return cfg
|
275 |
|
276 |
|
277 |
def FinalizeClusterDestroy(master): |
278 |
"""Execute the last steps of cluster destroy
|
279 |
|
280 |
This function shuts down all the daemons, completing the destroy
|
281 |
begun in cmdlib.LUDestroyOpcode.
|
282 |
|
283 |
"""
|
284 |
if not rpc.call_node_stop_master(master, True): |
285 |
logging.warning("Could not disable the master role")
|
286 |
if not rpc.call_node_leave_cluster(master): |
287 |
logging.warning("Could not shutdown the node daemon and cleanup the node")
|
288 |
|
289 |
|
290 |
def SetupNodeDaemon(node, ssh_key_check): |
291 |
"""Add a node to the cluster.
|
292 |
|
293 |
This function must be called before the actual opcode, and will ssh
|
294 |
to the remote node, copy the needed files, and start ganeti-noded,
|
295 |
allowing the master to do the rest via normal rpc calls.
|
296 |
|
297 |
Args:
|
298 |
node: fully qualified domain name for the new node
|
299 |
|
300 |
"""
|
301 |
cfg = ssconf.SimpleConfigReader() |
302 |
sshrunner = ssh.SshRunner(cfg.GetClusterName()) |
303 |
gntpass = utils.GetNodeDaemonPassword() |
304 |
if not re.match('^[a-zA-Z0-9.]{1,64}$', gntpass): |
305 |
raise errors.OpExecError("ganeti password corruption detected") |
306 |
f = open(constants.SSL_CERT_FILE)
|
307 |
try:
|
308 |
gntpem = f.read(8192)
|
309 |
finally:
|
310 |
f.close() |
311 |
# in the base64 pem encoding, neither '!' nor '.' are valid chars,
|
312 |
# so we use this to detect an invalid certificate; as long as the
|
313 |
# cert doesn't contain this, the here-document will be correctly
|
314 |
# parsed by the shell sequence below
|
315 |
if re.search('^!EOF\.', gntpem, re.MULTILINE): |
316 |
raise errors.OpExecError("invalid PEM encoding in the SSL certificate") |
317 |
if not gntpem.endswith("\n"): |
318 |
raise errors.OpExecError("PEM must end with newline") |
319 |
|
320 |
# set up inter-node password and certificate and restarts the node daemon
|
321 |
# and then connect with ssh to set password and start ganeti-noded
|
322 |
# note that all the below variables are sanitized at this point,
|
323 |
# either by being constants or by the checks above
|
324 |
mycommand = ("umask 077 && "
|
325 |
"echo '%s' > '%s' && "
|
326 |
"cat > '%s' << '!EOF.' && \n"
|
327 |
"%s!EOF.\n%s restart" %
|
328 |
(gntpass, constants.CLUSTER_PASSWORD_FILE, |
329 |
constants.SSL_CERT_FILE, gntpem, |
330 |
constants.NODE_INITD_SCRIPT)) |
331 |
|
332 |
result = sshrunner.Run(node, 'root', mycommand, batch=False, |
333 |
ask_key=ssh_key_check, |
334 |
use_cluster_key=False,
|
335 |
strict_host_check=ssh_key_check) |
336 |
if result.failed:
|
337 |
raise errors.OpExecError("Remote command on node %s, error: %s," |
338 |
" output: %s" %
|
339 |
(node, result.fail_reason, result.output)) |
340 |
|
341 |
return 0 |
342 |
|
343 |
|
344 |
def MasterFailover(): |
345 |
"""Failover the master node.
|
346 |
|
347 |
This checks that we are not already the master, and will cause the
|
348 |
current master to cease being master, and the non-master to become
|
349 |
new master.
|
350 |
|
351 |
"""
|
352 |
cfg = ssconf.SimpleConfigWriter() |
353 |
|
354 |
new_master = utils.HostInfo().name |
355 |
old_master = cfg.GetMasterNode() |
356 |
|
357 |
if old_master == new_master:
|
358 |
raise errors.OpPrereqError("This commands must be run on the node" |
359 |
" where you want the new master to be."
|
360 |
" %s is already the master" %
|
361 |
old_master) |
362 |
# end checks
|
363 |
|
364 |
rcode = 0
|
365 |
|
366 |
logging.info("setting master to %s, old master: %s", new_master, old_master)
|
367 |
|
368 |
if not rpc.call_node_stop_master(old_master, True): |
369 |
logging.error("could disable the master role on the old master"
|
370 |
" %s, please disable manually", old_master)
|
371 |
|
372 |
cfg.SetMasterNode(new_master) |
373 |
cfg.Save() |
374 |
|
375 |
# Here we have a phase where no master should be running
|
376 |
|
377 |
if not rpc.call_upload_file(cfg.GetNodeList(), |
378 |
constants.CLUSTER_CONF_FILE): |
379 |
logging.error("could not distribute the new simple store master file"
|
380 |
" to the other nodes, please check.")
|
381 |
|
382 |
if not rpc.call_node_start_master(new_master, True): |
383 |
logging.error("could not start the master role on the new master"
|
384 |
" %s, please check", new_master)
|
385 |
rcode = 1
|
386 |
|
387 |
return rcode
|