root / lib / bootstrap.py @ b119bccb
History | View | Annotate | Download (18.6 kB)
1 | a0c9f010 | Michael Hanselmann | #
|
---|---|---|---|
2 | a0c9f010 | Michael Hanselmann | #
|
3 | a0c9f010 | Michael Hanselmann | |
4 | a0c9f010 | Michael Hanselmann | # Copyright (C) 2006, 2007, 2008 Google Inc.
|
5 | a0c9f010 | Michael Hanselmann | #
|
6 | a0c9f010 | Michael Hanselmann | # This program is free software; you can redistribute it and/or modify
|
7 | a0c9f010 | Michael Hanselmann | # it under the terms of the GNU General Public License as published by
|
8 | a0c9f010 | Michael Hanselmann | # the Free Software Foundation; either version 2 of the License, or
|
9 | a0c9f010 | Michael Hanselmann | # (at your option) any later version.
|
10 | a0c9f010 | Michael Hanselmann | #
|
11 | a0c9f010 | Michael Hanselmann | # This program is distributed in the hope that it will be useful, but
|
12 | a0c9f010 | Michael Hanselmann | # WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 | a0c9f010 | Michael Hanselmann | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 | a0c9f010 | Michael Hanselmann | # General Public License for more details.
|
15 | a0c9f010 | Michael Hanselmann | #
|
16 | a0c9f010 | Michael Hanselmann | # You should have received a copy of the GNU General Public License
|
17 | a0c9f010 | Michael Hanselmann | # along with this program; if not, write to the Free Software
|
18 | a0c9f010 | Michael Hanselmann | # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
19 | a0c9f010 | Michael Hanselmann | # 02110-1301, USA.
|
20 | a0c9f010 | Michael Hanselmann | |
21 | a0c9f010 | Michael Hanselmann | |
22 | a0c9f010 | Michael Hanselmann | """Functions to bootstrap a new cluster.
|
23 | a0c9f010 | Michael Hanselmann |
|
24 | a0c9f010 | Michael Hanselmann | """
|
25 | a0c9f010 | Michael Hanselmann | |
26 | a0c9f010 | Michael Hanselmann | import os |
27 | a0c9f010 | Michael Hanselmann | import os.path |
28 | a0c9f010 | Michael Hanselmann | import re |
29 | b1b6ea87 | Iustin Pop | import logging |
30 | c4415fd5 | Michael Hanselmann | import tempfile |
31 | a0c9f010 | Michael Hanselmann | |
32 | a0c9f010 | Michael Hanselmann | from ganeti import rpc |
33 | a0c9f010 | Michael Hanselmann | from ganeti import ssh |
34 | a0c9f010 | Michael Hanselmann | from ganeti import utils |
35 | a0c9f010 | Michael Hanselmann | from ganeti import errors |
36 | a0c9f010 | Michael Hanselmann | from ganeti import config |
37 | a0c9f010 | Michael Hanselmann | from ganeti import constants |
38 | b9eeeb02 | Michael Hanselmann | from ganeti import objects |
39 | a0c9f010 | Michael Hanselmann | from ganeti import ssconf |
40 | a5728081 | Guido Trotter | from ganeti import hypervisor |
41 | a0c9f010 | Michael Hanselmann | |
42 | e38220e4 | Michael Hanselmann | |
43 | 531baf8e | Iustin Pop | def _InitSSHSetup(): |
44 | a0c9f010 | Michael Hanselmann | """Setup the SSH configuration for the cluster.
|
45 | a0c9f010 | Michael Hanselmann |
|
46 | a0c9f010 | Michael Hanselmann | This generates a dsa keypair for root, adds the pub key to the
|
47 | a0c9f010 | Michael Hanselmann | permitted hosts and adds the hostkey to its own known hosts.
|
48 | a0c9f010 | Michael Hanselmann |
|
49 | a0c9f010 | Michael Hanselmann | """
|
50 | a0c9f010 | Michael Hanselmann | priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS) |
51 | a0c9f010 | Michael Hanselmann | |
52 | a0c9f010 | Michael Hanselmann | for name in priv_key, pub_key: |
53 | a0c9f010 | Michael Hanselmann | if os.path.exists(name):
|
54 | a0c9f010 | Michael Hanselmann | utils.CreateBackup(name) |
55 | a0c9f010 | Michael Hanselmann | utils.RemoveFile(name) |
56 | a0c9f010 | Michael Hanselmann | |
57 | a0c9f010 | Michael Hanselmann | result = utils.RunCmd(["ssh-keygen", "-t", "dsa", |
58 | a0c9f010 | Michael Hanselmann | "-f", priv_key,
|
59 | a0c9f010 | Michael Hanselmann | "-q", "-N", ""]) |
60 | a0c9f010 | Michael Hanselmann | if result.failed:
|
61 | a0c9f010 | Michael Hanselmann | raise errors.OpExecError("Could not generate ssh keypair, error %s" % |
62 | a0c9f010 | Michael Hanselmann | result.output) |
63 | a0c9f010 | Michael Hanselmann | |
64 | a0c9f010 | Michael Hanselmann | f = open(pub_key, 'r') |
65 | a0c9f010 | Michael Hanselmann | try:
|
66 | a0c9f010 | Michael Hanselmann | utils.AddAuthorizedKey(auth_keys, f.read(8192))
|
67 | a0c9f010 | Michael Hanselmann | finally:
|
68 | a0c9f010 | Michael Hanselmann | f.close() |
69 | a0c9f010 | Michael Hanselmann | |
70 | a0c9f010 | Michael Hanselmann | |
71 | 40a97d80 | Michael Hanselmann | def _GenerateSelfSignedSslCert(file_name, validity=(365 * 5)): |
72 | 40a97d80 | Michael Hanselmann | """Generates a self-signed SSL certificate.
|
73 | a0c9f010 | Michael Hanselmann |
|
74 | 40a97d80 | Michael Hanselmann | @type file_name: str
|
75 | 40a97d80 | Michael Hanselmann | @param file_name: Path to output file
|
76 | 40a97d80 | Michael Hanselmann | @type validity: int
|
77 | 40a97d80 | Michael Hanselmann | @param validity: Validity for certificate in days
|
78 | a0c9f010 | Michael Hanselmann |
|
79 | a0c9f010 | Michael Hanselmann | """
|
80 | c4415fd5 | Michael Hanselmann | (fd, tmp_file_name) = tempfile.mkstemp(dir=os.path.dirname(file_name)) |
81 | c4415fd5 | Michael Hanselmann | try:
|
82 | 88828491 | Michael Hanselmann | try:
|
83 | 88828491 | Michael Hanselmann | # Set permissions before writing key
|
84 | 88828491 | Michael Hanselmann | os.chmod(tmp_file_name, 0600)
|
85 | 88828491 | Michael Hanselmann | |
86 | 88828491 | Michael Hanselmann | result = utils.RunCmd(["openssl", "req", "-new", "-newkey", "rsa:1024", |
87 | 88828491 | Michael Hanselmann | "-days", str(validity), "-nodes", "-x509", |
88 | 88828491 | Michael Hanselmann | "-keyout", tmp_file_name, "-out", tmp_file_name, |
89 | 88828491 | Michael Hanselmann | "-batch"])
|
90 | 88828491 | Michael Hanselmann | if result.failed:
|
91 | 88828491 | Michael Hanselmann | raise errors.OpExecError("Could not generate SSL certificate, command" |
92 | 88828491 | Michael Hanselmann | " %s had exitcode %s and error message %s" %
|
93 | 88828491 | Michael Hanselmann | (result.cmd, result.exit_code, result.output)) |
94 | 88828491 | Michael Hanselmann | |
95 | 88828491 | Michael Hanselmann | # Make read-only
|
96 | 88828491 | Michael Hanselmann | os.chmod(tmp_file_name, 0400)
|
97 | 88828491 | Michael Hanselmann | |
98 | 88828491 | Michael Hanselmann | os.rename(tmp_file_name, file_name) |
99 | 88828491 | Michael Hanselmann | finally:
|
100 | 88828491 | Michael Hanselmann | utils.RemoveFile(tmp_file_name) |
101 | c4415fd5 | Michael Hanselmann | finally:
|
102 | 88828491 | Michael Hanselmann | os.close(fd) |
103 | 40a97d80 | Michael Hanselmann | |
104 | 40a97d80 | Michael Hanselmann | |
105 | 40a97d80 | Michael Hanselmann | def _InitGanetiServerSetup(): |
106 | 40a97d80 | Michael Hanselmann | """Setup the necessary configuration for the initial node daemon.
|
107 | 40a97d80 | Michael Hanselmann |
|
108 | 40a97d80 | Michael Hanselmann | This creates the nodepass file containing the shared password for
|
109 | 40a97d80 | Michael Hanselmann | the cluster and also generates the SSL certificate.
|
110 | 40a97d80 | Michael Hanselmann |
|
111 | 40a97d80 | Michael Hanselmann | """
|
112 | 40a97d80 | Michael Hanselmann | _GenerateSelfSignedSslCert(constants.SSL_CERT_FILE) |
113 | a0c9f010 | Michael Hanselmann | |
114 | 61a08fa3 | Michael Hanselmann | # Don't overwrite existing file
|
115 | 61a08fa3 | Michael Hanselmann | if not os.path.exists(constants.RAPI_CERT_FILE): |
116 | 61a08fa3 | Michael Hanselmann | _GenerateSelfSignedSslCert(constants.RAPI_CERT_FILE) |
117 | 61a08fa3 | Michael Hanselmann | |
118 | a0c9f010 | Michael Hanselmann | result = utils.RunCmd([constants.NODE_INITD_SCRIPT, "restart"])
|
119 | a0c9f010 | Michael Hanselmann | |
120 | a0c9f010 | Michael Hanselmann | if result.failed:
|
121 | a0c9f010 | Michael Hanselmann | raise errors.OpExecError("Could not start the node daemon, command %s" |
122 | a0c9f010 | Michael Hanselmann | " had exitcode %s and error %s" %
|
123 | a0c9f010 | Michael Hanselmann | (result.cmd, result.exit_code, result.output)) |
124 | a0c9f010 | Michael Hanselmann | |
125 | a0c9f010 | Michael Hanselmann | |
126 | 4342e89b | Alexander Schreiber | def InitCluster(cluster_name, mac_prefix, def_bridge, |
127 | ce735215 | Guido Trotter | master_netdev, file_storage_dir, candidate_pool_size, |
128 | ce735215 | Guido Trotter | secondary_ip=None, vg_name=None, beparams=None, hvparams=None, |
129 | 02691904 | Alexander Schreiber | enabled_hypervisors=None, default_hypervisor=None): |
130 | a0c9f010 | Michael Hanselmann | """Initialise the cluster.
|
131 | a0c9f010 | Michael Hanselmann |
|
132 | ce735215 | Guido Trotter | @type candidate_pool_size: int
|
133 | ce735215 | Guido Trotter | @param candidate_pool_size: master candidate pool size
|
134 | ce735215 | Guido Trotter |
|
135 | a0c9f010 | Michael Hanselmann | """
|
136 | ce735215 | Guido Trotter | # TODO: complete the docstring
|
137 | a0c9f010 | Michael Hanselmann | if config.ConfigWriter.IsCluster():
|
138 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Cluster is already initialised") |
139 | a0c9f010 | Michael Hanselmann | |
140 | b119bccb | Guido Trotter | if not enabled_hypervisors: |
141 | b119bccb | Guido Trotter | raise errors.OpPrereqError("Enabled hypervisors list must contain at" |
142 | b119bccb | Guido Trotter | " least one member")
|
143 | b119bccb | Guido Trotter | invalid_hvs = set(enabled_hypervisors) - constants.HYPER_TYPES
|
144 | b119bccb | Guido Trotter | if invalid_hvs:
|
145 | b119bccb | Guido Trotter | raise errors.OpPrereqError("Enabled hypervisors contains invalid" |
146 | b119bccb | Guido Trotter | " entries: %s" % invalid_hvs)
|
147 | b119bccb | Guido Trotter | |
148 | a0c9f010 | Michael Hanselmann | hostname = utils.HostInfo() |
149 | a0c9f010 | Michael Hanselmann | |
150 | a0c9f010 | Michael Hanselmann | if hostname.ip.startswith("127."): |
151 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("This host's IP resolves to the private" |
152 | a0c9f010 | Michael Hanselmann | " range (%s). Please fix DNS or %s." %
|
153 | a0c9f010 | Michael Hanselmann | (hostname.ip, constants.ETC_HOSTS)) |
154 | a0c9f010 | Michael Hanselmann | |
155 | caad16e2 | Iustin Pop | if not utils.OwnIpAddress(hostname.ip): |
156 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Inconsistency: this host's name resolves" |
157 | a0c9f010 | Michael Hanselmann | " to %s,\nbut this ip address does not"
|
158 | a0c9f010 | Michael Hanselmann | " belong to this host."
|
159 | a0c9f010 | Michael Hanselmann | " Aborting." % hostname.ip)
|
160 | a0c9f010 | Michael Hanselmann | |
161 | a0c9f010 | Michael Hanselmann | clustername = utils.HostInfo(cluster_name) |
162 | a0c9f010 | Michael Hanselmann | |
163 | a0c9f010 | Michael Hanselmann | if utils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT,
|
164 | a0c9f010 | Michael Hanselmann | timeout=5):
|
165 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Cluster IP already active. Aborting.") |
166 | a0c9f010 | Michael Hanselmann | |
167 | a0c9f010 | Michael Hanselmann | if secondary_ip:
|
168 | a0c9f010 | Michael Hanselmann | if not utils.IsValidIP(secondary_ip): |
169 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Invalid secondary ip given") |
170 | a0c9f010 | Michael Hanselmann | if (secondary_ip != hostname.ip and |
171 | caad16e2 | Iustin Pop | not utils.OwnIpAddress(secondary_ip)):
|
172 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("You gave %s as secondary IP," |
173 | a0c9f010 | Michael Hanselmann | " but it does not belong to this host." %
|
174 | a0c9f010 | Michael Hanselmann | secondary_ip) |
175 | b9eeeb02 | Michael Hanselmann | else:
|
176 | b9eeeb02 | Michael Hanselmann | secondary_ip = hostname.ip |
177 | a0c9f010 | Michael Hanselmann | |
178 | a0c9f010 | Michael Hanselmann | if vg_name is not None: |
179 | a0c9f010 | Michael Hanselmann | # Check if volume group is valid
|
180 | a0c9f010 | Michael Hanselmann | vgstatus = utils.CheckVolumeGroupSize(utils.ListVolumeGroups(), vg_name, |
181 | a0c9f010 | Michael Hanselmann | constants.MIN_VG_SIZE) |
182 | a0c9f010 | Michael Hanselmann | if vgstatus:
|
183 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if" |
184 | a0c9f010 | Michael Hanselmann | " you are not using lvm" % vgstatus)
|
185 | a0c9f010 | Michael Hanselmann | |
186 | a0c9f010 | Michael Hanselmann | file_storage_dir = os.path.normpath(file_storage_dir) |
187 | a0c9f010 | Michael Hanselmann | |
188 | a0c9f010 | Michael Hanselmann | if not os.path.isabs(file_storage_dir): |
189 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("The file storage directory you passed is" |
190 | a0c9f010 | Michael Hanselmann | " not an absolute path.")
|
191 | a0c9f010 | Michael Hanselmann | |
192 | a0c9f010 | Michael Hanselmann | if not os.path.exists(file_storage_dir): |
193 | a0c9f010 | Michael Hanselmann | try:
|
194 | a0c9f010 | Michael Hanselmann | os.makedirs(file_storage_dir, 0750)
|
195 | a0c9f010 | Michael Hanselmann | except OSError, err: |
196 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Cannot create file storage directory" |
197 | a0c9f010 | Michael Hanselmann | " '%s': %s" %
|
198 | a0c9f010 | Michael Hanselmann | (file_storage_dir, err)) |
199 | a0c9f010 | Michael Hanselmann | |
200 | a0c9f010 | Michael Hanselmann | if not os.path.isdir(file_storage_dir): |
201 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("The file storage directory '%s' is not" |
202 | a0c9f010 | Michael Hanselmann | " a directory." % file_storage_dir)
|
203 | a0c9f010 | Michael Hanselmann | |
204 | a0c9f010 | Michael Hanselmann | if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", mac_prefix): |
205 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Invalid mac prefix given '%s'" % mac_prefix) |
206 | a0c9f010 | Michael Hanselmann | |
207 | a0c9f010 | Michael Hanselmann | result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev]) |
208 | a0c9f010 | Michael Hanselmann | if result.failed:
|
209 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" % |
210 | a0c9f010 | Michael Hanselmann | (master_netdev, |
211 | a0c9f010 | Michael Hanselmann | result.output.strip())) |
212 | a0c9f010 | Michael Hanselmann | |
213 | a0c9f010 | Michael Hanselmann | if not (os.path.isfile(constants.NODE_INITD_SCRIPT) and |
214 | a0c9f010 | Michael Hanselmann | os.access(constants.NODE_INITD_SCRIPT, os.X_OK)): |
215 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Init.d script '%s' missing or not" |
216 | a0c9f010 | Michael Hanselmann | " executable." % constants.NODE_INITD_SCRIPT)
|
217 | a0c9f010 | Michael Hanselmann | |
218 | 9dae41ad | Guido Trotter | dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE)] |
219 | 9dae41ad | Guido Trotter | utils.EnsureDirs(dirs) |
220 | 9dae41ad | Guido Trotter | |
221 | a5728081 | Guido Trotter | utils.ForceDictType(beparams, constants.BES_PARAMETER_TYPES) |
222 | a5728081 | Guido Trotter | # hvparams is a mapping of hypervisor->hvparams dict
|
223 | a5728081 | Guido Trotter | for hv_name, hv_params in hvparams.iteritems(): |
224 | a5728081 | Guido Trotter | utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES) |
225 | a5728081 | Guido Trotter | hv_class = hypervisor.GetHypervisor(hv_name) |
226 | a5728081 | Guido Trotter | hv_class.CheckParameterSyntax(hv_params) |
227 | d4b72030 | Guido Trotter | |
228 | a0c9f010 | Michael Hanselmann | # set up the inter-node password and certificate
|
229 | d23ef431 | Michael Hanselmann | _InitGanetiServerSetup() |
230 | a0c9f010 | Michael Hanselmann | |
231 | a0c9f010 | Michael Hanselmann | # set up ssh config and /etc/hosts
|
232 | a0c9f010 | Michael Hanselmann | f = open(constants.SSH_HOST_RSA_PUB, 'r') |
233 | a0c9f010 | Michael Hanselmann | try:
|
234 | a0c9f010 | Michael Hanselmann | sshline = f.read() |
235 | a0c9f010 | Michael Hanselmann | finally:
|
236 | a0c9f010 | Michael Hanselmann | f.close() |
237 | a0c9f010 | Michael Hanselmann | sshkey = sshline.split(" ")[1] |
238 | a0c9f010 | Michael Hanselmann | |
239 | a0c9f010 | Michael Hanselmann | utils.AddHostToEtcHosts(hostname.name) |
240 | 531baf8e | Iustin Pop | _InitSSHSetup() |
241 | a0c9f010 | Michael Hanselmann | |
242 | a0c9f010 | Michael Hanselmann | # init of cluster config file
|
243 | b9eeeb02 | Michael Hanselmann | cluster_config = objects.Cluster( |
244 | b9eeeb02 | Michael Hanselmann | serial_no=1,
|
245 | b9eeeb02 | Michael Hanselmann | rsahostkeypub=sshkey, |
246 | b9eeeb02 | Michael Hanselmann | highest_used_port=(constants.FIRST_DRBD_PORT - 1),
|
247 | b9eeeb02 | Michael Hanselmann | mac_prefix=mac_prefix, |
248 | b9eeeb02 | Michael Hanselmann | volume_group_name=vg_name, |
249 | b9eeeb02 | Michael Hanselmann | default_bridge=def_bridge, |
250 | b9eeeb02 | Michael Hanselmann | tcpudp_port_pool=set(),
|
251 | f6bd6e98 | Michael Hanselmann | master_node=hostname.name, |
252 | f6bd6e98 | Michael Hanselmann | master_ip=clustername.ip, |
253 | f6bd6e98 | Michael Hanselmann | master_netdev=master_netdev, |
254 | f6bd6e98 | Michael Hanselmann | cluster_name=clustername.name, |
255 | f6bd6e98 | Michael Hanselmann | file_storage_dir=file_storage_dir, |
256 | ea3a925f | Alexander Schreiber | enabled_hypervisors=enabled_hypervisors, |
257 | 02691904 | Alexander Schreiber | default_hypervisor=default_hypervisor, |
258 | ea3a925f | Alexander Schreiber | beparams={constants.BEGR_DEFAULT: beparams}, |
259 | ea3a925f | Alexander Schreiber | hvparams=hvparams, |
260 | ce735215 | Guido Trotter | candidate_pool_size=candidate_pool_size, |
261 | b9eeeb02 | Michael Hanselmann | ) |
262 | b9eeeb02 | Michael Hanselmann | master_node_config = objects.Node(name=hostname.name, |
263 | b9eeeb02 | Michael Hanselmann | primary_ip=hostname.ip, |
264 | b9222f32 | Guido Trotter | secondary_ip=secondary_ip, |
265 | c044f32c | Guido Trotter | serial_no=1,
|
266 | c044f32c | Guido Trotter | master_candidate=True,
|
267 | af64c0ea | Iustin Pop | offline=False, drained=False, |
268 | c044f32c | Guido Trotter | ) |
269 | a0c9f010 | Michael Hanselmann | |
270 | 05cc153f | Guido Trotter | sscfg = InitConfig(constants.CONFIG_VERSION, |
271 | 05cc153f | Guido Trotter | cluster_config, master_node_config) |
272 | 05cc153f | Guido Trotter | ssh.WriteKnownHostsFile(sscfg, constants.SSH_KNOWN_HOSTS_FILE) |
273 | 05cc153f | Guido Trotter | cfg = config.ConfigWriter() |
274 | 05cc153f | Guido Trotter | cfg.Update(cfg.GetClusterInfo()) |
275 | 827f753e | Guido Trotter | |
276 | b3f1cf6f | Iustin Pop | # start the master ip
|
277 | b3f1cf6f | Iustin Pop | # TODO: Review rpc call from bootstrap
|
278 | fda5f19f | Michael Hanselmann | rpc.RpcRunner.call_node_start_master(hostname.name, True)
|
279 | b3f1cf6f | Iustin Pop | |
280 | b1b6ea87 | Iustin Pop | |
281 | 02f99608 | Oleksiy Mishchenko | def InitConfig(version, cluster_config, master_node_config, |
282 | 02f99608 | Oleksiy Mishchenko | cfg_file=constants.CLUSTER_CONF_FILE): |
283 | 7b3a8fb5 | Iustin Pop | """Create the initial cluster configuration.
|
284 | 7b3a8fb5 | Iustin Pop |
|
285 | 7b3a8fb5 | Iustin Pop | It will contain the current node, which will also be the master
|
286 | 7b3a8fb5 | Iustin Pop | node, and no instances.
|
287 | 7b3a8fb5 | Iustin Pop |
|
288 | 7b3a8fb5 | Iustin Pop | @type version: int
|
289 | c41eea6e | Iustin Pop | @param version: configuration version
|
290 | c41eea6e | Iustin Pop | @type cluster_config: L{objects.Cluster}
|
291 | c41eea6e | Iustin Pop | @param cluster_config: cluster configuration
|
292 | c41eea6e | Iustin Pop | @type master_node_config: L{objects.Node}
|
293 | c41eea6e | Iustin Pop | @param master_node_config: master node configuration
|
294 | c41eea6e | Iustin Pop | @type cfg_file: string
|
295 | c41eea6e | Iustin Pop | @param cfg_file: configuration file path
|
296 | c41eea6e | Iustin Pop |
|
297 | c41eea6e | Iustin Pop | @rtype: L{ssconf.SimpleConfigWriter}
|
298 | 5fcc718f | Iustin Pop | @return: initialized config instance
|
299 | 7b3a8fb5 | Iustin Pop |
|
300 | 7b3a8fb5 | Iustin Pop | """
|
301 | 7b3a8fb5 | Iustin Pop | nodes = { |
302 | 7b3a8fb5 | Iustin Pop | master_node_config.name: master_node_config, |
303 | 7b3a8fb5 | Iustin Pop | } |
304 | 7b3a8fb5 | Iustin Pop | |
305 | 7b3a8fb5 | Iustin Pop | config_data = objects.ConfigData(version=version, |
306 | 7b3a8fb5 | Iustin Pop | cluster=cluster_config, |
307 | 7b3a8fb5 | Iustin Pop | nodes=nodes, |
308 | 7b3a8fb5 | Iustin Pop | instances={}, |
309 | 7b3a8fb5 | Iustin Pop | serial_no=1)
|
310 | 7b3a8fb5 | Iustin Pop | cfg = ssconf.SimpleConfigWriter.FromDict(config_data.ToDict(), cfg_file) |
311 | 7b3a8fb5 | Iustin Pop | cfg.Save() |
312 | 7b3a8fb5 | Iustin Pop | |
313 | 7b3a8fb5 | Iustin Pop | return cfg
|
314 | 02f99608 | Oleksiy Mishchenko | |
315 | 02f99608 | Oleksiy Mishchenko | |
316 | 140aa4a8 | Iustin Pop | def FinalizeClusterDestroy(master): |
317 | 140aa4a8 | Iustin Pop | """Execute the last steps of cluster destroy
|
318 | 140aa4a8 | Iustin Pop |
|
319 | 140aa4a8 | Iustin Pop | This function shuts down all the daemons, completing the destroy
|
320 | 140aa4a8 | Iustin Pop | begun in cmdlib.LUDestroyOpcode.
|
321 | 140aa4a8 | Iustin Pop |
|
322 | 140aa4a8 | Iustin Pop | """
|
323 | 781de953 | Iustin Pop | result = rpc.RpcRunner.call_node_stop_master(master, True)
|
324 | 781de953 | Iustin Pop | if result.failed or not result.data: |
325 | 140aa4a8 | Iustin Pop | logging.warning("Could not disable the master role")
|
326 | 781de953 | Iustin Pop | result = rpc.RpcRunner.call_node_leave_cluster(master) |
327 | 781de953 | Iustin Pop | if result.failed or not result.data: |
328 | 140aa4a8 | Iustin Pop | logging.warning("Could not shutdown the node daemon and cleanup the node")
|
329 | 140aa4a8 | Iustin Pop | |
330 | 140aa4a8 | Iustin Pop | |
331 | 87622829 | Iustin Pop | def SetupNodeDaemon(cluster_name, node, ssh_key_check): |
332 | 827f753e | Guido Trotter | """Add a node to the cluster.
|
333 | 827f753e | Guido Trotter |
|
334 | b1b6ea87 | Iustin Pop | This function must be called before the actual opcode, and will ssh
|
335 | b1b6ea87 | Iustin Pop | to the remote node, copy the needed files, and start ganeti-noded,
|
336 | b1b6ea87 | Iustin Pop | allowing the master to do the rest via normal rpc calls.
|
337 | 827f753e | Guido Trotter |
|
338 | 87622829 | Iustin Pop | @param cluster_name: the cluster name
|
339 | 87622829 | Iustin Pop | @param node: the name of the new node
|
340 | 87622829 | Iustin Pop | @param ssh_key_check: whether to do a strict key check
|
341 | 827f753e | Guido Trotter |
|
342 | 827f753e | Guido Trotter | """
|
343 | 87622829 | Iustin Pop | sshrunner = ssh.SshRunner(cluster_name) |
344 | 5557b04c | Michael Hanselmann | |
345 | 5557b04c | Michael Hanselmann | noded_cert = utils.ReadFile(constants.SSL_CERT_FILE) |
346 | 2438c157 | Michael Hanselmann | rapi_cert = utils.ReadFile(constants.RAPI_CERT_FILE) |
347 | 5557b04c | Michael Hanselmann | |
348 | 827f753e | Guido Trotter | # in the base64 pem encoding, neither '!' nor '.' are valid chars,
|
349 | 827f753e | Guido Trotter | # so we use this to detect an invalid certificate; as long as the
|
350 | 827f753e | Guido Trotter | # cert doesn't contain this, the here-document will be correctly
|
351 | 827f753e | Guido Trotter | # parsed by the shell sequence below
|
352 | 2438c157 | Michael Hanselmann | if (re.search('^!EOF\.', noded_cert, re.MULTILINE) or |
353 | 2438c157 | Michael Hanselmann | re.search('^!EOF\.', rapi_cert, re.MULTILINE)):
|
354 | 827f753e | Guido Trotter | raise errors.OpExecError("invalid PEM encoding in the SSL certificate") |
355 | 5557b04c | Michael Hanselmann | |
356 | 5557b04c | Michael Hanselmann | if not noded_cert.endswith("\n"): |
357 | 5557b04c | Michael Hanselmann | noded_cert += "\n"
|
358 | 2438c157 | Michael Hanselmann | if not rapi_cert.endswith("\n"): |
359 | 2438c157 | Michael Hanselmann | rapi_cert += "\n"
|
360 | 827f753e | Guido Trotter | |
361 | 827f753e | Guido Trotter | # set up inter-node password and certificate and restarts the node daemon
|
362 | 827f753e | Guido Trotter | # and then connect with ssh to set password and start ganeti-noded
|
363 | 827f753e | Guido Trotter | # note that all the below variables are sanitized at this point,
|
364 | 827f753e | Guido Trotter | # either by being constants or by the checks above
|
365 | 827f753e | Guido Trotter | mycommand = ("umask 077 && "
|
366 | 827f753e | Guido Trotter | "cat > '%s' << '!EOF.' && \n"
|
367 | 2438c157 | Michael Hanselmann | "%s!EOF.\n"
|
368 | 2438c157 | Michael Hanselmann | "cat > '%s' << '!EOF.' && \n"
|
369 | 2438c157 | Michael Hanselmann | "%s!EOF.\n"
|
370 | 5b099da9 | Michael Hanselmann | "chmod 0400 %s %s && "
|
371 | 2438c157 | Michael Hanselmann | "%s restart" %
|
372 | 5557b04c | Michael Hanselmann | (constants.SSL_CERT_FILE, noded_cert, |
373 | 2438c157 | Michael Hanselmann | constants.RAPI_CERT_FILE, rapi_cert, |
374 | 5b099da9 | Michael Hanselmann | constants.SSL_CERT_FILE, constants.RAPI_CERT_FILE, |
375 | 827f753e | Guido Trotter | constants.NODE_INITD_SCRIPT)) |
376 | 827f753e | Guido Trotter | |
377 | c4b6c29c | Michael Hanselmann | result = sshrunner.Run(node, 'root', mycommand, batch=False, |
378 | c4b6c29c | Michael Hanselmann | ask_key=ssh_key_check, |
379 | c4b6c29c | Michael Hanselmann | use_cluster_key=False,
|
380 | c4b6c29c | Michael Hanselmann | strict_host_check=ssh_key_check) |
381 | 827f753e | Guido Trotter | if result.failed:
|
382 | 827f753e | Guido Trotter | raise errors.OpExecError("Remote command on node %s, error: %s," |
383 | 827f753e | Guido Trotter | " output: %s" %
|
384 | 827f753e | Guido Trotter | (node, result.fail_reason, result.output)) |
385 | 827f753e | Guido Trotter | |
386 | b1b6ea87 | Iustin Pop | |
387 | 8e2524c3 | Guido Trotter | def MasterFailover(no_voting=False): |
388 | b1b6ea87 | Iustin Pop | """Failover the master node.
|
389 | b1b6ea87 | Iustin Pop |
|
390 | b1b6ea87 | Iustin Pop | This checks that we are not already the master, and will cause the
|
391 | b1b6ea87 | Iustin Pop | current master to cease being master, and the non-master to become
|
392 | b1b6ea87 | Iustin Pop | new master.
|
393 | b1b6ea87 | Iustin Pop |
|
394 | 8e2524c3 | Guido Trotter | @type no_voting: boolean
|
395 | 8e2524c3 | Guido Trotter | @param no_voting: force the operation without remote nodes agreement
|
396 | 8e2524c3 | Guido Trotter | (dangerous)
|
397 | 8e2524c3 | Guido Trotter |
|
398 | b1b6ea87 | Iustin Pop | """
|
399 | 8135a2db | Iustin Pop | sstore = ssconf.SimpleStore() |
400 | b1b6ea87 | Iustin Pop | |
401 | 8135a2db | Iustin Pop | old_master, new_master = ssconf.GetMasterAndMyself(sstore) |
402 | 8135a2db | Iustin Pop | node_list = sstore.GetNodeList() |
403 | 8135a2db | Iustin Pop | mc_list = sstore.GetMasterCandidates() |
404 | b1b6ea87 | Iustin Pop | |
405 | b1b6ea87 | Iustin Pop | if old_master == new_master:
|
406 | b1b6ea87 | Iustin Pop | raise errors.OpPrereqError("This commands must be run on the node" |
407 | b1b6ea87 | Iustin Pop | " where you want the new master to be."
|
408 | b1b6ea87 | Iustin Pop | " %s is already the master" %
|
409 | b1b6ea87 | Iustin Pop | old_master) |
410 | d5927e48 | Iustin Pop | |
411 | 8135a2db | Iustin Pop | if new_master not in mc_list: |
412 | 8135a2db | Iustin Pop | mc_no_master = [name for name in mc_list if name != old_master] |
413 | 8135a2db | Iustin Pop | raise errors.OpPrereqError("This node is not among the nodes marked" |
414 | 8135a2db | Iustin Pop | " as master candidates. Only these nodes"
|
415 | 8135a2db | Iustin Pop | " can become masters. Current list of"
|
416 | 8135a2db | Iustin Pop | " master candidates is:\n"
|
417 | 8135a2db | Iustin Pop | "%s" % ('\n'.join(mc_no_master))) |
418 | 8135a2db | Iustin Pop | |
419 | 8e2524c3 | Guido Trotter | if not no_voting: |
420 | 8e2524c3 | Guido Trotter | vote_list = GatherMasterVotes(node_list) |
421 | 8e2524c3 | Guido Trotter | |
422 | 8e2524c3 | Guido Trotter | if vote_list:
|
423 | 8e2524c3 | Guido Trotter | voted_master = vote_list[0][0] |
424 | 8e2524c3 | Guido Trotter | if voted_master is None: |
425 | 8e2524c3 | Guido Trotter | raise errors.OpPrereqError("Cluster is inconsistent, most nodes did" |
426 | 8e2524c3 | Guido Trotter | " not respond.")
|
427 | 8e2524c3 | Guido Trotter | elif voted_master != old_master:
|
428 | 8e2524c3 | Guido Trotter | raise errors.OpPrereqError("I have a wrong configuration, I believe" |
429 | 8e2524c3 | Guido Trotter | " the master is %s but the other nodes"
|
430 | 8e2524c3 | Guido Trotter | " voted %s. Please resync the configuration"
|
431 | 8e2524c3 | Guido Trotter | " of this node." %
|
432 | 8e2524c3 | Guido Trotter | (old_master, voted_master)) |
433 | b1b6ea87 | Iustin Pop | # end checks
|
434 | b1b6ea87 | Iustin Pop | |
435 | b1b6ea87 | Iustin Pop | rcode = 0
|
436 | b1b6ea87 | Iustin Pop | |
437 | d5927e48 | Iustin Pop | logging.info("Setting master to %s, old master: %s", new_master, old_master)
|
438 | b1b6ea87 | Iustin Pop | |
439 | 781de953 | Iustin Pop | result = rpc.RpcRunner.call_node_stop_master(old_master, True)
|
440 | 781de953 | Iustin Pop | if result.failed or not result.data: |
441 | d5927e48 | Iustin Pop | logging.error("Could not disable the master role on the old master"
|
442 | b1b6ea87 | Iustin Pop | " %s, please disable manually", old_master)
|
443 | b1b6ea87 | Iustin Pop | |
444 | d23ef431 | Michael Hanselmann | # Here we have a phase where no master should be running
|
445 | b1b6ea87 | Iustin Pop | |
446 | bbe19c17 | Iustin Pop | # instantiate a real config writer, as we now know we have the
|
447 | bbe19c17 | Iustin Pop | # configuration data
|
448 | bbe19c17 | Iustin Pop | cfg = config.ConfigWriter() |
449 | b1b6ea87 | Iustin Pop | |
450 | bbe19c17 | Iustin Pop | cluster_info = cfg.GetClusterInfo() |
451 | bbe19c17 | Iustin Pop | cluster_info.master_node = new_master |
452 | bbe19c17 | Iustin Pop | # this will also regenerate the ssconf files, since we updated the
|
453 | bbe19c17 | Iustin Pop | # cluster info
|
454 | bbe19c17 | Iustin Pop | cfg.Update(cluster_info) |
455 | d5927e48 | Iustin Pop | |
456 | 8e2524c3 | Guido Trotter | # 2.0.X: Don't start the master if no_voting is true
|
457 | 8e2524c3 | Guido Trotter | result = rpc.RpcRunner.call_node_start_master(new_master, not no_voting)
|
458 | 781de953 | Iustin Pop | if result.failed or not result.data: |
459 | d5927e48 | Iustin Pop | logging.error("Could not start the master role on the new master"
|
460 | b1b6ea87 | Iustin Pop | " %s, please check", new_master)
|
461 | b1b6ea87 | Iustin Pop | rcode = 1
|
462 | b1b6ea87 | Iustin Pop | |
463 | b1b6ea87 | Iustin Pop | return rcode
|
464 | d7cdb55d | Iustin Pop | |
465 | d7cdb55d | Iustin Pop | |
466 | 8eb148ae | Iustin Pop | def GetMaster(): |
467 | 8eb148ae | Iustin Pop | """Returns the current master node.
|
468 | 8eb148ae | Iustin Pop |
|
469 | 8eb148ae | Iustin Pop | This is a separate function in bootstrap since it's needed by
|
470 | 8eb148ae | Iustin Pop | gnt-cluster, and instead of importing directly ssconf, it's better
|
471 | 8eb148ae | Iustin Pop | to abstract it in bootstrap, where we do use ssconf in other
|
472 | 8eb148ae | Iustin Pop | functions too.
|
473 | 8eb148ae | Iustin Pop |
|
474 | 8eb148ae | Iustin Pop | """
|
475 | 8eb148ae | Iustin Pop | sstore = ssconf.SimpleStore() |
476 | 8eb148ae | Iustin Pop | |
477 | 8eb148ae | Iustin Pop | old_master, _ = ssconf.GetMasterAndMyself(sstore) |
478 | 8eb148ae | Iustin Pop | |
479 | 8eb148ae | Iustin Pop | return old_master
|
480 | 8eb148ae | Iustin Pop | |
481 | 8eb148ae | Iustin Pop | |
482 | d7cdb55d | Iustin Pop | def GatherMasterVotes(node_list): |
483 | d7cdb55d | Iustin Pop | """Check the agreement on who is the master.
|
484 | d7cdb55d | Iustin Pop |
|
485 | d7cdb55d | Iustin Pop | This function will return a list of (node, number of votes), ordered
|
486 | d7cdb55d | Iustin Pop | by the number of votes. Errors will be denoted by the key 'None'.
|
487 | d7cdb55d | Iustin Pop |
|
488 | d7cdb55d | Iustin Pop | Note that the sum of votes is the number of nodes this machine
|
489 | d7cdb55d | Iustin Pop | knows, whereas the number of entries in the list could be different
|
490 | d7cdb55d | Iustin Pop | (if some nodes vote for another master).
|
491 | d7cdb55d | Iustin Pop |
|
492 | d7cdb55d | Iustin Pop | We remove ourselves from the list since we know that (bugs aside)
|
493 | d7cdb55d | Iustin Pop | since we use the same source for configuration information for both
|
494 | d7cdb55d | Iustin Pop | backend and boostrap, we'll always vote for ourselves.
|
495 | d7cdb55d | Iustin Pop |
|
496 | d7cdb55d | Iustin Pop | @type node_list: list
|
497 | d7cdb55d | Iustin Pop | @param node_list: the list of nodes to query for master info; the current
|
498 | 5bbd3f7f | Michael Hanselmann | node will be removed if it is in the list
|
499 | d7cdb55d | Iustin Pop | @rtype: list
|
500 | d7cdb55d | Iustin Pop | @return: list of (node, votes)
|
501 | d7cdb55d | Iustin Pop |
|
502 | d7cdb55d | Iustin Pop | """
|
503 | d7cdb55d | Iustin Pop | myself = utils.HostInfo().name |
504 | d7cdb55d | Iustin Pop | try:
|
505 | d7cdb55d | Iustin Pop | node_list.remove(myself) |
506 | d7cdb55d | Iustin Pop | except ValueError: |
507 | d7cdb55d | Iustin Pop | pass
|
508 | d7cdb55d | Iustin Pop | if not node_list: |
509 | d7cdb55d | Iustin Pop | # no nodes left (eventually after removing myself)
|
510 | d7cdb55d | Iustin Pop | return []
|
511 | d7cdb55d | Iustin Pop | results = rpc.RpcRunner.call_master_info(node_list) |
512 | d7cdb55d | Iustin Pop | if not isinstance(results, dict): |
513 | d7cdb55d | Iustin Pop | # this should not happen (unless internal error in rpc)
|
514 | d7cdb55d | Iustin Pop | logging.critical("Can't complete rpc call, aborting master startup")
|
515 | d7cdb55d | Iustin Pop | return [(None, len(node_list))] |
516 | d7cdb55d | Iustin Pop | votes = {} |
517 | d7cdb55d | Iustin Pop | for node in results: |
518 | 781de953 | Iustin Pop | nres = results[node] |
519 | 781de953 | Iustin Pop | data = nres.data |
520 | 781de953 | Iustin Pop | if nres.failed or not isinstance(data, (tuple, list)) or len(data) < 3: |
521 | d7cdb55d | Iustin Pop | # here the rpc layer should have already logged errors
|
522 | d7cdb55d | Iustin Pop | if None not in votes: |
523 | d7cdb55d | Iustin Pop | votes[None] = 0 |
524 | d7cdb55d | Iustin Pop | votes[None] += 1 |
525 | d7cdb55d | Iustin Pop | continue
|
526 | 781de953 | Iustin Pop | master_node = data[2]
|
527 | d7cdb55d | Iustin Pop | if master_node not in votes: |
528 | d7cdb55d | Iustin Pop | votes[master_node] = 0
|
529 | d7cdb55d | Iustin Pop | votes[master_node] += 1
|
530 | d7cdb55d | Iustin Pop | |
531 | d7cdb55d | Iustin Pop | vote_list = [v for v in votes.items()] |
532 | d7cdb55d | Iustin Pop | # sort first on number of votes then on name, since we want None
|
533 | d7cdb55d | Iustin Pop | # sorted later if we have the half of the nodes not responding, and
|
534 | d7cdb55d | Iustin Pop | # half voting all for the same master
|
535 | d7cdb55d | Iustin Pop | vote_list.sort(key=lambda x: (x[1], x[0]), reverse=True) |
536 | d7cdb55d | Iustin Pop | |
537 | d7cdb55d | Iustin Pop | return vote_list |