root / lib / bootstrap.py @ ebeb600f
History | View | Annotate | Download (22.9 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 | d693c864 | Iustin Pop | import time |
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 | a33848a5 | Guido Trotter | from ganeti import serializer |
41 | a5728081 | Guido Trotter | from ganeti import hypervisor |
42 | a0c9f010 | Michael Hanselmann | |
43 | e38220e4 | Michael Hanselmann | |
44 | 531baf8e | Iustin Pop | def _InitSSHSetup(): |
45 | a0c9f010 | Michael Hanselmann | """Setup the SSH configuration for the cluster.
|
46 | a0c9f010 | Michael Hanselmann |
|
47 | a0c9f010 | Michael Hanselmann | This generates a dsa keypair for root, adds the pub key to the
|
48 | a0c9f010 | Michael Hanselmann | permitted hosts and adds the hostkey to its own known hosts.
|
49 | a0c9f010 | Michael Hanselmann |
|
50 | a0c9f010 | Michael Hanselmann | """
|
51 | a0c9f010 | Michael Hanselmann | priv_key, pub_key, auth_keys = ssh.GetUserFiles(constants.GANETI_RUNAS) |
52 | a0c9f010 | Michael Hanselmann | |
53 | a0c9f010 | Michael Hanselmann | for name in priv_key, pub_key: |
54 | a0c9f010 | Michael Hanselmann | if os.path.exists(name):
|
55 | a0c9f010 | Michael Hanselmann | utils.CreateBackup(name) |
56 | a0c9f010 | Michael Hanselmann | utils.RemoveFile(name) |
57 | a0c9f010 | Michael Hanselmann | |
58 | a0c9f010 | Michael Hanselmann | result = utils.RunCmd(["ssh-keygen", "-t", "dsa", |
59 | a0c9f010 | Michael Hanselmann | "-f", priv_key,
|
60 | a0c9f010 | Michael Hanselmann | "-q", "-N", ""]) |
61 | a0c9f010 | Michael Hanselmann | if result.failed:
|
62 | a0c9f010 | Michael Hanselmann | raise errors.OpExecError("Could not generate ssh keypair, error %s" % |
63 | a0c9f010 | Michael Hanselmann | result.output) |
64 | a0c9f010 | Michael Hanselmann | |
65 | 7a0156dc | Luca Bigliardi | utils.AddAuthorizedKey(auth_keys, utils.ReadFile(pub_key)) |
66 | a0c9f010 | Michael Hanselmann | |
67 | a0c9f010 | Michael Hanselmann | |
68 | c008906b | Michael Hanselmann | def GenerateHmacKey(file_name): |
69 | c008906b | Michael Hanselmann | """Writes a new HMAC key.
|
70 | c008906b | Michael Hanselmann |
|
71 | c008906b | Michael Hanselmann | @type file_name: str
|
72 | c008906b | Michael Hanselmann | @param file_name: Path to output file
|
73 | c008906b | Michael Hanselmann |
|
74 | c008906b | Michael Hanselmann | """
|
75 | 43575108 | Michael Hanselmann | utils.WriteFile(file_name, data="%s\n" % utils.GenerateSecret(), mode=0400, |
76 | 43575108 | Michael Hanselmann | backup=True)
|
77 | 43575108 | Michael Hanselmann | |
78 | 43575108 | Michael Hanselmann | |
79 | 6b7d5878 | Michael Hanselmann | def GenerateClusterCrypto(new_cluster_cert, new_rapi_cert, new_confd_hmac_key, |
80 | af2ae1c0 | Iustin Pop | new_cds, rapi_cert_pem=None, cds=None, |
81 | aeefe835 | Iustin Pop | nodecert_file=constants.NODED_CERT_FILE, |
82 | aeefe835 | Iustin Pop | rapicert_file=constants.RAPI_CERT_FILE, |
83 | aeefe835 | Iustin Pop | hmackey_file=constants.CONFD_HMAC_KEY): |
84 | 43575108 | Michael Hanselmann | """Updates the cluster certificates, keys and secrets.
|
85 | 43575108 | Michael Hanselmann |
|
86 | 43575108 | Michael Hanselmann | @type new_cluster_cert: bool
|
87 | 43575108 | Michael Hanselmann | @param new_cluster_cert: Whether to generate a new cluster certificate
|
88 | 43575108 | Michael Hanselmann | @type new_rapi_cert: bool
|
89 | 43575108 | Michael Hanselmann | @param new_rapi_cert: Whether to generate a new RAPI certificate
|
90 | 6b7d5878 | Michael Hanselmann | @type new_confd_hmac_key: bool
|
91 | 6b7d5878 | Michael Hanselmann | @param new_confd_hmac_key: Whether to generate a new HMAC key
|
92 | 3db3eb2a | Michael Hanselmann | @type new_cds: bool
|
93 | 3db3eb2a | Michael Hanselmann | @param new_cds: Whether to generate a new cluster domain secret
|
94 | 43575108 | Michael Hanselmann | @type rapi_cert_pem: string
|
95 | 43575108 | Michael Hanselmann | @param rapi_cert_pem: New RAPI certificate in PEM format
|
96 | 3db3eb2a | Michael Hanselmann | @type cds: string
|
97 | 3db3eb2a | Michael Hanselmann | @param cds: New cluster domain secret
|
98 | aeefe835 | Iustin Pop | @type nodecert_file: string
|
99 | aeefe835 | Iustin Pop | @param nodecert_file: optional override of the node cert file path
|
100 | aeefe835 | Iustin Pop | @type rapicert_file: string
|
101 | aeefe835 | Iustin Pop | @param rapicert_file: optional override of the rapi cert file path
|
102 | aeefe835 | Iustin Pop | @type hmackey_file: string
|
103 | aeefe835 | Iustin Pop | @param hmackey_file: optional override of the hmac key file path
|
104 | 43575108 | Michael Hanselmann |
|
105 | 43575108 | Michael Hanselmann | """
|
106 | 168c1de2 | Michael Hanselmann | # noded SSL certificate
|
107 | aeefe835 | Iustin Pop | cluster_cert_exists = os.path.exists(nodecert_file) |
108 | 43575108 | Michael Hanselmann | if new_cluster_cert or not cluster_cert_exists: |
109 | 43575108 | Michael Hanselmann | if cluster_cert_exists:
|
110 | aeefe835 | Iustin Pop | utils.CreateBackup(nodecert_file) |
111 | 43575108 | Michael Hanselmann | |
112 | aeefe835 | Iustin Pop | logging.debug("Generating new cluster certificate at %s", nodecert_file)
|
113 | af2ae1c0 | Iustin Pop | utils.GenerateSelfSignedSslCert(nodecert_file) |
114 | 43575108 | Michael Hanselmann | |
115 | 6b7d5878 | Michael Hanselmann | # confd HMAC key
|
116 | aeefe835 | Iustin Pop | if new_confd_hmac_key or not os.path.exists(hmackey_file): |
117 | aeefe835 | Iustin Pop | logging.debug("Writing new confd HMAC key to %s", hmackey_file)
|
118 | aeefe835 | Iustin Pop | GenerateHmacKey(hmackey_file) |
119 | 43575108 | Michael Hanselmann | |
120 | 43575108 | Michael Hanselmann | # RAPI
|
121 | aeefe835 | Iustin Pop | rapi_cert_exists = os.path.exists(rapicert_file) |
122 | 43575108 | Michael Hanselmann | |
123 | 43575108 | Michael Hanselmann | if rapi_cert_pem:
|
124 | 43575108 | Michael Hanselmann | # Assume rapi_pem contains a valid PEM-formatted certificate and key
|
125 | aeefe835 | Iustin Pop | logging.debug("Writing RAPI certificate at %s", rapicert_file)
|
126 | aeefe835 | Iustin Pop | utils.WriteFile(rapicert_file, data=rapi_cert_pem, backup=True)
|
127 | 43575108 | Michael Hanselmann | |
128 | 43575108 | Michael Hanselmann | elif new_rapi_cert or not rapi_cert_exists: |
129 | 43575108 | Michael Hanselmann | if rapi_cert_exists:
|
130 | aeefe835 | Iustin Pop | utils.CreateBackup(rapicert_file) |
131 | 43575108 | Michael Hanselmann | |
132 | aeefe835 | Iustin Pop | logging.debug("Generating new RAPI certificate at %s", rapicert_file)
|
133 | af2ae1c0 | Iustin Pop | utils.GenerateSelfSignedSslCert(rapicert_file) |
134 | c008906b | Michael Hanselmann | |
135 | 3db3eb2a | Michael Hanselmann | # Cluster domain secret
|
136 | 3db3eb2a | Michael Hanselmann | if cds:
|
137 | 3db3eb2a | Michael Hanselmann | logging.debug("Writing cluster domain secret to %s",
|
138 | 3db3eb2a | Michael Hanselmann | constants.CLUSTER_DOMAIN_SECRET_FILE) |
139 | 3db3eb2a | Michael Hanselmann | utils.WriteFile(constants.CLUSTER_DOMAIN_SECRET_FILE, |
140 | 3db3eb2a | Michael Hanselmann | data=cds, backup=True)
|
141 | 3db3eb2a | Michael Hanselmann | |
142 | 3db3eb2a | Michael Hanselmann | elif new_cds or not os.path.exists(constants.CLUSTER_DOMAIN_SECRET_FILE): |
143 | 3db3eb2a | Michael Hanselmann | logging.debug("Generating new cluster domain secret at %s",
|
144 | 3db3eb2a | Michael Hanselmann | constants.CLUSTER_DOMAIN_SECRET_FILE) |
145 | 3db3eb2a | Michael Hanselmann | GenerateHmacKey(constants.CLUSTER_DOMAIN_SECRET_FILE) |
146 | 3db3eb2a | Michael Hanselmann | |
147 | c008906b | Michael Hanselmann | |
148 | 8f215968 | Michael Hanselmann | def _InitGanetiServerSetup(master_name): |
149 | 40a97d80 | Michael Hanselmann | """Setup the necessary configuration for the initial node daemon.
|
150 | 40a97d80 | Michael Hanselmann |
|
151 | 40a97d80 | Michael Hanselmann | This creates the nodepass file containing the shared password for
|
152 | 40a97d80 | Michael Hanselmann | the cluster and also generates the SSL certificate.
|
153 | 40a97d80 | Michael Hanselmann |
|
154 | 40a97d80 | Michael Hanselmann | """
|
155 | 43575108 | Michael Hanselmann | # Generate cluster secrets
|
156 | 3db3eb2a | Michael Hanselmann | GenerateClusterCrypto(True, False, False, False) |
157 | 4a34c5cf | Guido Trotter | |
158 | f154a7a3 | Michael Hanselmann | result = utils.RunCmd([constants.DAEMON_UTIL, "start", constants.NODED])
|
159 | a0c9f010 | Michael Hanselmann | if result.failed:
|
160 | a0c9f010 | Michael Hanselmann | raise errors.OpExecError("Could not start the node daemon, command %s" |
161 | a0c9f010 | Michael Hanselmann | " had exitcode %s and error %s" %
|
162 | a0c9f010 | Michael Hanselmann | (result.cmd, result.exit_code, result.output)) |
163 | a0c9f010 | Michael Hanselmann | |
164 | 5627f375 | Michael Hanselmann | _WaitForNodeDaemon(master_name) |
165 | 5627f375 | Michael Hanselmann | |
166 | 5627f375 | Michael Hanselmann | |
167 | 5627f375 | Michael Hanselmann | def _WaitForNodeDaemon(node_name): |
168 | 5627f375 | Michael Hanselmann | """Wait for node daemon to become responsive.
|
169 | 5627f375 | Michael Hanselmann |
|
170 | 5627f375 | Michael Hanselmann | """
|
171 | d3833ebd | Michael Hanselmann | def _CheckNodeDaemon(): |
172 | 5627f375 | Michael Hanselmann | result = rpc.RpcRunner.call_version([node_name])[node_name] |
173 | d3833ebd | Michael Hanselmann | if result.fail_msg:
|
174 | d3833ebd | Michael Hanselmann | raise utils.RetryAgain()
|
175 | 8f215968 | Michael Hanselmann | |
176 | d3833ebd | Michael Hanselmann | try:
|
177 | d3833ebd | Michael Hanselmann | utils.Retry(_CheckNodeDaemon, 1.0, 10.0) |
178 | d3833ebd | Michael Hanselmann | except utils.RetryTimeout:
|
179 | 5627f375 | Michael Hanselmann | raise errors.OpExecError("Node daemon on %s didn't answer queries within" |
180 | 5627f375 | Michael Hanselmann | " 10 seconds" % node_name)
|
181 | 5627f375 | Michael Hanselmann | |
182 | a0c9f010 | Michael Hanselmann | |
183 | 0e3baaf3 | Iustin Pop | def _InitFileStorage(file_storage_dir): |
184 | 0e3baaf3 | Iustin Pop | """Initialize if needed the file storage.
|
185 | 0e3baaf3 | Iustin Pop |
|
186 | 0e3baaf3 | Iustin Pop | @param file_storage_dir: the user-supplied value
|
187 | 0e3baaf3 | Iustin Pop | @return: either empty string (if file storage was disabled at build
|
188 | 0e3baaf3 | Iustin Pop | time) or the normalized path to the storage directory
|
189 | 0e3baaf3 | Iustin Pop |
|
190 | 0e3baaf3 | Iustin Pop | """
|
191 | 0e3baaf3 | Iustin Pop | if not constants.ENABLE_FILE_STORAGE: |
192 | 0e3baaf3 | Iustin Pop | return "" |
193 | 0e3baaf3 | Iustin Pop | |
194 | 0e3baaf3 | Iustin Pop | file_storage_dir = os.path.normpath(file_storage_dir) |
195 | 0e3baaf3 | Iustin Pop | |
196 | 0e3baaf3 | Iustin Pop | if not os.path.isabs(file_storage_dir): |
197 | 0e3baaf3 | Iustin Pop | raise errors.OpPrereqError("The file storage directory you passed is" |
198 | 0e3baaf3 | Iustin Pop | " not an absolute path.", errors.ECODE_INVAL)
|
199 | 0e3baaf3 | Iustin Pop | |
200 | 0e3baaf3 | Iustin Pop | if not os.path.exists(file_storage_dir): |
201 | 0e3baaf3 | Iustin Pop | try:
|
202 | 0e3baaf3 | Iustin Pop | os.makedirs(file_storage_dir, 0750)
|
203 | 0e3baaf3 | Iustin Pop | except OSError, err: |
204 | 0e3baaf3 | Iustin Pop | raise errors.OpPrereqError("Cannot create file storage directory" |
205 | 0e3baaf3 | Iustin Pop | " '%s': %s" % (file_storage_dir, err),
|
206 | 0e3baaf3 | Iustin Pop | errors.ECODE_ENVIRON) |
207 | 0e3baaf3 | Iustin Pop | |
208 | 0e3baaf3 | Iustin Pop | if not os.path.isdir(file_storage_dir): |
209 | 0e3baaf3 | Iustin Pop | raise errors.OpPrereqError("The file storage directory '%s' is not" |
210 | 0e3baaf3 | Iustin Pop | " a directory." % file_storage_dir,
|
211 | 0e3baaf3 | Iustin Pop | errors.ECODE_ENVIRON) |
212 | 0e3baaf3 | Iustin Pop | return file_storage_dir
|
213 | 0e3baaf3 | Iustin Pop | |
214 | 0e3baaf3 | Iustin Pop | |
215 | ec0652ad | Guido Trotter | def InitCluster(cluster_name, mac_prefix, |
216 | ce735215 | Guido Trotter | master_netdev, file_storage_dir, candidate_pool_size, |
217 | b6a30b0d | Guido Trotter | secondary_ip=None, vg_name=None, beparams=None, |
218 | b6a30b0d | Guido Trotter | nicparams=None, hvparams=None, enabled_hypervisors=None, |
219 | 3953242f | Iustin Pop | modify_etc_hosts=True, modify_ssh_setup=True, |
220 | 39b0f0c2 | Balazs Lecz | maintain_node_health=False,
|
221 | 39b0f0c2 | Balazs Lecz | uid_pool=None):
|
222 | a0c9f010 | Michael Hanselmann | """Initialise the cluster.
|
223 | a0c9f010 | Michael Hanselmann |
|
224 | ce735215 | Guido Trotter | @type candidate_pool_size: int
|
225 | ce735215 | Guido Trotter | @param candidate_pool_size: master candidate pool size
|
226 | ce735215 | Guido Trotter |
|
227 | a0c9f010 | Michael Hanselmann | """
|
228 | ce735215 | Guido Trotter | # TODO: complete the docstring
|
229 | a0c9f010 | Michael Hanselmann | if config.ConfigWriter.IsCluster():
|
230 | debac808 | Iustin Pop | raise errors.OpPrereqError("Cluster is already initialised", |
231 | debac808 | Iustin Pop | errors.ECODE_STATE) |
232 | a0c9f010 | Michael Hanselmann | |
233 | b119bccb | Guido Trotter | if not enabled_hypervisors: |
234 | b119bccb | Guido Trotter | raise errors.OpPrereqError("Enabled hypervisors list must contain at" |
235 | debac808 | Iustin Pop | " least one member", errors.ECODE_INVAL)
|
236 | b119bccb | Guido Trotter | invalid_hvs = set(enabled_hypervisors) - constants.HYPER_TYPES
|
237 | b119bccb | Guido Trotter | if invalid_hvs:
|
238 | b119bccb | Guido Trotter | raise errors.OpPrereqError("Enabled hypervisors contains invalid" |
239 | debac808 | Iustin Pop | " entries: %s" % invalid_hvs,
|
240 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
241 | b119bccb | Guido Trotter | |
242 | 104f4ca1 | Iustin Pop | hostname = utils.GetHostInfo() |
243 | a0c9f010 | Michael Hanselmann | |
244 | a0c9f010 | Michael Hanselmann | if hostname.ip.startswith("127."): |
245 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("This host's IP resolves to the private" |
246 | a0c9f010 | Michael Hanselmann | " range (%s). Please fix DNS or %s." %
|
247 | debac808 | Iustin Pop | (hostname.ip, constants.ETC_HOSTS), |
248 | debac808 | Iustin Pop | errors.ECODE_ENVIRON) |
249 | a0c9f010 | Michael Hanselmann | |
250 | caad16e2 | Iustin Pop | if not utils.OwnIpAddress(hostname.ip): |
251 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Inconsistency: this host's name resolves" |
252 | a0c9f010 | Michael Hanselmann | " to %s,\nbut this ip address does not"
|
253 | debac808 | Iustin Pop | " belong to this host. Aborting." %
|
254 | debac808 | Iustin Pop | hostname.ip, errors.ECODE_ENVIRON) |
255 | a0c9f010 | Michael Hanselmann | |
256 | 44caf5a8 | Iustin Pop | clustername = utils.GetHostInfo(utils.HostInfo.NormalizeName(cluster_name)) |
257 | a0c9f010 | Michael Hanselmann | |
258 | a0c9f010 | Michael Hanselmann | if utils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT,
|
259 | a0c9f010 | Michael Hanselmann | timeout=5):
|
260 | debac808 | Iustin Pop | raise errors.OpPrereqError("Cluster IP already active. Aborting.", |
261 | debac808 | Iustin Pop | errors.ECODE_NOTUNIQUE) |
262 | a0c9f010 | Michael Hanselmann | |
263 | a0c9f010 | Michael Hanselmann | if secondary_ip:
|
264 | a0c9f010 | Michael Hanselmann | if not utils.IsValidIP(secondary_ip): |
265 | debac808 | Iustin Pop | raise errors.OpPrereqError("Invalid secondary ip given", |
266 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
267 | a0c9f010 | Michael Hanselmann | if (secondary_ip != hostname.ip and |
268 | caad16e2 | Iustin Pop | not utils.OwnIpAddress(secondary_ip)):
|
269 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("You gave %s as secondary IP," |
270 | a0c9f010 | Michael Hanselmann | " but it does not belong to this host." %
|
271 | debac808 | Iustin Pop | secondary_ip, errors.ECODE_ENVIRON) |
272 | b9eeeb02 | Michael Hanselmann | else:
|
273 | b9eeeb02 | Michael Hanselmann | secondary_ip = hostname.ip |
274 | a0c9f010 | Michael Hanselmann | |
275 | a0c9f010 | Michael Hanselmann | if vg_name is not None: |
276 | a0c9f010 | Michael Hanselmann | # Check if volume group is valid
|
277 | a0c9f010 | Michael Hanselmann | vgstatus = utils.CheckVolumeGroupSize(utils.ListVolumeGroups(), vg_name, |
278 | a0c9f010 | Michael Hanselmann | constants.MIN_VG_SIZE) |
279 | a0c9f010 | Michael Hanselmann | if vgstatus:
|
280 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if" |
281 | debac808 | Iustin Pop | " you are not using lvm" % vgstatus,
|
282 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
283 | a0c9f010 | Michael Hanselmann | |
284 | 0e3baaf3 | Iustin Pop | file_storage_dir = _InitFileStorage(file_storage_dir) |
285 | a0c9f010 | Michael Hanselmann | |
286 | a0c9f010 | Michael Hanselmann | if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", mac_prefix): |
287 | debac808 | Iustin Pop | raise errors.OpPrereqError("Invalid mac prefix given '%s'" % mac_prefix, |
288 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
289 | a0c9f010 | Michael Hanselmann | |
290 | a0c9f010 | Michael Hanselmann | result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev]) |
291 | a0c9f010 | Michael Hanselmann | if result.failed:
|
292 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" % |
293 | a0c9f010 | Michael Hanselmann | (master_netdev, |
294 | debac808 | Iustin Pop | result.output.strip()), errors.ECODE_INVAL) |
295 | a0c9f010 | Michael Hanselmann | |
296 | 9dae41ad | Guido Trotter | dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE)] |
297 | 9dae41ad | Guido Trotter | utils.EnsureDirs(dirs) |
298 | 9dae41ad | Guido Trotter | |
299 | a5728081 | Guido Trotter | utils.ForceDictType(beparams, constants.BES_PARAMETER_TYPES) |
300 | b6a30b0d | Guido Trotter | utils.ForceDictType(nicparams, constants.NICS_PARAMETER_TYPES) |
301 | b6a30b0d | Guido Trotter | objects.NIC.CheckParameterSyntax(nicparams) |
302 | b6a30b0d | Guido Trotter | |
303 | a5728081 | Guido Trotter | # hvparams is a mapping of hypervisor->hvparams dict
|
304 | a5728081 | Guido Trotter | for hv_name, hv_params in hvparams.iteritems(): |
305 | a5728081 | Guido Trotter | utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES) |
306 | a5728081 | Guido Trotter | hv_class = hypervisor.GetHypervisor(hv_name) |
307 | a5728081 | Guido Trotter | hv_class.CheckParameterSyntax(hv_params) |
308 | d4b72030 | Guido Trotter | |
309 | a0c9f010 | Michael Hanselmann | # set up the inter-node password and certificate
|
310 | 8f215968 | Michael Hanselmann | _InitGanetiServerSetup(hostname.name) |
311 | a0c9f010 | Michael Hanselmann | |
312 | a0c9f010 | Michael Hanselmann | # set up ssh config and /etc/hosts
|
313 | 13998ef2 | Michael Hanselmann | sshline = utils.ReadFile(constants.SSH_HOST_RSA_PUB) |
314 | a0c9f010 | Michael Hanselmann | sshkey = sshline.split(" ")[1] |
315 | a0c9f010 | Michael Hanselmann | |
316 | b86a6bcd | Guido Trotter | if modify_etc_hosts:
|
317 | b86a6bcd | Guido Trotter | utils.AddHostToEtcHosts(hostname.name) |
318 | b86a6bcd | Guido Trotter | |
319 | b989b9d9 | Ken Wehr | if modify_ssh_setup:
|
320 | b989b9d9 | Ken Wehr | _InitSSHSetup() |
321 | a0c9f010 | Michael Hanselmann | |
322 | 430b923c | Iustin Pop | now = time.time() |
323 | 430b923c | Iustin Pop | |
324 | a0c9f010 | Michael Hanselmann | # init of cluster config file
|
325 | b9eeeb02 | Michael Hanselmann | cluster_config = objects.Cluster( |
326 | b9eeeb02 | Michael Hanselmann | serial_no=1,
|
327 | b9eeeb02 | Michael Hanselmann | rsahostkeypub=sshkey, |
328 | b9eeeb02 | Michael Hanselmann | highest_used_port=(constants.FIRST_DRBD_PORT - 1),
|
329 | b9eeeb02 | Michael Hanselmann | mac_prefix=mac_prefix, |
330 | b9eeeb02 | Michael Hanselmann | volume_group_name=vg_name, |
331 | b9eeeb02 | Michael Hanselmann | tcpudp_port_pool=set(),
|
332 | f6bd6e98 | Michael Hanselmann | master_node=hostname.name, |
333 | f6bd6e98 | Michael Hanselmann | master_ip=clustername.ip, |
334 | f6bd6e98 | Michael Hanselmann | master_netdev=master_netdev, |
335 | f6bd6e98 | Michael Hanselmann | cluster_name=clustername.name, |
336 | f6bd6e98 | Michael Hanselmann | file_storage_dir=file_storage_dir, |
337 | ea3a925f | Alexander Schreiber | enabled_hypervisors=enabled_hypervisors, |
338 | 4ef7f423 | Guido Trotter | beparams={constants.PP_DEFAULT: beparams}, |
339 | b6a30b0d | Guido Trotter | nicparams={constants.PP_DEFAULT: nicparams}, |
340 | ea3a925f | Alexander Schreiber | hvparams=hvparams, |
341 | ce735215 | Guido Trotter | candidate_pool_size=candidate_pool_size, |
342 | 022c3a0b | Guido Trotter | modify_etc_hosts=modify_etc_hosts, |
343 | b989b9d9 | Ken Wehr | modify_ssh_setup=modify_ssh_setup, |
344 | 39b0f0c2 | Balazs Lecz | uid_pool=uid_pool, |
345 | 430b923c | Iustin Pop | ctime=now, |
346 | 430b923c | Iustin Pop | mtime=now, |
347 | 430b923c | Iustin Pop | uuid=utils.NewUUID(), |
348 | 3953242f | Iustin Pop | maintain_node_health=maintain_node_health, |
349 | b9eeeb02 | Michael Hanselmann | ) |
350 | b9eeeb02 | Michael Hanselmann | master_node_config = objects.Node(name=hostname.name, |
351 | b9eeeb02 | Michael Hanselmann | primary_ip=hostname.ip, |
352 | b9222f32 | Guido Trotter | secondary_ip=secondary_ip, |
353 | c044f32c | Guido Trotter | serial_no=1,
|
354 | c044f32c | Guido Trotter | master_candidate=True,
|
355 | af64c0ea | Iustin Pop | offline=False, drained=False, |
356 | c044f32c | Guido Trotter | ) |
357 | 9e1333b9 | Guido Trotter | InitConfig(constants.CONFIG_VERSION, cluster_config, master_node_config) |
358 | 05cc153f | Guido Trotter | cfg = config.ConfigWriter() |
359 | 9e1333b9 | Guido Trotter | ssh.WriteKnownHostsFile(cfg, constants.SSH_KNOWN_HOSTS_FILE) |
360 | a4eae71f | Michael Hanselmann | cfg.Update(cfg.GetClusterInfo(), logging.error) |
361 | 827f753e | Guido Trotter | |
362 | b3f1cf6f | Iustin Pop | # start the master ip
|
363 | b3f1cf6f | Iustin Pop | # TODO: Review rpc call from bootstrap
|
364 | b726aff0 | Iustin Pop | # TODO: Warn on failed start master
|
365 | 3583908a | Guido Trotter | rpc.RpcRunner.call_node_start_master(hostname.name, True, False) |
366 | b3f1cf6f | Iustin Pop | |
367 | b1b6ea87 | Iustin Pop | |
368 | 02f99608 | Oleksiy Mishchenko | def InitConfig(version, cluster_config, master_node_config, |
369 | 02f99608 | Oleksiy Mishchenko | cfg_file=constants.CLUSTER_CONF_FILE): |
370 | 7b3a8fb5 | Iustin Pop | """Create the initial cluster configuration.
|
371 | 7b3a8fb5 | Iustin Pop |
|
372 | 7b3a8fb5 | Iustin Pop | It will contain the current node, which will also be the master
|
373 | 7b3a8fb5 | Iustin Pop | node, and no instances.
|
374 | 7b3a8fb5 | Iustin Pop |
|
375 | 7b3a8fb5 | Iustin Pop | @type version: int
|
376 | c41eea6e | Iustin Pop | @param version: configuration version
|
377 | c41eea6e | Iustin Pop | @type cluster_config: L{objects.Cluster}
|
378 | c41eea6e | Iustin Pop | @param cluster_config: cluster configuration
|
379 | c41eea6e | Iustin Pop | @type master_node_config: L{objects.Node}
|
380 | c41eea6e | Iustin Pop | @param master_node_config: master node configuration
|
381 | c41eea6e | Iustin Pop | @type cfg_file: string
|
382 | c41eea6e | Iustin Pop | @param cfg_file: configuration file path
|
383 | c41eea6e | Iustin Pop |
|
384 | 7b3a8fb5 | Iustin Pop | """
|
385 | 7b3a8fb5 | Iustin Pop | nodes = { |
386 | 7b3a8fb5 | Iustin Pop | master_node_config.name: master_node_config, |
387 | 7b3a8fb5 | Iustin Pop | } |
388 | 7b3a8fb5 | Iustin Pop | |
389 | d693c864 | Iustin Pop | now = time.time() |
390 | 7b3a8fb5 | Iustin Pop | config_data = objects.ConfigData(version=version, |
391 | 7b3a8fb5 | Iustin Pop | cluster=cluster_config, |
392 | 7b3a8fb5 | Iustin Pop | nodes=nodes, |
393 | 7b3a8fb5 | Iustin Pop | instances={}, |
394 | d693c864 | Iustin Pop | serial_no=1,
|
395 | d693c864 | Iustin Pop | ctime=now, mtime=now) |
396 | a33848a5 | Guido Trotter | utils.WriteFile(cfg_file, |
397 | a33848a5 | Guido Trotter | data=serializer.Dump(config_data.ToDict()), |
398 | a33848a5 | Guido Trotter | mode=0600)
|
399 | 02f99608 | Oleksiy Mishchenko | |
400 | 02f99608 | Oleksiy Mishchenko | |
401 | 140aa4a8 | Iustin Pop | def FinalizeClusterDestroy(master): |
402 | 140aa4a8 | Iustin Pop | """Execute the last steps of cluster destroy
|
403 | 140aa4a8 | Iustin Pop |
|
404 | 140aa4a8 | Iustin Pop | This function shuts down all the daemons, completing the destroy
|
405 | 140aa4a8 | Iustin Pop | begun in cmdlib.LUDestroyOpcode.
|
406 | 140aa4a8 | Iustin Pop |
|
407 | 140aa4a8 | Iustin Pop | """
|
408 | b989b9d9 | Ken Wehr | cfg = config.ConfigWriter() |
409 | b989b9d9 | Ken Wehr | modify_ssh_setup = cfg.GetClusterInfo().modify_ssh_setup |
410 | 781de953 | Iustin Pop | result = rpc.RpcRunner.call_node_stop_master(master, True)
|
411 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
412 | 6c00d19a | Iustin Pop | if msg:
|
413 | 099c52ad | Iustin Pop | logging.warning("Could not disable the master role: %s", msg)
|
414 | b989b9d9 | Ken Wehr | result = rpc.RpcRunner.call_node_leave_cluster(master, modify_ssh_setup) |
415 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
416 | 0623d351 | Iustin Pop | if msg:
|
417 | 0623d351 | Iustin Pop | logging.warning("Could not shutdown the node daemon and cleanup"
|
418 | 0623d351 | Iustin Pop | " the node: %s", msg)
|
419 | 140aa4a8 | Iustin Pop | |
420 | 140aa4a8 | Iustin Pop | |
421 | 87622829 | Iustin Pop | def SetupNodeDaemon(cluster_name, node, ssh_key_check): |
422 | 827f753e | Guido Trotter | """Add a node to the cluster.
|
423 | 827f753e | Guido Trotter |
|
424 | b1b6ea87 | Iustin Pop | This function must be called before the actual opcode, and will ssh
|
425 | b1b6ea87 | Iustin Pop | to the remote node, copy the needed files, and start ganeti-noded,
|
426 | b1b6ea87 | Iustin Pop | allowing the master to do the rest via normal rpc calls.
|
427 | 827f753e | Guido Trotter |
|
428 | 87622829 | Iustin Pop | @param cluster_name: the cluster name
|
429 | 87622829 | Iustin Pop | @param node: the name of the new node
|
430 | 87622829 | Iustin Pop | @param ssh_key_check: whether to do a strict key check
|
431 | 827f753e | Guido Trotter |
|
432 | 827f753e | Guido Trotter | """
|
433 | 87622829 | Iustin Pop | sshrunner = ssh.SshRunner(cluster_name) |
434 | 5557b04c | Michael Hanselmann | |
435 | 168c1de2 | Michael Hanselmann | noded_cert = utils.ReadFile(constants.NODED_CERT_FILE) |
436 | 2438c157 | Michael Hanselmann | rapi_cert = utils.ReadFile(constants.RAPI_CERT_FILE) |
437 | 6b7d5878 | Michael Hanselmann | confd_hmac_key = utils.ReadFile(constants.CONFD_HMAC_KEY) |
438 | 5557b04c | Michael Hanselmann | |
439 | 827f753e | Guido Trotter | # in the base64 pem encoding, neither '!' nor '.' are valid chars,
|
440 | 827f753e | Guido Trotter | # so we use this to detect an invalid certificate; as long as the
|
441 | 827f753e | Guido Trotter | # cert doesn't contain this, the here-document will be correctly
|
442 | 77b076ca | Guido Trotter | # parsed by the shell sequence below. HMAC keys are hexadecimal strings,
|
443 | 77b076ca | Guido Trotter | # so the same restrictions apply.
|
444 | 6b7d5878 | Michael Hanselmann | for content in (noded_cert, rapi_cert, confd_hmac_key): |
445 | 77b076ca | Guido Trotter | if re.search('^!EOF\.', content, re.MULTILINE): |
446 | 77b076ca | Guido Trotter | raise errors.OpExecError("invalid SSL certificate or HMAC key") |
447 | 5557b04c | Michael Hanselmann | |
448 | 5557b04c | Michael Hanselmann | if not noded_cert.endswith("\n"): |
449 | 5557b04c | Michael Hanselmann | noded_cert += "\n"
|
450 | 2438c157 | Michael Hanselmann | if not rapi_cert.endswith("\n"): |
451 | 2438c157 | Michael Hanselmann | rapi_cert += "\n"
|
452 | 6b7d5878 | Michael Hanselmann | if not confd_hmac_key.endswith("\n"): |
453 | 6b7d5878 | Michael Hanselmann | confd_hmac_key += "\n"
|
454 | 827f753e | Guido Trotter | |
455 | 827f753e | Guido Trotter | # set up inter-node password and certificate and restarts the node daemon
|
456 | 827f753e | Guido Trotter | # and then connect with ssh to set password and start ganeti-noded
|
457 | 827f753e | Guido Trotter | # note that all the below variables are sanitized at this point,
|
458 | 827f753e | Guido Trotter | # either by being constants or by the checks above
|
459 | 3db3eb2a | Michael Hanselmann | # TODO: Could this command exceed a shell's maximum command length?
|
460 | 827f753e | Guido Trotter | mycommand = ("umask 077 && "
|
461 | 827f753e | Guido Trotter | "cat > '%s' << '!EOF.' && \n"
|
462 | 2438c157 | Michael Hanselmann | "%s!EOF.\n"
|
463 | 2438c157 | Michael Hanselmann | "cat > '%s' << '!EOF.' && \n"
|
464 | 2438c157 | Michael Hanselmann | "%s!EOF.\n"
|
465 | 77b076ca | Guido Trotter | "cat > '%s' << '!EOF.' && \n"
|
466 | 77b076ca | Guido Trotter | "%s!EOF.\n"
|
467 | 77b076ca | Guido Trotter | "chmod 0400 %s %s %s && "
|
468 | f154a7a3 | Michael Hanselmann | "%s start %s" %
|
469 | 168c1de2 | Michael Hanselmann | (constants.NODED_CERT_FILE, noded_cert, |
470 | 2438c157 | Michael Hanselmann | constants.RAPI_CERT_FILE, rapi_cert, |
471 | 6b7d5878 | Michael Hanselmann | constants.CONFD_HMAC_KEY, confd_hmac_key, |
472 | 168c1de2 | Michael Hanselmann | constants.NODED_CERT_FILE, constants.RAPI_CERT_FILE, |
473 | 6b7d5878 | Michael Hanselmann | constants.CONFD_HMAC_KEY, |
474 | f154a7a3 | Michael Hanselmann | constants.DAEMON_UTIL, constants.NODED)) |
475 | 827f753e | Guido Trotter | |
476 | c4b6c29c | Michael Hanselmann | result = sshrunner.Run(node, 'root', mycommand, batch=False, |
477 | c4b6c29c | Michael Hanselmann | ask_key=ssh_key_check, |
478 | c4b6c29c | Michael Hanselmann | use_cluster_key=False,
|
479 | c4b6c29c | Michael Hanselmann | strict_host_check=ssh_key_check) |
480 | 827f753e | Guido Trotter | if result.failed:
|
481 | 827f753e | Guido Trotter | raise errors.OpExecError("Remote command on node %s, error: %s," |
482 | 827f753e | Guido Trotter | " output: %s" %
|
483 | 827f753e | Guido Trotter | (node, result.fail_reason, result.output)) |
484 | 827f753e | Guido Trotter | |
485 | 5627f375 | Michael Hanselmann | _WaitForNodeDaemon(node) |
486 | 5627f375 | Michael Hanselmann | |
487 | b1b6ea87 | Iustin Pop | |
488 | 8e2524c3 | Guido Trotter | def MasterFailover(no_voting=False): |
489 | b1b6ea87 | Iustin Pop | """Failover the master node.
|
490 | b1b6ea87 | Iustin Pop |
|
491 | b1b6ea87 | Iustin Pop | This checks that we are not already the master, and will cause the
|
492 | b1b6ea87 | Iustin Pop | current master to cease being master, and the non-master to become
|
493 | b1b6ea87 | Iustin Pop | new master.
|
494 | b1b6ea87 | Iustin Pop |
|
495 | 8e2524c3 | Guido Trotter | @type no_voting: boolean
|
496 | 8e2524c3 | Guido Trotter | @param no_voting: force the operation without remote nodes agreement
|
497 | 8e2524c3 | Guido Trotter | (dangerous)
|
498 | 8e2524c3 | Guido Trotter |
|
499 | b1b6ea87 | Iustin Pop | """
|
500 | 8135a2db | Iustin Pop | sstore = ssconf.SimpleStore() |
501 | b1b6ea87 | Iustin Pop | |
502 | 8135a2db | Iustin Pop | old_master, new_master = ssconf.GetMasterAndMyself(sstore) |
503 | 8135a2db | Iustin Pop | node_list = sstore.GetNodeList() |
504 | 8135a2db | Iustin Pop | mc_list = sstore.GetMasterCandidates() |
505 | b1b6ea87 | Iustin Pop | |
506 | b1b6ea87 | Iustin Pop | if old_master == new_master:
|
507 | b1b6ea87 | Iustin Pop | raise errors.OpPrereqError("This commands must be run on the node" |
508 | b1b6ea87 | Iustin Pop | " where you want the new master to be."
|
509 | b1b6ea87 | Iustin Pop | " %s is already the master" %
|
510 | debac808 | Iustin Pop | old_master, errors.ECODE_INVAL) |
511 | d5927e48 | Iustin Pop | |
512 | 8135a2db | Iustin Pop | if new_master not in mc_list: |
513 | 8135a2db | Iustin Pop | mc_no_master = [name for name in mc_list if name != old_master] |
514 | 8135a2db | Iustin Pop | raise errors.OpPrereqError("This node is not among the nodes marked" |
515 | 8135a2db | Iustin Pop | " as master candidates. Only these nodes"
|
516 | 8135a2db | Iustin Pop | " can become masters. Current list of"
|
517 | 8135a2db | Iustin Pop | " master candidates is:\n"
|
518 | debac808 | Iustin Pop | "%s" % ('\n'.join(mc_no_master)), |
519 | debac808 | Iustin Pop | errors.ECODE_STATE) |
520 | 8135a2db | Iustin Pop | |
521 | 8e2524c3 | Guido Trotter | if not no_voting: |
522 | 8e2524c3 | Guido Trotter | vote_list = GatherMasterVotes(node_list) |
523 | 8e2524c3 | Guido Trotter | |
524 | 8e2524c3 | Guido Trotter | if vote_list:
|
525 | 8e2524c3 | Guido Trotter | voted_master = vote_list[0][0] |
526 | 8e2524c3 | Guido Trotter | if voted_master is None: |
527 | 8e2524c3 | Guido Trotter | raise errors.OpPrereqError("Cluster is inconsistent, most nodes did" |
528 | debac808 | Iustin Pop | " not respond.", errors.ECODE_ENVIRON)
|
529 | 8e2524c3 | Guido Trotter | elif voted_master != old_master:
|
530 | 8e2524c3 | Guido Trotter | raise errors.OpPrereqError("I have a wrong configuration, I believe" |
531 | 8e2524c3 | Guido Trotter | " the master is %s but the other nodes"
|
532 | 8e2524c3 | Guido Trotter | " voted %s. Please resync the configuration"
|
533 | 8e2524c3 | Guido Trotter | " of this node." %
|
534 | debac808 | Iustin Pop | (old_master, voted_master), |
535 | debac808 | Iustin Pop | errors.ECODE_STATE) |
536 | b1b6ea87 | Iustin Pop | # end checks
|
537 | b1b6ea87 | Iustin Pop | |
538 | b1b6ea87 | Iustin Pop | rcode = 0
|
539 | b1b6ea87 | Iustin Pop | |
540 | d5927e48 | Iustin Pop | logging.info("Setting master to %s, old master: %s", new_master, old_master)
|
541 | b1b6ea87 | Iustin Pop | |
542 | 781de953 | Iustin Pop | result = rpc.RpcRunner.call_node_stop_master(old_master, True)
|
543 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
544 | 6c00d19a | Iustin Pop | if msg:
|
545 | d5927e48 | Iustin Pop | logging.error("Could not disable the master role on the old master"
|
546 | 6c00d19a | Iustin Pop | " %s, please disable manually: %s", old_master, msg)
|
547 | b1b6ea87 | Iustin Pop | |
548 | d23ef431 | Michael Hanselmann | # Here we have a phase where no master should be running
|
549 | b1b6ea87 | Iustin Pop | |
550 | bbe19c17 | Iustin Pop | # instantiate a real config writer, as we now know we have the
|
551 | bbe19c17 | Iustin Pop | # configuration data
|
552 | bbe19c17 | Iustin Pop | cfg = config.ConfigWriter() |
553 | b1b6ea87 | Iustin Pop | |
554 | bbe19c17 | Iustin Pop | cluster_info = cfg.GetClusterInfo() |
555 | bbe19c17 | Iustin Pop | cluster_info.master_node = new_master |
556 | bbe19c17 | Iustin Pop | # this will also regenerate the ssconf files, since we updated the
|
557 | bbe19c17 | Iustin Pop | # cluster info
|
558 | a4eae71f | Michael Hanselmann | cfg.Update(cluster_info, logging.error) |
559 | d5927e48 | Iustin Pop | |
560 | 3583908a | Guido Trotter | result = rpc.RpcRunner.call_node_start_master(new_master, True, no_voting)
|
561 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
562 | b726aff0 | Iustin Pop | if msg:
|
563 | d5927e48 | Iustin Pop | logging.error("Could not start the master role on the new master"
|
564 | b726aff0 | Iustin Pop | " %s, please check: %s", new_master, msg)
|
565 | b1b6ea87 | Iustin Pop | rcode = 1
|
566 | b1b6ea87 | Iustin Pop | |
567 | b1b6ea87 | Iustin Pop | return rcode
|
568 | d7cdb55d | Iustin Pop | |
569 | d7cdb55d | Iustin Pop | |
570 | 8eb148ae | Iustin Pop | def GetMaster(): |
571 | 8eb148ae | Iustin Pop | """Returns the current master node.
|
572 | 8eb148ae | Iustin Pop |
|
573 | 8eb148ae | Iustin Pop | This is a separate function in bootstrap since it's needed by
|
574 | 8eb148ae | Iustin Pop | gnt-cluster, and instead of importing directly ssconf, it's better
|
575 | 8eb148ae | Iustin Pop | to abstract it in bootstrap, where we do use ssconf in other
|
576 | 8eb148ae | Iustin Pop | functions too.
|
577 | 8eb148ae | Iustin Pop |
|
578 | 8eb148ae | Iustin Pop | """
|
579 | 8eb148ae | Iustin Pop | sstore = ssconf.SimpleStore() |
580 | 8eb148ae | Iustin Pop | |
581 | 8eb148ae | Iustin Pop | old_master, _ = ssconf.GetMasterAndMyself(sstore) |
582 | 8eb148ae | Iustin Pop | |
583 | 8eb148ae | Iustin Pop | return old_master
|
584 | 8eb148ae | Iustin Pop | |
585 | 8eb148ae | Iustin Pop | |
586 | d7cdb55d | Iustin Pop | def GatherMasterVotes(node_list): |
587 | d7cdb55d | Iustin Pop | """Check the agreement on who is the master.
|
588 | d7cdb55d | Iustin Pop |
|
589 | d7cdb55d | Iustin Pop | This function will return a list of (node, number of votes), ordered
|
590 | d7cdb55d | Iustin Pop | by the number of votes. Errors will be denoted by the key 'None'.
|
591 | d7cdb55d | Iustin Pop |
|
592 | d7cdb55d | Iustin Pop | Note that the sum of votes is the number of nodes this machine
|
593 | d7cdb55d | Iustin Pop | knows, whereas the number of entries in the list could be different
|
594 | d7cdb55d | Iustin Pop | (if some nodes vote for another master).
|
595 | d7cdb55d | Iustin Pop |
|
596 | d7cdb55d | Iustin Pop | We remove ourselves from the list since we know that (bugs aside)
|
597 | d7cdb55d | Iustin Pop | since we use the same source for configuration information for both
|
598 | d7cdb55d | Iustin Pop | backend and boostrap, we'll always vote for ourselves.
|
599 | d7cdb55d | Iustin Pop |
|
600 | d7cdb55d | Iustin Pop | @type node_list: list
|
601 | d7cdb55d | Iustin Pop | @param node_list: the list of nodes to query for master info; the current
|
602 | 5bbd3f7f | Michael Hanselmann | node will be removed if it is in the list
|
603 | d7cdb55d | Iustin Pop | @rtype: list
|
604 | d7cdb55d | Iustin Pop | @return: list of (node, votes)
|
605 | d7cdb55d | Iustin Pop |
|
606 | d7cdb55d | Iustin Pop | """
|
607 | d7cdb55d | Iustin Pop | myself = utils.HostInfo().name |
608 | d7cdb55d | Iustin Pop | try:
|
609 | d7cdb55d | Iustin Pop | node_list.remove(myself) |
610 | d7cdb55d | Iustin Pop | except ValueError: |
611 | d7cdb55d | Iustin Pop | pass
|
612 | d7cdb55d | Iustin Pop | if not node_list: |
613 | d7cdb55d | Iustin Pop | # no nodes left (eventually after removing myself)
|
614 | d7cdb55d | Iustin Pop | return []
|
615 | d7cdb55d | Iustin Pop | results = rpc.RpcRunner.call_master_info(node_list) |
616 | d7cdb55d | Iustin Pop | if not isinstance(results, dict): |
617 | d7cdb55d | Iustin Pop | # this should not happen (unless internal error in rpc)
|
618 | d7cdb55d | Iustin Pop | logging.critical("Can't complete rpc call, aborting master startup")
|
619 | d7cdb55d | Iustin Pop | return [(None, len(node_list))] |
620 | d7cdb55d | Iustin Pop | votes = {} |
621 | d7cdb55d | Iustin Pop | for node in results: |
622 | 781de953 | Iustin Pop | nres = results[node] |
623 | 2a52a064 | Iustin Pop | data = nres.payload |
624 | 3cebe102 | Michael Hanselmann | msg = nres.fail_msg |
625 | 2a52a064 | Iustin Pop | fail = False
|
626 | 2a52a064 | Iustin Pop | if msg:
|
627 | 2a52a064 | Iustin Pop | logging.warning("Error contacting node %s: %s", node, msg)
|
628 | 2a52a064 | Iustin Pop | fail = True
|
629 | 2a52a064 | Iustin Pop | elif not isinstance(data, (tuple, list)) or len(data) < 3: |
630 | 2a52a064 | Iustin Pop | logging.warning("Invalid data received from node %s: %s", node, data)
|
631 | 2a52a064 | Iustin Pop | fail = True
|
632 | 2a52a064 | Iustin Pop | if fail:
|
633 | d7cdb55d | Iustin Pop | if None not in votes: |
634 | d7cdb55d | Iustin Pop | votes[None] = 0 |
635 | d7cdb55d | Iustin Pop | votes[None] += 1 |
636 | d7cdb55d | Iustin Pop | continue
|
637 | 781de953 | Iustin Pop | master_node = data[2]
|
638 | d7cdb55d | Iustin Pop | if master_node not in votes: |
639 | d7cdb55d | Iustin Pop | votes[master_node] = 0
|
640 | d7cdb55d | Iustin Pop | votes[master_node] += 1
|
641 | d7cdb55d | Iustin Pop | |
642 | d7cdb55d | Iustin Pop | vote_list = [v for v in votes.items()] |
643 | d7cdb55d | Iustin Pop | # sort first on number of votes then on name, since we want None
|
644 | d7cdb55d | Iustin Pop | # sorted later if we have the half of the nodes not responding, and
|
645 | d7cdb55d | Iustin Pop | # half voting all for the same master
|
646 | d7cdb55d | Iustin Pop | vote_list.sort(key=lambda x: (x[1], x[0]), reverse=True) |
647 | d7cdb55d | Iustin Pop | |
648 | d7cdb55d | Iustin Pop | return vote_list |