root / lib / bootstrap.py @ 28f34048
History | View | Annotate | Download (22.7 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 | fc0726b9 | Michael Hanselmann | hmackey_file=constants.CONFD_HMAC_KEY, |
84 | fc0726b9 | Michael Hanselmann | cds_file=constants.CLUSTER_DOMAIN_SECRET_FILE): |
85 | 43575108 | Michael Hanselmann | """Updates the cluster certificates, keys and secrets.
|
86 | 43575108 | Michael Hanselmann |
|
87 | 43575108 | Michael Hanselmann | @type new_cluster_cert: bool
|
88 | 43575108 | Michael Hanselmann | @param new_cluster_cert: Whether to generate a new cluster certificate
|
89 | 43575108 | Michael Hanselmann | @type new_rapi_cert: bool
|
90 | 43575108 | Michael Hanselmann | @param new_rapi_cert: Whether to generate a new RAPI certificate
|
91 | 6b7d5878 | Michael Hanselmann | @type new_confd_hmac_key: bool
|
92 | 6b7d5878 | Michael Hanselmann | @param new_confd_hmac_key: Whether to generate a new HMAC key
|
93 | 3db3eb2a | Michael Hanselmann | @type new_cds: bool
|
94 | 3db3eb2a | Michael Hanselmann | @param new_cds: Whether to generate a new cluster domain secret
|
95 | 43575108 | Michael Hanselmann | @type rapi_cert_pem: string
|
96 | 43575108 | Michael Hanselmann | @param rapi_cert_pem: New RAPI certificate in PEM format
|
97 | 3db3eb2a | Michael Hanselmann | @type cds: string
|
98 | 3db3eb2a | Michael Hanselmann | @param cds: New cluster domain secret
|
99 | aeefe835 | Iustin Pop | @type nodecert_file: string
|
100 | aeefe835 | Iustin Pop | @param nodecert_file: optional override of the node cert file path
|
101 | aeefe835 | Iustin Pop | @type rapicert_file: string
|
102 | aeefe835 | Iustin Pop | @param rapicert_file: optional override of the rapi cert file path
|
103 | aeefe835 | Iustin Pop | @type hmackey_file: string
|
104 | aeefe835 | Iustin Pop | @param hmackey_file: optional override of the hmac key file path
|
105 | 43575108 | Michael Hanselmann |
|
106 | 43575108 | Michael Hanselmann | """
|
107 | 168c1de2 | Michael Hanselmann | # noded SSL certificate
|
108 | aeefe835 | Iustin Pop | cluster_cert_exists = os.path.exists(nodecert_file) |
109 | 43575108 | Michael Hanselmann | if new_cluster_cert or not cluster_cert_exists: |
110 | 43575108 | Michael Hanselmann | if cluster_cert_exists:
|
111 | aeefe835 | Iustin Pop | utils.CreateBackup(nodecert_file) |
112 | 43575108 | Michael Hanselmann | |
113 | aeefe835 | Iustin Pop | logging.debug("Generating new cluster certificate at %s", nodecert_file)
|
114 | af2ae1c0 | Iustin Pop | utils.GenerateSelfSignedSslCert(nodecert_file) |
115 | 43575108 | Michael Hanselmann | |
116 | 6b7d5878 | Michael Hanselmann | # confd HMAC key
|
117 | aeefe835 | Iustin Pop | if new_confd_hmac_key or not os.path.exists(hmackey_file): |
118 | aeefe835 | Iustin Pop | logging.debug("Writing new confd HMAC key to %s", hmackey_file)
|
119 | aeefe835 | Iustin Pop | GenerateHmacKey(hmackey_file) |
120 | 43575108 | Michael Hanselmann | |
121 | 43575108 | Michael Hanselmann | # RAPI
|
122 | aeefe835 | Iustin Pop | rapi_cert_exists = os.path.exists(rapicert_file) |
123 | 43575108 | Michael Hanselmann | |
124 | 43575108 | Michael Hanselmann | if rapi_cert_pem:
|
125 | 43575108 | Michael Hanselmann | # Assume rapi_pem contains a valid PEM-formatted certificate and key
|
126 | aeefe835 | Iustin Pop | logging.debug("Writing RAPI certificate at %s", rapicert_file)
|
127 | aeefe835 | Iustin Pop | utils.WriteFile(rapicert_file, data=rapi_cert_pem, backup=True)
|
128 | 43575108 | Michael Hanselmann | |
129 | 43575108 | Michael Hanselmann | elif new_rapi_cert or not rapi_cert_exists: |
130 | 43575108 | Michael Hanselmann | if rapi_cert_exists:
|
131 | aeefe835 | Iustin Pop | utils.CreateBackup(rapicert_file) |
132 | 43575108 | Michael Hanselmann | |
133 | aeefe835 | Iustin Pop | logging.debug("Generating new RAPI certificate at %s", rapicert_file)
|
134 | af2ae1c0 | Iustin Pop | utils.GenerateSelfSignedSslCert(rapicert_file) |
135 | c008906b | Michael Hanselmann | |
136 | 3db3eb2a | Michael Hanselmann | # Cluster domain secret
|
137 | 3db3eb2a | Michael Hanselmann | if cds:
|
138 | fc0726b9 | Michael Hanselmann | logging.debug("Writing cluster domain secret to %s", cds_file)
|
139 | fc0726b9 | Michael Hanselmann | utils.WriteFile(cds_file, data=cds, backup=True)
|
140 | fc0726b9 | Michael Hanselmann | |
141 | fc0726b9 | Michael Hanselmann | elif new_cds or not os.path.exists(cds_file): |
142 | fc0726b9 | Michael Hanselmann | logging.debug("Generating new cluster domain secret at %s", cds_file)
|
143 | fc0726b9 | Michael Hanselmann | GenerateHmacKey(cds_file) |
144 | 3db3eb2a | Michael Hanselmann | |
145 | c008906b | Michael Hanselmann | |
146 | 8f215968 | Michael Hanselmann | def _InitGanetiServerSetup(master_name): |
147 | 40a97d80 | Michael Hanselmann | """Setup the necessary configuration for the initial node daemon.
|
148 | 40a97d80 | Michael Hanselmann |
|
149 | 40a97d80 | Michael Hanselmann | This creates the nodepass file containing the shared password for
|
150 | 40a97d80 | Michael Hanselmann | the cluster and also generates the SSL certificate.
|
151 | 40a97d80 | Michael Hanselmann |
|
152 | 40a97d80 | Michael Hanselmann | """
|
153 | 43575108 | Michael Hanselmann | # Generate cluster secrets
|
154 | 3db3eb2a | Michael Hanselmann | GenerateClusterCrypto(True, False, False, False) |
155 | 4a34c5cf | Guido Trotter | |
156 | f154a7a3 | Michael Hanselmann | result = utils.RunCmd([constants.DAEMON_UTIL, "start", constants.NODED])
|
157 | a0c9f010 | Michael Hanselmann | if result.failed:
|
158 | a0c9f010 | Michael Hanselmann | raise errors.OpExecError("Could not start the node daemon, command %s" |
159 | a0c9f010 | Michael Hanselmann | " had exitcode %s and error %s" %
|
160 | a0c9f010 | Michael Hanselmann | (result.cmd, result.exit_code, result.output)) |
161 | a0c9f010 | Michael Hanselmann | |
162 | 5627f375 | Michael Hanselmann | _WaitForNodeDaemon(master_name) |
163 | 5627f375 | Michael Hanselmann | |
164 | 5627f375 | Michael Hanselmann | |
165 | 5627f375 | Michael Hanselmann | def _WaitForNodeDaemon(node_name): |
166 | 5627f375 | Michael Hanselmann | """Wait for node daemon to become responsive.
|
167 | 5627f375 | Michael Hanselmann |
|
168 | 5627f375 | Michael Hanselmann | """
|
169 | d3833ebd | Michael Hanselmann | def _CheckNodeDaemon(): |
170 | 5627f375 | Michael Hanselmann | result = rpc.RpcRunner.call_version([node_name])[node_name] |
171 | d3833ebd | Michael Hanselmann | if result.fail_msg:
|
172 | d3833ebd | Michael Hanselmann | raise utils.RetryAgain()
|
173 | 8f215968 | Michael Hanselmann | |
174 | d3833ebd | Michael Hanselmann | try:
|
175 | d3833ebd | Michael Hanselmann | utils.Retry(_CheckNodeDaemon, 1.0, 10.0) |
176 | d3833ebd | Michael Hanselmann | except utils.RetryTimeout:
|
177 | 5627f375 | Michael Hanselmann | raise errors.OpExecError("Node daemon on %s didn't answer queries within" |
178 | 5627f375 | Michael Hanselmann | " 10 seconds" % node_name)
|
179 | 5627f375 | Michael Hanselmann | |
180 | a0c9f010 | Michael Hanselmann | |
181 | 0e3baaf3 | Iustin Pop | def _InitFileStorage(file_storage_dir): |
182 | 0e3baaf3 | Iustin Pop | """Initialize if needed the file storage.
|
183 | 0e3baaf3 | Iustin Pop |
|
184 | 0e3baaf3 | Iustin Pop | @param file_storage_dir: the user-supplied value
|
185 | 0e3baaf3 | Iustin Pop | @return: either empty string (if file storage was disabled at build
|
186 | 0e3baaf3 | Iustin Pop | time) or the normalized path to the storage directory
|
187 | 0e3baaf3 | Iustin Pop |
|
188 | 0e3baaf3 | Iustin Pop | """
|
189 | 0e3baaf3 | Iustin Pop | if not constants.ENABLE_FILE_STORAGE: |
190 | 0e3baaf3 | Iustin Pop | return "" |
191 | 0e3baaf3 | Iustin Pop | |
192 | 0e3baaf3 | Iustin Pop | file_storage_dir = os.path.normpath(file_storage_dir) |
193 | 0e3baaf3 | Iustin Pop | |
194 | 0e3baaf3 | Iustin Pop | if not os.path.isabs(file_storage_dir): |
195 | 0e3baaf3 | Iustin Pop | raise errors.OpPrereqError("The file storage directory you passed is" |
196 | 0e3baaf3 | Iustin Pop | " not an absolute path.", errors.ECODE_INVAL)
|
197 | 0e3baaf3 | Iustin Pop | |
198 | 0e3baaf3 | Iustin Pop | if not os.path.exists(file_storage_dir): |
199 | 0e3baaf3 | Iustin Pop | try:
|
200 | 0e3baaf3 | Iustin Pop | os.makedirs(file_storage_dir, 0750)
|
201 | 0e3baaf3 | Iustin Pop | except OSError, err: |
202 | 0e3baaf3 | Iustin Pop | raise errors.OpPrereqError("Cannot create file storage directory" |
203 | 0e3baaf3 | Iustin Pop | " '%s': %s" % (file_storage_dir, err),
|
204 | 0e3baaf3 | Iustin Pop | errors.ECODE_ENVIRON) |
205 | 0e3baaf3 | Iustin Pop | |
206 | 0e3baaf3 | Iustin Pop | if not os.path.isdir(file_storage_dir): |
207 | 0e3baaf3 | Iustin Pop | raise errors.OpPrereqError("The file storage directory '%s' is not" |
208 | 0e3baaf3 | Iustin Pop | " a directory." % file_storage_dir,
|
209 | 0e3baaf3 | Iustin Pop | errors.ECODE_ENVIRON) |
210 | 0e3baaf3 | Iustin Pop | return file_storage_dir
|
211 | 0e3baaf3 | Iustin Pop | |
212 | 0e3baaf3 | Iustin Pop | |
213 | ec0652ad | Guido Trotter | def InitCluster(cluster_name, mac_prefix, |
214 | ce735215 | Guido Trotter | master_netdev, file_storage_dir, candidate_pool_size, |
215 | b6a30b0d | Guido Trotter | secondary_ip=None, vg_name=None, beparams=None, |
216 | b6a30b0d | Guido Trotter | nicparams=None, hvparams=None, enabled_hypervisors=None, |
217 | 3953242f | Iustin Pop | modify_etc_hosts=True, modify_ssh_setup=True, |
218 | 39b0f0c2 | Balazs Lecz | maintain_node_health=False,
|
219 | 39b0f0c2 | Balazs Lecz | uid_pool=None):
|
220 | a0c9f010 | Michael Hanselmann | """Initialise the cluster.
|
221 | a0c9f010 | Michael Hanselmann |
|
222 | ce735215 | Guido Trotter | @type candidate_pool_size: int
|
223 | ce735215 | Guido Trotter | @param candidate_pool_size: master candidate pool size
|
224 | ce735215 | Guido Trotter |
|
225 | a0c9f010 | Michael Hanselmann | """
|
226 | ce735215 | Guido Trotter | # TODO: complete the docstring
|
227 | a0c9f010 | Michael Hanselmann | if config.ConfigWriter.IsCluster():
|
228 | debac808 | Iustin Pop | raise errors.OpPrereqError("Cluster is already initialised", |
229 | debac808 | Iustin Pop | errors.ECODE_STATE) |
230 | a0c9f010 | Michael Hanselmann | |
231 | b119bccb | Guido Trotter | if not enabled_hypervisors: |
232 | b119bccb | Guido Trotter | raise errors.OpPrereqError("Enabled hypervisors list must contain at" |
233 | debac808 | Iustin Pop | " least one member", errors.ECODE_INVAL)
|
234 | b119bccb | Guido Trotter | invalid_hvs = set(enabled_hypervisors) - constants.HYPER_TYPES
|
235 | b119bccb | Guido Trotter | if invalid_hvs:
|
236 | b119bccb | Guido Trotter | raise errors.OpPrereqError("Enabled hypervisors contains invalid" |
237 | debac808 | Iustin Pop | " entries: %s" % invalid_hvs,
|
238 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
239 | b119bccb | Guido Trotter | |
240 | 104f4ca1 | Iustin Pop | hostname = utils.GetHostInfo() |
241 | a0c9f010 | Michael Hanselmann | |
242 | a0c9f010 | Michael Hanselmann | if hostname.ip.startswith("127."): |
243 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("This host's IP resolves to the private" |
244 | a0c9f010 | Michael Hanselmann | " range (%s). Please fix DNS or %s." %
|
245 | debac808 | Iustin Pop | (hostname.ip, constants.ETC_HOSTS), |
246 | debac808 | Iustin Pop | errors.ECODE_ENVIRON) |
247 | a0c9f010 | Michael Hanselmann | |
248 | caad16e2 | Iustin Pop | if not utils.OwnIpAddress(hostname.ip): |
249 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Inconsistency: this host's name resolves" |
250 | a0c9f010 | Michael Hanselmann | " to %s,\nbut this ip address does not"
|
251 | debac808 | Iustin Pop | " belong to this host. Aborting." %
|
252 | debac808 | Iustin Pop | hostname.ip, errors.ECODE_ENVIRON) |
253 | a0c9f010 | Michael Hanselmann | |
254 | 44caf5a8 | Iustin Pop | clustername = utils.GetHostInfo(utils.HostInfo.NormalizeName(cluster_name)) |
255 | a0c9f010 | Michael Hanselmann | |
256 | a0c9f010 | Michael Hanselmann | if utils.TcpPing(clustername.ip, constants.DEFAULT_NODED_PORT,
|
257 | a0c9f010 | Michael Hanselmann | timeout=5):
|
258 | debac808 | Iustin Pop | raise errors.OpPrereqError("Cluster IP already active. Aborting.", |
259 | debac808 | Iustin Pop | errors.ECODE_NOTUNIQUE) |
260 | a0c9f010 | Michael Hanselmann | |
261 | a0c9f010 | Michael Hanselmann | if secondary_ip:
|
262 | a0c9f010 | Michael Hanselmann | if not utils.IsValidIP(secondary_ip): |
263 | debac808 | Iustin Pop | raise errors.OpPrereqError("Invalid secondary ip given", |
264 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
265 | a0c9f010 | Michael Hanselmann | if (secondary_ip != hostname.ip and |
266 | caad16e2 | Iustin Pop | not utils.OwnIpAddress(secondary_ip)):
|
267 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("You gave %s as secondary IP," |
268 | a0c9f010 | Michael Hanselmann | " but it does not belong to this host." %
|
269 | debac808 | Iustin Pop | secondary_ip, errors.ECODE_ENVIRON) |
270 | b9eeeb02 | Michael Hanselmann | else:
|
271 | b9eeeb02 | Michael Hanselmann | secondary_ip = hostname.ip |
272 | a0c9f010 | Michael Hanselmann | |
273 | a0c9f010 | Michael Hanselmann | if vg_name is not None: |
274 | a0c9f010 | Michael Hanselmann | # Check if volume group is valid
|
275 | a0c9f010 | Michael Hanselmann | vgstatus = utils.CheckVolumeGroupSize(utils.ListVolumeGroups(), vg_name, |
276 | a0c9f010 | Michael Hanselmann | constants.MIN_VG_SIZE) |
277 | a0c9f010 | Michael Hanselmann | if vgstatus:
|
278 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Error: %s\nspecify --no-lvm-storage if" |
279 | debac808 | Iustin Pop | " you are not using lvm" % vgstatus,
|
280 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
281 | a0c9f010 | Michael Hanselmann | |
282 | 0e3baaf3 | Iustin Pop | file_storage_dir = _InitFileStorage(file_storage_dir) |
283 | a0c9f010 | Michael Hanselmann | |
284 | a0c9f010 | Michael Hanselmann | if not re.match("^[0-9a-z]{2}:[0-9a-z]{2}:[0-9a-z]{2}$", mac_prefix): |
285 | debac808 | Iustin Pop | raise errors.OpPrereqError("Invalid mac prefix given '%s'" % mac_prefix, |
286 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
287 | a0c9f010 | Michael Hanselmann | |
288 | a0c9f010 | Michael Hanselmann | result = utils.RunCmd(["ip", "link", "show", "dev", master_netdev]) |
289 | a0c9f010 | Michael Hanselmann | if result.failed:
|
290 | a0c9f010 | Michael Hanselmann | raise errors.OpPrereqError("Invalid master netdev given (%s): '%s'" % |
291 | a0c9f010 | Michael Hanselmann | (master_netdev, |
292 | debac808 | Iustin Pop | result.output.strip()), errors.ECODE_INVAL) |
293 | a0c9f010 | Michael Hanselmann | |
294 | 9dae41ad | Guido Trotter | dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE)] |
295 | 9dae41ad | Guido Trotter | utils.EnsureDirs(dirs) |
296 | 9dae41ad | Guido Trotter | |
297 | a5728081 | Guido Trotter | utils.ForceDictType(beparams, constants.BES_PARAMETER_TYPES) |
298 | b6a30b0d | Guido Trotter | utils.ForceDictType(nicparams, constants.NICS_PARAMETER_TYPES) |
299 | b6a30b0d | Guido Trotter | objects.NIC.CheckParameterSyntax(nicparams) |
300 | b6a30b0d | Guido Trotter | |
301 | a5728081 | Guido Trotter | # hvparams is a mapping of hypervisor->hvparams dict
|
302 | a5728081 | Guido Trotter | for hv_name, hv_params in hvparams.iteritems(): |
303 | a5728081 | Guido Trotter | utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES) |
304 | a5728081 | Guido Trotter | hv_class = hypervisor.GetHypervisor(hv_name) |
305 | a5728081 | Guido Trotter | hv_class.CheckParameterSyntax(hv_params) |
306 | d4b72030 | Guido Trotter | |
307 | a0c9f010 | Michael Hanselmann | # set up the inter-node password and certificate
|
308 | 8f215968 | Michael Hanselmann | _InitGanetiServerSetup(hostname.name) |
309 | a0c9f010 | Michael Hanselmann | |
310 | a0c9f010 | Michael Hanselmann | # set up ssh config and /etc/hosts
|
311 | 13998ef2 | Michael Hanselmann | sshline = utils.ReadFile(constants.SSH_HOST_RSA_PUB) |
312 | a0c9f010 | Michael Hanselmann | sshkey = sshline.split(" ")[1] |
313 | a0c9f010 | Michael Hanselmann | |
314 | b86a6bcd | Guido Trotter | if modify_etc_hosts:
|
315 | b86a6bcd | Guido Trotter | utils.AddHostToEtcHosts(hostname.name) |
316 | b86a6bcd | Guido Trotter | |
317 | b989b9d9 | Ken Wehr | if modify_ssh_setup:
|
318 | b989b9d9 | Ken Wehr | _InitSSHSetup() |
319 | a0c9f010 | Michael Hanselmann | |
320 | 430b923c | Iustin Pop | now = time.time() |
321 | 430b923c | Iustin Pop | |
322 | a0c9f010 | Michael Hanselmann | # init of cluster config file
|
323 | b9eeeb02 | Michael Hanselmann | cluster_config = objects.Cluster( |
324 | b9eeeb02 | Michael Hanselmann | serial_no=1,
|
325 | b9eeeb02 | Michael Hanselmann | rsahostkeypub=sshkey, |
326 | b9eeeb02 | Michael Hanselmann | highest_used_port=(constants.FIRST_DRBD_PORT - 1),
|
327 | b9eeeb02 | Michael Hanselmann | mac_prefix=mac_prefix, |
328 | b9eeeb02 | Michael Hanselmann | volume_group_name=vg_name, |
329 | b9eeeb02 | Michael Hanselmann | tcpudp_port_pool=set(),
|
330 | f6bd6e98 | Michael Hanselmann | master_node=hostname.name, |
331 | f6bd6e98 | Michael Hanselmann | master_ip=clustername.ip, |
332 | f6bd6e98 | Michael Hanselmann | master_netdev=master_netdev, |
333 | f6bd6e98 | Michael Hanselmann | cluster_name=clustername.name, |
334 | f6bd6e98 | Michael Hanselmann | file_storage_dir=file_storage_dir, |
335 | ea3a925f | Alexander Schreiber | enabled_hypervisors=enabled_hypervisors, |
336 | 4ef7f423 | Guido Trotter | beparams={constants.PP_DEFAULT: beparams}, |
337 | b6a30b0d | Guido Trotter | nicparams={constants.PP_DEFAULT: nicparams}, |
338 | ea3a925f | Alexander Schreiber | hvparams=hvparams, |
339 | ce735215 | Guido Trotter | candidate_pool_size=candidate_pool_size, |
340 | 022c3a0b | Guido Trotter | modify_etc_hosts=modify_etc_hosts, |
341 | b989b9d9 | Ken Wehr | modify_ssh_setup=modify_ssh_setup, |
342 | 39b0f0c2 | Balazs Lecz | uid_pool=uid_pool, |
343 | 430b923c | Iustin Pop | ctime=now, |
344 | 430b923c | Iustin Pop | mtime=now, |
345 | 430b923c | Iustin Pop | uuid=utils.NewUUID(), |
346 | 3953242f | Iustin Pop | maintain_node_health=maintain_node_health, |
347 | b9eeeb02 | Michael Hanselmann | ) |
348 | b9eeeb02 | Michael Hanselmann | master_node_config = objects.Node(name=hostname.name, |
349 | b9eeeb02 | Michael Hanselmann | primary_ip=hostname.ip, |
350 | b9222f32 | Guido Trotter | secondary_ip=secondary_ip, |
351 | c044f32c | Guido Trotter | serial_no=1,
|
352 | c044f32c | Guido Trotter | master_candidate=True,
|
353 | af64c0ea | Iustin Pop | offline=False, drained=False, |
354 | c044f32c | Guido Trotter | ) |
355 | 9e1333b9 | Guido Trotter | InitConfig(constants.CONFIG_VERSION, cluster_config, master_node_config) |
356 | 05cc153f | Guido Trotter | cfg = config.ConfigWriter() |
357 | 9e1333b9 | Guido Trotter | ssh.WriteKnownHostsFile(cfg, constants.SSH_KNOWN_HOSTS_FILE) |
358 | a4eae71f | Michael Hanselmann | cfg.Update(cfg.GetClusterInfo(), logging.error) |
359 | 827f753e | Guido Trotter | |
360 | b3f1cf6f | Iustin Pop | # start the master ip
|
361 | b3f1cf6f | Iustin Pop | # TODO: Review rpc call from bootstrap
|
362 | b726aff0 | Iustin Pop | # TODO: Warn on failed start master
|
363 | 3583908a | Guido Trotter | rpc.RpcRunner.call_node_start_master(hostname.name, True, False) |
364 | b3f1cf6f | Iustin Pop | |
365 | b1b6ea87 | Iustin Pop | |
366 | 02f99608 | Oleksiy Mishchenko | def InitConfig(version, cluster_config, master_node_config, |
367 | 02f99608 | Oleksiy Mishchenko | cfg_file=constants.CLUSTER_CONF_FILE): |
368 | 7b3a8fb5 | Iustin Pop | """Create the initial cluster configuration.
|
369 | 7b3a8fb5 | Iustin Pop |
|
370 | 7b3a8fb5 | Iustin Pop | It will contain the current node, which will also be the master
|
371 | 7b3a8fb5 | Iustin Pop | node, and no instances.
|
372 | 7b3a8fb5 | Iustin Pop |
|
373 | 7b3a8fb5 | Iustin Pop | @type version: int
|
374 | c41eea6e | Iustin Pop | @param version: configuration version
|
375 | c41eea6e | Iustin Pop | @type cluster_config: L{objects.Cluster}
|
376 | c41eea6e | Iustin Pop | @param cluster_config: cluster configuration
|
377 | c41eea6e | Iustin Pop | @type master_node_config: L{objects.Node}
|
378 | c41eea6e | Iustin Pop | @param master_node_config: master node configuration
|
379 | c41eea6e | Iustin Pop | @type cfg_file: string
|
380 | c41eea6e | Iustin Pop | @param cfg_file: configuration file path
|
381 | c41eea6e | Iustin Pop |
|
382 | 7b3a8fb5 | Iustin Pop | """
|
383 | 7b3a8fb5 | Iustin Pop | nodes = { |
384 | 7b3a8fb5 | Iustin Pop | master_node_config.name: master_node_config, |
385 | 7b3a8fb5 | Iustin Pop | } |
386 | 7b3a8fb5 | Iustin Pop | |
387 | d693c864 | Iustin Pop | now = time.time() |
388 | 7b3a8fb5 | Iustin Pop | config_data = objects.ConfigData(version=version, |
389 | 7b3a8fb5 | Iustin Pop | cluster=cluster_config, |
390 | 7b3a8fb5 | Iustin Pop | nodes=nodes, |
391 | 7b3a8fb5 | Iustin Pop | instances={}, |
392 | d693c864 | Iustin Pop | serial_no=1,
|
393 | d693c864 | Iustin Pop | ctime=now, mtime=now) |
394 | a33848a5 | Guido Trotter | utils.WriteFile(cfg_file, |
395 | a33848a5 | Guido Trotter | data=serializer.Dump(config_data.ToDict()), |
396 | a33848a5 | Guido Trotter | mode=0600)
|
397 | 02f99608 | Oleksiy Mishchenko | |
398 | 02f99608 | Oleksiy Mishchenko | |
399 | 140aa4a8 | Iustin Pop | def FinalizeClusterDestroy(master): |
400 | 140aa4a8 | Iustin Pop | """Execute the last steps of cluster destroy
|
401 | 140aa4a8 | Iustin Pop |
|
402 | 140aa4a8 | Iustin Pop | This function shuts down all the daemons, completing the destroy
|
403 | 140aa4a8 | Iustin Pop | begun in cmdlib.LUDestroyOpcode.
|
404 | 140aa4a8 | Iustin Pop |
|
405 | 140aa4a8 | Iustin Pop | """
|
406 | b989b9d9 | Ken Wehr | cfg = config.ConfigWriter() |
407 | b989b9d9 | Ken Wehr | modify_ssh_setup = cfg.GetClusterInfo().modify_ssh_setup |
408 | 781de953 | Iustin Pop | result = rpc.RpcRunner.call_node_stop_master(master, True)
|
409 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
410 | 6c00d19a | Iustin Pop | if msg:
|
411 | 099c52ad | Iustin Pop | logging.warning("Could not disable the master role: %s", msg)
|
412 | b989b9d9 | Ken Wehr | result = rpc.RpcRunner.call_node_leave_cluster(master, modify_ssh_setup) |
413 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
414 | 0623d351 | Iustin Pop | if msg:
|
415 | 0623d351 | Iustin Pop | logging.warning("Could not shutdown the node daemon and cleanup"
|
416 | 0623d351 | Iustin Pop | " the node: %s", msg)
|
417 | 140aa4a8 | Iustin Pop | |
418 | 140aa4a8 | Iustin Pop | |
419 | 87622829 | Iustin Pop | def SetupNodeDaemon(cluster_name, node, ssh_key_check): |
420 | 827f753e | Guido Trotter | """Add a node to the cluster.
|
421 | 827f753e | Guido Trotter |
|
422 | b1b6ea87 | Iustin Pop | This function must be called before the actual opcode, and will ssh
|
423 | b1b6ea87 | Iustin Pop | to the remote node, copy the needed files, and start ganeti-noded,
|
424 | b1b6ea87 | Iustin Pop | allowing the master to do the rest via normal rpc calls.
|
425 | 827f753e | Guido Trotter |
|
426 | 87622829 | Iustin Pop | @param cluster_name: the cluster name
|
427 | 87622829 | Iustin Pop | @param node: the name of the new node
|
428 | 87622829 | Iustin Pop | @param ssh_key_check: whether to do a strict key check
|
429 | 827f753e | Guido Trotter |
|
430 | 827f753e | Guido Trotter | """
|
431 | 87622829 | Iustin Pop | sshrunner = ssh.SshRunner(cluster_name) |
432 | 5557b04c | Michael Hanselmann | |
433 | 168c1de2 | Michael Hanselmann | noded_cert = utils.ReadFile(constants.NODED_CERT_FILE) |
434 | 2438c157 | Michael Hanselmann | rapi_cert = utils.ReadFile(constants.RAPI_CERT_FILE) |
435 | 6b7d5878 | Michael Hanselmann | confd_hmac_key = utils.ReadFile(constants.CONFD_HMAC_KEY) |
436 | 5557b04c | Michael Hanselmann | |
437 | 827f753e | Guido Trotter | # in the base64 pem encoding, neither '!' nor '.' are valid chars,
|
438 | 827f753e | Guido Trotter | # so we use this to detect an invalid certificate; as long as the
|
439 | 827f753e | Guido Trotter | # cert doesn't contain this, the here-document will be correctly
|
440 | 77b076ca | Guido Trotter | # parsed by the shell sequence below. HMAC keys are hexadecimal strings,
|
441 | 77b076ca | Guido Trotter | # so the same restrictions apply.
|
442 | 6b7d5878 | Michael Hanselmann | for content in (noded_cert, rapi_cert, confd_hmac_key): |
443 | 77b076ca | Guido Trotter | if re.search('^!EOF\.', content, re.MULTILINE): |
444 | 77b076ca | Guido Trotter | raise errors.OpExecError("invalid SSL certificate or HMAC key") |
445 | 5557b04c | Michael Hanselmann | |
446 | 5557b04c | Michael Hanselmann | if not noded_cert.endswith("\n"): |
447 | 5557b04c | Michael Hanselmann | noded_cert += "\n"
|
448 | 2438c157 | Michael Hanselmann | if not rapi_cert.endswith("\n"): |
449 | 2438c157 | Michael Hanselmann | rapi_cert += "\n"
|
450 | 6b7d5878 | Michael Hanselmann | if not confd_hmac_key.endswith("\n"): |
451 | 6b7d5878 | Michael Hanselmann | confd_hmac_key += "\n"
|
452 | 827f753e | Guido Trotter | |
453 | 827f753e | Guido Trotter | # set up inter-node password and certificate and restarts the node daemon
|
454 | 827f753e | Guido Trotter | # and then connect with ssh to set password and start ganeti-noded
|
455 | 827f753e | Guido Trotter | # note that all the below variables are sanitized at this point,
|
456 | 827f753e | Guido Trotter | # either by being constants or by the checks above
|
457 | 3db3eb2a | Michael Hanselmann | # TODO: Could this command exceed a shell's maximum command length?
|
458 | 827f753e | Guido Trotter | mycommand = ("umask 077 && "
|
459 | 827f753e | Guido Trotter | "cat > '%s' << '!EOF.' && \n"
|
460 | 2438c157 | Michael Hanselmann | "%s!EOF.\n"
|
461 | 2438c157 | Michael Hanselmann | "cat > '%s' << '!EOF.' && \n"
|
462 | 2438c157 | Michael Hanselmann | "%s!EOF.\n"
|
463 | 77b076ca | Guido Trotter | "cat > '%s' << '!EOF.' && \n"
|
464 | 77b076ca | Guido Trotter | "%s!EOF.\n"
|
465 | 77b076ca | Guido Trotter | "chmod 0400 %s %s %s && "
|
466 | f154a7a3 | Michael Hanselmann | "%s start %s" %
|
467 | 168c1de2 | Michael Hanselmann | (constants.NODED_CERT_FILE, noded_cert, |
468 | 2438c157 | Michael Hanselmann | constants.RAPI_CERT_FILE, rapi_cert, |
469 | 6b7d5878 | Michael Hanselmann | constants.CONFD_HMAC_KEY, confd_hmac_key, |
470 | 168c1de2 | Michael Hanselmann | constants.NODED_CERT_FILE, constants.RAPI_CERT_FILE, |
471 | 6b7d5878 | Michael Hanselmann | constants.CONFD_HMAC_KEY, |
472 | f154a7a3 | Michael Hanselmann | constants.DAEMON_UTIL, constants.NODED)) |
473 | 827f753e | Guido Trotter | |
474 | c4b6c29c | Michael Hanselmann | result = sshrunner.Run(node, 'root', mycommand, batch=False, |
475 | c4b6c29c | Michael Hanselmann | ask_key=ssh_key_check, |
476 | c4b6c29c | Michael Hanselmann | use_cluster_key=False,
|
477 | c4b6c29c | Michael Hanselmann | strict_host_check=ssh_key_check) |
478 | 827f753e | Guido Trotter | if result.failed:
|
479 | 827f753e | Guido Trotter | raise errors.OpExecError("Remote command on node %s, error: %s," |
480 | 827f753e | Guido Trotter | " output: %s" %
|
481 | 827f753e | Guido Trotter | (node, result.fail_reason, result.output)) |
482 | 827f753e | Guido Trotter | |
483 | 5627f375 | Michael Hanselmann | _WaitForNodeDaemon(node) |
484 | 5627f375 | Michael Hanselmann | |
485 | b1b6ea87 | Iustin Pop | |
486 | 8e2524c3 | Guido Trotter | def MasterFailover(no_voting=False): |
487 | b1b6ea87 | Iustin Pop | """Failover the master node.
|
488 | b1b6ea87 | Iustin Pop |
|
489 | b1b6ea87 | Iustin Pop | This checks that we are not already the master, and will cause the
|
490 | b1b6ea87 | Iustin Pop | current master to cease being master, and the non-master to become
|
491 | b1b6ea87 | Iustin Pop | new master.
|
492 | b1b6ea87 | Iustin Pop |
|
493 | 8e2524c3 | Guido Trotter | @type no_voting: boolean
|
494 | 8e2524c3 | Guido Trotter | @param no_voting: force the operation without remote nodes agreement
|
495 | 8e2524c3 | Guido Trotter | (dangerous)
|
496 | 8e2524c3 | Guido Trotter |
|
497 | b1b6ea87 | Iustin Pop | """
|
498 | 8135a2db | Iustin Pop | sstore = ssconf.SimpleStore() |
499 | b1b6ea87 | Iustin Pop | |
500 | 8135a2db | Iustin Pop | old_master, new_master = ssconf.GetMasterAndMyself(sstore) |
501 | 8135a2db | Iustin Pop | node_list = sstore.GetNodeList() |
502 | 8135a2db | Iustin Pop | mc_list = sstore.GetMasterCandidates() |
503 | b1b6ea87 | Iustin Pop | |
504 | b1b6ea87 | Iustin Pop | if old_master == new_master:
|
505 | b1b6ea87 | Iustin Pop | raise errors.OpPrereqError("This commands must be run on the node" |
506 | b1b6ea87 | Iustin Pop | " where you want the new master to be."
|
507 | b1b6ea87 | Iustin Pop | " %s is already the master" %
|
508 | debac808 | Iustin Pop | old_master, errors.ECODE_INVAL) |
509 | d5927e48 | Iustin Pop | |
510 | 8135a2db | Iustin Pop | if new_master not in mc_list: |
511 | 8135a2db | Iustin Pop | mc_no_master = [name for name in mc_list if name != old_master] |
512 | 8135a2db | Iustin Pop | raise errors.OpPrereqError("This node is not among the nodes marked" |
513 | 8135a2db | Iustin Pop | " as master candidates. Only these nodes"
|
514 | 8135a2db | Iustin Pop | " can become masters. Current list of"
|
515 | 8135a2db | Iustin Pop | " master candidates is:\n"
|
516 | debac808 | Iustin Pop | "%s" % ('\n'.join(mc_no_master)), |
517 | debac808 | Iustin Pop | errors.ECODE_STATE) |
518 | 8135a2db | Iustin Pop | |
519 | 8e2524c3 | Guido Trotter | if not no_voting: |
520 | 8e2524c3 | Guido Trotter | vote_list = GatherMasterVotes(node_list) |
521 | 8e2524c3 | Guido Trotter | |
522 | 8e2524c3 | Guido Trotter | if vote_list:
|
523 | 8e2524c3 | Guido Trotter | voted_master = vote_list[0][0] |
524 | 8e2524c3 | Guido Trotter | if voted_master is None: |
525 | 8e2524c3 | Guido Trotter | raise errors.OpPrereqError("Cluster is inconsistent, most nodes did" |
526 | debac808 | Iustin Pop | " not respond.", errors.ECODE_ENVIRON)
|
527 | 8e2524c3 | Guido Trotter | elif voted_master != old_master:
|
528 | 8e2524c3 | Guido Trotter | raise errors.OpPrereqError("I have a wrong configuration, I believe" |
529 | 8e2524c3 | Guido Trotter | " the master is %s but the other nodes"
|
530 | 8e2524c3 | Guido Trotter | " voted %s. Please resync the configuration"
|
531 | 8e2524c3 | Guido Trotter | " of this node." %
|
532 | debac808 | Iustin Pop | (old_master, voted_master), |
533 | debac808 | Iustin Pop | errors.ECODE_STATE) |
534 | b1b6ea87 | Iustin Pop | # end checks
|
535 | b1b6ea87 | Iustin Pop | |
536 | b1b6ea87 | Iustin Pop | rcode = 0
|
537 | b1b6ea87 | Iustin Pop | |
538 | d5927e48 | Iustin Pop | logging.info("Setting master to %s, old master: %s", new_master, old_master)
|
539 | b1b6ea87 | Iustin Pop | |
540 | 781de953 | Iustin Pop | result = rpc.RpcRunner.call_node_stop_master(old_master, True)
|
541 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
542 | 6c00d19a | Iustin Pop | if msg:
|
543 | d5927e48 | Iustin Pop | logging.error("Could not disable the master role on the old master"
|
544 | 6c00d19a | Iustin Pop | " %s, please disable manually: %s", old_master, msg)
|
545 | b1b6ea87 | Iustin Pop | |
546 | d23ef431 | Michael Hanselmann | # Here we have a phase where no master should be running
|
547 | b1b6ea87 | Iustin Pop | |
548 | bbe19c17 | Iustin Pop | # instantiate a real config writer, as we now know we have the
|
549 | bbe19c17 | Iustin Pop | # configuration data
|
550 | bbe19c17 | Iustin Pop | cfg = config.ConfigWriter() |
551 | b1b6ea87 | Iustin Pop | |
552 | bbe19c17 | Iustin Pop | cluster_info = cfg.GetClusterInfo() |
553 | bbe19c17 | Iustin Pop | cluster_info.master_node = new_master |
554 | bbe19c17 | Iustin Pop | # this will also regenerate the ssconf files, since we updated the
|
555 | bbe19c17 | Iustin Pop | # cluster info
|
556 | a4eae71f | Michael Hanselmann | cfg.Update(cluster_info, logging.error) |
557 | d5927e48 | Iustin Pop | |
558 | 3583908a | Guido Trotter | result = rpc.RpcRunner.call_node_start_master(new_master, True, no_voting)
|
559 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
560 | b726aff0 | Iustin Pop | if msg:
|
561 | d5927e48 | Iustin Pop | logging.error("Could not start the master role on the new master"
|
562 | b726aff0 | Iustin Pop | " %s, please check: %s", new_master, msg)
|
563 | b1b6ea87 | Iustin Pop | rcode = 1
|
564 | b1b6ea87 | Iustin Pop | |
565 | b1b6ea87 | Iustin Pop | return rcode
|
566 | d7cdb55d | Iustin Pop | |
567 | d7cdb55d | Iustin Pop | |
568 | 8eb148ae | Iustin Pop | def GetMaster(): |
569 | 8eb148ae | Iustin Pop | """Returns the current master node.
|
570 | 8eb148ae | Iustin Pop |
|
571 | 8eb148ae | Iustin Pop | This is a separate function in bootstrap since it's needed by
|
572 | 8eb148ae | Iustin Pop | gnt-cluster, and instead of importing directly ssconf, it's better
|
573 | 8eb148ae | Iustin Pop | to abstract it in bootstrap, where we do use ssconf in other
|
574 | 8eb148ae | Iustin Pop | functions too.
|
575 | 8eb148ae | Iustin Pop |
|
576 | 8eb148ae | Iustin Pop | """
|
577 | 8eb148ae | Iustin Pop | sstore = ssconf.SimpleStore() |
578 | 8eb148ae | Iustin Pop | |
579 | 8eb148ae | Iustin Pop | old_master, _ = ssconf.GetMasterAndMyself(sstore) |
580 | 8eb148ae | Iustin Pop | |
581 | 8eb148ae | Iustin Pop | return old_master
|
582 | 8eb148ae | Iustin Pop | |
583 | 8eb148ae | Iustin Pop | |
584 | d7cdb55d | Iustin Pop | def GatherMasterVotes(node_list): |
585 | d7cdb55d | Iustin Pop | """Check the agreement on who is the master.
|
586 | d7cdb55d | Iustin Pop |
|
587 | d7cdb55d | Iustin Pop | This function will return a list of (node, number of votes), ordered
|
588 | d7cdb55d | Iustin Pop | by the number of votes. Errors will be denoted by the key 'None'.
|
589 | d7cdb55d | Iustin Pop |
|
590 | d7cdb55d | Iustin Pop | Note that the sum of votes is the number of nodes this machine
|
591 | d7cdb55d | Iustin Pop | knows, whereas the number of entries in the list could be different
|
592 | d7cdb55d | Iustin Pop | (if some nodes vote for another master).
|
593 | d7cdb55d | Iustin Pop |
|
594 | d7cdb55d | Iustin Pop | We remove ourselves from the list since we know that (bugs aside)
|
595 | d7cdb55d | Iustin Pop | since we use the same source for configuration information for both
|
596 | d7cdb55d | Iustin Pop | backend and boostrap, we'll always vote for ourselves.
|
597 | d7cdb55d | Iustin Pop |
|
598 | d7cdb55d | Iustin Pop | @type node_list: list
|
599 | d7cdb55d | Iustin Pop | @param node_list: the list of nodes to query for master info; the current
|
600 | 5bbd3f7f | Michael Hanselmann | node will be removed if it is in the list
|
601 | d7cdb55d | Iustin Pop | @rtype: list
|
602 | d7cdb55d | Iustin Pop | @return: list of (node, votes)
|
603 | d7cdb55d | Iustin Pop |
|
604 | d7cdb55d | Iustin Pop | """
|
605 | d7cdb55d | Iustin Pop | myself = utils.HostInfo().name |
606 | d7cdb55d | Iustin Pop | try:
|
607 | d7cdb55d | Iustin Pop | node_list.remove(myself) |
608 | d7cdb55d | Iustin Pop | except ValueError: |
609 | d7cdb55d | Iustin Pop | pass
|
610 | d7cdb55d | Iustin Pop | if not node_list: |
611 | d7cdb55d | Iustin Pop | # no nodes left (eventually after removing myself)
|
612 | d7cdb55d | Iustin Pop | return []
|
613 | d7cdb55d | Iustin Pop | results = rpc.RpcRunner.call_master_info(node_list) |
614 | d7cdb55d | Iustin Pop | if not isinstance(results, dict): |
615 | d7cdb55d | Iustin Pop | # this should not happen (unless internal error in rpc)
|
616 | d7cdb55d | Iustin Pop | logging.critical("Can't complete rpc call, aborting master startup")
|
617 | d7cdb55d | Iustin Pop | return [(None, len(node_list))] |
618 | d7cdb55d | Iustin Pop | votes = {} |
619 | d7cdb55d | Iustin Pop | for node in results: |
620 | 781de953 | Iustin Pop | nres = results[node] |
621 | 2a52a064 | Iustin Pop | data = nres.payload |
622 | 3cebe102 | Michael Hanselmann | msg = nres.fail_msg |
623 | 2a52a064 | Iustin Pop | fail = False
|
624 | 2a52a064 | Iustin Pop | if msg:
|
625 | 2a52a064 | Iustin Pop | logging.warning("Error contacting node %s: %s", node, msg)
|
626 | 2a52a064 | Iustin Pop | fail = True
|
627 | 2a52a064 | Iustin Pop | elif not isinstance(data, (tuple, list)) or len(data) < 3: |
628 | 2a52a064 | Iustin Pop | logging.warning("Invalid data received from node %s: %s", node, data)
|
629 | 2a52a064 | Iustin Pop | fail = True
|
630 | 2a52a064 | Iustin Pop | if fail:
|
631 | d7cdb55d | Iustin Pop | if None not in votes: |
632 | d7cdb55d | Iustin Pop | votes[None] = 0 |
633 | d7cdb55d | Iustin Pop | votes[None] += 1 |
634 | d7cdb55d | Iustin Pop | continue
|
635 | 781de953 | Iustin Pop | master_node = data[2]
|
636 | d7cdb55d | Iustin Pop | if master_node not in votes: |
637 | d7cdb55d | Iustin Pop | votes[master_node] = 0
|
638 | d7cdb55d | Iustin Pop | votes[master_node] += 1
|
639 | d7cdb55d | Iustin Pop | |
640 | d7cdb55d | Iustin Pop | vote_list = [v for v in votes.items()] |
641 | d7cdb55d | Iustin Pop | # sort first on number of votes then on name, since we want None
|
642 | d7cdb55d | Iustin Pop | # sorted later if we have the half of the nodes not responding, and
|
643 | d7cdb55d | Iustin Pop | # half voting all for the same master
|
644 | d7cdb55d | Iustin Pop | vote_list.sort(key=lambda x: (x[1], x[0]), reverse=True) |
645 | d7cdb55d | Iustin Pop | |
646 | d7cdb55d | Iustin Pop | return vote_list |