root / lib / server / masterd.py @ cb4d3314
History | View | Annotate | Download (23.6 kB)
1 | 69cf3abd | Michael Hanselmann | #
|
---|---|---|---|
2 | ffeffa1d | Iustin Pop | #
|
3 | ffeffa1d | Iustin Pop | |
4 | f2af0bec | Iustin Pop | # Copyright (C) 2006, 2007, 2010, 2011 Google Inc.
|
5 | ffeffa1d | Iustin Pop | #
|
6 | ffeffa1d | Iustin Pop | # This program is free software; you can redistribute it and/or modify
|
7 | ffeffa1d | Iustin Pop | # it under the terms of the GNU General Public License as published by
|
8 | ffeffa1d | Iustin Pop | # the Free Software Foundation; either version 2 of the License, or
|
9 | ffeffa1d | Iustin Pop | # (at your option) any later version.
|
10 | ffeffa1d | Iustin Pop | #
|
11 | ffeffa1d | Iustin Pop | # This program is distributed in the hope that it will be useful, but
|
12 | ffeffa1d | Iustin Pop | # WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 | ffeffa1d | Iustin Pop | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 | ffeffa1d | Iustin Pop | # General Public License for more details.
|
15 | ffeffa1d | Iustin Pop | #
|
16 | ffeffa1d | Iustin Pop | # You should have received a copy of the GNU General Public License
|
17 | ffeffa1d | Iustin Pop | # along with this program; if not, write to the Free Software
|
18 | ffeffa1d | Iustin Pop | # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
19 | ffeffa1d | Iustin Pop | # 02110-1301, USA.
|
20 | ffeffa1d | Iustin Pop | |
21 | ffeffa1d | Iustin Pop | |
22 | ffeffa1d | Iustin Pop | """Master daemon program.
|
23 | ffeffa1d | Iustin Pop |
|
24 | ffeffa1d | Iustin Pop | Some classes deviates from the standard style guide since the
|
25 | ffeffa1d | Iustin Pop | inheritance from parent classes requires it.
|
26 | ffeffa1d | Iustin Pop |
|
27 | ffeffa1d | Iustin Pop | """
|
28 | ffeffa1d | Iustin Pop | |
29 | b459a848 | Andrea Spadaccini | # pylint: disable=C0103
|
30 | 7260cfbe | Iustin Pop | # C0103: Invalid name ganeti-masterd
|
31 | ffeffa1d | Iustin Pop | |
32 | bbfd0568 | René Nussbaumer | import grp |
33 | bbfd0568 | René Nussbaumer | import os |
34 | bbfd0568 | René Nussbaumer | import pwd |
35 | c1f2901b | Iustin Pop | import sys |
36 | cdd7f900 | Guido Trotter | import socket |
37 | ffeffa1d | Iustin Pop | import time |
38 | bbfd0568 | René Nussbaumer | import tempfile |
39 | 96cb3986 | Michael Hanselmann | import logging |
40 | ffeffa1d | Iustin Pop | |
41 | c1f2901b | Iustin Pop | from optparse import OptionParser |
42 | ffeffa1d | Iustin Pop | |
43 | 39dcf2ef | Guido Trotter | from ganeti import config |
44 | ffeffa1d | Iustin Pop | from ganeti import constants |
45 | 04ccf5e9 | Guido Trotter | from ganeti import daemon |
46 | ffeffa1d | Iustin Pop | from ganeti import mcpu |
47 | ffeffa1d | Iustin Pop | from ganeti import opcodes |
48 | ffeffa1d | Iustin Pop | from ganeti import jqueue |
49 | 39dcf2ef | Guido Trotter | from ganeti import locking |
50 | ffeffa1d | Iustin Pop | from ganeti import luxi |
51 | ffeffa1d | Iustin Pop | from ganeti import utils |
52 | c1f2901b | Iustin Pop | from ganeti import errors |
53 | c1f2901b | Iustin Pop | from ganeti import ssconf |
54 | 23e50d39 | Michael Hanselmann | from ganeti import workerpool |
55 | b1b6ea87 | Iustin Pop | from ganeti import rpc |
56 | d7cdb55d | Iustin Pop | from ganeti import bootstrap |
57 | a744b676 | Manuel Franceschini | from ganeti import netutils |
58 | 28b71a76 | Michael Hanselmann | from ganeti import objects |
59 | 24d16f76 | Michael Hanselmann | from ganeti import query |
60 | c1f2901b | Iustin Pop | |
61 | c1f2901b | Iustin Pop | |
62 | 23e50d39 | Michael Hanselmann | CLIENT_REQUEST_WORKERS = 16
|
63 | 23e50d39 | Michael Hanselmann | |
64 | c1f2901b | Iustin Pop | EXIT_NOTMASTER = constants.EXIT_NOTMASTER |
65 | c1f2901b | Iustin Pop | EXIT_NODESETUP_ERROR = constants.EXIT_NODESETUP_ERROR |
66 | ffeffa1d | Iustin Pop | |
67 | ffeffa1d | Iustin Pop | |
68 | 23e50d39 | Michael Hanselmann | class ClientRequestWorker(workerpool.BaseWorker): |
69 | b459a848 | Andrea Spadaccini | # pylint: disable=W0221
|
70 | 7e5a6e86 | Guido Trotter | def RunTask(self, server, message, client): |
71 | 23e50d39 | Michael Hanselmann | """Process the request.
|
72 | 23e50d39 | Michael Hanselmann |
|
73 | 23e50d39 | Michael Hanselmann | """
|
74 | 7e5a6e86 | Guido Trotter | client_ops = ClientOps(server) |
75 | 7e5a6e86 | Guido Trotter | |
76 | 23e50d39 | Michael Hanselmann | try:
|
77 | e986f20c | Michael Hanselmann | (method, args, version) = luxi.ParseRequest(message) |
78 | 7e5a6e86 | Guido Trotter | except luxi.ProtocolError, err:
|
79 | 7e5a6e86 | Guido Trotter | logging.error("Protocol Error: %s", err)
|
80 | 7e5a6e86 | Guido Trotter | client.close_log() |
81 | 7e5a6e86 | Guido Trotter | return
|
82 | 7e5a6e86 | Guido Trotter | |
83 | 7e5a6e86 | Guido Trotter | success = False
|
84 | 7e5a6e86 | Guido Trotter | try:
|
85 | e986f20c | Michael Hanselmann | # Verify client's version if there was one in the request
|
86 | e986f20c | Michael Hanselmann | if version is not None and version != constants.LUXI_VERSION: |
87 | e986f20c | Michael Hanselmann | raise errors.LuxiError("LUXI version mismatch, server %s, request %s" % |
88 | e986f20c | Michael Hanselmann | (constants.LUXI_VERSION, version)) |
89 | e986f20c | Michael Hanselmann | |
90 | 7e5a6e86 | Guido Trotter | result = client_ops.handle_request(method, args) |
91 | 7e5a6e86 | Guido Trotter | success = True
|
92 | 7e5a6e86 | Guido Trotter | except errors.GenericError, err:
|
93 | 7e5a6e86 | Guido Trotter | logging.exception("Unexpected exception")
|
94 | 7e5a6e86 | Guido Trotter | success = False
|
95 | 7e5a6e86 | Guido Trotter | result = errors.EncodeException(err) |
96 | 7e5a6e86 | Guido Trotter | except:
|
97 | 7e5a6e86 | Guido Trotter | logging.exception("Unexpected exception")
|
98 | 7e5a6e86 | Guido Trotter | err = sys.exc_info() |
99 | 7e5a6e86 | Guido Trotter | result = "Caught exception: %s" % str(err[1]) |
100 | 7e5a6e86 | Guido Trotter | |
101 | 7e5a6e86 | Guido Trotter | try:
|
102 | 7e5a6e86 | Guido Trotter | reply = luxi.FormatResponse(success, result) |
103 | 7e5a6e86 | Guido Trotter | client.send_message(reply) |
104 | 7e5a6e86 | Guido Trotter | # awake the main thread so that it can write out the data.
|
105 | 7e5a6e86 | Guido Trotter | server.awaker.signal() |
106 | b459a848 | Andrea Spadaccini | except: # pylint: disable=W0702 |
107 | 7e5a6e86 | Guido Trotter | logging.exception("Send error")
|
108 | 7e5a6e86 | Guido Trotter | client.close_log() |
109 | 7e5a6e86 | Guido Trotter | |
110 | 7e5a6e86 | Guido Trotter | |
111 | 7e5a6e86 | Guido Trotter | class MasterClientHandler(daemon.AsyncTerminatedMessageStream): |
112 | 7e5a6e86 | Guido Trotter | """Handler for master peers.
|
113 | 7e5a6e86 | Guido Trotter |
|
114 | 7e5a6e86 | Guido Trotter | """
|
115 | 7e5a6e86 | Guido Trotter | _MAX_UNHANDLED = 1
|
116 | e687ec01 | Michael Hanselmann | |
117 | 7e5a6e86 | Guido Trotter | def __init__(self, server, connected_socket, client_address, family): |
118 | 7e5a6e86 | Guido Trotter | daemon.AsyncTerminatedMessageStream.__init__(self, connected_socket,
|
119 | 7e5a6e86 | Guido Trotter | client_address, |
120 | 7e5a6e86 | Guido Trotter | constants.LUXI_EOM, |
121 | 7e5a6e86 | Guido Trotter | family, self._MAX_UNHANDLED)
|
122 | 7e5a6e86 | Guido Trotter | self.server = server
|
123 | 7e5a6e86 | Guido Trotter | |
124 | 7e5a6e86 | Guido Trotter | def handle_message(self, message, _): |
125 | b2e8a4d9 | Michael Hanselmann | self.server.request_workers.AddTask((self.server, message, self)) |
126 | 23e50d39 | Michael Hanselmann | |
127 | 23e50d39 | Michael Hanselmann | |
128 | 5483fd73 | Michael Hanselmann | class _MasterShutdownCheck: |
129 | 5483fd73 | Michael Hanselmann | """Logic for master daemon shutdown.
|
130 | 5483fd73 | Michael Hanselmann |
|
131 | 5483fd73 | Michael Hanselmann | """
|
132 | 5483fd73 | Michael Hanselmann | #: How long to wait between checks
|
133 | 5483fd73 | Michael Hanselmann | _CHECK_INTERVAL = 5.0
|
134 | 5483fd73 | Michael Hanselmann | |
135 | 5483fd73 | Michael Hanselmann | #: How long to wait after all jobs are done (e.g. to give clients time to
|
136 | 5483fd73 | Michael Hanselmann | #: retrieve the job status)
|
137 | 5483fd73 | Michael Hanselmann | _SHUTDOWN_LINGER = 5.0
|
138 | 5483fd73 | Michael Hanselmann | |
139 | 5483fd73 | Michael Hanselmann | def __init__(self): |
140 | 5483fd73 | Michael Hanselmann | """Initializes this class.
|
141 | 5483fd73 | Michael Hanselmann |
|
142 | 5483fd73 | Michael Hanselmann | """
|
143 | 5483fd73 | Michael Hanselmann | self._had_active_jobs = None |
144 | 5483fd73 | Michael Hanselmann | self._linger_timeout = None |
145 | 5483fd73 | Michael Hanselmann | |
146 | 5483fd73 | Michael Hanselmann | def __call__(self, jq_prepare_result): |
147 | 5483fd73 | Michael Hanselmann | """Determines if master daemon is ready for shutdown.
|
148 | 5483fd73 | Michael Hanselmann |
|
149 | 5483fd73 | Michael Hanselmann | @param jq_prepare_result: Result of L{jqueue.JobQueue.PrepareShutdown}
|
150 | 5483fd73 | Michael Hanselmann | @rtype: None or number
|
151 | 5483fd73 | Michael Hanselmann | @return: None if master daemon is ready, timeout if the check must be
|
152 | 5483fd73 | Michael Hanselmann | repeated
|
153 | 5483fd73 | Michael Hanselmann |
|
154 | 5483fd73 | Michael Hanselmann | """
|
155 | 5483fd73 | Michael Hanselmann | if jq_prepare_result:
|
156 | 5483fd73 | Michael Hanselmann | # Check again shortly
|
157 | 5483fd73 | Michael Hanselmann | logging.info("Job queue has been notified for shutdown but is still"
|
158 | 5483fd73 | Michael Hanselmann | " busy; next check in %s seconds", self._CHECK_INTERVAL) |
159 | 5483fd73 | Michael Hanselmann | self._had_active_jobs = True |
160 | 5483fd73 | Michael Hanselmann | return self._CHECK_INTERVAL |
161 | 5483fd73 | Michael Hanselmann | |
162 | 5483fd73 | Michael Hanselmann | if not self._had_active_jobs: |
163 | 5483fd73 | Michael Hanselmann | # Can shut down as there were no active jobs on the first check
|
164 | 5483fd73 | Michael Hanselmann | return None |
165 | 5483fd73 | Michael Hanselmann | |
166 | 5483fd73 | Michael Hanselmann | # No jobs are running anymore, but maybe some clients want to collect some
|
167 | 5483fd73 | Michael Hanselmann | # information. Give them a short amount of time.
|
168 | 5483fd73 | Michael Hanselmann | if self._linger_timeout is None: |
169 | 5483fd73 | Michael Hanselmann | self._linger_timeout = utils.RunningTimeout(self._SHUTDOWN_LINGER, True) |
170 | 5483fd73 | Michael Hanselmann | |
171 | 5483fd73 | Michael Hanselmann | remaining = self._linger_timeout.Remaining()
|
172 | 5483fd73 | Michael Hanselmann | |
173 | 5483fd73 | Michael Hanselmann | logging.info("Job queue no longer busy; shutting down master daemon"
|
174 | 5483fd73 | Michael Hanselmann | " in %s seconds", remaining)
|
175 | 5483fd73 | Michael Hanselmann | |
176 | 5483fd73 | Michael Hanselmann | # TODO: Should the master daemon socket be closed at this point? Doing so
|
177 | 5483fd73 | Michael Hanselmann | # wouldn't affect existing connections.
|
178 | 5483fd73 | Michael Hanselmann | |
179 | 5483fd73 | Michael Hanselmann | if remaining < 0: |
180 | 5483fd73 | Michael Hanselmann | return None |
181 | 5483fd73 | Michael Hanselmann | else:
|
182 | 5483fd73 | Michael Hanselmann | return remaining
|
183 | 5483fd73 | Michael Hanselmann | |
184 | 5483fd73 | Michael Hanselmann | |
185 | cdd7f900 | Guido Trotter | class MasterServer(daemon.AsyncStreamServer): |
186 | cdd7f900 | Guido Trotter | """Master Server.
|
187 | ffeffa1d | Iustin Pop |
|
188 | cdd7f900 | Guido Trotter | This is the main asynchronous master server. It handles connections to the
|
189 | cdd7f900 | Guido Trotter | master socket.
|
190 | ffeffa1d | Iustin Pop |
|
191 | ffeffa1d | Iustin Pop | """
|
192 | 7e5a6e86 | Guido Trotter | family = socket.AF_UNIX |
193 | 7e5a6e86 | Guido Trotter | |
194 | e8a701f6 | Michael Hanselmann | def __init__(self, address, uid, gid): |
195 | cdd7f900 | Guido Trotter | """MasterServer constructor
|
196 | ce862cd5 | Guido Trotter |
|
197 | cdd7f900 | Guido Trotter | @param address: the unix socket address to bind the MasterServer to
|
198 | bbfd0568 | René Nussbaumer | @param uid: The uid of the owner of the socket
|
199 | bbfd0568 | René Nussbaumer | @param gid: The gid of the owner of the socket
|
200 | ce862cd5 | Guido Trotter |
|
201 | ce862cd5 | Guido Trotter | """
|
202 | bbfd0568 | René Nussbaumer | temp_name = tempfile.mktemp(dir=os.path.dirname(address)) |
203 | 7e5a6e86 | Guido Trotter | daemon.AsyncStreamServer.__init__(self, self.family, temp_name) |
204 | bbfd0568 | René Nussbaumer | os.chmod(temp_name, 0770)
|
205 | bbfd0568 | René Nussbaumer | os.chown(temp_name, uid, gid) |
206 | bbfd0568 | René Nussbaumer | os.rename(temp_name, address) |
207 | bbfd0568 | René Nussbaumer | |
208 | 7e5a6e86 | Guido Trotter | self.awaker = daemon.AsyncAwaker()
|
209 | 50a3fbb2 | Michael Hanselmann | |
210 | 50a3fbb2 | Michael Hanselmann | # We'll only start threads once we've forked.
|
211 | 9113300d | Michael Hanselmann | self.context = None |
212 | 23e50d39 | Michael Hanselmann | self.request_workers = None |
213 | 50a3fbb2 | Michael Hanselmann | |
214 | 5483fd73 | Michael Hanselmann | self._shutdown_check = None |
215 | 5483fd73 | Michael Hanselmann | |
216 | cdd7f900 | Guido Trotter | def handle_connection(self, connected_socket, client_address): |
217 | 7e5a6e86 | Guido Trotter | # TODO: add connection count and limit the number of open connections to a
|
218 | 7e5a6e86 | Guido Trotter | # maximum number to avoid breaking for lack of file descriptors or memory.
|
219 | 7e5a6e86 | Guido Trotter | MasterClientHandler(self, connected_socket, client_address, self.family) |
220 | cdd7f900 | Guido Trotter | |
221 | 50a3fbb2 | Michael Hanselmann | def setup_queue(self): |
222 | 9113300d | Michael Hanselmann | self.context = GanetiContext()
|
223 | 89e2b4d2 | Michael Hanselmann | self.request_workers = workerpool.WorkerPool("ClientReq", |
224 | 89e2b4d2 | Michael Hanselmann | CLIENT_REQUEST_WORKERS, |
225 | 23e50d39 | Michael Hanselmann | ClientRequestWorker) |
226 | ffeffa1d | Iustin Pop | |
227 | 5483fd73 | Michael Hanselmann | def WaitForShutdown(self): |
228 | 5483fd73 | Michael Hanselmann | """Prepares server for shutdown.
|
229 | 5483fd73 | Michael Hanselmann |
|
230 | 5483fd73 | Michael Hanselmann | """
|
231 | 5483fd73 | Michael Hanselmann | if self._shutdown_check is None: |
232 | 5483fd73 | Michael Hanselmann | self._shutdown_check = _MasterShutdownCheck()
|
233 | 5483fd73 | Michael Hanselmann | |
234 | 5483fd73 | Michael Hanselmann | return self._shutdown_check(self.context.jobqueue.PrepareShutdown()) |
235 | 5483fd73 | Michael Hanselmann | |
236 | c1f2901b | Iustin Pop | def server_cleanup(self): |
237 | c1f2901b | Iustin Pop | """Cleanup the server.
|
238 | c1f2901b | Iustin Pop |
|
239 | c1f2901b | Iustin Pop | This involves shutting down the processor threads and the master
|
240 | c1f2901b | Iustin Pop | socket.
|
241 | c1f2901b | Iustin Pop |
|
242 | c1f2901b | Iustin Pop | """
|
243 | 50a3fbb2 | Michael Hanselmann | try:
|
244 | cdd7f900 | Guido Trotter | self.close()
|
245 | 50a3fbb2 | Michael Hanselmann | finally:
|
246 | 23e50d39 | Michael Hanselmann | if self.request_workers: |
247 | 36088c4c | Michael Hanselmann | self.request_workers.TerminateWorkers()
|
248 | 9113300d | Michael Hanselmann | if self.context: |
249 | 9113300d | Michael Hanselmann | self.context.jobqueue.Shutdown()
|
250 | ffeffa1d | Iustin Pop | |
251 | ffeffa1d | Iustin Pop | |
252 | ffeffa1d | Iustin Pop | class ClientOps: |
253 | ffeffa1d | Iustin Pop | """Class holding high-level client operations."""
|
254 | ffeffa1d | Iustin Pop | def __init__(self, server): |
255 | ffeffa1d | Iustin Pop | self.server = server
|
256 | ffeffa1d | Iustin Pop | |
257 | b459a848 | Andrea Spadaccini | def handle_request(self, method, args): # pylint: disable=R0911 |
258 | 9113300d | Michael Hanselmann | queue = self.server.context.jobqueue
|
259 | 0bbe448c | Michael Hanselmann | |
260 | 0bbe448c | Michael Hanselmann | # TODO: Parameter validation
|
261 | a629ecb9 | Iustin Pop | if not isinstance(args, (tuple, list)): |
262 | a629ecb9 | Iustin Pop | logging.info("Received invalid arguments of type '%s'", type(args)) |
263 | a629ecb9 | Iustin Pop | raise ValueError("Invalid arguments type '%s'" % type(args)) |
264 | 0bbe448c | Michael Hanselmann | |
265 | 7260cfbe | Iustin Pop | # TODO: Rewrite to not exit in each 'if/elif' branch
|
266 | 7260cfbe | Iustin Pop | |
267 | 0bbe448c | Michael Hanselmann | if method == luxi.REQ_SUBMIT_JOB:
|
268 | e566ddbd | Iustin Pop | logging.info("Received new job")
|
269 | 0bbe448c | Michael Hanselmann | ops = [opcodes.OpCode.LoadOpCode(state) for state in args] |
270 | 4c848b18 | Michael Hanselmann | return queue.SubmitJob(ops)
|
271 | ffeffa1d | Iustin Pop | |
272 | 2971c913 | Iustin Pop | if method == luxi.REQ_SUBMIT_MANY_JOBS:
|
273 | 2971c913 | Iustin Pop | logging.info("Received multiple jobs")
|
274 | 2971c913 | Iustin Pop | jobs = [] |
275 | 2971c913 | Iustin Pop | for ops in args: |
276 | 2971c913 | Iustin Pop | jobs.append([opcodes.OpCode.LoadOpCode(state) for state in ops]) |
277 | 2971c913 | Iustin Pop | return queue.SubmitManyJobs(jobs)
|
278 | 2971c913 | Iustin Pop | |
279 | 0bbe448c | Michael Hanselmann | elif method == luxi.REQ_CANCEL_JOB:
|
280 | a629ecb9 | Iustin Pop | (job_id, ) = args |
281 | e566ddbd | Iustin Pop | logging.info("Received job cancel request for %s", job_id)
|
282 | 0bbe448c | Michael Hanselmann | return queue.CancelJob(job_id)
|
283 | ffeffa1d | Iustin Pop | |
284 | 0bbe448c | Michael Hanselmann | elif method == luxi.REQ_ARCHIVE_JOB:
|
285 | a629ecb9 | Iustin Pop | (job_id, ) = args |
286 | e566ddbd | Iustin Pop | logging.info("Received job archive request for %s", job_id)
|
287 | 0bbe448c | Michael Hanselmann | return queue.ArchiveJob(job_id)
|
288 | 0bbe448c | Michael Hanselmann | |
289 | 07cd723a | Iustin Pop | elif method == luxi.REQ_AUTOARCHIVE_JOBS:
|
290 | f8ad5591 | Michael Hanselmann | (age, timeout) = args |
291 | e566ddbd | Iustin Pop | logging.info("Received job autoarchive request for age %s, timeout %s",
|
292 | e566ddbd | Iustin Pop | age, timeout) |
293 | f8ad5591 | Michael Hanselmann | return queue.AutoArchiveJobs(age, timeout)
|
294 | 07cd723a | Iustin Pop | |
295 | dfe57c22 | Michael Hanselmann | elif method == luxi.REQ_WAIT_FOR_JOB_CHANGE:
|
296 | 5c735209 | Iustin Pop | (job_id, fields, prev_job_info, prev_log_serial, timeout) = args |
297 | e566ddbd | Iustin Pop | logging.info("Received job poll request for %s", job_id)
|
298 | 6c5a7090 | Michael Hanselmann | return queue.WaitForJobChanges(job_id, fields, prev_job_info,
|
299 | 5c735209 | Iustin Pop | prev_log_serial, timeout) |
300 | dfe57c22 | Michael Hanselmann | |
301 | 28b71a76 | Michael Hanselmann | elif method == luxi.REQ_QUERY:
|
302 | a629ecb9 | Iustin Pop | (what, fields, qfilter) = args |
303 | a629ecb9 | Iustin Pop | req = objects.QueryRequest(what=what, fields=fields, qfilter=qfilter) |
304 | 28b71a76 | Michael Hanselmann | |
305 | abd66bf8 | Michael Hanselmann | if req.what in constants.QR_VIA_OP: |
306 | 28b71a76 | Michael Hanselmann | result = self._Query(opcodes.OpQuery(what=req.what, fields=req.fields,
|
307 | 2e5c33db | Iustin Pop | qfilter=req.qfilter)) |
308 | 24d16f76 | Michael Hanselmann | elif req.what == constants.QR_LOCK:
|
309 | 2e5c33db | Iustin Pop | if req.qfilter is not None: |
310 | 24d16f76 | Michael Hanselmann | raise errors.OpPrereqError("Lock queries can't be filtered") |
311 | 24d16f76 | Michael Hanselmann | return self.server.context.glm.QueryLocks(req.fields) |
312 | abd66bf8 | Michael Hanselmann | elif req.what in constants.QR_VIA_LUXI: |
313 | 28b71a76 | Michael Hanselmann | raise NotImplementedError |
314 | 28b71a76 | Michael Hanselmann | else:
|
315 | 28b71a76 | Michael Hanselmann | raise errors.OpPrereqError("Resource type '%s' unknown" % req.what, |
316 | 28b71a76 | Michael Hanselmann | errors.ECODE_INVAL) |
317 | 28b71a76 | Michael Hanselmann | |
318 | 28b71a76 | Michael Hanselmann | return result
|
319 | 28b71a76 | Michael Hanselmann | |
320 | 28b71a76 | Michael Hanselmann | elif method == luxi.REQ_QUERY_FIELDS:
|
321 | a629ecb9 | Iustin Pop | (what, fields) = args |
322 | a629ecb9 | Iustin Pop | req = objects.QueryFieldsRequest(what=what, fields=fields) |
323 | 28b71a76 | Michael Hanselmann | |
324 | c1391810 | Michael Hanselmann | try:
|
325 | c1391810 | Michael Hanselmann | fielddefs = query.ALL_FIELDS[req.what] |
326 | c1391810 | Michael Hanselmann | except KeyError: |
327 | 28b71a76 | Michael Hanselmann | raise errors.OpPrereqError("Resource type '%s' unknown" % req.what, |
328 | 28b71a76 | Michael Hanselmann | errors.ECODE_INVAL) |
329 | 28b71a76 | Michael Hanselmann | |
330 | c1391810 | Michael Hanselmann | return query.QueryFields(fielddefs, req.fields)
|
331 | 28b71a76 | Michael Hanselmann | |
332 | 0bbe448c | Michael Hanselmann | elif method == luxi.REQ_QUERY_JOBS:
|
333 | 0bbe448c | Michael Hanselmann | (job_ids, fields) = args |
334 | e566ddbd | Iustin Pop | if isinstance(job_ids, (tuple, list)) and job_ids: |
335 | 1f864b60 | Iustin Pop | msg = utils.CommaJoin(job_ids) |
336 | e566ddbd | Iustin Pop | else:
|
337 | e566ddbd | Iustin Pop | msg = str(job_ids)
|
338 | e566ddbd | Iustin Pop | logging.info("Received job query request for %s", msg)
|
339 | 0bbe448c | Michael Hanselmann | return queue.QueryJobs(job_ids, fields)
|
340 | 0bbe448c | Michael Hanselmann | |
341 | ee6c7b94 | Michael Hanselmann | elif method == luxi.REQ_QUERY_INSTANCES:
|
342 | ec79568d | Iustin Pop | (names, fields, use_locking) = args |
343 | e566ddbd | Iustin Pop | logging.info("Received instance query request for %s", names)
|
344 | 77921a95 | Iustin Pop | if use_locking:
|
345 | debac808 | Iustin Pop | raise errors.OpPrereqError("Sync queries are not allowed", |
346 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
347 | f2af0bec | Iustin Pop | op = opcodes.OpInstanceQuery(names=names, output_fields=fields, |
348 | f2af0bec | Iustin Pop | use_locking=use_locking) |
349 | ee6c7b94 | Michael Hanselmann | return self._Query(op) |
350 | ee6c7b94 | Michael Hanselmann | |
351 | 02f7fe54 | Michael Hanselmann | elif method == luxi.REQ_QUERY_NODES:
|
352 | ec79568d | Iustin Pop | (names, fields, use_locking) = args |
353 | e566ddbd | Iustin Pop | logging.info("Received node query request for %s", names)
|
354 | 77921a95 | Iustin Pop | if use_locking:
|
355 | debac808 | Iustin Pop | raise errors.OpPrereqError("Sync queries are not allowed", |
356 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
357 | 2237687b | Iustin Pop | op = opcodes.OpNodeQuery(names=names, output_fields=fields, |
358 | 2237687b | Iustin Pop | use_locking=use_locking) |
359 | 02f7fe54 | Michael Hanselmann | return self._Query(op) |
360 | 02f7fe54 | Michael Hanselmann | |
361 | a79ef2a5 | Adeodato Simo | elif method == luxi.REQ_QUERY_GROUPS:
|
362 | a79ef2a5 | Adeodato Simo | (names, fields, use_locking) = args |
363 | a79ef2a5 | Adeodato Simo | logging.info("Received group query request for %s", names)
|
364 | a79ef2a5 | Adeodato Simo | if use_locking:
|
365 | a79ef2a5 | Adeodato Simo | raise errors.OpPrereqError("Sync queries are not allowed", |
366 | a79ef2a5 | Adeodato Simo | errors.ECODE_INVAL) |
367 | d4d654bd | Iustin Pop | op = opcodes.OpGroupQuery(names=names, output_fields=fields) |
368 | a79ef2a5 | Adeodato Simo | return self._Query(op) |
369 | a79ef2a5 | Adeodato Simo | |
370 | 32f93223 | Michael Hanselmann | elif method == luxi.REQ_QUERY_EXPORTS:
|
371 | a629ecb9 | Iustin Pop | (nodes, use_locking) = args |
372 | 77921a95 | Iustin Pop | if use_locking:
|
373 | debac808 | Iustin Pop | raise errors.OpPrereqError("Sync queries are not allowed", |
374 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
375 | e566ddbd | Iustin Pop | logging.info("Received exports query request")
|
376 | 7ca2d4d8 | Iustin Pop | op = opcodes.OpBackupQuery(nodes=nodes, use_locking=use_locking) |
377 | 32f93223 | Michael Hanselmann | return self._Query(op) |
378 | 32f93223 | Michael Hanselmann | |
379 | ae5849b5 | Michael Hanselmann | elif method == luxi.REQ_QUERY_CONFIG_VALUES:
|
380 | a629ecb9 | Iustin Pop | (fields, ) = args |
381 | e566ddbd | Iustin Pop | logging.info("Received config values query request for %s", fields)
|
382 | 2f093ea0 | Iustin Pop | op = opcodes.OpClusterConfigQuery(output_fields=fields) |
383 | ae5849b5 | Michael Hanselmann | return self._Query(op) |
384 | ae5849b5 | Michael Hanselmann | |
385 | 66baeccc | Iustin Pop | elif method == luxi.REQ_QUERY_CLUSTER_INFO:
|
386 | e566ddbd | Iustin Pop | logging.info("Received cluster info query request")
|
387 | a2f7ab92 | Iustin Pop | op = opcodes.OpClusterQuery() |
388 | 66baeccc | Iustin Pop | return self._Query(op) |
389 | 66baeccc | Iustin Pop | |
390 | 7699c3af | Iustin Pop | elif method == luxi.REQ_QUERY_TAGS:
|
391 | a629ecb9 | Iustin Pop | (kind, name) = args |
392 | 7699c3af | Iustin Pop | logging.info("Received tags query request")
|
393 | c6afb1ca | Iustin Pop | op = opcodes.OpTagsGet(kind=kind, name=name) |
394 | 7699c3af | Iustin Pop | return self._Query(op) |
395 | 7699c3af | Iustin Pop | |
396 | 19b9ba9a | Michael Hanselmann | elif method == luxi.REQ_QUERY_LOCKS:
|
397 | 19b9ba9a | Michael Hanselmann | (fields, sync) = args |
398 | 19b9ba9a | Michael Hanselmann | logging.info("Received locks query request")
|
399 | 24d16f76 | Michael Hanselmann | if sync:
|
400 | 24d16f76 | Michael Hanselmann | raise NotImplementedError("Synchronous queries are not implemented") |
401 | 24d16f76 | Michael Hanselmann | return self.server.context.glm.OldStyleQueryLocks(fields) |
402 | 19b9ba9a | Michael Hanselmann | |
403 | 3ccafd0e | Iustin Pop | elif method == luxi.REQ_QUEUE_SET_DRAIN_FLAG:
|
404 | a629ecb9 | Iustin Pop | (drain_flag, ) = args |
405 | e566ddbd | Iustin Pop | logging.info("Received queue drain flag change request to %s",
|
406 | e566ddbd | Iustin Pop | drain_flag) |
407 | 3ccafd0e | Iustin Pop | return queue.SetDrainFlag(drain_flag)
|
408 | 3ccafd0e | Iustin Pop | |
409 | 05e50653 | Michael Hanselmann | elif method == luxi.REQ_SET_WATCHER_PAUSE:
|
410 | 05e50653 | Michael Hanselmann | (until, ) = args |
411 | 05e50653 | Michael Hanselmann | |
412 | 05e50653 | Michael Hanselmann | if until is None: |
413 | 05e50653 | Michael Hanselmann | logging.info("Received request to no longer pause the watcher")
|
414 | 05e50653 | Michael Hanselmann | else:
|
415 | 05e50653 | Michael Hanselmann | if not isinstance(until, (int, float)): |
416 | 05e50653 | Michael Hanselmann | raise TypeError("Duration must be an integer or float") |
417 | 05e50653 | Michael Hanselmann | |
418 | 05e50653 | Michael Hanselmann | if until < time.time():
|
419 | 05e50653 | Michael Hanselmann | raise errors.GenericError("Unable to set pause end time in the past") |
420 | 05e50653 | Michael Hanselmann | |
421 | 05e50653 | Michael Hanselmann | logging.info("Received request to pause the watcher until %s", until)
|
422 | 05e50653 | Michael Hanselmann | |
423 | 05e50653 | Michael Hanselmann | return _SetWatcherPause(until)
|
424 | 05e50653 | Michael Hanselmann | |
425 | 0bbe448c | Michael Hanselmann | else:
|
426 | e566ddbd | Iustin Pop | logging.info("Received invalid request '%s'", method)
|
427 | e566ddbd | Iustin Pop | raise ValueError("Invalid operation '%s'" % method) |
428 | ffeffa1d | Iustin Pop | |
429 | ee6c7b94 | Michael Hanselmann | def _Query(self, op): |
430 | ee6c7b94 | Michael Hanselmann | """Runs the specified opcode and returns the result.
|
431 | ee6c7b94 | Michael Hanselmann |
|
432 | ee6c7b94 | Michael Hanselmann | """
|
433 | adfa97e3 | Guido Trotter | # Queries don't have a job id
|
434 | adfa97e3 | Guido Trotter | proc = mcpu.Processor(self.server.context, None) |
435 | 26d3fd2f | Michael Hanselmann | |
436 | 26d3fd2f | Michael Hanselmann | # TODO: Executing an opcode using locks will acquire them in blocking mode.
|
437 | 26d3fd2f | Michael Hanselmann | # Consider using a timeout for retries.
|
438 | 031a3e57 | Michael Hanselmann | return proc.ExecOpCode(op, None) |
439 | ee6c7b94 | Michael Hanselmann | |
440 | ffeffa1d | Iustin Pop | |
441 | 39dcf2ef | Guido Trotter | class GanetiContext(object): |
442 | 39dcf2ef | Guido Trotter | """Context common to all ganeti threads.
|
443 | 39dcf2ef | Guido Trotter |
|
444 | 39dcf2ef | Guido Trotter | This class creates and holds common objects shared by all threads.
|
445 | 39dcf2ef | Guido Trotter |
|
446 | 39dcf2ef | Guido Trotter | """
|
447 | b459a848 | Andrea Spadaccini | # pylint: disable=W0212
|
448 | 7260cfbe | Iustin Pop | # we do want to ensure a singleton here
|
449 | 39dcf2ef | Guido Trotter | _instance = None
|
450 | 39dcf2ef | Guido Trotter | |
451 | 39dcf2ef | Guido Trotter | def __init__(self): |
452 | 39dcf2ef | Guido Trotter | """Constructs a new GanetiContext object.
|
453 | 39dcf2ef | Guido Trotter |
|
454 | 39dcf2ef | Guido Trotter | There should be only a GanetiContext object at any time, so this
|
455 | 39dcf2ef | Guido Trotter | function raises an error if this is not the case.
|
456 | 39dcf2ef | Guido Trotter |
|
457 | 39dcf2ef | Guido Trotter | """
|
458 | 39dcf2ef | Guido Trotter | assert self.__class__._instance is None, "double GanetiContext instance" |
459 | 39dcf2ef | Guido Trotter | |
460 | 9113300d | Michael Hanselmann | # Create global configuration object
|
461 | 39dcf2ef | Guido Trotter | self.cfg = config.ConfigWriter()
|
462 | 9113300d | Michael Hanselmann | |
463 | 9113300d | Michael Hanselmann | # Locking manager
|
464 | 984f7c32 | Guido Trotter | self.glm = locking.GanetiLockManager(
|
465 | 39dcf2ef | Guido Trotter | self.cfg.GetNodeList(),
|
466 | 819ca990 | Guido Trotter | self.cfg.GetNodeGroupList(),
|
467 | 39dcf2ef | Guido Trotter | self.cfg.GetInstanceList())
|
468 | 39dcf2ef | Guido Trotter | |
469 | b2acdbdc | Michael Hanselmann | self.cfg.SetContext(self) |
470 | b2acdbdc | Michael Hanselmann | |
471 | 87b3cb26 | Michael Hanselmann | # RPC runner
|
472 | 87b3cb26 | Michael Hanselmann | self.rpc = rpc.RpcRunner(self) |
473 | 87b3cb26 | Michael Hanselmann | |
474 | cb4d3314 | Michael Hanselmann | # Job queue
|
475 | cb4d3314 | Michael Hanselmann | self.jobqueue = jqueue.JobQueue(self) |
476 | cb4d3314 | Michael Hanselmann | |
477 | 39dcf2ef | Guido Trotter | # setting this also locks the class against attribute modifications
|
478 | 39dcf2ef | Guido Trotter | self.__class__._instance = self |
479 | 39dcf2ef | Guido Trotter | |
480 | 39dcf2ef | Guido Trotter | def __setattr__(self, name, value): |
481 | 39dcf2ef | Guido Trotter | """Setting GanetiContext attributes is forbidden after initialization.
|
482 | 39dcf2ef | Guido Trotter |
|
483 | 39dcf2ef | Guido Trotter | """
|
484 | 39dcf2ef | Guido Trotter | assert self.__class__._instance is None, "Attempt to modify Ganeti Context" |
485 | 39dcf2ef | Guido Trotter | object.__setattr__(self, name, value) |
486 | 39dcf2ef | Guido Trotter | |
487 | 0debfb35 | Guido Trotter | def AddNode(self, node, ec_id): |
488 | d8470559 | Michael Hanselmann | """Adds a node to the configuration and lock manager.
|
489 | d8470559 | Michael Hanselmann |
|
490 | d8470559 | Michael Hanselmann | """
|
491 | d8470559 | Michael Hanselmann | # Add it to the configuration
|
492 | 0debfb35 | Guido Trotter | self.cfg.AddNode(node, ec_id)
|
493 | d8470559 | Michael Hanselmann | |
494 | c36176cc | Michael Hanselmann | # If preseeding fails it'll not be added
|
495 | 99aabbed | Iustin Pop | self.jobqueue.AddNode(node)
|
496 | c36176cc | Michael Hanselmann | |
497 | d8470559 | Michael Hanselmann | # Add the new node to the Ganeti Lock Manager
|
498 | d8470559 | Michael Hanselmann | self.glm.add(locking.LEVEL_NODE, node.name)
|
499 | 4e070776 | Michael Hanselmann | self.glm.add(locking.LEVEL_NODE_RES, node.name)
|
500 | d8470559 | Michael Hanselmann | |
501 | d8470559 | Michael Hanselmann | def ReaddNode(self, node): |
502 | d8470559 | Michael Hanselmann | """Updates a node that's already in the configuration
|
503 | d8470559 | Michael Hanselmann |
|
504 | d8470559 | Michael Hanselmann | """
|
505 | c36176cc | Michael Hanselmann | # Synchronize the queue again
|
506 | 99aabbed | Iustin Pop | self.jobqueue.AddNode(node)
|
507 | d8470559 | Michael Hanselmann | |
508 | d8470559 | Michael Hanselmann | def RemoveNode(self, name): |
509 | d8470559 | Michael Hanselmann | """Removes a node from the configuration and lock manager.
|
510 | d8470559 | Michael Hanselmann |
|
511 | d8470559 | Michael Hanselmann | """
|
512 | d8470559 | Michael Hanselmann | # Remove node from configuration
|
513 | d8470559 | Michael Hanselmann | self.cfg.RemoveNode(name)
|
514 | d8470559 | Michael Hanselmann | |
515 | c36176cc | Michael Hanselmann | # Notify job queue
|
516 | c36176cc | Michael Hanselmann | self.jobqueue.RemoveNode(name)
|
517 | c36176cc | Michael Hanselmann | |
518 | d8470559 | Michael Hanselmann | # Remove the node from the Ganeti Lock Manager
|
519 | d8470559 | Michael Hanselmann | self.glm.remove(locking.LEVEL_NODE, name)
|
520 | 4e070776 | Michael Hanselmann | self.glm.remove(locking.LEVEL_NODE_RES, name)
|
521 | d8470559 | Michael Hanselmann | |
522 | 39dcf2ef | Guido Trotter | |
523 | 05e50653 | Michael Hanselmann | def _SetWatcherPause(until): |
524 | 05e50653 | Michael Hanselmann | """Creates or removes the watcher pause file.
|
525 | 05e50653 | Michael Hanselmann |
|
526 | 05e50653 | Michael Hanselmann | @type until: None or int
|
527 | 05e50653 | Michael Hanselmann | @param until: Unix timestamp saying until when the watcher shouldn't run
|
528 | 05e50653 | Michael Hanselmann |
|
529 | 05e50653 | Michael Hanselmann | """
|
530 | 05e50653 | Michael Hanselmann | if until is None: |
531 | 05e50653 | Michael Hanselmann | utils.RemoveFile(constants.WATCHER_PAUSEFILE) |
532 | 05e50653 | Michael Hanselmann | else:
|
533 | 05e50653 | Michael Hanselmann | utils.WriteFile(constants.WATCHER_PAUSEFILE, |
534 | 05e50653 | Michael Hanselmann | data="%d\n" % (until, ))
|
535 | 05e50653 | Michael Hanselmann | |
536 | 28b498cd | Michael Hanselmann | return until
|
537 | 28b498cd | Michael Hanselmann | |
538 | 05e50653 | Michael Hanselmann | |
539 | e0e916fe | Iustin Pop | @rpc.RunWithRPC
|
540 | 36205981 | Iustin Pop | def CheckAgreement(): |
541 | 36205981 | Iustin Pop | """Check the agreement on who is the master.
|
542 | 36205981 | Iustin Pop |
|
543 | 36205981 | Iustin Pop | The function uses a very simple algorithm: we must get more positive
|
544 | 36205981 | Iustin Pop | than negative answers. Since in most of the cases we are the master,
|
545 | 36205981 | Iustin Pop | we'll use our own config file for getting the node list. In the
|
546 | 36205981 | Iustin Pop | future we could collect the current node list from our (possibly
|
547 | 36205981 | Iustin Pop | obsolete) known nodes.
|
548 | 36205981 | Iustin Pop |
|
549 | d7cdb55d | Iustin Pop | In order to account for cold-start of all nodes, we retry for up to
|
550 | d7cdb55d | Iustin Pop | a minute until we get a real answer as the top-voted one. If the
|
551 | d7cdb55d | Iustin Pop | nodes are more out-of-sync, for now manual startup of the master
|
552 | d7cdb55d | Iustin Pop | should be attempted.
|
553 | d7cdb55d | Iustin Pop |
|
554 | d7cdb55d | Iustin Pop | Note that for a even number of nodes cluster, we need at least half
|
555 | d7cdb55d | Iustin Pop | of the nodes (beside ourselves) to vote for us. This creates a
|
556 | d7cdb55d | Iustin Pop | problem on two-node clusters, since in this case we require the
|
557 | d7cdb55d | Iustin Pop | other node to be up too to confirm our status.
|
558 | d7cdb55d | Iustin Pop |
|
559 | 36205981 | Iustin Pop | """
|
560 | b705c7a6 | Manuel Franceschini | myself = netutils.Hostname.GetSysName() |
561 | 36205981 | Iustin Pop | #temp instantiation of a config writer, used only to get the node list
|
562 | 36205981 | Iustin Pop | cfg = config.ConfigWriter() |
563 | 36205981 | Iustin Pop | node_list = cfg.GetNodeList() |
564 | 36205981 | Iustin Pop | del cfg
|
565 | d7cdb55d | Iustin Pop | retries = 6
|
566 | d7cdb55d | Iustin Pop | while retries > 0: |
567 | d7cdb55d | Iustin Pop | votes = bootstrap.GatherMasterVotes(node_list) |
568 | d7cdb55d | Iustin Pop | if not votes: |
569 | d7cdb55d | Iustin Pop | # empty node list, this is a one node cluster
|
570 | d7cdb55d | Iustin Pop | return True |
571 | d7cdb55d | Iustin Pop | if votes[0][0] is None: |
572 | d7cdb55d | Iustin Pop | retries -= 1
|
573 | d7cdb55d | Iustin Pop | time.sleep(10)
|
574 | 36205981 | Iustin Pop | continue
|
575 | d7cdb55d | Iustin Pop | break
|
576 | d7cdb55d | Iustin Pop | if retries == 0: |
577 | e09fdcfa | Iustin Pop | logging.critical("Cluster inconsistent, most of the nodes didn't answer"
|
578 | e09fdcfa | Iustin Pop | " after multiple retries. Aborting startup")
|
579 | d8f5a37d | Iustin Pop | logging.critical("Use the --no-voting option if you understand what"
|
580 | d8f5a37d | Iustin Pop | " effects it has on the cluster state")
|
581 | e09fdcfa | Iustin Pop | return False |
582 | d7cdb55d | Iustin Pop | # here a real node is at the top of the list
|
583 | d7cdb55d | Iustin Pop | all_votes = sum(item[1] for item in votes) |
584 | d7cdb55d | Iustin Pop | top_node, top_votes = votes[0]
|
585 | 8a20c732 | Michael Hanselmann | |
586 | d7cdb55d | Iustin Pop | result = False
|
587 | d7cdb55d | Iustin Pop | if top_node != myself:
|
588 | d7cdb55d | Iustin Pop | logging.critical("It seems we are not the master (top-voted node"
|
589 | bbe19c17 | Iustin Pop | " is %s with %d out of %d votes)", top_node, top_votes,
|
590 | bbe19c17 | Iustin Pop | all_votes) |
591 | d7cdb55d | Iustin Pop | elif top_votes < all_votes - top_votes:
|
592 | 36205981 | Iustin Pop | logging.critical("It seems we are not the master (%d votes for,"
|
593 | d7cdb55d | Iustin Pop | " %d votes against)", top_votes, all_votes - top_votes)
|
594 | d7cdb55d | Iustin Pop | else:
|
595 | d7cdb55d | Iustin Pop | result = True
|
596 | d7cdb55d | Iustin Pop | |
597 | d7cdb55d | Iustin Pop | return result
|
598 | 36205981 | Iustin Pop | |
599 | 6c948699 | Michael Hanselmann | |
600 | 340f4757 | Iustin Pop | @rpc.RunWithRPC
|
601 | 340f4757 | Iustin Pop | def ActivateMasterIP(): |
602 | 340f4757 | Iustin Pop | # activate ip
|
603 | 8da2bd43 | Andrea Spadaccini | cfg = config.ConfigWriter() |
604 | f9d20654 | Andrea Spadaccini | master_params = cfg.GetMasterNetworkParameters() |
605 | 57c7bc57 | Andrea Spadaccini | ems = cfg.GetUseExternalMipScript() |
606 | 8da2bd43 | Andrea Spadaccini | runner = rpc.BootstrapRunner() |
607 | f9d20654 | Andrea Spadaccini | result = runner.call_node_activate_master_ip(master_params.name, |
608 | 57c7bc57 | Andrea Spadaccini | master_params, ems) |
609 | 8da2bd43 | Andrea Spadaccini | |
610 | 340f4757 | Iustin Pop | msg = result.fail_msg |
611 | 340f4757 | Iustin Pop | if msg:
|
612 | 340f4757 | Iustin Pop | logging.error("Can't activate master IP address: %s", msg)
|
613 | 340f4757 | Iustin Pop | |
614 | 340f4757 | Iustin Pop | |
615 | ed0efaa5 | Michael Hanselmann | def CheckMasterd(options, args): |
616 | ed0efaa5 | Michael Hanselmann | """Initial checks whether to run or exit with a failure.
|
617 | ed0efaa5 | Michael Hanselmann |
|
618 | ed0efaa5 | Michael Hanselmann | """
|
619 | f93427cd | Iustin Pop | if args: # masterd doesn't take any arguments |
620 | f93427cd | Iustin Pop | print >> sys.stderr, ("Usage: %s [-f] [-d]" % sys.argv[0]) |
621 | f93427cd | Iustin Pop | sys.exit(constants.EXIT_FAILURE) |
622 | f93427cd | Iustin Pop | |
623 | ed0efaa5 | Michael Hanselmann | ssconf.CheckMaster(options.debug) |
624 | ed0efaa5 | Michael Hanselmann | |
625 | bbfd0568 | René Nussbaumer | try:
|
626 | bbfd0568 | René Nussbaumer | options.uid = pwd.getpwnam(constants.MASTERD_USER).pw_uid |
627 | bbfd0568 | René Nussbaumer | options.gid = grp.getgrnam(constants.DAEMONS_GROUP).gr_gid |
628 | bbfd0568 | René Nussbaumer | except KeyError: |
629 | bbfd0568 | René Nussbaumer | print >> sys.stderr, ("User or group not existing on system: %s:%s" % |
630 | bbfd0568 | René Nussbaumer | (constants.MASTERD_USER, constants.DAEMONS_GROUP)) |
631 | bbfd0568 | René Nussbaumer | sys.exit(constants.EXIT_FAILURE) |
632 | bbfd0568 | René Nussbaumer | |
633 | 4b63dc7a | Iustin Pop | # Check the configuration is sane before anything else
|
634 | 4b63dc7a | Iustin Pop | try:
|
635 | 4b63dc7a | Iustin Pop | config.ConfigWriter() |
636 | 4b63dc7a | Iustin Pop | except errors.ConfigVersionMismatch, err:
|
637 | 4b63dc7a | Iustin Pop | v1 = "%s.%s.%s" % constants.SplitVersion(err.args[0]) |
638 | 4b63dc7a | Iustin Pop | v2 = "%s.%s.%s" % constants.SplitVersion(err.args[1]) |
639 | 4b63dc7a | Iustin Pop | print >> sys.stderr, \
|
640 | 4b63dc7a | Iustin Pop | ("Configuration version mismatch. The current Ganeti software"
|
641 | 4b63dc7a | Iustin Pop | " expects version %s, but the on-disk configuration file has"
|
642 | 4b63dc7a | Iustin Pop | " version %s. This is likely the result of upgrading the"
|
643 | 4b63dc7a | Iustin Pop | " software without running the upgrade procedure. Please contact"
|
644 | 4b63dc7a | Iustin Pop | " your cluster administrator or complete the upgrade using the"
|
645 | 4b63dc7a | Iustin Pop | " cfgupgrade utility, after reading the upgrade notes." %
|
646 | 4b63dc7a | Iustin Pop | (v1, v2)) |
647 | 4b63dc7a | Iustin Pop | sys.exit(constants.EXIT_FAILURE) |
648 | 4b63dc7a | Iustin Pop | except errors.ConfigurationError, err:
|
649 | 4b63dc7a | Iustin Pop | print >> sys.stderr, \
|
650 | 4b63dc7a | Iustin Pop | ("Configuration error while opening the configuration file: %s\n"
|
651 | 4b63dc7a | Iustin Pop | "This might be caused by an incomplete software upgrade or"
|
652 | 4b63dc7a | Iustin Pop | " by a corrupted configuration file. Until the problem is fixed"
|
653 | 4b63dc7a | Iustin Pop | " the master daemon cannot start." % str(err)) |
654 | 4b63dc7a | Iustin Pop | sys.exit(constants.EXIT_FAILURE) |
655 | bbfd0568 | René Nussbaumer | |
656 | ed0efaa5 | Michael Hanselmann | # If CheckMaster didn't fail we believe we are the master, but we have to
|
657 | ed0efaa5 | Michael Hanselmann | # confirm with the other nodes.
|
658 | ed0efaa5 | Michael Hanselmann | if options.no_voting:
|
659 | 675e2bf5 | Iustin Pop | if not options.yes_do_it: |
660 | 675e2bf5 | Iustin Pop | sys.stdout.write("The 'no voting' option has been selected.\n")
|
661 | 675e2bf5 | Iustin Pop | sys.stdout.write("This is dangerous, please confirm by"
|
662 | 675e2bf5 | Iustin Pop | " typing uppercase 'yes': ")
|
663 | 675e2bf5 | Iustin Pop | sys.stdout.flush() |
664 | ed0efaa5 | Michael Hanselmann | |
665 | 675e2bf5 | Iustin Pop | confirmation = sys.stdin.readline().strip() |
666 | 675e2bf5 | Iustin Pop | if confirmation != "YES": |
667 | 675e2bf5 | Iustin Pop | print >> sys.stderr, "Aborting." |
668 | 675e2bf5 | Iustin Pop | sys.exit(constants.EXIT_FAILURE) |
669 | ed0efaa5 | Michael Hanselmann | |
670 | 675e2bf5 | Iustin Pop | else:
|
671 | 675e2bf5 | Iustin Pop | # CheckAgreement uses RPC and threads, hence it needs to be run in
|
672 | 675e2bf5 | Iustin Pop | # a separate process before we call utils.Daemonize in the current
|
673 | 675e2bf5 | Iustin Pop | # process.
|
674 | 675e2bf5 | Iustin Pop | if not utils.RunInSeparateProcess(CheckAgreement): |
675 | ed0efaa5 | Michael Hanselmann | sys.exit(constants.EXIT_FAILURE) |
676 | ed0efaa5 | Michael Hanselmann | |
677 | 340f4757 | Iustin Pop | # ActivateMasterIP also uses RPC/threads, so we run it again via a
|
678 | 340f4757 | Iustin Pop | # separate process.
|
679 | 340f4757 | Iustin Pop | |
680 | 340f4757 | Iustin Pop | # TODO: decide whether failure to activate the master IP is a fatal error
|
681 | 340f4757 | Iustin Pop | utils.RunInSeparateProcess(ActivateMasterIP) |
682 | 340f4757 | Iustin Pop | |
683 | ed0efaa5 | Michael Hanselmann | |
684 | 3ee53f1f | Iustin Pop | def PrepMasterd(options, _): |
685 | 3ee53f1f | Iustin Pop | """Prep master daemon function, executed with the PID file held.
|
686 | 3b316acb | Iustin Pop |
|
687 | 04ccf5e9 | Guido Trotter | """
|
688 | 04ccf5e9 | Guido Trotter | # This is safe to do as the pid file guarantees against
|
689 | 04ccf5e9 | Guido Trotter | # concurrent execution.
|
690 | 04ccf5e9 | Guido Trotter | utils.RemoveFile(constants.MASTER_SOCKET) |
691 | b1b6ea87 | Iustin Pop | |
692 | cdd7f900 | Guido Trotter | mainloop = daemon.Mainloop() |
693 | e8a701f6 | Michael Hanselmann | master = MasterServer(constants.MASTER_SOCKET, options.uid, options.gid) |
694 | 3ee53f1f | Iustin Pop | return (mainloop, master)
|
695 | 3ee53f1f | Iustin Pop | |
696 | 3ee53f1f | Iustin Pop | |
697 | b459a848 | Andrea Spadaccini | def ExecMasterd(options, args, prep_data): # pylint: disable=W0613 |
698 | 3ee53f1f | Iustin Pop | """Main master daemon function, executed with the PID file held.
|
699 | 3ee53f1f | Iustin Pop |
|
700 | 3ee53f1f | Iustin Pop | """
|
701 | 3ee53f1f | Iustin Pop | (mainloop, master) = prep_data |
702 | 04ccf5e9 | Guido Trotter | try:
|
703 | 15486fa7 | Michael Hanselmann | rpc.Init() |
704 | 4331f6cd | Michael Hanselmann | try:
|
705 | 15486fa7 | Michael Hanselmann | master.setup_queue() |
706 | 15486fa7 | Michael Hanselmann | try:
|
707 | 5483fd73 | Michael Hanselmann | mainloop.Run(shutdown_wait_fn=master.WaitForShutdown) |
708 | 15486fa7 | Michael Hanselmann | finally:
|
709 | 15486fa7 | Michael Hanselmann | master.server_cleanup() |
710 | 4331f6cd | Michael Hanselmann | finally:
|
711 | 15486fa7 | Michael Hanselmann | rpc.Shutdown() |
712 | a4af651e | Iustin Pop | finally:
|
713 | 227647ac | Guido Trotter | utils.RemoveFile(constants.MASTER_SOCKET) |
714 | a4af651e | Iustin Pop | |
715 | 5483fd73 | Michael Hanselmann | logging.info("Clean master daemon shutdown")
|
716 | 5483fd73 | Michael Hanselmann | |
717 | ffeffa1d | Iustin Pop | |
718 | 29d91329 | Michael Hanselmann | def Main(): |
719 | 04ccf5e9 | Guido Trotter | """Main function"""
|
720 | 04ccf5e9 | Guido Trotter | parser = OptionParser(description="Ganeti master daemon",
|
721 | 04ccf5e9 | Guido Trotter | usage="%prog [-f] [-d]",
|
722 | 04ccf5e9 | Guido Trotter | version="%%prog (ganeti) %s" %
|
723 | 04ccf5e9 | Guido Trotter | constants.RELEASE_VERSION) |
724 | 04ccf5e9 | Guido Trotter | parser.add_option("--no-voting", dest="no_voting", |
725 | 04ccf5e9 | Guido Trotter | help="Do not check that the nodes agree on this node"
|
726 | 04ccf5e9 | Guido Trotter | " being the master and start the daemon unconditionally",
|
727 | 04ccf5e9 | Guido Trotter | default=False, action="store_true") |
728 | 04ccf5e9 | Guido Trotter | parser.add_option("--yes-do-it", dest="yes_do_it", |
729 | 04ccf5e9 | Guido Trotter | help="Override interactive check for --no-voting",
|
730 | 04ccf5e9 | Guido Trotter | default=False, action="store_true") |
731 | 3ee53f1f | Iustin Pop | daemon.GenericMain(constants.MASTERD, parser, CheckMasterd, PrepMasterd, |
732 | b42ea9ed | Iustin Pop | ExecMasterd, multithreaded=True) |