root / daemons / ganeti-masterd @ 7e5a6e86
History | View | Annotate | Download (17.6 kB)
1 | 834f8b67 | Iustin Pop | #!/usr/bin/python |
---|---|---|---|
2 | ffeffa1d | Iustin Pop | # |
3 | ffeffa1d | Iustin Pop | |
4 | ffeffa1d | Iustin Pop | # Copyright (C) 2006, 2007 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 | 7260cfbe | Iustin Pop | # pylint: disable-msg=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 | c1f2901b | Iustin Pop | |
58 | c1f2901b | Iustin Pop | |
59 | 23e50d39 | Michael Hanselmann | CLIENT_REQUEST_WORKERS = 16 |
60 | 23e50d39 | Michael Hanselmann | |
61 | c1f2901b | Iustin Pop | EXIT_NOTMASTER = constants.EXIT_NOTMASTER |
62 | c1f2901b | Iustin Pop | EXIT_NODESETUP_ERROR = constants.EXIT_NODESETUP_ERROR |
63 | ffeffa1d | Iustin Pop | |
64 | ffeffa1d | Iustin Pop | |
65 | 23e50d39 | Michael Hanselmann | class ClientRequestWorker(workerpool.BaseWorker): |
66 | 7260cfbe | Iustin Pop | # pylint: disable-msg=W0221 |
67 | 7e5a6e86 | Guido Trotter | def RunTask(self, server, message, client): |
68 | 23e50d39 | Michael Hanselmann | """Process the request. |
69 | 23e50d39 | Michael Hanselmann | |
70 | 23e50d39 | Michael Hanselmann | """ |
71 | 7e5a6e86 | Guido Trotter | client_ops = ClientOps(server) |
72 | 7e5a6e86 | Guido Trotter | |
73 | 23e50d39 | Michael Hanselmann | try: |
74 | 7e5a6e86 | Guido Trotter | (method, args) = luxi.ParseRequest(message) |
75 | 7e5a6e86 | Guido Trotter | except luxi.ProtocolError, err: |
76 | 7e5a6e86 | Guido Trotter | logging.error("Protocol Error: %s", err) |
77 | 7e5a6e86 | Guido Trotter | client.close_log() |
78 | 7e5a6e86 | Guido Trotter | return |
79 | 7e5a6e86 | Guido Trotter | |
80 | 7e5a6e86 | Guido Trotter | success = False |
81 | 7e5a6e86 | Guido Trotter | try: |
82 | 7e5a6e86 | Guido Trotter | result = client_ops.handle_request(method, args) |
83 | 7e5a6e86 | Guido Trotter | success = True |
84 | 7e5a6e86 | Guido Trotter | except errors.GenericError, err: |
85 | 7e5a6e86 | Guido Trotter | logging.exception("Unexpected exception") |
86 | 7e5a6e86 | Guido Trotter | success = False |
87 | 7e5a6e86 | Guido Trotter | result = errors.EncodeException(err) |
88 | 7e5a6e86 | Guido Trotter | except: |
89 | 7e5a6e86 | Guido Trotter | logging.exception("Unexpected exception") |
90 | 7e5a6e86 | Guido Trotter | err = sys.exc_info() |
91 | 7e5a6e86 | Guido Trotter | result = "Caught exception: %s" % str(err[1]) |
92 | 7e5a6e86 | Guido Trotter | |
93 | 7e5a6e86 | Guido Trotter | try: |
94 | 7e5a6e86 | Guido Trotter | reply = luxi.FormatResponse(success, result) |
95 | 7e5a6e86 | Guido Trotter | client.send_message(reply) |
96 | 7e5a6e86 | Guido Trotter | # awake the main thread so that it can write out the data. |
97 | 7e5a6e86 | Guido Trotter | server.awaker.signal() |
98 | 7e5a6e86 | Guido Trotter | except: |
99 | 7e5a6e86 | Guido Trotter | logging.exception("Send error") |
100 | 7e5a6e86 | Guido Trotter | client.close_log() |
101 | 7e5a6e86 | Guido Trotter | |
102 | 7e5a6e86 | Guido Trotter | |
103 | 7e5a6e86 | Guido Trotter | class MasterClientHandler(daemon.AsyncTerminatedMessageStream): |
104 | 7e5a6e86 | Guido Trotter | """Handler for master peers. |
105 | 7e5a6e86 | Guido Trotter | |
106 | 7e5a6e86 | Guido Trotter | """ |
107 | 7e5a6e86 | Guido Trotter | _MAX_UNHANDLED = 1 |
108 | 7e5a6e86 | Guido Trotter | def __init__(self, server, connected_socket, client_address, family): |
109 | 7e5a6e86 | Guido Trotter | daemon.AsyncTerminatedMessageStream.__init__(self, connected_socket, |
110 | 7e5a6e86 | Guido Trotter | client_address, |
111 | 7e5a6e86 | Guido Trotter | constants.LUXI_EOM, |
112 | 7e5a6e86 | Guido Trotter | family, self._MAX_UNHANDLED) |
113 | 7e5a6e86 | Guido Trotter | self.server = server |
114 | 7e5a6e86 | Guido Trotter | |
115 | 7e5a6e86 | Guido Trotter | def handle_message(self, message, _): |
116 | 7e5a6e86 | Guido Trotter | self.server.request_workers.AddTask(self.server, message, self) |
117 | 23e50d39 | Michael Hanselmann | |
118 | 23e50d39 | Michael Hanselmann | |
119 | cdd7f900 | Guido Trotter | class MasterServer(daemon.AsyncStreamServer): |
120 | cdd7f900 | Guido Trotter | """Master Server. |
121 | ffeffa1d | Iustin Pop | |
122 | cdd7f900 | Guido Trotter | This is the main asynchronous master server. It handles connections to the |
123 | cdd7f900 | Guido Trotter | master socket. |
124 | ffeffa1d | Iustin Pop | |
125 | ffeffa1d | Iustin Pop | """ |
126 | 7e5a6e86 | Guido Trotter | family = socket.AF_UNIX |
127 | 7e5a6e86 | Guido Trotter | |
128 | 7e5a6e86 | Guido Trotter | def __init__(self, mainloop, address, uid, gid): |
129 | cdd7f900 | Guido Trotter | """MasterServer constructor |
130 | ce862cd5 | Guido Trotter | |
131 | cdd7f900 | Guido Trotter | @type mainloop: ganeti.daemon.Mainloop |
132 | cdd7f900 | Guido Trotter | @param mainloop: Mainloop used to poll for I/O events |
133 | cdd7f900 | Guido Trotter | @param address: the unix socket address to bind the MasterServer to |
134 | bbfd0568 | René Nussbaumer | @param uid: The uid of the owner of the socket |
135 | bbfd0568 | René Nussbaumer | @param gid: The gid of the owner of the socket |
136 | ce862cd5 | Guido Trotter | |
137 | ce862cd5 | Guido Trotter | """ |
138 | bbfd0568 | René Nussbaumer | temp_name = tempfile.mktemp(dir=os.path.dirname(address)) |
139 | 7e5a6e86 | Guido Trotter | daemon.AsyncStreamServer.__init__(self, self.family, temp_name) |
140 | bbfd0568 | René Nussbaumer | os.chmod(temp_name, 0770) |
141 | bbfd0568 | René Nussbaumer | os.chown(temp_name, uid, gid) |
142 | bbfd0568 | René Nussbaumer | os.rename(temp_name, address) |
143 | bbfd0568 | René Nussbaumer | |
144 | cdd7f900 | Guido Trotter | self.mainloop = mainloop |
145 | 7e5a6e86 | Guido Trotter | self.awaker = daemon.AsyncAwaker() |
146 | 50a3fbb2 | Michael Hanselmann | |
147 | 50a3fbb2 | Michael Hanselmann | # We'll only start threads once we've forked. |
148 | 9113300d | Michael Hanselmann | self.context = None |
149 | 23e50d39 | Michael Hanselmann | self.request_workers = None |
150 | 50a3fbb2 | Michael Hanselmann | |
151 | cdd7f900 | Guido Trotter | def handle_connection(self, connected_socket, client_address): |
152 | 7e5a6e86 | Guido Trotter | # TODO: add connection count and limit the number of open connections to a |
153 | 7e5a6e86 | Guido Trotter | # maximum number to avoid breaking for lack of file descriptors or memory. |
154 | 7e5a6e86 | Guido Trotter | MasterClientHandler(self, connected_socket, client_address, self.family) |
155 | cdd7f900 | Guido Trotter | |
156 | 50a3fbb2 | Michael Hanselmann | def setup_queue(self): |
157 | 9113300d | Michael Hanselmann | self.context = GanetiContext() |
158 | 89e2b4d2 | Michael Hanselmann | self.request_workers = workerpool.WorkerPool("ClientReq", |
159 | 89e2b4d2 | Michael Hanselmann | CLIENT_REQUEST_WORKERS, |
160 | 23e50d39 | Michael Hanselmann | ClientRequestWorker) |
161 | ffeffa1d | Iustin Pop | |
162 | c1f2901b | Iustin Pop | def server_cleanup(self): |
163 | c1f2901b | Iustin Pop | """Cleanup the server. |
164 | c1f2901b | Iustin Pop | |
165 | c1f2901b | Iustin Pop | This involves shutting down the processor threads and the master |
166 | c1f2901b | Iustin Pop | socket. |
167 | c1f2901b | Iustin Pop | |
168 | c1f2901b | Iustin Pop | """ |
169 | 50a3fbb2 | Michael Hanselmann | try: |
170 | cdd7f900 | Guido Trotter | self.close() |
171 | 50a3fbb2 | Michael Hanselmann | finally: |
172 | 23e50d39 | Michael Hanselmann | if self.request_workers: |
173 | 36088c4c | Michael Hanselmann | self.request_workers.TerminateWorkers() |
174 | 9113300d | Michael Hanselmann | if self.context: |
175 | 9113300d | Michael Hanselmann | self.context.jobqueue.Shutdown() |
176 | ffeffa1d | Iustin Pop | |
177 | ffeffa1d | Iustin Pop | |
178 | ffeffa1d | Iustin Pop | class ClientOps: |
179 | ffeffa1d | Iustin Pop | """Class holding high-level client operations.""" |
180 | ffeffa1d | Iustin Pop | def __init__(self, server): |
181 | ffeffa1d | Iustin Pop | self.server = server |
182 | ffeffa1d | Iustin Pop | |
183 | 7260cfbe | Iustin Pop | def handle_request(self, method, args): # pylint: disable-msg=R0911 |
184 | 9113300d | Michael Hanselmann | queue = self.server.context.jobqueue |
185 | 0bbe448c | Michael Hanselmann | |
186 | 0bbe448c | Michael Hanselmann | # TODO: Parameter validation |
187 | 0bbe448c | Michael Hanselmann | |
188 | 7260cfbe | Iustin Pop | # TODO: Rewrite to not exit in each 'if/elif' branch |
189 | 7260cfbe | Iustin Pop | |
190 | 0bbe448c | Michael Hanselmann | if method == luxi.REQ_SUBMIT_JOB: |
191 | e566ddbd | Iustin Pop | logging.info("Received new job") |
192 | 0bbe448c | Michael Hanselmann | ops = [opcodes.OpCode.LoadOpCode(state) for state in args] |
193 | 4c848b18 | Michael Hanselmann | return queue.SubmitJob(ops) |
194 | ffeffa1d | Iustin Pop | |
195 | 2971c913 | Iustin Pop | if method == luxi.REQ_SUBMIT_MANY_JOBS: |
196 | 2971c913 | Iustin Pop | logging.info("Received multiple jobs") |
197 | 2971c913 | Iustin Pop | jobs = [] |
198 | 2971c913 | Iustin Pop | for ops in args: |
199 | 2971c913 | Iustin Pop | jobs.append([opcodes.OpCode.LoadOpCode(state) for state in ops]) |
200 | 2971c913 | Iustin Pop | return queue.SubmitManyJobs(jobs) |
201 | 2971c913 | Iustin Pop | |
202 | 0bbe448c | Michael Hanselmann | elif method == luxi.REQ_CANCEL_JOB: |
203 | 3a2c7775 | Michael Hanselmann | job_id = args |
204 | e566ddbd | Iustin Pop | logging.info("Received job cancel request for %s", job_id) |
205 | 0bbe448c | Michael Hanselmann | return queue.CancelJob(job_id) |
206 | ffeffa1d | Iustin Pop | |
207 | 0bbe448c | Michael Hanselmann | elif method == luxi.REQ_ARCHIVE_JOB: |
208 | 3a2c7775 | Michael Hanselmann | job_id = args |
209 | e566ddbd | Iustin Pop | logging.info("Received job archive request for %s", job_id) |
210 | 0bbe448c | Michael Hanselmann | return queue.ArchiveJob(job_id) |
211 | 0bbe448c | Michael Hanselmann | |
212 | 07cd723a | Iustin Pop | elif method == luxi.REQ_AUTOARCHIVE_JOBS: |
213 | f8ad5591 | Michael Hanselmann | (age, timeout) = args |
214 | e566ddbd | Iustin Pop | logging.info("Received job autoarchive request for age %s, timeout %s", |
215 | e566ddbd | Iustin Pop | age, timeout) |
216 | f8ad5591 | Michael Hanselmann | return queue.AutoArchiveJobs(age, timeout) |
217 | 07cd723a | Iustin Pop | |
218 | dfe57c22 | Michael Hanselmann | elif method == luxi.REQ_WAIT_FOR_JOB_CHANGE: |
219 | 5c735209 | Iustin Pop | (job_id, fields, prev_job_info, prev_log_serial, timeout) = args |
220 | e566ddbd | Iustin Pop | logging.info("Received job poll request for %s", job_id) |
221 | 6c5a7090 | Michael Hanselmann | return queue.WaitForJobChanges(job_id, fields, prev_job_info, |
222 | 5c735209 | Iustin Pop | prev_log_serial, timeout) |
223 | dfe57c22 | Michael Hanselmann | |
224 | 0bbe448c | Michael Hanselmann | elif method == luxi.REQ_QUERY_JOBS: |
225 | 0bbe448c | Michael Hanselmann | (job_ids, fields) = args |
226 | e566ddbd | Iustin Pop | if isinstance(job_ids, (tuple, list)) and job_ids: |
227 | 1f864b60 | Iustin Pop | msg = utils.CommaJoin(job_ids) |
228 | e566ddbd | Iustin Pop | else: |
229 | e566ddbd | Iustin Pop | msg = str(job_ids) |
230 | e566ddbd | Iustin Pop | logging.info("Received job query request for %s", msg) |
231 | 0bbe448c | Michael Hanselmann | return queue.QueryJobs(job_ids, fields) |
232 | 0bbe448c | Michael Hanselmann | |
233 | ee6c7b94 | Michael Hanselmann | elif method == luxi.REQ_QUERY_INSTANCES: |
234 | ec79568d | Iustin Pop | (names, fields, use_locking) = args |
235 | e566ddbd | Iustin Pop | logging.info("Received instance query request for %s", names) |
236 | 77921a95 | Iustin Pop | if use_locking: |
237 | debac808 | Iustin Pop | raise errors.OpPrereqError("Sync queries are not allowed", |
238 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
239 | ec79568d | Iustin Pop | op = opcodes.OpQueryInstances(names=names, output_fields=fields, |
240 | ec79568d | Iustin Pop | use_locking=use_locking) |
241 | ee6c7b94 | Michael Hanselmann | return self._Query(op) |
242 | ee6c7b94 | Michael Hanselmann | |
243 | 02f7fe54 | Michael Hanselmann | elif method == luxi.REQ_QUERY_NODES: |
244 | ec79568d | Iustin Pop | (names, fields, use_locking) = args |
245 | e566ddbd | Iustin Pop | logging.info("Received node query request for %s", names) |
246 | 77921a95 | Iustin Pop | if use_locking: |
247 | debac808 | Iustin Pop | raise errors.OpPrereqError("Sync queries are not allowed", |
248 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
249 | ec79568d | Iustin Pop | op = opcodes.OpQueryNodes(names=names, output_fields=fields, |
250 | ec79568d | Iustin Pop | use_locking=use_locking) |
251 | 02f7fe54 | Michael Hanselmann | return self._Query(op) |
252 | 02f7fe54 | Michael Hanselmann | |
253 | 32f93223 | Michael Hanselmann | elif method == luxi.REQ_QUERY_EXPORTS: |
254 | ec79568d | Iustin Pop | nodes, use_locking = args |
255 | 77921a95 | Iustin Pop | if use_locking: |
256 | debac808 | Iustin Pop | raise errors.OpPrereqError("Sync queries are not allowed", |
257 | debac808 | Iustin Pop | errors.ECODE_INVAL) |
258 | e566ddbd | Iustin Pop | logging.info("Received exports query request") |
259 | ec79568d | Iustin Pop | op = opcodes.OpQueryExports(nodes=nodes, use_locking=use_locking) |
260 | 32f93223 | Michael Hanselmann | return self._Query(op) |
261 | 32f93223 | Michael Hanselmann | |
262 | ae5849b5 | Michael Hanselmann | elif method == luxi.REQ_QUERY_CONFIG_VALUES: |
263 | ae5849b5 | Michael Hanselmann | fields = args |
264 | e566ddbd | Iustin Pop | logging.info("Received config values query request for %s", fields) |
265 | ae5849b5 | Michael Hanselmann | op = opcodes.OpQueryConfigValues(output_fields=fields) |
266 | ae5849b5 | Michael Hanselmann | return self._Query(op) |
267 | ae5849b5 | Michael Hanselmann | |
268 | 66baeccc | Iustin Pop | elif method == luxi.REQ_QUERY_CLUSTER_INFO: |
269 | e566ddbd | Iustin Pop | logging.info("Received cluster info query request") |
270 | 66baeccc | Iustin Pop | op = opcodes.OpQueryClusterInfo() |
271 | 66baeccc | Iustin Pop | return self._Query(op) |
272 | 66baeccc | Iustin Pop | |
273 | 7699c3af | Iustin Pop | elif method == luxi.REQ_QUERY_TAGS: |
274 | 7699c3af | Iustin Pop | kind, name = args |
275 | 7699c3af | Iustin Pop | logging.info("Received tags query request") |
276 | 7699c3af | Iustin Pop | op = opcodes.OpGetTags(kind=kind, name=name) |
277 | 7699c3af | Iustin Pop | return self._Query(op) |
278 | 7699c3af | Iustin Pop | |
279 | 3ccafd0e | Iustin Pop | elif method == luxi.REQ_QUEUE_SET_DRAIN_FLAG: |
280 | 3ccafd0e | Iustin Pop | drain_flag = args |
281 | e566ddbd | Iustin Pop | logging.info("Received queue drain flag change request to %s", |
282 | e566ddbd | Iustin Pop | drain_flag) |
283 | 3ccafd0e | Iustin Pop | return queue.SetDrainFlag(drain_flag) |
284 | 3ccafd0e | Iustin Pop | |
285 | 05e50653 | Michael Hanselmann | elif method == luxi.REQ_SET_WATCHER_PAUSE: |
286 | 05e50653 | Michael Hanselmann | (until, ) = args |
287 | 05e50653 | Michael Hanselmann | |
288 | 05e50653 | Michael Hanselmann | if until is None: |
289 | 05e50653 | Michael Hanselmann | logging.info("Received request to no longer pause the watcher") |
290 | 05e50653 | Michael Hanselmann | else: |
291 | 05e50653 | Michael Hanselmann | if not isinstance(until, (int, float)): |
292 | 05e50653 | Michael Hanselmann | raise TypeError("Duration must be an integer or float") |
293 | 05e50653 | Michael Hanselmann | |
294 | 05e50653 | Michael Hanselmann | if until < time.time(): |
295 | 05e50653 | Michael Hanselmann | raise errors.GenericError("Unable to set pause end time in the past") |
296 | 05e50653 | Michael Hanselmann | |
297 | 05e50653 | Michael Hanselmann | logging.info("Received request to pause the watcher until %s", until) |
298 | 05e50653 | Michael Hanselmann | |
299 | 05e50653 | Michael Hanselmann | return _SetWatcherPause(until) |
300 | 05e50653 | Michael Hanselmann | |
301 | 0bbe448c | Michael Hanselmann | else: |
302 | e566ddbd | Iustin Pop | logging.info("Received invalid request '%s'", method) |
303 | e566ddbd | Iustin Pop | raise ValueError("Invalid operation '%s'" % method) |
304 | ffeffa1d | Iustin Pop | |
305 | ee6c7b94 | Michael Hanselmann | def _Query(self, op): |
306 | ee6c7b94 | Michael Hanselmann | """Runs the specified opcode and returns the result. |
307 | ee6c7b94 | Michael Hanselmann | |
308 | ee6c7b94 | Michael Hanselmann | """ |
309 | adfa97e3 | Guido Trotter | # Queries don't have a job id |
310 | adfa97e3 | Guido Trotter | proc = mcpu.Processor(self.server.context, None) |
311 | 031a3e57 | Michael Hanselmann | return proc.ExecOpCode(op, None) |
312 | ee6c7b94 | Michael Hanselmann | |
313 | ffeffa1d | Iustin Pop | |
314 | 39dcf2ef | Guido Trotter | class GanetiContext(object): |
315 | 39dcf2ef | Guido Trotter | """Context common to all ganeti threads. |
316 | 39dcf2ef | Guido Trotter | |
317 | 39dcf2ef | Guido Trotter | This class creates and holds common objects shared by all threads. |
318 | 39dcf2ef | Guido Trotter | |
319 | 39dcf2ef | Guido Trotter | """ |
320 | 7260cfbe | Iustin Pop | # pylint: disable-msg=W0212 |
321 | 7260cfbe | Iustin Pop | # we do want to ensure a singleton here |
322 | 39dcf2ef | Guido Trotter | _instance = None |
323 | 39dcf2ef | Guido Trotter | |
324 | 39dcf2ef | Guido Trotter | def __init__(self): |
325 | 39dcf2ef | Guido Trotter | """Constructs a new GanetiContext object. |
326 | 39dcf2ef | Guido Trotter | |
327 | 39dcf2ef | Guido Trotter | There should be only a GanetiContext object at any time, so this |
328 | 39dcf2ef | Guido Trotter | function raises an error if this is not the case. |
329 | 39dcf2ef | Guido Trotter | |
330 | 39dcf2ef | Guido Trotter | """ |
331 | 39dcf2ef | Guido Trotter | assert self.__class__._instance is None, "double GanetiContext instance" |
332 | 39dcf2ef | Guido Trotter | |
333 | 9113300d | Michael Hanselmann | # Create global configuration object |
334 | 39dcf2ef | Guido Trotter | self.cfg = config.ConfigWriter() |
335 | 9113300d | Michael Hanselmann | |
336 | 9113300d | Michael Hanselmann | # Locking manager |
337 | 984f7c32 | Guido Trotter | self.glm = locking.GanetiLockManager( |
338 | 39dcf2ef | Guido Trotter | self.cfg.GetNodeList(), |
339 | 39dcf2ef | Guido Trotter | self.cfg.GetInstanceList()) |
340 | 39dcf2ef | Guido Trotter | |
341 | 9113300d | Michael Hanselmann | # Job queue |
342 | 9113300d | Michael Hanselmann | self.jobqueue = jqueue.JobQueue(self) |
343 | 9113300d | Michael Hanselmann | |
344 | 39dcf2ef | Guido Trotter | # setting this also locks the class against attribute modifications |
345 | 39dcf2ef | Guido Trotter | self.__class__._instance = self |
346 | 39dcf2ef | Guido Trotter | |
347 | 39dcf2ef | Guido Trotter | def __setattr__(self, name, value): |
348 | 39dcf2ef | Guido Trotter | """Setting GanetiContext attributes is forbidden after initialization. |
349 | 39dcf2ef | Guido Trotter | |
350 | 39dcf2ef | Guido Trotter | """ |
351 | 39dcf2ef | Guido Trotter | assert self.__class__._instance is None, "Attempt to modify Ganeti Context" |
352 | 39dcf2ef | Guido Trotter | object.__setattr__(self, name, value) |
353 | 39dcf2ef | Guido Trotter | |
354 | 0debfb35 | Guido Trotter | def AddNode(self, node, ec_id): |
355 | d8470559 | Michael Hanselmann | """Adds a node to the configuration and lock manager. |
356 | d8470559 | Michael Hanselmann | |
357 | d8470559 | Michael Hanselmann | """ |
358 | d8470559 | Michael Hanselmann | # Add it to the configuration |
359 | 0debfb35 | Guido Trotter | self.cfg.AddNode(node, ec_id) |
360 | d8470559 | Michael Hanselmann | |
361 | c36176cc | Michael Hanselmann | # If preseeding fails it'll not be added |
362 | 99aabbed | Iustin Pop | self.jobqueue.AddNode(node) |
363 | c36176cc | Michael Hanselmann | |
364 | d8470559 | Michael Hanselmann | # Add the new node to the Ganeti Lock Manager |
365 | d8470559 | Michael Hanselmann | self.glm.add(locking.LEVEL_NODE, node.name) |
366 | d8470559 | Michael Hanselmann | |
367 | d8470559 | Michael Hanselmann | def ReaddNode(self, node): |
368 | d8470559 | Michael Hanselmann | """Updates a node that's already in the configuration |
369 | d8470559 | Michael Hanselmann | |
370 | d8470559 | Michael Hanselmann | """ |
371 | c36176cc | Michael Hanselmann | # Synchronize the queue again |
372 | 99aabbed | Iustin Pop | self.jobqueue.AddNode(node) |
373 | d8470559 | Michael Hanselmann | |
374 | d8470559 | Michael Hanselmann | def RemoveNode(self, name): |
375 | d8470559 | Michael Hanselmann | """Removes a node from the configuration and lock manager. |
376 | d8470559 | Michael Hanselmann | |
377 | d8470559 | Michael Hanselmann | """ |
378 | d8470559 | Michael Hanselmann | # Remove node from configuration |
379 | d8470559 | Michael Hanselmann | self.cfg.RemoveNode(name) |
380 | d8470559 | Michael Hanselmann | |
381 | c36176cc | Michael Hanselmann | # Notify job queue |
382 | c36176cc | Michael Hanselmann | self.jobqueue.RemoveNode(name) |
383 | c36176cc | Michael Hanselmann | |
384 | d8470559 | Michael Hanselmann | # Remove the node from the Ganeti Lock Manager |
385 | d8470559 | Michael Hanselmann | self.glm.remove(locking.LEVEL_NODE, name) |
386 | d8470559 | Michael Hanselmann | |
387 | 39dcf2ef | Guido Trotter | |
388 | 05e50653 | Michael Hanselmann | def _SetWatcherPause(until): |
389 | 05e50653 | Michael Hanselmann | """Creates or removes the watcher pause file. |
390 | 05e50653 | Michael Hanselmann | |
391 | 05e50653 | Michael Hanselmann | @type until: None or int |
392 | 05e50653 | Michael Hanselmann | @param until: Unix timestamp saying until when the watcher shouldn't run |
393 | 05e50653 | Michael Hanselmann | |
394 | 05e50653 | Michael Hanselmann | """ |
395 | 05e50653 | Michael Hanselmann | if until is None: |
396 | 05e50653 | Michael Hanselmann | utils.RemoveFile(constants.WATCHER_PAUSEFILE) |
397 | 05e50653 | Michael Hanselmann | else: |
398 | 05e50653 | Michael Hanselmann | utils.WriteFile(constants.WATCHER_PAUSEFILE, |
399 | 05e50653 | Michael Hanselmann | data="%d\n" % (until, )) |
400 | 05e50653 | Michael Hanselmann | |
401 | 28b498cd | Michael Hanselmann | return until |
402 | 28b498cd | Michael Hanselmann | |
403 | 05e50653 | Michael Hanselmann | |
404 | 36205981 | Iustin Pop | def CheckAgreement(): |
405 | 36205981 | Iustin Pop | """Check the agreement on who is the master. |
406 | 36205981 | Iustin Pop | |
407 | 36205981 | Iustin Pop | The function uses a very simple algorithm: we must get more positive |
408 | 36205981 | Iustin Pop | than negative answers. Since in most of the cases we are the master, |
409 | 36205981 | Iustin Pop | we'll use our own config file for getting the node list. In the |
410 | 36205981 | Iustin Pop | future we could collect the current node list from our (possibly |
411 | 36205981 | Iustin Pop | obsolete) known nodes. |
412 | 36205981 | Iustin Pop | |
413 | d7cdb55d | Iustin Pop | In order to account for cold-start of all nodes, we retry for up to |
414 | d7cdb55d | Iustin Pop | a minute until we get a real answer as the top-voted one. If the |
415 | d7cdb55d | Iustin Pop | nodes are more out-of-sync, for now manual startup of the master |
416 | d7cdb55d | Iustin Pop | should be attempted. |
417 | d7cdb55d | Iustin Pop | |
418 | d7cdb55d | Iustin Pop | Note that for a even number of nodes cluster, we need at least half |
419 | d7cdb55d | Iustin Pop | of the nodes (beside ourselves) to vote for us. This creates a |
420 | d7cdb55d | Iustin Pop | problem on two-node clusters, since in this case we require the |
421 | d7cdb55d | Iustin Pop | other node to be up too to confirm our status. |
422 | d7cdb55d | Iustin Pop | |
423 | 36205981 | Iustin Pop | """ |
424 | 36205981 | Iustin Pop | myself = utils.HostInfo().name |
425 | 36205981 | Iustin Pop | #temp instantiation of a config writer, used only to get the node list |
426 | 36205981 | Iustin Pop | cfg = config.ConfigWriter() |
427 | 36205981 | Iustin Pop | node_list = cfg.GetNodeList() |
428 | 36205981 | Iustin Pop | del cfg |
429 | d7cdb55d | Iustin Pop | retries = 6 |
430 | d7cdb55d | Iustin Pop | while retries > 0: |
431 | d7cdb55d | Iustin Pop | votes = bootstrap.GatherMasterVotes(node_list) |
432 | d7cdb55d | Iustin Pop | if not votes: |
433 | d7cdb55d | Iustin Pop | # empty node list, this is a one node cluster |
434 | d7cdb55d | Iustin Pop | return True |
435 | d7cdb55d | Iustin Pop | if votes[0][0] is None: |
436 | d7cdb55d | Iustin Pop | retries -= 1 |
437 | d7cdb55d | Iustin Pop | time.sleep(10) |
438 | 36205981 | Iustin Pop | continue |
439 | d7cdb55d | Iustin Pop | break |
440 | d7cdb55d | Iustin Pop | if retries == 0: |
441 | e09fdcfa | Iustin Pop | logging.critical("Cluster inconsistent, most of the nodes didn't answer" |
442 | e09fdcfa | Iustin Pop | " after multiple retries. Aborting startup") |
443 | d8f5a37d | Iustin Pop | logging.critical("Use the --no-voting option if you understand what" |
444 | d8f5a37d | Iustin Pop | " effects it has on the cluster state") |
445 | e09fdcfa | Iustin Pop | return False |
446 | d7cdb55d | Iustin Pop | # here a real node is at the top of the list |
447 | d7cdb55d | Iustin Pop | all_votes = sum(item[1] for item in votes) |
448 | d7cdb55d | Iustin Pop | top_node, top_votes = votes[0] |
449 | 8a20c732 | Michael Hanselmann | |
450 | d7cdb55d | Iustin Pop | result = False |
451 | d7cdb55d | Iustin Pop | if top_node != myself: |
452 | d7cdb55d | Iustin Pop | logging.critical("It seems we are not the master (top-voted node" |
453 | bbe19c17 | Iustin Pop | " is %s with %d out of %d votes)", top_node, top_votes, |
454 | bbe19c17 | Iustin Pop | all_votes) |
455 | d7cdb55d | Iustin Pop | elif top_votes < all_votes - top_votes: |
456 | 36205981 | Iustin Pop | logging.critical("It seems we are not the master (%d votes for," |
457 | d7cdb55d | Iustin Pop | " %d votes against)", top_votes, all_votes - top_votes) |
458 | d7cdb55d | Iustin Pop | else: |
459 | d7cdb55d | Iustin Pop | result = True |
460 | d7cdb55d | Iustin Pop | |
461 | d7cdb55d | Iustin Pop | return result |
462 | 36205981 | Iustin Pop | |
463 | 6c948699 | Michael Hanselmann | |
464 | ed0efaa5 | Michael Hanselmann | def CheckAgreementWithRpc(): |
465 | 4331f6cd | Michael Hanselmann | rpc.Init() |
466 | 4331f6cd | Michael Hanselmann | try: |
467 | ed0efaa5 | Michael Hanselmann | return CheckAgreement() |
468 | 4331f6cd | Michael Hanselmann | finally: |
469 | 4331f6cd | Michael Hanselmann | rpc.Shutdown() |
470 | ffeffa1d | Iustin Pop | |
471 | c1f2901b | Iustin Pop | |
472 | ed0efaa5 | Michael Hanselmann | def CheckMasterd(options, args): |
473 | ed0efaa5 | Michael Hanselmann | """Initial checks whether to run or exit with a failure. |
474 | ed0efaa5 | Michael Hanselmann | |
475 | ed0efaa5 | Michael Hanselmann | """ |
476 | f93427cd | Iustin Pop | if args: # masterd doesn't take any arguments |
477 | f93427cd | Iustin Pop | print >> sys.stderr, ("Usage: %s [-f] [-d]" % sys.argv[0]) |
478 | f93427cd | Iustin Pop | sys.exit(constants.EXIT_FAILURE) |
479 | f93427cd | Iustin Pop | |
480 | ed0efaa5 | Michael Hanselmann | ssconf.CheckMaster(options.debug) |
481 | ed0efaa5 | Michael Hanselmann | |
482 | bbfd0568 | René Nussbaumer | try: |
483 | bbfd0568 | René Nussbaumer | options.uid = pwd.getpwnam(constants.MASTERD_USER).pw_uid |
484 | bbfd0568 | René Nussbaumer | options.gid = grp.getgrnam(constants.DAEMONS_GROUP).gr_gid |
485 | bbfd0568 | René Nussbaumer | except KeyError: |
486 | bbfd0568 | René Nussbaumer | print >> sys.stderr, ("User or group not existing on system: %s:%s" % |
487 | bbfd0568 | René Nussbaumer | (constants.MASTERD_USER, constants.DAEMONS_GROUP)) |
488 | bbfd0568 | René Nussbaumer | sys.exit(constants.EXIT_FAILURE) |
489 | bbfd0568 | René Nussbaumer | |
490 | bbfd0568 | René Nussbaumer | |
491 | ed0efaa5 | Michael Hanselmann | # If CheckMaster didn't fail we believe we are the master, but we have to |
492 | ed0efaa5 | Michael Hanselmann | # confirm with the other nodes. |
493 | ed0efaa5 | Michael Hanselmann | if options.no_voting: |
494 | ed0efaa5 | Michael Hanselmann | if options.yes_do_it: |
495 | ed0efaa5 | Michael Hanselmann | return |
496 | ed0efaa5 | Michael Hanselmann | |
497 | ed0efaa5 | Michael Hanselmann | sys.stdout.write("The 'no voting' option has been selected.\n") |
498 | ed0efaa5 | Michael Hanselmann | sys.stdout.write("This is dangerous, please confirm by" |
499 | ed0efaa5 | Michael Hanselmann | " typing uppercase 'yes': ") |
500 | ed0efaa5 | Michael Hanselmann | sys.stdout.flush() |
501 | ed0efaa5 | Michael Hanselmann | |
502 | ed0efaa5 | Michael Hanselmann | confirmation = sys.stdin.readline().strip() |
503 | ed0efaa5 | Michael Hanselmann | if confirmation != "YES": |
504 | 7260cfbe | Iustin Pop | print >> sys.stderr, "Aborting." |
505 | ed0efaa5 | Michael Hanselmann | sys.exit(constants.EXIT_FAILURE) |
506 | ed0efaa5 | Michael Hanselmann | |
507 | ed0efaa5 | Michael Hanselmann | return |
508 | ed0efaa5 | Michael Hanselmann | |
509 | ed0efaa5 | Michael Hanselmann | # CheckAgreement uses RPC and threads, hence it needs to be run in a separate |
510 | ed0efaa5 | Michael Hanselmann | # process before we call utils.Daemonize in the current process. |
511 | eb58f7bd | Michael Hanselmann | if not utils.RunInSeparateProcess(CheckAgreementWithRpc): |
512 | ed0efaa5 | Michael Hanselmann | sys.exit(constants.EXIT_FAILURE) |
513 | ed0efaa5 | Michael Hanselmann | |
514 | ed0efaa5 | Michael Hanselmann | |
515 | 2d54e29c | Iustin Pop | def ExecMasterd (options, args): # pylint: disable-msg=W0613 |
516 | 6c948699 | Michael Hanselmann | """Main master daemon function, executed with the PID file held. |
517 | 3b316acb | Iustin Pop | |
518 | 04ccf5e9 | Guido Trotter | """ |
519 | 04ccf5e9 | Guido Trotter | # This is safe to do as the pid file guarantees against |
520 | 04ccf5e9 | Guido Trotter | # concurrent execution. |
521 | 04ccf5e9 | Guido Trotter | utils.RemoveFile(constants.MASTER_SOCKET) |
522 | b1b6ea87 | Iustin Pop | |
523 | cdd7f900 | Guido Trotter | mainloop = daemon.Mainloop() |
524 | 7e5a6e86 | Guido Trotter | master = MasterServer(mainloop, constants.MASTER_SOCKET, |
525 | bbfd0568 | René Nussbaumer | options.uid, options.gid) |
526 | 04ccf5e9 | Guido Trotter | try: |
527 | 15486fa7 | Michael Hanselmann | rpc.Init() |
528 | 4331f6cd | Michael Hanselmann | try: |
529 | 15486fa7 | Michael Hanselmann | # activate ip |
530 | b2890442 | Guido Trotter | master_node = ssconf.SimpleStore().GetMasterNode() |
531 | 3583908a | Guido Trotter | result = rpc.RpcRunner.call_node_start_master(master_node, False, False) |
532 | 3cebe102 | Michael Hanselmann | msg = result.fail_msg |
533 | b726aff0 | Iustin Pop | if msg: |
534 | b726aff0 | Iustin Pop | logging.error("Can't activate master IP address: %s", msg) |
535 | 15486fa7 | Michael Hanselmann | |
536 | 15486fa7 | Michael Hanselmann | master.setup_queue() |
537 | 15486fa7 | Michael Hanselmann | try: |
538 | cdd7f900 | Guido Trotter | mainloop.Run() |
539 | 15486fa7 | Michael Hanselmann | finally: |
540 | 15486fa7 | Michael Hanselmann | master.server_cleanup() |
541 | 4331f6cd | Michael Hanselmann | finally: |
542 | 15486fa7 | Michael Hanselmann | rpc.Shutdown() |
543 | a4af651e | Iustin Pop | finally: |
544 | 227647ac | Guido Trotter | utils.RemoveFile(constants.MASTER_SOCKET) |
545 | a4af651e | Iustin Pop | |
546 | ffeffa1d | Iustin Pop | |
547 | 04ccf5e9 | Guido Trotter | def main(): |
548 | 04ccf5e9 | Guido Trotter | """Main function""" |
549 | 04ccf5e9 | Guido Trotter | parser = OptionParser(description="Ganeti master daemon", |
550 | 04ccf5e9 | Guido Trotter | usage="%prog [-f] [-d]", |
551 | 04ccf5e9 | Guido Trotter | version="%%prog (ganeti) %s" % |
552 | 04ccf5e9 | Guido Trotter | constants.RELEASE_VERSION) |
553 | 04ccf5e9 | Guido Trotter | parser.add_option("--no-voting", dest="no_voting", |
554 | 04ccf5e9 | Guido Trotter | help="Do not check that the nodes agree on this node" |
555 | 04ccf5e9 | Guido Trotter | " being the master and start the daemon unconditionally", |
556 | 04ccf5e9 | Guido Trotter | default=False, action="store_true") |
557 | 04ccf5e9 | Guido Trotter | parser.add_option("--yes-do-it", dest="yes_do_it", |
558 | 04ccf5e9 | Guido Trotter | help="Override interactive check for --no-voting", |
559 | 04ccf5e9 | Guido Trotter | default=False, action="store_true") |
560 | 04ccf5e9 | Guido Trotter | dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE), |
561 | 04ccf5e9 | Guido Trotter | (constants.SOCKET_DIR, constants.SOCKET_DIR_MODE), |
562 | 04ccf5e9 | Guido Trotter | ] |
563 | 04ccf5e9 | Guido Trotter | daemon.GenericMain(constants.MASTERD, parser, dirs, |
564 | 30dabd03 | Michael Hanselmann | CheckMasterd, ExecMasterd, |
565 | 30dabd03 | Michael Hanselmann | multithreaded=True) |
566 | 6c948699 | Michael Hanselmann | |
567 | 04ccf5e9 | Guido Trotter | |
568 | ffeffa1d | Iustin Pop | if __name__ == "__main__": |
569 | ffeffa1d | Iustin Pop | main() |