Statistics
| Branch: | Tag: | Revision:

root / daemons / ganeti-masterd @ 340f4757

History | View | Annotate | Download (17.8 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 a744b676 Manuel Franceschini
from ganeti import netutils
58 c1f2901b Iustin Pop
59 c1f2901b Iustin Pop
60 23e50d39 Michael Hanselmann
CLIENT_REQUEST_WORKERS = 16
61 23e50d39 Michael Hanselmann
62 c1f2901b Iustin Pop
EXIT_NOTMASTER = constants.EXIT_NOTMASTER
63 c1f2901b Iustin Pop
EXIT_NODESETUP_ERROR = constants.EXIT_NODESETUP_ERROR
64 ffeffa1d Iustin Pop
65 ffeffa1d Iustin Pop
66 23e50d39 Michael Hanselmann
class ClientRequestWorker(workerpool.BaseWorker):
67 e0dbb89b Guido Trotter
  # pylint: disable-msg=W0221
68 7e5a6e86 Guido Trotter
  def RunTask(self, server, message, client):
69 23e50d39 Michael Hanselmann
    """Process the request.
70 23e50d39 Michael Hanselmann
71 23e50d39 Michael Hanselmann
    """
72 7e5a6e86 Guido Trotter
    client_ops = ClientOps(server)
73 7e5a6e86 Guido Trotter
74 23e50d39 Michael Hanselmann
    try:
75 7e5a6e86 Guido Trotter
      (method, args) = luxi.ParseRequest(message)
76 7e5a6e86 Guido Trotter
    except luxi.ProtocolError, err:
77 7e5a6e86 Guido Trotter
      logging.error("Protocol Error: %s", err)
78 7e5a6e86 Guido Trotter
      client.close_log()
79 7e5a6e86 Guido Trotter
      return
80 7e5a6e86 Guido Trotter
81 7e5a6e86 Guido Trotter
    success = False
82 7e5a6e86 Guido Trotter
    try:
83 7e5a6e86 Guido Trotter
      result = client_ops.handle_request(method, args)
84 7e5a6e86 Guido Trotter
      success = True
85 7e5a6e86 Guido Trotter
    except errors.GenericError, err:
86 7e5a6e86 Guido Trotter
      logging.exception("Unexpected exception")
87 7e5a6e86 Guido Trotter
      success = False
88 7e5a6e86 Guido Trotter
      result = errors.EncodeException(err)
89 7e5a6e86 Guido Trotter
    except:
90 7e5a6e86 Guido Trotter
      logging.exception("Unexpected exception")
91 7e5a6e86 Guido Trotter
      err = sys.exc_info()
92 7e5a6e86 Guido Trotter
      result = "Caught exception: %s" % str(err[1])
93 7e5a6e86 Guido Trotter
94 7e5a6e86 Guido Trotter
    try:
95 7e5a6e86 Guido Trotter
      reply = luxi.FormatResponse(success, result)
96 7e5a6e86 Guido Trotter
      client.send_message(reply)
97 7e5a6e86 Guido Trotter
      # awake the main thread so that it can write out the data.
98 7e5a6e86 Guido Trotter
      server.awaker.signal()
99 e0dbb89b Guido Trotter
    except: # pylint: disable-msg=W0702
100 7e5a6e86 Guido Trotter
      logging.exception("Send error")
101 7e5a6e86 Guido Trotter
      client.close_log()
102 7e5a6e86 Guido Trotter
103 7e5a6e86 Guido Trotter
104 7e5a6e86 Guido Trotter
class MasterClientHandler(daemon.AsyncTerminatedMessageStream):
105 7e5a6e86 Guido Trotter
  """Handler for master peers.
106 7e5a6e86 Guido Trotter
107 7e5a6e86 Guido Trotter
  """
108 7e5a6e86 Guido Trotter
  _MAX_UNHANDLED = 1
109 7e5a6e86 Guido Trotter
  def __init__(self, server, connected_socket, client_address, family):
110 7e5a6e86 Guido Trotter
    daemon.AsyncTerminatedMessageStream.__init__(self, connected_socket,
111 7e5a6e86 Guido Trotter
                                                 client_address,
112 7e5a6e86 Guido Trotter
                                                 constants.LUXI_EOM,
113 7e5a6e86 Guido Trotter
                                                 family, self._MAX_UNHANDLED)
114 7e5a6e86 Guido Trotter
    self.server = server
115 7e5a6e86 Guido Trotter
116 7e5a6e86 Guido Trotter
  def handle_message(self, message, _):
117 7e5a6e86 Guido Trotter
    self.server.request_workers.AddTask(self.server, message, self)
118 23e50d39 Michael Hanselmann
119 23e50d39 Michael Hanselmann
120 cdd7f900 Guido Trotter
class MasterServer(daemon.AsyncStreamServer):
121 cdd7f900 Guido Trotter
  """Master Server.
122 ffeffa1d Iustin Pop
123 cdd7f900 Guido Trotter
  This is the main asynchronous master server. It handles connections to the
124 cdd7f900 Guido Trotter
  master socket.
125 ffeffa1d Iustin Pop
126 ffeffa1d Iustin Pop
  """
127 7e5a6e86 Guido Trotter
  family = socket.AF_UNIX
128 7e5a6e86 Guido Trotter
129 7e5a6e86 Guido Trotter
  def __init__(self, mainloop, address, uid, gid):
130 cdd7f900 Guido Trotter
    """MasterServer constructor
131 ce862cd5 Guido Trotter
132 cdd7f900 Guido Trotter
    @type mainloop: ganeti.daemon.Mainloop
133 cdd7f900 Guido Trotter
    @param mainloop: Mainloop used to poll for I/O events
134 cdd7f900 Guido Trotter
    @param address: the unix socket address to bind the MasterServer to
135 bbfd0568 René Nussbaumer
    @param uid: The uid of the owner of the socket
136 bbfd0568 René Nussbaumer
    @param gid: The gid of the owner of the socket
137 ce862cd5 Guido Trotter
138 ce862cd5 Guido Trotter
    """
139 bbfd0568 René Nussbaumer
    temp_name = tempfile.mktemp(dir=os.path.dirname(address))
140 7e5a6e86 Guido Trotter
    daemon.AsyncStreamServer.__init__(self, self.family, temp_name)
141 bbfd0568 René Nussbaumer
    os.chmod(temp_name, 0770)
142 bbfd0568 René Nussbaumer
    os.chown(temp_name, uid, gid)
143 bbfd0568 René Nussbaumer
    os.rename(temp_name, address)
144 bbfd0568 René Nussbaumer
145 cdd7f900 Guido Trotter
    self.mainloop = mainloop
146 7e5a6e86 Guido Trotter
    self.awaker = daemon.AsyncAwaker()
147 50a3fbb2 Michael Hanselmann
148 50a3fbb2 Michael Hanselmann
    # We'll only start threads once we've forked.
149 9113300d Michael Hanselmann
    self.context = None
150 23e50d39 Michael Hanselmann
    self.request_workers = None
151 50a3fbb2 Michael Hanselmann
152 cdd7f900 Guido Trotter
  def handle_connection(self, connected_socket, client_address):
153 7e5a6e86 Guido Trotter
    # TODO: add connection count and limit the number of open connections to a
154 7e5a6e86 Guido Trotter
    # maximum number to avoid breaking for lack of file descriptors or memory.
155 7e5a6e86 Guido Trotter
    MasterClientHandler(self, connected_socket, client_address, self.family)
156 cdd7f900 Guido Trotter
157 50a3fbb2 Michael Hanselmann
  def setup_queue(self):
158 9113300d Michael Hanselmann
    self.context = GanetiContext()
159 89e2b4d2 Michael Hanselmann
    self.request_workers = workerpool.WorkerPool("ClientReq",
160 89e2b4d2 Michael Hanselmann
                                                 CLIENT_REQUEST_WORKERS,
161 23e50d39 Michael Hanselmann
                                                 ClientRequestWorker)
162 ffeffa1d Iustin Pop
163 c1f2901b Iustin Pop
  def server_cleanup(self):
164 c1f2901b Iustin Pop
    """Cleanup the server.
165 c1f2901b Iustin Pop
166 c1f2901b Iustin Pop
    This involves shutting down the processor threads and the master
167 c1f2901b Iustin Pop
    socket.
168 c1f2901b Iustin Pop
169 c1f2901b Iustin Pop
    """
170 50a3fbb2 Michael Hanselmann
    try:
171 cdd7f900 Guido Trotter
      self.close()
172 50a3fbb2 Michael Hanselmann
    finally:
173 23e50d39 Michael Hanselmann
      if self.request_workers:
174 36088c4c Michael Hanselmann
        self.request_workers.TerminateWorkers()
175 9113300d Michael Hanselmann
      if self.context:
176 9113300d Michael Hanselmann
        self.context.jobqueue.Shutdown()
177 ffeffa1d Iustin Pop
178 ffeffa1d Iustin Pop
179 ffeffa1d Iustin Pop
class ClientOps:
180 ffeffa1d Iustin Pop
  """Class holding high-level client operations."""
181 ffeffa1d Iustin Pop
  def __init__(self, server):
182 ffeffa1d Iustin Pop
    self.server = server
183 ffeffa1d Iustin Pop
184 7260cfbe Iustin Pop
  def handle_request(self, method, args): # pylint: disable-msg=R0911
185 9113300d Michael Hanselmann
    queue = self.server.context.jobqueue
186 0bbe448c Michael Hanselmann
187 0bbe448c Michael Hanselmann
    # TODO: Parameter validation
188 0bbe448c Michael Hanselmann
189 7260cfbe Iustin Pop
    # TODO: Rewrite to not exit in each 'if/elif' branch
190 7260cfbe Iustin Pop
191 0bbe448c Michael Hanselmann
    if method == luxi.REQ_SUBMIT_JOB:
192 e566ddbd Iustin Pop
      logging.info("Received new job")
193 0bbe448c Michael Hanselmann
      ops = [opcodes.OpCode.LoadOpCode(state) for state in args]
194 4c848b18 Michael Hanselmann
      return queue.SubmitJob(ops)
195 ffeffa1d Iustin Pop
196 2971c913 Iustin Pop
    if method == luxi.REQ_SUBMIT_MANY_JOBS:
197 2971c913 Iustin Pop
      logging.info("Received multiple jobs")
198 2971c913 Iustin Pop
      jobs = []
199 2971c913 Iustin Pop
      for ops in args:
200 2971c913 Iustin Pop
        jobs.append([opcodes.OpCode.LoadOpCode(state) for state in ops])
201 2971c913 Iustin Pop
      return queue.SubmitManyJobs(jobs)
202 2971c913 Iustin Pop
203 0bbe448c Michael Hanselmann
    elif method == luxi.REQ_CANCEL_JOB:
204 3a2c7775 Michael Hanselmann
      job_id = args
205 e566ddbd Iustin Pop
      logging.info("Received job cancel request for %s", job_id)
206 0bbe448c Michael Hanselmann
      return queue.CancelJob(job_id)
207 ffeffa1d Iustin Pop
208 0bbe448c Michael Hanselmann
    elif method == luxi.REQ_ARCHIVE_JOB:
209 3a2c7775 Michael Hanselmann
      job_id = args
210 e566ddbd Iustin Pop
      logging.info("Received job archive request for %s", job_id)
211 0bbe448c Michael Hanselmann
      return queue.ArchiveJob(job_id)
212 0bbe448c Michael Hanselmann
213 07cd723a Iustin Pop
    elif method == luxi.REQ_AUTOARCHIVE_JOBS:
214 f8ad5591 Michael Hanselmann
      (age, timeout) = args
215 e566ddbd Iustin Pop
      logging.info("Received job autoarchive request for age %s, timeout %s",
216 e566ddbd Iustin Pop
                   age, timeout)
217 f8ad5591 Michael Hanselmann
      return queue.AutoArchiveJobs(age, timeout)
218 07cd723a Iustin Pop
219 dfe57c22 Michael Hanselmann
    elif method == luxi.REQ_WAIT_FOR_JOB_CHANGE:
220 5c735209 Iustin Pop
      (job_id, fields, prev_job_info, prev_log_serial, timeout) = args
221 e566ddbd Iustin Pop
      logging.info("Received job poll request for %s", job_id)
222 6c5a7090 Michael Hanselmann
      return queue.WaitForJobChanges(job_id, fields, prev_job_info,
223 5c735209 Iustin Pop
                                     prev_log_serial, timeout)
224 dfe57c22 Michael Hanselmann
225 0bbe448c Michael Hanselmann
    elif method == luxi.REQ_QUERY_JOBS:
226 0bbe448c Michael Hanselmann
      (job_ids, fields) = args
227 e566ddbd Iustin Pop
      if isinstance(job_ids, (tuple, list)) and job_ids:
228 1f864b60 Iustin Pop
        msg = utils.CommaJoin(job_ids)
229 e566ddbd Iustin Pop
      else:
230 e566ddbd Iustin Pop
        msg = str(job_ids)
231 e566ddbd Iustin Pop
      logging.info("Received job query request for %s", msg)
232 0bbe448c Michael Hanselmann
      return queue.QueryJobs(job_ids, fields)
233 0bbe448c Michael Hanselmann
234 ee6c7b94 Michael Hanselmann
    elif method == luxi.REQ_QUERY_INSTANCES:
235 ec79568d Iustin Pop
      (names, fields, use_locking) = args
236 e566ddbd Iustin Pop
      logging.info("Received instance query request for %s", names)
237 77921a95 Iustin Pop
      if use_locking:
238 debac808 Iustin Pop
        raise errors.OpPrereqError("Sync queries are not allowed",
239 debac808 Iustin Pop
                                   errors.ECODE_INVAL)
240 ec79568d Iustin Pop
      op = opcodes.OpQueryInstances(names=names, output_fields=fields,
241 ec79568d Iustin Pop
                                    use_locking=use_locking)
242 ee6c7b94 Michael Hanselmann
      return self._Query(op)
243 ee6c7b94 Michael Hanselmann
244 02f7fe54 Michael Hanselmann
    elif method == luxi.REQ_QUERY_NODES:
245 ec79568d Iustin Pop
      (names, fields, use_locking) = args
246 e566ddbd Iustin Pop
      logging.info("Received node query request for %s", names)
247 77921a95 Iustin Pop
      if use_locking:
248 debac808 Iustin Pop
        raise errors.OpPrereqError("Sync queries are not allowed",
249 debac808 Iustin Pop
                                   errors.ECODE_INVAL)
250 ec79568d Iustin Pop
      op = opcodes.OpQueryNodes(names=names, output_fields=fields,
251 ec79568d Iustin Pop
                                use_locking=use_locking)
252 02f7fe54 Michael Hanselmann
      return self._Query(op)
253 02f7fe54 Michael Hanselmann
254 32f93223 Michael Hanselmann
    elif method == luxi.REQ_QUERY_EXPORTS:
255 ec79568d Iustin Pop
      nodes, use_locking = args
256 77921a95 Iustin Pop
      if use_locking:
257 debac808 Iustin Pop
        raise errors.OpPrereqError("Sync queries are not allowed",
258 debac808 Iustin Pop
                                   errors.ECODE_INVAL)
259 e566ddbd Iustin Pop
      logging.info("Received exports query request")
260 ec79568d Iustin Pop
      op = opcodes.OpQueryExports(nodes=nodes, use_locking=use_locking)
261 32f93223 Michael Hanselmann
      return self._Query(op)
262 32f93223 Michael Hanselmann
263 ae5849b5 Michael Hanselmann
    elif method == luxi.REQ_QUERY_CONFIG_VALUES:
264 ae5849b5 Michael Hanselmann
      fields = args
265 e566ddbd Iustin Pop
      logging.info("Received config values query request for %s", fields)
266 ae5849b5 Michael Hanselmann
      op = opcodes.OpQueryConfigValues(output_fields=fields)
267 ae5849b5 Michael Hanselmann
      return self._Query(op)
268 ae5849b5 Michael Hanselmann
269 66baeccc Iustin Pop
    elif method == luxi.REQ_QUERY_CLUSTER_INFO:
270 e566ddbd Iustin Pop
      logging.info("Received cluster info query request")
271 66baeccc Iustin Pop
      op = opcodes.OpQueryClusterInfo()
272 66baeccc Iustin Pop
      return self._Query(op)
273 66baeccc Iustin Pop
274 7699c3af Iustin Pop
    elif method == luxi.REQ_QUERY_TAGS:
275 7699c3af Iustin Pop
      kind, name = args
276 7699c3af Iustin Pop
      logging.info("Received tags query request")
277 7699c3af Iustin Pop
      op = opcodes.OpGetTags(kind=kind, name=name)
278 7699c3af Iustin Pop
      return self._Query(op)
279 7699c3af Iustin Pop
280 3ccafd0e Iustin Pop
    elif method == luxi.REQ_QUEUE_SET_DRAIN_FLAG:
281 3ccafd0e Iustin Pop
      drain_flag = args
282 e566ddbd Iustin Pop
      logging.info("Received queue drain flag change request to %s",
283 e566ddbd Iustin Pop
                   drain_flag)
284 3ccafd0e Iustin Pop
      return queue.SetDrainFlag(drain_flag)
285 3ccafd0e Iustin Pop
286 05e50653 Michael Hanselmann
    elif method == luxi.REQ_SET_WATCHER_PAUSE:
287 05e50653 Michael Hanselmann
      (until, ) = args
288 05e50653 Michael Hanselmann
289 05e50653 Michael Hanselmann
      if until is None:
290 05e50653 Michael Hanselmann
        logging.info("Received request to no longer pause the watcher")
291 05e50653 Michael Hanselmann
      else:
292 05e50653 Michael Hanselmann
        if not isinstance(until, (int, float)):
293 05e50653 Michael Hanselmann
          raise TypeError("Duration must be an integer or float")
294 05e50653 Michael Hanselmann
295 05e50653 Michael Hanselmann
        if until < time.time():
296 05e50653 Michael Hanselmann
          raise errors.GenericError("Unable to set pause end time in the past")
297 05e50653 Michael Hanselmann
298 05e50653 Michael Hanselmann
        logging.info("Received request to pause the watcher until %s", until)
299 05e50653 Michael Hanselmann
300 05e50653 Michael Hanselmann
      return _SetWatcherPause(until)
301 05e50653 Michael Hanselmann
302 0bbe448c Michael Hanselmann
    else:
303 e566ddbd Iustin Pop
      logging.info("Received invalid request '%s'", method)
304 e566ddbd Iustin Pop
      raise ValueError("Invalid operation '%s'" % method)
305 ffeffa1d Iustin Pop
306 ee6c7b94 Michael Hanselmann
  def _Query(self, op):
307 ee6c7b94 Michael Hanselmann
    """Runs the specified opcode and returns the result.
308 ee6c7b94 Michael Hanselmann
309 ee6c7b94 Michael Hanselmann
    """
310 adfa97e3 Guido Trotter
    # Queries don't have a job id
311 adfa97e3 Guido Trotter
    proc = mcpu.Processor(self.server.context, None)
312 031a3e57 Michael Hanselmann
    return proc.ExecOpCode(op, None)
313 ee6c7b94 Michael Hanselmann
314 ffeffa1d Iustin Pop
315 39dcf2ef Guido Trotter
class GanetiContext(object):
316 39dcf2ef Guido Trotter
  """Context common to all ganeti threads.
317 39dcf2ef Guido Trotter
318 39dcf2ef Guido Trotter
  This class creates and holds common objects shared by all threads.
319 39dcf2ef Guido Trotter
320 39dcf2ef Guido Trotter
  """
321 7260cfbe Iustin Pop
  # pylint: disable-msg=W0212
322 7260cfbe Iustin Pop
  # we do want to ensure a singleton here
323 39dcf2ef Guido Trotter
  _instance = None
324 39dcf2ef Guido Trotter
325 39dcf2ef Guido Trotter
  def __init__(self):
326 39dcf2ef Guido Trotter
    """Constructs a new GanetiContext object.
327 39dcf2ef Guido Trotter
328 39dcf2ef Guido Trotter
    There should be only a GanetiContext object at any time, so this
329 39dcf2ef Guido Trotter
    function raises an error if this is not the case.
330 39dcf2ef Guido Trotter
331 39dcf2ef Guido Trotter
    """
332 39dcf2ef Guido Trotter
    assert self.__class__._instance is None, "double GanetiContext instance"
333 39dcf2ef Guido Trotter
334 9113300d Michael Hanselmann
    # Create global configuration object
335 39dcf2ef Guido Trotter
    self.cfg = config.ConfigWriter()
336 9113300d Michael Hanselmann
337 9113300d Michael Hanselmann
    # Locking manager
338 984f7c32 Guido Trotter
    self.glm = locking.GanetiLockManager(
339 39dcf2ef Guido Trotter
                self.cfg.GetNodeList(),
340 39dcf2ef Guido Trotter
                self.cfg.GetInstanceList())
341 39dcf2ef Guido Trotter
342 9113300d Michael Hanselmann
    # Job queue
343 9113300d Michael Hanselmann
    self.jobqueue = jqueue.JobQueue(self)
344 9113300d Michael Hanselmann
345 39dcf2ef Guido Trotter
    # setting this also locks the class against attribute modifications
346 39dcf2ef Guido Trotter
    self.__class__._instance = self
347 39dcf2ef Guido Trotter
348 39dcf2ef Guido Trotter
  def __setattr__(self, name, value):
349 39dcf2ef Guido Trotter
    """Setting GanetiContext attributes is forbidden after initialization.
350 39dcf2ef Guido Trotter
351 39dcf2ef Guido Trotter
    """
352 39dcf2ef Guido Trotter
    assert self.__class__._instance is None, "Attempt to modify Ganeti Context"
353 39dcf2ef Guido Trotter
    object.__setattr__(self, name, value)
354 39dcf2ef Guido Trotter
355 0debfb35 Guido Trotter
  def AddNode(self, node, ec_id):
356 d8470559 Michael Hanselmann
    """Adds a node to the configuration and lock manager.
357 d8470559 Michael Hanselmann
358 d8470559 Michael Hanselmann
    """
359 d8470559 Michael Hanselmann
    # Add it to the configuration
360 0debfb35 Guido Trotter
    self.cfg.AddNode(node, ec_id)
361 d8470559 Michael Hanselmann
362 c36176cc Michael Hanselmann
    # If preseeding fails it'll not be added
363 99aabbed Iustin Pop
    self.jobqueue.AddNode(node)
364 c36176cc Michael Hanselmann
365 d8470559 Michael Hanselmann
    # Add the new node to the Ganeti Lock Manager
366 d8470559 Michael Hanselmann
    self.glm.add(locking.LEVEL_NODE, node.name)
367 d8470559 Michael Hanselmann
368 d8470559 Michael Hanselmann
  def ReaddNode(self, node):
369 d8470559 Michael Hanselmann
    """Updates a node that's already in the configuration
370 d8470559 Michael Hanselmann
371 d8470559 Michael Hanselmann
    """
372 c36176cc Michael Hanselmann
    # Synchronize the queue again
373 99aabbed Iustin Pop
    self.jobqueue.AddNode(node)
374 d8470559 Michael Hanselmann
375 d8470559 Michael Hanselmann
  def RemoveNode(self, name):
376 d8470559 Michael Hanselmann
    """Removes a node from the configuration and lock manager.
377 d8470559 Michael Hanselmann
378 d8470559 Michael Hanselmann
    """
379 d8470559 Michael Hanselmann
    # Remove node from configuration
380 d8470559 Michael Hanselmann
    self.cfg.RemoveNode(name)
381 d8470559 Michael Hanselmann
382 c36176cc Michael Hanselmann
    # Notify job queue
383 c36176cc Michael Hanselmann
    self.jobqueue.RemoveNode(name)
384 c36176cc Michael Hanselmann
385 d8470559 Michael Hanselmann
    # Remove the node from the Ganeti Lock Manager
386 d8470559 Michael Hanselmann
    self.glm.remove(locking.LEVEL_NODE, name)
387 d8470559 Michael Hanselmann
388 39dcf2ef Guido Trotter
389 05e50653 Michael Hanselmann
def _SetWatcherPause(until):
390 05e50653 Michael Hanselmann
  """Creates or removes the watcher pause file.
391 05e50653 Michael Hanselmann
392 05e50653 Michael Hanselmann
  @type until: None or int
393 05e50653 Michael Hanselmann
  @param until: Unix timestamp saying until when the watcher shouldn't run
394 05e50653 Michael Hanselmann
395 05e50653 Michael Hanselmann
  """
396 05e50653 Michael Hanselmann
  if until is None:
397 05e50653 Michael Hanselmann
    utils.RemoveFile(constants.WATCHER_PAUSEFILE)
398 05e50653 Michael Hanselmann
  else:
399 05e50653 Michael Hanselmann
    utils.WriteFile(constants.WATCHER_PAUSEFILE,
400 05e50653 Michael Hanselmann
                    data="%d\n" % (until, ))
401 05e50653 Michael Hanselmann
402 28b498cd Michael Hanselmann
  return until
403 28b498cd Michael Hanselmann
404 05e50653 Michael Hanselmann
405 e0e916fe Iustin Pop
@rpc.RunWithRPC
406 36205981 Iustin Pop
def CheckAgreement():
407 36205981 Iustin Pop
  """Check the agreement on who is the master.
408 36205981 Iustin Pop
409 36205981 Iustin Pop
  The function uses a very simple algorithm: we must get more positive
410 36205981 Iustin Pop
  than negative answers. Since in most of the cases we are the master,
411 36205981 Iustin Pop
  we'll use our own config file for getting the node list. In the
412 36205981 Iustin Pop
  future we could collect the current node list from our (possibly
413 36205981 Iustin Pop
  obsolete) known nodes.
414 36205981 Iustin Pop
415 d7cdb55d Iustin Pop
  In order to account for cold-start of all nodes, we retry for up to
416 d7cdb55d Iustin Pop
  a minute until we get a real answer as the top-voted one. If the
417 d7cdb55d Iustin Pop
  nodes are more out-of-sync, for now manual startup of the master
418 d7cdb55d Iustin Pop
  should be attempted.
419 d7cdb55d Iustin Pop
420 d7cdb55d Iustin Pop
  Note that for a even number of nodes cluster, we need at least half
421 d7cdb55d Iustin Pop
  of the nodes (beside ourselves) to vote for us. This creates a
422 d7cdb55d Iustin Pop
  problem on two-node clusters, since in this case we require the
423 d7cdb55d Iustin Pop
  other node to be up too to confirm our status.
424 d7cdb55d Iustin Pop
425 36205981 Iustin Pop
  """
426 a744b676 Manuel Franceschini
  myself = netutils.HostInfo().name
427 36205981 Iustin Pop
  #temp instantiation of a config writer, used only to get the node list
428 36205981 Iustin Pop
  cfg = config.ConfigWriter()
429 36205981 Iustin Pop
  node_list = cfg.GetNodeList()
430 36205981 Iustin Pop
  del cfg
431 d7cdb55d Iustin Pop
  retries = 6
432 d7cdb55d Iustin Pop
  while retries > 0:
433 d7cdb55d Iustin Pop
    votes = bootstrap.GatherMasterVotes(node_list)
434 d7cdb55d Iustin Pop
    if not votes:
435 d7cdb55d Iustin Pop
      # empty node list, this is a one node cluster
436 d7cdb55d Iustin Pop
      return True
437 d7cdb55d Iustin Pop
    if votes[0][0] is None:
438 d7cdb55d Iustin Pop
      retries -= 1
439 d7cdb55d Iustin Pop
      time.sleep(10)
440 36205981 Iustin Pop
      continue
441 d7cdb55d Iustin Pop
    break
442 d7cdb55d Iustin Pop
  if retries == 0:
443 e09fdcfa Iustin Pop
    logging.critical("Cluster inconsistent, most of the nodes didn't answer"
444 e09fdcfa Iustin Pop
                     " after multiple retries. Aborting startup")
445 d8f5a37d Iustin Pop
    logging.critical("Use the --no-voting option if you understand what"
446 d8f5a37d Iustin Pop
                     " effects it has on the cluster state")
447 e09fdcfa Iustin Pop
    return False
448 d7cdb55d Iustin Pop
  # here a real node is at the top of the list
449 d7cdb55d Iustin Pop
  all_votes = sum(item[1] for item in votes)
450 d7cdb55d Iustin Pop
  top_node, top_votes = votes[0]
451 8a20c732 Michael Hanselmann
452 d7cdb55d Iustin Pop
  result = False
453 d7cdb55d Iustin Pop
  if top_node != myself:
454 d7cdb55d Iustin Pop
    logging.critical("It seems we are not the master (top-voted node"
455 bbe19c17 Iustin Pop
                     " is %s with %d out of %d votes)", top_node, top_votes,
456 bbe19c17 Iustin Pop
                     all_votes)
457 d7cdb55d Iustin Pop
  elif top_votes < all_votes - top_votes:
458 36205981 Iustin Pop
    logging.critical("It seems we are not the master (%d votes for,"
459 d7cdb55d Iustin Pop
                     " %d votes against)", top_votes, all_votes - top_votes)
460 d7cdb55d Iustin Pop
  else:
461 d7cdb55d Iustin Pop
    result = True
462 d7cdb55d Iustin Pop
463 d7cdb55d Iustin Pop
  return result
464 36205981 Iustin Pop
465 6c948699 Michael Hanselmann
466 340f4757 Iustin Pop
@rpc.RunWithRPC
467 340f4757 Iustin Pop
def ActivateMasterIP():
468 340f4757 Iustin Pop
  # activate ip
469 340f4757 Iustin Pop
  master_node = ssconf.SimpleStore().GetMasterNode()
470 340f4757 Iustin Pop
  result = rpc.RpcRunner.call_node_start_master(master_node, False, False)
471 340f4757 Iustin Pop
  msg = result.fail_msg
472 340f4757 Iustin Pop
  if msg:
473 340f4757 Iustin Pop
    logging.error("Can't activate master IP address: %s", msg)
474 340f4757 Iustin Pop
475 340f4757 Iustin Pop
476 ed0efaa5 Michael Hanselmann
def CheckMasterd(options, args):
477 ed0efaa5 Michael Hanselmann
  """Initial checks whether to run or exit with a failure.
478 ed0efaa5 Michael Hanselmann
479 ed0efaa5 Michael Hanselmann
  """
480 f93427cd Iustin Pop
  if args: # masterd doesn't take any arguments
481 f93427cd Iustin Pop
    print >> sys.stderr, ("Usage: %s [-f] [-d]" % sys.argv[0])
482 f93427cd Iustin Pop
    sys.exit(constants.EXIT_FAILURE)
483 f93427cd Iustin Pop
484 ed0efaa5 Michael Hanselmann
  ssconf.CheckMaster(options.debug)
485 ed0efaa5 Michael Hanselmann
486 bbfd0568 René Nussbaumer
  try:
487 bbfd0568 René Nussbaumer
    options.uid = pwd.getpwnam(constants.MASTERD_USER).pw_uid
488 bbfd0568 René Nussbaumer
    options.gid = grp.getgrnam(constants.DAEMONS_GROUP).gr_gid
489 bbfd0568 René Nussbaumer
  except KeyError:
490 bbfd0568 René Nussbaumer
    print >> sys.stderr, ("User or group not existing on system: %s:%s" %
491 bbfd0568 René Nussbaumer
                          (constants.MASTERD_USER, constants.DAEMONS_GROUP))
492 bbfd0568 René Nussbaumer
    sys.exit(constants.EXIT_FAILURE)
493 bbfd0568 René Nussbaumer
494 bbfd0568 René Nussbaumer
495 ed0efaa5 Michael Hanselmann
  # If CheckMaster didn't fail we believe we are the master, but we have to
496 ed0efaa5 Michael Hanselmann
  # confirm with the other nodes.
497 ed0efaa5 Michael Hanselmann
  if options.no_voting:
498 ed0efaa5 Michael Hanselmann
    if options.yes_do_it:
499 ed0efaa5 Michael Hanselmann
      return
500 ed0efaa5 Michael Hanselmann
501 ed0efaa5 Michael Hanselmann
    sys.stdout.write("The 'no voting' option has been selected.\n")
502 ed0efaa5 Michael Hanselmann
    sys.stdout.write("This is dangerous, please confirm by"
503 ed0efaa5 Michael Hanselmann
                     " typing uppercase 'yes': ")
504 ed0efaa5 Michael Hanselmann
    sys.stdout.flush()
505 ed0efaa5 Michael Hanselmann
506 ed0efaa5 Michael Hanselmann
    confirmation = sys.stdin.readline().strip()
507 ed0efaa5 Michael Hanselmann
    if confirmation != "YES":
508 7260cfbe Iustin Pop
      print >> sys.stderr, "Aborting."
509 ed0efaa5 Michael Hanselmann
      sys.exit(constants.EXIT_FAILURE)
510 ed0efaa5 Michael Hanselmann
511 ed0efaa5 Michael Hanselmann
    return
512 ed0efaa5 Michael Hanselmann
513 ed0efaa5 Michael Hanselmann
  # CheckAgreement uses RPC and threads, hence it needs to be run in a separate
514 ed0efaa5 Michael Hanselmann
  # process before we call utils.Daemonize in the current process.
515 e0e916fe Iustin Pop
  if not utils.RunInSeparateProcess(CheckAgreement):
516 ed0efaa5 Michael Hanselmann
    sys.exit(constants.EXIT_FAILURE)
517 ed0efaa5 Michael Hanselmann
518 340f4757 Iustin Pop
  # ActivateMasterIP also uses RPC/threads, so we run it again via a
519 340f4757 Iustin Pop
  # separate process.
520 340f4757 Iustin Pop
521 340f4757 Iustin Pop
  # TODO: decide whether failure to activate the master IP is a fatal error
522 340f4757 Iustin Pop
  utils.RunInSeparateProcess(ActivateMasterIP)
523 340f4757 Iustin Pop
524 ed0efaa5 Michael Hanselmann
525 33231500 Michael Hanselmann
def ExecMasterd(options, args): # pylint: disable-msg=W0613
526 6c948699 Michael Hanselmann
  """Main master daemon function, executed with the PID file held.
527 3b316acb Iustin Pop
528 04ccf5e9 Guido Trotter
  """
529 04ccf5e9 Guido Trotter
  # This is safe to do as the pid file guarantees against
530 04ccf5e9 Guido Trotter
  # concurrent execution.
531 04ccf5e9 Guido Trotter
  utils.RemoveFile(constants.MASTER_SOCKET)
532 b1b6ea87 Iustin Pop
533 cdd7f900 Guido Trotter
  mainloop = daemon.Mainloop()
534 7e5a6e86 Guido Trotter
  master = MasterServer(mainloop, constants.MASTER_SOCKET,
535 bbfd0568 René Nussbaumer
                        options.uid, options.gid)
536 04ccf5e9 Guido Trotter
  try:
537 15486fa7 Michael Hanselmann
    rpc.Init()
538 4331f6cd Michael Hanselmann
    try:
539 15486fa7 Michael Hanselmann
      master.setup_queue()
540 15486fa7 Michael Hanselmann
      try:
541 cdd7f900 Guido Trotter
        mainloop.Run()
542 15486fa7 Michael Hanselmann
      finally:
543 15486fa7 Michael Hanselmann
        master.server_cleanup()
544 4331f6cd Michael Hanselmann
    finally:
545 15486fa7 Michael Hanselmann
      rpc.Shutdown()
546 a4af651e Iustin Pop
  finally:
547 227647ac Guido Trotter
    utils.RemoveFile(constants.MASTER_SOCKET)
548 a4af651e Iustin Pop
549 ffeffa1d Iustin Pop
550 04ccf5e9 Guido Trotter
def main():
551 04ccf5e9 Guido Trotter
  """Main function"""
552 04ccf5e9 Guido Trotter
  parser = OptionParser(description="Ganeti master daemon",
553 04ccf5e9 Guido Trotter
                        usage="%prog [-f] [-d]",
554 04ccf5e9 Guido Trotter
                        version="%%prog (ganeti) %s" %
555 04ccf5e9 Guido Trotter
                        constants.RELEASE_VERSION)
556 04ccf5e9 Guido Trotter
  parser.add_option("--no-voting", dest="no_voting",
557 04ccf5e9 Guido Trotter
                    help="Do not check that the nodes agree on this node"
558 04ccf5e9 Guido Trotter
                    " being the master and start the daemon unconditionally",
559 04ccf5e9 Guido Trotter
                    default=False, action="store_true")
560 04ccf5e9 Guido Trotter
  parser.add_option("--yes-do-it", dest="yes_do_it",
561 04ccf5e9 Guido Trotter
                    help="Override interactive check for --no-voting",
562 04ccf5e9 Guido Trotter
                    default=False, action="store_true")
563 04ccf5e9 Guido Trotter
  dirs = [(constants.RUN_GANETI_DIR, constants.RUN_DIRS_MODE),
564 04ccf5e9 Guido Trotter
          (constants.SOCKET_DIR, constants.SOCKET_DIR_MODE),
565 04ccf5e9 Guido Trotter
         ]
566 04ccf5e9 Guido Trotter
  daemon.GenericMain(constants.MASTERD, parser, dirs,
567 30dabd03 Michael Hanselmann
                     CheckMasterd, ExecMasterd,
568 30dabd03 Michael Hanselmann
                     multithreaded=True)
569 6c948699 Michael Hanselmann
570 04ccf5e9 Guido Trotter
571 ffeffa1d Iustin Pop
if __name__ == "__main__":
572 ffeffa1d Iustin Pop
  main()