src/collecty/daemon.py

   1 #!/usr/bin/python3
   2 ###############################################################################
   3 #                                                                             #
   4 # collecty - A system statistics collection daemon for IPFire                 #
   5 # Copyright (C) 2012 IPFire development team                                  #
   6 #                                                                             #
   7 # This program is free software: you can redistribute it and/or modify        #
   8 # it under the terms of the GNU General Public License as published by        #
   9 # the Free Software Foundation, either version 3 of the License, or           #
  10 # (at your option) any later version.                                         #
  11 #                                                                             #
  12 # This program is distributed in the hope that it will be useful,             #
  13 # but WITHOUT ANY WARRANTY; without even the implied warranty of              #
  14 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               #
  15 # GNU General Public License for more details.                                #
  16 #                                                                             #
  17 # You should have received a copy of the GNU General Public License           #
  18 # along with this program.  If not, see <http://www.gnu.org/licenses/>.       #
  19 #                                                                             #
  20 ###############################################################################
  21
  22 import datetime
  23 import multiprocessing
  24 import os
  25 import queue
  26 import rrdtool
  27 import signal
  28 import threading
  29 import time
  30
  31 from . import bus
  32 from . import plugins
  33
  34 from .constants import *
  35 from .i18n import _
  36
  37 import logging
  38 log = logging.getLogger("collecty")
  39
  40 class Collecty(object):
  41         # The default interval, when all data is written to disk.
  42         SUBMIT_INTERVAL = 300
  43
  44         HEARTBEAT = 1
  45
  46         def __init__(self, debug=False):
  47                 self.debug = debug
  48
  49                 # Reset timezone to UTC
  50                 # rrdtool is reading that from the environment
  51                 os.environ["TZ"] = "UTC"
  52
  53                 # Enable debug logging when running in debug mode
  54                 if self.debug:
  55                         log.setLevel(logging.DEBUG)
  56
  57                 self.plugins = []
  58
  59                 # Indicates whether this process should be running or not.
  60                 self.running = True
  61
  62                 # The write queue holds all collected pieces of data which
  63                 # will be written to disk later.
  64                 self.write_queue = WriteQueue(self, self.SUBMIT_INTERVAL)
  65
  66                 # Create worker threads
  67                 self.worker_threads = self.create_worker_threads()
  68
  69                 self._timer_queue = queue.PriorityQueue()
  70                 self._worker_queue = queue.Queue()
  71
  72                 # Create a thread that connects to dbus and processes requests we
  73                 # get from there.
  74                 self.bus = bus.Bus(self)
  75
  76                 # Add all plugins
  77                 for plugin in plugins.get():
  78                         self.add_plugin(plugin)
  79
  80                 log.debug(_("Collecty successfully initialized with %s plugins") \
  81                         % len(self.plugins))
  82
  83         def add_plugin(self, plugin_class):
  84                 # Try initialising a new plugin. If that fails, we will log the
  85                 # error and try to go on.
  86                 try:
  87                         plugin = plugin_class(self)
  88                 except:
  89                         log.critical(_("Plugin %s could not be initialised") % plugin_class, exc_info=True)
  90                         return
  91
  92                 self.plugins.append(plugin)
  93
  94         @property
  95         def templates(self):
  96                 for plugin in self.plugins:
  97                         for template in plugin.templates:
  98                                 yield template
  99
 100         def run(self):
 101                 # Register signal handlers.
 102                 self.register_signal_handler()
 103
 104                 # Cannot do anything if no plugins have been initialised
 105                 if not self.plugins:
 106                         log.critical(_("No plugins have been initialised"))
 107                         return
 108
 109                 # Start the bus
 110                 self.bus.start()
 111
 112                 # Initialise the timer queue
 113                 self.initialise_timer_queue()
 114
 115                 # Start worker threads
 116                 for w in self.worker_threads:
 117                         w.start()
 118
 119                 # Run the write queue thread
 120                 self.write_queue.start()
 121
 122                 # Regularly submit all data to disk.
 123                 while self.running:
 124                         try:
 125                                 # Try processing one event from the queue. If that succeeded
 126                                 # we will retry immediately.
 127                                 if self.process_timer_queue():
 128                                         continue
 129
 130                                 # Otherwise we will sleep for a bit
 131                                 time.sleep(self.HEARTBEAT)
 132
 133                                 # Log warnings if the worker queue is filling up
 134                                 queue_size = self._worker_queue.qsize()
 135                                 if queue_size >= 5:
 136                                         log.warning(_("Worker queue is filling up with %s events") % queue_size)
 137
 138                         except KeyboardInterrupt:
 139                                 self.shutdown()
 140                                 break
 141
 142                 # Wait until all worker threads are finished
 143                 for w in self.worker_threads:
 144                         w.join()
 145
 146                 # Stop the bus thread
 147                 self.bus.shutdown()
 148
 149                 # Write all collected data to disk before ending the main thread
 150                 self.write_queue.shutdown()
 151
 152                 log.debug(_("Main thread exited"))
 153
 154         def shutdown(self):
 155                 if not self.running:
 156                         return
 157
 158                 log.info(_("Received shutdown signal"))
 159                 self.running = False
 160
 161                 # Propagating shutdown to all threads.
 162                 for w in self.worker_threads:
 163                         w.shutdown()
 164
 165         def register_signal_handler(self):
 166                 for s in (signal.SIGTERM, signal.SIGINT, signal.SIGUSR1):
 167                         log.debug(_("Registering signal %d") % s)
 168
 169                 signal.signal(s, self.signal_handler)
 170
 171         def signal_handler(self, sig, *args, **kwargs):
 172                 log.info(_("Caught signal %d") % sig)
 173
 174                 if sig in (signal.SIGTERM, signal.SIGINT):
 175                         # Shutdown this application.
 176                         self.shutdown()
 177
 178                 elif sig == signal.SIGUSR1:
 179                         # Commit all data.
 180                         self.write_queue.commit()
 181
 182         def get_plugin_from_template(self, template_name):
 183                 for plugin in self.plugins:
 184                         if not template_name in [t.name for t in plugin.templates]:
 185                                 continue
 186
 187                         return plugin
 188
 189         def generate_graph(self, template_name, *args, **kwargs):
 190                 plugin = self.get_plugin_from_template(template_name)
 191                 if not plugin:
 192                         raise RuntimeError("Could not find template %s" % template_name)
 193
 194                 return plugin.generate_graph(template_name, *args, **kwargs)
 195
 196         def create_worker_threads(self, num=None):
 197                 """
 198                         Creates a number of worker threads
 199                 """
 200                 # If no number of threads is given, we will create as many as we have
 201                 # active processor cores but never less than two.
 202                 if num is None:
 203                         num = max(multiprocessing.cpu_count(), 2)
 204
 205                 worker_threads = []
 206
 207                 for id in range(num):
 208                         worker_thread = WorkerThread(self, id)
 209                         worker_threads.append(worker_thread)
 210
 211                 return worker_threads
 212
 213         def initialise_timer_queue(self):
 214                 for p in self.plugins:
 215                         timer = PluginTimer(p)
 216
 217                         self._timer_queue.put(timer)
 218
 219         def process_timer_queue(self):
 220                 # Take the item from the timer queue that is to be due first
 221                 timer = self._timer_queue.get()
 222
 223                 try:
 224                         # If the timer event is to be executed, we will put the plugin
 225                         # into the worker queue and reset the timer
 226                         if timer.is_due():
 227                                 self._worker_queue.put(timer.plugin)
 228                                 timer.reset_deadline()
 229
 230                                 return timer
 231                 finally:
 232                         # Put the timer back into the timer queue.
 233                         self._timer_queue.put(timer)
 234
 235
 236 class WorkerThread(threading.Thread):
 237         HEARTBEAT = 2.5
 238
 239         def __init__(self, collecty, id):
 240                 threading.Thread.__init__(self)
 241                 self.daemon = True
 242
 243                 self.log = logging.getLogger("collecty.worker")
 244                 self.log.propagate = 1
 245
 246                 self.collecty = collecty
 247                 self.id = id
 248
 249                 self.log.debug(_("Worker thread %s has been initialised") % self.id)
 250
 251         @property
 252         def queue(self):
 253                 """
 254                         The queue this thread is getting events from
 255                 """
 256                 return self.collecty._worker_queue
 257
 258         def run(self):
 259                 self.log.debug(_("Worker thread %s has been started") % self.id)
 260                 self.running = True
 261
 262                 while self.running:
 263                         try:
 264                                 plugin = self.queue.get(block=True, timeout=self.HEARTBEAT)
 265
 266                         # If the queue has been empty we just retry
 267                         except queue.Empty:
 268                                 continue
 269
 270                         # Execute the collect operation for this plugin
 271                         plugin.collect()
 272
 273                 self.log.debug(_("Worker thread %s has been terminated") % self.id)
 274
 275         def shutdown(self):
 276                 self.running = False
 277
 278
 279 class WriteQueue(threading.Thread):
 280         def __init__(self, collecty, submit_interval):
 281                 threading.Thread.__init__(self)
 282                 self.daemon = True
 283
 284                 self.collecty = collecty
 285
 286                 self.log = logging.getLogger("collecty.queue")
 287                 self.log.propagate = 1
 288
 289                 self.timer = plugins.Timer(submit_interval)
 290                 self._queue = queue.PriorityQueue()
 291
 292                 self.log.debug(_("Initialised write queue"))
 293
 294         def run(self):
 295                 self.log.debug(_("Write queue process started"))
 296                 self.running = True
 297
 298                 while self.running:
 299                         # Reset the timer.
 300                         self.timer.reset()
 301
 302                         # Wait until the timer has successfully elapsed.
 303                         if self.timer.wait():
 304                                 self.commit()
 305
 306                 self.commit()
 307                 self.log.debug(_("Write queue process stopped"))
 308
 309         def shutdown(self):
 310                 self.running = False
 311                 self.timer.cancel()
 312
 313                 # Wait until all data has been written.
 314                 self.join()
 315
 316         def add(self, object, time, data):
 317                 result = QueueObject(object.file, time, data)
 318                 self._queue.put(result)
 319
 320         def commit(self):
 321                 """
 322                         Flushes the read data to disk.
 323                 """
 324                 # There is nothing to do if the queue is empty
 325                 if self._queue.empty():
 326                         self.log.debug(_("No data to commit"))
 327                         return
 328
 329                 time_start = time.time()
 330
 331                 self.log.debug(_("Submitting data to the databases..."))
 332
 333                 # Get all objects from the queue and group them by the RRD file
 334                 # to commit them all at once
 335                 results = {}
 336                 while not self._queue.empty():
 337                         result = self._queue.get()
 338
 339                         try:
 340                                 results[result.file].append(result)
 341                         except KeyError:
 342                                 results[result.file] = [result]
 343
 344                 # Write the collected data to disk
 345                 for filename, results in list(results.items()):
 346                         self._commit_file(filename, results)
 347
 348                 duration = time.time() - time_start
 349                 self.log.debug(_("Emptied write queue in %.2fs") % duration)
 350
 351         def _commit_file(self, filename, results):
 352                 self.log.debug(_("Committing %(counter)s entries to %(filename)s") \
 353                         % { "counter" : len(results), "filename" : filename })
 354
 355                 for result in results:
 356                         self.log.debug("  %s: %s" % (result.time, result.data))
 357
 358                 try:
 359                         rrdtool.update(filename, *["%s" % r for r in results])
 360
 361                 # Catch operational errors like unreadable/unwritable RRD databases
 362                 # or those where the format has changed. The collected data will be lost.
 363                 except rrdtool.OperationalError as e:
 364                         self.log.critical(_("Could not update RRD database %s: %s") \
 365                                 % (filename, e))
 366
 367
 368 class QueueObject(object):
 369         def __init__(self, file, time, data):
 370                 self.file = file
 371                 self.time = time
 372                 self.data = data
 373
 374         def __str__(self):
 375                 return "%s:%s" % (self.time.strftime("%s"), self.data)
 376
 377         def __lt__(self, other):
 378                 return self.time < other.time
 379
 380
 381 class PluginTimer(object):
 382         def __init__(self, plugin):
 383                 self.plugin = plugin
 384
 385                 self.deadline = datetime.datetime.utcnow()
 386
 387         def __repr__(self):
 388                 return "<%s %s>" % (self.__class__.__name__, self.deadline)
 389
 390         def __lt__(self, other):
 391                 return self.deadline < other.deadline
 392
 393         def reset_deadline(self):
 394                 self.deadline = datetime.datetime.utcnow() \
 395                         + datetime.timedelta(seconds=self.plugin.interval)
 396
 397         def is_due(self):
 398                 return datetime.datetime.utcnow() >= self.deadline