htsworkflow/automation/spoolwatcher.py

   1 #!/usr/bin/env python
   2 import logging
   3 import os
   4 import re
   5 import shlex
   6 import sys
   7 import time
   8
   9 from htsworkflow.util import mount
  10
  11 # this uses pyinotify
  12 import pyinotify
  13 from pyinotify import EventsCodes
  14
  15 from benderjab import rpc
  16
  17 def is_runfolder(name):
  18     """
  19     Is it a runfolder?
  20
  21     >>> print is_runfolder('090630_HWUSI-EAS999_0006_30LNFAAXX')
  22     True
  23     >>> print is_runfolder('hello')
  24     False
  25     """
  26     if re.match("[0-9]{6}_.*", name):
  27         return True
  28     else:
  29         return False
  30
  31 def get_top_dir(root, path):
  32     """
  33     Return the directory in path that is a subdirectory of root.
  34     e.g.
  35
  36     >>> print get_top_dir('/a/b/c', '/a/b/c/d/e/f')
  37     d
  38     >>> print get_top_dir('/a/b/c/', '/a/b/c/d/e/f')
  39     d
  40     >>> print get_top_dir('/a/b/c', '/g/e/f')
  41     None
  42     >>> print get_top_dir('/a/b/c', '/a/b/c')
  43     <BLANKLINE>
  44     """
  45     if path.startswith(root):
  46         subpath = path[len(root):]
  47         if subpath.startswith('/'):
  48             subpath = subpath[1:]
  49         return subpath.split(os.path.sep)[0]
  50     else:
  51         return None
  52
  53 class WatcherEvent(object):
  54     """
  55     Track information about a file event
  56
  57     Currently its time, and if it was an indication we've completed the run.
  58     """
  59     def __init__(self, event_root=None):
  60         self.time = time.time()
  61         self.event_root = event_root
  62         self.complete = False
  63
  64     def __unicode__(self):
  65         if self.complete:
  66            complete = "(completed)"
  67         else:
  68            complete = ""
  69         return u"<WatchEvent: %s %s %s>" % (time.ctime(self.time), self.event_root, complete)
  70
  71 class Handler(pyinotify.ProcessEvent):
  72     def __init__(self, watchmanager, bot, completion_file=None):
  73         """
  74         Completion file contains current "completion" filename
  75         """
  76         self.last_event = {}
  77         self.watchmanager = watchmanager
  78         self.bot = bot
  79         if completion_file is not None:
  80             completion_file = completion_file.lower()
  81         self.completion_file = completion_file
  82
  83     def process_IN_CREATE(self, event):
  84         for wdd in self.bot.wdds:
  85             for watch_path in self.bot.watchdirs:
  86                 run_already_complete = False
  87                 # I only care about things created inside the watch directory, so
  88                 # the event path needs to be longer than the watch path in addition to
  89                 # starting with the watch_path
  90                 if len(event.path) > len(watch_path) and event.path.startswith(watch_path):
  91                     # compute name of the top level directory that had an event
  92                     # in the current watch path
  93                     target = get_top_dir(watch_path, event.path)
  94                     runfolder = os.path.join(watch_path, target)
  95
  96                     if not is_runfolder(target):
  97                         logging.debug("Skipping %s, not a runfolder" % (target,))
  98                         continue
  99
 100                     # grab the previous events for this watch path
 101                     watch_path_events = self.last_event.setdefault(watch_path, {})
 102
 103                     # if we've already seen an event in this directory (AKA runfolder)
 104                     # keep track if its already hit the "completed" flag
 105                     if watch_path_events.has_key(target):
 106                        run_already_complete = watch_path_events[target].complete
 107
 108                     watch_path_events[target] = WatcherEvent(target)
 109                     #self.last_event.setdefault(watch_path, {})[target] = WatcherEvent(target)
 110
 111                     msg = "Create: %s %s %s %s" % (watch_path, target, event.path, event.name)
 112
 113                     # the ReadPrep step uses some of the same file completion flags as the
 114                     # main analysis, which means this completion code might get tripped because of it
 115                     # so we need to make sure we're getting the completion file in the root of the
 116                     # runfolder
 117                     if (self.completion_file == event.name.lower() and event.path == runfolder) \
 118                       or run_already_complete:
 119                         self.last_event[watch_path][target].complete = True
 120                         msg += "(completed)"
 121
 122                     logging.debug(msg)
 123
 124     def process_IN_DELETE(self, event):
 125         logging.debug("Remove: %s" %  os.path.join(event.path, event.name))
 126         pass
 127
 128     def process_IN_UNMOUNT(self, event):
 129         pathname = os.path.join(event.path, event.name)
 130         logging.debug("IN_UNMOUNT: %s" % (pathname,))
 131         self.bot.unmount_watch(event.path)
 132
 133 class SpoolWatcher(rpc.XmlRpcBot):
 134     """
 135     Watch a directory and send a message when another process is done writing.
 136
 137     This monitors a directory tree using inotify (linux specific) and
 138     after some files having been written will send a message after <timeout>
 139     seconds of no file writing.
 140
 141     (Basically when the solexa machine finishes dumping a round of data
 142     this'll hopefully send out a message saying hey look theres data available
 143
 144     """
 145     # these params need to be in the config file
 146     # I wonder where I should put the documentation
 147     #:Parameters:
 148     #    `watchdirs` - list of directories to monitor for modifications
 149     #    `profile` - specify which .htsworkflow profile to use
 150     #    `write_timeout` - how many seconds to wait for writes to finish to
 151     #                      the spool
 152     #    `notify_timeout` - how often to timeout from notify
 153     #    `completion_file` - what file indicates we've finished sequencing
 154     #                        defaults to: netcopy_complete.txt
 155
 156     def __init__(self, section=None, configfile=None):
 157         #if configfile is None:
 158         #    self.configfile = "~/.htsworkflow"
 159         super(SpoolWatcher, self).__init__(section, configfile)
 160
 161         self.cfg['watchdirs'] = None
 162         self.cfg['write_timeout'] = 10
 163         self.cfg['notify_users'] = None
 164         self.cfg['notify_runner'] = None
 165         self.cfg['completion_file'] = 'netcopy_complete.txt'
 166
 167         self.watchdirs = []
 168         self.watchdir_url_map = {}
 169         self.notify_timeout = 0.001
 170
 171         self.wm = None
 172         self.notify_users = None
 173         self.notify_runner = None
 174         self.wdds = []
 175
 176         # keep track if the specified mount point is currently mounted
 177         self.mounted_points = {}
 178         # keep track of which mount points tie to which watch directories
 179         # so maybe we can remount them.
 180         self.mounts_to_watches = {}
 181
 182         self.eventTasks.append(self.process_notify)
 183
 184     def read_config(self, section=None, configfile=None):
 185         # Don't give in to the temptation to use logging functions here,
 186         # need to wait until after we detach in start
 187         super(SpoolWatcher, self).read_config(section, configfile)
 188
 189         self.watchdirs = shlex.split(self._check_required_option('watchdirs'))
 190         # see if there's an alternate url that should be used for the watchdir
 191         for watchdir in self.watchdirs:
 192             self.watchdir_url_map[watchdir] = self.cfg.get(watchdir, watchdir)
 193
 194         self.write_timeout = int(self.cfg['write_timeout'])
 195         self.completion_file = self.cfg['completion_file']
 196
 197         self.notify_users = self._parse_user_list(self.cfg['notify_users'])
 198         try:
 199           self.notify_runner = \
 200              self._parse_user_list(self.cfg['notify_runner'],
 201                                    require_resource=True)
 202         except bot.JIDMissingResource:
 203             msg = 'need a full jabber ID + resource for xml-rpc destinations'
 204             raise bot.JIDMissingResource(msg)
 205
 206         self.handler = None
 207         self.notifier = None
 208
 209     def add_watch(self, watchdirs=None):
 210         """
 211         start watching watchdir or self.watchdir
 212         we're currently limited to watching one directory tree.
 213         """
 214         # create the watch managers if we need them
 215         if self.wm is None:
 216             self.wm = pyinotify.WatchManager()
 217             self.handler = Handler(self.wm, self, self.completion_file)
 218             self.notifier = pyinotify.Notifier(self.wm, self.handler)
 219
 220         # the one tree limit is mostly because self.wdd is a single item
 221         # but managing it as a list might be a bit more annoying
 222         if watchdirs is None:
 223             watchdirs = self.watchdirs
 224
 225         mask = EventsCodes.IN_CREATE | EventsCodes.IN_UNMOUNT
 226         # rec traverses the tree and adds all the directories that are there
 227         # at the start.
 228         # auto_add will add in new directories as they are created
 229         for w in watchdirs:
 230             mount_location = mount.find_mount_point_for(w)
 231             self.mounted_points[mount_location] = True
 232             mounts = self.mounts_to_watches.get(mount_location, [])
 233             if w not in mounts:
 234                 mounts.append(w)
 235                 self.mounts_to_watches[mount_location] = mounts
 236
 237             logging.info(u"Watching:"+unicode(w))
 238             self.wdds.append(self.wm.add_watch(w, mask, rec=True, auto_add=True))
 239
 240     def unmount_watch(self, event_path):
 241         # remove backwards so we don't get weirdness from
 242         # the list getting shorter
 243         for i in range(len(self.wdds),0, -1):
 244             wdd = self.wdds[i]
 245             logging.info(u'unmounting: '+unicode(wdd.items()))
 246             self.wm.rm_watch(wdd.values())
 247             del self.wdds[i]
 248         self.mounted = False
 249
 250     def make_copy_url(self, watchdir, list_event_dir):
 251         root_copy_url = self.watchdir_url_map[watchdir]
 252         if root_copy_url[-1] != '/':
 253             root_copy_url += '/'
 254         copy_url = root_copy_url + list_event_dir
 255         logging.debug('Copy url: %s' % (copy_url,))
 256         return copy_url
 257
 258     def process_notify(self, *args):
 259         if self.notifier is None:
 260             # nothing to do yet
 261             return
 262         # process the queue of events as explained above
 263         self.notifier.process_events()
 264         #check events waits timeout
 265         if self.notifier.check_events(self.notify_timeout):
 266             # read notified events and enqeue them
 267             self.notifier.read_events()
 268             # should we do something?
 269         # has something happened?
 270         for watchdir, last_events in self.handler.last_event.items():
 271             for last_event_dir, last_event_detail in last_events.items():
 272                 time_delta = time.time() - last_event_detail.time
 273                 if time_delta > self.write_timeout:
 274                     print "timeout", unicode(last_event_detail)
 275                     copy_url = self.make_copy_url(watchdir, last_event_dir)
 276                     self.startCopy(copy_url)
 277                     if last_event_detail.complete:
 278                         self.sequencingFinished(last_event_detail.event_root)
 279
 280                     self.handler.last_event[watchdir] = {}
 281         # handle unmounted filesystems
 282         for mount_point, was_mounted in self.mounted_points.items():
 283             if not was_mounted and mount.is_mounted(mount_point):
 284                 # we've been remounted. Huzzah!
 285                 # restart the watch
 286                 for watch in self.mounts_to_watches[mount_point]:
 287                     self.add_watch(watch)
 288                     logging.info(
 289                         "%s was remounted, restarting watch" % \
 290                             (mount_point)
 291                     )
 292                 self.mounted_points[mount_point] = True
 293
 294     def _parser(self, msg, who):
 295         """
 296         Parse xmpp chat messages
 297         """
 298         help = u"I can send [copy] message, or squencer [finished]"
 299         if re.match(u"help", msg):
 300             reply = help
 301         elif re.match("copy", msg):
 302             self.startCopy(msg)
 303             reply = u"sent copy message"
 304         elif re.match(u"finished", msg):
 305             words = msg.split()
 306             if len(words) == 2:
 307                 self.sequencingFinished(words[1])
 308                 reply = u"sending sequencing finished for %s" % (words[1])
 309             else:
 310                 reply = u"need runfolder name"
 311         else:
 312             reply = u"I didn't understand '%s'" %(msg)
 313         return reply
 314
 315     def run(self):
 316         """
 317         Start application
 318         """
 319         # we have to configure pyinotify after BenderJab.start is called
 320         # as weird things happen to pyinotify if the stdio is closed
 321         # after it's initialized.
 322         self.add_watch()
 323         super(SpoolWatcher, self).run()
 324
 325     def stop(self):
 326         """
 327         shutdown application
 328         """
 329         # destroy the inotify's instance on this interrupt (stop monitoring)
 330         if self.notifier is not None:
 331             self.notifier.stop()
 332         super(SpoolWatcher, self).stop()
 333
 334     def startCopy(self, copy_url=None):
 335         logging.debug("writes seem to have stopped")
 336         if self.notify_runner is not None:
 337             for r in self.notify_runner:
 338                 self.rpc_send(r, tuple([copy_url]), 'startCopy')
 339         if self.notify_users is not None:
 340             for u in self.notify_users:
 341                 self.send(u, 'startCopy %s.' % (copy_url,))
 342
 343     def sequencingFinished(self, run_dir):
 344         # need to strip off self.watchdirs from rundir I suspect.
 345         logging.info("run.completed in " + str(run_dir))
 346         for watch in self.watchdirs:
 347             if not run_dir.startswith(watch):
 348                 print "%s didn't start with %s" % (run_dir, watch)
 349                 continue
 350             if watch[-1] != os.path.sep:
 351                 watch += os.path.sep
 352             stripped_run_dir = re.sub(watch, "", run_dir)
 353         else:
 354             stripped_run_dir = run_dir
 355
 356         logging.debug("stripped to " + stripped_run_dir)
 357         if self.notify_users is not None:
 358             for u in self.notify_users:
 359                 self.send(u, 'Sequencing run %s finished' % \
 360                           (stripped_run_dir))
 361         if self.notify_runner is not None:
 362             for r in self.notify_runner:
 363                 self.rpc_send(r, (stripped_run_dir,), 'sequencingFinished')
 364
 365 def main(args=None):
 366     bot = SpoolWatcher()
 367     return bot.main(args)
 368
 369 if __name__ == "__main__":
 370     ret = main(sys.argv[1:])
 371     #sys.exit(ret)
 372
 373 # TODO:
 374 # send messages to copier specifying which mount to copy