state.py 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318
  1. import time
  2. import heapq
  3. from threading import Lock
  4. from kombu.utils import partition
  5. from celery import states
  6. from celery.datastructures import AttributeDict, LocalCache
  7. from celery.utils import kwdict
  8. #: Hartbeat expiry time in seconds. The worker will be considered offline
  9. #: if no heartbeat is received within this time.
  10. #: Default is 2:30 minutes.
  11. HEARTBEAT_EXPIRE = 150
  12. class Element(AttributeDict):
  13. """Base class for worker state elements."""
  14. class Worker(Element):
  15. """Worker State."""
  16. heartbeat_max = 4
  17. def __init__(self, **fields):
  18. super(Worker, self).__init__(**fields)
  19. self.heartbeats = []
  20. def on_online(self, timestamp=None, **kwargs):
  21. """Callback for the `worker-online` event."""
  22. self._heartpush(timestamp)
  23. def on_offline(self, **kwargs):
  24. """Callback for the `worker-offline` event."""
  25. self.heartbeats = []
  26. def on_heartbeat(self, timestamp=None, **kwargs):
  27. """Callback for the `worker-heartbeat` event."""
  28. self._heartpush(timestamp)
  29. def _heartpush(self, timestamp):
  30. if timestamp:
  31. heapq.heappush(self.heartbeats, timestamp)
  32. if len(self.heartbeats) > self.heartbeat_max:
  33. self.heartbeats = self.heartbeats[self.heartbeat_max:]
  34. def __repr__(self):
  35. return "<Worker: %s (%s)" % (self.hostname,
  36. self.alive and "ONLINE" or "OFFLINE")
  37. @property
  38. def alive(self):
  39. return (self.heartbeats and
  40. time.time() < self.heartbeats[-1] + HEARTBEAT_EXPIRE)
  41. class Task(Element):
  42. """Task State."""
  43. #: How to merge out of order events.
  44. #: Disorder is detected by logical ordering (e.g. task-received must have
  45. #: happened before a task-failed event).
  46. #:
  47. #: A merge rule consists of a state and a list of fields to keep from
  48. #: that state. ``(RECEIVED, ("name", "args")``, means the name and args
  49. #: fields are always taken from the RECEIVED state, and any values for
  50. #: these fields received before or after is simply ignored.
  51. merge_rules = {states.RECEIVED: ("name", "args", "kwargs",
  52. "retries", "eta", "expires")}
  53. #: meth:`info` displays these fields by default.
  54. _info_fields = ("args", "kwargs", "retries", "result",
  55. "eta", "runtime", "expires", "exception")
  56. #: Default values.
  57. _defaults = dict(uuid=None, name=None, state=states.PENDING,
  58. received=False, sent=False, started=False,
  59. succeeded=False, failed=False, retried=False,
  60. revoked=False, args=None, kwargs=None, eta=None,
  61. expires=None, retries=None, worker=None, result=None,
  62. exception=None, timestamp=None, runtime=None,
  63. traceback=None)
  64. def __init__(self, **fields):
  65. super(Task, self).__init__(**dict(self._defaults, **fields))
  66. def update(self, state, timestamp, fields):
  67. """Update state from new event.
  68. :param state: State from event.
  69. :param timestamp: Timestamp from event.
  70. :param fields: Event data.
  71. """
  72. if self.worker:
  73. self.worker.on_heartbeat(timestamp=timestamp)
  74. if state != states.RETRY and self.state != states.RETRY and \
  75. states.state(state) < states.state(self.state):
  76. # this state logically happens-before the current state, so merge.
  77. self.merge(state, timestamp, fields)
  78. else:
  79. self.state = state
  80. self.timestamp = timestamp
  81. super(Task, self).update(fields)
  82. def merge(self, state, timestamp, fields):
  83. """Merge with out of order event."""
  84. keep = self.merge_rules.get(state)
  85. if keep is not None:
  86. fields = dict((key, fields.get(key)) for key in keep)
  87. super(Task, self).update(fields)
  88. def on_sent(self, timestamp=None, **fields):
  89. """Callback for the ``task-sent`` event."""
  90. self.sent = timestamp
  91. self.update(states.PENDING, timestamp, fields)
  92. def on_received(self, timestamp=None, **fields):
  93. """Callback for the ``task-received`` event."""
  94. self.received = timestamp
  95. self.update(states.RECEIVED, timestamp, fields)
  96. def on_started(self, timestamp=None, **fields):
  97. """Callback for the ``task-started`` event."""
  98. self.started = timestamp
  99. self.update(states.STARTED, timestamp, fields)
  100. def on_failed(self, timestamp=None, **fields):
  101. """Callback for the ``task-failed`` event."""
  102. self.failed = timestamp
  103. self.update(states.FAILURE, timestamp, fields)
  104. def on_retried(self, timestamp=None, **fields):
  105. """Callback for the ``task-retried`` event."""
  106. self.retried = timestamp
  107. self.update(states.RETRY, timestamp, fields)
  108. def on_succeeded(self, timestamp=None, **fields):
  109. """Callback for the ``task-succeeded`` event."""
  110. self.succeeded = timestamp
  111. self.update(states.SUCCESS, timestamp, fields)
  112. def on_revoked(self, timestamp=None, **fields):
  113. """Callback for the ``task-revoked`` event."""
  114. self.revoked = timestamp
  115. self.update(states.REVOKED, timestamp, fields)
  116. def info(self, fields=None, extra=[]):
  117. """Information about this task suitable for on-screen display."""
  118. if fields is None:
  119. fields = self._info_fields
  120. return dict((key, getattr(self, key, None))
  121. for key in list(fields) + list(extra)
  122. if getattr(self, key, None) is not None)
  123. def __repr__(self):
  124. return "<Task: %s(%s) %s>" % (self.name, self.uuid, self.state)
  125. @property
  126. def ready(self):
  127. return self.state in states.READY_STATES
  128. class State(object):
  129. """Records clusters state."""
  130. event_count = 0
  131. task_count = 0
  132. def __init__(self, callback=None,
  133. max_workers_in_memory=5000, max_tasks_in_memory=10000):
  134. self.workers = LocalCache(max_workers_in_memory)
  135. self.tasks = LocalCache(max_tasks_in_memory)
  136. self.event_callback = callback
  137. self.group_handlers = {"worker": self.worker_event,
  138. "task": self.task_event}
  139. self._mutex = Lock()
  140. def freeze_while(self, fun, *args, **kwargs):
  141. clear_after = kwargs.pop("clear_after", False)
  142. self._mutex.acquire()
  143. try:
  144. return fun(*args, **kwargs)
  145. finally:
  146. if clear_after:
  147. self._clear()
  148. self._mutex.release()
  149. def clear_tasks(self, ready=True):
  150. self._mutex.acquire()
  151. try:
  152. return self._clear_tasks(ready)
  153. finally:
  154. self._mutex.release()
  155. def _clear_tasks(self, ready=True):
  156. if ready:
  157. self.tasks = dict((uuid, task)
  158. for uuid, task in self.tasks.items()
  159. if task.state not in states.READY_STATES)
  160. else:
  161. self.tasks.clear()
  162. def _clear(self, ready=True):
  163. self.workers.clear()
  164. self._clear_tasks(ready)
  165. self.event_count = 0
  166. self.task_count = 0
  167. def clear(self, ready=True):
  168. self._mutex.acquire()
  169. try:
  170. return self._clear(ready)
  171. finally:
  172. self._mutex.release()
  173. def get_or_create_worker(self, hostname, **kwargs):
  174. """Get or create worker by hostname."""
  175. try:
  176. worker = self.workers[hostname]
  177. worker.update(kwargs)
  178. except KeyError:
  179. worker = self.workers[hostname] = Worker(
  180. hostname=hostname, **kwargs)
  181. return worker
  182. def get_or_create_task(self, uuid):
  183. """Get or create task by uuid."""
  184. try:
  185. return self.tasks[uuid]
  186. except KeyError:
  187. task = self.tasks[uuid] = Task(uuid=uuid)
  188. return task
  189. def worker_event(self, type, fields):
  190. """Process worker event."""
  191. hostname = fields.pop("hostname", None)
  192. if hostname:
  193. worker = self.get_or_create_worker(hostname)
  194. handler = getattr(worker, "on_%s" % type, None)
  195. if handler:
  196. handler(**fields)
  197. def task_event(self, type, fields):
  198. """Process task event."""
  199. uuid = fields.pop("uuid")
  200. hostname = fields.pop("hostname")
  201. worker = self.get_or_create_worker(hostname)
  202. task = self.get_or_create_task(uuid)
  203. handler = getattr(task, "on_%s" % type, None)
  204. if type == "received":
  205. self.task_count += 1
  206. if handler:
  207. handler(**fields)
  208. task.worker = worker
  209. def event(self, event):
  210. self._mutex.acquire()
  211. try:
  212. return self._dispatch_event(event)
  213. finally:
  214. self._mutex.release()
  215. def _dispatch_event(self, event):
  216. self.event_count += 1
  217. event = kwdict(event)
  218. group, _, type = partition(event.pop("type"), "-")
  219. self.group_handlers[group](type, event)
  220. if self.event_callback:
  221. self.event_callback(self, event)
  222. def tasks_by_timestamp(self, limit=None):
  223. """Get tasks by timestamp.
  224. Returns a list of `(uuid, task)` tuples.
  225. """
  226. return self._sort_tasks_by_time(self.tasks.items()[:limit])
  227. def _sort_tasks_by_time(self, tasks):
  228. """Sort task items by time."""
  229. return sorted(tasks, key=lambda t: t[1].timestamp,
  230. reverse=True)
  231. def tasks_by_type(self, name, limit=None):
  232. """Get all tasks by type.
  233. Returns a list of `(uuid, task)` tuples.
  234. """
  235. return self._sort_tasks_by_time([(uuid, task)
  236. for uuid, task in self.tasks.items()[:limit]
  237. if task.name == name])
  238. def tasks_by_worker(self, hostname, limit=None):
  239. """Get all tasks by worker.
  240. Returns a list of `(uuid, task)` tuples.
  241. """
  242. return self._sort_tasks_by_time([(uuid, task)
  243. for uuid, task in self.tasks.items()[:limit]
  244. if task.worker.hostname == hostname])
  245. def task_types(self):
  246. """Returns a list of all seen task types."""
  247. return list(sorted(set(task.name for task in self.tasks.values())))
  248. def alive_workers(self):
  249. """Returns a list of (seemingly) alive workers."""
  250. return [w for w in self.workers.values() if w.alive]
  251. def __repr__(self):
  252. return "<ClusterState: events=%s tasks=%s>" % (self.event_count,
  253. self.task_count)
  254. state = State()