from __future__ import absolute_import from django.conf import settings from collections import deque import os import time import socket import logging import ujson import requests import cPickle as pickle import atexit import sys import signal import tornado import random from zerver.lib.utils import statsd from zerver.middleware import async_request_restart # The idle timeout used to be a week, but we found that in that # situation, queues from dead browser sessions would grow quite large # due to the accumulation of message data in those queues. IDLE_EVENT_QUEUE_TIMEOUT_SECS = 60 * 10 EVENT_QUEUE_GC_FREQ_MSECS = 1000 * 60 * 5 # Capped limit for how long a client can request an event queue # to live MAX_QUEUE_TIMEOUT_SECS = 7 * 24 * 60 * 60 # The heartbeats effectively act as a server-side timeout for # get_events(). The actual timeout value is randomized for each # client connection based on the below value. We ensure that the # maximum timeout value is 55 seconds, to deal with crappy home # wireless routers that kill "inactive" http connections. HEARTBEAT_MIN_FREQ_SECS = 45 class ClientDescriptor(object): def __init__(self, user_profile_id, realm_id, id, event_types, client_type, apply_markdown=True, all_public_streams=False, lifespan_secs=0): # These objects are pickled on shutdown and restored on restart. # If fields are added or semantics are changed, temporary code must be # added to load_event_queues() to update the restored objects self.user_profile_id = user_profile_id self.realm_id = realm_id self.current_handler = None self.event_queue = EventQueue(id) self.queue_timeout = lifespan_secs self.event_types = event_types self.last_connection_time = time.time() self.apply_markdown = apply_markdown self.all_public_streams = all_public_streams self.client_type = client_type self._timeout_handle = None # Clamp queue_timeout to between minimum and maximum timeouts self.queue_timeout = max(IDLE_EVENT_QUEUE_TIMEOUT_SECS, min(self.queue_timeout, MAX_QUEUE_TIMEOUT_SECS)) def prepare_for_pickling(self): self.current_handler = None self._timeout_handle = None def add_event(self, event): if self.current_handler is not None: async_request_restart(self.current_handler._request) self.event_queue.push(event) if self.current_handler is not None: try: self.current_handler.zulip_finish(dict(result='success', msg='', events=[event], queue_id=self.event_queue.id), self.current_handler._request, apply_markdown=self.apply_markdown) except socket.error: pass self.disconnect_handler() def accepts_event_type(self, type): if self.event_types is None: return True return type in self.event_types def idle(self, now): if not hasattr(self, 'queue_timeout'): self.queue_timeout = IDLE_EVENT_QUEUE_TIMEOUT_SECS return (self.current_handler is None and now - self.last_connection_time >= self.queue_timeout) def connect_handler(self, handler): self.current_handler = handler self.last_connection_time = time.time() def timeout_callback(): self._timeout_handle = None # All clients get heartbeat events self.add_event(dict(type='heartbeat')) ioloop = tornado.ioloop.IOLoop.instance() heartbeat_time = time.time() + HEARTBEAT_MIN_FREQ_SECS + random.randint(0, 10) self._timeout_handle = ioloop.add_timeout(heartbeat_time, timeout_callback) def disconnect_handler(self): self.current_handler = None if self._timeout_handle is not None: ioloop = tornado.ioloop.IOLoop.instance() ioloop.remove_timeout(self._timeout_handle) self._timeout_handle = None class EventQueue(object): def __init__(self, id): self.queue = deque() self.next_event_id = 0 self.id = id def push(self, event): event['id'] = self.next_event_id self.next_event_id += 1 self.queue.append(event) def pop(self): return self.queue.popleft() def empty(self): return len(self.queue) == 0 def prune(self, through_id): while not self.empty() and self.queue[0]['id'] <= through_id: self.pop() def contents(self): return list(self.queue) # maps queue ids to client descriptors clients = {} # maps user id to list of client descriptors user_clients = {} # maps realm id to list of client descriptors with all_public_streams=True realm_clients_all_streams = {} # list of registered gc hooks. # each one will be called with a user profile id, queue, and bool # last_for_client that is true if this is the last queue pertaining # to this user_profile_id # that is about to be deleted gc_hooks = [] next_queue_id = 0 def add_client_gc_hook(hook): gc_hooks.append(hook) def get_client_descriptor(queue_id): return clients.get(queue_id) def get_client_descriptors_for_user(user_profile_id): return user_clients.get(user_profile_id, []) def get_client_descriptors_for_realm_all_streams(realm_id): return realm_clients_all_streams.get(realm_id, []) def allocate_client_descriptor(user_profile_id, realm_id, event_types, client_type, apply_markdown, all_public_streams, lifespan_secs): global next_queue_id id = str(settings.SERVER_GENERATION) + ':' + str(next_queue_id) next_queue_id += 1 client = ClientDescriptor(user_profile_id, realm_id, id, event_types, client_type, apply_markdown, all_public_streams, lifespan_secs) clients[id] = client user_clients.setdefault(user_profile_id, []).append(client) if all_public_streams: realm_clients_all_streams.setdefault(realm_id, []).append(client) return client def gc_event_queues(): start = time.time() to_remove = set() affected_users = set() affected_realms = set() for (id, client) in clients.iteritems(): if client.idle(start): to_remove.add(id) affected_users.add(client.user_profile_id) affected_realms.add(client.realm_id) def filter_client_dict(client_dict, key): if key not in client_dict: return new_client_list = filter(lambda c: c.event_queue.id not in to_remove, client_dict[key]) if len(new_client_list) == 0: del client_dict[key] else: client_dict[key] = new_client_list for user_id in affected_users: filter_client_dict(user_clients, user_id) for realm_id in affected_realms: filter_client_dict(realm_clients_all_streams, realm_id) for id in to_remove: for cb in gc_hooks: cb(clients[id].user_profile_id, clients[id], clients[id].user_profile_id not in user_clients) del clients[id] logging.info(('Tornado removed %d idle event queues owned by %d users in %.3fs.' + ' Now %d active queues') % (len(to_remove), len(affected_users), time.time() - start, len(clients))) statsd.gauge('tornado.active_queues', len(clients)) statsd.gauge('tornado.active_users', len(user_clients)) def dump_event_queues(): start = time.time() # Remove unpickle-able attributes for client in clients.itervalues(): client.prepare_for_pickling() with file(settings.PERSISTENT_QUEUE_FILENAME, "w") as stored_queues: pickle.dump(clients, stored_queues) logging.info('Tornado dumped %d event queues in %.3fs' % (len(clients), time.time() - start)) def load_event_queues(): global clients start = time.time() try: with file(settings.PERSISTENT_QUEUE_FILENAME, "r") as stored_queues: clients = pickle.load(stored_queues) except (IOError, EOFError): pass for client in clients.itervalues(): # The following block can be dropped once we've # cleared out all our old event queues if not hasattr(client, 'realm_id') or not hasattr(client, 'all_public_streams'): from zerver.models import get_user_profile_by_id client.realm_id = get_user_profile_by_id(client.user_profile_id).realm.id client.all_public_streams = False logging.info("Tornado updated a queue") user_clients.setdefault(client.user_profile_id, []).append(client) if client.all_public_streams: realm_clients_all_streams.setdefault(client.realm_id, []).append(client) logging.info('Tornado loaded %d event queues in %.3fs' % (len(clients), time.time() - start)) def send_restart_events(): event = dict(type='restart', server_generation=settings.SERVER_GENERATION) for client in clients.itervalues(): # All clients get restart events client.add_event(event.copy()) def setup_event_queue(): load_event_queues() atexit.register(dump_event_queues) # Make sure we dump event queues even if we exit via signal signal.signal(signal.SIGTERM, lambda signum, stack: sys.exit(1)) try: os.remove(settings.PERSISTENT_QUEUE_FILENAME) except OSError: pass # Set up event queue garbage collection ioloop = tornado.ioloop.IOLoop.instance() pc = tornado.ioloop.PeriodicCallback(gc_event_queues, EVENT_QUEUE_GC_FREQ_MSECS, ioloop) pc.start() send_restart_events() # The following functions are called from Django # Workaround to support the Python-requests 1.0 transition of .json # from a property to a function requests_json_is_function = callable(requests.Response.json) def extract_json_response(resp): if requests_json_is_function: return resp.json() else: return resp.json def request_event_queue(user_profile, user_client, apply_markdown, queue_lifespan_secs, event_types=None, all_public_streams=False): if settings.TORNADO_SERVER: req = {'dont_block' : 'true', 'apply_markdown': ujson.dumps(apply_markdown), 'all_public_streams': ujson.dumps(all_public_streams), 'client' : 'internal', 'user_client' : user_client.name, 'lifespan_secs' : queue_lifespan_secs} if event_types is not None: req['event_types'] = ujson.dumps(event_types) resp = requests.get(settings.TORNADO_SERVER + '/api/v1/events', auth=requests.auth.HTTPBasicAuth(user_profile.email, user_profile.api_key), params=req) resp.raise_for_status() return extract_json_response(resp)['queue_id'] return None def get_user_events(user_profile, queue_id, last_event_id): if settings.TORNADO_SERVER: resp = requests.get(settings.TORNADO_SERVER + '/api/v1/events', auth=requests.auth.HTTPBasicAuth(user_profile.email, user_profile.api_key), params={'queue_id' : queue_id, 'last_event_id': last_event_id, 'dont_block' : 'true', 'client' : 'internal'}) resp.raise_for_status() return extract_json_response(resp)['events']