MatrixSynapse/synapse/appservice/scheduler.py

257 lines
9.2 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
2016-01-07 05:26:29 +01:00
# Copyright 2015, 2016 OpenMarket Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
This module controls the reliability for application service transactions.
The nominal flow through this module looks like:
__________
1---ASa[e]-->| Service |--> Queue ASa[f]
2----ASb[e]->| Queuer |
3--ASa[f]--->|__________|-----------+ ASa[e], ASb[e]
V
-````````- +------------+
|````````|<--StoreTxn-|Transaction |
|Database| | Controller |---> SEND TO AS
`--------` +------------+
What happens on SEND TO AS depends on the state of the Application Service:
- If the AS is marked as DOWN, do nothing.
- If the AS is marked as UP, send the transaction.
* SUCCESS : Increment where the AS is up to txn-wise and nuke the txn
contents from the db.
* FAILURE : Marked AS as DOWN and start Recoverer.
Recoverer attempts to recover ASes who have died. The flow for this looks like:
,--------------------- backoff++ --------------.
V |
START ---> Wait exp ------> Get oldest txn ID from ----> FAILURE
backoff DB and try to send it
^ |___________
Mark AS as | V
UP & quit +---------- YES SUCCESS
| | |
NO <--- Have more txns? <------ Mark txn success & nuke <-+
from db; incr AS pos.
2015-03-05 18:45:52 +01:00
Reset backoff.
This is all tied together by the AppServiceScheduler which DIs the required
components.
"""
2016-08-17 13:03:04 +02:00
from twisted.internet import defer
2015-03-06 16:12:24 +01:00
from synapse.appservice import ApplicationServiceState
from synapse.util.logcontext import run_in_background
2016-08-17 13:03:04 +02:00
from synapse.util.metrics import Measure
2015-03-06 16:12:24 +01:00
import logging
logger = logging.getLogger(__name__)
2015-03-06 12:50:27 +01:00
class ApplicationServiceScheduler(object):
""" Public facing API for this module. Does the required DI to tie the
components together. This also serves as the "event_pool", which in this
case is a simple array.
"""
def __init__(self, hs):
self.clock = hs.get_clock()
self.store = hs.get_datastore()
self.as_api = hs.get_application_service_api()
def create_recoverer(service, callback):
return _Recoverer(self.clock, self.store, self.as_api, service, callback)
self.txn_ctrl = _TransactionController(
self.clock, self.store, self.as_api, create_recoverer
)
2016-08-17 13:03:04 +02:00
self.queuer = _ServiceQueuer(self.txn_ctrl, self.clock)
@defer.inlineCallbacks
def start(self):
logger.info("Starting appservice scheduler")
# check for any DOWN ASes and start recoverers for them.
recoverers = yield _Recoverer.start(
self.clock, self.store, self.as_api, self.txn_ctrl.on_recovered
)
self.txn_ctrl.add_recoverers(recoverers)
def submit_event_for_as(self, service, event):
self.queuer.enqueue(service, event)
class _ServiceQueuer(object):
"""Queues events for the same application service together, sending
transactions as soon as possible. Once a transaction is sent successfully,
this schedules any other events in the queue to run.
"""
2016-08-17 13:03:04 +02:00
def __init__(self, txn_ctrl, clock):
2015-03-16 15:03:16 +01:00
self.queued_events = {} # dict of {service_id: [events]}
2016-08-17 13:03:04 +02:00
self.requests_in_flight = set()
self.txn_ctrl = txn_ctrl
2016-08-17 13:03:04 +02:00
self.clock = clock
2015-03-16 11:16:59 +01:00
def enqueue(self, service, event):
2015-03-16 15:03:16 +01:00
# if this service isn't being sent something
2016-08-17 13:03:04 +02:00
self.queued_events.setdefault(service.id, []).append(event)
run_in_background(self._send_request, service)
2016-08-17 13:03:04 +02:00
@defer.inlineCallbacks
def _send_request(self, service):
if service.id in self.requests_in_flight:
return
2016-08-17 15:52:26 +02:00
self.requests_in_flight.add(service.id)
try:
while True:
events = self.queued_events.pop(service.id, [])
if not events:
return
2016-08-17 16:05:50 +02:00
with Measure(self.clock, "servicequeuer.send"):
2016-08-17 13:03:04 +02:00
try:
yield self.txn_ctrl.send(service, events)
except Exception:
2016-08-17 13:03:04 +02:00
logger.exception("AS request failed")
2016-08-17 15:52:26 +02:00
finally:
self.requests_in_flight.discard(service.id)
class _TransactionController(object):
def __init__(self, clock, store, as_api, recoverer_fn):
self.clock = clock
self.store = store
self.as_api = as_api
self.recoverer_fn = recoverer_fn
# keep track of how many recoverers there are
self.recoverers = []
@defer.inlineCallbacks
def send(self, service, events):
try:
txn = yield self.store.create_appservice_txn(
service=service,
events=events
)
service_is_up = yield self._is_service_up(service)
if service_is_up:
sent = yield txn.send(self.as_api)
if sent:
2016-08-23 16:23:39 +02:00
yield txn.complete(self.store)
else:
run_in_background(self._start_recoverer, service)
except Exception:
logger.exception("Error creating appservice transaction")
run_in_background(self._start_recoverer, service)
2015-03-06 16:12:24 +01:00
@defer.inlineCallbacks
def on_recovered(self, recoverer):
2015-03-06 16:17:50 +01:00
self.recoverers.remove(recoverer)
logger.info("Successfully recovered application service AS ID %s",
recoverer.service.id)
logger.info("Remaining active recoverers: %s", len(self.recoverers))
yield self.store.set_appservice_state(
2015-03-06 16:12:24 +01:00
recoverer.service,
ApplicationServiceState.UP
)
def add_recoverers(self, recoverers):
for r in recoverers:
self.recoverers.append(r)
2015-03-06 16:17:50 +01:00
if len(recoverers) > 0:
logger.info("New active recoverers: %s", len(self.recoverers))
2015-03-06 16:12:24 +01:00
@defer.inlineCallbacks
def _start_recoverer(self, service):
try:
yield self.store.set_appservice_state(
service,
ApplicationServiceState.DOWN
)
logger.info(
"Application service falling behind. Starting recoverer. AS ID %s",
service.id
)
recoverer = self.recoverer_fn(service, self.on_recovered)
self.add_recoverers([recoverer])
recoverer.recover()
except Exception:
logger.exception("Error starting AS recoverer")
@defer.inlineCallbacks
def _is_service_up(self, service):
state = yield self.store.get_appservice_state(service)
defer.returnValue(state == ApplicationServiceState.UP or state is None)
class _Recoverer(object):
@staticmethod
@defer.inlineCallbacks
def start(clock, store, as_api, callback):
2015-03-06 16:12:24 +01:00
services = yield store.get_appservices_by_state(
ApplicationServiceState.DOWN
)
recoverers = [
_Recoverer(clock, store, as_api, s, callback) for s in services
]
for r in recoverers:
logger.info("Starting recoverer for AS ID %s which was marked as "
"DOWN", r.service.id)
r.recover()
defer.returnValue(recoverers)
def __init__(self, clock, store, as_api, service, callback):
self.clock = clock
self.store = store
self.as_api = as_api
self.service = service
self.callback = callback
self.backoff_counter = 1
def recover(self):
self.clock.call_later((2 ** self.backoff_counter), self.retry)
def _backoff(self):
# cap the backoff to be around 8.5min => (2^9) = 512 secs
if self.backoff_counter < 9:
self.backoff_counter += 1
self.recover()
2015-03-06 12:50:27 +01:00
@defer.inlineCallbacks
def retry(self):
try:
txn = yield self.store.get_oldest_unsent_txn(self.service)
if txn:
logger.info("Retrying transaction %s for AS ID %s",
txn.id, txn.service.id)
sent = yield txn.send(self.as_api)
if sent:
yield txn.complete(self.store)
# reset the backoff counter and retry immediately
self.backoff_counter = 1
yield self.retry()
else:
self._backoff()
else:
self._set_service_recovered()
except Exception as e:
logger.exception(e)
self._backoff()
def _set_service_recovered(self):
2015-03-06 16:12:24 +01:00
self.callback(self)