oslo.messaging/oslo_messaging/_drivers/amqp1_driver/controller.py
John Eckersberg 02a38f507d amqp1: fix race when reconnecting
Currently this is how reconnect works:

- pyngus detects failure and invokes callback
  Controller.connection_failed() which in turn calls
  Controller._handle_connection_loss()

- The first thing that _handle_connection_loss does is to set
  self.addresser to None (important later)

- Then it defers _do_reconnect after a delay (normally 1 second)

- (1 second passes)

- _do_reconnect calls _hard_reset which resets the controller state

However, there is a race here.  This can happen:

- The above, up until it defers and waits for 1 second

- Controller.send() is invoked on a task

- A new Sender is created, and critically because self.reply_link
  still exists and is active, we call sender.attach and pass in
  self.addresser.  Remember _handle_connection_loss sets
  self.addresser to None.

- Eventually Sender.attach throws an AttributeError because it
  attempts to call addresser.resolve() but addresser is None

The reason this happens is because although the connection is dead,
the controller state is still half-alive because _hard_reset hasn't
been called yet since it's deferred one second in _do_reconnect.

The fix here is to move _hard_reset out of _do_reconnect and directly
into _handle_connection_loss.  The eventloop is woken up immediately
to process _hard_reset but _do_reconnect is still deferred as before
so as to retain the desired reconnect backoff behavior.

Closes-Bug: #1941652
Change-Id: Ife62a7d76022908f0dc6a77f1ad607cb2fbd3e8f
2021-11-09 15:59:54 -05:00

1336 lines
54 KiB
Python

# Copyright 2014, Red Hat, Inc.
#
# 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.
"""
Controller that manages the interface between the driver and the messaging
service.
This module defines a Controller class that is responsible for performing
messaging-related operations (Tasks) requested by the driver, and for managing
the connection to the messaging service. The Controller creates a background
thread which performs all messaging operations and socket I/O. The
Controller's messaging logic is executed in the background thread via lambda
functions scheduled by the Controller.
"""
import abc
import collections
import logging
import os
import platform
import queue
import random
import sys
import threading
import time
import uuid
from oslo_utils import eventletutils
import proton
import pyngus
from oslo_messaging._drivers.amqp1_driver.addressing import AddresserFactory
from oslo_messaging._drivers.amqp1_driver.addressing import keyify
from oslo_messaging._drivers.amqp1_driver.addressing import SERVICE_NOTIFY
from oslo_messaging._drivers.amqp1_driver.addressing import SERVICE_RPC
from oslo_messaging._drivers.amqp1_driver import eventloop
from oslo_messaging import exceptions
from oslo_messaging.target import Target
from oslo_messaging import transport
LOG = logging.getLogger(__name__)
class Task(object):
"""Run a command on the eventloop thread, wait until it completes
"""
@abc.abstractmethod
def wait(self):
"""Called by the client thread to wait for the operation to
complete. The implementation may optionally return a value.
"""
@abc.abstractmethod
def _execute(self, controller):
"""This method will be run on the eventloop thread to perform the
messaging operation.
"""
class SubscribeTask(Task):
"""A task that creates a subscription to the given target. Messages
arriving from the target are given to the listener.
"""
def __init__(self, target, listener, notifications=False):
super(SubscribeTask, self).__init__()
self._target = target() # mutable - need a copy
self._subscriber_id = listener.id
self._in_queue = listener.incoming
self._service = SERVICE_NOTIFY if notifications else SERVICE_RPC
self._wakeup = eventletutils.Event()
def wait(self):
self._wakeup.wait()
def _execute(self, controller):
controller.subscribe(self)
self._wakeup.set()
class SendTask(Task):
"""This is the class used by the Controller to send messages to a given
destination.
"""
def __init__(self, name, message, target, deadline, retry,
wait_for_ack, notification=False):
super(SendTask, self).__init__()
self.name = name
# note: target can be either a Target class or a string
# target is mutable - make copy
self.target = target() if isinstance(target, Target) else target
self.message = message
self.deadline = deadline
self.wait_for_ack = wait_for_ack
self.service = SERVICE_NOTIFY if notification else SERVICE_RPC
self.timer = None
self._retry = None if retry is None or retry < 0 else retry
self._wakeup = eventletutils.Event()
self._error = None
self._sender = None
def wait(self):
self._wakeup.wait()
return self._error
def _execute(self, controller):
if self.deadline:
# time out the send
self.timer = controller.processor.alarm(self._on_timeout,
self.deadline)
controller.send(self)
def _prepare(self, sender):
"""Called immediately before the message is handed off to the i/o
system. This implies that the sender link is up.
"""
self._sender = sender
def _on_ack(self, state, info):
"""If wait_for_ack is True, this is called by the eventloop thread when
the ack/nack is received from the peer. If wait_for_ack is False this
is called by the eventloop right after the message is written to the
link. In the last case state will always be set to ACCEPTED.
"""
if state != pyngus.SenderLink.ACCEPTED:
msg = ("{name} message send to {target} failed: remote"
" disposition: {disp}, info:"
"{info}".format(name=self.name,
target=self.target,
disp=state,
info=info))
self._error = exceptions.MessageDeliveryFailure(msg)
LOG.warning("%s", msg)
self._cleanup()
self._wakeup.set()
def _on_timeout(self):
"""Invoked by the eventloop when our timer expires
"""
self.timer = None
self._sender and self._sender.cancel_send(self)
msg = ("{name} message sent to {target} failed: timed"
" out".format(name=self.name, target=self.target))
LOG.warning("%s", msg)
# Only raise a MessagingTimeout if the caller has explicitly specified
# a timeout.
self._error = exceptions.MessagingTimeout(msg) \
if self.message.ttl else \
exceptions.MessageDeliveryFailure(msg)
self._cleanup()
self._wakeup.set()
def _on_error(self, description):
"""Invoked by the eventloop if the send operation fails for reasons
other than timeout and nack.
"""
msg = ("{name} message sent to {target} failed:"
" {reason}".format(name=self.name,
target=self.target,
reason=description))
LOG.warning("%s", msg)
self._error = exceptions.MessageDeliveryFailure(msg)
self._cleanup()
self._wakeup.set()
def _cleanup(self):
self._sender = None
if self.timer:
self.timer.cancel()
self.timer = None
@property
def _can_retry(self):
# has the retry count expired?
if self._retry is not None:
self._retry -= 1
if self._retry < 0:
return False
return True
class RPCCallTask(SendTask):
"""Performs an RPC Call. Sends the request and waits for a response from
the destination.
"""
def __init__(self, target, message, deadline, retry, wait_for_ack):
super(RPCCallTask, self).__init__("RPC Call", message, target,
deadline, retry, wait_for_ack)
self._reply_link = None
self._reply_msg = None
self._msg_id = None
def wait(self):
error = super(RPCCallTask, self).wait()
return error or self._reply_msg
def _prepare(self, sender):
super(RPCCallTask, self)._prepare(sender)
# reserve a message id for mapping the received response
if self._msg_id:
# already set so this is a re-transmit. To be safe cancel the old
# msg_id and allocate a fresh one.
self._reply_link.cancel_response(self._msg_id)
self._reply_link = sender._reply_link
rl = self._reply_link
self._msg_id = rl.prepare_for_response(self.message, self._on_reply)
def _on_reply(self, message):
# called if/when the reply message arrives
self._reply_msg = message
self._cleanup()
self._wakeup.set()
def _on_ack(self, state, info):
if state != pyngus.SenderLink.ACCEPTED:
super(RPCCallTask, self)._on_ack(state, info)
# must wait for reply if ACCEPTED
def _cleanup(self):
if self._msg_id:
self._reply_link.cancel_response(self._msg_id)
self._msg_id = None
self._reply_link = None
super(RPCCallTask, self)._cleanup()
class RPCMonitoredCallTask(RPCCallTask):
"""An RPC call which expects a periodic heartbeat until the response is
received. There are two timeouts:
deadline - overall hard timeout, implemented in RPCCallTask
call_monitor_timeout - keep alive timeout, reset when heartbeat arrives
"""
def __init__(self, target, message, deadline, call_monitor_timeout,
retry, wait_for_ack):
super(RPCMonitoredCallTask, self).__init__(target, message, deadline,
retry, wait_for_ack)
assert call_monitor_timeout is not None # nosec
self._monitor_timeout = call_monitor_timeout
self._monitor_timer = None
self._set_alarm = None
def _execute(self, controller):
self._set_alarm = controller.processor.defer
self._monitor_timer = self._set_alarm(self._call_timeout,
self._monitor_timeout)
super(RPCMonitoredCallTask, self)._execute(controller)
def _call_timeout(self):
# monitor_timeout expired
self._monitor_timer = None
self._sender and self._sender.cancel_send(self)
msg = ("{name} message sent to {target} failed: call monitor timed"
" out".format(name=self.name, target=self.target))
LOG.warning("%s", msg)
self._error = exceptions.MessagingTimeout(msg)
self._cleanup()
self._wakeup.set()
def _on_reply(self, message):
# if reply is null, then this is the call monitor heartbeat
if message.body is None:
self._monitor_timer.cancel()
self._monitor_timer = self._set_alarm(self._call_timeout,
self._monitor_timeout)
else:
super(RPCMonitoredCallTask, self)._on_reply(message)
def _cleanup(self):
self._set_alarm = None
if self._monitor_timer:
self._monitor_timer.cancel()
self._monitor_timer = None
super(RPCMonitoredCallTask, self)._cleanup()
class MessageDispositionTask(Task):
"""A task that updates the message disposition as accepted or released
for a Server
"""
def __init__(self, disposition, released=False):
super(MessageDispositionTask, self).__init__()
self._disposition = disposition
self._released = released
def wait(self):
# disposition update does not have to block the sender since there is
# no result to pend for. This avoids a thread context switch with
# every RPC call
pass
def _execute(self, controller):
try:
self._disposition(self._released)
except Exception as e:
# there's really nothing we can do about a failed disposition.
LOG.exception("Message acknowledgment failed: %s", e)
class Sender(pyngus.SenderEventHandler):
"""A link for sending to a particular destination on the message bus.
"""
def __init__(self, destination, scheduler, delay, service):
super(Sender, self).__init__()
self._destination = destination
self._service = service
self._address = None
self._link = None
self._scheduler = scheduler
self._delay = delay # for re-connecting/re-transmitting
# holds all pending SendTasks
self._pending_sends = collections.deque()
# holds all messages sent but not yet acked
self._unacked = set()
self._reply_link = None
self._connection = None
self._resend_timer = None
@property
def pending_messages(self):
return len(self._pending_sends)
@property
def unacked_messages(self):
return len(self._unacked)
def attach(self, connection, reply_link, addresser):
"""Open the link. Called by the Controller when the AMQP connection
becomes active.
"""
self._connection = connection
self._reply_link = reply_link
self._address = addresser.resolve(self._destination, self._service)
LOG.debug("Sender %s attached", self._address)
self._link = self._open_link()
def detach(self):
"""Close the link. Called by the controller when shutting down or in
response to a close requested by the remote. May be re-attached later
(after a reset is done)
"""
LOG.debug("Sender %s detached", self._address)
self._connection = None
self._reply_link = None
if self._resend_timer:
self._resend_timer.cancel()
self._resend_timer = None
if self._link:
self._link.close()
def reset(self, reason="Link reset"):
"""Called by the controller on connection failover. Release all link
resources, abort any in-flight messages, and check the retry limit on
all pending send requests.
"""
self._address = None
self._connection = None
self._reply_link = None
if self._link:
self._link.destroy()
self._link = None
self._abort_unacked(reason)
self._check_retry_limit(reason)
def destroy(self, reason="Link destroyed"):
"""Destroy the sender and all pending messages. Called on driver
shutdown.
"""
LOG.debug("Sender %s destroyed", self._address)
self.reset(reason)
self._abort_pending(reason)
def send_message(self, send_task):
"""Send a message out the link.
"""
if not self._can_send or self._pending_sends:
self._pending_sends.append(send_task)
else:
self._send(send_task)
def cancel_send(self, send_task):
"""Attempts to cancel a send request. It is possible that the send has
already completed, so this is best-effort.
"""
# may be in either list, or none
self._unacked.discard(send_task)
try:
self._pending_sends.remove(send_task)
except ValueError:
pass
# Pyngus callbacks:
def sender_active(self, sender_link):
LOG.debug("Sender %s active", self._address)
self._send_pending()
def credit_granted(self, sender_link):
pass
def sender_remote_closed(self, sender_link, pn_condition):
# The remote has initiated a close. This could happen when the message
# bus is shutting down, or it detected an error
LOG.warning("Sender %(addr)s failed due to remote initiated close:"
" condition=%(cond)s",
{'addr': self._address, 'cond': pn_condition})
self._link.close()
# sender_closed() will be called once the link completes closing
def sender_closed(self, sender_link):
self._handle_sender_closed()
def sender_failed(self, sender_link, error):
"""Protocol error occurred."""
LOG.warning("Sender %(addr)s failed error=%(error)s",
{'addr': self._address, 'error': error})
self._handle_sender_closed(str(error))
# end Pyngus callbacks
def _handle_sender_closed(self, reason="Sender closed"):
self._abort_unacked(reason)
if self._connection:
# still attached, so attempt to restart the link
self._check_retry_limit(reason)
self._scheduler.defer(self._reopen_link, self._delay)
def _check_retry_limit(self, reason):
# Called on recoverable connection or link failure. Remove any pending
# sends that have exhausted their retry count:
expired = set()
for send_task in self._pending_sends:
if not send_task._can_retry:
expired.add(send_task)
send_task._on_error("Message send failed: %s" % reason)
while expired:
self._pending_sends.remove(expired.pop())
def _abort_unacked(self, error):
# fail all messages that have been sent to the message bus and have not
# been acked yet
while self._unacked:
send_task = self._unacked.pop()
send_task._on_error("Message send failed: %s" % error)
def _abort_pending(self, error):
# fail all messages that have yet to be sent to the message bus
while self._pending_sends:
send_task = self._pending_sends.popleft()
send_task._on_error("Message send failed: %s" % error)
@property
def _can_send(self):
return self._link and self._link.active
# acknowledge status
_TIMED_OUT = pyngus.SenderLink.TIMED_OUT
_ACCEPTED = pyngus.SenderLink.ACCEPTED
_RELEASED = pyngus.SenderLink.RELEASED
_MODIFIED = pyngus.SenderLink.MODIFIED
def _send(self, send_task):
send_task._prepare(self)
send_task.message.address = self._address
if send_task.wait_for_ack:
self._unacked.add(send_task)
def pyngus_callback(link, handle, state, info):
# invoked when the message bus (n)acks this message
if state == Sender._TIMED_OUT:
# ignore pyngus timeout - we maintain our own timer
# which will properly deal with this case
return
self._unacked.discard(send_task)
if state == Sender._ACCEPTED:
send_task._on_ack(Sender._ACCEPTED, info)
elif (state == Sender._RELEASED or
(state == Sender._MODIFIED and
# assuming delivery-failed means in-doubt:
not info.get("delivery-failed") and
not info.get("undeliverable-here"))):
# These states indicate that the message was never
# forwarded beyond the next hop so they can be
# re-transmitted without risk of duplication
self._resend(send_task)
else:
# some error - let task figure it out...
send_task._on_ack(state, info)
self._link.send(send_task.message,
delivery_callback=pyngus_callback,
handle=self,
deadline=send_task.deadline)
else: # do not wait for ack
self._link.send(send_task.message,
delivery_callback=None,
handle=self,
deadline=send_task.deadline)
send_task._on_ack(pyngus.SenderLink.ACCEPTED, {})
def _resend(self, send_task):
# the message bus returned the message without forwarding it. Wait a
# bit for other outstanding sends to finish - most likely ending up
# here since they are all going to the same destination - then resend
# this message
if send_task._can_retry:
# note well: once there is something on the pending list no further
# messages will be sent (they will all queue up behind this one).
self._pending_sends.append(send_task)
if self._resend_timer is None:
sched = self._scheduler
# this will get the pending sends going again
self._resend_timer = sched.defer(self._resend_pending,
self._delay)
else:
send_task._on_error("Send retries exhausted")
def _resend_pending(self):
# run from the _resend_timer, attempt to resend pending messages
self._resend_timer = None
self._send_pending()
def _send_pending(self):
# flush all pending messages out
if self._can_send:
while self._pending_sends:
self._send(self._pending_sends.popleft())
def _open_link(self):
name = "openstack.org/om/sender/[%s]/%s" % (self._address,
uuid.uuid4().hex)
link = self._connection.create_sender(name=name,
source_address=self._address,
target_address=self._address,
event_handler=self)
link.open()
return link
def _reopen_link(self):
if self._connection:
if self._link:
self._link.destroy()
self._link = self._open_link()
class Replies(pyngus.ReceiverEventHandler):
"""This is the receiving link for all RPC reply messages. Messages are
routed to the proper incoming queue using the correlation-id header in the
message.
"""
def __init__(self, connection, on_ready, on_down, capacity):
self._correlation = {} # map of correlation-id to response queue
self._on_ready = on_ready
self._on_down = on_down
rname = ("openstack.org/om/receiver/[rpc-response]/%s"
% uuid.uuid4().hex)
self._receiver = connection.create_receiver("rpc-response",
event_handler=self,
name=rname)
# capacity determines the maximum number of reply messages this link is
# willing to receive. As messages are received and capacity is
# consumed, this driver will 'top up' the capacity back to max
# capacity. This number should be large enough to avoid needlessly
# flow-controlling the replies.
self._capacity = capacity
self._capacity_low = (capacity + 1) / 2
self._receiver.open()
def detach(self):
# close the link
if self._receiver:
self._receiver.close()
def destroy(self):
self._correlation.clear()
if self._receiver:
self._receiver.destroy()
self._receiver = None
def prepare_for_response(self, request, callback):
"""Apply a unique message identifier to this request message. This will
be used to identify messages received in reply. The identifier is
placed in the 'id' field of the request message. It is expected that
the identifier will appear in the 'correlation-id' field of the
corresponding response message.
When the caller is done receiving replies, it must call cancel_response
"""
request.id = uuid.uuid4().hex
# reply is placed on reply_queue
self._correlation[request.id] = callback
request.reply_to = self._receiver.source_address
return request.id
def cancel_response(self, msg_id):
"""Abort waiting for the response message corresponding to msg_id.
This can be used if the request fails and no reply is expected.
"""
try:
del self._correlation[msg_id]
except KeyError:
pass
@property
def active(self):
return self._receiver and self._receiver.active
# Pyngus ReceiverLink event callbacks:
def receiver_active(self, receiver_link):
"""This is a Pyngus callback, invoked by Pyngus when the receiver_link
has transitioned to the open state and is able to receive incoming
messages.
"""
LOG.debug("Replies link active src=%s", self._receiver.source_address)
receiver_link.add_capacity(self._capacity)
self._on_ready()
def receiver_remote_closed(self, receiver, pn_condition):
"""This is a Pyngus callback, invoked by Pyngus when the peer of this
receiver link has initiated closing the connection.
"""
if pn_condition:
LOG.error("Reply subscription closed by peer: %s",
pn_condition)
receiver.close()
def receiver_failed(self, receiver_link, error):
"""Protocol error occurred."""
LOG.error("Link to reply queue failed. error=%(error)s",
{"error": error})
self._on_down()
def receiver_closed(self, receiver_link):
self._on_down()
def message_received(self, receiver, message, handle):
"""This is a Pyngus callback, invoked by Pyngus when a new message
arrives on this receiver link from the peer.
"""
key = message.correlation_id
try:
self._correlation[key](message)
receiver.message_accepted(handle)
except KeyError:
LOG.warning("Can't find receiver for response msg id=%s, "
"dropping!", key)
receiver.message_modified(handle, True, True, None)
# ensure we have enough credit
if receiver.capacity <= self._capacity_low:
receiver.add_capacity(self._capacity - receiver.capacity)
class Server(pyngus.ReceiverEventHandler):
"""A group of links that receive messages from a set of addresses derived
from a given target. Messages arriving on the links are placed on the
'incoming' queue.
"""
def __init__(self, target, incoming, scheduler, delay, capacity):
self._target = target
self._incoming = incoming
self._addresses = []
self._capacity = capacity # credit per each link
self._capacity_low = (capacity + 1) / 2
self._receivers = []
self._scheduler = scheduler
self._delay = delay # for link re-attach
self._connection = None
self._reopen_scheduled = False
def attach(self, connection):
"""Create receiver links over the given connection for all the
configured addresses.
"""
self._connection = connection
for a in self._addresses:
name = "openstack.org/om/receiver/[%s]/%s" % (a, uuid.uuid4().hex)
r = self._open_link(a, name)
self._receivers.append(r)
def detach(self):
"""Attempt a clean shutdown of the links"""
self._connection = None
self._addresses = []
for receiver in self._receivers:
receiver.close()
def reset(self):
# destroy the links, but keep the addresses around since we may be
# failing over. Since links are destroyed, this cannot be called from
# any of the following ReceiverLink callbacks.
self._connection = None
self._addresses = []
self._reopen_scheduled = False
for r in self._receivers:
r.destroy()
self._receivers = []
# Pyngus ReceiverLink event callbacks. Note that all of the Server's links
# share this handler
def receiver_remote_closed(self, receiver, pn_condition):
"""This is a Pyngus callback, invoked by Pyngus when the peer of this
receiver link has initiated closing the connection.
"""
LOG.debug("Server subscription to %s remote detach",
receiver.source_address)
if pn_condition:
vals = {
"addr": receiver.source_address or receiver.target_address,
"err_msg": pn_condition
}
LOG.error("Server subscription %(addr)s closed "
"by peer: %(err_msg)s", vals)
receiver.close()
def receiver_failed(self, receiver_link, error):
"""Protocol error occurred."""
LOG.error("Listener link queue failed. error=%(error)s",
{"error": error})
self.receiver_closed(receiver_link)
def receiver_closed(self, receiver_link):
LOG.debug("Server subscription to %s closed",
receiver_link.source_address)
# If still attached, attempt to re-start link
if self._connection and not self._reopen_scheduled:
LOG.debug("Server subscription reopen scheduled")
self._reopen_scheduled = True
self._scheduler.defer(self._reopen_links, self._delay)
def message_received(self, receiver, message, handle):
"""This is a Pyngus callback, invoked by Pyngus when a new message
arrives on this receiver link from the peer.
"""
def message_disposition(released=False):
if receiver in self._receivers and not receiver.closed:
if released:
receiver.message_released(handle)
else:
receiver.message_accepted(handle)
if receiver.capacity <= self._capacity_low:
receiver.add_capacity(self._capacity - receiver.capacity)
else:
LOG.debug("Can't find receiver for settlement")
qentry = {"message": message, "disposition": message_disposition}
self._incoming.put(qentry)
def _open_link(self, address, name):
props = {"snd-settle-mode": "mixed"}
r = self._connection.create_receiver(source_address=address,
target_address=address,
event_handler=self,
name=name,
properties=props)
r.add_capacity(self._capacity)
r.open()
return r
def _reopen_links(self):
# attempt to re-establish any closed links
LOG.debug("Server subscription reopening")
self._reopen_scheduled = False
if self._connection:
for i in range(len(self._receivers)):
link = self._receivers[i]
if link.closed:
addr = link.target_address
name = link.name
link.destroy()
self._receivers[i] = self._open_link(addr, name)
class RPCServer(Server):
"""Subscribes to RPC addresses"""
def __init__(self, target, incoming, scheduler, delay, capacity):
super(RPCServer, self).__init__(target, incoming, scheduler, delay,
capacity)
def attach(self, connection, addresser):
# Generate the AMQP 1.0 addresses for the base class
self._addresses = [
addresser.unicast_address(self._target, SERVICE_RPC),
addresser.multicast_address(self._target, SERVICE_RPC),
addresser.anycast_address(self._target, SERVICE_RPC)
]
# now invoke the base class with the generated addresses
super(RPCServer, self).attach(connection)
class NotificationServer(Server):
"""Subscribes to Notification addresses"""
def __init__(self, target, incoming, scheduler, delay, capacity):
super(NotificationServer, self).__init__(target, incoming, scheduler,
delay, capacity)
def attach(self, connection, addresser):
# Generate the AMQP 1.0 addresses for the base class
self._addresses = [
addresser.anycast_address(self._target, SERVICE_NOTIFY)
]
# now invoke the base class with the generated addresses
super(NotificationServer, self).attach(connection)
class Hosts(object):
"""An order list of TransportHost addresses. Connection failover progresses
from one host to the next. The default realm comes from the configuration
and is only used if no realm is present in the URL.
"""
def __init__(self, url, default_realm=None):
self.virtual_host = url.virtual_host
if url.hosts:
self._entries = url.hosts[:]
else:
self._entries = [transport.TransportHost(hostname="localhost",
port=5672)]
for entry in self._entries:
entry.port = entry.port or 5672
entry.username = entry.username
entry.password = entry.password
if default_realm and entry.username and '@' not in entry.username:
entry.username = entry.username + '@' + default_realm
self._current = random.randint(0, len(self._entries) - 1) # nosec
@property
def current(self):
return self._entries[self._current]
def next(self):
if len(self._entries) > 1:
self._current = (self._current + 1) % len(self._entries)
return self.current
def __repr__(self):
return '<Hosts ' + str(self) + '>'
def __str__(self):
r = ', vhost=%s' % self.virtual_host if self.virtual_host else ''
return ", ".join(["%r" % th for th in self._entries]) + r
class Controller(pyngus.ConnectionEventHandler):
"""Controls the connection to the AMQP messaging service. This object is
the 'brains' of the driver. It maintains the logic for addressing, sending
and receiving messages, and managing the connection. All messaging and I/O
work is done on the Eventloop thread, allowing the driver to run
asynchronously from the messaging clients.
"""
def __init__(self, url, default_exchange, config):
self.processor = None
self._socket_connection = None
self._node = platform.node() or "<UNKNOWN>"
self._command = os.path.basename(sys.argv[0])
self._pid = os.getpid()
# queue of drivertask objects to execute on the eventloop thread
self._tasks = queue.Queue(maxsize=500)
# limit the number of Task()'s to execute per call to _process_tasks().
# This allows the eventloop main thread to return to servicing socket
# I/O in a timely manner
self._max_task_batch = 50
# cache of all Sender links indexed by address:
self._all_senders = {}
# active Sender links indexed by address:
self._active_senders = set()
# closing Sender links indexed by address:
self._purged_senders = []
# Servers indexed by target. Each entry is a map indexed by the
# specific ProtonListener's identifier:
self._servers = {}
self._container_name = config.oslo_messaging_amqp.container_name
self.idle_timeout = config.oslo_messaging_amqp.idle_timeout
self.trace_protocol = config.oslo_messaging_amqp.trace
self.ssl = config.oslo_messaging_amqp.ssl
self.ssl_ca_file = config.oslo_messaging_amqp.ssl_ca_file
self.ssl_cert_file = config.oslo_messaging_amqp.ssl_cert_file
self.ssl_key_file = config.oslo_messaging_amqp.ssl_key_file
self.ssl_key_password = config.oslo_messaging_amqp.ssl_key_password
self.ssl_verify_vhost = config.oslo_messaging_amqp.ssl_verify_vhost
self.pseudo_vhost = config.oslo_messaging_amqp.pseudo_vhost
self.sasl_mechanisms = config.oslo_messaging_amqp.sasl_mechanisms
self.sasl_config_dir = config.oslo_messaging_amqp.sasl_config_dir
self.sasl_config_name = config.oslo_messaging_amqp.sasl_config_name
self.hosts = Hosts(url, config.oslo_messaging_amqp.sasl_default_realm)
self.conn_retry_interval = \
config.oslo_messaging_amqp.connection_retry_interval
self.conn_retry_backoff = \
config.oslo_messaging_amqp.connection_retry_backoff
self.conn_retry_interval_max = \
config.oslo_messaging_amqp.connection_retry_interval_max
self.link_retry_delay = config.oslo_messaging_amqp.link_retry_delay
_opts = config.oslo_messaging_amqp
factory_args = {"legacy_server_prefix": _opts.server_request_prefix,
"legacy_broadcast_prefix": _opts.broadcast_prefix,
"legacy_group_prefix": _opts.group_request_prefix,
"rpc_prefix": _opts.rpc_address_prefix,
"notify_prefix": _opts.notify_address_prefix,
"multicast": _opts.multicast_address,
"unicast": _opts.unicast_address,
"anycast": _opts.anycast_address,
"notify_exchange": _opts.default_notification_exchange,
"rpc_exchange": _opts.default_rpc_exchange}
self.addresser_factory = AddresserFactory(default_exchange,
_opts.addressing_mode,
**factory_args)
self.addresser = None
# cannot send an RPC request until the replies link is active, as we
# need the peer assigned address, so need to delay sending any RPC
# requests until this link is active:
self.reply_link = None
# Set True when the driver is shutting down
self._closing = False
# only schedule one outstanding reconnect attempt at a time
self._reconnecting = False
self._delay = self.conn_retry_interval # seconds between retries
# prevent queuing up multiple requests to run _process_tasks()
self._process_tasks_scheduled = False
self._process_tasks_lock = threading.Lock()
# credit levels for incoming links
self._reply_credit = _opts.reply_link_credit
self._rpc_credit = _opts.rpc_server_credit
self._notify_credit = _opts.notify_server_credit
# sender link maintenance timer and interval
self._link_maint_timer = None
self._link_maint_timeout = _opts.default_sender_link_timeout
def connect(self):
"""Connect to the messaging service."""
self.processor = eventloop.Thread(self._container_name, self._node,
self._command, self._pid)
self.processor.wakeup(lambda: self._do_connect())
def add_task(self, task):
"""Add a Task for execution on processor thread."""
self._tasks.put(task)
self._schedule_task_processing()
def shutdown(self, timeout=30):
"""Shutdown the messaging service."""
LOG.info("Shutting down the AMQP 1.0 connection")
if self.processor:
self.processor.wakeup(self._start_shutdown)
LOG.debug("Waiting for eventloop to exit")
self.processor.join(timeout)
self._hard_reset("Shutting down")
for sender in self._all_senders.values():
sender.destroy()
self._all_senders.clear()
self._servers.clear()
self.processor.destroy()
self.processor = None
LOG.debug("Eventloop exited, driver shut down")
# The remaining methods are reserved to run from the eventloop thread only!
# They must not be invoked directly!
# methods executed by Tasks created by the driver:
def send(self, send_task):
if send_task.deadline and send_task.deadline <= time.monotonic():
send_task._on_timeout()
return
key = keyify(send_task.target, send_task.service)
sender = self._all_senders.get(key)
if not sender:
sender = Sender(send_task.target, self.processor,
self.link_retry_delay, send_task.service)
self._all_senders[key] = sender
if self.reply_link and self.reply_link.active:
sender.attach(self._socket_connection.pyngus_conn,
self.reply_link, self.addresser)
self._active_senders.add(key)
sender.send_message(send_task)
def subscribe(self, subscribe_task):
"""Subscribe to a given target"""
if subscribe_task._service == SERVICE_NOTIFY:
t = "notification"
server = NotificationServer(subscribe_task._target,
subscribe_task._in_queue,
self.processor,
self.link_retry_delay,
self._notify_credit)
else:
t = "RPC"
server = RPCServer(subscribe_task._target,
subscribe_task._in_queue,
self.processor,
self.link_retry_delay,
self._rpc_credit)
LOG.debug("Subscribing to %(type)s target %(target)s",
{'type': t, 'target': subscribe_task._target})
key = keyify(subscribe_task._target, subscribe_task._service)
servers = self._servers.get(key)
if servers is None:
servers = {}
self._servers[key] = servers
servers[subscribe_task._subscriber_id] = server
if self._active:
server.attach(self._socket_connection.pyngus_conn,
self.addresser)
# commands executed on the processor (eventloop) via 'wakeup()':
def _do_connect(self):
"""Establish connection and reply subscription on processor thread."""
host = self.hosts.current
conn_props = {'properties': {'process': self._command,
'pid': self._pid,
'node': self._node}}
# only set hostname in the AMQP 1.0 Open performative if the message
# bus can interpret it as the virtual host. We leave it unspecified
# since apparently noone can agree on how it should be used otherwise!
if self.hosts.virtual_host and not self.pseudo_vhost:
conn_props['hostname'] = self.hosts.virtual_host
if self.idle_timeout:
conn_props["idle-time-out"] = float(self.idle_timeout)
if self.trace_protocol:
conn_props["x-trace-protocol"] = self.trace_protocol
# SSL configuration
ssl_enabled = False
if self.ssl:
ssl_enabled = True
conn_props["x-ssl"] = self.ssl
if self.ssl_ca_file:
conn_props["x-ssl-ca-file"] = self.ssl_ca_file
ssl_enabled = True
if self.ssl_cert_file:
ssl_enabled = True
conn_props["x-ssl-identity"] = (self.ssl_cert_file,
self.ssl_key_file,
self.ssl_key_password)
if ssl_enabled:
# Set the identity of the remote server for SSL to use when
# verifying the received certificate. Typically this is the DNS
# name used to set up the TCP connections. However some servers
# may provide a certificate for the virtual host instead. If that
# is the case we need to use the virtual hostname instead.
# Refer to SSL Server Name Indication (SNI) for the entire story:
# https://tools.ietf.org/html/rfc6066
if self.ssl_verify_vhost:
if self.hosts.virtual_host:
conn_props['x-ssl-peer-name'] = self.hosts.virtual_host
else:
conn_props['x-ssl-peer-name'] = host.hostname
# SASL configuration:
if self.sasl_mechanisms:
conn_props["x-sasl-mechs"] = self.sasl_mechanisms
if self.sasl_config_dir:
conn_props["x-sasl-config-dir"] = self.sasl_config_dir
if self.sasl_config_name:
conn_props["x-sasl-config-name"] = self.sasl_config_name
self._socket_connection = self.processor.connect(host,
handler=self,
properties=conn_props)
LOG.debug("Connection initiated")
def _process_tasks(self):
"""Execute Task objects in the context of the processor thread."""
with self._process_tasks_lock:
self._process_tasks_scheduled = False
count = 0
while (not self._tasks.empty() and
count < self._max_task_batch):
try:
self._tasks.get(False)._execute(self)
except Exception as e:
LOG.exception("Error processing task: %s", e)
count += 1
# if we hit _max_task_batch, resume task processing later:
if not self._tasks.empty():
self._schedule_task_processing()
def _schedule_task_processing(self):
"""_process_tasks() helper: prevent queuing up multiple requests for
task processing. This method is called both by the application thread
and the processing thread.
"""
if self.processor:
with self._process_tasks_lock:
already_scheduled = self._process_tasks_scheduled
self._process_tasks_scheduled = True
if not already_scheduled:
self.processor.wakeup(lambda: self._process_tasks())
def _start_shutdown(self):
"""Called when the application is closing the transport.
Attempt to cleanly flush/close all links.
"""
self._closing = True
if self._active:
# try a clean shutdown
self._detach_senders()
self._detach_servers()
self.reply_link.detach()
self._socket_connection.pyngus_conn.close()
else:
# don't wait for a close from the remote, may never happen
self.processor.shutdown()
# reply link callbacks:
def _reply_link_ready(self):
"""Invoked when the Replies reply link has become active. At this
point, we are ready to receive messages, so start all pending RPC
requests.
"""
LOG.info("Messaging is active (%(hostname)s:%(port)s%(vhost)s)",
{'hostname': self.hosts.current.hostname,
'port': self.hosts.current.port,
'vhost': ("/" + self.hosts.virtual_host
if self.hosts.virtual_host else "")})
for sender in self._all_senders.values():
sender.attach(self._socket_connection.pyngus_conn,
self.reply_link, self.addresser)
def _reply_link_down(self):
# Treat it as a recoverable failure because the RPC reply address is
# now invalid for all in-flight RPC requests.
if not self._closing:
self._detach_senders()
self._detach_servers()
self._socket_connection.pyngus_conn.close()
# once closed, _handle_connection_loss() will initiate reconnect
# callback from eventloop on socket error
def socket_error(self, error):
"""Called by eventloop when a socket error occurs."""
LOG.error("Socket failure: %s", error)
self._handle_connection_loss(str(error))
# Pyngus connection event callbacks (and their helpers), all invoked from
# the eventloop thread:
def connection_failed(self, connection, error):
"""This is a Pyngus callback, invoked by Pyngus when a non-recoverable
error occurs on the connection.
"""
if connection is not self._socket_connection.pyngus_conn:
# pyngus bug: ignore failure callback on destroyed connections
return
LOG.debug("AMQP Connection failure: %s", error)
self._handle_connection_loss(str(error))
def connection_active(self, connection):
"""This is a Pyngus callback, invoked by Pyngus when the connection to
the peer is up. At this point, the driver will activate all subscriber
links (server) and the reply link.
"""
LOG.debug("Connection active (%(hostname)s:%(port)s), subscribing...",
{'hostname': self.hosts.current.hostname,
'port': self.hosts.current.port})
# allocate an addresser based on the advertised properties of the
# message bus
props = connection.remote_properties or {}
self.addresser = self.addresser_factory(props,
self.hosts.virtual_host
if self.pseudo_vhost else None)
for servers in self._servers.values():
for server in servers.values():
server.attach(self._socket_connection.pyngus_conn,
self.addresser)
self.reply_link = Replies(self._socket_connection.pyngus_conn,
self._reply_link_ready,
self._reply_link_down,
self._reply_credit)
self._delay = self.conn_retry_interval # reset
# schedule periodic maintenance of sender links
self._link_maint_timer = self.processor.defer(self._purge_sender_links,
self._link_maint_timeout)
def connection_closed(self, connection):
"""This is a Pyngus callback, invoked by Pyngus when the connection has
cleanly closed. This occurs after the driver closes the connection
locally, and the peer has acknowledged the close. At this point, the
shutdown of the driver's connection is complete.
"""
LOG.debug("AMQP connection closed.")
# if the driver isn't being shutdown, failover and reconnect
self._handle_connection_loss("AMQP connection closed.")
def connection_remote_closed(self, connection, reason):
"""This is a Pyngus callback, invoked by Pyngus when the peer has
requested that the connection be closed.
"""
# The messaging service/broker is trying to shut down the
# connection. Acknowledge the close, and try to reconnect/failover
# later once the connection has closed (connection_closed is called).
if reason:
LOG.info("Connection closed by peer: %s", reason)
self._detach_senders()
self._detach_servers()
self.reply_link.detach()
self._socket_connection.pyngus_conn.close()
def sasl_done(self, connection, pn_sasl, outcome):
"""This is a Pyngus callback invoked when the SASL handshake
has completed. The outcome of the handshake is passed in the outcome
argument.
"""
if outcome == proton.SASL.OK:
return
LOG.error("AUTHENTICATION FAILURE: Cannot connect to "
"%(hostname)s:%(port)s as user %(username)s",
{'hostname': self.hosts.current.hostname,
'port': self.hosts.current.port,
'username': self.hosts.current.username})
# pyngus will invoke connection_failed() eventually
def _handle_connection_loss(self, reason):
"""The connection to the messaging service has been lost. Try to
reestablish the connection/failover if not shutting down the driver.
"""
self.addresser = None
self._socket_connection.close()
if self._closing:
# we're in the middle of shutting down the driver anyways,
# just consider it done:
self.processor.shutdown()
else:
# for some reason, we've lost the connection to the messaging
# service. Try to re-establish the connection:
if not self._reconnecting:
self._reconnecting = True
self.processor.wakeup(lambda: self._hard_reset(reason))
LOG.info("Delaying reconnect attempt for %d seconds",
self._delay)
self.processor.defer(lambda: self._do_reconnect(reason),
self._delay)
self._delay = min(self._delay * self.conn_retry_backoff,
self.conn_retry_interval_max)
if self._link_maint_timer:
self._link_maint_timer.cancel()
self._link_maint_timer = None
def _do_reconnect(self, reason):
"""Invoked on connection/socket failure, failover and re-connect to the
messaging service.
"""
self._reconnecting = False
if not self._closing:
host = self.hosts.next()
LOG.info("Reconnecting to: %(hostname)s:%(port)s",
{'hostname': host.hostname, 'port': host.port})
self.processor.wakeup(lambda: self._do_connect())
def _hard_reset(self, reason):
"""Reset the controller to its pre-connection state"""
# note well: since this method destroys the connection, it cannot be
# invoked directly from a pyngus callback. Use processor.defer() to
# run this method on the main loop instead.
for sender in self._purged_senders:
sender.destroy(reason)
del self._purged_senders[:]
self._active_senders.clear()
unused = []
for key, sender in self._all_senders.items():
# clean up any sender links that no longer have messages to send
if sender.pending_messages == 0:
unused.append(key)
else:
sender.reset(reason)
self._active_senders.add(key)
for key in unused:
self._all_senders[key].destroy(reason)
del self._all_senders[key]
for servers in self._servers.values():
for server in servers.values():
server.reset()
if self.reply_link:
self.reply_link.destroy()
self.reply_link = None
if self._socket_connection:
self._socket_connection.reset()
def _detach_senders(self):
"""Close all sender links"""
for sender in self._all_senders.values():
sender.detach()
def _detach_servers(self):
"""Close all listener links"""
for servers in self._servers.values():
for server in servers.values():
server.detach()
def _purge_sender_links(self):
"""Purge inactive sender links"""
if not self._closing:
# destroy links that have already been closed
for sender in self._purged_senders:
sender.destroy("Idle link purged")
del self._purged_senders[:]
# determine next set to purge
purge = set(self._all_senders.keys()) - self._active_senders
for key in purge:
sender = self._all_senders[key]
if not sender.pending_messages and not sender.unacked_messages:
sender.detach()
self._purged_senders.append(self._all_senders.pop(key))
self._active_senders.clear()
self._link_maint_timer = \
self.processor.defer(self._purge_sender_links,
self._link_maint_timeout)
@property
def _active(self):
# Is the connection up
return (self._socket_connection and
self._socket_connection.pyngus_conn and
self._socket_connection.pyngus_conn.active)