Restore static direct connections which may be useful
for services running RPC on controllers only and talking
to agents via other means (ssh for example).
Change-Id: Icbe45978cb4a8ba5db74e5593ea7be23cba3f44e
ZeroMQ driver options are current stored into the DEFAULT group.
This change makes the zmq configuration clearer by putting its
options into oslo_messaging_zmq group.
Change-Id: Ia00fda005b1664750d2646f8c82ebdf295b156fb
Closes-bug: #1417040
Co-Authored-By: Oleksii Zamiatin <ozamiatin@mirantis.com>
This patch fixes the problem of sending messages in the following
configuration: use_router_proxy=True and use_pub_sub=False. Right
now we have incorrect both cast and fanout messages: cast acts as
fanout (instead of round-robin) and fanout duplicate each message
twice.
Change-Id: I2b039aa462ddde300fd50b3754e4fe66abf6e635
In this change router was redesigned in a way most
appropriate for routing concept of zmq.ROUTER socket.
DEALER(cli)-ROUTER(proxy)-DEALER(srv) instead of
DEALER-ROUTER-DEALER-ROUTER (3 layers instead of 4)
The main reason is to use zmq.DEALER identity in message
routing. For this reason DealerConsumer was introduced
server-side. RouterConsumer is left for peer-to-peer
DEALER-ROUTER deployment option.
Also handled assertions in receive-methods in order
to not stop server when received message with wrong format.
Change-Id: If25edf500fa8d220d4233bb13d67121824e841c6
Closes-Bug: #1558601
Related-Bug: #1555007
In this change 'use_router_proxy' option was added
to switch between direct connections and proxy.
Proxy was reimplemented and splitted onto two
types of proxies:
* PUBLISHER proxy for fanout pattern
* ROUTER proxy for direct messaging
Each type of proxy is configured over command line
argument --type.
Deployment guide was updated accordingly to the change.
Change-Id: If36e9c26e2a8ebe622cfa7e9f2a07b1a69aabe34
Closes-Bug: #1555007
The main issue with eventlet.green.zmq is that libzmq as a C-library
is completely monkey-patch unfriendly. So any blocking call inside
the native library makes calling process stuck. We can't avoid this
actually in an absolutely normal situation when a client appears
earlier than listener we have all client process get stuck until listener
raised. If the listener for example is also blocked awaiting for some
other service to appear we have a chain of locks which may occasionally
result in a dead-lock. The other situation with Notifier is quite similar.
For that reason zmq-broker was restored, but now it serves as an outgoing
queue on a client side. Servers remained the same dynamically port-binded.
Now all clients can still use green-zmq, but presence of the broker-queue
on a host guarantees that green threads will never blocked in a client
because all messages will wait their listeners inside the broker queue.
The broker process's modules are not monkey-patched, they make use of native
threading and native zmq.
Possibility to run without broker also remains. The option zmq_use_broker
introduced for that reason.
Closes-Bug: #1497315
Change-Id: I786b100fd6ee1cf4b99139db0ca044d358d36345
As far as we use redis as a name service we don't need a proxy,
becase we can pass binded port over name service too.
Change-Id: I59bbe2b34dcedfeef113ef06d6a988e1c413405e
- Fixed universal proxy to not get stuck with multiple backends
- Fixed threading pollers/executors (proxy side)
- Driver option to switch green/no-green impl.
- Swtiched to no-green in real-world proxy (green left for unit tests)
- Minor names fixes in serializer
Change-Id: Id6508101521d8914228c639ed58ecd29db0ef456
This adds a setup script for each tox functional target to start a
rabbitmq-server, qpidd or redis daemon dedicated for the functional
testing.
This script is responsible to spawn a preconfigured daemon needed for
the functional tests.
This also changes the gate script to just install the required packages
instead of setup a devstack.
This also fixes the zmq config options loading in tests
Closes-bug: #1442612
Change-Id: I27eb2c1d3d0ca67aa361c83e41372138e03d9bdd
ozamiatin