Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


A modular OpenStack notification event processor/broadcaster written in Python.

Yagi is designed to efficiently gather amqp messages in the json format used by OpenStack projects notification busses, from a large and configurable number of queues, and proccess them through an extensible set of simple handlers.

Handlers are simple to write, and can be chained in a WSGI-like architecture. Yagi handles fetching messages, a batch at a time, and passes fetched messages to each handler, handling AMQP message semantics so the handlers can concentrate on the task at hand.

In addition, a feed daemon is included that can generate a paged Atom feed of notification events that have been persisted in a datastore.

Available Handlers

  • AtomPub: Formats notifications in Atom format, and pushes them to a feed server using the AtomPub protocol. Useful with feed servers such as AtomHopper (
  • Redis: Persists notifications to a Redis database. Can be used with Yagi's feed daemon.
  • PubSubHubub: Pings a pubsubhubub hub when notifications arrive. Together with a hub, and yagi's feed daemon, this can enable publish/subscribe subscriptions to notification events.
  • StackTackPing: Works with the StackTach openstack monitoring tool to monitor event feeds. If you are using Yagi to provide feeds of openstack notifications, this will ping stacktach when those feeds are updated, informing it of the success or failure of the updates, letting you catch if the feed server is down, or some system is dropping events.
  • Elasticsearch: Persists distilled compute.instance.exists, .verified, and .failed events to Elasticsearch. Additionally if the CufPub handler is present earlier in the chain pushes synthetic .cuf events as well in response to successful CufPub events.

Installation and running

The current version of Yagi can be fetched from the code repository at:

cd to the yagi directory and run:

sudo python install

The launch the yagi process:


An altername config file may be passed to yagi like this:

yagi-event -c /path/to/config/file

Yagi does not daemonize. use your favorite daemon manager to do that.


A sample yagi.conf can be found in the etc directory.

Sections to note:

  • rabbit_broker: Your rabbit connection info goes here.
  • event_feed: If using the feed daemon, remember to set the feed_host to the name of the host it is running on. This allows yagi to correctly construct links in the feed.
  • persistence: If using the redis handler, put your reddisc connection info here.
  • consumers: the 'queues' config variable lists the queues yagi should listen on.
  • consumer:$queue_name: For each queue Yagi is listening on there should be a consumer section in the config file (for example if you have a queue named some.queue listed in the [consumers] section, there should be a [consumer:some.queue] section with configuration for that specific queue.) This should list properties for the queue, such as if itshould be durable. Important variables here are 'apps', which is a comma separated list of handlers that messages from that queues should be passed to, and 'max_messages' which is the maximum number of messages that Yagi will pull from that queue at one time. (it will then go to the next queue, eventually coming back around, if there are still messages waiting)

Handlers may also have their own, additional configuration. This is usually found in a section named after the handler (all lowercase, one word)


Yagi is designed to scale by running multiple processes. Simply launch as many yagi-event processes as you need to handle your load. (yagi-event is fairly lightweight)


  • anyjson
  • argparse
  • feedgenerator
  • httplib2
  • requests (eventually httplib2 will be replaced with requests)
  • redis (if using redis handler)
  • webob
  • eventlet
  • python-dateutil
  • daemon
  • pubsubhubbub_publish (if using pubsubhubub handler) (available under the publisher_clients folder after checking out the project from Google Code NOTE: the plan is to replace this dependency later with our implementation
  • carrot (if using Rabbit)
  • routes

Running the feed daemon

Simply run: yagi-feed

Setting up a hub (if using PubSubHubub)

Download the Google App Engine SDK for Linux and add it to your path

Then checkout the reference hub.

svn checkout pubsubhubbub-read-only

Install pubsubhubbub_publisher for python

cd pubsubhubbub-read-only/publisher_clients/python
sudo python install

Start the hub

cd pubsubhubbub-read-only hub/ -p<port number specified in yagi.conf>

Testing subscriptions for PubSubHubub.

cd yagi

# You'll want to run this in multiple screen windows or terminal 
# sessions, as the callback process won't daemonize

python subscriber/ <sub_port>
python subscriber/ <topic> <callback> <hub>

# I usually load other/ in an iPython session
cd other
import push_rabbit

# the cast below is assuming you setting up yagi to listen 
# on a queue named 'notifications.warn'
push_rabbit.cast(dict(a=3), 'instance', 'notifications', 'warn')

You should see XML content being pushed to your window running, above.

Brokers, Consumers and Handlers(Apps)

The top-level object for Yagi is the Event Broker, which is defined in the configuration file as:

event_driver =

The Broker will create a Consumer object for each input queue defined.

queues = queue1, queue2

apps = yagi.handler.handler1, myhandlers.handler
routing_key = notifications.warn

apps = ...

The Consumer grabs the notifications as they come in and give them to the handlers for processing. Handlers are also known as apps for legacy reasons. Handlers are chained together, and it's the responability of the handler to call child handlers if they think it's appropriate (the default behavior). However, some handler may wish to kill the chain on an error some other condition. Each handler also gets to process the messages themself. First notifications are checked against the filter list for that handler and, if accepted, the handle_messages() method is called. Child handlers always get called unfiltered.

cufpub = compute.instance.exists.verified,compute.instance.exists

Currently, filters are applied to all handlers, but this should change to a per-handler filter list.

Look at for details.


A modular OpenStack notification event processor/broadcaster written in Python.






No packages published